mirrors
/
micropython
mirror of https://github.com/micropython/micropython.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
micropython/py/emitnative.c

2803 lines
115 KiB
C
Raw Normal View History

Add license header to (almost) all files. Blanket wide to all .c and .h files. Some files originating from ST are difficult to deal with (license wise) so it was left out of those. Also merged modpyb.h, modos.h, modstm.h and modtime.h in stmhal/.
2014-05-03 23:27:38 +01:00
/*
all: Use the name MicroPython consistently in comments There were several different spellings of MicroPython present in comments, when there should be only one.
2017-06-30 08:22:17 +01:00
* This file is part of the MicroPython project, http://micropython.org/
Add license header to (almost) all files. Blanket wide to all .c and .h files. Some files originating from ST are difficult to deal with (license wise) so it was left out of those. Also merged modpyb.h, modos.h, modstm.h and modtime.h in stmhal/.
2014-05-03 23:27:38 +01:00
*
* The MIT License (MIT)
*
* Copyright (c) 2013, 2014 Damien P. George
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*/
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
// Essentially normal Python has 1 type: Python objects
// Viper has more than 1 type, and is just a more complicated (a superset of) Python.
// If you declare everything in Viper as a Python object (ie omit type decls) then
// it should in principle be exactly the same as Python native.
// Having types means having more opcodes, like binary_op_nat_nat, binary_op_nat_obj etc.
// In practice we won't have a VM but rather do this in asm which is actually very minimal.
// Because it breaks strict Python equivalence it should be a completely separate
// decorator. It breaks equivalence because overflow on integers wraps around.
// It shouldn't break equivalence if you don't use the new types, but since the
// type decls might be used in normal Python for other reasons, it's probably safest,
// cleanest and clearest to make it a separate decorator.
// Actually, it does break equivalence because integers default to native integers,
// not Python objects.
// for x in l[0:8]: can be compiled into a native loop if l has pointer type
#include <stdio.h>
#include <string.h>
#include <assert.h>
py: Move to guarded includes, everywhere in py/ core. Addresses issue #1022.
2015-01-01 20:27:54 +00:00
#include "py/emit.h"
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
#include "py/bc.h"
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
py: Add verbose debug compile-time flag MICROPY_DEBUG_VERBOSE. It enables all the DEBUG_printf outputs in the py/ source code.
2017-07-24 17:55:14 +01:00
#if MICROPY_DEBUG_VERBOSE // print debugging info
py: Implement more features in native emitter. On x64, native emitter now passes 70 of the tests.
2014-04-06 12:58:40 +01:00
#define DEBUG_PRINT (1)
#define DEBUG_printf DEBUG_printf
#else // don't print debugging info
#define DEBUG_printf(...) (void)0
#endif
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
// wrapper around everything in this file
py: Refactor how native emitter code is compiled with a file per arch. Instead of emitnative.c having configuration code for each supported architecture, and then compiling this file multiple times with different macros defined, this patch adds a file per architecture with the necessary code to configure the native emitter. These files then #include the emitnative.c file. This simplifies emitnative.c (which is already very large), and simplifies the build system because emitnative.c no longer needs special handling for compilation and qstr extraction.
2018-04-10 06:06:47 +01:00
#if N_X64 || N_X86 || N_THUMB || N_ARM || N_XTENSA
py: Move arch-specific assembler macros from emitnative to asmXXX.h.
2016-12-08 06:47:17 +00:00
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
// C stack layout for native functions:
py/emitnative: Reorder native state on C stack so nlr_buf_t is first. The nlr_buf_t doesn't need to be part of the Python value stack (as it was before this commit), it's simpler to have it separated as auxiliary state that lives on the C stack. This will help adding yield support because in that case the nlr_buf_t and Python value stack live in separate memory areas (C stack and heap respectively).
2018-10-01 03:36:21 +01:00
// 0: nlr_buf_t [optional]
// emit->code_state_start: mp_code_state_t
// emit->stack_start: Python object stack | emit->n_state
// locals (reversed, L0 at end) |
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
//
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
// C stack layout for native generator functions:
// 0=emit->stack_start: nlr_buf_t
//
// Then REG_GENERATOR_STATE points to:
// 0=emit->code_state_start: mp_code_state_t
// emit->stack_start: Python object stack | emit->n_state
// locals (reversed, L0 at end) |
//
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
// C stack layout for viper functions:
py/emitnative: Reorder native state on C stack so nlr_buf_t is first. The nlr_buf_t doesn't need to be part of the Python value stack (as it was before this commit), it's simpler to have it separated as auxiliary state that lives on the C stack. This will help adding yield support because in that case the nlr_buf_t and Python value stack live in separate memory areas (C stack and heap respectively).
2018-10-01 03:36:21 +01:00
// 0: nlr_buf_t [optional]
// emit->code_state_start: fun_obj, old_globals [optional]
// emit->stack_start: Python object stack | emit->n_state
// locals (reversed, L0 at end) |
// (L0-L2 may be in regs instead)
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
// Word index of nlr_buf_t.ret_val
#define NLR_BUF_IDX_RET_VAL (1)
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
// Whether the viper function needs access to fun_obj
#define NEED_FUN_OBJ(emit) ((emit)->scope->exc_stack_size > 0 \
|| ((emit)->scope->scope_flags & (MP_SCOPE_FLAG_REFGLOBALS | MP_SCOPE_FLAG_HASCONSTS)))
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
// Whether the native/viper function needs to be wrapped in an exception handler
py: Fix native functions so they run with their correct globals context. Prior to this commit a function compiled with the native decorator @micropython.native would not work correctly when accessing global variables, because the globals dict was not being set upon function entry. This commit fixes this problem by, upon function entry, setting as the current globals dict the globals dict context the function was defined within, as per normal Python semantics, and as bytecode does. Upon function exit the original globals dict is restored. In order to restore the globals dict when an exception is raised the native function must guard its internals with an nlr_push/nlr_pop pair. Because this push/pop is relatively expensive, in both C stack usage for the nlr_buf_t and CPU execution time, the implementation here optimises things as much as possible. First, the compiler keeps track of whether a function even needs to access global variables. Using this information the native emitter then generates three different kinds of code: 1. no globals used, no exception handlers: no nlr handling code and no setting of the globals dict. 2. globals used, no exception handlers: an nlr_buf_t is allocated on the C stack but it is not used if the globals dict is unchanged, saving execution time because nlr_push/nlr_pop don't need to run. 3. function has exception handlers, may use globals: an nlr_buf_t is allocated and nlr_push/nlr_pop are always called. In the end, native functions that don't access globals and don't have exception handlers will run more efficiently than those that do. Fixes issue #1573.
2018-09-13 13:03:48 +01:00
#define NEED_GLOBAL_EXC_HANDLER(emit) ((emit)->scope->exc_stack_size > 0 \
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
|| ((emit)->scope->scope_flags & (MP_SCOPE_FLAG_GENERATOR | MP_SCOPE_FLAG_REFGLOBALS)))
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
// Whether registers can be used to store locals (only true if there are no
// exception handlers, because otherwise an nlr_jump will restore registers to
// their state at the start of the function and updates to locals will be lost)
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
#define CAN_USE_REGS_FOR_LOCALS(emit) ((emit)->scope->exc_stack_size == 0 && !(emit->scope->scope_flags & MP_SCOPE_FLAG_GENERATOR))
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
// Indices within the local C stack for various variables
py/emitnative: Reorder native state on C stack so nlr_buf_t is first. The nlr_buf_t doesn't need to be part of the Python value stack (as it was before this commit), it's simpler to have it separated as auxiliary state that lives on the C stack. This will help adding yield support because in that case the nlr_buf_t and Python value stack live in separate memory areas (C stack and heap respectively).
2018-10-01 03:36:21 +01:00
#define LOCAL_IDX_EXC_VAL(emit) (NLR_BUF_IDX_RET_VAL)
#define LOCAL_IDX_EXC_HANDLER_PC(emit) (NLR_BUF_IDX_LOCAL_1)
#define LOCAL_IDX_EXC_HANDLER_UNWIND(emit) (NLR_BUF_IDX_LOCAL_2)
#define LOCAL_IDX_RET_VAL(emit) (NLR_BUF_IDX_LOCAL_3)
#define LOCAL_IDX_FUN_OBJ(emit) ((emit)->code_state_start + offsetof(mp_code_state_t, fun_bc) / sizeof(uintptr_t))
#define LOCAL_IDX_OLD_GLOBALS(emit) ((emit)->code_state_start + offsetof(mp_code_state_t, ip) / sizeof(uintptr_t))
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
#define LOCAL_IDX_GEN_PC(emit) ((emit)->code_state_start + offsetof(mp_code_state_t, ip) / sizeof(uintptr_t))
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
#define LOCAL_IDX_LOCAL_VAR(emit, local_num) ((emit)->stack_start + (emit)->n_state - 1 - (local_num))
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
#define REG_GENERATOR_STATE (REG_LOCAL_3)
py: Make viper codegen raise proper exception (ViperTypeError) on error. This fixes a long standing problem that viper code generation gave terrible error messages, and actually no errors on pyboard where assertions are disabled. Now all compile-time errors are raised as proper Python exceptions, and are of type ViperTypeError. Addresses issue #940.
2015-04-20 14:29:31 +01:00
#define EMIT_NATIVE_VIPER_TYPE_ERROR(emit, ...) do { \
*emit->error_slot = mp_obj_new_exception_msg_varg(&mp_type_ViperTypeError, __VA_ARGS__); \
} while (0)
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
typedef enum {
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
STACK_VALUE,
STACK_REG,
STACK_IMM,
} stack_info_kind_t;
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
py: Extend native type-sig to use 4 bits, so uint is separate to ptr. Before this patch, the native types for uint and ptr/ptr8/ptr16/ptr32 all overlapped and it was possible to make a mistake in casting. Now, these types are all separate and any coding mistakes will be raised as runtime errors.
2016-02-02 23:16:05 +00:00
// these enums must be distinct and the bottom 4 bits
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
// must correspond to the correct MP_NATIVE_TYPE_xxx value
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
typedef enum {
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
VTYPE_PYOBJ = 0x00 | MP_NATIVE_TYPE_OBJ,
VTYPE_BOOL = 0x00 | MP_NATIVE_TYPE_BOOL,
VTYPE_INT = 0x00 | MP_NATIVE_TYPE_INT,
VTYPE_UINT = 0x00 | MP_NATIVE_TYPE_UINT,
py: Extend native type-sig to use 4 bits, so uint is separate to ptr. Before this patch, the native types for uint and ptr/ptr8/ptr16/ptr32 all overlapped and it was possible to make a mistake in casting. Now, these types are all separate and any coding mistakes will be raised as runtime errors.
2016-02-02 23:16:05 +00:00
VTYPE_PTR = 0x00 | MP_NATIVE_TYPE_PTR,
VTYPE_PTR8 = 0x00 | MP_NATIVE_TYPE_PTR8,
VTYPE_PTR16 = 0x00 | MP_NATIVE_TYPE_PTR16,
VTYPE_PTR32 = 0x00 | MP_NATIVE_TYPE_PTR32,
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
py: Extend native type-sig to use 4 bits, so uint is separate to ptr. Before this patch, the native types for uint and ptr/ptr8/ptr16/ptr32 all overlapped and it was possible to make a mistake in casting. Now, these types are all separate and any coding mistakes will be raised as runtime errors.
2016-02-02 23:16:05 +00:00
VTYPE_PTR_NONE = 0x50 | MP_NATIVE_TYPE_PTR,
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
py: Implement ptr32 load and store in viper emitter.
2015-10-13 00:50:17 +01:00
VTYPE_UNBOUND = 0x60 | MP_NATIVE_TYPE_OBJ,
VTYPE_BUILTIN_CAST = 0x70 | MP_NATIVE_TYPE_OBJ,
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
} vtype_kind_t;
py: Make viper codegen raise proper exception (ViperTypeError) on error. This fixes a long standing problem that viper code generation gave terrible error messages, and actually no errors on pyboard where assertions are disabled. Now all compile-time errors are raised as proper Python exceptions, and are of type ViperTypeError. Addresses issue #940.
2015-04-20 14:29:31 +01:00
STATIC qstr vtype_to_qstr(vtype_kind_t vtype) {
switch (vtype) {
case VTYPE_PYOBJ: return MP_QSTR_object;
case VTYPE_BOOL: return MP_QSTR_bool;
case VTYPE_INT: return MP_QSTR_int;
case VTYPE_UINT: return MP_QSTR_uint;
case VTYPE_PTR: return MP_QSTR_ptr;
case VTYPE_PTR8: return MP_QSTR_ptr8;
case VTYPE_PTR16: return MP_QSTR_ptr16;
py: Implement ptr32 load and store in viper emitter.
2015-10-13 00:50:17 +01:00
case VTYPE_PTR32: return MP_QSTR_ptr32;
py: Make viper codegen raise proper exception (ViperTypeError) on error. This fixes a long standing problem that viper code generation gave terrible error messages, and actually no errors on pyboard where assertions are disabled. Now all compile-time errors are raised as proper Python exceptions, and are of type ViperTypeError. Addresses issue #940.
2015-04-20 14:29:31 +01:00
case VTYPE_PTR_NONE: default: return MP_QSTR_None;
}
}
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
typedef struct _stack_info_t {
vtype_kind_t vtype;
stack_info_kind_t kind;
union {
int u_reg;
Rename machine_(u)int_t to mp_(u)int_t. See discussion in issue #50.
2014-07-03 13:25:24 +01:00
mp_int_t u_imm;
py: Don't use anonymous unions, name them instead. This makes the code (more) compatible with the C99 standard.
2015-01-24 23:14:12 +00:00
} data;
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
} stack_info_t;
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
#define UNWIND_LABEL_UNUSED (0x7fff)
#define UNWIND_LABEL_DO_FINAL_UNWIND (0x7ffe)
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
typedef struct _exc_stack_entry_t {
uint16_t label : 15;
uint16_t is_finally : 1;
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
uint16_t unwind_label : 15;
uint16_t is_active : 1;
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
} exc_stack_entry_t;
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
struct _emit_t {
py: Make viper codegen raise proper exception (ViperTypeError) on error. This fixes a long standing problem that viper code generation gave terrible error messages, and actually no errors on pyboard where assertions are disabled. Now all compile-time errors are raised as proper Python exceptions, and are of type ViperTypeError. Addresses issue #940.
2015-04-20 14:29:31 +01:00
mp_obj_t *error_slot;
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
uint *label_slot;
uint exit_label;
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
int pass;
bool do_viper_types;
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
mp_uint_t local_vtype_alloc;
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
vtype_kind_t *local_vtype;
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
mp_uint_t stack_info_alloc;
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
stack_info_t *stack_info;
py: Partially fix viper multi-comparison; add test for it.
2014-10-19 19:00:51 +01:00
vtype_kind_t saved_stack_vtype;
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
size_t exc_stack_alloc;
size_t exc_stack_size;
exc_stack_entry_t *exc_stack;
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
int prelude_offset;
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
int start_offset;
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
int n_state;
py/emitnative: Reorder native state on C stack so nlr_buf_t is first. The nlr_buf_t doesn't need to be part of the Python value stack (as it was before this commit), it's simpler to have it separated as auxiliary state that lives on the C stack. This will help adding yield support because in that case the nlr_buf_t and Python value stack live in separate memory areas (C stack and heap respectively).
2018-10-01 03:36:21 +01:00
uint16_t code_state_start;
uint16_t stack_start;
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
int stack_size;
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
uint16_t const_table_cur_obj;
uint16_t const_table_num_obj;
uint16_t const_table_cur_raw_code;
py/emitnative: Change type of const_table from uintptr_t to mp_uint_t. This matches how bytecode does it, and matches the signature of mp_emit_glue_assign_native. Since the native emitter doesn't support nan-boxing uintptr_t and mp_uint_t are anyway the same bit-width.
2018-09-27 15:04:10 +01:00
mp_uint_t *const_table;
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
py: Add support to save native, viper and asm code to .mpy files. This commit adds support for saving and loading .mpy files that contain native code (native, viper and inline-asm). A lot of the ground work was already done for this in the form of removing pointers from generated native code. The changes here are mainly to link in qstr values to the native code, and change the format of .mpy files to contain native code blocks (possibly mixed with bytecode). A top-level summary: - @micropython.native, @micropython.viper and @micropython.asm_thumb/ asm_xtensa are now allowed in .py files when compiling to .mpy, and they work transparently to the user. - Entire .py files can be compiled to native via mpy-cross -X emit=native and for the most part the generated .mpy files should work the same as their bytecode version. - The .mpy file format is changed to 1) specify in the header if the file contains native code and if so the architecture (eg x86, ARMV7M, Xtensa); 2) for each function block the kind of code is specified (bytecode, native, viper, asm). - When native code is loaded from a .mpy file the native code must be modified (in place) to link qstr values in, just like bytecode (see py/persistentcode.c:arch_link_qstr() function). In addition, this now defines a public, native ABI for dynamically loadable native code generated by other languages, like C.
2019-02-21 04:18:33 +00:00
#if MICROPY_PERSISTENT_CODE_SAVE
uint16_t qstr_link_cur;
mp_qstr_link_entry_t *qstr_link;
#endif
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
bool last_emit_was_return_value;
scope_t *scope;
py: Add support for emitting native x86 machine code.
2014-09-06 23:06:36 +01:00
ASM_T *as;
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
};
py/emitnative: Use small tables to simplify handling of local regs.
2018-08-15 01:52:38 +01:00
STATIC const uint8_t reg_local_table[REG_LOCAL_NUM] = {REG_LOCAL_1, REG_LOCAL_2, REG_LOCAL_3};
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
STATIC void emit_native_global_exc_entry(emit_t *emit);
STATIC void emit_native_global_exc_exit(emit_t *emit);
py: Add support to save native, viper and asm code to .mpy files. This commit adds support for saving and loading .mpy files that contain native code (native, viper and inline-asm). A lot of the ground work was already done for this in the form of removing pointers from generated native code. The changes here are mainly to link in qstr values to the native code, and change the format of .mpy files to contain native code blocks (possibly mixed with bytecode). A top-level summary: - @micropython.native, @micropython.viper and @micropython.asm_thumb/ asm_xtensa are now allowed in .py files when compiling to .mpy, and they work transparently to the user. - Entire .py files can be compiled to native via mpy-cross -X emit=native and for the most part the generated .mpy files should work the same as their bytecode version. - The .mpy file format is changed to 1) specify in the header if the file contains native code and if so the architecture (eg x86, ARMV7M, Xtensa); 2) for each function block the kind of code is specified (bytecode, native, viper, asm). - When native code is loaded from a .mpy file the native code must be modified (in place) to link qstr values in, just like bytecode (see py/persistentcode.c:arch_link_qstr() function). In addition, this now defines a public, native ABI for dynamically loadable native code generated by other languages, like C.
2019-02-21 04:18:33 +00:00
STATIC void emit_native_load_const_obj(emit_t *emit, mp_obj_t obj);
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
emit_t *EXPORT_FUN(new)(mp_obj_t *error_slot, uint *label_slot, mp_uint_t max_num_labels) {
py, compiler: Improve passes; add an extra pass for native emitter.
2014-05-07 17:24:22 +01:00
emit_t *emit = m_new0(emit_t, 1);
py: Make viper codegen raise proper exception (ViperTypeError) on error. This fixes a long standing problem that viper code generation gave terrible error messages, and actually no errors on pyboard where assertions are disabled. Now all compile-time errors are raised as proper Python exceptions, and are of type ViperTypeError. Addresses issue #940.
2015-04-20 14:29:31 +01:00
emit->error_slot = error_slot;
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
emit->label_slot = label_slot;
py/emitnative: Allocate space for local stack info as it's needed.
2018-08-07 07:05:44 +01:00
emit->stack_info_alloc = 8;
emit->stack_info = m_new(stack_info_t, emit->stack_info_alloc);
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
emit->exc_stack_alloc = 8;
emit->exc_stack = m_new(exc_stack_entry_t, emit->exc_stack_alloc);
py: Factor out common code from assemblers into asmbase.[ch]. All assemblers should "derive" from mp_asm_base_t.
2016-11-27 22:24:50 +00:00
emit->as = m_new0(ASM_T, 1);
mp_asm_base_init(&emit->as->base, max_num_labels);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
return emit;
}
py: Improve freeing of emitters in mp_compile. There can be multiple emitters allocated during compile (eg byte code and native).
2014-01-24 22:42:28 +00:00
void EXPORT_FUN(free)(emit_t *emit) {
py: Factor out common code from assemblers into asmbase.[ch]. All assemblers should "derive" from mp_asm_base_t.
2016-11-27 22:24:50 +00:00
mp_asm_base_deinit(&emit->as->base, false);
m_del_obj(ASM_T, emit->as);
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
m_del(exc_stack_entry_t, emit->exc_stack, emit->exc_stack_alloc);
py, compiler: Improve passes; add an extra pass for native emitter.
2014-05-07 17:24:22 +01:00
m_del(vtype_kind_t, emit->local_vtype, emit->local_vtype_alloc);
m_del(stack_info_t, emit->stack_info, emit->stack_info_alloc);
Add support for freeing code emitter objects at the end of compilation.
2014-01-24 14:20:11 +00:00
m_del_obj(emit_t, emit);
}
py: Make viper functions have the same entry signature as native. This commit makes viper functions have the same signature as native functions, at the level of the emitter/assembler. This means that viper functions can now be wrapped in the same uPy object as native functions. Viper functions are now responsible for parsing their arguments (before it was done by the runtime), and this makes calling them more efficient (in most cases) because the viper entry code can be custom generated to suit the signature of the function. This change also opens the way forward for viper functions to take arbitrary numbers of arguments, and for them to handle globals correctly, among other things.
2018-09-14 08:40:59 +01:00
STATIC void emit_call_with_imm_arg(emit_t *emit, mp_fun_kind_t fun_kind, mp_int_t arg_val, int arg_reg);
py: Fix bug in native emitter when closing over an argument.
2015-04-03 16:03:24 +01:00
py/emitnative: Put None/False/True in global native const table. So these constant objects can be loaded by dereferencing the REG_FUN_TABLE pointer instead of loading immediate values. This reduces the size of generated native code (when such constants are used), and means that pointers to these constants are no longer stored in the assembly code.
2018-10-14 13:56:46 +01:00
STATIC void emit_native_mov_reg_const(emit_t *emit, int reg_dest, int const_val) {
ASM_LOAD_REG_REG_OFFSET(emit->as, reg_dest, REG_FUN_TABLE, const_val);
}
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
STATIC void emit_native_mov_state_reg(emit_t *emit, int local_num, int reg_src) {
if (emit->scope->scope_flags & MP_SCOPE_FLAG_GENERATOR) {
ASM_STORE_REG_REG_OFFSET(emit->as, reg_src, REG_GENERATOR_STATE, local_num);
} else {
ASM_MOV_LOCAL_REG(emit->as, local_num, reg_src);
}
}
STATIC void emit_native_mov_reg_state(emit_t *emit, int reg_dest, int local_num) {
if (emit->scope->scope_flags & MP_SCOPE_FLAG_GENERATOR) {
ASM_LOAD_REG_REG_OFFSET(emit->as, reg_dest, REG_GENERATOR_STATE, local_num);
} else {
ASM_MOV_REG_LOCAL(emit->as, reg_dest, local_num);
}
}
STATIC void emit_native_mov_reg_state_addr(emit_t *emit, int reg_dest, int local_num) {
if (emit->scope->scope_flags & MP_SCOPE_FLAG_GENERATOR) {
ASM_MOV_REG_IMM(emit->as, reg_dest, local_num * ASM_WORD_SIZE);
ASM_ADD_REG_REG(emit->as, reg_dest, REG_GENERATOR_STATE);
} else {
ASM_MOV_REG_LOCAL_ADDR(emit->as, reg_dest, local_num);
}
}
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
STATIC void emit_native_mov_reg_qstr(emit_t *emit, int arg_reg, qstr qst) {
py: Add support to save native, viper and asm code to .mpy files. This commit adds support for saving and loading .mpy files that contain native code (native, viper and inline-asm). A lot of the ground work was already done for this in the form of removing pointers from generated native code. The changes here are mainly to link in qstr values to the native code, and change the format of .mpy files to contain native code blocks (possibly mixed with bytecode). A top-level summary: - @micropython.native, @micropython.viper and @micropython.asm_thumb/ asm_xtensa are now allowed in .py files when compiling to .mpy, and they work transparently to the user. - Entire .py files can be compiled to native via mpy-cross -X emit=native and for the most part the generated .mpy files should work the same as their bytecode version. - The .mpy file format is changed to 1) specify in the header if the file contains native code and if so the architecture (eg x86, ARMV7M, Xtensa); 2) for each function block the kind of code is specified (bytecode, native, viper, asm). - When native code is loaded from a .mpy file the native code must be modified (in place) to link qstr values in, just like bytecode (see py/persistentcode.c:arch_link_qstr() function). In addition, this now defines a public, native ABI for dynamically loadable native code generated by other languages, like C.
2019-02-21 04:18:33 +00:00
#if MICROPY_PERSISTENT_CODE_SAVE
size_t loc = ASM_MOV_REG_IMM_FIX_U16(emit->as, arg_reg, qst);
size_t link_idx = emit->qstr_link_cur++;
if (emit->pass == MP_PASS_EMIT) {
emit->qstr_link[link_idx].off = loc << 2 | 1;
emit->qstr_link[link_idx].qst = qst;
}
#else
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
ASM_MOV_REG_IMM(emit->as, arg_reg, qst);
py: Add support to save native, viper and asm code to .mpy files. This commit adds support for saving and loading .mpy files that contain native code (native, viper and inline-asm). A lot of the ground work was already done for this in the form of removing pointers from generated native code. The changes here are mainly to link in qstr values to the native code, and change the format of .mpy files to contain native code blocks (possibly mixed with bytecode). A top-level summary: - @micropython.native, @micropython.viper and @micropython.asm_thumb/ asm_xtensa are now allowed in .py files when compiling to .mpy, and they work transparently to the user. - Entire .py files can be compiled to native via mpy-cross -X emit=native and for the most part the generated .mpy files should work the same as their bytecode version. - The .mpy file format is changed to 1) specify in the header if the file contains native code and if so the architecture (eg x86, ARMV7M, Xtensa); 2) for each function block the kind of code is specified (bytecode, native, viper, asm). - When native code is loaded from a .mpy file the native code must be modified (in place) to link qstr values in, just like bytecode (see py/persistentcode.c:arch_link_qstr() function). In addition, this now defines a public, native ABI for dynamically loadable native code generated by other languages, like C.
2019-02-21 04:18:33 +00:00
#endif
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
}
STATIC void emit_native_mov_reg_qstr_obj(emit_t *emit, int reg_dest, qstr qst) {
py: Add support to save native, viper and asm code to .mpy files. This commit adds support for saving and loading .mpy files that contain native code (native, viper and inline-asm). A lot of the ground work was already done for this in the form of removing pointers from generated native code. The changes here are mainly to link in qstr values to the native code, and change the format of .mpy files to contain native code blocks (possibly mixed with bytecode). A top-level summary: - @micropython.native, @micropython.viper and @micropython.asm_thumb/ asm_xtensa are now allowed in .py files when compiling to .mpy, and they work transparently to the user. - Entire .py files can be compiled to native via mpy-cross -X emit=native and for the most part the generated .mpy files should work the same as their bytecode version. - The .mpy file format is changed to 1) specify in the header if the file contains native code and if so the architecture (eg x86, ARMV7M, Xtensa); 2) for each function block the kind of code is specified (bytecode, native, viper, asm). - When native code is loaded from a .mpy file the native code must be modified (in place) to link qstr values in, just like bytecode (see py/persistentcode.c:arch_link_qstr() function). In addition, this now defines a public, native ABI for dynamically loadable native code generated by other languages, like C.
2019-02-21 04:18:33 +00:00
#if MICROPY_PERSISTENT_CODE_SAVE
size_t loc = ASM_MOV_REG_IMM_FIX_WORD(emit->as, reg_dest, (mp_uint_t)MP_OBJ_NEW_QSTR(qst));
size_t link_idx = emit->qstr_link_cur++;
if (emit->pass == MP_PASS_EMIT) {
emit->qstr_link[link_idx].off = loc << 2 | 2;
emit->qstr_link[link_idx].qst = qst;
}
#else
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
ASM_MOV_REG_IMM(emit->as, reg_dest, (mp_uint_t)MP_OBJ_NEW_QSTR(qst));
py: Add support to save native, viper and asm code to .mpy files. This commit adds support for saving and loading .mpy files that contain native code (native, viper and inline-asm). A lot of the ground work was already done for this in the form of removing pointers from generated native code. The changes here are mainly to link in qstr values to the native code, and change the format of .mpy files to contain native code blocks (possibly mixed with bytecode). A top-level summary: - @micropython.native, @micropython.viper and @micropython.asm_thumb/ asm_xtensa are now allowed in .py files when compiling to .mpy, and they work transparently to the user. - Entire .py files can be compiled to native via mpy-cross -X emit=native and for the most part the generated .mpy files should work the same as their bytecode version. - The .mpy file format is changed to 1) specify in the header if the file contains native code and if so the architecture (eg x86, ARMV7M, Xtensa); 2) for each function block the kind of code is specified (bytecode, native, viper, asm). - When native code is loaded from a .mpy file the native code must be modified (in place) to link qstr values in, just like bytecode (see py/persistentcode.c:arch_link_qstr() function). In addition, this now defines a public, native ABI for dynamically loadable native code generated by other languages, like C.
2019-02-21 04:18:33 +00:00
#endif
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
}
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
#define emit_native_mov_state_imm_via(emit, local_num, imm, reg_temp) \
do { \
ASM_MOV_REG_IMM((emit)->as, (reg_temp), (imm)); \
emit_native_mov_state_reg((emit), (local_num), (reg_temp)); \
} while (false)
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_native_start_pass(emit_t *emit, pass_kind_t pass, scope_t *scope) {
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
DEBUG_printf("start_pass(pass=%u, scope=%p)\n", pass, scope);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit->pass = pass;
py/emit: Remove need to call set_native_type to set native/viper mode. The native emitter can easily determine the mode via scope->emit_options.
2018-09-15 03:17:14 +01:00
emit->do_viper_types = scope->emit_options == MP_EMIT_OPT_VIPER;
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit->stack_size = 0;
py/emitnative: Adjust accounting of size of const_table. n_obj no longer includes a count for mp_fun_table to make it a bit simpler.
2019-02-25 04:57:35 +00:00
emit->const_table_cur_obj = 0;
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
emit->const_table_cur_raw_code = 0;
py: Add support to save native, viper and asm code to .mpy files. This commit adds support for saving and loading .mpy files that contain native code (native, viper and inline-asm). A lot of the ground work was already done for this in the form of removing pointers from generated native code. The changes here are mainly to link in qstr values to the native code, and change the format of .mpy files to contain native code blocks (possibly mixed with bytecode). A top-level summary: - @micropython.native, @micropython.viper and @micropython.asm_thumb/ asm_xtensa are now allowed in .py files when compiling to .mpy, and they work transparently to the user. - Entire .py files can be compiled to native via mpy-cross -X emit=native and for the most part the generated .mpy files should work the same as their bytecode version. - The .mpy file format is changed to 1) specify in the header if the file contains native code and if so the architecture (eg x86, ARMV7M, Xtensa); 2) for each function block the kind of code is specified (bytecode, native, viper, asm). - When native code is loaded from a .mpy file the native code must be modified (in place) to link qstr values in, just like bytecode (see py/persistentcode.c:arch_link_qstr() function). In addition, this now defines a public, native ABI for dynamically loadable native code generated by other languages, like C.
2019-02-21 04:18:33 +00:00
#if MICROPY_PERSISTENT_CODE_SAVE
emit->qstr_link_cur = 0;
#endif
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit->last_emit_was_return_value = false;
emit->scope = scope;
py, compiler: Improve passes; add an extra pass for native emitter.
2014-05-07 17:24:22 +01:00
// allocate memory for keeping track of the types of locals
if (emit->local_vtype_alloc < scope->num_locals) {
emit->local_vtype = m_renew(vtype_kind_t, emit->local_vtype, emit->local_vtype_alloc, scope->num_locals);
emit->local_vtype_alloc = scope->num_locals;
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
}
py, compiler: Improve passes; add an extra pass for native emitter.
