micropython · jimmo · Jan 25, 2024 · Jan 25, 2024 · Jan 25, 2024 · Jan 25, 2024
diff --git a/ports/bare-arm/Makefile b/ports/bare-arm/Makefile
@@ -16,7 +16,6 @@ PYDFU ?= $(TOP)/tools/pydfu.py
 CFLAGS += -I. -I$(TOP) -I$(BUILD)
 CFLAGS += -Wall -Werror -std=c99 -nostdlib
 CFLAGS += -mthumb -mtune=cortex-m4 -mcpu=cortex-m4 -msoft-float
-CSUPEROPT = -Os # save some code space for performance-critical code
 
 # Select debugging or optimisation build.
 ifeq ($(DEBUG), 1)

diff --git a/ports/bare-arm/mpconfigport.h b/ports/bare-arm/mpconfigport.h
@@ -37,6 +37,9 @@
 // Python internal features
 #define MICROPY_ERROR_REPORTING                 (MICROPY_ERROR_REPORTING_NONE)
 
+// Just use -Os for everything to generate the smallest possible binary.
+#define MICROPY_APPLY_COMPILER_EXTRA_OPTIMISATIONS(f) f
+
 // Type definitions for the specific machine
 
 typedef int32_t mp_int_t; // must be pointer size

diff --git a/ports/esp32/mpconfigport.h b/ports/esp32/mpconfigport.h
@@ -231,18 +231,12 @@ void *esp_native_code_commit(void *, size_t, void *);
 #endif
 
 // Functions that should go in IRAM
+#define MICROPY_PERFORMANCE_CRITICAL_LEVEL_1(f) IRAM_ATTR MICROPY_APPLY_COMPILER_OPTIMISATIONS(f)
+#if !(CONFIG_IDF_TARGET_ESP32 && CONFIG_SPIRAM && CONFIG_SPIRAM_CACHE_WORKAROUND)
 // For ESP32 with SPIRAM workaround, firmware is larger and uses more static IRAM,
 // so in that configuration don't put too many functions in IRAM.
-#if !(CONFIG_IDF_TARGET_ESP32 && CONFIG_SPIRAM && CONFIG_SPIRAM_CACHE_WORKAROUND)
-#define MICROPY_WRAP_MP_BINARY_OP(f) IRAM_ATTR f
+#define MICROPY_PERFORMANCE_CRITICAL_LEVEL_2(f) IRAM_ATTR MICROPY_APPLY_COMPILER_OPTIMISATIONS(f)
 #endif
-#define MICROPY_WRAP_MP_EXECUTE_BYTECODE(f) IRAM_ATTR f
-#define MICROPY_WRAP_MP_LOAD_GLOBAL(f) IRAM_ATTR f
-#define MICROPY_WRAP_MP_LOAD_NAME(f) IRAM_ATTR f
-#define MICROPY_WRAP_MP_MAP_LOOKUP(f) IRAM_ATTR f
-#define MICROPY_WRAP_MP_OBJ_GET_TYPE(f) IRAM_ATTR f
-#define MICROPY_WRAP_MP_SCHED_EXCEPTION(f) IRAM_ATTR f
-#define MICROPY_WRAP_MP_SCHED_KEYBOARD_INTERRUPT(f) IRAM_ATTR f
 
 #define UINT_FMT "%u"
 #define INT_FMT "%d"

diff --git a/ports/minimal/Makefile b/ports/minimal/Makefile
@@ -31,8 +31,6 @@ CFLAGS += $(INC) -Wall -Werror -Wdouble-promotion -Wfloat-conversion -std=c99 $(
 LDFLAGS += -Wl,-Map=$@.map,--cref -Wl,--gc-sections
 endif
 
-CSUPEROPT = -Os # save some code space
-
 # Tune for Debugging or Optimization
 CFLAGS += -g  # always include debug info in the ELF
 ifeq ($(DEBUG), 1)

diff --git a/ports/minimal/mpconfigport.h b/ports/minimal/mpconfigport.h
@@ -21,6 +21,9 @@
 // Use the minimum headroom in the chunk allocator for parse nodes.
 #define MICROPY_ALLOC_PARSE_CHUNK_INIT    (16)
 
