Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

feat: add dedicated staging uncached allocator back & other fixes #83

Merged
merged 1 commit into from
Jan 23, 2025
Merged

feat: add dedicated staging uncached allocator back & other fixes #83

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 5 additions & 2 deletions extensions/pl_debug_ext.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -231,12 +231,13 @@ pl__show_memory_allocations(bool* bValue)

size_t szOriginalAllocationCount = 0;
plAllocationEntry* atAllocations = gptMemory->get_allocations(&szOriginalAllocationCount);

plUiClipper tClipper = {(uint32_t)szOriginalAllocationCount};
while(gptUI->step_clipper(&tClipper))
{
for(uint32_t i = tClipper.uDisplayStart; i < tClipper.uDisplayEnd; i++)
{
size_t szUnused = 0;
atAllocations = gptMemory->get_allocations(&szUnused);
plAllocationEntry tEntry = atAllocations[i];
strncpy(pcFile, tEntry.pcFile, 1024);
gptUI->text("%i", i);
Expand Down Expand Up @@ -878,6 +879,7 @@ pl__show_device_memory(bool* bValue)
const plDeviceMemoryAllocatorI atAllocators[] = {
*gptGpuAllocators->get_local_buddy_allocator(gptDebugCtx->ptDevice),
*gptGpuAllocators->get_local_dedicated_allocator(gptDebugCtx->ptDevice),
*gptGpuAllocators->get_staging_uncached_buddy_allocator(gptDebugCtx->ptDevice),
*gptGpuAllocators->get_staging_uncached_allocator(gptDebugCtx->ptDevice),
*gptGpuAllocators->get_staging_cached_allocator(gptDebugCtx->ptDevice)
};
Expand All @@ -886,13 +888,14 @@ pl__show_device_memory(bool* bValue)
"Device Memory: Local Buddy",
"Device Memory: Local Dedicated",
"Device Memory: Staging Uncached Buddy",
"Device Memory: Staging Uncached",
"Device Memory: Staging Cached"
};

gptUI->push_theme_color(PL_UI_COLOR_BUTTON, tButtonColor);
gptUI->push_theme_color(PL_UI_COLOR_BUTTON_ACTIVE, tButtonColor);
gptUI->push_theme_color(PL_UI_COLOR_BUTTON_HOVERED, tButtonColor);
for(uint32_t uAllocatorIndex = 0; uAllocatorIndex < 4; uAllocatorIndex++)
for(uint32_t uAllocatorIndex = 0; uAllocatorIndex < 5; uAllocatorIndex++)
{
uint32_t uBlockCount = 0;
uint32_t uRangeCount = 0;
Expand Down
128 changes: 121 additions & 7 deletions extensions/pl_gpu_allocators_ext.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -552,6 +552,75 @@ pl_free_buddy(struct plDeviceMemoryAllocatorO* ptInst, plDeviceMemoryAllocation*
strncpy(ptNode->acName, "not used", PL_MAX_NAME_LENGTH);
}

static plDeviceMemoryAllocation
pl_allocate_staging_uncached(
struct plDeviceMemoryAllocatorO* ptInst, uint32_t uTypeFilter,
uint64_t ulSize, uint64_t ulAlignment, const char* pcName)
{
plDeviceAllocatorData* ptData = (plDeviceAllocatorData*)ptInst;

plDeviceMemoryAllocation tBlock = gptGfx->allocate_memory(ptData->ptDevice, ulSize,
PL_MEMORY_GPU_CPU, uTypeFilter, pcName);

plDeviceMemoryAllocation tAllocation = {
.pHostMapped = tBlock.pHostMapped,
.uHandle = tBlock.uHandle,
.ulOffset = 0,
.ulSize = ulSize,
.ptAllocator = ptData->ptAllocator,
.tMemoryMode = PL_MEMORY_GPU_CPU
};

uint32_t uBlockIndex = pl_sb_size(ptData->sbtBlocks);
if(pl_sb_size(ptData->sbtFreeBlockIndices) > 0)
uBlockIndex = pl_sb_pop(ptData->sbtFreeBlockIndices);
else
pl_sb_add(ptData->sbtBlocks);

plDeviceAllocationRange tRange = {
.ulOffset = 0,
.ulTotalSize = ulSize,
.ulUsedSize = ulSize,
.ulBlockIndex = uBlockIndex
};
pl_sprintf(tRange.acName, "%s", pcName);

pl_sb_push(ptData->sbtNodes, tRange);
ptData->sbtBlocks[uBlockIndex] = tBlock;
return tAllocation;
}