2014-05-07 17:24:22 +01:00
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
// set default type for arguments
mp_uint_t num_args = emit->scope->num_pos_args + emit->scope->num_kwonly_args;
if (scope->scope_flags & MP_SCOPE_FLAG_VARARGS) {
num_args += 1;
}
if (scope->scope_flags & MP_SCOPE_FLAG_VARKEYWORDS) {
num_args += 1;
}
for (mp_uint_t i = 0; i < num_args; i++) {
py: Allow viper to have type annotations. Viper functions can now be annotated with the type of their arguments and return value. Eg: @micropython.viper def f(x:int) -> int: return x + 1
2014-08-15 16:45:41 +01:00
emit->local_vtype[i] = VTYPE_PYOBJ;
}
py: Fix typing of viper locals; allow default types in annotation.
2014-08-15 22:39:08 +01:00
py/emit: Completely remove set_native_type, arg type is set in compiler. In viper mode, the type of the argument is now stored in id_info->flags.
2018-09-15 04:00:11 +01:00
// Set viper type for arguments
if (emit->do_viper_types) {
for (int i = 0; i < emit->scope->id_info_len; ++i) {
id_info_t *id = &emit->scope->id_info[i];
if (id->flags & ID_FLAG_IS_PARAM) {
assert(id->local_num < emit->local_vtype_alloc);
emit->local_vtype[id->local_num] = id->flags >> ID_FLAG_VIPER_TYPE_POS;
}
}
}
py: Fix typing of viper locals; allow default types in annotation.
2014-08-15 22:39:08 +01:00
// local variables begin unbound, and have unknown type
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
for (mp_uint_t i = num_args; i < emit->local_vtype_alloc; i++) {
py: Fix typing of viper locals; allow default types in annotation.
2014-08-15 22:39:08 +01:00
emit->local_vtype[i] = VTYPE_UNBOUND;
}
// values on stack begin unbound
for (mp_uint_t i = 0; i < emit->stack_info_alloc; i++) {
py: Allow viper to have type annotations. Viper functions can now be annotated with the type of their arguments and return value. Eg: @micropython.viper def f(x:int) -> int: return x + 1
2014-08-15 16:45:41 +01:00
emit->stack_info[i].kind = STACK_VALUE;
py: Fix typing of viper locals; allow default types in annotation.
2014-08-15 22:39:08 +01:00
emit->stack_info[i].vtype = VTYPE_UNBOUND;
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Factor out common code from assemblers into asmbase.[ch]. All assemblers should "derive" from mp_asm_base_t.
2016-11-27 22:24:50 +00:00
mp_asm_base_start_pass(&emit->as->base, pass == MP_PASS_EMIT ? MP_ASM_PASS_EMIT : MP_ASM_PASS_COMPUTE);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
// generate code for entry to function
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
py/emitnative: Reorder native state on C stack so nlr_buf_t is first. The nlr_buf_t doesn't need to be part of the Python value stack (as it was before this commit), it's simpler to have it separated as auxiliary state that lives on the C stack. This will help adding yield support because in that case the nlr_buf_t and Python value stack live in separate memory areas (C stack and heap respectively).
2018-10-01 03:36:21 +01:00
// Work out start of code state (mp_code_state_t or reduced version for viper)
emit->code_state_start = 0;
if (NEED_GLOBAL_EXC_HANDLER(emit)) {
emit->code_state_start = sizeof(nlr_buf_t) / sizeof(uintptr_t);
}
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
if (emit->do_viper_types) {
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
// Work out size of state (locals plus stack)
// n_state counts all stack and locals, even those in registers
emit->n_state = scope->num_locals + scope->stack_size;
int num_locals_in_regs = 0;
if (CAN_USE_REGS_FOR_LOCALS(emit)) {
num_locals_in_regs = scope->num_locals;
if (num_locals_in_regs > REG_LOCAL_NUM) {
num_locals_in_regs = REG_LOCAL_NUM;
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
}
py: Make viper functions have the same entry signature as native. This commit makes viper functions have the same signature as native functions, at the level of the emitter/assembler. This means that viper functions can now be wrapped in the same uPy object as native functions. Viper functions are now responsible for parsing their arguments (before it was done by the runtime), and this makes calling them more efficient (in most cases) because the viper entry code can be custom generated to suit the signature of the function. This change also opens the way forward for viper functions to take arbitrary numbers of arguments, and for them to handle globals correctly, among other things.
2018-09-14 08:40:59 +01:00
// Need a spot for REG_LOCAL_3 if 4 or more args (see below)
if (scope->num_pos_args >= 4) {
--num_locals_in_regs;
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
py/emitnative: Make viper funcs run with their correct globals context. Viper functions will now capture the globals at the point they were defined and use these globals when executing.
2018-09-15 13:37:07 +01:00
// Work out where the locals and Python stack start within the C stack
if (NEED_GLOBAL_EXC_HANDLER(emit)) {
// Reserve 2 words for function object and old globals
py/emitnative: Reorder native state on C stack so nlr_buf_t is first. The nlr_buf_t doesn't need to be part of the Python value stack (as it was before this commit), it's simpler to have it separated as auxiliary state that lives on the C stack. This will help adding yield support because in that case the nlr_buf_t and Python value stack live in separate memory areas (C stack and heap respectively).
2018-10-01 03:36:21 +01:00
emit->stack_start = emit->code_state_start + 2;
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
} else if (scope->scope_flags & MP_SCOPE_FLAG_HASCONSTS) {
// Reserve 1 word for function object, to access const table
py/emitnative: Reorder native state on C stack so nlr_buf_t is first. The nlr_buf_t doesn't need to be part of the Python value stack (as it was before this commit), it's simpler to have it separated as auxiliary state that lives on the C stack. This will help adding yield support because in that case the nlr_buf_t and Python value stack live in separate memory areas (C stack and heap respectively).
2018-10-01 03:36:21 +01:00
emit->stack_start = emit->code_state_start + 1;
py/emitnative: Make viper funcs run with their correct globals context. Viper functions will now capture the globals at the point they were defined and use these globals when executing.
2018-09-15 13:37:07 +01:00
} else {
py/emitnative: Reorder native state on C stack so nlr_buf_t is first. The nlr_buf_t doesn't need to be part of the Python value stack (as it was before this commit), it's simpler to have it separated as auxiliary state that lives on the C stack. This will help adding yield support because in that case the nlr_buf_t and Python value stack live in separate memory areas (C stack and heap respectively).
2018-10-01 03:36:21 +01:00
emit->stack_start = emit->code_state_start + 0;
py/emitnative: Make viper funcs run with their correct globals context. Viper functions will now capture the globals at the point they were defined and use these globals when executing.
2018-09-15 13:37:07 +01:00
}
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
// Entry to function
ASM_ENTRY(emit->as, emit->stack_start + emit->n_state - num_locals_in_regs);
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
py/emitnative: Load native fun table ptr from const table for all archs. All architectures now have a dedicated register to hold the pointer to the native function table mp_fun_table, and so they all need to load this register at the start of the native function. This commit makes the loading of this register uniform across architectures by passing the pointer in the constant table for the native function, and then loading the register from the constant table. Doing it this way means that the pointer is not stored in the assembly code, helping to make the code more portable.
2018-10-13 04:53:35 +01:00
#if N_X86
asm_x86_mov_arg_to_r32(emit->as, 0, REG_ARG_1);
py: In native ARM emitter, load r7 with table earlier in func prelude. r7 may be needed to set up code state, so it must be loaded before the set-up function is called.
2015-08-12 23:31:19 +01:00
#endif
py/emitnative: Load native fun table ptr from const table for all archs. All architectures now have a dedicated register to hold the pointer to the native function table mp_fun_table, and so they all need to load this register at the start of the native function. This commit makes the loading of this register uniform across architectures by passing the pointer in the constant table for the native function, and then loading the register from the constant table. Doing it this way means that the pointer is not stored in the assembly code, helping to make the code more portable.
2018-10-13 04:53:35 +01:00
// Load REG_FUN_TABLE with a pointer to mp_fun_table, found in the const_table
ASM_LOAD_REG_REG_OFFSET(emit->as, REG_LOCAL_3, REG_ARG_1, offsetof(mp_obj_fun_bc_t, const_table) / sizeof(uintptr_t));
ASM_LOAD_REG_REG_OFFSET(emit->as, REG_FUN_TABLE, REG_LOCAL_3, 0);
py/emitnative: Make viper funcs run with their correct globals context. Viper functions will now capture the globals at the point they were defined and use these globals when executing.
2018-09-15 13:37:07 +01:00
// Store function object (passed as first arg) to stack if needed
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
if (NEED_FUN_OBJ(emit)) {
py/emitnative: Make viper funcs run with their correct globals context. Viper functions will now capture the globals at the point they were defined and use these globals when executing.
2018-09-15 13:37:07 +01:00
ASM_MOV_LOCAL_REG(emit->as, LOCAL_IDX_FUN_OBJ(emit), REG_ARG_1);
}
py: Make viper functions have the same entry signature as native. This commit makes viper functions have the same signature as native functions, at the level of the emitter/assembler. This means that viper functions can now be wrapped in the same uPy object as native functions. Viper functions are now responsible for parsing their arguments (before it was done by the runtime), and this makes calling them more efficient (in most cases) because the viper entry code can be custom generated to suit the signature of the function. This change also opens the way forward for viper functions to take arbitrary numbers of arguments, and for them to handle globals correctly, among other things.
2018-09-14 08:40:59 +01:00
// Put n_args in REG_ARG_1, n_kw in REG_ARG_2, args array in REG_LOCAL_3
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
#if N_X86
py: Make viper functions have the same entry signature as native. This commit makes viper functions have the same signature as native functions, at the level of the emitter/assembler. This means that viper functions can now be wrapped in the same uPy object as native functions. Viper functions are now responsible for parsing their arguments (before it was done by the runtime), and this makes calling them more efficient (in most cases) because the viper entry code can be custom generated to suit the signature of the function. This change also opens the way forward for viper functions to take arbitrary numbers of arguments, and for them to handle globals correctly, among other things.
2018-09-14 08:40:59 +01:00
asm_x86_mov_arg_to_r32(emit->as, 1, REG_ARG_1);
asm_x86_mov_arg_to_r32(emit->as, 2, REG_ARG_2);
asm_x86_mov_arg_to_r32(emit->as, 3, REG_LOCAL_3);
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
#else
py: Make viper functions have the same entry signature as native. This commit makes viper functions have the same signature as native functions, at the level of the emitter/assembler. This means that viper functions can now be wrapped in the same uPy object as native functions. Viper functions are now responsible for parsing their arguments (before it was done by the runtime), and this makes calling them more efficient (in most cases) because the viper entry code can be custom generated to suit the signature of the function. This change also opens the way forward for viper functions to take arbitrary numbers of arguments, and for them to handle globals correctly, among other things.
2018-09-14 08:40:59 +01:00
ASM_MOV_REG_REG(emit->as, REG_ARG_1, REG_ARG_2);
ASM_MOV_REG_REG(emit->as, REG_ARG_2, REG_ARG_3);
ASM_MOV_REG_REG(emit->as, REG_LOCAL_3, REG_ARG_4);
#endif
// Check number of args matches this function, and call mp_arg_check_num_sig if not
ASM_JUMP_IF_REG_NONZERO(emit->as, REG_ARG_2, *emit->label_slot + 4, true);
ASM_MOV_REG_IMM(emit->as, REG_ARG_3, scope->num_pos_args);
ASM_JUMP_IF_REG_EQ(emit->as, REG_ARG_1, REG_ARG_3, *emit->label_slot + 5);
mp_asm_base_label_assign(&emit->as->base, *emit->label_slot + 4);
ASM_MOV_REG_IMM(emit->as, REG_ARG_3, MP_OBJ_FUN_MAKE_SIG(scope->num_pos_args, scope->num_pos_args, false));
py/emitnative: Remove unused ptr argument from ASM_CALL_IND macro.
2018-10-13 05:13:55 +01:00
ASM_CALL_IND(emit->as, MP_F_ARG_CHECK_NUM_SIG);
py: Make viper functions have the same entry signature as native. This commit makes viper functions have the same signature as native functions, at the level of the emitter/assembler. This means that viper functions can now be wrapped in the same uPy object as native functions. Viper functions are now responsible for parsing their arguments (before it was done by the runtime), and this makes calling them more efficient (in most cases) because the viper entry code can be custom generated to suit the signature of the function. This change also opens the way forward for viper functions to take arbitrary numbers of arguments, and for them to handle globals correctly, among other things.
2018-09-14 08:40:59 +01:00
mp_asm_base_label_assign(&emit->as->base, *emit->label_slot + 5);
// Store arguments into locals (reg or stack), converting to native if needed
for (int i = 0; i < emit->scope->num_pos_args; i++) {
int r = REG_ARG_1;
ASM_LOAD_REG_REG_OFFSET(emit->as, REG_ARG_1, REG_LOCAL_3, i);
if (emit->local_vtype[i] != VTYPE_PYOBJ) {
emit_call_with_imm_arg(emit, MP_F_CONVERT_OBJ_TO_NATIVE, emit->local_vtype[i], REG_ARG_2);
r = REG_RET;
}
// REG_LOCAL_3 points to the args array so be sure not to overwrite it if it's still needed
if (i < REG_LOCAL_NUM && CAN_USE_REGS_FOR_LOCALS(emit) && (i != 2 || emit->scope->num_pos_args == 3)) {
ASM_MOV_REG_REG(emit->as, reg_local_table[i], r);
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
} else {
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
emit_native_mov_state_reg(emit, LOCAL_IDX_LOCAL_VAR(emit, i), r);
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
}
}
py: Make viper functions have the same entry signature as native. This commit makes viper functions have the same signature as native functions, at the level of the emitter/assembler. This means that viper functions can now be wrapped in the same uPy object as native functions. Viper functions are now responsible for parsing their arguments (before it was done by the runtime), and this makes calling them more efficient (in most cases) because the viper entry code can be custom generated to suit the signature of the function. This change also opens the way forward for viper functions to take arbitrary numbers of arguments, and for them to handle globals correctly, among other things.
2018-09-14 08:40:59 +01:00
// Get 3rd local from the stack back into REG_LOCAL_3 if this reg couldn't be written to above
if (emit->scope->num_pos_args >= 4 && CAN_USE_REGS_FOR_LOCALS(emit)) {
ASM_MOV_REG_LOCAL(emit->as, REG_LOCAL_3, LOCAL_IDX_LOCAL_VAR(emit, 2));
}
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
emit_native_global_exc_entry(emit);
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
} else {
// work out size of state (locals plus stack)
emit->n_state = scope->num_locals + scope->stack_size;
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
if (emit->scope->scope_flags & MP_SCOPE_FLAG_GENERATOR) {
emit->code_state_start = 0;
emit->stack_start = sizeof(mp_code_state_t) / sizeof(mp_uint_t);
mp_asm_base_data(&emit->as->base, ASM_WORD_SIZE, (uintptr_t)emit->prelude_offset);
mp_asm_base_data(&emit->as->base, ASM_WORD_SIZE, (uintptr_t)emit->start_offset);
ASM_ENTRY(emit->as, sizeof(nlr_buf_t) / sizeof(uintptr_t));
// Put address of code_state into REG_GENERATOR_STATE
#if N_X86
asm_x86_mov_arg_to_r32(emit->as, 0, REG_GENERATOR_STATE);
#else
ASM_MOV_REG_REG(emit->as, REG_GENERATOR_STATE, REG_ARG_1);
#endif
// Put throw value into LOCAL_IDX_EXC_VAL slot, for yield/yield-from
#if N_X86
asm_x86_mov_arg_to_r32(emit->as, 1, REG_ARG_2);
#endif
ASM_MOV_LOCAL_REG(emit->as, LOCAL_IDX_EXC_VAL(emit), REG_ARG_2);
py/emitnative: Load native fun table ptr from const table for all archs. All architectures now have a dedicated register to hold the pointer to the native function table mp_fun_table, and so they all need to load this register at the start of the native function. This commit makes the loading of this register uniform across architectures by passing the pointer in the constant table for the native function, and then loading the register from the constant table. Doing it this way means that the pointer is not stored in the assembly code, helping to make the code more portable.
2018-10-13 04:53:35 +01:00
// Load REG_FUN_TABLE with a pointer to mp_fun_table, found in the const_table
ASM_LOAD_REG_REG_OFFSET(emit->as, REG_TEMP0, REG_GENERATOR_STATE, LOCAL_IDX_FUN_OBJ(emit));
ASM_LOAD_REG_REG_OFFSET(emit->as, REG_TEMP0, REG_TEMP0, offsetof(mp_obj_fun_bc_t, const_table) / sizeof(uintptr_t));
ASM_LOAD_REG_REG_OFFSET(emit->as, REG_FUN_TABLE, REG_TEMP0, emit->scope->num_pos_args + emit->scope->num_kwonly_args);
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
} else {
// The locals and stack start after the code_state structure
emit->stack_start = emit->code_state_start + sizeof(mp_code_state_t) / sizeof(mp_uint_t);
py/emitnative: Fix native locals stack to start at correct location. A native function allocates space on its C stack for mp_code_state_t, followed by its Python stack, then its locals. This patch makes sure that the native function actually starts at the start of its Python stack, rather than at the start of mp_code_state_t (which didn't lead to any issues so far because the mp_code_state_t is unused after the native function sets itself up).
2018-08-04 13:41:35 +01:00
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
// Allocate space on C-stack for code_state structure, which includes state
ASM_ENTRY(emit->as, emit->stack_start + emit->n_state);
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
// Prepare incoming arguments for call to mp_setup_code_state
py: Reduce size of mp_code_state_t structure. Instead of caching data that is constant (code_info, const_table and n_state), store just a pointer to the underlying function object from which this data can be derived. This helps reduce stack usage for the case when the mp_code_state_t structure is stored on the stack, as well as heap usage when it's stored on the heap. The downside is that the VM becomes a little more complex because it now needs to derive the data from the underlying function object. But this doesn't impact the performance by much (if at all) because most of the decoding of data is done outside the main opcode loop. Measurements using pystone show that little to no performance is lost. This patch also fixes a nasty bug whereby the bytecode can be reclaimed by the GC during execution. With this patch there is always a pointer to the function object held by the VM during execution, since it's stored in the mp_code_state_t structure.
2017-03-17 03:54:53 +00:00
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
#if N_X86
asm_x86_mov_arg_to_r32(emit->as, 0, REG_ARG_1);
asm_x86_mov_arg_to_r32(emit->as, 1, REG_ARG_2);
asm_x86_mov_arg_to_r32(emit->as, 2, REG_ARG_3);
asm_x86_mov_arg_to_r32(emit->as, 3, REG_ARG_4);
#endif
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
py/emitnative: Load native fun table ptr from const table for all archs. All architectures now have a dedicated register to hold the pointer to the native function table mp_fun_table, and so they all need to load this register at the start of the native function. This commit makes the loading of this register uniform across architectures by passing the pointer in the constant table for the native function, and then loading the register from the constant table. Doing it this way means that the pointer is not stored in the assembly code, helping to make the code more portable.
2018-10-13 04:53:35 +01:00
// Load REG_FUN_TABLE with a pointer to mp_fun_table, found in the const_table
ASM_LOAD_REG_REG_OFFSET(emit->as, REG_LOCAL_3, REG_ARG_1, offsetof(mp_obj_fun_bc_t, const_table) / sizeof(uintptr_t));
ASM_LOAD_REG_REG_OFFSET(emit->as, REG_FUN_TABLE, REG_LOCAL_3, emit->scope->num_pos_args + emit->scope->num_kwonly_args);
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
// Set code_state.fun_bc
ASM_MOV_LOCAL_REG(emit->as, LOCAL_IDX_FUN_OBJ(emit), REG_ARG_1);
py: Reduce size of mp_code_state_t structure. Instead of caching data that is constant (code_info, const_table and n_state), store just a pointer to the underlying function object from which this data can be derived. This helps reduce stack usage for the case when the mp_code_state_t structure is stored on the stack, as well as heap usage when it's stored on the heap. The downside is that the VM becomes a little more complex because it now needs to derive the data from the underlying function object. But this doesn't impact the performance by much (if at all) because most of the decoding of data is done outside the main opcode loop. Measurements using pystone show that little to no performance is lost. This patch also fixes a nasty bug whereby the bytecode can be reclaimed by the GC during execution. With this patch there is always a pointer to the function object held by the VM during execution, since it's stored in the mp_code_state_t structure.
2017-03-17 03:54:53 +00:00
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
// Set code_state.ip (offset from start of this function to prelude info)
// TODO this encoding may change size in the final pass, need to make it fixed
emit_native_mov_state_imm_via(emit, emit->code_state_start + offsetof(mp_code_state_t, ip) / sizeof(uintptr_t), emit->prelude_offset, REG_ARG_1);
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
// Put address of code_state into first arg
ASM_MOV_REG_LOCAL_ADDR(emit->as, REG_ARG_1, emit->code_state_start);
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
// Call mp_setup_code_state to prepare code_state structure
#if N_THUMB
py/asmthumb: Remove unused fun_ptr arg from asm_thumb_bl_ind function.
2018-10-13 05:08:31 +01:00
asm_thumb_bl_ind(emit->as, MP_F_SETUP_CODE_STATE, ASM_THUMB_REG_R4);
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
#elif N_ARM
py/asmarm: Simplify asm_arm_bl_ind to only load via index, not literal. The maximum index into mp_fun_table is currently less than 1024 and should stay that way to keep things efficient for all architectures, so there is no need to handle loading the pointer directly via a literal in this function.
2018-10-13 05:03:29 +01:00
asm_arm_bl_ind(emit->as, MP_F_SETUP_CODE_STATE, ASM_ARM_REG_R4);
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
#else
py/emitnative: Remove unused ptr argument from ASM_CALL_IND macro.
2018-10-13 05:13:55 +01:00
ASM_CALL_IND(emit->as, MP_F_SETUP_CODE_STATE);
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
#endif
}
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
emit_native_global_exc_entry(emit);
// cache some locals in registers, but only if no exception handlers
if (CAN_USE_REGS_FOR_LOCALS(emit)) {
for (int i = 0; i < REG_LOCAL_NUM && i < scope->num_locals; ++i) {
ASM_MOV_REG_LOCAL(emit->as, reg_local_table[i], LOCAL_IDX_LOCAL_VAR(emit, i));
}
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
}
// set the type of closed over variables
for (mp_uint_t i = 0; i < scope->id_info_len; i++) {
id_info_t *id = &scope->id_info[i];
if (id->kind == ID_INFO_KIND_CELL) {
emit->local_vtype[id->local_num] = VTYPE_PYOBJ;
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
if (pass == MP_PASS_EMIT) {
// write argument names as qstr objects
// see comment in corresponding part of emitbc.c about the logic here
for (int i = 0; i < scope->num_pos_args + scope->num_kwonly_args; i++) {
qstr qst = MP_QSTR__star_;
for (int j = 0; j < scope->id_info_len; ++j) {
id_info_t *id = &scope->id_info[j];
if ((id->flags & ID_FLAG_IS_PARAM) && id->local_num == i) {
qst = id->qst;
break;
}
}
emit->const_table[i] = (mp_uint_t)MP_OBJ_NEW_QSTR(qst);
}
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_native_end_pass(emit_t *emit) {
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
emit_native_global_exc_exit(emit);
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
if (!emit->do_viper_types) {
py: Factor out common code from assemblers into asmbase.[ch]. All assemblers should "derive" from mp_asm_base_t.
2016-11-27 22:24:50 +00:00
emit->prelude_offset = mp_asm_base_get_code_pos(&emit->as->base);
py: Reduce size of mp_code_state_t structure. Instead of caching data that is constant (code_info, const_table and n_state), store just a pointer to the underlying function object from which this data can be derived. This helps reduce stack usage for the case when the mp_code_state_t structure is stored on the stack, as well as heap usage when it's stored on the heap. The downside is that the VM becomes a little more complex because it now needs to derive the data from the underlying function object. But this doesn't impact the performance by much (if at all) because most of the decoding of data is done outside the main opcode loop. Measurements using pystone show that little to no performance is lost. This patch also fixes a nasty bug whereby the bytecode can be reclaimed by the GC during execution. With this patch there is always a pointer to the function object held by the VM during execution, since it's stored in the mp_code_state_t structure.
2017-03-17 03:54:53 +00:00
mp_asm_base_data(&emit->as->base, 1, 0x80 | ((emit->n_state >> 7) & 0x7f));
mp_asm_base_data(&emit->as->base, 1, emit->n_state & 0x7f);
mp_asm_base_data(&emit->as->base, 1, 0); // n_exc_stack
py: Factor out common code from assemblers into asmbase.[ch]. All assemblers should "derive" from mp_asm_base_t.
2016-11-27 22:24:50 +00:00
mp_asm_base_data(&emit->as->base, 1, emit->scope->scope_flags);
mp_asm_base_data(&emit->as->base, 1, emit->scope->num_pos_args);
mp_asm_base_data(&emit->as->base, 1, emit->scope->num_kwonly_args);
mp_asm_base_data(&emit->as->base, 1, emit->scope->num_def_pos_args);
py: Add constant table to bytecode. Contains just argument names at the moment but makes it easy to add arbitrary constants.
2015-10-23 01:23:11 +01:00
py: Add MICROPY_PERSISTENT_CODE so code can persist beyond the runtime. Main changes when MICROPY_PERSISTENT_CODE is enabled are: - qstrs are encoded as 2-byte fixed width in the bytecode - all pointers are removed from bytecode and put in const_table (this includes const objects and raw code pointers) Ultimately this option will enable persistence for not just bytecode but also native code.
2015-11-02 17:27:18 +00:00
// write code info
#if MICROPY_PERSISTENT_CODE
py: Factor out common code from assemblers into asmbase.[ch]. All assemblers should "derive" from mp_asm_base_t.
2016-11-27 22:24:50 +00:00
mp_asm_base_data(&emit->as->base, 1, 5);
mp_asm_base_data(&emit->as->base, 1, emit->scope->simple_name);
mp_asm_base_data(&emit->as->base, 1, emit->scope->simple_name >> 8);
mp_asm_base_data(&emit->as->base, 1, emit->scope->source_file);
mp_asm_base_data(&emit->as->base, 1, emit->scope->source_file >> 8);
py: Add MICROPY_PERSISTENT_CODE so code can persist beyond the runtime. Main changes when MICROPY_PERSISTENT_CODE is enabled are: - qstrs are encoded as 2-byte fixed width in the bytecode - all pointers are removed from bytecode and put in const_table (this includes const objects and raw code pointers) Ultimately this option will enable persistence for not just bytecode but also native code.
2015-11-02 17:27:18 +00:00
#else
py: Factor out common code from assemblers into asmbase.[ch]. All assemblers should "derive" from mp_asm_base_t.
2016-11-27 22:24:50 +00:00
mp_asm_base_data(&emit->as->base, 1, 1);
py: Add MICROPY_PERSISTENT_CODE so code can persist beyond the runtime. Main changes when MICROPY_PERSISTENT_CODE is enabled are: - qstrs are encoded as 2-byte fixed width in the bytecode - all pointers are removed from bytecode and put in const_table (this includes const objects and raw code pointers) Ultimately this option will enable persistence for not just bytecode but also native code.
2015-11-02 17:27:18 +00:00
#endif
py: Add constant table to bytecode. Contains just argument names at the moment but makes it easy to add arbitrary constants.
2015-10-23 01:23:11 +01:00
// bytecode prelude: initialise closed over variables
for (int i = 0; i < emit->scope->id_info_len; i++) {
id_info_t *id = &emit->scope->id_info[i];
if (id->kind == ID_INFO_KIND_CELL) {
assert(id->local_num < 255);
py: Factor out common code from assemblers into asmbase.[ch]. All assemblers should "derive" from mp_asm_base_t.
2016-11-27 22:24:50 +00:00
mp_asm_base_data(&emit->as->base, 1, id->local_num); // write the local which should be converted to a cell
py: Add constant table to bytecode. Contains just argument names at the moment but makes it easy to add arbitrary constants.
2015-10-23 01:23:11 +01:00
}
}
py: Factor out common code from assemblers into asmbase.[ch]. All assemblers should "derive" from mp_asm_base_t.
2016-11-27 22:24:50 +00:00
mp_asm_base_data(&emit->as->base, 1, 255); // end of list sentinel
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
}
py: Add support for emitting native x86 machine code.
2014-09-06 23:06:36 +01:00
ASM_END_PASS(emit->as);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
// check stack is back to zero size
py/emitnative: Use assertions and mp_not_implemented correctly. Assertions are used to check expressions that should always be true, and mp_not_implemented is used for code that can be reached.
2017-03-14 12:00:53 +00:00
assert(emit->stack_size == 0);
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
assert(emit->exc_stack_size == 0);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
// Deal with const table accounting
assert(emit->pass <= MP_PASS_STACK_SIZE || (emit->const_table_num_obj == emit->const_table_cur_obj));
emit->const_table_num_obj = emit->const_table_cur_obj;
if (emit->pass == MP_PASS_CODE_SIZE) {
py/emitnative: Adjust accounting of size of const_table. n_obj no longer includes a count for mp_fun_table to make it a bit simpler.
2019-02-25 04:57:35 +00:00
size_t const_table_alloc = 1 + emit->const_table_num_obj + emit->const_table_cur_raw_code;
py/emitnative: Load native fun table ptr from const table for all archs. All architectures now have a dedicated register to hold the pointer to the native function table mp_fun_table, and so they all need to load this register at the start of the native function. This commit makes the loading of this register uniform across architectures by passing the pointer in the constant table for the native function, and then loading the register from the constant table. Doing it this way means that the pointer is not stored in the assembly code, helping to make the code more portable.
2018-10-13 04:53:35 +01:00
size_t nqstr = 0;
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
if (!emit->do_viper_types) {
// Add room for qstr names of arguments
py/emitnative: Load native fun table ptr from const table for all archs. All architectures now have a dedicated register to hold the pointer to the native function table mp_fun_table, and so they all need to load this register at the start of the native function. This commit makes the loading of this register uniform across architectures by passing the pointer in the constant table for the native function, and then loading the register from the constant table. Doing it this way means that the pointer is not stored in the assembly code, helping to make the code more portable.
2018-10-13 04:53:35 +01:00
nqstr = emit->scope->num_pos_args + emit->scope->num_kwonly_args;
const_table_alloc += nqstr;
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
}
py/emitnative: Change type of const_table from uintptr_t to mp_uint_t. This matches how bytecode does it, and matches the signature of mp_emit_glue_assign_native. Since the native emitter doesn't support nan-boxing uintptr_t and mp_uint_t are anyway the same bit-width.
2018-09-27 15:04:10 +01:00
emit->const_table = m_new(mp_uint_t, const_table_alloc);
py/nativeglue: Remove dependency on mp_fun_table in dyn-compiler mode. mpy-cross uses MICROPY_DYNAMIC_COMPILER and MICROPY_EMIT_NATIVE but does not actually need to execute native functions, and does not need mp_fun_table. This commit makes it so mp_fun_table and all its entries are not built when MICROPY_DYNAMIC_COMPILER is enabled, significantly reducing the size of the mpy-cross executable and allowing it to be built on more machines/OS's.
2019-05-29 12:17:29 +01:00
#if !MICROPY_DYNAMIC_COMPILER
py/emitnative: Load native fun table ptr from const table for all archs. All architectures now have a dedicated register to hold the pointer to the native function table mp_fun_table, and so they all need to load this register at the start of the native function. This commit makes the loading of this register uniform across architectures by passing the pointer in the constant table for the native function, and then loading the register from the constant table. Doing it this way means that the pointer is not stored in the assembly code, helping to make the code more portable.
2018-10-13 04:53:35 +01:00
// Store mp_fun_table pointer just after qstrs
py/nativeglue: Remove dependency on mp_fun_table in dyn-compiler mode. mpy-cross uses MICROPY_DYNAMIC_COMPILER and MICROPY_EMIT_NATIVE but does not actually need to execute native functions, and does not need mp_fun_table. This commit makes it so mp_fun_table and all its entries are not built when MICROPY_DYNAMIC_COMPILER is enabled, significantly reducing the size of the mpy-cross executable and allowing it to be built on more machines/OS's.
