Half-Life-RTX/xash3d-fwgs
1
0
Fork 0
mirror of https://github.com/w23/xash3d-fwgs synced 2024-12-14 21:20:26 +01:00
Code Issues Projects Releases Wiki Activity

vk: profiler: devmem: take into account alignment holes

Browse Source 
Also improve `VK_DevMemFree` debug printing: include usage type, size,
alignment, and unusable alignment hole size.
This commit is contained in:
nilsoncore 2023-09-25 23:43:45 +03:00
parent 985ebd23c2
commit 95223077b6
2 changed files with 80 additions and 44 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

47
ref/vk/vk_devmem.c
View File

@ -31,6 +31,10 @@ typedef struct vk_devmem_allocation_stats_s {
int allocated_total; // Total size of allocated memory.
int frees_total; // Total number of memory deallocations (frees).
int freed_total; // Total size of deallocated (freed) memory.
int align_holes_current; // Current number of alignment holes in active (not freed) allocations.
int align_holes_size_current; // Current size of alignment holes in active (not freed) allocations.
int align_holes_total; // Total number of alignment holes in all of allocations made.
int align_holes_size_total; // Total size of alignment holes in all of allocations made.
} vk_devmem_allocation_stats_t;
static struct {
@ -49,28 +53,47 @@ static struct {
// Register allocation in overall stats and for the corresponding type too.
// This is "scalable" approach, which can be simplified if needed.
#define REGISTER_ALLOCATION( type, size ) \
#define REGISTER_ALLOCATION( type, size, alignment ) { \
g_vk_devmem.stats[VK_DEVMEM_USAGE_TYPE_ALL].allocations_current += 1; \
g_vk_devmem.stats[VK_DEVMEM_USAGE_TYPE_ALL].allocated_current += size; \
g_vk_devmem.stats[VK_DEVMEM_USAGE_TYPE_ALL].allocations_total += 1; \
g_vk_devmem.stats[VK_DEVMEM_USAGE_TYPE_ALL].allocated_total += size; \
int alignment_hole = size % alignment; \
if ( alignment_hole > 0 ) { \
g_vk_devmem.stats[VK_DEVMEM_USAGE_TYPE_ALL].align_holes_current += 1; \
g_vk_devmem.stats[VK_DEVMEM_USAGE_TYPE_ALL].align_holes_size_current += alignment_hole; \
g_vk_devmem.stats[VK_DEVMEM_USAGE_TYPE_ALL].align_holes_total += 1; \
g_vk_devmem.stats[VK_DEVMEM_USAGE_TYPE_ALL].align_holes_size_total += alignment_hole; \
} \
for ( int type_idx = VK_DEVMEM_USAGE_TYPE_ALL + 1; type_idx < VK_DEVMEM_USAGE_TYPES_COUNT; type_idx += 1 ) { \
if ( type_idx == type ) { \
g_vk_devmem.stats[type_idx].allocations_current += 1; \
g_vk_devmem.stats[type_idx].allocated_current += size; \
g_vk_devmem.stats[type_idx].allocations_total += 1; \
g_vk_devmem.stats[type_idx].allocated_total += size; \
if ( alignment_hole > 0 ) { \
g_vk_devmem.stats[type_idx].align_holes_current += 1; \
g_vk_devmem.stats[type_idx].align_holes_size_current += alignment_hole; \
g_vk_devmem.stats[type_idx].align_holes_total += 1; \
g_vk_devmem.stats[type_idx].align_holes_size_total += alignment_hole; \
} \
break; \
} \
}
} \
}
// Register deallocation (freeing) in overall stats and for the corresponding type too.
// This is "scalable" approach, which can be simplified if needed.
#define REGISTER_FREE( type, size ) \
#define REGISTER_FREE( type, size, alignment ) \
g_vk_devmem.stats[VK_DEVMEM_USAGE_TYPE_ALL].allocations_current -= 1; \
g_vk_devmem.stats[VK_DEVMEM_USAGE_TYPE_ALL].allocated_current -= size; \
g_vk_devmem.stats[VK_DEVMEM_USAGE_TYPE_ALL].frees_total += 1; \
g_vk_devmem.stats[VK_DEVMEM_USAGE_TYPE_ALL].freed_total += size; \
int alignment_hole = size % alignment; \
if ( alignment_hole > 0 ) { \
g_vk_devmem.stats[VK_DEVMEM_USAGE_TYPE_ALL].align_holes_current -= 1; \