+// Just use -Os for everything to generate the smallest possible binary.
+#define MICROPY_APPLY_COMPILER_EXTRA_OPTIMISATIONS(f) f
+
 // type definitions for the specific machine
 
 typedef intptr_t mp_int_t; // must be pointer size

diff --git a/ports/stm32/Makefile b/ports/stm32/Makefile
@@ -284,7 +284,6 @@ SRC_O += \
 	$(SYSTEM_FILE)
 
 ifeq ($(MCU_SERIES),$(filter $(MCU_SERIES),f0 g0 l0))
-CSUPEROPT = -Os # save some code space
 SRC_O += \
 	resethandler_m0.o \
 	shared/runtime/gchelper_thumb1.o

diff --git a/ports/stm32/mpconfigport.h b/ports/stm32/mpconfigport.h
@@ -55,6 +55,12 @@
 #define MICROPY_OPT_MAP_LOOKUP_CACHE (__CORTEX_M > 0)
 #endif
 
+#if __CORTEX_M == 0
+// Just use -Os for everything to avoid using extra flash. Using M0 as a stand-in
+// for "likely has small flash", this could be moved to the board config instead.
+#define MICROPY_APPLY_COMPILER_EXTRA_OPTIMISATIONS(f) f
+#endif
+
 // emitters
 #define MICROPY_PERSISTENT_CODE_LOAD (1)
 #ifndef MICROPY_EMIT_THUMB

diff --git a/py/gc.c b/py/gc.c
@@ -723,7 +723,12 @@ void gc_info(gc_info_t *info) {
     GC_EXIT();
 }
 
-void *gc_alloc(size_t n_bytes, unsigned int alloc_flags) {
+#ifndef MICROPY_WRAP_GC_ALLOC
+// Optimising gc for speed; 5ms down to 4ms on pybv2
+#define MICROPY_WRAP_GC_ALLOC(f) MICROPY_PERFORMANCE_CRITICAL_LEVEL_3(f)
+#endif
+
+void *MICROPY_WRAP_GC_ALLOC(gc_alloc)(size_t n_bytes, unsigned int alloc_flags) {
     bool has_finaliser = alloc_flags & GC_ALLOC_FLAG_HAS_FINALISER;
     size_t n_blocks = ((n_bytes + BYTES_PER_BLOCK - 1) & (~(BYTES_PER_BLOCK - 1))) / BYTES_PER_BLOCK;
     DEBUG_printf("gc_alloc(" UINT_FMT " bytes -> " UINT_FMT " blocks)\n", n_bytes, n_blocks);
@@ -890,9 +895,13 @@ void *gc_alloc_with_finaliser(mp_uint_t n_bytes) {
 }
 */
 
+#ifndef MICROPY_WRAP_GC_FREE
+#define MICROPY_WRAP_GC_FREE(f) MICROPY_PERFORMANCE_CRITICAL_LEVEL_3(f)
+#endif
+
 // force the freeing of a piece of memory
 // TODO: freeing here does not call finaliser
-void gc_free(void *ptr) {
+void MICROPY_WRAP_GC_FREE(gc_free)(void *ptr) {
     if (MP_STATE_THREAD(gc_lock_depth) > 0) {
         // Cannot free while the GC is locked. However free is an optimisation
         // to reclaim the memory immediately, this means it will now be left
@@ -1021,7 +1030,11 @@ void *gc_realloc(void *ptr, mp_uint_t n_bytes) {
 
 #else // Alternative gc_realloc impl
 
-void *gc_realloc(void *ptr_in, size_t n_bytes, bool allow_move) {
+#ifndef MICROPY_WRAP_GC_REALLOC
+#define MICROPY_WRAP_GC_REALLOC(f) MICROPY_PERFORMANCE_CRITICAL_LEVEL_3(f)
+#endif
+
+void *MICROPY_WRAP_GC_REALLOC(gc_realloc)(void *ptr_in, size_t n_bytes, bool allow_move) {
     // check for pure allocation
     if (ptr_in == NULL) {
         return gc_alloc(n_bytes, false);

diff --git a/py/map.c b/py/map.c
@@ -147,6 +147,10 @@ STATIC void mp_map_rehash(mp_map_t *map) {
     m_del(mp_map_elem_t, old_table, old_alloc);
 }
 