2019-05-29 12:17:29 +01:00
// (but in dynamic-compiler mode eliminate dependency on mp_fun_table)
py/emitnative: Load native fun table ptr from const table for all archs. All architectures now have a dedicated register to hold the pointer to the native function table mp_fun_table, and so they all need to load this register at the start of the native function. This commit makes the loading of this register uniform across architectures by passing the pointer in the constant table for the native function, and then loading the register from the constant table. Doing it this way means that the pointer is not stored in the assembly code, helping to make the code more portable.
2018-10-13 04:53:35 +01:00
emit->const_table[nqstr] = (mp_uint_t)(uintptr_t)mp_fun_table;
py/nativeglue: Remove dependency on mp_fun_table in dyn-compiler mode. mpy-cross uses MICROPY_DYNAMIC_COMPILER and MICROPY_EMIT_NATIVE but does not actually need to execute native functions, and does not need mp_fun_table. This commit makes it so mp_fun_table and all its entries are not built when MICROPY_DYNAMIC_COMPILER is enabled, significantly reducing the size of the mpy-cross executable and allowing it to be built on more machines/OS's.
2019-05-29 12:17:29 +01:00
#endif
py: Add support to save native, viper and asm code to .mpy files. This commit adds support for saving and loading .mpy files that contain native code (native, viper and inline-asm). A lot of the ground work was already done for this in the form of removing pointers from generated native code. The changes here are mainly to link in qstr values to the native code, and change the format of .mpy files to contain native code blocks (possibly mixed with bytecode). A top-level summary: - @micropython.native, @micropython.viper and @micropython.asm_thumb/ asm_xtensa are now allowed in .py files when compiling to .mpy, and they work transparently to the user. - Entire .py files can be compiled to native via mpy-cross -X emit=native and for the most part the generated .mpy files should work the same as their bytecode version. - The .mpy file format is changed to 1) specify in the header if the file contains native code and if so the architecture (eg x86, ARMV7M, Xtensa); 2) for each function block the kind of code is specified (bytecode, native, viper, asm). - When native code is loaded from a .mpy file the native code must be modified (in place) to link qstr values in, just like bytecode (see py/persistentcode.c:arch_link_qstr() function). In addition, this now defines a public, native ABI for dynamically loadable native code generated by other languages, like C.
2019-02-21 04:18:33 +00:00
#if MICROPY_PERSISTENT_CODE_SAVE
size_t qstr_link_alloc = emit->qstr_link_cur;
if (qstr_link_alloc > 0) {
emit->qstr_link = m_new(mp_qstr_link_entry_t, qstr_link_alloc);
}
#endif
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
}
py, compiler: Improve passes; add an extra pass for native emitter.
2014-05-07 17:24:22 +01:00
if (emit->pass == MP_PASS_EMIT) {
py: Factor out common code from assemblers into asmbase.[ch]. All assemblers should "derive" from mp_asm_base_t.
2016-11-27 22:24:50 +00:00
void *f = mp_asm_base_get_code(&emit->as->base);
mp_uint_t f_len = mp_asm_base_get_code_size(&emit->as->base);
py: Allow viper to have type annotations. Viper functions can now be annotated with the type of their arguments and return value. Eg: @micropython.viper def f(x:int) -> int: return x + 1
2014-08-15 16:45:41 +01:00
py: Implement full func arg passing for native emitter. This patch gets full function argument passing working with native emitter. Includes named args, keyword args, default args, var args and var keyword args. Fully Python compliant. It reuses the bytecode mp_setup_code_state function to do all the hard work. This function is slightly adjusted to accommodate native calls, and the native emitter is forced a bit to emit similar prelude and code-info as bytecode.
2015-04-06 22:38:53 +01:00
mp_emit_glue_assign_native(emit->scope->raw_code,
emit->do_viper_types ? MP_CODE_NATIVE_VIPER : MP_CODE_NATIVE_PY,
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
f, f_len, emit->const_table,
py: Add support to save native, viper and asm code to .mpy files. This commit adds support for saving and loading .mpy files that contain native code (native, viper and inline-asm). A lot of the ground work was already done for this in the form of removing pointers from generated native code. The changes here are mainly to link in qstr values to the native code, and change the format of .mpy files to contain native code blocks (possibly mixed with bytecode). A top-level summary: - @micropython.native, @micropython.viper and @micropython.asm_thumb/ asm_xtensa are now allowed in .py files when compiling to .mpy, and they work transparently to the user. - Entire .py files can be compiled to native via mpy-cross -X emit=native and for the most part the generated .mpy files should work the same as their bytecode version. - The .mpy file format is changed to 1) specify in the header if the file contains native code and if so the architecture (eg x86, ARMV7M, Xtensa); 2) for each function block the kind of code is specified (bytecode, native, viper, asm). - When native code is loaded from a .mpy file the native code must be modified (in place) to link qstr values in, just like bytecode (see py/persistentcode.c:arch_link_qstr() function). In addition, this now defines a public, native ABI for dynamically loadable native code generated by other languages, like C.
2019-02-21 04:18:33 +00:00
#if MICROPY_PERSISTENT_CODE_SAVE
emit->prelude_offset,
emit->const_table_cur_obj, emit->const_table_cur_raw_code,
emit->qstr_link_cur, emit->qstr_link,
#endif
py: Make viper functions have the same entry signature as native. This commit makes viper functions have the same signature as native functions, at the level of the emitter/assembler. This means that viper functions can now be wrapped in the same uPy object as native functions. Viper functions are now responsible for parsing their arguments (before it was done by the runtime), and this makes calling them more efficient (in most cases) because the viper entry code can be custom generated to suit the signature of the function. This change also opens the way forward for viper functions to take arbitrary numbers of arguments, and for them to handle globals correctly, among other things.
2018-09-14 08:40:59 +01:00
emit->scope->num_pos_args, emit->scope->scope_flags, 0);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC bool emit_native_last_emit_was_return_value(emit_t *emit) {
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
return emit->last_emit_was_return_value;
}
py/emitnative: Allocate space for local stack info as it's needed.
2018-08-07 07:05:44 +01:00
STATIC void ensure_extra_stack(emit_t *emit, size_t delta) {
if (emit->stack_size + delta > emit->stack_info_alloc) {
size_t new_alloc = (emit->stack_size + delta + 8) & ~3;
emit->stack_info = m_renew(stack_info_t, emit->stack_info, emit->stack_info_alloc, new_alloc);
emit->stack_info_alloc = new_alloc;
}
}
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
STATIC void adjust_stack(emit_t *emit, mp_int_t stack_size_delta) {
assert((mp_int_t)emit->stack_size + stack_size_delta >= 0);
py/emitnative: Allocate space for local stack info as it's needed.
2018-08-07 07:05:44 +01:00
assert((mp_int_t)emit->stack_size + stack_size_delta <= (mp_int_t)emit->stack_info_alloc);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit->stack_size += stack_size_delta;
py, compiler: Improve passes; add an extra pass for native emitter.
2014-05-07 17:24:22 +01:00
if (emit->pass > MP_PASS_SCOPE && emit->stack_size > emit->scope->stack_size) {
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit->scope->stack_size = emit->stack_size;
}
py: Partially fix viper multi-comparison; add test for it.
2014-10-19 19:00:51 +01:00
#ifdef DEBUG_PRINT
DEBUG_printf(" adjust_stack; stack_size=%d+%d; stack now:", emit->stack_size - stack_size_delta, stack_size_delta);
for (int i = 0; i < emit->stack_size; i++) {
stack_info_t *si = &emit->stack_info[i];
py: Don't use anonymous unions, name them instead. This makes the code (more) compatible with the C99 standard.
2015-01-24 23:14:12 +00:00
DEBUG_printf(" (v=%d k=%d %d)", si->vtype, si->kind, si->data.u_reg);
py: Partially fix viper multi-comparison; add test for it.
2014-10-19 19:00:51 +01:00
}
DEBUG_printf("\n");
#endif
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
STATIC void emit_native_adjust_stack_size(emit_t *emit, mp_int_t delta) {
py: Partially fix viper multi-comparison; add test for it.
2014-10-19 19:00:51 +01:00
DEBUG_printf("adjust_stack_size(" INT_FMT ")\n", delta);
py/emitnative: Allocate space for local stack info as it's needed.
2018-08-07 07:05:44 +01:00
if (delta > 0) {
ensure_extra_stack(emit, delta);
}
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
// If we are adjusting the stack in a positive direction (pushing) then we
// need to fill in values for the stack kind and vtype of the newly-pushed
// entries. These should be set to "value" (ie not reg or imm) because we
// should only need to adjust the stack due to a jump to this part in the
// code (and hence we have settled the stack before the jump).
for (mp_int_t i = 0; i < delta; i++) {
stack_info_t *si = &emit->stack_info[emit->stack_size + i];
si->kind = STACK_VALUE;
py: Partially fix viper multi-comparison; add test for it.
2014-10-19 19:00:51 +01:00
// TODO we don't know the vtype to use here. At the moment this is a
// hack to get the case of multi comparison working.
if (delta == 1) {
si->vtype = emit->saved_stack_vtype;
} else {
si->vtype = VTYPE_PYOBJ;
}
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
}
adjust_stack(emit, delta);
}
STATIC void emit_native_set_source_line(emit_t *emit, mp_uint_t source_line) {
py, unix: Allow to compile with -Wunused-parameter. See issue #699.
2015-01-20 12:47:20 +00:00
(void)emit;
(void)source_line;
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
}
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
// this must be called at start of emit functions
py: Rename emit_pre so they have globally unique names.
2014-03-27 23:30:26 +00:00
STATIC void emit_native_pre(emit_t *emit) {
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
emit->last_emit_was_return_value = false;
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
// depth==0 is top, depth==1 is before top, etc
STATIC stack_info_t *peek_stack(emit_t *emit, mp_uint_t depth) {
return &emit->stack_info[emit->stack_size - 1 - depth];
}
// depth==0 is top, depth==1 is before top, etc
STATIC vtype_kind_t peek_vtype(emit_t *emit, mp_uint_t depth) {
py/emitnative: Consolidate use of stacked immediate values to one func. This commit adds the helper function load_reg_stack_imm() which deals with constant immediate values and converting them to Python objects if needed.
2018-10-14 12:58:27 +01:00
if (emit->do_viper_types) {
return peek_stack(emit, depth)->vtype;
} else {
// Type is always PYOBJ even if the intermediate stored value is not
return VTYPE_PYOBJ;
}
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
// pos=1 is TOS, pos=2 is next, etc
// use pos=0 for no skipping
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void need_reg_single(emit_t *emit, int reg_needed, int skip_stack_pos) {
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
skip_stack_pos = emit->stack_size - skip_stack_pos;
for (int i = 0; i < emit->stack_size; i++) {
if (i != skip_stack_pos) {
stack_info_t *si = &emit->stack_info[i];
py: Don't use anonymous unions, name them instead. This makes the code (more) compatible with the C99 standard.
2015-01-24 23:14:12 +00:00
if (si->kind == STACK_REG && si->data.u_reg == reg_needed) {
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
si->kind = STACK_VALUE;
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
emit_native_mov_state_reg(emit, emit->stack_start + i, si->data.u_reg);
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
}
}
}
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void need_reg_all(emit_t *emit) {
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
for (int i = 0; i < emit->stack_size; i++) {
stack_info_t *si = &emit->stack_info[i];
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
if (si->kind == STACK_REG) {
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
si->kind = STACK_VALUE;
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
emit_native_mov_state_reg(emit, emit->stack_start + i, si->data.u_reg);
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
}
}
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
py/emitnative: Consolidate use of stacked immediate values to one func. This commit adds the helper function load_reg_stack_imm() which deals with constant immediate values and converting them to Python objects if needed.
2018-10-14 12:58:27 +01:00
STATIC vtype_kind_t load_reg_stack_imm(emit_t *emit, int reg_dest, const stack_info_t *si, bool convert_to_pyobj) {
if (!convert_to_pyobj && emit->do_viper_types) {
ASM_MOV_REG_IMM(emit->as, reg_dest, si->data.u_imm);
return si->vtype;
} else {
if (si->vtype == VTYPE_PYOBJ) {
ASM_MOV_REG_IMM(emit->as, reg_dest, si->data.u_imm);
} else if (si->vtype == VTYPE_BOOL) {
py/emitnative: Put None/False/True in global native const table. So these constant objects can be loaded by dereferencing the REG_FUN_TABLE pointer instead of loading immediate values. This reduces the size of generated native code (when such constants are used), and means that pointers to these constants are no longer stored in the assembly code.
2018-10-14 13:56:46 +01:00
emit_native_mov_reg_const(emit, reg_dest, MP_F_CONST_FALSE_OBJ + si->data.u_imm);
py/emitnative: Consolidate use of stacked immediate values to one func. This commit adds the helper function load_reg_stack_imm() which deals with constant immediate values and converting them to Python objects if needed.
2018-10-14 12:58:27 +01:00
} else if (si->vtype == VTYPE_INT || si->vtype == VTYPE_UINT) {
ASM_MOV_REG_IMM(emit->as, reg_dest, (uintptr_t)MP_OBJ_NEW_SMALL_INT(si->data.u_imm));
} else if (si->vtype == VTYPE_PTR_NONE) {
py/emitnative: Put None/False/True in global native const table. So these constant objects can be loaded by dereferencing the REG_FUN_TABLE pointer instead of loading immediate values. This reduces the size of generated native code (when such constants are used), and means that pointers to these constants are no longer stored in the assembly code.
2018-10-14 13:56:46 +01:00
emit_native_mov_reg_const(emit, reg_dest, MP_F_CONST_NONE_OBJ);
py/emitnative: Consolidate use of stacked immediate values to one func. This commit adds the helper function load_reg_stack_imm() which deals with constant immediate values and converting them to Python objects if needed.
2018-10-14 12:58:27 +01:00
} else {
mp_raise_NotImplementedError("conversion to object");
}
return VTYPE_PYOBJ;
}
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void need_stack_settled(emit_t *emit) {
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
DEBUG_printf(" need_stack_settled; stack_size=%d\n", emit->stack_size);
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
for (int i = 0; i < emit->stack_size; i++) {
stack_info_t *si = &emit->stack_info[i];
if (si->kind == STACK_REG) {
py: Don't use anonymous unions, name them instead. This makes the code (more) compatible with the C99 standard.
2015-01-24 23:14:12 +00:00
DEBUG_printf(" reg(%u) to local(%u)\n", si->data.u_reg, emit->stack_start + i);
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
si->kind = STACK_VALUE;
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
emit_native_mov_state_reg(emit, emit->stack_start + i, si->data.u_reg);
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
for (int i = 0; i < emit->stack_size; i++) {
stack_info_t *si = &emit->stack_info[i];
if (si->kind == STACK_IMM) {
py: Don't use anonymous unions, name them instead. This makes the code (more) compatible with the C99 standard.
2015-01-24 23:14:12 +00:00
DEBUG_printf(" imm(" INT_FMT ") to local(%u)\n", si->data.u_imm, emit->stack_start + i);
py: Fix 2 bugs in native emitter: jump_or_pop and stack settling. Addresses issue #838.
2014-08-29 20:05:32 +01:00
si->kind = STACK_VALUE;
py/emitnative: Consolidate use of stacked immediate values to one func. This commit adds the helper function load_reg_stack_imm() which deals with constant immediate values and converting them to Python objects if needed.
2018-10-14 12:58:27 +01:00
si->vtype = load_reg_stack_imm(emit, REG_TEMP0, si, false);
emit_native_mov_state_reg(emit, emit->stack_start + i, REG_TEMP0);
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
}
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
// pos=1 is TOS, pos=2 is next, etc
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_access_stack(emit_t *emit, int pos, vtype_kind_t *vtype, int reg_dest) {
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
need_reg_single(emit, reg_dest, pos);
stack_info_t *si = &emit->stack_info[emit->stack_size - pos];
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
*vtype = si->vtype;
switch (si->kind) {
case STACK_VALUE:
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
emit_native_mov_reg_state(emit, reg_dest, emit->stack_start + emit->stack_size - pos);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
break;
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
case STACK_REG:
py: Don't use anonymous unions, name them instead. This makes the code (more) compatible with the C99 standard.
2015-01-24 23:14:12 +00:00
if (si->data.u_reg != reg_dest) {
ASM_MOV_REG_REG(emit->as, reg_dest, si->data.u_reg);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
break;
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
case STACK_IMM:
py/emitnative: Consolidate use of stacked immediate values to one func. This commit adds the helper function load_reg_stack_imm() which deals with constant immediate values and converting them to Python objects if needed.
2018-10-14 12:58:27 +01:00
*vtype = load_reg_stack_imm(emit, reg_dest, si, false);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
break;
}
}
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
// does an efficient X=pop(); discard(); push(X)
// needs a (non-temp) register in case the poped element was stored in the stack
STATIC void emit_fold_stack_top(emit_t *emit, int reg_dest) {
stack_info_t *si = &emit->stack_info[emit->stack_size - 2];
si[0] = si[1];
if (si->kind == STACK_VALUE) {
// if folded element was on the stack we need to put it in a register
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
emit_native_mov_reg_state(emit, reg_dest, emit->stack_start + emit->stack_size - 1);
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
si->kind = STACK_REG;
py: Don't use anonymous unions, name them instead. This makes the code (more) compatible with the C99 standard.
2015-01-24 23:14:12 +00:00
si->data.u_reg = reg_dest;
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
}
adjust_stack(emit, -1);
}
// If stacked value is in a register and the register is not r1 or r2, then
// *reg_dest is set to that register. Otherwise the value is put in *reg_dest.
STATIC void emit_pre_pop_reg_flexible(emit_t *emit, vtype_kind_t *vtype, int *reg_dest, int not_r1, int not_r2) {
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
emit->last_emit_was_return_value = false;
stack_info_t *si = peek_stack(emit, 0);
py: Don't use anonymous unions, name them instead. This makes the code (more) compatible with the C99 standard.
2015-01-24 23:14:12 +00:00
if (si->kind == STACK_REG && si->data.u_reg != not_r1 && si->data.u_reg != not_r2) {
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
*vtype = si->vtype;
py: Don't use anonymous unions, name them instead. This makes the code (more) compatible with the C99 standard.
2015-01-24 23:14:12 +00:00
*reg_dest = si->data.u_reg;
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
need_reg_single(emit, *reg_dest, 1);
} else {
emit_access_stack(emit, 1, vtype, *reg_dest);
}
adjust_stack(emit, -1);
}
py: Native emitter now supports delete name & global, and end finally.
2014-09-06 18:38:20 +01:00
STATIC void emit_pre_pop_discard(emit_t *emit) {
py: Combine native emitters to 1 glue function; distinguish viper. This patch simplifies the glue between native emitter and runtime, and handles viper code like inline assember: return values are converted to Python objects. Fixes issue #531.
2014-05-10 13:40:46 +01:00
emit->last_emit_was_return_value = false;
adjust_stack(emit, -1);
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_pre_pop_reg(emit_t *emit, vtype_kind_t *vtype, int reg_dest) {
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
emit->last_emit_was_return_value = false;
emit_access_stack(emit, 1, vtype, reg_dest);
adjust_stack(emit, -1);
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_pre_pop_reg_reg(emit_t *emit, vtype_kind_t *vtypea, int rega, vtype_kind_t *vtypeb, int regb) {
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit_pre_pop_reg(emit, vtypea, rega);
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
emit_pre_pop_reg(emit, vtypeb, regb);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_pre_pop_reg_reg_reg(emit_t *emit, vtype_kind_t *vtypea, int rega, vtype_kind_t *vtypeb, int regb, vtype_kind_t *vtypec, int regc) {
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit_pre_pop_reg(emit, vtypea, rega);
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
emit_pre_pop_reg(emit, vtypeb, regb);
emit_pre_pop_reg(emit, vtypec, regc);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_post(emit_t *emit) {
py, unix: Allow to compile with -Wunused-parameter. See issue #699.
2015-01-20 12:47:20 +00:00
(void)emit;
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
STATIC void emit_post_top_set_vtype(emit_t *emit, vtype_kind_t new_vtype) {
stack_info_t *si = &emit->stack_info[emit->stack_size - 1];
si->vtype = new_vtype;
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_post_push_reg(emit_t *emit, vtype_kind_t vtype, int reg) {
py/emitnative: Allocate space for local stack info as it's needed.
2018-08-07 07:05:44 +01:00
ensure_extra_stack(emit, 1);
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
stack_info_t *si = &emit->stack_info[emit->stack_size];
si->vtype = vtype;
si->kind = STACK_REG;
py: Don't use anonymous unions, name them instead. This makes the code (more) compatible with the C99 standard.
2015-01-24 23:14:12 +00:00
si->data.u_reg = reg;
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
adjust_stack(emit, 1);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Rename machine_(u)int_t to mp_(u)int_t. See discussion in issue #50.
2014-07-03 13:25:24 +01:00
STATIC void emit_post_push_imm(emit_t *emit, vtype_kind_t vtype, mp_int_t imm) {
py/emitnative: Allocate space for local stack info as it's needed.
2018-08-07 07:05:44 +01:00
ensure_extra_stack(emit, 1);
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
stack_info_t *si = &emit->stack_info[emit->stack_size];
si->vtype = vtype;
si->kind = STACK_IMM;
py: Don't use anonymous unions, name them instead. This makes the code (more) compatible with the C99 standard.
2015-01-24 23:14:12 +00:00
si->data.u_imm = imm;
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
adjust_stack(emit, 1);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_post_push_reg_reg(emit_t *emit, vtype_kind_t vtypea, int rega, vtype_kind_t vtypeb, int regb) {
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
emit_post_push_reg(emit, vtypea, rega);
emit_post_push_reg(emit, vtypeb, regb);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_post_push_reg_reg_reg(emit_t *emit, vtype_kind_t vtypea, int rega, vtype_kind_t vtypeb, int regb, vtype_kind_t vtypec, int regc) {
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
emit_post_push_reg(emit, vtypea, rega);
emit_post_push_reg(emit, vtypeb, regb);
emit_post_push_reg(emit, vtypec, regc);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_post_push_reg_reg_reg_reg(emit_t *emit, vtype_kind_t vtypea, int rega, vtype_kind_t vtypeb, int regb, vtype_kind_t vtypec, int regc, vtype_kind_t vtyped, int regd) {
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
emit_post_push_reg(emit, vtypea, rega);
emit_post_push_reg(emit, vtypeb, regb);
emit_post_push_reg(emit, vtypec, regc);
emit_post_push_reg(emit, vtyped, regd);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Code clean-up in native emitter; improve thumb native calls.
2014-08-16 22:31:57 +01:00
STATIC void emit_call(emit_t *emit, mp_fun_kind_t fun_kind) {
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
need_reg_all(emit);
py/emitnative: Remove unused ptr argument from ASM_CALL_IND macro.
2018-10-13 05:13:55 +01:00
ASM_CALL_IND(emit->as, fun_kind);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Code clean-up in native emitter; improve thumb native calls.
2014-08-16 22:31:57 +01:00
STATIC void emit_call_with_imm_arg(emit_t *emit, mp_fun_kind_t fun_kind, mp_int_t arg_val, int arg_reg) {
Simplify and improve function & method calling.
2013-10-10 22:06:54 +01:00
need_reg_all(emit);
py/emitnative: Clean up asm macro names so they have dest as first arg. All the asm macro names that convert a particular architecture to a generic interface now follow the convention whereby the "destination" (usually a register) is specified first.
2017-11-15 00:46:49 +00:00
ASM_MOV_REG_IMM(emit->as, arg_reg, arg_val);
py/emitnative: Remove unused ptr argument from ASM_CALL_IND macro.
2018-10-13 05:13:55 +01:00
ASM_CALL_IND(emit->as, fun_kind);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Code clean-up in native emitter; improve thumb native calls.
2014-08-16 22:31:57 +01:00
STATIC void emit_call_with_2_imm_args(emit_t *emit, mp_fun_kind_t fun_kind, mp_int_t arg_val1, int arg_reg1, mp_int_t arg_val2, int arg_reg2) {
py: Partially fix native emitter to work with latest runtime. Native emitter has been broken since stack order has changed from reverse to standard. This fix gets it partially working.
2014-02-02 13:11:48 +00:00
need_reg_all(emit);
py/emitnative: Clean up asm macro names so they have dest as first arg. All the asm macro names that convert a particular architecture to a generic interface now follow the convention whereby the "destination" (usually a register) is specified first.
2017-11-15 00:46:49 +00:00
ASM_MOV_REG_IMM(emit->as, arg_reg1, arg_val1);
ASM_MOV_REG_IMM(emit->as, arg_reg2, arg_val2);
py/emitnative: Remove unused ptr argument from ASM_CALL_IND macro.
2018-10-13 05:13:55 +01:00
ASM_CALL_IND(emit->as, fun_kind);
py: Partially fix native emitter to work with latest runtime. Native emitter has been broken since stack order has changed from reverse to standard. This fix gets it partially working.
2014-02-02 13:11:48 +00:00
}
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
STATIC void emit_call_with_qstr_arg(emit_t *emit, mp_fun_kind_t fun_kind, qstr qst, int arg_reg) {
need_reg_all(emit);
emit_native_mov_reg_qstr(emit, arg_reg, qst);
ASM_CALL_IND(emit->as, fun_kind);
}
py: Viper can call functions with native types, and raise exceptions.
2014-08-16 22:06:11 +01:00
// vtype of all n_pop objects is VTYPE_PYOBJ
// Will convert any items that are not VTYPE_PYOBJ to this type and put them back on the stack.
// If any conversions of non-immediate values are needed, then it uses REG_ARG_1, REG_ARG_2 and REG_RET.
// Otherwise, it does not use any temporary registers (but may use reg_dest before loading it with stack pointer).
STATIC void emit_get_stack_pointer_to_reg_for_pop(emit_t *emit, mp_uint_t reg_dest, mp_uint_t n_pop) {
need_reg_all(emit);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
py: Viper can call functions with native types, and raise exceptions.
2014-08-16 22:06:11 +01:00
// First, store any immediate values to their respective place on the stack.
for (mp_uint_t i = 0; i < n_pop; i++) {
stack_info_t *si = &emit->stack_info[emit->stack_size - 1 - i];
// must push any imm's to stack
// must convert them to VTYPE_PYOBJ for viper code
if (si->kind == STACK_IMM) {
si->kind = STACK_VALUE;
py/emitnative: Consolidate use of stacked immediate values to one func. This commit adds the helper function load_reg_stack_imm() which deals with constant immediate values and converting them to Python objects if needed.
2018-10-14 12:58:27 +01:00
si->vtype = load_reg_stack_imm(emit, reg_dest, si, true);
emit_native_mov_state_reg(emit, emit->stack_start + emit->stack_size - 1 - i, reg_dest);
py: Viper can call functions with native types, and raise exceptions.
2014-08-16 22:06:11 +01:00
}
// verify that this value is on the stack
assert(si->kind == STACK_VALUE);
}
// Second, convert any non-VTYPE_PYOBJ to that type.
for (mp_uint_t i = 0; i < n_pop; i++) {
stack_info_t *si = &emit->stack_info[emit->stack_size - 1 - i];
if (si->vtype != VTYPE_PYOBJ) {
mp_uint_t local_num = emit->stack_start + emit->stack_size - 1 - i;
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
emit_native_mov_reg_state(emit, REG_ARG_1, local_num);
py: Code clean-up in native emitter; improve thumb native calls.
2014-08-16 22:31:57 +01:00
emit_call_with_imm_arg(emit, MP_F_CONVERT_NATIVE_TO_OBJ, si->vtype, REG_ARG_2); // arg2 = type
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
emit_native_mov_state_reg(emit, local_num, REG_RET);
py: Viper can call functions with native types, and raise exceptions.
2014-08-16 22:06:11 +01:00
si->vtype = VTYPE_PYOBJ;
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
DEBUG_printf(" convert_native_to_obj(local_num=" UINT_FMT ")\n", local_num);
py: Viper can call functions with native types, and raise exceptions.
2014-08-16 22:06:11 +01:00
}
}
// Adujust the stack for a pop of n_pop items, and load the stack pointer into reg_dest.
adjust_stack(emit, -n_pop);
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
emit_native_mov_reg_state_addr(emit, reg_dest, emit->stack_start + emit->stack_size);
py: Viper can call functions with native types, and raise exceptions.
2014-08-16 22:06:11 +01:00
}
// vtype of all n_push objects is VTYPE_PYOBJ
STATIC void emit_get_stack_pointer_to_reg_for_push(emit_t *emit, mp_uint_t reg_dest, mp_uint_t n_push) {
need_reg_all(emit);
py/emitnative: Allocate space for local stack info as it's needed.
2018-08-07 07:05:44 +01:00
ensure_extra_stack(emit, n_push);
py: Viper can call functions with native types, and raise exceptions.
2014-08-16 22:06:11 +01:00
for (mp_uint_t i = 0; i < n_push; i++) {
emit->stack_info[emit->stack_size + i].kind = STACK_VALUE;
emit->stack_info[emit->stack_size + i].vtype = VTYPE_PYOBJ;
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
emit_native_mov_reg_state_addr(emit, reg_dest, emit->stack_start + emit->stack_size);
py: Viper can call functions with native types, and raise exceptions.
2014-08-16 22:06:11 +01:00
adjust_stack(emit, n_push);
}
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
STATIC void emit_native_push_exc_stack(emit_t *emit, uint label, bool is_finally) {
if (emit->exc_stack_size + 1 > emit->exc_stack_alloc) {
size_t new_alloc = emit->exc_stack_alloc + 4;
emit->exc_stack = m_renew(exc_stack_entry_t, emit->exc_stack, emit->exc_stack_alloc, new_alloc);
emit->exc_stack_alloc = new_alloc;
}
exc_stack_entry_t *e = &emit->exc_stack[emit->exc_stack_size++];
e->label = label;
e->is_finally = is_finally;
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
e->unwind_label = UNWIND_LABEL_UNUSED;
e->is_active = true;
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
ASM_MOV_REG_PCREL(emit->as, REG_RET, label);
ASM_MOV_LOCAL_REG(emit->as, LOCAL_IDX_EXC_HANDLER_PC(emit), REG_RET);
}
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
STATIC void emit_native_leave_exc_stack(emit_t *emit, bool start_of_handler) {
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
assert(emit->exc_stack_size > 0);
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
// Get current exception handler and deactivate it
exc_stack_entry_t *e = &emit->exc_stack[emit->exc_stack_size - 1];
e->is_active = false;
// Find next innermost active exception handler, to restore as current handler
for (--e; e >= emit->exc_stack && !e->is_active; --e) {
}
// Update the PC of the new exception handler
if (e < emit->exc_stack) {
// No active handler, clear handler PC to zero
if (start_of_handler) {
// Optimisation: PC is already cleared by global exc handler
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
return;
}
ASM_XOR_REG_REG(emit->as, REG_RET, REG_RET);
} else {
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
// Found new active handler, get its PC
ASM_MOV_REG_PCREL(emit->as, REG_RET, e->label);
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
}
ASM_MOV_LOCAL_REG(emit->as, LOCAL_IDX_EXC_HANDLER_PC(emit), REG_RET);
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
}
STATIC exc_stack_entry_t *emit_native_pop_exc_stack(emit_t *emit) {
assert(emit->exc_stack_size > 0);
exc_stack_entry_t *e = &emit->exc_stack[--emit->exc_stack_size];
assert(e->is_active == false);
return e;
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
}
py/emitnative: Change type of const_table from uintptr_t to mp_uint_t. This matches how bytecode does it, and matches the signature of mp_emit_glue_assign_native. Since the native emitter doesn't support nan-boxing uintptr_t and mp_uint_t are anyway the same bit-width.
2018-09-27 15:04:10 +01:00
STATIC void emit_load_reg_with_ptr(emit_t *emit, int reg, mp_uint_t ptr, size_t table_off) {
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
if (!emit->do_viper_types) {
// Skip qstr names of arguments
table_off += emit->scope->num_pos_args + emit->scope->num_kwonly_args;
}
if (emit->pass == MP_PASS_EMIT) {
emit->const_table[table_off] = ptr;
}
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
emit_native_mov_reg_state(emit, REG_TEMP0, LOCAL_IDX_FUN_OBJ(emit));
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
ASM_LOAD_REG_REG_OFFSET(emit->as, REG_TEMP0, REG_TEMP0, offsetof(mp_obj_fun_bc_t, const_table) / sizeof(uintptr_t));
ASM_LOAD_REG_REG_OFFSET(emit->as, reg, REG_TEMP0, table_off);
}
STATIC void emit_load_reg_with_object(emit_t *emit, int reg, mp_obj_t obj) {
py/emitnative: Adjust accounting of size of const_table. n_obj no longer includes a count for mp_fun_table to make it a bit simpler.