g_vk_devmem.stats[VK_DEVMEM_USAGE_TYPE_ALL].align_holes_size_current -= alignment_hole; \
} \
for ( int type_idx = VK_DEVMEM_USAGE_TYPE_ALL + 1; type_idx < VK_DEVMEM_USAGE_TYPES_COUNT; type_idx += 1 ) { \
if ( type_idx == type ) { \
g_vk_devmem.stats[type_idx].allocations_current -= 1; \
@ -78,6 +101,10 @@ static struct {
g_vk_devmem.stats[type_idx].frees_total += 1; \
g_vk_devmem.stats[type_idx].freed_total += size; \
break; \
if ( alignment_hole > 0 ) { \
g_vk_devmem.stats[type_idx].align_holes_current -= 1; \
g_vk_devmem.stats[type_idx].align_holes_size_current -= alignment_hole; \
} \
} \
}
@ -293,8 +320,9 @@ vk_devmem_t VK_DevMemAllocate(const char *name, vk_devmem_usage_type_t usage_typ
devmem._slot_index = slot_index;
devmem._block_index = block.index;
devmem._block_size = block.size;
devmem._block_alignment = req.alignment;
REGISTER_ALLOCATION( usage_type, block.size );
REGISTER_ALLOCATION( usage_type, block.size, req.alignment );
return devmem;
}
@ -309,12 +337,15 @@ void VK_DevMemFree(const vk_devmem_t *mem) {
ASSERT( mem->device_memory == slot->device_memory );
if ( g_vk_devmem.verbose ) {
gEngine.Con_Reportf( "^2VK_DevMemFree:^7 { slot: %d, block: %d }\n", slot_index, mem->_block_index );
const char *usage_type = VK_DevMemUsageTypeString( mem->usage_type );
int align_hole = mem->_block_size % mem->_block_alignment;
gEngine.Con_Reportf( "^2VK_DevMemFree:^7 { slot: %d, block: %d, usage: %s, size: %d, alignment: %d, hole: %d }\n",
slot_index, mem->_block_index, usage_type, mem->_block_size, mem->_block_alignment, align_hole );
}
aloPoolFree( slot->allocator, mem->_block_index );
REGISTER_FREE( mem->usage_type, mem->_block_size );
REGISTER_FREE( mem->usage_type, mem->_block_size, mem->_block_alignment );
slot->refcount--;
}
@ -337,6 +368,10 @@ void VK_DevMemFree(const vk_devmem_t *mem) {
REGISTER_STATS_METRIC( g_vk_devmem.stats[usage_type].allocated_total, STRING( allocated_total##usage_suffix ), STRING( g_vk_devmem.stats[usage_suffix].allocated_total ), kSpeedsMetricBytes ); \
REGISTER_STATS_METRIC( g_vk_devmem.stats[usage_type].frees_total, STRING( frees_total##usage_suffix ), STRING( g_vk_devmem.stats[usage_suffix].frees_total ), kSpeedsMetricCount ); \
REGISTER_STATS_METRIC( g_vk_devmem.stats[usage_type].freed_total, STRING( freed_total##usage_suffix ), STRING( g_vk_devmem.stats[usage_suffix].freed_total ), kSpeedsMetricBytes ); \
REGISTER_STATS_METRIC( g_vk_devmem.stats[usage_type].align_holes_current, STRING( align_holes_current##usage_suffix ), STRING( g_vk_devmem.stats[usage_suffix].align_holes_current ), kSpeedsMetricCount ); \
REGISTER_STATS_METRIC( g_vk_devmem.stats[usage_type].align_holes_size_current, STRING( align_holes_size_current##usage_suffix ), STRING( g_vk_devmem.stats[usage_suffix].align_holes_size_current ), kSpeedsMetricBytes ); \
REGISTER_STATS_METRIC( g_vk_devmem.stats[usage_type].align_holes_total, STRING( align_holes_total##usage_suffix ), STRING( g_vk_devmem.stats[usage_suffix].align_holes_total ), kSpeedsMetricCount ); \
REGISTER_STATS_METRIC( g_vk_devmem.stats[usage_type].align_holes_size_total, STRING( align_holes_size_total##usage_suffix ), STRING( g_vk_devmem.stats[usage_suffix].align_holes_size_total ), kSpeedsMetricBytes ); \
}
qboolean VK_DevMemInit( void ) {

1
ref/vk/vk_devmem.h
View File

@ -31,6 +31,7 @@ typedef struct vk_devmem_s {
int _slot_index;
int _block_index;
int _block_size;
int _block_alignment;
} vk_devmem_t;
typedef struct vk_devmem_allocate_args_s {