+#ifndef MICROPY_WRAP_MP_MAP_LOOKUP
+#define MICROPY_WRAP_MP_MAP_LOOKUP(f) MICROPY_PERFORMANCE_CRITICAL_LEVEL_1(f)
+#endif
+
 // MP_MAP_LOOKUP behaviour:
 //  - returns NULL if not found, else the slot it was found in with key,value non-null
 // MP_MAP_LOOKUP_ADD_IF_NOT_FOUND behaviour:

diff --git a/py/mpconfig.h b/py/mpconfig.h
@@ -1784,46 +1784,39 @@ typedef double mp_float_t;
 #endif
 
 /*****************************************************************************/
-/* Hooks for a port to wrap functions with attributes                        */
+/* Hooks for a port to wrap functions with performance-tuning attributes     */
 
-#ifndef MICROPY_WRAP_MP_BINARY_OP
-#define MICROPY_WRAP_MP_BINARY_OP(f) f
-#endif
-
-#ifndef MICROPY_WRAP_MP_EXECUTE_BYTECODE
-#define MICROPY_WRAP_MP_EXECUTE_BYTECODE(f) f
-#endif
-
-#ifndef MICROPY_WRAP_MP_LOAD_GLOBAL
-#define MICROPY_WRAP_MP_LOAD_GLOBAL(f) f
-#endif
-
-#ifndef MICROPY_WRAP_MP_LOAD_NAME
-#define MICROPY_WRAP_MP_LOAD_NAME(f) f
-#endif
-
-#ifndef MICROPY_WRAP_MP_MAP_LOOKUP
-#define MICROPY_WRAP_MP_MAP_LOOKUP(f) f
+#ifndef MICROPY_APPLY_COMPILER_OPTIMISATIONS
+#if defined(__GNUC__) && !defined(__clang__)
+// Enable -O3 optimisations.
+#define MICROPY_APPLY_COMPILER_OPTIMISATIONS(f) __attribute__((optimize("O2"))) f
+#else
+// Unsupported on other compilers, will use global optimisation setting (typically -Os).
+#define MICROPY_APPLY_COMPILER_OPTIMISATIONS(f) f
 #endif
-
-#ifndef MICROPY_WRAP_MP_OBJ_GET_TYPE
-#define MICROPY_WRAP_MP_OBJ_GET_TYPE(f) f
 #endif
 
-#ifndef MICROPY_WRAP_MP_SCHED_EXCEPTION
-#define MICROPY_WRAP_MP_SCHED_EXCEPTION(f) f
+// Ideally apply full compiler optimisations and place in RAM.
+// Use this on small functions that need the highest possible performance.
+#ifndef MICROPY_PERFORMANCE_CRITICAL_LEVEL_1
+#define MICROPY_PERFORMANCE_CRITICAL_LEVEL_1(f) MICROPY_APPLY_COMPILER_OPTIMISATIONS(f)
 #endif
 
-#ifndef MICROPY_WRAP_MP_SCHED_KEYBOARD_INTERRUPT
-#define MICROPY_WRAP_MP_SCHED_KEYBOARD_INTERRUPT(f) f
+// Ideally apply full compiler optimisations and optionally place in RAM (if IRAM available).
+// Use this on larger functions that should go in RAM if possible.
+#ifndef MICROPY_PERFORMANCE_CRITICAL_LEVEL_2
+#define MICROPY_PERFORMANCE_CRITICAL_LEVEL_2(f) MICROPY_APPLY_COMPILER_OPTIMISATIONS(f)
 #endif
 
-#ifndef MICROPY_WRAP_MP_SCHED_SCHEDULE
-#define MICROPY_WRAP_MP_SCHED_SCHEDULE(f) f
+// Ideally apply full compiler optimisation if flash available.
+// Use this on functions that are not important enough to place in RAM.
+#ifndef MICROPY_PERFORMANCE_CRITICAL_LEVEL_3
+#define MICROPY_PERFORMANCE_CRITICAL_LEVEL_3(f) MICROPY_APPLY_COMPILER_OPTIMISATIONS(f)
 #endif
 
-#ifndef MICROPY_WRAP_MP_SCHED_VM_ABORT
-#define MICROPY_WRAP_MP_SCHED_VM_ABORT(f) f
+// Ideally apply full compiler optimisation if flash available (but lower priority than level 3).
+#ifndef MICROPY_PERFORMANCE_CRITICAL_LEVEL_4
+#define MICROPY_PERFORMANCE_CRITICAL_LEVEL_4(f) f
 #endif
 
 /*****************************************************************************/