2019-02-25 04:57:35 +00:00
// First entry is for mp_fun_table
size_t table_off = 1 + emit->const_table_cur_obj++;
py/emitnative: Change type of const_table from uintptr_t to mp_uint_t. This matches how bytecode does it, and matches the signature of mp_emit_glue_assign_native. Since the native emitter doesn't support nan-boxing uintptr_t and mp_uint_t are anyway the same bit-width.
2018-09-27 15:04:10 +01:00
emit_load_reg_with_ptr(emit, reg, (mp_uint_t)obj, table_off);
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
}
STATIC void emit_load_reg_with_raw_code(emit_t *emit, int reg, mp_raw_code_t *rc) {
py/emitnative: Adjust accounting of size of const_table. n_obj no longer includes a count for mp_fun_table to make it a bit simpler.
2019-02-25 04:57:35 +00:00
// First entry is for mp_fun_table, then constant objects
size_t table_off = 1 + emit->const_table_num_obj + emit->const_table_cur_raw_code++;
py/emitnative: Change type of const_table from uintptr_t to mp_uint_t. This matches how bytecode does it, and matches the signature of mp_emit_glue_assign_native. Since the native emitter doesn't support nan-boxing uintptr_t and mp_uint_t are anyway the same bit-width.
2018-09-27 15:04:10 +01:00
emit_load_reg_with_ptr(emit, reg, (mp_uint_t)rc, table_off);
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
}
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
STATIC void emit_native_label_assign(emit_t *emit, mp_uint_t l) {
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
DEBUG_printf("label_assign(" UINT_FMT ")\n", l);
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
bool is_finally = false;
if (emit->exc_stack_size > 0) {
exc_stack_entry_t *e = &emit->exc_stack[emit->exc_stack_size - 1];
is_finally = e->is_finally && e->label == l;
}
if (is_finally) {
// Label is at start of finally handler: store TOS into exception slot
vtype_kind_t vtype;
emit_pre_pop_reg(emit, &vtype, REG_TEMP0);
ASM_MOV_LOCAL_REG(emit->as, LOCAL_IDX_EXC_VAL(emit), REG_TEMP0);
}
py: Rename emit_pre so they have globally unique names.
2014-03-27 23:30:26 +00:00
emit_native_pre(emit);
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
// need to commit stack because we can jump here from elsewhere
need_stack_settled(emit);
py: Factor out common code from assemblers into asmbase.[ch]. All assemblers should "derive" from mp_asm_base_t.
2016-11-27 22:24:50 +00:00
mp_asm_base_label_assign(&emit->as->base, l);
Fix bug: emit native didn't clear last_was_return in label_assign.
2013-11-02 20:34:54 +00:00
emit_post(emit);
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
if (is_finally) {
// Label is at start of finally handler: pop exception stack
py/emitnative: Fix try-finally in outer scope, so finally is cancelled.
2018-09-11 06:33:25 +01:00
emit_native_leave_exc_stack(emit, false);
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
}
}
STATIC void emit_native_global_exc_entry(emit_t *emit) {
// Note: 4 labels are reserved for this function, starting at *emit->label_slot
emit->exit_label = *emit->label_slot;
if (NEED_GLOBAL_EXC_HANDLER(emit)) {
mp_uint_t nlr_label = *emit->label_slot + 1;
mp_uint_t start_label = *emit->label_slot + 2;
mp_uint_t global_except_label = *emit->label_slot + 3;
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
if (!(emit->scope->scope_flags & MP_SCOPE_FLAG_GENERATOR)) {
// Set new globals
emit_native_mov_reg_state(emit, REG_ARG_1, LOCAL_IDX_FUN_OBJ(emit));
ASM_LOAD_REG_REG_OFFSET(emit->as, REG_ARG_1, REG_ARG_1, offsetof(mp_obj_fun_bc_t, globals) / sizeof(uintptr_t));
emit_call(emit, MP_F_NATIVE_SWAP_GLOBALS);
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
// Save old globals (or NULL if globals didn't change)
emit_native_mov_state_reg(emit, LOCAL_IDX_OLD_GLOBALS(emit), REG_RET);
}
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
py: Fix native functions so they run with their correct globals context. Prior to this commit a function compiled with the native decorator @micropython.native would not work correctly when accessing global variables, because the globals dict was not being set upon function entry. This commit fixes this problem by, upon function entry, setting as the current globals dict the globals dict context the function was defined within, as per normal Python semantics, and as bytecode does. Upon function exit the original globals dict is restored. In order to restore the globals dict when an exception is raised the native function must guard its internals with an nlr_push/nlr_pop pair. Because this push/pop is relatively expensive, in both C stack usage for the nlr_buf_t and CPU execution time, the implementation here optimises things as much as possible. First, the compiler keeps track of whether a function even needs to access global variables. Using this information the native emitter then generates three different kinds of code: 1. no globals used, no exception handlers: no nlr handling code and no setting of the globals dict. 2. globals used, no exception handlers: an nlr_buf_t is allocated on the C stack but it is not used if the globals dict is unchanged, saving execution time because nlr_push/nlr_pop don't need to run. 3. function has exception handlers, may use globals: an nlr_buf_t is allocated and nlr_push/nlr_pop are always called. In the end, native functions that don't access globals and don't have exception handlers will run more efficiently than those that do. Fixes issue #1573.
2018-09-13 13:03:48 +01:00
if (emit->scope->exc_stack_size == 0) {
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
if (!(emit->scope->scope_flags & MP_SCOPE_FLAG_GENERATOR)) {
// Optimisation: if globals didn't change don't push the nlr context
ASM_JUMP_IF_REG_ZERO(emit->as, REG_RET, start_label, false);
}
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
py: Fix native functions so they run with their correct globals context. Prior to this commit a function compiled with the native decorator @micropython.native would not work correctly when accessing global variables, because the globals dict was not being set upon function entry. This commit fixes this problem by, upon function entry, setting as the current globals dict the globals dict context the function was defined within, as per normal Python semantics, and as bytecode does. Upon function exit the original globals dict is restored. In order to restore the globals dict when an exception is raised the native function must guard its internals with an nlr_push/nlr_pop pair. Because this push/pop is relatively expensive, in both C stack usage for the nlr_buf_t and CPU execution time, the implementation here optimises things as much as possible. First, the compiler keeps track of whether a function even needs to access global variables. Using this information the native emitter then generates three different kinds of code: 1. no globals used, no exception handlers: no nlr handling code and no setting of the globals dict. 2. globals used, no exception handlers: an nlr_buf_t is allocated on the C stack but it is not used if the globals dict is unchanged, saving execution time because nlr_push/nlr_pop don't need to run. 3. function has exception handlers, may use globals: an nlr_buf_t is allocated and nlr_push/nlr_pop are always called. In the end, native functions that don't access globals and don't have exception handlers will run more efficiently than those that do. Fixes issue #1573.
2018-09-13 13:03:48 +01:00
// Wrap everything in an nlr context
py/emitnative: Reorder native state on C stack so nlr_buf_t is first. The nlr_buf_t doesn't need to be part of the Python value stack (as it was before this commit), it's simpler to have it separated as auxiliary state that lives on the C stack. This will help adding yield support because in that case the nlr_buf_t and Python value stack live in separate memory areas (C stack and heap respectively).
2018-10-01 03:36:21 +01:00
ASM_MOV_REG_LOCAL_ADDR(emit->as, REG_ARG_1, 0);
py: Fix native functions so they run with their correct globals context. Prior to this commit a function compiled with the native decorator @micropython.native would not work correctly when accessing global variables, because the globals dict was not being set upon function entry. This commit fixes this problem by, upon function entry, setting as the current globals dict the globals dict context the function was defined within, as per normal Python semantics, and as bytecode does. Upon function exit the original globals dict is restored. In order to restore the globals dict when an exception is raised the native function must guard its internals with an nlr_push/nlr_pop pair. Because this push/pop is relatively expensive, in both C stack usage for the nlr_buf_t and CPU execution time, the implementation here optimises things as much as possible. First, the compiler keeps track of whether a function even needs to access global variables. Using this information the native emitter then generates three different kinds of code: 1. no globals used, no exception handlers: no nlr handling code and no setting of the globals dict. 2. globals used, no exception handlers: an nlr_buf_t is allocated on the C stack but it is not used if the globals dict is unchanged, saving execution time because nlr_push/nlr_pop don't need to run. 3. function has exception handlers, may use globals: an nlr_buf_t is allocated and nlr_push/nlr_pop are always called. In the end, native functions that don't access globals and don't have exception handlers will run more efficiently than those that do. Fixes issue #1573.
2018-09-13 13:03:48 +01:00
emit_call(emit, MP_F_NLR_PUSH);
ASM_JUMP_IF_REG_ZERO(emit->as, REG_RET, start_label, true);
} else {
// Clear the unwind state
ASM_XOR_REG_REG(emit->as, REG_TEMP0, REG_TEMP0);
ASM_MOV_LOCAL_REG(emit->as, LOCAL_IDX_EXC_HANDLER_UNWIND(emit), REG_TEMP0);
// Put PC of start code block into REG_LOCAL_1
ASM_MOV_REG_PCREL(emit->as, REG_LOCAL_1, start_label);
// Wrap everything in an nlr context
emit_native_label_assign(emit, nlr_label);
ASM_MOV_REG_LOCAL(emit->as, REG_LOCAL_2, LOCAL_IDX_EXC_HANDLER_UNWIND(emit));
py/emitnative: Reorder native state on C stack so nlr_buf_t is first. The nlr_buf_t doesn't need to be part of the Python value stack (as it was before this commit), it's simpler to have it separated as auxiliary state that lives on the C stack. This will help adding yield support because in that case the nlr_buf_t and Python value stack live in separate memory areas (C stack and heap respectively).
2018-10-01 03:36:21 +01:00
ASM_MOV_REG_LOCAL_ADDR(emit->as, REG_ARG_1, 0);
py: Fix native functions so they run with their correct globals context. Prior to this commit a function compiled with the native decorator @micropython.native would not work correctly when accessing global variables, because the globals dict was not being set upon function entry. This commit fixes this problem by, upon function entry, setting as the current globals dict the globals dict context the function was defined within, as per normal Python semantics, and as bytecode does. Upon function exit the original globals dict is restored. In order to restore the globals dict when an exception is raised the native function must guard its internals with an nlr_push/nlr_pop pair. Because this push/pop is relatively expensive, in both C stack usage for the nlr_buf_t and CPU execution time, the implementation here optimises things as much as possible. First, the compiler keeps track of whether a function even needs to access global variables. Using this information the native emitter then generates three different kinds of code: 1. no globals used, no exception handlers: no nlr handling code and no setting of the globals dict. 2. globals used, no exception handlers: an nlr_buf_t is allocated on the C stack but it is not used if the globals dict is unchanged, saving execution time because nlr_push/nlr_pop don't need to run. 3. function has exception handlers, may use globals: an nlr_buf_t is allocated and nlr_push/nlr_pop are always called. In the end, native functions that don't access globals and don't have exception handlers will run more efficiently than those that do. Fixes issue #1573.
2018-09-13 13:03:48 +01:00
emit_call(emit, MP_F_NLR_PUSH);
ASM_MOV_LOCAL_REG(emit->as, LOCAL_IDX_EXC_HANDLER_UNWIND(emit), REG_LOCAL_2);
ASM_JUMP_IF_REG_NONZERO(emit->as, REG_RET, global_except_label, true);
// Clear PC of current code block, and jump there to resume execution
ASM_XOR_REG_REG(emit->as, REG_TEMP0, REG_TEMP0);
ASM_MOV_LOCAL_REG(emit->as, LOCAL_IDX_EXC_HANDLER_PC(emit), REG_TEMP0);
ASM_JUMP_REG(emit->as, REG_LOCAL_1);
// Global exception handler: check for valid exception handler
emit_native_label_assign(emit, global_except_label);
ASM_MOV_REG_LOCAL(emit->as, REG_LOCAL_1, LOCAL_IDX_EXC_HANDLER_PC(emit));
ASM_JUMP_IF_REG_NONZERO(emit->as, REG_LOCAL_1, nlr_label, false);
}
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
if (!(emit->scope->scope_flags & MP_SCOPE_FLAG_GENERATOR)) {
// Restore old globals
emit_native_mov_reg_state(emit, REG_ARG_1, LOCAL_IDX_OLD_GLOBALS(emit));
emit_call(emit, MP_F_NATIVE_SWAP_GLOBALS);
}
if (emit->scope->scope_flags & MP_SCOPE_FLAG_GENERATOR) {
// Store return value in state[0]
ASM_MOV_REG_LOCAL(emit->as, REG_TEMP0, LOCAL_IDX_EXC_VAL(emit));
ASM_STORE_REG_REG_OFFSET(emit->as, REG_TEMP0, REG_GENERATOR_STATE, offsetof(mp_code_state_t, state) / sizeof(uintptr_t));
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
// Load return kind
ASM_MOV_REG_IMM(emit->as, REG_RET, MP_VM_RETURN_EXCEPTION);
ASM_EXIT(emit->as);
} else {
// Re-raise exception out to caller
ASM_MOV_REG_LOCAL(emit->as, REG_ARG_1, LOCAL_IDX_EXC_VAL(emit));
emit_call(emit, MP_F_NATIVE_RAISE);
}
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
// Label for start of function
emit_native_label_assign(emit, start_label);
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
if (emit->scope->scope_flags & MP_SCOPE_FLAG_GENERATOR) {
emit_native_mov_reg_state(emit, REG_TEMP0, LOCAL_IDX_GEN_PC(emit));
ASM_JUMP_REG(emit->as, REG_TEMP0);
emit->start_offset = mp_asm_base_get_code_pos(&emit->as->base);
// This is the first entry of the generator
// Check LOCAL_IDX_EXC_VAL for any injected value
ASM_MOV_REG_LOCAL(emit->as, REG_ARG_1, LOCAL_IDX_EXC_VAL(emit));
emit_call(emit, MP_F_NATIVE_RAISE);
}
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
}
}
STATIC void emit_native_global_exc_exit(emit_t *emit) {
// Label for end of function
emit_native_label_assign(emit, emit->exit_label);
if (NEED_GLOBAL_EXC_HANDLER(emit)) {
py/emitnative: Make viper funcs run with their correct globals context. Viper functions will now capture the globals at the point they were defined and use these globals when executing.
2018-09-15 13:37:07 +01:00
// Get old globals
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
if (!(emit->scope->scope_flags & MP_SCOPE_FLAG_GENERATOR)) {
emit_native_mov_reg_state(emit, REG_ARG_1, LOCAL_IDX_OLD_GLOBALS(emit));
py: Fix native functions so they run with their correct globals context. Prior to this commit a function compiled with the native decorator @micropython.native would not work correctly when accessing global variables, because the globals dict was not being set upon function entry. This commit fixes this problem by, upon function entry, setting as the current globals dict the globals dict context the function was defined within, as per normal Python semantics, and as bytecode does. Upon function exit the original globals dict is restored. In order to restore the globals dict when an exception is raised the native function must guard its internals with an nlr_push/nlr_pop pair. Because this push/pop is relatively expensive, in both C stack usage for the nlr_buf_t and CPU execution time, the implementation here optimises things as much as possible. First, the compiler keeps track of whether a function even needs to access global variables. Using this information the native emitter then generates three different kinds of code: 1. no globals used, no exception handlers: no nlr handling code and no setting of the globals dict. 2. globals used, no exception handlers: an nlr_buf_t is allocated on the C stack but it is not used if the globals dict is unchanged, saving execution time because nlr_push/nlr_pop don't need to run. 3. function has exception handlers, may use globals: an nlr_buf_t is allocated and nlr_push/nlr_pop are always called. In the end, native functions that don't access globals and don't have exception handlers will run more efficiently than those that do. Fixes issue #1573.
2018-09-13 13:03:48 +01:00
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
if (emit->scope->exc_stack_size == 0) {
// Optimisation: if globals didn't change then don't restore them and don't do nlr_pop
ASM_JUMP_IF_REG_ZERO(emit->as, REG_ARG_1, emit->exit_label + 1, false);
}
py: Fix native functions so they run with their correct globals context. Prior to this commit a function compiled with the native decorator @micropython.native would not work correctly when accessing global variables, because the globals dict was not being set upon function entry. This commit fixes this problem by, upon function entry, setting as the current globals dict the globals dict context the function was defined within, as per normal Python semantics, and as bytecode does. Upon function exit the original globals dict is restored. In order to restore the globals dict when an exception is raised the native function must guard its internals with an nlr_push/nlr_pop pair. Because this push/pop is relatively expensive, in both C stack usage for the nlr_buf_t and CPU execution time, the implementation here optimises things as much as possible. First, the compiler keeps track of whether a function even needs to access global variables. Using this information the native emitter then generates three different kinds of code: 1. no globals used, no exception handlers: no nlr handling code and no setting of the globals dict. 2. globals used, no exception handlers: an nlr_buf_t is allocated on the C stack but it is not used if the globals dict is unchanged, saving execution time because nlr_push/nlr_pop don't need to run. 3. function has exception handlers, may use globals: an nlr_buf_t is allocated and nlr_push/nlr_pop are always called. In the end, native functions that don't access globals and don't have exception handlers will run more efficiently than those that do. Fixes issue #1573.
2018-09-13 13:03:48 +01:00
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
// Restore old globals
emit_call(emit, MP_F_NATIVE_SWAP_GLOBALS);
}
py/emitnative: Make viper funcs run with their correct globals context. Viper functions will now capture the globals at the point they were defined and use these globals when executing.
2018-09-15 13:37:07 +01:00
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
// Pop the nlr context
emit_call(emit, MP_F_NLR_POP);
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
if (!(emit->scope->scope_flags & MP_SCOPE_FLAG_GENERATOR)) {
if (emit->scope->exc_stack_size == 0) {
// Destination label for above optimisation
emit_native_label_assign(emit, emit->exit_label + 1);
}
py: Fix native functions so they run with their correct globals context. Prior to this commit a function compiled with the native decorator @micropython.native would not work correctly when accessing global variables, because the globals dict was not being set upon function entry. This commit fixes this problem by, upon function entry, setting as the current globals dict the globals dict context the function was defined within, as per normal Python semantics, and as bytecode does. Upon function exit the original globals dict is restored. In order to restore the globals dict when an exception is raised the native function must guard its internals with an nlr_push/nlr_pop pair. Because this push/pop is relatively expensive, in both C stack usage for the nlr_buf_t and CPU execution time, the implementation here optimises things as much as possible. First, the compiler keeps track of whether a function even needs to access global variables. Using this information the native emitter then generates three different kinds of code: 1. no globals used, no exception handlers: no nlr handling code and no setting of the globals dict. 2. globals used, no exception handlers: an nlr_buf_t is allocated on the C stack but it is not used if the globals dict is unchanged, saving execution time because nlr_push/nlr_pop don't need to run. 3. function has exception handlers, may use globals: an nlr_buf_t is allocated and nlr_push/nlr_pop are always called. In the end, native functions that don't access globals and don't have exception handlers will run more efficiently than those that do. Fixes issue #1573.
2018-09-13 13:03:48 +01:00
}
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
// Load return value
ASM_MOV_REG_LOCAL(emit->as, REG_RET, LOCAL_IDX_RET_VAL(emit));
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
}
ASM_EXIT(emit->as);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Implement more features in native emitter. On x64, native emitter now passes 70 of the tests.
2014-04-06 12:58:40 +01:00
STATIC void emit_native_import_name(emit_t *emit, qstr qst) {
DEBUG_printf("import_name %s\n", qstr_str(qst));
py/emitnative: Fix native import emitter when in viper mode.
2016-11-10 11:53:04 +00:00
// get arguments from stack: arg2 = fromlist, arg3 = level
py/emitnative: Simplify viper mode handling in emit_native_import_name.
2018-10-14 13:06:09 +01:00
// If using viper types these arguments must be converted to proper objects, and
// to accomplish this viper types are turned off for the emit_pre_pop_reg_reg call.
bool orig_do_viper_types = emit->do_viper_types;
emit->do_viper_types = false;
vtype_kind_t vtype_fromlist;
vtype_kind_t vtype_level;
emit_pre_pop_reg_reg(emit, &vtype_fromlist, REG_ARG_2, &vtype_level, REG_ARG_3);
assert(vtype_fromlist == VTYPE_PYOBJ);
assert(vtype_level == VTYPE_PYOBJ);
emit->do_viper_types = orig_do_viper_types;
py/emitnative: Fix native import emitter when in viper mode.
2016-11-10 11:53:04 +00:00
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
emit_call_with_qstr_arg(emit, MP_F_IMPORT_NAME, qst, REG_ARG_1); // arg1 = import name
py: Implement more features in native emitter. On x64, native emitter now passes 70 of the tests.
2014-04-06 12:58:40 +01:00
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Implement more features in native emitter. On x64, native emitter now passes 70 of the tests.
2014-04-06 12:58:40 +01:00
STATIC void emit_native_import_from(emit_t *emit, qstr qst) {
DEBUG_printf("import_from %s\n", qstr_str(qst));
emit_native_pre(emit);
vtype_kind_t vtype_module;
emit_access_stack(emit, 1, &vtype_module, REG_ARG_1); // arg1 = module
assert(vtype_module == VTYPE_PYOBJ);
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
emit_call_with_qstr_arg(emit, MP_F_IMPORT_FROM, qst, REG_ARG_2); // arg2 = import name
py: Implement more features in native emitter. On x64, native emitter now passes 70 of the tests.
2014-04-06 12:58:40 +01:00
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_native_import_star(emit_t *emit) {
py: Implement more features in native emitter. On x64, native emitter now passes 70 of the tests.
2014-04-06 12:58:40 +01:00
DEBUG_printf("import_star\n");
vtype_kind_t vtype_module;
emit_pre_pop_reg(emit, &vtype_module, REG_ARG_1); // arg1 = module
assert(vtype_module == VTYPE_PYOBJ);
py: Code clean-up in native emitter; improve thumb native calls.
2014-08-16 22:31:57 +01:00
emit_call(emit, MP_F_IMPORT_ALL);
py: Implement more features in native emitter. On x64, native emitter now passes 70 of the tests.
2014-04-06 12:58:40 +01:00
emit_post(emit);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py/emit: Combine import from/name/star into one emit function. Change in code size is: bare-arm: +4 minimal x86: -88 unix x64: -456 unix nanbox: -88 stm32: -44 cc3200: +0 esp8266: -104 esp32: +8
2018-05-22 12:58:25 +01:00
STATIC void emit_native_import(emit_t *emit, qstr qst, int kind) {
if (kind == MP_EMIT_IMPORT_NAME) {
emit_native_import_name(emit, qst);
} else if (kind == MP_EMIT_IMPORT_FROM) {
emit_native_import_from(emit, qst);
} else {
emit_native_import_star(emit);
}
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_native_load_const_tok(emit_t *emit, mp_token_kind_t tok) {
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
DEBUG_printf("load_const_tok(tok=%u)\n", tok);
py/emitnative: Consolidate use of stacked immediate values to one func. This commit adds the helper function load_reg_stack_imm() which deals with constant immediate values and converting them to Python objects if needed.
2018-10-14 12:58:27 +01:00
if (tok == MP_TOKEN_ELLIPSIS) {
py: Add support to save native, viper and asm code to .mpy files. This commit adds support for saving and loading .mpy files that contain native code (native, viper and inline-asm). A lot of the ground work was already done for this in the form of removing pointers from generated native code. The changes here are mainly to link in qstr values to the native code, and change the format of .mpy files to contain native code blocks (possibly mixed with bytecode). A top-level summary: - @micropython.native, @micropython.viper and @micropython.asm_thumb/ asm_xtensa are now allowed in .py files when compiling to .mpy, and they work transparently to the user. - Entire .py files can be compiled to native via mpy-cross -X emit=native and for the most part the generated .mpy files should work the same as their bytecode version. - The .mpy file format is changed to 1) specify in the header if the file contains native code and if so the architecture (eg x86, ARMV7M, Xtensa); 2) for each function block the kind of code is specified (bytecode, native, viper, asm). - When native code is loaded from a .mpy file the native code must be modified (in place) to link qstr values in, just like bytecode (see py/persistentcode.c:arch_link_qstr() function). In addition, this now defines a public, native ABI for dynamically loadable native code generated by other languages, like C.
2019-02-21 04:18:33 +00:00
#if MICROPY_PERSISTENT_CODE_SAVE
emit_native_load_const_obj(emit, MP_OBJ_FROM_PTR(&mp_const_ellipsis_obj));
#else
py/emitnative: Consolidate use of stacked immediate values to one func. This commit adds the helper function load_reg_stack_imm() which deals with constant immediate values and converting them to Python objects if needed.
2018-10-14 12:58:27 +01:00
emit_post_push_imm(emit, VTYPE_PYOBJ, (mp_uint_t)MP_OBJ_FROM_PTR(&mp_const_ellipsis_obj));
py: Add support to save native, viper and asm code to .mpy files. This commit adds support for saving and loading .mpy files that contain native code (native, viper and inline-asm). A lot of the ground work was already done for this in the form of removing pointers from generated native code. The changes here are mainly to link in qstr values to the native code, and change the format of .mpy files to contain native code blocks (possibly mixed with bytecode). A top-level summary: - @micropython.native, @micropython.viper and @micropython.asm_thumb/ asm_xtensa are now allowed in .py files when compiling to .mpy, and they work transparently to the user. - Entire .py files can be compiled to native via mpy-cross -X emit=native and for the most part the generated .mpy files should work the same as their bytecode version. - The .mpy file format is changed to 1) specify in the header if the file contains native code and if so the architecture (eg x86, ARMV7M, Xtensa); 2) for each function block the kind of code is specified (bytecode, native, viper, asm). - When native code is loaded from a .mpy file the native code must be modified (in place) to link qstr values in, just like bytecode (see py/persistentcode.c:arch_link_qstr() function). In addition, this now defines a public, native ABI for dynamically loadable native code generated by other languages, like C.
2019-02-21 04:18:33 +00:00
#endif
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
} else {
py/emitnative: Consolidate use of stacked immediate values to one func. This commit adds the helper function load_reg_stack_imm() which deals with constant immediate values and converting them to Python objects if needed.
2018-10-14 12:58:27 +01:00
emit_native_pre(emit);
if (tok == MP_TOKEN_KW_NONE) {
emit_post_push_imm(emit, VTYPE_PTR_NONE, 0);
} else {
emit_post_push_imm(emit, VTYPE_BOOL, tok == MP_TOKEN_KW_FALSE ? 0 : 1);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
}
}
Rename machine_(u)int_t to mp_(u)int_t. See discussion in issue #50.
2014-07-03 13:25:24 +01:00
STATIC void emit_native_load_const_small_int(emit_t *emit, mp_int_t arg) {
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
DEBUG_printf("load_const_small_int(int=" INT_FMT ")\n", arg);
py: Rename emit_pre so they have globally unique names.
2014-03-27 23:30:26 +00:00
emit_native_pre(emit);
py/emitnative: Consolidate use of stacked immediate values to one func. This commit adds the helper function load_reg_stack_imm() which deals with constant immediate values and converting them to Python objects if needed.
2018-10-14 12:58:27 +01:00
emit_post_push_imm(emit, VTYPE_INT, arg);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Remove mp_load_const_bytes and instead load precreated bytes object. Previous to this patch each time a bytes object was referenced a new instance (with the same data) was created. With this patch a single bytes object is created in the compiler and is loaded directly at execute time as a true constant (similar to loading bignum and float objects). This saves on allocating RAM and means that bytes objects can now be used when the memory manager is locked (eg in interrupts). The MP_BC_LOAD_CONST_BYTES bytecode was removed as part of this. Generated bytecode is slightly larger due to storing a pointer to the bytes object instead of the qstr identifier. Code size is reduced by about 60 bytes on Thumb2 architectures.
2015-06-25 15:42:13 +01:00
STATIC void emit_native_load_const_str(emit_t *emit, qstr qst) {
py: Rename emit_pre so they have globally unique names.
2014-03-27 23:30:26 +00:00
emit_native_pre(emit);
py: Load strings as objects when compiling viper. Eventually, viper wants to be able to use raw pointers to strings and arrays for efficient access. But for now, let's just load strings as a Python object so they can be used as normal. This will anyway be compatible with eventual intended viper behaviour. Addresses issue #857.
2014-09-12 23:15:06 +01:00
// TODO: Eventually we want to be able to work with raw pointers in viper to
// do native array access. For now we just load them as any other object.
/*
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
if (emit->do_viper_types) {
// load a pointer to the asciiz string?
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
emit_post_push_imm(emit, VTYPE_PTR, (mp_uint_t)qstr_str(qst));
py: Load strings as objects when compiling viper. Eventually, viper wants to be able to use raw pointers to strings and arrays for efficient access. But for now, let's just load strings as a Python object so they can be used as normal. This will anyway be compatible with eventual intended viper behaviour. Addresses issue #857.
2014-09-12 23:15:06 +01:00
} else
*/
{
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
need_reg_single(emit, REG_TEMP0, 0);
emit_native_mov_reg_qstr_obj(emit, REG_TEMP0, qst);
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_TEMP0);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
}
py/emit: Change type of arg of load_const_obj from void* to mp_obj_t.
2015-11-27 12:41:25 +00:00
STATIC void emit_native_load_const_obj(emit_t *emit, mp_obj_t obj) {
py/emitnative: Consolidate where HASCONSTS is set to load-const-obj fun. Simplifies the code and fixes handling of the Ellipsis const in native code generation (which also needs the constant table so must set this flag).
2019-03-07 11:46:04 +00:00
emit->scope->scope_flags |= MP_SCOPE_FLAG_HASCONSTS;
py: Add load_const_obj to emitter, add LOAD_CONST_OBJ to bytecode. This allows to directly load a Python object to the Python stack. See issue #722 for background.
2015-01-13 15:55:54 +00:00
emit_native_pre(emit);
py, unix: Trace root pointers with native emitter under unix port. Native code has GC-heap pointers in it so it must be scanned. But on unix port memory for native functions is mmap'd, and so it must have explicit code to scan it for root pointers.
2015-01-14 00:11:09 +00:00
need_reg_single(emit, REG_RET, 0);
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
emit_load_reg_with_object(emit, REG_RET, obj);
py: Add load_const_obj to emitter, add LOAD_CONST_OBJ to bytecode. This allows to directly load a Python object to the Python stack. See issue #722 for background.
2015-01-13 15:55:54 +00:00
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
}
py: Making closures now passes pointer to stack, not a tuple for vars. Closed over variables are now passed on the stack, instead of creating a tuple and passing that. This way memory for the closed over variables can be allocated within the closure object itself. See issue #510 for background.
2014-04-20 17:50:40 +01:00
STATIC void emit_native_load_null(emit_t *emit) {
emit_native_pre(emit);
emit_post_push_imm(emit, VTYPE_PYOBJ, 0);
}
py: Remove unnecessary id_flags argument from emitter's load_fast. Saves 24 bytes in bare-arm.
2015-01-16 12:24:49 +00:00
STATIC void emit_native_load_fast(emit_t *emit, qstr qst, mp_uint_t local_num) {
DEBUG_printf("load_fast(%s, " UINT_FMT ")\n", qstr_str(qst), local_num);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
vtype_kind_t vtype = emit->local_vtype[local_num];
if (vtype == VTYPE_UNBOUND) {
py: Make viper codegen raise proper exception (ViperTypeError) on error. This fixes a long standing problem that viper code generation gave terrible error messages, and actually no errors on pyboard where assertions are disabled. Now all compile-time errors are raised as proper Python exceptions, and are of type ViperTypeError. Addresses issue #940.
2015-04-20 14:29:31 +01:00
EMIT_NATIVE_VIPER_TYPE_ERROR(emit, "local '%q' used before type known", qst);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Rename emit_pre so they have globally unique names.
2014-03-27 23:30:26 +00:00
emit_native_pre(emit);
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
if (local_num < REG_LOCAL_NUM && CAN_USE_REGS_FOR_LOCALS(emit)) {
py/emitnative: Use small tables to simplify handling of local regs.
2018-08-15 01:52:38 +01:00
emit_post_push_reg(emit, vtype, reg_local_table[local_num]);
Basic native ARM emitter
2014-08-16 21:55:53 +01:00
} else {
py: Make macro names in assemblers consistent, and tidy up a bit.
