py: Improve memory usage debugging; better GC AT dumping.
In unix port, mem_info(1) now prints pretty GC alloc table.
diff --git a/py/gc.c b/py/gc.c
index 8e71307..ce2fa3a 100644
--- a/py/gc.c
+++ b/py/gc.c
@@ -492,7 +492,7 @@
}
}
-mp_uint_t gc_nbytes(void *ptr_in) {
+mp_uint_t gc_nbytes(const void *ptr_in) {
mp_uint_t ptr = (mp_uint_t)ptr_in;
if (VERIFY_PTR(ptr)) {
@@ -681,31 +681,32 @@
for (mp_uint_t bl = 0; bl < gc_alloc_table_byte_len * BLOCKS_PER_ATB; bl++) {
if (bl % DUMP_BYTES_PER_LINE == 0) {
// a new line of blocks
- #if EXTENSIVE_HEAP_PROFILING
{
// check if this line contains only free blocks
- bool only_free_blocks = true;
- for (mp_uint_t bl2 = bl; bl2 < gc_alloc_table_byte_len * BLOCKS_PER_ATB && bl2 < bl + DUMP_BYTES_PER_LINE; bl2++) {
- if (ATB_GET_KIND(bl2) != AT_FREE) {
-
- only_free_blocks = false;
+ mp_uint_t bl2 = bl;
+ while (bl2 < gc_alloc_table_byte_len * BLOCKS_PER_ATB && ATB_GET_KIND(bl2) == AT_FREE) {
+ bl2++;
+ }
+ if (bl2 - bl >= 2 * DUMP_BYTES_PER_LINE) {
+ // there are at least 2 lines containing only free blocks, so abbreviate their printing
+ printf("\n (" UINT_FMT " lines all free)", (bl2 - bl) / DUMP_BYTES_PER_LINE);
+ bl = bl2 & (~(DUMP_BYTES_PER_LINE - 1));
+ if (bl >= gc_alloc_table_byte_len * BLOCKS_PER_ATB) {
+ // got to end of heap
break;
}
}
- if (only_free_blocks) {
- // line contains only free blocks, so skip printing it
- bl += DUMP_BYTES_PER_LINE - 1;
- continue;
- }
}
- #endif
// print header for new line of blocks
- printf("\n%04x: ", (uint)bl);
+ #if EXTENSIVE_HEAP_PROFILING
+ printf("\n%05x: ", (uint)(bl * BYTES_PER_BLOCK) & 0xfffff);
+ #else
+ printf("\n%05x: ", (uint)PTR_FROM_BLOCK(bl) & 0xfffff);
+ #endif
}
int c = ' ';
switch (ATB_GET_KIND(bl)) {
case AT_FREE: c = '.'; break;
- case AT_HEAD: c = 'h'; break;
/* this prints out if the object is reachable from BSS or STACK (for unix only)
case AT_HEAD: {
extern char __bss_start, _end;
@@ -734,7 +735,7 @@
break;
}
*/
- /* this prints the uPy object type of the head block
+ /* this prints the uPy object type of the head block */
case AT_HEAD: {
mp_uint_t *ptr = gc_pool_start + bl * WORDS_PER_BLOCK;
if (*ptr == (mp_uint_t)&mp_type_tuple) { c = 'T'; }
@@ -742,10 +743,10 @@
else if (*ptr == (mp_uint_t)&mp_type_dict) { c = 'D'; }
else if (*ptr == (mp_uint_t)&mp_type_float) { c = 'F'; }
else if (*ptr == (mp_uint_t)&mp_type_fun_bc) { c = 'B'; }
+ else if (*ptr == (mp_uint_t)&mp_type_module) { c = 'M'; }
else { c = 'h'; }
break;
}
- */
case AT_TAIL: c = 't'; break;
case AT_MARK: c = 'm'; break;
}
diff --git a/py/gc.h b/py/gc.h
index fd48d26..dc276dd 100644
--- a/py/gc.h
+++ b/py/gc.h
@@ -40,7 +40,7 @@
void *gc_alloc(mp_uint_t n_bytes, bool has_finaliser);
void gc_free(void *ptr);
-mp_uint_t gc_nbytes(void *ptr);
+mp_uint_t gc_nbytes(const void *ptr);
void *gc_realloc(void *ptr, mp_uint_t n_bytes);
typedef struct _gc_info_t {
diff --git a/py/qstr.c b/py/qstr.c
index f841f1d..c2cfda8 100644
--- a/py/qstr.c
+++ b/py/qstr.c
@@ -30,6 +30,7 @@
#include "mpconfig.h"
#include "misc.h"
#include "qstr.h"
+#include "gc.h"
// NOTE: we are using linear arrays to store and search for qstr's (unique strings, interned strings)
// ultimately we will replace this with a static hash table of some kind
@@ -220,9 +221,17 @@
*n_pool += 1;
*n_qstr += pool->len;
for (const byte **q = pool->qstrs, **q_top = pool->qstrs + pool->len; q < q_top; q++) {
+ #if MICROPY_ENABLE_GC
+ *n_str_data_bytes += gc_nbytes(*q); // this counts actual bytes used in heap
+ #else
*n_str_data_bytes += Q_GET_ALLOC(*q);
+ #endif
}
+ #if MICROPY_ENABLE_GC
+ *n_total_bytes += gc_nbytes(pool); // this counts actual bytes used in heap
+ #else
*n_total_bytes += sizeof(qstr_pool_t) + sizeof(qstr) * pool->alloc;
+ #endif
}
*n_total_bytes += *n_str_data_bytes;
}