Merge pull request #6397 from jepler/extra-memset

Some flash size optimizations related to string0.c (implementation of str/mem functions)

Author: dhalbert

py/gc.c

@@ -138,7 +138,6 @@ void gc_init(void *start, void *end) {
     MP_STATE_MEM(gc_alloc_table_start) = (byte *)start;
 
     #if MICROPY_ENABLE_FINALISER
-    size_t gc_finaliser_table_byte_len = (MP_STATE_MEM(gc_alloc_table_byte_len) * BLOCKS_PER_ATB + BLOCKS_PER_FTB - 1) / BLOCKS_PER_FTB;
     MP_STATE_MEM(gc_finaliser_table_start) = MP_STATE_MEM(gc_alloc_table_start) + MP_STATE_MEM(gc_alloc_table_byte_len) + 1;
     #endif
 
@@ -147,18 +146,16 @@ void gc_init(void *start, void *end) {
     MP_STATE_MEM(gc_pool_end) = end;
 
     #if MICROPY_ENABLE_FINALISER
+    size_t gc_finaliser_table_byte_len = (MP_STATE_MEM(gc_alloc_table_byte_len) * BLOCKS_PER_ATB + BLOCKS_PER_FTB - 1) / BLOCKS_PER_FTB;
+    (void)gc_finaliser_table_byte_len; // avoid unused variable diagnostic if asserts are disabled
     assert(MP_STATE_MEM(gc_pool_start) >= MP_STATE_MEM(gc_finaliser_table_start) + gc_finaliser_table_byte_len);
     #endif
 
-    // Clear ATBs plus one more byte. The extra byte might be read when we read the final ATB and
-    // then try to count its tail. Clearing the byte ensures it is 0 and ends the chain. Without an
-    // FTB, it'll just clear the pool byte early.
-    memset(MP_STATE_MEM(gc_alloc_table_start), 0, MP_STATE_MEM(gc_alloc_table_byte_len) + 1);
-
-    #if MICROPY_ENABLE_FINALISER
-    // clear FTBs
-    memset(MP_STATE_MEM(gc_finaliser_table_start), 0, gc_finaliser_table_byte_len);
-    #endif
+    // Clear ATBs & finalisers (if enabled). This also clears the extra byte
+    // which appears between ATBs and finalisers that ensures every chain in
+    // the ATB terminates, rather than erroneously using bits from the
+    // finalisers.
+    memset(MP_STATE_MEM(gc_alloc_table_start), 0, MP_STATE_MEM(gc_pool_start) - MP_STATE_MEM(gc_alloc_table_start));
 
     // Set first free ATB index to the start of the heap.
     for (size_t i = 0; i < MICROPY_ATB_INDICES; i++) {

shared/libc/string0.c

@@ -28,13 +28,16 @@
 #include <stddef.h>
 #include <string.h>
 
+#include "py/mpconfig.h"
+
 #ifndef likely
 #define likely(x) __builtin_expect((x), 1)
 #endif
 
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wcast-align"
 void *memcpy(void *dst, const void *src, size_t n) {
+#if CIRCUITPY_FULL_BUILD
     if (likely(!(((uintptr_t)dst) & 3) && !(((uintptr_t)src) & 3))) {
         // pointers aligned
         uint32_t *d = dst;
@@ -56,7 +59,9 @@ void *memcpy(void *dst, const void *src, size_t n) {
             // copy byte
             *((uint8_t*)d) = *((const uint8_t*)s);
         }
-    } else {
+    } else
+#endif
+    {
         // unaligned access, copy bytes
         uint8_t *d = dst;
         const uint8_t *s = src;
@@ -93,6 +98,7 @@ void *memmove(void *dest, const void *src, size_t n) {
 }
 
 void *memset(void *s, int c, size_t n) {
+#if CIRCUITPY_FULL_BUILD
     if (c == 0 && ((uintptr_t)s & 3) == 0) {
         // aligned store of 0
         uint32_t *s32 = s;
@@ -106,7 +112,9 @@ void *memset(void *s, int c, size_t n) {
         if (n & 1) {
             *((uint8_t*)s32) = 0;
         }
-    } else {
+    } else
+#endif
+    {
         uint8_t *s2 = s;
         for (; n > 0; n--) {
             *s2++ = c;