2014-09-29 16:25:04 +01:00
need_reg_single(emit, REG_TEMP0, 0);
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
emit_native_mov_reg_state(emit, REG_TEMP0, LOCAL_IDX_LOCAL_VAR(emit, local_num));
py: Make macro names in assemblers consistent, and tidy up a bit.
2014-09-29 16:25:04 +01:00
emit_post_push_reg(emit, vtype, REG_TEMP0);
Basic native ARM emitter
2014-08-16 21:55:53 +01:00
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
STATIC void emit_native_load_deref(emit_t *emit, qstr qst, mp_uint_t local_num) {
py: Implement closures in native code generator. Currently supports only x64 and Thumb2 archs.
2015-01-15 14:41:41 +00:00
DEBUG_printf("load_deref(%s, " UINT_FMT ")\n", qstr_str(qst), local_num);
need_reg_single(emit, REG_RET, 0);
emit_native_load_fast(emit, qst, local_num);
vtype_kind_t vtype;
int reg_base = REG_RET;
emit_pre_pop_reg_flexible(emit, &vtype, &reg_base, -1, -1);
ASM_LOAD_REG_REG_OFFSET(emit->as, REG_RET, reg_base, 1);
// closed over vars are always Python objects
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
py: work towards working closures.
2013-12-11 00:41:43 +00:00
}
py/emit: Combine fast and deref into one function for load/store/delete. Reduces code size by: bare-arm: -16 minimal x86: -208 unix x64: -408 unix nanbox: -248 stm32: -12 cc3200: -24 esp8266: -96 esp32: -44
2018-05-18 15:11:04 +01:00
STATIC void emit_native_load_local(emit_t *emit, qstr qst, mp_uint_t local_num, int kind) {
if (kind == MP_EMIT_IDOP_LOCAL_FAST) {
emit_native_load_fast(emit, qst, local_num);
} else {
emit_native_load_deref(emit, qst, local_num);
}
}
py/emit: Combine name and global into one func for load/store/delete. Reduces code size by: bare-arm: -56 minimal x86: -300 unix x64: -576 unix nanbox: -300 stm32: -164 cc3200: -56 esp8266: -236 esp32: -76
2018-05-22 12:16:30 +01:00
STATIC void emit_native_load_global(emit_t *emit, qstr qst, int kind) {
MP_STATIC_ASSERT(MP_F_LOAD_NAME + MP_EMIT_IDOP_GLOBAL_NAME == MP_F_LOAD_NAME);
MP_STATIC_ASSERT(MP_F_LOAD_NAME + MP_EMIT_IDOP_GLOBAL_GLOBAL == MP_F_LOAD_GLOBAL);
py: Rename emit_pre so they have globally unique names.
2014-03-27 23:30:26 +00:00
emit_native_pre(emit);
py/emit: Combine name and global into one func for load/store/delete. Reduces code size by: bare-arm: -56 minimal x86: -300 unix x64: -576 unix nanbox: -300 stm32: -164 cc3200: -56 esp8266: -236 esp32: -76
2018-05-22 12:16:30 +01:00
if (kind == MP_EMIT_IDOP_GLOBAL_NAME) {
DEBUG_printf("load_name(%s)\n", qstr_str(qst));
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
} else {
py/emit: Combine name and global into one func for load/store/delete. Reduces code size by: bare-arm: -56 minimal x86: -300 unix x64: -576 unix nanbox: -300 stm32: -164 cc3200: -56 esp8266: -236 esp32: -76
2018-05-22 12:16:30 +01:00
DEBUG_printf("load_global(%s)\n", qstr_str(qst));
if (emit->do_viper_types) {
// check for builtin casting operators
py/emitnative: Reuse mp_native_type_from_qstr when searching for a cast.
2018-09-15 04:50:04 +01:00
int native_type = mp_native_type_from_qstr(qst);
py/native: Improve support for bool type in viper functions. Variables with type bool now act more like an int, and there is proper casting to/from Python objects.
2019-05-03 14:18:30 +01:00
if (native_type >= MP_NATIVE_TYPE_BOOL) {
py/emitnative: Reuse mp_native_type_from_qstr when searching for a cast.
2018-09-15 04:50:04 +01:00
emit_post_push_imm(emit, VTYPE_BUILTIN_CAST, native_type);
py/emit: Combine name and global into one func for load/store/delete. Reduces code size by: bare-arm: -56 minimal x86: -300 unix x64: -576 unix nanbox: -300 stm32: -164 cc3200: -56 esp8266: -236 esp32: -76
2018-05-22 12:16:30 +01:00
return;
}
}
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
}
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
emit_call_with_qstr_arg(emit, MP_F_LOAD_NAME + kind, qst, REG_ARG_1);
py/emit: Combine name and global into one func for load/store/delete. Reduces code size by: bare-arm: -56 minimal x86: -300 unix x64: -576 unix nanbox: -300 stm32: -164 cc3200: -56 esp8266: -236 esp32: -76
2018-05-22 12:16:30 +01:00
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
STATIC void emit_native_load_attr(emit_t *emit, qstr qst) {
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
// depends on type of subject:
// - integer, function, pointer to integers: error
// - pointer to structure: get member, quite easy
Rename rt_* to mp_*. Mostly just a global search and replace. Except rt_is_true which becomes mp_obj_is_true. Still would like to tidy up some of the names, but this will do for now.
2014-03-30 13:35:08 +01:00
// - Python object: call mp_load_attr, and needs to be typed to convert result
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
vtype_kind_t vtype_base;
emit_pre_pop_reg(emit, &vtype_base, REG_ARG_1); // arg1 = base
assert(vtype_base == VTYPE_PYOBJ);
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
emit_call_with_qstr_arg(emit, MP_F_LOAD_ATTR, qst, REG_ARG_2); // arg2 = attribute name
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
}
py: Add LOAD_SUPER_METHOD bytecode to allow heap-free super meth calls. This patch allows the following code to run without allocating on the heap: super().foo(...) Before this patch such a call would allocate a super object on the heap and then load the foo method and call it right away. The super object is only needed to perform the lookup of the method and not needed after that. This patch makes an optimisation to allocate the super object on the C stack and discard it right after use. Changes in code size due to this patch are: bare-arm: +128 minimal: +232 unix x64: +416 unix nanbox: +364 stmhal: +184 esp8266: +340 cc3200: +128
2017-04-19 00:45:59 +01:00
STATIC void emit_native_load_method(emit_t *emit, qstr qst, bool is_super) {
if (is_super) {
emit_get_stack_pointer_to_reg_for_pop(emit, REG_ARG_2, 3); // arg2 = dest ptr
emit_get_stack_pointer_to_reg_for_push(emit, REG_ARG_2, 2); // arg2 = dest ptr
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
emit_call_with_qstr_arg(emit, MP_F_LOAD_SUPER_METHOD, qst, REG_ARG_1); // arg1 = method name
py: Add LOAD_SUPER_METHOD bytecode to allow heap-free super meth calls. This patch allows the following code to run without allocating on the heap: super().foo(...) Before this patch such a call would allocate a super object on the heap and then load the foo method and call it right away. The super object is only needed to perform the lookup of the method and not needed after that. This patch makes an optimisation to allocate the super object on the C stack and discard it right after use. Changes in code size due to this patch are: bare-arm: +128 minimal: +232 unix x64: +416 unix nanbox: +364 stmhal: +184 esp8266: +340 cc3200: +128
2017-04-19 00:45:59 +01:00
} else {
vtype_kind_t vtype_base;
emit_pre_pop_reg(emit, &vtype_base, REG_ARG_1); // arg1 = base
assert(vtype_base == VTYPE_PYOBJ);
emit_get_stack_pointer_to_reg_for_push(emit, REG_ARG_3, 2); // arg3 = dest ptr
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
emit_call_with_qstr_arg(emit, MP_F_LOAD_METHOD, qst, REG_ARG_2); // arg2 = method name
py: Add LOAD_SUPER_METHOD bytecode to allow heap-free super meth calls. This patch allows the following code to run without allocating on the heap: super().foo(...) Before this patch such a call would allocate a super object on the heap and then load the foo method and call it right away. The super object is only needed to perform the lookup of the method and not needed after that. This patch makes an optimisation to allocate the super object on the C stack and discard it right after use. Changes in code size due to this patch are: bare-arm: +128 minimal: +232 unix x64: +416 unix nanbox: +364 stmhal: +184 esp8266: +340 cc3200: +128
2017-04-19 00:45:59 +01:00
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_native_load_build_class(emit_t *emit) {
py: Rename emit_pre so they have globally unique names.
2014-03-27 23:30:26 +00:00
emit_native_pre(emit);
py: Code clean-up in native emitter; improve thumb native calls.
2014-08-16 22:31:57 +01:00
emit_call(emit, MP_F_LOAD_BUILD_CLASS);
Implement basic class/object in native code.
2013-10-10 11:24:39 +01:00
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Merge BINARY_OP_SUBSCR and store_subscr (w/ delete) into subscr. mp_obj_t->subscr now does load/store/delete.
2014-04-17 22:10:53 +01:00
STATIC void emit_native_load_subscr(emit_t *emit) {
py: Implement native load for viper. Viper can now do: ptr8(buf)[0], which loads a byte from a buffer using machine instructions.
2014-10-12 16:59:29 +01:00
DEBUG_printf("load_subscr\n");
// need to compile: base[index]
// pop: index, base
// optimise case where index is an immediate
vtype_kind_t vtype_base = peek_vtype(emit, 1);
if (vtype_base == VTYPE_PYOBJ) {
py: Implement implicit cast to obj for viper load/store index/value. This allows to do "ar[i]" and "ar[i] = val" in viper when ar is a Python object and i and/or val are native viper types (eg ints). Patch also includes tests for this feature.
2015-06-04 11:52:16 +01:00
// standard Python subscr
// TODO factor this implicit cast code with other uses of it
vtype_kind_t vtype_index = peek_vtype(emit, 0);
if (vtype_index == VTYPE_PYOBJ) {
emit_pre_pop_reg(emit, &vtype_index, REG_ARG_2);
} else {
emit_pre_pop_reg(emit, &vtype_index, REG_ARG_1);
emit_call_with_imm_arg(emit, MP_F_CONVERT_NATIVE_TO_OBJ, vtype_index, REG_ARG_2); // arg2 = type
ASM_MOV_REG_REG(emit->as, REG_ARG_2, REG_RET);
}
emit_pre_pop_reg(emit, &vtype_base, REG_ARG_1);
py: Code clean-up in native emitter; improve thumb native calls.
2014-08-16 22:31:57 +01:00
emit_call_with_imm_arg(emit, MP_F_OBJ_SUBSCR, (mp_uint_t)MP_OBJ_SENTINEL, REG_ARG_3);
py: Merge BINARY_OP_SUBSCR and store_subscr (w/ delete) into subscr. mp_obj_t->subscr now does load/store/delete.
2014-04-17 22:10:53 +01:00
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
} else {
py: Implement native load for viper. Viper can now do: ptr8(buf)[0], which loads a byte from a buffer using machine instructions.
2014-10-12 16:59:29 +01:00
// viper load
// TODO The different machine architectures have very different
// capabilities and requirements for loads, so probably best to
// write a completely separate load-optimiser for each one.
stack_info_t *top = peek_stack(emit, 0);
if (top->vtype == VTYPE_INT && top->kind == STACK_IMM) {
// index is an immediate
py: Don't use anonymous unions, name them instead. This makes the code (more) compatible with the C99 standard.
2015-01-24 23:14:12 +00:00
mp_int_t index_value = top->data.u_imm;
py: Implement native load for viper. Viper can now do: ptr8(buf)[0], which loads a byte from a buffer using machine instructions.
2014-10-12 16:59:29 +01:00
emit_pre_pop_discard(emit); // discard index
int reg_base = REG_ARG_1;
int reg_index = REG_ARG_2;
emit_pre_pop_reg_flexible(emit, &vtype_base, &reg_base, reg_index, reg_index);
switch (vtype_base) {
case VTYPE_PTR8: {
// pointer to 8-bit memory
// TODO optimise to use thumb ldrb r1, [r2, r3]
if (index_value != 0) {
// index is non-zero
#if N_THUMB
if (index_value > 0 && index_value < 32) {
asm_thumb_ldrb_rlo_rlo_i5(emit->as, REG_RET, reg_base, index_value);
break;
}
#endif
py/emitnative: Clean up asm macro names so they have dest as first arg. All the asm macro names that convert a particular architecture to a generic interface now follow the convention whereby the "destination" (usually a register) is specified first.
2017-11-15 00:46:49 +00:00
ASM_MOV_REG_IMM(emit->as, reg_index, index_value);
py: Implement native load for viper. Viper can now do: ptr8(buf)[0], which loads a byte from a buffer using machine instructions.
2014-10-12 16:59:29 +01:00
ASM_ADD_REG_REG(emit->as, reg_index, reg_base); // add index to base
reg_base = reg_index;
}
ASM_LOAD8_REG_REG(emit->as, REG_RET, reg_base); // load from (base+index)
break;
}
case VTYPE_PTR16: {
// pointer to 16-bit memory
if (index_value != 0) {
// index is a non-zero immediate
#if N_THUMB
if (index_value > 0 && index_value < 32) {
asm_thumb_ldrh_rlo_rlo_i5(emit->as, REG_RET, reg_base, index_value);
break;
}
#endif
py/emitnative: Clean up asm macro names so they have dest as first arg. All the asm macro names that convert a particular architecture to a generic interface now follow the convention whereby the "destination" (usually a register) is specified first.
2017-11-15 00:46:49 +00:00
ASM_MOV_REG_IMM(emit->as, reg_index, index_value << 1);
py: Implement native load for viper. Viper can now do: ptr8(buf)[0], which loads a byte from a buffer using machine instructions.
2014-10-12 16:59:29 +01:00
ASM_ADD_REG_REG(emit->as, reg_index, reg_base); // add 2*index to base
reg_base = reg_index;
}
ASM_LOAD16_REG_REG(emit->as, REG_RET, reg_base); // load from (base+2*index)
break;
}
py: Implement ptr32 load and store in viper emitter.
2015-10-13 00:50:17 +01:00
case VTYPE_PTR32: {
// pointer to 32-bit memory
if (index_value != 0) {
// index is a non-zero immediate
#if N_THUMB
if (index_value > 0 && index_value < 32) {
asm_thumb_ldr_rlo_rlo_i5(emit->as, REG_RET, reg_base, index_value);
break;
}
#endif
py/emitnative: Clean up asm macro names so they have dest as first arg. All the asm macro names that convert a particular architecture to a generic interface now follow the convention whereby the "destination" (usually a register) is specified first.
2017-11-15 00:46:49 +00:00
ASM_MOV_REG_IMM(emit->as, reg_index, index_value << 2);
py: Implement ptr32 load and store in viper emitter.
2015-10-13 00:50:17 +01:00
ASM_ADD_REG_REG(emit->as, reg_index, reg_base); // add 4*index to base
reg_base = reg_index;
}
ASM_LOAD32_REG_REG(emit->as, REG_RET, reg_base); // load from (base+4*index)
break;
}
py: Implement native load for viper. Viper can now do: ptr8(buf)[0], which loads a byte from a buffer using machine instructions.
2014-10-12 16:59:29 +01:00
default:
py: Make viper codegen raise proper exception (ViperTypeError) on error. This fixes a long standing problem that viper code generation gave terrible error messages, and actually no errors on pyboard where assertions are disabled. Now all compile-time errors are raised as proper Python exceptions, and are of type ViperTypeError. Addresses issue #940.
2015-04-20 14:29:31 +01:00
EMIT_NATIVE_VIPER_TYPE_ERROR(emit,
"can't load from '%q'", vtype_to_qstr(vtype_base));
py: Implement native load for viper. Viper can now do: ptr8(buf)[0], which loads a byte from a buffer using machine instructions.
2014-10-12 16:59:29 +01:00
}
} else {
// index is not an immediate
vtype_kind_t vtype_index;
int reg_index = REG_ARG_2;
emit_pre_pop_reg_flexible(emit, &vtype_index, &reg_index, REG_ARG_1, REG_ARG_1);
emit_pre_pop_reg(emit, &vtype_base, REG_ARG_1);
py/viper: Allow uint as index to load/store, and give better error msg.
2016-02-09 13:46:49 +00:00
if (vtype_index != VTYPE_INT && vtype_index != VTYPE_UINT) {
EMIT_NATIVE_VIPER_TYPE_ERROR(emit,
"can't load with '%q' index", vtype_to_qstr(vtype_index));
}
py: Implement native load for viper. Viper can now do: ptr8(buf)[0], which loads a byte from a buffer using machine instructions.
2014-10-12 16:59:29 +01:00
switch (vtype_base) {
case VTYPE_PTR8: {
// pointer to 8-bit memory
// TODO optimise to use thumb ldrb r1, [r2, r3]
ASM_ADD_REG_REG(emit->as, REG_ARG_1, reg_index); // add index to base
ASM_LOAD8_REG_REG(emit->as, REG_RET, REG_ARG_1); // store value to (base+index)
break;
}
case VTYPE_PTR16: {
// pointer to 16-bit memory
ASM_ADD_REG_REG(emit->as, REG_ARG_1, reg_index); // add index to base
ASM_ADD_REG_REG(emit->as, REG_ARG_1, reg_index); // add index to base
ASM_LOAD16_REG_REG(emit->as, REG_RET, REG_ARG_1); // load from (base+2*index)
break;
}
py: Implement ptr32 load and store in viper emitter.
2015-10-13 00:50:17 +01:00
case VTYPE_PTR32: {
// pointer to word-size memory
ASM_ADD_REG_REG(emit->as, REG_ARG_1, reg_index); // add index to base
ASM_ADD_REG_REG(emit->as, REG_ARG_1, reg_index); // add index to base
ASM_ADD_REG_REG(emit->as, REG_ARG_1, reg_index); // add index to base
ASM_ADD_REG_REG(emit->as, REG_ARG_1, reg_index); // add index to base
ASM_LOAD32_REG_REG(emit->as, REG_RET, REG_ARG_1); // load from (base+4*index)
break;
}
py: Implement native load for viper. Viper can now do: ptr8(buf)[0], which loads a byte from a buffer using machine instructions.
2014-10-12 16:59:29 +01:00
default:
py: Make viper codegen raise proper exception (ViperTypeError) on error. This fixes a long standing problem that viper code generation gave terrible error messages, and actually no errors on pyboard where assertions are disabled. Now all compile-time errors are raised as proper Python exceptions, and are of type ViperTypeError. Addresses issue #940.
2015-04-20 14:29:31 +01:00
EMIT_NATIVE_VIPER_TYPE_ERROR(emit,
"can't load from '%q'", vtype_to_qstr(vtype_base));
py: Implement native load for viper. Viper can now do: ptr8(buf)[0], which loads a byte from a buffer using machine instructions.
2014-10-12 16:59:29 +01:00
}
}
emit_post_push_reg(emit, VTYPE_INT, REG_RET);
py: Merge BINARY_OP_SUBSCR and store_subscr (w/ delete) into subscr. mp_obj_t->subscr now does load/store/delete.
2014-04-17 22:10:53 +01:00
}
}
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
STATIC void emit_native_store_fast(emit_t *emit, qstr qst, mp_uint_t local_num) {
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
vtype_kind_t vtype;
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
if (local_num < REG_LOCAL_NUM && CAN_USE_REGS_FOR_LOCALS(emit)) {
py/emitnative: Use small tables to simplify handling of local regs.
2018-08-15 01:52:38 +01:00
emit_pre_pop_reg(emit, &vtype, reg_local_table[local_num]);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
} else {
py: Make macro names in assemblers consistent, and tidy up a bit.
2014-09-29 16:25:04 +01:00
emit_pre_pop_reg(emit, &vtype, REG_TEMP0);
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
emit_native_mov_state_reg(emit, LOCAL_IDX_LOCAL_VAR(emit, local_num), REG_TEMP0);
Basic native ARM emitter
2014-08-16 21:55:53 +01:00
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit_post(emit);
// check types
if (emit->local_vtype[local_num] == VTYPE_UNBOUND) {
// first time this local is assigned, so give it a type of the object stored in it
emit->local_vtype[local_num] = vtype;
} else if (emit->local_vtype[local_num] != vtype) {
// type of local is not the same as object stored in it
py: Make viper codegen raise proper exception (ViperTypeError) on error. This fixes a long standing problem that viper code generation gave terrible error messages, and actually no errors on pyboard where assertions are disabled. Now all compile-time errors are raised as proper Python exceptions, and are of type ViperTypeError. Addresses issue #940.
2015-04-20 14:29:31 +01:00
EMIT_NATIVE_VIPER_TYPE_ERROR(emit,
"local '%q' has type '%q' but source is '%q'",
qst, vtype_to_qstr(emit->local_vtype[local_num]), vtype_to_qstr(vtype));
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
}
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
STATIC void emit_native_store_deref(emit_t *emit, qstr qst, mp_uint_t local_num) {
py: Implement closures in native code generator. Currently supports only x64 and Thumb2 archs.
2015-01-15 14:41:41 +00:00
DEBUG_printf("store_deref(%s, " UINT_FMT ")\n", qstr_str(qst), local_num);
need_reg_single(emit, REG_TEMP0, 0);
need_reg_single(emit, REG_TEMP1, 0);
emit_native_load_fast(emit, qst, local_num);
vtype_kind_t vtype;
int reg_base = REG_TEMP0;
emit_pre_pop_reg_flexible(emit, &vtype, &reg_base, -1, -1);
int reg_src = REG_TEMP1;
emit_pre_pop_reg_flexible(emit, &vtype, &reg_src, reg_base, reg_base);
ASM_STORE_REG_REG_OFFSET(emit->as, reg_src, reg_base, 1);
emit_post(emit);
py: work towards working closures.
2013-12-11 00:41:43 +00:00
}
py/emit: Combine fast and deref into one function for load/store/delete. Reduces code size by: bare-arm: -16 minimal x86: -208 unix x64: -408 unix nanbox: -248 stm32: -12 cc3200: -24 esp8266: -96 esp32: -44
2018-05-18 15:11:04 +01:00
STATIC void emit_native_store_local(emit_t *emit, qstr qst, mp_uint_t local_num, int kind) {
if (kind == MP_EMIT_IDOP_LOCAL_FAST) {
emit_native_store_fast(emit, qst, local_num);
} else {
emit_native_store_deref(emit, qst, local_num);
}
}
py/emit: Combine name and global into one func for load/store/delete. Reduces code size by: bare-arm: -56 minimal x86: -300 unix x64: -576 unix nanbox: -300 stm32: -164 cc3200: -56 esp8266: -236 esp32: -76
2018-05-22 12:16:30 +01:00
STATIC void emit_native_store_global(emit_t *emit, qstr qst, int kind) {
MP_STATIC_ASSERT(MP_F_STORE_NAME + MP_EMIT_IDOP_GLOBAL_NAME == MP_F_STORE_NAME);
MP_STATIC_ASSERT(MP_F_STORE_NAME + MP_EMIT_IDOP_GLOBAL_GLOBAL == MP_F_STORE_GLOBAL);
if (kind == MP_EMIT_IDOP_GLOBAL_NAME) {
// mp_store_name, but needs conversion of object (maybe have mp_viper_store_name(obj, type))
vtype_kind_t vtype;
py: Viper can now store to global.
2014-08-15 23:47:59 +01:00
emit_pre_pop_reg(emit, &vtype, REG_ARG_2);
py/emit: Combine name and global into one func for load/store/delete. Reduces code size by: bare-arm: -56 minimal x86: -300 unix x64: -576 unix nanbox: -300 stm32: -164 cc3200: -56 esp8266: -236 esp32: -76
2018-05-22 12:16:30 +01:00
assert(vtype == VTYPE_PYOBJ);
py: Viper can now store to global.
2014-08-15 23:47:59 +01:00
} else {
py/emit: Combine name and global into one func for load/store/delete. Reduces code size by: bare-arm: -56 minimal x86: -300 unix x64: -576 unix nanbox: -300 stm32: -164 cc3200: -56 esp8266: -236 esp32: -76
2018-05-22 12:16:30 +01:00
vtype_kind_t vtype = peek_vtype(emit, 0);
if (vtype == VTYPE_PYOBJ) {
emit_pre_pop_reg(emit, &vtype, REG_ARG_2);
} else {
emit_pre_pop_reg(emit, &vtype, REG_ARG_1);
emit_call_with_imm_arg(emit, MP_F_CONVERT_NATIVE_TO_OBJ, vtype, REG_ARG_2); // arg2 = type
ASM_MOV_REG_REG(emit->as, REG_ARG_2, REG_RET);
}
py: Viper can now store to global.
2014-08-15 23:47:59 +01:00
}
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
emit_call_with_qstr_arg(emit, MP_F_STORE_NAME + kind, qst, REG_ARG_1); // arg1 = name
py: Viper can now store to global.
2014-08-15 23:47:59 +01:00
emit_post(emit);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
STATIC void emit_native_store_attr(emit_t *emit, qstr qst) {
Implement basic class/object in native code.
2013-10-10 11:24:39 +01:00
vtype_kind_t vtype_base, vtype_val;
emit_pre_pop_reg_reg(emit, &vtype_base, REG_ARG_1, &vtype_val, REG_ARG_3); // arg1 = base, arg3 = value
assert(vtype_base == VTYPE_PYOBJ);
assert(vtype_val == VTYPE_PYOBJ);
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
emit_call_with_qstr_arg(emit, MP_F_STORE_ATTR, qst, REG_ARG_2); // arg2 = attribute name
Implement basic class/object in native code.
2013-10-10 11:24:39 +01:00
emit_post(emit);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_native_store_subscr(emit_t *emit) {
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
DEBUG_printf("store_subscr\n");
// need to compile: base[index] = value
// pop: index, base, value
// optimise case where index is an immediate
vtype_kind_t vtype_base = peek_vtype(emit, 1);
if (vtype_base == VTYPE_PYOBJ) {
py: Implement implicit cast to obj for viper load/store index/value. This allows to do "ar[i]" and "ar[i] = val" in viper when ar is a Python object and i and/or val are native viper types (eg ints). Patch also includes tests for this feature.
2015-06-04 11:52:16 +01:00
// standard Python subscr
vtype_kind_t vtype_index = peek_vtype(emit, 0);
vtype_kind_t vtype_value = peek_vtype(emit, 2);
if (vtype_index != VTYPE_PYOBJ || vtype_value != VTYPE_PYOBJ) {
// need to implicitly convert non-objects to objects
// TODO do this properly
emit_get_stack_pointer_to_reg_for_pop(emit, REG_ARG_1, 3);
adjust_stack(emit, 3);
}
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
emit_pre_pop_reg_reg_reg(emit, &vtype_index, REG_ARG_2, &vtype_base, REG_ARG_1, &vtype_value, REG_ARG_3);
emit_call(emit, MP_F_OBJ_SUBSCR);
} else {
py: Fix viper store on x86; add tests for viper ptr16.
2014-09-29 22:41:41 +01:00
// viper store
// TODO The different machine architectures have very different
// capabilities and requirements for stores, so probably best to
// write a completely separate store-optimiser for each one.
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
stack_info_t *top = peek_stack(emit, 0);
if (top->vtype == VTYPE_INT && top->kind == STACK_IMM) {
// index is an immediate
py: Don't use anonymous unions, name them instead. This makes the code (more) compatible with the C99 standard.
2015-01-24 23:14:12 +00:00
mp_int_t index_value = top->data.u_imm;
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
emit_pre_pop_discard(emit); // discard index
vtype_kind_t vtype_value;
int reg_base = REG_ARG_1;
int reg_index = REG_ARG_2;
int reg_value = REG_ARG_3;
emit_pre_pop_reg_flexible(emit, &vtype_base, &reg_base, reg_index, reg_value);
py: Fix viper store on x86; add tests for viper ptr16.
2014-09-29 22:41:41 +01:00
#if N_X86
// special case: x86 needs byte stores to be from lower 4 regs (REG_ARG_3 is EDX)
emit_pre_pop_reg(emit, &vtype_value, reg_value);
#else
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
emit_pre_pop_reg_flexible(emit, &vtype_value, &reg_value, reg_base, reg_index);
py: Fix viper store on x86; add tests for viper ptr16.
2014-09-29 22:41:41 +01:00
#endif
py/emitnative: Add check that RHS of viper store is of integral type.
2016-02-15 09:38:02 +00:00
if (vtype_value != VTYPE_BOOL && vtype_value != VTYPE_INT && vtype_value != VTYPE_UINT) {
EMIT_NATIVE_VIPER_TYPE_ERROR(emit,
"can't store '%q'", vtype_to_qstr(vtype_value));
}
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
switch (vtype_base) {
case VTYPE_PTR8: {
// pointer to 8-bit memory
// TODO optimise to use thumb strb r1, [r2, r3]
if (index_value != 0) {
// index is non-zero
#if N_THUMB
if (index_value > 0 && index_value < 32) {
asm_thumb_strb_rlo_rlo_i5(emit->as, reg_value, reg_base, index_value);
break;
}
#endif
py/emitnative: Clean up asm macro names so they have dest as first arg. All the asm macro names that convert a particular architecture to a generic interface now follow the convention whereby the "destination" (usually a register) is specified first.
2017-11-15 00:46:49 +00:00
ASM_MOV_REG_IMM(emit->as, reg_index, index_value);
Implement missing ARM emitter functions for viper
2014-10-03 23:53:46 +01:00
#if N_ARM
asm_arm_strb_reg_reg_reg(emit->as, reg_value, reg_base, reg_index);
return;
#endif
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
ASM_ADD_REG_REG(emit->as, reg_index, reg_base); // add index to base
reg_base = reg_index;
}
ASM_STORE8_REG_REG(emit->as, reg_value, reg_base); // store value to (base+index)
break;
}
case VTYPE_PTR16: {
// pointer to 16-bit memory
if (index_value != 0) {
// index is a non-zero immediate
#if N_THUMB
if (index_value > 0 && index_value < 32) {
asm_thumb_strh_rlo_rlo_i5(emit->as, reg_value, reg_base, index_value);
break;
}
#endif
py/emitnative: Clean up asm macro names so they have dest as first arg. All the asm macro names that convert a particular architecture to a generic interface now follow the convention whereby the "destination" (usually a register) is specified first.
2017-11-15 00:46:49 +00:00
ASM_MOV_REG_IMM(emit->as, reg_index, index_value << 1);
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
ASM_ADD_REG_REG(emit->as, reg_index, reg_base); // add 2*index to base
reg_base = reg_index;
}
ASM_STORE16_REG_REG(emit->as, reg_value, reg_base); // store value to (base+2*index)
break;
}
py: Implement ptr32 load and store in viper emitter.
2015-10-13 00:50:17 +01:00
case VTYPE_PTR32: {
// pointer to 32-bit memory
if (index_value != 0) {
// index is a non-zero immediate
#if N_THUMB
if (index_value > 0 && index_value < 32) {
asm_thumb_str_rlo_rlo_i5(emit->as, reg_value, reg_base, index_value);
break;
}
#endif
#if N_ARM
py/emitnative: Fix bug with store of 16 and 32 values in viper ARM mode.
2018-08-16 16:12:05 +01:00
ASM_MOV_REG_IMM(emit->as, reg_index, index_value);
py: Implement ptr32 load and store in viper emitter.
2015-10-13 00:50:17 +01:00
asm_arm_str_reg_reg_reg(emit->as, reg_value, reg_base, reg_index);
return;
#endif
py/emitnative: Fix bug with store of 16 and 32 values in viper ARM mode.
2018-08-16 16:12:05 +01:00
ASM_MOV_REG_IMM(emit->as, reg_index, index_value << 2);
py: Implement ptr32 load and store in viper emitter.
2015-10-13 00:50:17 +01:00
ASM_ADD_REG_REG(emit->as, reg_index, reg_base); // add 4*index to base
reg_base = reg_index;
}
ASM_STORE32_REG_REG(emit->as, reg_value, reg_base); // store value to (base+4*index)
break;
}
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
default:
py: Make viper codegen raise proper exception (ViperTypeError) on error. This fixes a long standing problem that viper code generation gave terrible error messages, and actually no errors on pyboard where assertions are disabled. Now all compile-time errors are raised as proper Python exceptions, and are of type ViperTypeError. Addresses issue #940.
2015-04-20 14:29:31 +01:00
EMIT_NATIVE_VIPER_TYPE_ERROR(emit,
"can't store to '%q'", vtype_to_qstr(vtype_base));
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
}
} else {
// index is not an immediate
vtype_kind_t vtype_index, vtype_value;
int reg_index = REG_ARG_2;
int reg_value = REG_ARG_3;
emit_pre_pop_reg_flexible(emit, &vtype_index, &reg_index, REG_ARG_1, reg_value);
emit_pre_pop_reg(emit, &vtype_base, REG_ARG_1);
py/viper: Allow uint as index to load/store, and give better error msg.
2016-02-09 13:46:49 +00:00
if (vtype_index != VTYPE_INT && vtype_index != VTYPE_UINT) {
EMIT_NATIVE_VIPER_TYPE_ERROR(emit,
"can't store with '%q' index", vtype_to_qstr(vtype_index));
}
py: Fix viper store on x86; add tests for viper ptr16.
2014-09-29 22:41:41 +01:00
#if N_X86
// special case: x86 needs byte stores to be from lower 4 regs (REG_ARG_3 is EDX)
emit_pre_pop_reg(emit, &vtype_value, reg_value);
#else
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
emit_pre_pop_reg_flexible(emit, &vtype_value, &reg_value, REG_ARG_1, reg_index);
py: Fix viper store on x86; add tests for viper ptr16.
2014-09-29 22:41:41 +01:00
#endif
py/emitnative: Add check that RHS of viper store is of integral type.
2016-02-15 09:38:02 +00:00
if (vtype_value != VTYPE_BOOL && vtype_value != VTYPE_INT && vtype_value != VTYPE_UINT) {
EMIT_NATIVE_VIPER_TYPE_ERROR(emit,
"can't store '%q'", vtype_to_qstr(vtype_value));
}
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
switch (vtype_base) {
case VTYPE_PTR8: {
// pointer to 8-bit memory
// TODO optimise to use thumb strb r1, [r2, r3]
Implement missing ARM emitter functions for viper
2014-10-03 23:53:46 +01:00
#if N_ARM
asm_arm_strb_reg_reg_reg(emit->as, reg_value, REG_ARG_1, reg_index);
break;
#endif
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
ASM_ADD_REG_REG(emit->as, REG_ARG_1, reg_index); // add index to base
ASM_STORE8_REG_REG(emit->as, reg_value, REG_ARG_1); // store value to (base+index)
break;
}
case VTYPE_PTR16: {
// pointer to 16-bit memory
Implement missing ARM emitter functions for viper
2014-10-03 23:53:46 +01:00
#if N_ARM
asm_arm_strh_reg_reg_reg(emit->as, reg_value, REG_ARG_1, reg_index);
break;
#endif
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
ASM_ADD_REG_REG(emit->as, REG_ARG_1, reg_index); // add index to base
ASM_ADD_REG_REG(emit->as, REG_ARG_1, reg_index); // add index to base
ASM_STORE16_REG_REG(emit->as, reg_value, REG_ARG_1); // store value to (base+2*index)
break;
}
py: Implement ptr32 load and store in viper emitter.
2015-10-13 00:50:17 +01:00
case VTYPE_PTR32: {
// pointer to 32-bit memory
#if N_ARM
asm_arm_str_reg_reg_reg(emit->as, reg_value, REG_ARG_1, reg_index);
break;
#endif
ASM_ADD_REG_REG(emit->as, REG_ARG_1, reg_index); // add index to base
ASM_ADD_REG_REG(emit->as, REG_ARG_1, reg_index); // add index to base
ASM_ADD_REG_REG(emit->as, REG_ARG_1, reg_index); // add index to base
ASM_ADD_REG_REG(emit->as, REG_ARG_1, reg_index); // add index to base
ASM_STORE32_REG_REG(emit->as, reg_value, REG_ARG_1); // store value to (base+4*index)
break;
}
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
default:
py: Make viper codegen raise proper exception (ViperTypeError) on error. This fixes a long standing problem that viper code generation gave terrible error messages, and actually no errors on pyboard where assertions are disabled. Now all compile-time errors are raised as proper Python exceptions, and are of type ViperTypeError. Addresses issue #940.
2015-04-20 14:29:31 +01:00
EMIT_NATIVE_VIPER_TYPE_ERROR(emit,
"can't store to '%q'", vtype_to_qstr(vtype_base));
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
}
}
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py/emit: Combine fast and deref into one function for load/store/delete. Reduces code size by: bare-arm: -16 minimal x86: -208 unix x64: -408 unix nanbox: -248 stm32: -12 cc3200: -24 esp8266: -96 esp32: -44
2018-05-18 15:11:04 +01:00
STATIC void emit_native_delete_local(emit_t *emit, qstr qst, mp_uint_t local_num, int kind) {
if (kind == MP_EMIT_IDOP_LOCAL_FAST) {
// TODO: This is not compliant implementation. We could use MP_OBJ_SENTINEL
// to mark deleted vars but then every var would need to be checked on
// each access. Very inefficient, so just set value to None to enable GC.
emit_native_load_const_tok(emit, MP_TOKEN_KW_NONE);
emit_native_store_fast(emit, qst, local_num);
} else {
// TODO implement me!
}
py: work towards working closures.
2013-12-11 00:41:43 +00:00
}
py/emit: Combine name and global into one func for load/store/delete. Reduces code size by: bare-arm: -56 minimal x86: -300 unix x64: -576 unix nanbox: -300 stm32: -164 cc3200: -56 esp8266: -236 esp32: -76
2018-05-22 12:16:30 +01:00
STATIC void emit_native_delete_global(emit_t *emit, qstr qst, int kind) {
MP_STATIC_ASSERT(MP_F_DELETE_NAME + MP_EMIT_IDOP_GLOBAL_NAME == MP_F_DELETE_NAME);
MP_STATIC_ASSERT(MP_F_DELETE_NAME + MP_EMIT_IDOP_GLOBAL_GLOBAL == MP_F_DELETE_GLOBAL);
py: Native emitter now supports delete name & global, and end finally.
2014-09-06 18:38:20 +01:00
emit_native_pre(emit);
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
emit_call_with_qstr_arg(emit, MP_F_DELETE_NAME + kind, qst, REG_ARG_1);
py: Native emitter now supports delete name & global, and end finally.
2014-09-06 18:38:20 +01:00
emit_post(emit);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Native emitter now supports delete name & global, and end finally.
2014-09-06 18:38:20 +01:00
STATIC void emit_native_delete_attr(emit_t *emit, qstr qst) {
py: Implement delete_attr in native emitter.
2014-06-22 18:35:04 +01:00
vtype_kind_t vtype_base;
emit_pre_pop_reg(emit, &vtype_base, REG_ARG_1); // arg1 = base
assert(vtype_base == VTYPE_PYOBJ);
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
ASM_XOR_REG_REG(emit->as, REG_ARG_3, REG_ARG_3); // arg3 = value (null for delete)
emit_call_with_qstr_arg(emit, MP_F_STORE_ATTR, qst, REG_ARG_2); // arg2 = attribute name
py: Implement delete_attr in native emitter.
2014-06-22 18:35:04 +01:00
emit_post(emit);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_native_delete_subscr(emit_t *emit) {
py: Merge BINARY_OP_SUBSCR and store_subscr (w/ delete) into subscr. mp_obj_t->subscr now does load/store/delete.
2014-04-17 22:10:53 +01:00
vtype_kind_t vtype_index, vtype_base;
emit_pre_pop_reg_reg(emit, &vtype_index, REG_ARG_2, &vtype_base, REG_ARG_1); // index, base
assert(vtype_index == VTYPE_PYOBJ);
assert(vtype_base == VTYPE_PYOBJ);
py: Code clean-up in native emitter; improve thumb native calls.
2014-08-16 22:31:57 +01:00
emit_call_with_imm_arg(emit, MP_F_OBJ_SUBSCR, (mp_uint_t)MP_OBJ_NULL, REG_ARG_3);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py/emit: Combine load/store/delete subscr into one emit function. Reduces code size by: bare-arm: -8 minimal x86: -104 unix x64: -312 unix nanbox: -120 stm32: -60 cc3200: -16 esp8266: -92 esp32: -24
2018-05-22 12:31:56 +01:00
STATIC void emit_native_subscr(emit_t *emit, int kind) {
if (kind == MP_EMIT_SUBSCR_LOAD) {
emit_native_load_subscr(emit);
} else if (kind == MP_EMIT_SUBSCR_STORE) {
emit_native_store_subscr(emit);
} else {
emit_native_delete_subscr(emit);
}
}
py/emit: Combine load/store/delete attr into one emit function. Reduces code size by: bare-arm: -20 minimal x86: -140 unix x64: -408 unix nanbox: -140 stm32: -68 cc3200: -16 esp8266: -80 esp32: -32
2018-05-22 12:43:41 +01:00
STATIC void emit_native_attr(emit_t *emit, qstr qst, int kind) {
if (kind == MP_EMIT_ATTR_LOAD) {
emit_native_load_attr(emit, qst);
} else if (kind == MP_EMIT_ATTR_STORE) {
emit_native_store_attr(emit, qst);
} else {
emit_native_delete_attr(emit, qst);
}
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_native_dup_top(emit_t *emit) {
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
DEBUG_printf("dup_top\n");
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
vtype_kind_t vtype;
py: Partially fix viper multi-comparison; add test for it.
2014-10-19 19:00:51 +01:00
int reg = REG_TEMP0;
emit_pre_pop_reg_flexible(emit, &vtype, &reg, -1, -1);
emit_post_push_reg_reg(emit, vtype, reg, vtype, reg);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_native_dup_top_two(emit_t *emit) {
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
vtype_kind_t vtype0, vtype1;
emit_pre_pop_reg_reg(emit, &vtype0, REG_TEMP0, &vtype1, REG_TEMP1);
emit_post_push_reg_reg_reg_reg(emit, vtype1, REG_TEMP1, vtype0, REG_TEMP0, vtype1, REG_TEMP1, vtype0, REG_TEMP0);
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_native_pop_top(emit_t *emit) {
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
DEBUG_printf("pop_top\n");
py: Native emitter now supports delete name & global, and end finally.
2014-09-06 18:38:20 +01:00
emit_pre_pop_discard(emit);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit_post(emit);
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_native_rot_two(emit_t *emit) {
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
DEBUG_printf("rot_two\n");
Improve push/pop and register usage for native emit.
2013-10-08 22:18:32 +01:00
vtype_kind_t vtype0, vtype1;
emit_pre_pop_reg_reg(emit, &vtype0, REG_TEMP0, &vtype1, REG_TEMP1);
emit_post_push_reg_reg(emit, vtype0, REG_TEMP0, vtype1, REG_TEMP1);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_native_rot_three(emit_t *emit) {
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
DEBUG_printf("rot_three\n");
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
vtype_kind_t vtype0, vtype1, vtype2;
emit_pre_pop_reg_reg_reg(emit, &vtype0, REG_TEMP0, &vtype1, REG_TEMP1, &vtype2, REG_TEMP2);
emit_post_push_reg_reg_reg(emit, vtype0, REG_TEMP0, vtype2, REG_TEMP2, vtype1, REG_TEMP1);
}
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
STATIC void emit_native_jump(emit_t *emit, mp_uint_t label) {
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
DEBUG_printf("jump(label=" UINT_FMT ")\n", label);
py: Rename emit_pre so they have globally unique names.
2014-03-27 23:30:26 +00:00
emit_native_pre(emit);
py: Improve native emitter; now supports more opcodes.
2014-05-07 18:30:52 +01:00
// need to commit stack because we are jumping elsewhere
need_stack_settled(emit);
py: Add support for emitting native x86 machine code.
2014-09-06 23:06:36 +01:00
ASM_JUMP(emit->as, label);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit_post(emit);
}
py/emitnative: Factor common code for native jump helper.
2018-08-04 12:45:24 +01:00
STATIC void emit_native_jump_helper(emit_t *emit, bool cond, mp_uint_t label, bool pop) {
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
vtype_kind_t vtype = peek_vtype(emit, 0);
py: Make viper codegen raise proper exception (ViperTypeError) on error. This fixes a long standing problem that viper code generation gave terrible error messages, and actually no errors on pyboard where assertions are disabled. Now all compile-time errors are raised as proper Python exceptions, and are of type ViperTypeError. Addresses issue #940.
2015-04-20 14:29:31 +01:00
if (vtype == VTYPE_PYOBJ) {
emit_pre_pop_reg(emit, &vtype, REG_ARG_1);
if (!pop) {
adjust_stack(emit, 1);
}
emit_call(emit, MP_F_OBJ_IS_TRUE);
} else {
emit_pre_pop_reg(emit, &vtype, REG_RET);
if (!pop) {
adjust_stack(emit, 1);
}
if (!(vtype == VTYPE_BOOL || vtype == VTYPE_INT || vtype == VTYPE_UINT)) {
EMIT_NATIVE_VIPER_TYPE_ERROR(emit,
"can't implicitly convert '%q' to 'bool'", vtype_to_qstr(vtype));
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Partially fix viper multi-comparison; add test for it.
2014-10-19 19:00:51 +01:00
// For non-pop need to save the vtype so that emit_native_adjust_stack_size
// can use it. This is a bit of a hack.
if (!pop) {
emit->saved_stack_vtype = vtype;
}
py: Improve native emitter; now supports more opcodes.
2014-05-07 18:30:52 +01:00
// need to commit stack because we may jump elsewhere
need_stack_settled(emit);
py/emitnative: Factor common code for native jump helper.
2018-08-04 12:45:24 +01:00
// Emit the jump
py: Combine emit functions for jump true/false to reduce code size. Saves 116 bytes for stmhal and 56 bytes for cc3200 port.
2015-02-28 15:04:06 +00:00
if (cond) {
py/emitnative: Fix x86 native zero checks by comparing full word. On x86 archs (both 32 and 64 bit) a bool return value only sets the 8-bit al register, and the higher bits of the ax register have an undefined value. When testing the return value of such cases it is required to just test al for zero/non-zero. On the other hand, checking for truth or zero/non-zero on an integer return value requires checking all bits of the register. These two cases must be distinguished and handled correctly in generated native code. This patch makes sure of this. For other supported native archs (ARM, Thumb2, Xtensa) there is no such distinction and this patch does not change anything for them.
2018-08-04 13:03:49 +01:00
ASM_JUMP_IF_REG_NONZERO(emit->as, REG_RET, label, vtype == VTYPE_PYOBJ);
py: Combine emit functions for jump true/false to reduce code size. Saves 116 bytes for stmhal and 56 bytes for cc3200 port.
2015-02-28 15:04:06 +00:00
} else {
py/emitnative: Fix x86 native zero checks by comparing full word. On x86 archs (both 32 and 64 bit) a bool return value only sets the 8-bit al register, and the higher bits of the ax register have an undefined value. When testing the return value of such cases it is required to just test al for zero/non-zero. On the other hand, checking for truth or zero/non-zero on an integer return value requires checking all bits of the register. These two cases must be distinguished and handled correctly in generated native code. This patch makes sure of this. For other supported native archs (ARM, Thumb2, Xtensa) there is no such distinction and this patch does not change anything for them.
2018-08-04 13:03:49 +01:00
ASM_JUMP_IF_REG_ZERO(emit->as, REG_RET, label, vtype == VTYPE_PYOBJ);
py: Combine emit functions for jump true/false to reduce code size. Saves 116 bytes for stmhal and 56 bytes for cc3200 port.
2015-02-28 15:04:06 +00:00
}
py/emitnative: Factor common code for native jump helper.
2018-08-04 12:45:24 +01:00
if (!pop) {
adjust_stack(emit, -1);
}
Implement more thumb branch instructions.
2013-11-03 13:58:19 +00:00
emit_post(emit);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Implement more thumb branch instructions.
2013-11-03 13:58:19 +00:00
py/emitnative: Factor common code for native jump helper.
2018-08-04 12:45:24 +01:00
STATIC void emit_native_pop_jump_if(emit_t *emit, bool cond, mp_uint_t label) {
DEBUG_printf("pop_jump_if(cond=%u, label=" UINT_FMT ")\n", cond, label);
emit_native_jump_helper(emit, cond, label, true);
}
py: Combine emit functions for jump true/false to reduce code size. Saves 116 bytes for stmhal and 56 bytes for cc3200 port.
2015-02-28 15:04:06 +00:00
STATIC void emit_native_jump_if_or_pop(emit_t *emit, bool cond, mp_uint_t label) {
DEBUG_printf("jump_if_or_pop(cond=%u, label=" UINT_FMT ")\n", cond, label);
py/emitnative: Factor common code for native jump helper.
2018-08-04 12:45:24 +01:00
emit_native_jump_helper(emit, cond, label, false);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py/emit: Combine break_loop and continue_loop into one emit function. Reduces code size by: bare-arm: +0 minimal x86: +0 unix x64: -80 unix nanbox: +0 stm32: -12 cc3200: +0 esp8266: -28 esp32: +0
2018-05-22 12:50:22 +01:00
STATIC void emit_native_unwind_jump(emit_t *emit, mp_uint_t label, mp_uint_t except_depth) {
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
if (except_depth > 0) {
exc_stack_entry_t *first_finally = NULL;
exc_stack_entry_t *prev_finally = NULL;
exc_stack_entry_t *e = &emit->exc_stack[emit->exc_stack_size - 1];
for (; except_depth > 0; --except_depth, --e) {
if (e->is_finally && e->is_active) {
// Found an active finally handler
if (first_finally == NULL) {
first_finally = e;
}
if (prev_finally != NULL) {
// Mark prev finally as needed to unwind a jump
prev_finally->unwind_label = e->label;
}
prev_finally = e;
}
}
if (prev_finally == NULL) {
// No finally, handle the jump ourselves
// First, restore the exception handler address for the jump
if (e < emit->exc_stack) {
ASM_XOR_REG_REG(emit->as, REG_RET, REG_RET);
} else {
ASM_MOV_REG_PCREL(emit->as, REG_RET, e->label);
}
ASM_MOV_LOCAL_REG(emit->as, LOCAL_IDX_EXC_HANDLER_PC(emit), REG_RET);
} else {
// Last finally should do our jump for us
// Mark finally as needing to decide the type of jump
prev_finally->unwind_label = UNWIND_LABEL_DO_FINAL_UNWIND;
ASM_MOV_REG_PCREL(emit->as, REG_RET, label & ~MP_EMIT_BREAK_FROM_FOR);
ASM_MOV_LOCAL_REG(emit->as, LOCAL_IDX_EXC_HANDLER_UNWIND(emit), REG_RET);
py: Replace POP_BLOCK and POP_EXCEPT opcodes with POP_EXCEPT_JUMP. POP_BLOCK and POP_EXCEPT are now the same, and are always followed by a JUMP. So this optimisation reduces code size, and RAM usage of bytecode by two bytes for each try-except handler.
2019-02-15 01:18:59 +00:00
// Cancel any active exception (see also emit_native_pop_except_jump)
py/emitnative: Put None/False/True in global native const table. So these constant objects can be loaded by dereferencing the REG_FUN_TABLE pointer instead of loading immediate values. This reduces the size of generated native code (when such constants are used), and means that pointers to these constants are no longer stored in the assembly code.
2018-10-14 13:56:46 +01:00
emit_native_mov_reg_const(emit, REG_RET, MP_F_CONST_NONE_OBJ);
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
ASM_MOV_LOCAL_REG(emit->as, LOCAL_IDX_EXC_VAL(emit), REG_RET);
// Jump to the innermost active finally
label = first_finally->label;
}
}
emit_native_jump(emit, label & ~MP_EMIT_BREAK_FROM_FOR);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Improve native emitter; now supports more opcodes.
2014-05-07 18:30:52 +01:00
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
STATIC void emit_native_setup_with(emit_t *emit, mp_uint_t label) {
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
// the context manager is on the top of the stack
// stack: (..., ctx_mgr)
// get __exit__ method
vtype_kind_t vtype;
emit_access_stack(emit, 1, &vtype, REG_ARG_1); // arg1 = ctx_mgr
assert(vtype == VTYPE_PYOBJ);
emit_get_stack_pointer_to_reg_for_push(emit, REG_ARG_3, 2); // arg3 = dest ptr
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
emit_call_with_qstr_arg(emit, MP_F_LOAD_METHOD, MP_QSTR___exit__, REG_ARG_2);
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
// stack: (..., ctx_mgr, __exit__, self)
emit_pre_pop_reg(emit, &vtype, REG_ARG_3); // self
emit_pre_pop_reg(emit, &vtype, REG_ARG_2); // __exit__
emit_pre_pop_reg(emit, &vtype, REG_ARG_1); // ctx_mgr
emit_post_push_reg(emit, vtype, REG_ARG_2); // __exit__
emit_post_push_reg(emit, vtype, REG_ARG_3); // self
// stack: (..., __exit__, self)
// REG_ARG_1=ctx_mgr
// get __enter__ method
emit_get_stack_pointer_to_reg_for_push(emit, REG_ARG_3, 2); // arg3 = dest ptr
py/emitnative: Provide concentrated points of qstr emit.
2019-02-03 01:25:50 +00:00
emit_call_with_qstr_arg(emit, MP_F_LOAD_METHOD, MP_QSTR___enter__, REG_ARG_2); // arg2 = method name
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
// stack: (..., __exit__, self, __enter__, self)
// call __enter__ method
emit_get_stack_pointer_to_reg_for_pop(emit, REG_ARG_3, 2); // pointer to items, including meth and self
emit_call_with_2_imm_args(emit, MP_F_CALL_METHOD_N_KW, 0, REG_ARG_1, 0, REG_ARG_2);
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET); // push return value of __enter__
// stack: (..., __exit__, self, as_value)
// need to commit stack because we may jump elsewhere
need_stack_settled(emit);
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
emit_native_push_exc_stack(emit, label, true);
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
emit_native_dup_top(emit);
// stack: (..., __exit__, self, as_value, as_value)
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Improvements to native emitter. Native emitter can now compile try/except blocks using nlr_push/nlr_pop. It probably only works for 1 level of exception handling. It doesn't work on Thumb (only x64). Native emitter can also handle some additional op codes. With this patch, 198 tests now pass using "-X emit=native" option to micropython.
2014-06-30 05:17:25 +01:00
py/emit: Combine setup with/except/finally into one emit function. This patch reduces code size by: bare-arm: -16 minimal x86: -156 unix x64: -288 unix nanbox: -184 stm32: -48 cc3200: -16 esp8266: -96 esp32: -16 The last 10 patches combined reduce code size by: bare-arm: -164 minimal x86: -1260 unix x64: -3416 unix nanbox: -1616 stm32: -676 cc3200: -232 esp8266: -1144 esp32: -268
2018-05-22 13:33:26 +01:00
STATIC void emit_native_setup_block(emit_t *emit, mp_uint_t label, int kind) {
if (kind == MP_EMIT_SETUP_BLOCK_WITH) {
emit_native_setup_with(emit, label);
} else {
// Set up except and finally
emit_native_pre(emit);
need_stack_settled(emit);
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
emit_native_push_exc_stack(emit, label, kind == MP_EMIT_SETUP_BLOCK_FINALLY);
py/emit: Combine setup with/except/finally into one emit function. This patch reduces code size by: bare-arm: -16 minimal x86: -156 unix x64: -288 unix nanbox: -184 stm32: -48 cc3200: -16 esp8266: -96 esp32: -16 The last 10 patches combined reduce code size by: bare-arm: -164 minimal x86: -1260 unix x64: -3416 unix nanbox: -1616 stm32: -676 cc3200: -232 esp8266: -1144 esp32: -268
2018-05-22 13:33:26 +01:00
emit_post(emit);
}
}
py: Combine continuous block of emit steps into with_cleanup emit call. Because different emitters need to handle with-cleanup in different ways.
2016-04-07 08:50:38 +01:00
STATIC void emit_native_with_cleanup(emit_t *emit, mp_uint_t label) {
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
// Note: 3 labels are reserved for this function, starting at *emit->label_slot
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
// stack: (..., __exit__, self, as_value)
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
emit_native_pre(emit);
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
emit_native_leave_exc_stack(emit, false);
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
adjust_stack(emit, -1);
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
// stack: (..., __exit__, self)
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
// Label for case where __exit__ is called from an unwind jump
emit_native_label_assign(emit, *emit->label_slot + 2);
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
// call __exit__
py/emitnative: Push internal None rather than const obj where possible. This shifts the work of loading the constant None object on to load_reg_stack_imm(), making the handling of None more centralised.
2018-10-14 13:23:25 +01:00
emit_post_push_imm(emit, VTYPE_PTR_NONE, 0);
emit_post_push_imm(emit, VTYPE_PTR_NONE, 0);
emit_post_push_imm(emit, VTYPE_PTR_NONE, 0);
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
emit_get_stack_pointer_to_reg_for_pop(emit, REG_ARG_3, 5);
emit_call_with_2_imm_args(emit, MP_F_CALL_METHOD_N_KW, 3, REG_ARG_1, 0, REG_ARG_2);
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
// Replace exc with None and finish
emit_native_jump(emit, *emit->label_slot);
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
// nlr_catch
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
// Don't use emit_native_label_assign because this isn't a real finally label
mp_asm_base_label_assign(&emit->as->base, label);
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
// Leave with's exception handler
emit_native_leave_exc_stack(emit, true);
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
// Adjust stack counter for: __exit__, self (implicitly discard as_value which is above self)
emit_native_adjust_stack_size(emit, 2);
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
// stack: (..., __exit__, self)
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
ASM_MOV_REG_LOCAL(emit->as, REG_ARG_1, LOCAL_IDX_EXC_VAL(emit)); // get exc
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
// Check if exc is None and jump to non-exc handler if it is
py/emitnative: Put None/False/True in global native const table. So these constant objects can be loaded by dereferencing the REG_FUN_TABLE pointer instead of loading immediate values. This reduces the size of generated native code (when such constants are used), and means that pointers to these constants are no longer stored in the assembly code.
2018-10-14 13:56:46 +01:00
emit_native_mov_reg_const(emit, REG_ARG_2, MP_F_CONST_NONE_OBJ);
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
ASM_JUMP_IF_REG_EQ(emit->as, REG_ARG_1, REG_ARG_2, *emit->label_slot + 2);
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
ASM_LOAD_REG_REG_OFFSET(emit->as, REG_ARG_2, REG_ARG_1, 0); // get type(exc)
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_ARG_2); // push type(exc)
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_ARG_1); // push exc value
py/emitnative: Push internal None rather than const obj where possible. This shifts the work of loading the constant None object on to load_reg_stack_imm(), making the handling of None more centralised.
2018-10-14 13:23:25 +01:00
emit_post_push_imm(emit, VTYPE_PTR_NONE, 0); // traceback info
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
// Stack: (..., __exit__, self, type(exc), exc, traceback)
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
// call __exit__ method
emit_get_stack_pointer_to_reg_for_pop(emit, REG_ARG_3, 5);
emit_call_with_2_imm_args(emit, MP_F_CALL_METHOD_N_KW, 3, REG_ARG_1, 0, REG_ARG_2);
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
// Stack: (...)
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
// If REG_RET is true then we need to replace exception with None (swallow exception)
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
if (REG_ARG_1 != REG_RET) {
ASM_MOV_REG_REG(emit->as, REG_ARG_1, REG_RET);
}
emit_call(emit, MP_F_OBJ_IS_TRUE);
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
ASM_JUMP_IF_REG_ZERO(emit->as, REG_RET, *emit->label_slot + 1, true);
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
// Replace exception with None
emit_native_label_assign(emit, *emit->label_slot);
py/emitnative: Put None/False/True in global native const table. So these constant objects can be loaded by dereferencing the REG_FUN_TABLE pointer instead of loading immediate values. This reduces the size of generated native code (when such constants are used), and means that pointers to these constants are no longer stored in the assembly code.
2018-10-14 13:56:46 +01:00
emit_native_mov_reg_const(emit, REG_TEMP0, MP_F_CONST_NONE_OBJ);
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
ASM_MOV_LOCAL_REG(emit->as, LOCAL_IDX_EXC_VAL(emit), REG_TEMP0);
py: Implement basic with support in native emitter.
2016-04-07 08:53:24 +01:00
// end of with cleanup nlr_catch block
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
emit_native_label_assign(emit, *emit->label_slot + 1);
// Exception is in nlr_buf.ret_val slot
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Improvements to native emitter. Native emitter can now compile try/except blocks using nlr_push/nlr_pop. It probably only works for 1 level of exception handling. It doesn't work on Thumb (only x64). Native emitter can also handle some additional op codes. With this patch, 198 tests now pass using "-X emit=native" option to micropython.
2014-06-30 05:17:25 +01:00
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_native_end_finally(emit_t *emit) {
py: Implement proper re-raising in native codegen's finally handler. This allows an exception to propagate correctly through a finally handler.
2015-01-21 17:00:01 +00:00
// logic:
// exc = pop_stack
// if exc == None: pass
// else: raise exc
// the check if exc is None is done in the MP_F_NATIVE_RAISE stub
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
emit_native_pre(emit);
ASM_MOV_REG_LOCAL(emit->as, REG_ARG_1, LOCAL_IDX_EXC_VAL(emit));
py: Implement proper re-raising in native codegen's finally handler. This allows an exception to propagate correctly through a finally handler.
2015-01-21 17:00:01 +00:00
emit_call(emit, MP_F_NATIVE_RAISE);
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
// Get state for this finally and see if we need to unwind
exc_stack_entry_t *e = emit_native_pop_exc_stack(emit);
if (e->unwind_label != UNWIND_LABEL_UNUSED) {
ASM_MOV_REG_LOCAL(emit->as, REG_RET, LOCAL_IDX_EXC_HANDLER_UNWIND(emit));
ASM_JUMP_IF_REG_ZERO(emit->as, REG_RET, *emit->label_slot, false);
if (e->unwind_label == UNWIND_LABEL_DO_FINAL_UNWIND) {
ASM_JUMP_REG(emit->as, REG_RET);
} else {
emit_native_jump(emit, e->unwind_label);
}
emit_native_label_assign(emit, *emit->label_slot);
}
py: Native emitter now supports delete name & global, and end finally.
2014-09-06 18:38:20 +01:00
emit_post(emit);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
py: Allow bytecode/native to put iter_buf on stack for simple for loops. So that the "for x in it: ..." statement can now work without using the heap (so long as the iterator argument fits in an iter_buf structure).
2016-01-09 23:59:52 +00:00
STATIC void emit_native_get_iter(emit_t *emit, bool use_stack) {
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
// perhaps the difficult one, as we want to rewrite for loops using native code
// in cases where we iterate over a Python object, can we use normal runtime calls?
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
vtype_kind_t vtype;
emit_pre_pop_reg(emit, &vtype, REG_ARG_1);
assert(vtype == VTYPE_PYOBJ);
py: Allow bytecode/native to put iter_buf on stack for simple for loops. So that the "for x in it: ..." statement can now work without using the heap (so long as the iterator argument fits in an iter_buf structure).
2016-01-09 23:59:52 +00:00
if (use_stack) {
py: Define and use MP_OBJ_ITER_BUF_NSLOTS to get size of stack iter buf. It improves readability of code and reduces the chance to make a mistake. This patch also fixes a bug with nan-boxing builds by rounding up the calculation of the new NSLOTS variable, giving the correct number of slots (being 4) even if mp_obj_t is larger than the native machine size.
2017-03-23 05:36:08 +00:00
emit_get_stack_pointer_to_reg_for_push(emit, REG_ARG_2, MP_OBJ_ITER_BUF_NSLOTS);
py: Optimise storage of iterator so it takes only 4 slots on Py stack.
2017-01-17 04:27:37 +00:00
emit_call(emit, MP_F_NATIVE_GETITER);
py: Allow bytecode/native to put iter_buf on stack for simple for loops. So that the "for x in it: ..." statement can now work without using the heap (so long as the iterator argument fits in an iter_buf structure).
2016-01-09 23:59:52 +00:00
} else {
// mp_getiter will allocate the iter_buf on the heap
py/emitnative: Clean up asm macro names so they have dest as first arg. All the asm macro names that convert a particular architecture to a generic interface now follow the convention whereby the "destination" (usually a register) is specified first.
2017-11-15 00:46:49 +00:00
ASM_MOV_REG_IMM(emit->as, REG_ARG_2, 0);
py: Optimise storage of iterator so it takes only 4 slots on Py stack.
2017-01-17 04:27:37 +00:00
emit_call(emit, MP_F_NATIVE_GETITER);
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
py: Allow bytecode/native to put iter_buf on stack for simple for loops. So that the "for x in it: ..." statement can now work without using the heap (so long as the iterator argument fits in an iter_buf structure).
2016-01-09 23:59:52 +00:00
}
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
}
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
STATIC void emit_native_for_iter(emit_t *emit, mp_uint_t label) {
py: Rename emit_pre so they have globally unique names.
2014-03-27 23:30:26 +00:00
emit_native_pre(emit);
py: Define and use MP_OBJ_ITER_BUF_NSLOTS to get size of stack iter buf. It improves readability of code and reduces the chance to make a mistake. This patch also fixes a bug with nan-boxing builds by rounding up the calculation of the new NSLOTS variable, giving the correct number of slots (being 4) even if mp_obj_t is larger than the native machine size.
2017-03-23 05:36:08 +00:00
emit_get_stack_pointer_to_reg_for_pop(emit, REG_ARG_1, MP_OBJ_ITER_BUF_NSLOTS);
adjust_stack(emit, MP_OBJ_ITER_BUF_NSLOTS);
py: Optimise storage of iterator so it takes only 4 slots on Py stack.
2017-01-17 04:27:37 +00:00
emit_call(emit, MP_F_NATIVE_ITERNEXT);
py: Add independent config for debugging sentinel object values. The new compile-time option is MICROPY_DEBUG_MP_OBJ_SENTINELS, disabled by default. This is to allow finer control of whether this debugging feature is enabled or not (because, for example, this setting must be the same for mpy-cross and the MicroPython main code when using native code generation).
2019-03-08 04:48:20 +00:00
#if MICROPY_DEBUG_MP_OBJ_SENTINELS
py/emitnative: Clean up asm macro names so they have dest as first arg. All the asm macro names that convert a particular architecture to a generic interface now follow the convention whereby the "destination" (usually a register) is specified first.
2017-11-15 00:46:49 +00:00
ASM_MOV_REG_IMM(emit->as, REG_TEMP1, (mp_uint_t)MP_OBJ_STOP_ITERATION);
py: Add support for emitting native x86 machine code.
2014-09-06 23:06:36 +01:00
ASM_JUMP_IF_REG_EQ(emit->as, REG_RET, REG_TEMP1, label);
py: Add independent config for debugging sentinel object values. The new compile-time option is MICROPY_DEBUG_MP_OBJ_SENTINELS, disabled by default. This is to allow finer control of whether this debugging feature is enabled or not (because, for example, this setting must be the same for mpy-cross and the MicroPython main code when using native code generation).
2019-03-08 04:48:20 +00:00
#else
MP_STATIC_ASSERT(MP_OBJ_STOP_ITERATION == 0);
ASM_JUMP_IF_REG_ZERO(emit->as, REG_RET, label, false);
py/emitnative: Optimise for iteration asm code for non-debug build. In non-debug mode MP_OBJ_STOP_ITERATION is zero and comparing something to zero can be done more efficiently in assembler than comparing to a non-zero value.
2018-07-12 09:08:01 +01:00
#endif
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
py: Remove unused "use_stack" argument from for_iter_end emit function.
2017-01-17 04:30:18 +00:00
STATIC void emit_native_for_iter_end(emit_t *emit) {
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
// adjust stack counter (we get here from for_iter ending, which popped the value for us)
py: Rename emit_pre so they have globally unique names.
2014-03-27 23:30:26 +00:00
emit_native_pre(emit);
py: Define and use MP_OBJ_ITER_BUF_NSLOTS to get size of stack iter buf. It improves readability of code and reduces the chance to make a mistake. This patch also fixes a bug with nan-boxing builds by rounding up the calculation of the new NSLOTS variable, giving the correct number of slots (being 4) even if mp_obj_t is larger than the native machine size.
2017-03-23 05:36:08 +00:00
adjust_stack(emit, -MP_OBJ_ITER_BUF_NSLOTS);
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
emit_post(emit);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Replace POP_BLOCK and POP_EXCEPT opcodes with POP_EXCEPT_JUMP. POP_BLOCK and POP_EXCEPT are now the same, and are always followed by a JUMP. So this optimisation reduces code size, and RAM usage of bytecode by two bytes for each try-except handler.
2019-02-15 01:18:59 +00:00
STATIC void emit_native_pop_except_jump(emit_t *emit, mp_uint_t label, bool within_exc_handler) {
if (within_exc_handler) {
// Cancel any active exception so subsequent handlers don't see it
emit_native_mov_reg_const(emit, REG_TEMP0, MP_F_CONST_NONE_OBJ);
ASM_MOV_LOCAL_REG(emit->as, LOCAL_IDX_EXC_VAL(emit), REG_TEMP0);
} else {
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
emit_native_leave_exc_stack(emit, false);
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
}
py: Replace POP_BLOCK and POP_EXCEPT opcodes with POP_EXCEPT_JUMP. POP_BLOCK and POP_EXCEPT are now the same, and are always followed by a JUMP. So this optimisation reduces code size, and RAM usage of bytecode by two bytes for each try-except handler.
2019-02-15 01:18:59 +00:00
emit_native_jump(emit, label);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Rename rt_* to mp_*. Mostly just a global search and replace. Except rt_is_true which becomes mp_obj_is_true. Still would like to tidy up some of the names, but this will do for now.
2014-03-30 13:35:08 +01:00
STATIC void emit_native_unary_op(emit_t *emit, mp_unary_op_t op) {
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
vtype_kind_t vtype;
emit_pre_pop_reg(emit, &vtype, REG_ARG_2);
py/emitnative: Raise ViperTypeError for unsupported unary ops.
2015-10-08 13:08:59 +01:00
if (vtype == VTYPE_PYOBJ) {
emit_call_with_imm_arg(emit, MP_F_UNARY_OP, op, REG_ARG_1);
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
} else {
adjust_stack(emit, 1);
EMIT_NATIVE_VIPER_TYPE_ERROR(emit,
"unary op %q not implemented", mp_unary_op_method_name[op]);
py: Improvements to native emitter. Native emitter can now compile try/except blocks using nlr_push/nlr_pop. It probably only works for 1 level of exception handling. It doesn't work on Thumb (only x64). Native emitter can also handle some additional op codes. With this patch, 198 tests now pass using "-X emit=native" option to micropython.
2014-06-30 05:17:25 +01:00
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Rename rt_* to mp_*. Mostly just a global search and replace. Except rt_is_true which becomes mp_obj_is_true. Still would like to tidy up some of the names, but this will do for now.
2014-03-30 13:35:08 +01:00
STATIC void emit_native_binary_op(emit_t *emit, mp_binary_op_t op) {
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
DEBUG_printf("binary_op(" UINT_FMT ")\n", op);
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
vtype_kind_t vtype_lhs = peek_vtype(emit, 1);
vtype_kind_t vtype_rhs = peek_vtype(emit, 0);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
if (vtype_lhs == VTYPE_INT && vtype_rhs == VTYPE_INT) {
py/emitnative: Simplify binary op emitter, no need to check inplace ops.
2017-10-12 02:26:49 +01:00
// for integers, inplace and normal ops are equivalent, so use just normal ops
if (MP_BINARY_OP_INPLACE_OR <= op && op <= MP_BINARY_OP_INPLACE_POWER) {
op += MP_BINARY_OP_OR - MP_BINARY_OP_INPLACE_OR;
}
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
#if N_X64 || N_X86
// special cases for x86 and shifting
py/emitnative: Simplify binary op emitter, no need to check inplace ops.
2017-10-12 02:26:49 +01:00
if (op == MP_BINARY_OP_LSHIFT || op == MP_BINARY_OP_RSHIFT) {
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
#if N_X64
emit_pre_pop_reg_reg(emit, &vtype_rhs, ASM_X64_REG_RCX, &vtype_lhs, REG_RET);
#else
emit_pre_pop_reg_reg(emit, &vtype_rhs, ASM_X86_REG_ECX, &vtype_lhs, REG_RET);
#endif
py/emitnative: Simplify binary op emitter, no need to check inplace ops.
2017-10-12 02:26:49 +01:00
if (op == MP_BINARY_OP_LSHIFT) {
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
ASM_LSL_REG(emit->as, REG_RET);
} else {
ASM_ASR_REG(emit->as, REG_RET);
}
emit_post_push_reg(emit, VTYPE_INT, REG_RET);
return;
}
#endif
py/emitnative: Implement floor-division and modulo for viper emitter.
2017-10-11 08:54:34 +01:00
// special cases for floor-divide and module because we dispatch to helper functions
py/emitnative: Simplify binary op emitter, no need to check inplace ops.
2017-10-12 02:26:49 +01:00
if (op == MP_BINARY_OP_FLOOR_DIVIDE || op == MP_BINARY_OP_MODULO) {
py/emitnative: Implement floor-division and modulo for viper emitter.
2017-10-11 08:54:34 +01:00
emit_pre_pop_reg_reg(emit, &vtype_rhs, REG_ARG_2, &vtype_lhs, REG_ARG_1);
py/emitnative: Simplify binary op emitter, no need to check inplace ops.
2017-10-12 02:26:49 +01:00
if (op == MP_BINARY_OP_FLOOR_DIVIDE) {
py/emitnative: Implement floor-division and modulo for viper emitter.
2017-10-11 08:54:34 +01:00
emit_call(emit, MP_F_SMALL_INT_FLOOR_DIVIDE);
} else {
emit_call(emit, MP_F_SMALL_INT_MODULO);
}
emit_post_push_reg(emit, VTYPE_INT, REG_RET);
return;
}
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
int reg_rhs = REG_ARG_3;
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
emit_pre_pop_reg_flexible(emit, &vtype_rhs, &reg_rhs, REG_RET, REG_ARG_2);
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
emit_pre_pop_reg(emit, &vtype_lhs, REG_ARG_2);
#if !(N_X64 || N_X86)
py: remove "if (0)" and "if (false)" branches. Prior to this commit, building the unix port with `DEBUG=1` and `-finstrument-functions` the compilation would fail with an error like "control reaches end of non-void function". This change fixes this by removing the problematic "if (0)" branches. Not all branches affect compilation, but they are all removed for consistency.
2019-05-05 04:29:43 +01:00
if (op == MP_BINARY_OP_LSHIFT) {
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
ASM_LSL_REG_REG(emit->as, REG_ARG_2, reg_rhs);
py: Fix emitcpy and emitnative's binary_op.
2014-01-11 09:47:06 +00:00
emit_post_push_reg(emit, VTYPE_INT, REG_ARG_2);
py/emitnative: Simplify binary op emitter, no need to check inplace ops.
2017-10-12 02:26:49 +01:00
} else if (op == MP_BINARY_OP_RSHIFT) {
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
ASM_ASR_REG_REG(emit->as, REG_ARG_2, reg_rhs);
emit_post_push_reg(emit, VTYPE_INT, REG_ARG_2);
py: remove "if (0)" and "if (false)" branches. Prior to this commit, building the unix port with `DEBUG=1` and `-finstrument-functions` the compilation would fail with an error like "control reaches end of non-void function". This change fixes this by removing the problematic "if (0)" branches. Not all branches affect compilation, but they are all removed for consistency.
2019-05-05 04:29:43 +01:00
} else
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
#endif
py: remove "if (0)" and "if (false)" branches. Prior to this commit, building the unix port with `DEBUG=1` and `-finstrument-functions` the compilation would fail with an error like "control reaches end of non-void function". This change fixes this by removing the problematic "if (0)" branches. Not all branches affect compilation, but they are all removed for consistency.
2019-05-05 04:29:43 +01:00
if (op == MP_BINARY_OP_OR) {
py: Implement and,or,xor native ops for viper.
2014-10-12 14:21:06 +01:00
ASM_OR_REG_REG(emit->as, REG_ARG_2, reg_rhs);
emit_post_push_reg(emit, VTYPE_INT, REG_ARG_2);
py/emitnative: Simplify binary op emitter, no need to check inplace ops.
2017-10-12 02:26:49 +01:00
} else if (op == MP_BINARY_OP_XOR) {
py: Implement and,or,xor native ops for viper.
2014-10-12 14:21:06 +01:00
ASM_XOR_REG_REG(emit->as, REG_ARG_2, reg_rhs);
emit_post_push_reg(emit, VTYPE_INT, REG_ARG_2);
py/emitnative: Simplify binary op emitter, no need to check inplace ops.
2017-10-12 02:26:49 +01:00
} else if (op == MP_BINARY_OP_AND) {
py: Implement and,or,xor native ops for viper.
2014-10-12 14:21:06 +01:00
ASM_AND_REG_REG(emit->as, REG_ARG_2, reg_rhs);
emit_post_push_reg(emit, VTYPE_INT, REG_ARG_2);
py/emitnative: Simplify binary op emitter, no need to check inplace ops.
2017-10-12 02:26:49 +01:00
} else if (op == MP_BINARY_OP_ADD) {
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
ASM_ADD_REG_REG(emit->as, REG_ARG_2, reg_rhs);
emit_post_push_reg(emit, VTYPE_INT, REG_ARG_2);
py/emitnative: Simplify binary op emitter, no need to check inplace ops.
2017-10-12 02:26:49 +01:00
} else if (op == MP_BINARY_OP_SUBTRACT) {
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
ASM_SUB_REG_REG(emit->as, REG_ARG_2, reg_rhs);
emit_post_push_reg(emit, VTYPE_INT, REG_ARG_2);
py/emitnative: Simplify binary op emitter, no need to check inplace ops.
2017-10-12 02:26:49 +01:00
} else if (op == MP_BINARY_OP_MULTIPLY) {
py: Implement native multiply operation in viper emitter.
2015-06-04 15:00:29 +01:00
ASM_MUL_REG_REG(emit->as, REG_ARG_2, reg_rhs);
emit_post_push_reg(emit, VTYPE_INT, REG_ARG_2);
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
} else if (MP_BINARY_OP_LESS <= op && op <= MP_BINARY_OP_NOT_EQUAL) {
// comparison ops are (in enum order):
// MP_BINARY_OP_LESS
// MP_BINARY_OP_MORE
// MP_BINARY_OP_EQUAL
// MP_BINARY_OP_LESS_EQUAL
// MP_BINARY_OP_MORE_EQUAL
// MP_BINARY_OP_NOT_EQUAL
py: Partially fix viper multi-comparison; add test for it.
2014-10-19 19:00:51 +01:00
need_reg_single(emit, REG_RET, 0);
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
#if N_X64
asm_x64_xor_r64_r64(emit->as, REG_RET, REG_RET);
asm_x64_cmp_r64_with_r64(emit->as, reg_rhs, REG_ARG_2);
static byte ops[6] = {
ASM_X64_CC_JL,
ASM_X64_CC_JG,
ASM_X64_CC_JE,
ASM_X64_CC_JLE,
ASM_X64_CC_JGE,
ASM_X64_CC_JNE,
};
asm_x64_setcc_r8(emit->as, ops[op - MP_BINARY_OP_LESS], REG_RET);
#elif N_X86
asm_x86_xor_r32_r32(emit->as, REG_RET, REG_RET);
asm_x86_cmp_r32_with_r32(emit->as, reg_rhs, REG_ARG_2);
static byte ops[6] = {
ASM_X86_CC_JL,
ASM_X86_CC_JG,
ASM_X86_CC_JE,
ASM_X86_CC_JLE,
ASM_X86_CC_JGE,
ASM_X86_CC_JNE,
};
asm_x86_setcc_r8(emit->as, ops[op - MP_BINARY_OP_LESS], REG_RET);
#elif N_THUMB
asm_thumb_cmp_rlo_rlo(emit->as, REG_ARG_2, reg_rhs);
static uint16_t ops[6] = {
ASM_THUMB_OP_ITE_GE,
ASM_THUMB_OP_ITE_GT,
ASM_THUMB_OP_ITE_EQ,
ASM_THUMB_OP_ITE_GT,
ASM_THUMB_OP_ITE_GE,
ASM_THUMB_OP_ITE_EQ,
};
static byte ret[6] = { 0, 1, 1, 0, 1, 0, };
asm_thumb_op16(emit->as, ops[op - MP_BINARY_OP_LESS]);
asm_thumb_mov_rlo_i8(emit->as, REG_RET, ret[op - MP_BINARY_OP_LESS]);
asm_thumb_mov_rlo_i8(emit->as, REG_RET, ret[op - MP_BINARY_OP_LESS] ^ 1);
#elif N_ARM
Implement missing ARM emitter functions for viper
2014-10-03 23:53:46 +01:00
asm_arm_cmp_reg_reg(emit->as, REG_ARG_2, reg_rhs);
static uint ccs[6] = {
ASM_ARM_CC_LT,
ASM_ARM_CC_GT,
ASM_ARM_CC_EQ,
ASM_ARM_CC_LE,
ASM_ARM_CC_GE,
ASM_ARM_CC_NE,
};
asm_arm_setcc_reg(emit->as, REG_RET, ccs[op - MP_BINARY_OP_LESS]);
py: Integrate Xtensa assembler into native emitter. The config option MICROPY_EMIT_XTENSA can now be enabled to target the Xtensa architecture with @micropython.native and @micropython.viper decorators.
2016-12-09 05:39:39 +00:00
#elif N_XTENSA
static uint8_t ccs[6] = {
ASM_XTENSA_CC_LT,
0x80 | ASM_XTENSA_CC_LT, // for GT we'll swap args
ASM_XTENSA_CC_EQ,
0x80 | ASM_XTENSA_CC_GE, // for LE we'll swap args
ASM_XTENSA_CC_GE,
ASM_XTENSA_CC_NE,
};
uint8_t cc = ccs[op - MP_BINARY_OP_LESS];
if ((cc & 0x80) == 0) {
asm_xtensa_setcc_reg_reg_reg(emit->as, cc, REG_RET, REG_ARG_2, reg_rhs);
} else {
asm_xtensa_setcc_reg_reg_reg(emit->as, cc & ~0x80, REG_RET, reg_rhs, REG_ARG_2);
}
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
#else
#error not implemented
#endif
py: Fix emitcpy and emitnative's binary_op.
2014-01-11 09:47:06 +00:00
emit_post_push_reg(emit, VTYPE_BOOL, REG_RET);
} else {
// TODO other ops not yet implemented
py: Implement implicit cast to obj for viper load/store index/value. This allows to do "ar[i]" and "ar[i] = val" in viper when ar is a Python object and i and/or val are native viper types (eg ints). Patch also includes tests for this feature.
2015-06-04 11:52:16 +01:00
adjust_stack(emit, 1);
EMIT_NATIVE_VIPER_TYPE_ERROR(emit,
"binary op %q not implemented", mp_binary_op_method_name[op]);
py: Fix emitcpy and emitnative's binary_op.
2014-01-11 09:47:06 +00:00
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
} else if (vtype_lhs == VTYPE_PYOBJ && vtype_rhs == VTYPE_PYOBJ) {
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
emit_pre_pop_reg_reg(emit, &vtype_rhs, REG_ARG_3, &vtype_lhs, REG_ARG_2);
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
bool invert = false;
if (op == MP_BINARY_OP_NOT_IN) {
invert = true;
op = MP_BINARY_OP_IN;
} else if (op == MP_BINARY_OP_IS_NOT) {
invert = true;
op = MP_BINARY_OP_IS;
}
py: Code clean-up in native emitter; improve thumb native calls.
2014-08-16 22:31:57 +01:00
emit_call_with_imm_arg(emit, MP_F_BINARY_OP, op, REG_ARG_1);
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
if (invert) {
py: Implement more binary ops for viper emitter. This included a bit of restructuring of the assembler backends. Note that the ARM backend is missing a few functions and won't compile.
2014-09-29 18:45:42 +01:00
ASM_MOV_REG_REG(emit->as, REG_ARG_2, REG_RET);
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
emit_call_with_imm_arg(emit, MP_F_UNARY_OP, MP_UNARY_OP_NOT, REG_ARG_1);
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
} else {
py/emitnative.c: Fix stack adjustment when erroring on binary op.
2015-04-22 23:16:03 +01:00
adjust_stack(emit, -1);
py: Make viper codegen raise proper exception (ViperTypeError) on error. This fixes a long standing problem that viper code generation gave terrible error messages, and actually no errors on pyboard where assertions are disabled. Now all compile-time errors are raised as proper Python exceptions, and are of type ViperTypeError. Addresses issue #940.
2015-04-20 14:29:31 +01:00
EMIT_NATIVE_VIPER_TYPE_ERROR(emit,
"can't do binary op between '%q' and '%q'",
vtype_to_qstr(vtype_lhs), vtype_to_qstr(vtype_rhs));
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
}
py/emit: Merge build set/slice into existing build emit function. Reduces code size by: bare-arm: +0 minimal x86: +0 unix x64: -368 unix nanbox: -248 stm32: -128 cc3200: -48 esp8266: -184 esp32: -40
2018-05-22 13:18:42 +01:00
#if MICROPY_PY_BUILTINS_SLICE
STATIC void emit_native_build_slice(emit_t *emit, mp_uint_t n_args);
#endif
py/emit: Combine build tuple/list/map emit funcs into one. Reduces code size by: bare-arm: -24 minimal x86: -192 unix x64: -288 unix nanbox: -184 stm32: -72 cc3200: -16 esp8266: -148 esp32: -32
2018-05-18 15:41:40 +01:00
STATIC void emit_native_build(emit_t *emit, mp_uint_t n_args, int kind) {
Add iterators and comprehension to emitnative.
2013-10-16 23:58:48 +01:00
// for viper: call runtime, with types of args
// if wrapped in byte_array, or something, allocates memory and fills it
py/emit: Combine build tuple/list/map emit funcs into one. Reduces code size by: bare-arm: -24 minimal x86: -192 unix x64: -288 unix nanbox: -184 stm32: -72 cc3200: -16 esp8266: -148 esp32: -32
2018-05-18 15:41:40 +01:00
MP_STATIC_ASSERT(MP_F_BUILD_TUPLE + MP_EMIT_BUILD_TUPLE == MP_F_BUILD_TUPLE);
MP_STATIC_ASSERT(MP_F_BUILD_TUPLE + MP_EMIT_BUILD_LIST == MP_F_BUILD_LIST);
MP_STATIC_ASSERT(MP_F_BUILD_TUPLE + MP_EMIT_BUILD_MAP == MP_F_BUILD_MAP);
py/emit: Merge build set/slice into existing build emit function. Reduces code size by: bare-arm: +0 minimal x86: +0 unix x64: -368 unix nanbox: -248 stm32: -128 cc3200: -48 esp8266: -184 esp32: -40
2018-05-22 13:18:42 +01:00
MP_STATIC_ASSERT(MP_F_BUILD_TUPLE + MP_EMIT_BUILD_SET == MP_F_BUILD_SET);
#if MICROPY_PY_BUILTINS_SLICE
if (kind == MP_EMIT_BUILD_SLICE) {
emit_native_build_slice(emit, n_args);
return;
}
#endif
py: Rename emit_pre so they have globally unique names.
2014-03-27 23:30:26 +00:00
emit_native_pre(emit);
py/emit: Merge build set/slice into existing build emit function. Reduces code size by: bare-arm: +0 minimal x86: +0 unix x64: -368 unix nanbox: -248 stm32: -128 cc3200: -48 esp8266: -184 esp32: -40
2018-05-22 13:18:42 +01:00
if (kind == MP_EMIT_BUILD_TUPLE || kind == MP_EMIT_BUILD_LIST || kind == MP_EMIT_BUILD_SET) {
py/emit: Combine build tuple/list/map emit funcs into one. Reduces code size by: bare-arm: -24 minimal x86: -192 unix x64: -288 unix nanbox: -184 stm32: -72 cc3200: -16 esp8266: -148 esp32: -32
2018-05-18 15:41:40 +01:00
emit_get_stack_pointer_to_reg_for_pop(emit, REG_ARG_2, n_args); // pointer to items
}
emit_call_with_imm_arg(emit, MP_F_BUILD_TUPLE + kind, n_args, REG_ARG_1);
py/emit: Merge build set/slice into existing build emit function. Reduces code size by: bare-arm: +0 minimal x86: +0 unix x64: -368 unix nanbox: -248 stm32: -128 cc3200: -48 esp8266: -184 esp32: -40
2018-05-22 13:18:42 +01:00
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET); // new tuple/list/map/set
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_native_store_map(emit_t *emit) {
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
vtype_kind_t vtype_key, vtype_value, vtype_map;
emit_pre_pop_reg_reg_reg(emit, &vtype_key, REG_ARG_2, &vtype_value, REG_ARG_3, &vtype_map, REG_ARG_1); // key, value, map
assert(vtype_key == VTYPE_PYOBJ);
assert(vtype_value == VTYPE_PYOBJ);
assert(vtype_map == VTYPE_PYOBJ);
py: Code clean-up in native emitter; improve thumb native calls.
2014-08-16 22:31:57 +01:00
emit_call(emit, MP_F_STORE_MAP);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET); // map
}
py: Allow to properly disable builtin slice operation. This patch makes the MICROPY_PY_BUILTINS_SLICE compile-time option fully disable the builtin slice operation (when set to 0). This includes removing the slice sytanx from the grammar. Now, enabling slice costs 4228 bytes on unix x64, and 1816 bytes on stmhal.
2014-12-27 17:20:41 +00:00
#if MICROPY_PY_BUILTINS_SLICE
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
STATIC void emit_native_build_slice(emit_t *emit, mp_uint_t n_args) {
py: Implement more features in native emitter. On x64, native emitter now passes 70 of the tests.
2014-04-06 12:58:40 +01:00
DEBUG_printf("build_slice %d\n", n_args);
py: Improvements to native emitter. Native emitter can now compile try/except blocks using nlr_push/nlr_pop. It probably only works for 1 level of exception handling. It doesn't work on Thumb (only x64). Native emitter can also handle some additional op codes. With this patch, 198 tests now pass using "-X emit=native" option to micropython.
2014-06-30 05:17:25 +01:00
if (n_args == 2) {
vtype_kind_t vtype_start, vtype_stop;
emit_pre_pop_reg_reg(emit, &vtype_stop, REG_ARG_2, &vtype_start, REG_ARG_1); // arg1 = start, arg2 = stop
assert(vtype_start == VTYPE_PYOBJ);
assert(vtype_stop == VTYPE_PYOBJ);
py/emitnative: Put None/False/True in global native const table. So these constant objects can be loaded by dereferencing the REG_FUN_TABLE pointer instead of loading immediate values. This reduces the size of generated native code (when such constants are used), and means that pointers to these constants are no longer stored in the assembly code.
2018-10-14 13:56:46 +01:00
emit_native_mov_reg_const(emit, REG_ARG_3, MP_F_CONST_NONE_OBJ); // arg3 = step
py: Improvements to native emitter. Native emitter can now compile try/except blocks using nlr_push/nlr_pop. It probably only works for 1 level of exception handling. It doesn't work on Thumb (only x64). Native emitter can also handle some additional op codes. With this patch, 198 tests now pass using "-X emit=native" option to micropython.
2014-06-30 05:17:25 +01:00
} else {
assert(n_args == 3);
vtype_kind_t vtype_start, vtype_stop, vtype_step;
emit_pre_pop_reg_reg_reg(emit, &vtype_step, REG_ARG_3, &vtype_stop, REG_ARG_2, &vtype_start, REG_ARG_1); // arg1 = start, arg2 = stop, arg3 = step
assert(vtype_start == VTYPE_PYOBJ);
assert(vtype_stop == VTYPE_PYOBJ);
assert(vtype_step == VTYPE_PYOBJ);
}
py/emitnative: Put None/False/True in global native const table. So these constant objects can be loaded by dereferencing the REG_FUN_TABLE pointer instead of loading immediate values. This reduces the size of generated native code (when such constants are used), and means that pointers to these constants are no longer stored in the assembly code.
2018-10-14 13:56:46 +01:00
emit_call(emit, MP_F_NEW_SLICE);
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Allow to properly disable builtin slice operation. This patch makes the MICROPY_PY_BUILTINS_SLICE compile-time option fully disable the builtin slice operation (when set to 0). This includes removing the slice sytanx from the grammar. Now, enabling slice costs 4228 bytes on unix x64, and 1816 bytes on stmhal.
2014-12-27 17:20:41 +00:00
#endif
py: Implement more features in native emitter. On x64, native emitter now passes 70 of the tests.
2014-04-06 12:58:40 +01:00
py: Combine 3 comprehension emit functions (list/dict/set) into 1. The 3 kinds of comprehensions are similar enough that merging their emit functions reduces code size. Decreases in code size in bytes are: bare-arm:24, minimal:96, unix(NDEBUG,x86-64):328, stmhal:80, esp8266:76.
2016-09-18 14:59:47 +01:00
STATIC void emit_native_store_comp(emit_t *emit, scope_kind_t kind, mp_uint_t collection_index) {
mp_fun_kind_t f;
if (kind == SCOPE_LIST_COMP) {
vtype_kind_t vtype_item;
emit_pre_pop_reg(emit, &vtype_item, REG_ARG_2);
assert(vtype_item == VTYPE_PYOBJ);
f = MP_F_LIST_APPEND;
#if MICROPY_PY_BUILTINS_SET
} else if (kind == SCOPE_SET_COMP) {
vtype_kind_t vtype_item;
emit_pre_pop_reg(emit, &vtype_item, REG_ARG_2);
assert(vtype_item == VTYPE_PYOBJ);
f = MP_F_STORE_SET;
#endif
} else {
// SCOPE_DICT_COMP
vtype_kind_t vtype_key, vtype_value;
emit_pre_pop_reg_reg(emit, &vtype_key, REG_ARG_2, &vtype_value, REG_ARG_3);
assert(vtype_key == VTYPE_PYOBJ);
assert(vtype_value == VTYPE_PYOBJ);
f = MP_F_STORE_MAP;
}
vtype_kind_t vtype_collection;
emit_access_stack(emit, collection_index, &vtype_collection, REG_ARG_1);
assert(vtype_collection == VTYPE_PYOBJ);
emit_call(emit, f);
emit_post(emit);
}
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
STATIC void emit_native_unpack_sequence(emit_t *emit, mp_uint_t n_args) {
py: Implement more features in native emitter. On x64, native emitter now passes 70 of the tests.
2014-04-06 12:58:40 +01:00
DEBUG_printf("unpack_sequence %d\n", n_args);
vtype_kind_t vtype_base;
emit_pre_pop_reg(emit, &vtype_base, REG_ARG_1); // arg1 = seq
assert(vtype_base == VTYPE_PYOBJ);
emit_get_stack_pointer_to_reg_for_push(emit, REG_ARG_3, n_args); // arg3 = dest ptr
py: Code clean-up in native emitter; improve thumb native calls.
2014-08-16 22:31:57 +01:00
emit_call_with_imm_arg(emit, MP_F_UNPACK_SEQUENCE, n_args, REG_ARG_2); // arg2 = n_args
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Implement more features in native emitter. On x64, native emitter now passes 70 of the tests.
2014-04-06 12:58:40 +01:00
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
STATIC void emit_native_unpack_ex(emit_t *emit, mp_uint_t n_left, mp_uint_t n_right) {
py: Improve native emitter; now supports more opcodes.
2014-05-07 18:30:52 +01:00
DEBUG_printf("unpack_ex %d %d\n", n_left, n_right);
vtype_kind_t vtype_base;
emit_pre_pop_reg(emit, &vtype_base, REG_ARG_1); // arg1 = seq
assert(vtype_base == VTYPE_PYOBJ);
py: Improvements to native emitter. Native emitter can now compile try/except blocks using nlr_push/nlr_pop. It probably only works for 1 level of exception handling. It doesn't work on Thumb (only x64). Native emitter can also handle some additional op codes. With this patch, 198 tests now pass using "-X emit=native" option to micropython.
2014-06-30 05:17:25 +01:00
emit_get_stack_pointer_to_reg_for_push(emit, REG_ARG_3, n_left + n_right + 1); // arg3 = dest ptr
py: Code clean-up in native emitter; improve thumb native calls.
2014-08-16 22:31:57 +01:00
emit_call_with_imm_arg(emit, MP_F_UNPACK_EX, n_left | (n_right << 8), REG_ARG_2); // arg2 = n_left + n_right
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
STATIC void emit_native_make_function(emit_t *emit, scope_t *scope, mp_uint_t n_pos_defaults, mp_uint_t n_kw_defaults) {
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
// call runtime, with type info for args, or don't support dict/default params, or only support Python objects for them
py: Rename emit_pre so they have globally unique names.
2014-03-27 23:30:26 +00:00
emit_native_pre(emit);
py: Improve native emitter; now supports more opcodes.
2014-05-07 18:30:52 +01:00
if (n_pos_defaults == 0 && n_kw_defaults == 0) {
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
need_reg_all(emit);
ASM_MOV_REG_IMM(emit->as, REG_ARG_2, (mp_uint_t)MP_OBJ_NULL);
ASM_MOV_REG_IMM(emit->as, REG_ARG_3, (mp_uint_t)MP_OBJ_NULL);
py: Improve native emitter; now supports more opcodes.
2014-05-07 18:30:52 +01:00
} else {
vtype_kind_t vtype_def_tuple, vtype_def_dict;
emit_pre_pop_reg_reg(emit, &vtype_def_dict, REG_ARG_3, &vtype_def_tuple, REG_ARG_2);
assert(vtype_def_tuple == VTYPE_PYOBJ);
assert(vtype_def_dict == VTYPE_PYOBJ);
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
need_reg_all(emit);
py: Improve native emitter; now supports more opcodes.
2014-05-07 18:30:52 +01:00
}
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
emit_load_reg_with_raw_code(emit, REG_ARG_1, scope->raw_code);
py/emitnative: Remove unused ptr argument from ASM_CALL_IND macro.
2018-10-13 05:13:55 +01:00
ASM_CALL_IND(emit->as, MP_F_MAKE_FUNCTION_FROM_RAW_CODE);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
}
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
STATIC void emit_native_make_closure(emit_t *emit, scope_t *scope, mp_uint_t n_closed_over, mp_uint_t n_pos_defaults, mp_uint_t n_kw_defaults) {
py: Implement closures in native code generator. Currently supports only x64 and Thumb2 archs.
2015-01-15 14:41:41 +00:00
emit_native_pre(emit);
if (n_pos_defaults == 0 && n_kw_defaults == 0) {
emit_get_stack_pointer_to_reg_for_pop(emit, REG_ARG_3, n_closed_over);
py/emitnative: Clean up asm macro names so they have dest as first arg. All the asm macro names that convert a particular architecture to a generic interface now follow the convention whereby the "destination" (usually a register) is specified first.
2017-11-15 00:46:49 +00:00
ASM_MOV_REG_IMM(emit->as, REG_ARG_2, n_closed_over);
py: Implement closures in native code generator. Currently supports only x64 and Thumb2 archs.
2015-01-15 14:41:41 +00:00
} else {
emit_get_stack_pointer_to_reg_for_pop(emit, REG_ARG_3, n_closed_over + 2);
py/emitnative: Clean up asm macro names so they have dest as first arg. All the asm macro names that convert a particular architecture to a generic interface now follow the convention whereby the "destination" (usually a register) is specified first.
2017-11-15 00:46:49 +00:00
ASM_MOV_REG_IMM(emit->as, REG_ARG_2, 0x100 | n_closed_over);
py: Implement closures in native code generator. Currently supports only x64 and Thumb2 archs.
2015-01-15 14:41:41 +00:00
}
py/emitnative: Place const objs for native code in separate const table. This commit changes native code to handle constant objects like bytecode: instead of storing the pointers inside the native code they are now stored in a separate constant table (such pointers include objects like bignum, bytes, and raw code for nested functions). This removes the need for the GC to scan native code for root pointers, and takes a step towards making native code independent of the runtime (eg so it can be compiled offline by mpy-cross). Note that the changes to the struct scope_t did not increase its size: on a 32-bit architecture it is still 48 bytes, and on a 64-bit architecture it decreased from 80 to 72 bytes.
2018-09-27 14:27:53 +01:00
emit_load_reg_with_raw_code(emit, REG_ARG_1, scope->raw_code);
py/emitnative: Remove unused ptr argument from ASM_CALL_IND macro.
2018-10-13 05:13:55 +01:00
ASM_CALL_IND(emit->as, MP_F_MAKE_CLOSURE_FROM_RAW_CODE);
py: Implement closures in native code generator. Currently supports only x64 and Thumb2 archs.
2015-01-15 14:41:41 +00:00
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
STATIC void emit_native_call_function(emit_t *emit, mp_uint_t n_positional, mp_uint_t n_keyword, mp_uint_t star_flags) {
py: Make native emitter handle multi-compare and not/is not/not in ops.
2014-09-23 15:10:03 +01:00
DEBUG_printf("call_function(n_pos=" UINT_FMT ", n_kw=" UINT_FMT ", star_flags=" UINT_FMT ")\n", n_positional, n_keyword, star_flags);
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
// TODO: in viper mode, call special runtime routine with type info for args,
// and wanted type info for return, to remove need for boxing/unboxing
emit_native_pre(emit);
vtype_kind_t vtype_fun = peek_vtype(emit, n_positional + 2 * n_keyword);
if (vtype_fun == VTYPE_BUILTIN_CAST) {
// casting operator
assert(n_positional == 1 && n_keyword == 0);
py: Implement calling functions with *args in native emitter.
2015-04-06 22:48:21 +01:00
assert(!star_flags);
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
DEBUG_printf(" cast to %d\n", vtype_fun);
py: Don't use anonymous unions, name them instead. This makes the code (more) compatible with the C99 standard.
2015-01-24 23:14:12 +00:00
vtype_kind_t vtype_cast = peek_stack(emit, 1)->data.u_imm;
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
switch (peek_vtype(emit, 0)) {
case VTYPE_PYOBJ: {
vtype_kind_t vtype;
emit_pre_pop_reg(emit, &vtype, REG_ARG_1);
emit_pre_pop_discard(emit);
py: Extend native type-sig to use 4 bits, so uint is separate to ptr. Before this patch, the native types for uint and ptr/ptr8/ptr16/ptr32 all overlapped and it was possible to make a mistake in casting. Now, these types are all separate and any coding mistakes will be raised as runtime errors.
2016-02-02 23:16:05 +00:00
emit_call_with_imm_arg(emit, MP_F_CONVERT_OBJ_TO_NATIVE, vtype_cast, REG_ARG_2); // arg2 = type
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
emit_post_push_reg(emit, vtype_cast, REG_RET);
break;
}
case VTYPE_BOOL:
case VTYPE_INT:
case VTYPE_UINT:
case VTYPE_PTR:
case VTYPE_PTR8:
case VTYPE_PTR16:
py: Implement ptr32 load and store in viper emitter.
2015-10-13 00:50:17 +01:00
case VTYPE_PTR32:
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
case VTYPE_PTR_NONE:
emit_fold_stack_top(emit, REG_ARG_1);
emit_post_top_set_vtype(emit, vtype_cast);
break;
default:
py/emitnative: Use assertions and mp_not_implemented correctly. Assertions are used to check expressions that should always be true, and mp_not_implemented is used for code that can be reached.
2017-03-14 12:00:53 +00:00
// this can happen when casting a cast: int(int)
all: Raise exceptions via mp_raise_XXX - Changed: ValueError, TypeError, NotImplementedError - OSError invocations unchanged, because the corresponding utility function takes ints, not strings like the long form invocation. - OverflowError, IndexError and RuntimeError etc. not changed for now until we decide whether to add new utility functions.
2017-08-09 05:40:45 +01:00
mp_raise_NotImplementedError("casting");
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
}
} else {
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
assert(vtype_fun == VTYPE_PYOBJ);
py: Implement calling functions with *args in native emitter.
2015-04-06 22:48:21 +01:00
if (star_flags) {
emit_get_stack_pointer_to_reg_for_pop(emit, REG_ARG_3, n_positional + 2 * n_keyword + 3); // pointer to args
emit_call_with_2_imm_args(emit, MP_F_CALL_METHOD_N_KW_VAR, 0, REG_ARG_1, n_positional | (n_keyword << 8), REG_ARG_2);
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
} else {
if (n_positional != 0 || n_keyword != 0) {
emit_get_stack_pointer_to_reg_for_pop(emit, REG_ARG_3, n_positional + 2 * n_keyword); // pointer to args
}
emit_pre_pop_reg(emit, &vtype_fun, REG_ARG_1); // the function
emit_call_with_imm_arg(emit, MP_F_NATIVE_CALL_FUNCTION_N_KW, n_positional | (n_keyword << 8), REG_ARG_2);
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
}
py: Partially fix native emitter to work with latest runtime. Native emitter has been broken since stack order has changed from reverse to standard. This fix gets it partially working.
2014-02-02 13:11:48 +00:00
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
STATIC void emit_native_call_method(emit_t *emit, mp_uint_t n_positional, mp_uint_t n_keyword, mp_uint_t star_flags) {
py: Implement calling functions with *args in native emitter.
2015-04-06 22:48:21 +01:00
if (star_flags) {
emit_get_stack_pointer_to_reg_for_pop(emit, REG_ARG_3, n_positional + 2 * n_keyword + 4); // pointer to args
emit_call_with_2_imm_args(emit, MP_F_CALL_METHOD_N_KW_VAR, 1, REG_ARG_1, n_positional | (n_keyword << 8), REG_ARG_2);
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
} else {
emit_native_pre(emit);
emit_get_stack_pointer_to_reg_for_pop(emit, REG_ARG_3, 2 + n_positional + 2 * n_keyword); // pointer to items, including meth and self
emit_call_with_2_imm_args(emit, MP_F_CALL_METHOD_N_KW, n_positional, REG_ARG_1, n_keyword, REG_ARG_2);
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
Replace global "static" -> "STATIC", to allow "analysis builds". Part 2.
2014-02-12 16:31:30 +00:00
STATIC void emit_native_return_value(emit_t *emit) {
py: Implement more features in native emitter. On x64, native emitter now passes 70 of the tests.
2014-04-06 12:58:40 +01:00
DEBUG_printf("return_value\n");
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
if (emit->scope->scope_flags & MP_SCOPE_FLAG_GENERATOR) {
// Save pointer to current stack position for caller to access return value
emit_get_stack_pointer_to_reg_for_pop(emit, REG_TEMP0, 1);
emit_native_mov_state_reg(emit, offsetof(mp_code_state_t, sp) / sizeof(uintptr_t), REG_TEMP0);
// Put return type in return value slot
ASM_MOV_REG_IMM(emit->as, REG_TEMP0, MP_VM_RETURN_NORMAL);
ASM_MOV_LOCAL_REG(emit->as, LOCAL_IDX_RET_VAL(emit), REG_TEMP0);
// Do the unwinding jump to get to the return handler
emit_native_unwind_jump(emit, emit->exit_label, emit->exc_stack_size);
emit->last_emit_was_return_value = true;
return;
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
if (emit->do_viper_types) {
py/emit: Remove need to call set_native_type to set viper return type. Instead this return type is now stored in the scope_flags.
2018-09-15 03:41:25 +01:00
vtype_kind_t return_vtype = emit->scope->scope_flags >> MP_SCOPE_FLAG_VIPERRET_POS;
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
if (peek_vtype(emit, 0) == VTYPE_PTR_NONE) {
emit_pre_pop_discard(emit);
py/emit: Remove need to call set_native_type to set viper return type. Instead this return type is now stored in the scope_flags.
2018-09-15 03:41:25 +01:00
if (return_vtype == VTYPE_PYOBJ) {
py/emitnative: Put None/False/True in global native const table. So these constant objects can be loaded by dereferencing the REG_FUN_TABLE pointer instead of loading immediate values. This reduces the size of generated native code (when such constants are used), and means that pointers to these constants are no longer stored in the assembly code.
2018-10-14 13:56:46 +01:00
emit_native_mov_reg_const(emit, REG_RET, MP_F_CONST_NONE_OBJ);
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
} else {
py: Make viper functions have the same entry signature as native. This commit makes viper functions have the same signature as native functions, at the level of the emitter/assembler. This means that viper functions can now be wrapped in the same uPy object as native functions. Viper functions are now responsible for parsing their arguments (before it was done by the runtime), and this makes calling them more efficient (in most cases) because the viper entry code can be custom generated to suit the signature of the function. This change also opens the way forward for viper functions to take arbitrary numbers of arguments, and for them to handle globals correctly, among other things.
2018-09-14 08:40:59 +01:00
ASM_MOV_REG_IMM(emit->as, REG_ARG_1, 0);
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
}
} else {
vtype_kind_t vtype;
py: Make viper functions have the same entry signature as native. This commit makes viper functions have the same signature as native functions, at the level of the emitter/assembler. This means that viper functions can now be wrapped in the same uPy object as native functions. Viper functions are now responsible for parsing their arguments (before it was done by the runtime), and this makes calling them more efficient (in most cases) because the viper entry code can be custom generated to suit the signature of the function. This change also opens the way forward for viper functions to take arbitrary numbers of arguments, and for them to handle globals correctly, among other things.
2018-09-14 08:40:59 +01:00
emit_pre_pop_reg(emit, &vtype, return_vtype == VTYPE_PYOBJ ? REG_RET : REG_ARG_1);
py/emit: Remove need to call set_native_type to set viper return type. Instead this return type is now stored in the scope_flags.
2018-09-15 03:41:25 +01:00
if (vtype != return_vtype) {
py: Make viper codegen raise proper exception (ViperTypeError) on error. This fixes a long standing problem that viper code generation gave terrible error messages, and actually no errors on pyboard where assertions are disabled. Now all compile-time errors are raised as proper Python exceptions, and are of type ViperTypeError. Addresses issue #940.
2015-04-20 14:29:31 +01:00
EMIT_NATIVE_VIPER_TYPE_ERROR(emit,
"return expected '%q' but got '%q'",
py/emit: Remove need to call set_native_type to set viper return type. Instead this return type is now stored in the scope_flags.
2018-09-15 03:41:25 +01:00
vtype_to_qstr(return_vtype), vtype_to_qstr(vtype));
py: Viper can now store to global.
2014-08-15 23:47:59 +01:00
}
py: Allow viper to have type annotations. Viper functions can now be annotated with the type of their arguments and return value. Eg: @micropython.viper def f(x:int) -> int: return x + 1
2014-08-15 16:45:41 +01:00
}
py: Make viper functions have the same entry signature as native. This commit makes viper functions have the same signature as native functions, at the level of the emitter/assembler. This means that viper functions can now be wrapped in the same uPy object as native functions. Viper functions are now responsible for parsing their arguments (before it was done by the runtime), and this makes calling them more efficient (in most cases) because the viper entry code can be custom generated to suit the signature of the function. This change also opens the way forward for viper functions to take arbitrary numbers of arguments, and for them to handle globals correctly, among other things.
2018-09-14 08:40:59 +01:00
if (return_vtype != VTYPE_PYOBJ) {
emit_call_with_imm_arg(emit, MP_F_CONVERT_NATIVE_TO_OBJ, return_vtype, REG_ARG_2);
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
} else {
py: Add casting to viper; add native mem stores to viper. Viper can now do the following: def store(p:ptr8, c:int): p[0] = c This does a store of c to the memory pointed to by p using a machine instructions inline in the code.
2014-09-29 22:10:41 +01:00
vtype_kind_t vtype;
emit_pre_pop_reg(emit, &vtype, REG_RET);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
assert(vtype == VTYPE_PYOBJ);
}
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
if (NEED_GLOBAL_EXC_HANDLER(emit)) {
// Save return value for the global exception handler to use
ASM_MOV_LOCAL_REG(emit->as, LOCAL_IDX_RET_VAL(emit), REG_RET);
}
emit_native_unwind_jump(emit, emit->exit_label, emit->exc_stack_size);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit->last_emit_was_return_value = true;
}
py: Convert [u]int to mp_[u]int_t in emit.h and associated .c files. Towards resolving issue #50.
2014-09-08 23:05:16 +01:00
STATIC void emit_native_raise_varargs(emit_t *emit, mp_uint_t n_args) {
py: Eliminate warnings about unused arguments when debugging disabled.
2019-02-25 03:52:36 +00:00
(void)n_args;
py: Improvements to native emitter. Native emitter can now compile try/except blocks using nlr_push/nlr_pop. It probably only works for 1 level of exception handling. It doesn't work on Thumb (only x64). Native emitter can also handle some additional op codes. With this patch, 198 tests now pass using "-X emit=native" option to micropython.
2014-06-30 05:17:25 +01:00
assert(n_args == 1);
py: Viper can call functions with native types, and raise exceptions.
2014-08-16 22:06:11 +01:00
vtype_kind_t vtype_exc;
emit_pre_pop_reg(emit, &vtype_exc, REG_ARG_1); // arg1 = object to raise
if (vtype_exc != VTYPE_PYOBJ) {
py: Make viper codegen raise proper exception (ViperTypeError) on error. This fixes a long standing problem that viper code generation gave terrible error messages, and actually no errors on pyboard where assertions are disabled. Now all compile-time errors are raised as proper Python exceptions, and are of type ViperTypeError. Addresses issue #940.
2015-04-20 14:29:31 +01:00
EMIT_NATIVE_VIPER_TYPE_ERROR(emit, "must raise an object");
py: Viper can call functions with native types, and raise exceptions.
2014-08-16 22:06:11 +01:00
}
// TODO probably make this 1 call to the runtime (which could even call convert, native_raise(obj, type))
py: Code clean-up in native emitter; improve thumb native calls.
2014-08-16 22:31:57 +01:00
emit_call(emit, MP_F_NATIVE_RAISE);
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Improvements to native emitter. Native emitter can now compile try/except blocks using nlr_push/nlr_pop. It probably only works for 1 level of exception handling. It doesn't work on Thumb (only x64). Native emitter can also handle some additional op codes. With this patch, 198 tests now pass using "-X emit=native" option to micropython.
2014-06-30 05:17:25 +01:00
py/emit: Combine yield value and yield-from emit funcs into one. Reduces code size by: bare-arm: -24 minimal x86: -72 unix x64: -200 unix nanbox: -72 stm32: -52 cc3200: -32 esp8266: -84 esp32: -24
2018-05-18 15:30:42 +01:00
STATIC void emit_native_yield(emit_t *emit, int kind) {
py/emitnative: Implement yield and yield-from in native emitter. This commit adds first class support for yield and yield-from in the native emitter, including send and throw support, and yields enclosed in exception handlers (which requires pulling down the NLR stack before yielding, then rebuilding it when resuming). This has been fully tested and is working on unix x86 and x86-64, and stm32. Also basic tests have been done with the esp8266 port. Performance of existing native code is unchanged.
2018-10-01 04:07:04 +01:00
// Note: 1 (yield) or 3 (yield from) labels are reserved for this function, starting at *emit->label_slot
if (emit->do_viper_types) {
mp_raise_NotImplementedError("native yield");
}
emit->scope->scope_flags |= MP_SCOPE_FLAG_GENERATOR;
need_stack_settled(emit);
if (kind == MP_EMIT_YIELD_FROM) {
// Top of yield-from loop, conceptually implementing:
// for item in generator:
// yield item
// Jump to start of loop
emit_native_jump(emit, *emit->label_slot + 2);
// Label for top of loop
emit_native_label_assign(emit, *emit->label_slot + 1);
}
// Save pointer to current stack position for caller to access yielded value
emit_get_stack_pointer_to_reg_for_pop(emit, REG_TEMP0, 1);
emit_native_mov_state_reg(emit, offsetof(mp_code_state_t, sp) / sizeof(uintptr_t), REG_TEMP0);
// Put return type in return value slot
ASM_MOV_REG_IMM(emit->as, REG_TEMP0, MP_VM_RETURN_YIELD);
ASM_MOV_LOCAL_REG(emit->as, LOCAL_IDX_RET_VAL(emit), REG_TEMP0);
// Save re-entry PC
ASM_MOV_REG_PCREL(emit->as, REG_TEMP0, *emit->label_slot);
emit_native_mov_state_reg(emit, LOCAL_IDX_GEN_PC(emit), REG_TEMP0);
// Jump to exit handler
ASM_JUMP(emit->as, emit->exit_label);
// Label re-entry point
mp_asm_base_label_assign(&emit->as->base, *emit->label_slot);
// Re-open any active exception handler
if (emit->exc_stack_size > 0) {
// Find innermost active exception handler, to restore as current handler
exc_stack_entry_t *e = &emit->exc_stack[emit->exc_stack_size - 1];
for (; e >= emit->exc_stack; --e) {
if (e->is_active) {
// Found active handler, get its PC
ASM_MOV_REG_PCREL(emit->as, REG_RET, e->label);
ASM_MOV_LOCAL_REG(emit->as, LOCAL_IDX_EXC_HANDLER_PC(emit), REG_RET);
}
}
}
emit_native_adjust_stack_size(emit, 1); // send_value
if (kind == MP_EMIT_YIELD_VALUE) {
// Check LOCAL_IDX_EXC_VAL for any injected value
ASM_MOV_REG_LOCAL(emit->as, REG_ARG_1, LOCAL_IDX_EXC_VAL(emit));
emit_call(emit, MP_F_NATIVE_RAISE);
} else {
// Label loop entry
emit_native_label_assign(emit, *emit->label_slot + 2);
// Get the next item from the delegate generator
vtype_kind_t vtype;
emit_pre_pop_reg(emit, &vtype, REG_ARG_2); // send_value
emit_access_stack(emit, 1, &vtype, REG_ARG_1); // generator
ASM_MOV_REG_LOCAL(emit->as, REG_ARG_3, LOCAL_IDX_EXC_VAL(emit)); // throw_value
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_ARG_3);
emit_get_stack_pointer_to_reg_for_pop(emit, REG_ARG_3, 1); // ret_value
emit_call(emit, MP_F_NATIVE_YIELD_FROM);
// If returned non-zero then generator continues
ASM_JUMP_IF_REG_NONZERO(emit->as, REG_RET, *emit->label_slot + 1, true);
// Pop exhausted gen, replace with ret_value
emit_native_adjust_stack_size(emit, 1); // ret_value
emit_fold_stack_top(emit, REG_ARG_1);
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
}
py: Improvements to native emitter. Native emitter can now compile try/except blocks using nlr_push/nlr_pop. It probably only works for 1 level of exception handling. It doesn't work on Thumb (only x64). Native emitter can also handle some additional op codes. With this patch, 198 tests now pass using "-X emit=native" option to micropython.
2014-06-30 05:17:25 +01:00
STATIC void emit_native_start_except_handler(emit_t *emit) {
py/emitnative: Add support for return/break/continue in try and with. This patch adds full support for unwinding jumps to the native emitter. This means that return/break/continue can be used in try-except, try-finally and with statements. For code that doesn't use unwinding jumps there is almost no overhead added to the generated code.
2018-09-04 05:31:28 +01:00
// Protected block has finished so leave the current exception handler
emit_native_leave_exc_stack(emit, true);
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
// Get and push nlr_buf.ret_val
ASM_MOV_REG_LOCAL(emit->as, REG_TEMP0, LOCAL_IDX_EXC_VAL(emit));
emit_post_push_reg(emit, VTYPE_PYOBJ, REG_TEMP0);
py: Improvements to native emitter. Native emitter can now compile try/except blocks using nlr_push/nlr_pop. It probably only works for 1 level of exception handling. It doesn't work on Thumb (only x64). Native emitter can also handle some additional op codes. With this patch, 198 tests now pass using "-X emit=native" option to micropython.
2014-06-30 05:17:25 +01:00
}
STATIC void emit_native_end_except_handler(emit_t *emit) {
py/emitnative: Optimise and improve exception handling in native code. Prior to this patch, native code would use a full nlr_buf_t for each exception handler (try-except, try-finally, with). For nested exception handlers this would use a lot of C stack and be rather inefficient. This patch changes how exceptions are handled in native code by setting up only a single nlr_buf_t context for the entire function, and then manages a state machine (using the PC) to work out which exception handler to run when an exception is raised by an nlr_jump. This keeps the C stack usage at a constant level regardless of the depth of Python exception blocks. The patch also fixes an existing bug when local variables are written to within an exception handler, then their value was incorrectly restored if an exception was raised (since the nlr_jump would restore register values, back to the point of the nlr_push). And it also gets nested try-finally+with working with the viper emitter. Broadly speaking, efficiency of executing native code that doesn't use any exception blocks is unchanged, and emitted code size is only slightly increased for such function. C stack usage of all native functions is either equal or less than before. Emitted code size for native functions that use exception blocks is increased by roughly 10% (due in part to fixing of above-mentioned bugs). But, most importantly, this patch allows to implement more Python features in native code, like unwind jumps and yielding from within nested exception blocks.
2018-08-16 04:56:36 +01:00
adjust_stack(emit, -1); // pop the exception (end_finally didn't use it)
py: Improvements to native emitter. Native emitter can now compile try/except blocks using nlr_push/nlr_pop. It probably only works for 1 level of exception handling. It doesn't work on Thumb (only x64). Native emitter can also handle some additional op codes. With this patch, 198 tests now pass using "-X emit=native" option to micropython.
2014-06-30 05:17:25 +01:00
}
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
const emit_method_table_t EXPORT_FUN(method_table) = {
py/compile: Add support to select the native emitter at runtime.
2019-03-08 23:59:25 +00:00
#if MICROPY_DYNAMIC_COMPILER
EXPORT_FUN(new),
EXPORT_FUN(free),
#endif
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit_native_start_pass,
emit_native_end_pass,
emit_native_last_emit_was_return_value,
py: Simplify stack get/set to become stack adjust in emitters. Can do this now that the stack size calculation is improved.
2014-04-10 18:28:54 +01:00
emit_native_adjust_stack_size,
Add source file name and line number to error messages. Byte code has a map from byte-code offset to source-code line number, used to give better error messages.
2014-01-18 23:24:36 +00:00
emit_native_set_source_line,
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
py, compiler: Refactor load/store/delete_id logic to reduce code size. Saves around 230 bytes on Thumb2 and 750 bytes on x86.
2015-03-26 14:42:40 +00:00
{
py/emit: Combine fast and deref into one function for load/store/delete. Reduces code size by: bare-arm: -16 minimal x86: -208 unix x64: -408 unix nanbox: -248 stm32: -12 cc3200: -24 esp8266: -96 esp32: -44
2018-05-18 15:11:04 +01:00
emit_native_load_local,
py, compiler: Refactor load/store/delete_id logic to reduce code size. Saves around 230 bytes on Thumb2 and 750 bytes on x86.
2015-03-26 14:42:40 +00:00
emit_native_load_global,
},
{
py/emit: Combine fast and deref into one function for load/store/delete. Reduces code size by: bare-arm: -16 minimal x86: -208 unix x64: -408 unix nanbox: -248 stm32: -12 cc3200: -24 esp8266: -96 esp32: -44
2018-05-18 15:11:04 +01:00
emit_native_store_local,
py, compiler: Refactor load/store/delete_id logic to reduce code size. Saves around 230 bytes on Thumb2 and 750 bytes on x86.
2015-03-26 14:42:40 +00:00
emit_native_store_global,
},
{
py/emit: Combine fast and deref into one function for load/store/delete. Reduces code size by: bare-arm: -16 minimal x86: -208 unix x64: -408 unix nanbox: -248 stm32: -12 cc3200: -24 esp8266: -96 esp32: -44
2018-05-18 15:11:04 +01:00
emit_native_delete_local,
py, compiler: Refactor load/store/delete_id logic to reduce code size. Saves around 230 bytes on Thumb2 and 750 bytes on x86.
2015-03-26 14:42:40 +00:00
emit_native_delete_global,
},
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit_native_label_assign,
py/emit: Combine import from/name/star into one emit function. Change in code size is: bare-arm: +4 minimal x86: -88 unix x64: -456 unix nanbox: -88 stm32: -44 cc3200: +0 esp8266: -104 esp32: +8
2018-05-22 12:58:25 +01:00
emit_native_import,
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit_native_load_const_tok,
emit_native_load_const_small_int,
emit_native_load_const_str,
py: Add load_const_obj to emitter, add LOAD_CONST_OBJ to bytecode. This allows to directly load a Python object to the Python stack. See issue #722 for background.
2015-01-13 15:55:54 +00:00
emit_native_load_const_obj,
py: Making closures now passes pointer to stack, not a tuple for vars. Closed over variables are now passed on the stack, instead of creating a tuple and passing that. This way memory for the closed over variables can be allocated within the closure object itself. See issue #510 for background.
2014-04-20 17:50:40 +01:00
emit_native_load_null,
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit_native_load_method,
emit_native_load_build_class,
py/emit: Combine load/store/delete subscr into one emit function. Reduces code size by: bare-arm: -8 minimal x86: -104 unix x64: -312 unix nanbox: -120 stm32: -60 cc3200: -16 esp8266: -92 esp32: -24
2018-05-22 12:31:56 +01:00
emit_native_subscr,
py/emit: Combine load/store/delete attr into one emit function. Reduces code size by: bare-arm: -20 minimal x86: -140 unix x64: -408 unix nanbox: -140 stm32: -68 cc3200: -16 esp8266: -80 esp32: -32
2018-05-22 12:43:41 +01:00
emit_native_attr,
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit_native_dup_top,
emit_native_dup_top_two,
emit_native_pop_top,
emit_native_rot_two,
emit_native_rot_three,
emit_native_jump,
py: Combine emit functions for jump true/false to reduce code size. Saves 116 bytes for stmhal and 56 bytes for cc3200 port.
2015-02-28 15:04:06 +00:00
emit_native_pop_jump_if,
emit_native_jump_if_or_pop,
py/emit: Combine break_loop and continue_loop into one emit function. Reduces code size by: bare-arm: +0 minimal x86: +0 unix x64: -80 unix nanbox: +0 stm32: -12 cc3200: +0 esp8266: -28 esp32: +0
2018-05-22 12:50:22 +01:00
emit_native_unwind_jump,
py/emit: Combine setup with/except/finally into one emit function. This patch reduces code size by: bare-arm: -16 minimal x86: -156 unix x64: -288 unix nanbox: -184 stm32: -48 cc3200: -16 esp8266: -96 esp32: -16 The last 10 patches combined reduce code size by: bare-arm: -164 minimal x86: -1260 unix x64: -3416 unix nanbox: -1616 stm32: -676 cc3200: -232 esp8266: -1144 esp32: -268
2018-05-22 13:33:26 +01:00
emit_native_setup_block,
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit_native_with_cleanup,
emit_native_end_finally,
emit_native_get_iter,
emit_native_for_iter,
emit_native_for_iter_end,
py: Replace POP_BLOCK and POP_EXCEPT opcodes with POP_EXCEPT_JUMP. POP_BLOCK and POP_EXCEPT are now the same, and are always followed by a JUMP. So this optimisation reduces code size, and RAM usage of bytecode by two bytes for each try-except handler.
2019-02-15 01:18:59 +00:00
emit_native_pop_except_jump,
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit_native_unary_op,
emit_native_binary_op,
py/emit: Combine build tuple/list/map emit funcs into one. Reduces code size by: bare-arm: -24 minimal x86: -192 unix x64: -288 unix nanbox: -184 stm32: -72 cc3200: -16 esp8266: -148 esp32: -32
2018-05-18 15:41:40 +01:00
emit_native_build,
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit_native_store_map,
py: Combine 3 comprehension emit functions (list/dict/set) into 1. The 3 kinds of comprehensions are similar enough that merging their emit functions reduces code size. Decreases in code size in bytes are: bare-arm:24, minimal:96, unix(NDEBUG,x86-64):328, stmhal:80, esp8266:76.
2016-09-18 14:59:47 +01:00
emit_native_store_comp,
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
emit_native_unpack_sequence,
emit_native_unpack_ex,
emit_native_make_function,
emit_native_make_closure,
emit_native_call_function,
emit_native_call_method,
emit_native_return_value,
emit_native_raise_varargs,
py/emit: Combine yield value and yield-from emit funcs into one. Reduces code size by: bare-arm: -24 minimal x86: -72 unix x64: -200 unix nanbox: -72 stm32: -52 cc3200: -32 esp8266: -84 esp32: -24
2018-05-18 15:30:42 +01:00
emit_native_yield,
py: Improvements to native emitter. Native emitter can now compile try/except blocks using nlr_push/nlr_pop. It probably only works for 1 level of exception handling. It doesn't work on Thumb (only x64). Native emitter can also handle some additional op codes. With this patch, 198 tests now pass using "-X emit=native" option to micropython.
2014-06-30 05:17:25 +01:00
emit_native_start_except_handler,
emit_native_end_except_handler,
Native Python and Viper support for x64 and thumb all together.
2013-10-08 09:05:10 +01:00
};
py: Add support for emitting native x86 machine code.
2014-09-06 23:06:36 +01:00
#endif