static void
pl_free_staging_uncached(struct plDeviceMemoryAllocatorO* ptInst, plDeviceMemoryAllocation* ptAllocation)
{
plDeviceAllocatorData* ptData = (plDeviceAllocatorData*)ptInst;

uint32_t uBlockIndex = 0;
uint32_t uNodeIndex = 0;
for(uint32_t i = 0; i < pl_sb_size(ptData->sbtNodes); i++)
{
plDeviceAllocationRange* ptNode = &ptData->sbtNodes[i];
plDeviceMemoryAllocation* ptBlock = &ptData->sbtBlocks[ptNode->ulBlockIndex];

if(ptBlock->uHandle == ptAllocation->uHandle)
{
uNodeIndex = i;
uBlockIndex = (uint32_t)ptNode->ulBlockIndex;
ptBlock->ulSize = 0;
break;
}
}
pl_sb_del_swap(ptData->sbtNodes, uNodeIndex);
pl_sb_push(ptData->sbtFreeBlockIndices, uBlockIndex);

gptGfx->free_memory(ptData->ptDevice, &ptData->sbtBlocks[uBlockIndex]);

ptAllocation->pHostMapped = NULL;
ptAllocation->uHandle = 0;
ptAllocation->ulOffset = 0;
ptAllocation->ulSize = 0;
}

static plDeviceMemoryAllocation
pl_allocate_staging_uncached_buddy(
struct plDeviceMemoryAllocatorO* ptInst, uint32_t uTypeFilter,
Expand Down Expand Up @@ -763,6 +832,39 @@ pl_get_staging_uncached_allocator(plDevice* ptDevice)

ptAllocatorData = &gtAllocatorData;
ptAllocator = &gtAllocator;
}
}
ptAllocator->allocate = pl_allocate_staging_uncached;
ptAllocator->free = pl_free_staging_uncached;
return ptAllocator;
}

static plDeviceMemoryAllocatorI*
pl_get_staging_uncached_allocator_buddy(plDevice* ptDevice)
{
static plDeviceAllocatorData* ptAllocatorData = NULL;
static plDeviceMemoryAllocatorI* ptAllocator = NULL;

if(ptAllocator == NULL)
{
ptAllocator = gptDataRegistry->get_data("StagingUncachedAllocatorBuddy");

if(ptAllocator)
ptAllocatorData = gptDataRegistry->get_data("StagingUncachedAllocatorBuddyData");

else
{

static plDeviceMemoryAllocatorI gtAllocator = {0};
static plDeviceAllocatorData gtAllocatorData = {0};
gptDataRegistry->set_data("StagingUncachedAllocatorBuddy", &gtAllocator);
gptDataRegistry->set_data("StagingUncachedAllocatorBuddyData", &gtAllocatorData);
gtAllocatorData.ptDevice = ptDevice;
gtAllocatorData.ptAllocator = &gtAllocator;
gtAllocator.ptInst = (struct plDeviceMemoryAllocatorO*)&gtAllocatorData;

ptAllocatorData = &gtAllocatorData;
ptAllocator = &gtAllocator;

if(ptAllocatorData->auFreeList[0] == 0)
{
Expand Down Expand Up @@ -856,6 +958,17 @@ pl_cleanup_allocators(plDevice* ptDevice)
pl_sb_free(ptAllocatorData->sbtBlocks);
pl_sb_free(ptAllocatorData->sbtNodes);
pl_sb_free(ptAllocatorData->sbtFreeBlockIndices);

ptAllocator = pl_get_staging_uncached_allocator_buddy(ptDevice);
ptAllocatorData = (plDeviceAllocatorData*)ptAllocator->ptInst;
for(uint32_t i = 0; i < pl_sb_size(ptAllocatorData->sbtBlocks); i++)
{
if(ptAllocatorData->sbtBlocks[i].uHandle)
gptGfx->free_memory(ptDevice, &ptAllocatorData->sbtBlocks[i]);
}
pl_sb_free(ptAllocatorData->sbtBlocks);
pl_sb_free(ptAllocatorData->sbtNodes);
pl_sb_free(ptAllocatorData->sbtFreeBlockIndices);
}

//-----------------------------------------------------------------------------
Expand All @@ -866,13 +979,14 @@ PL_EXPORT void
pl_load_gpu_allocators_ext(plApiRegistryI* ptApiRegistry, bool bReload)
{
const plGPUAllocatorsI tApi = {
.get_local_dedicated_allocator = pl_get_local_dedicated_allocator,
.get_local_buddy_allocator = pl_get_local_buddy_allocator,
.get_staging_uncached_allocator = pl_get_staging_uncached_allocator,
.get_staging_cached_allocator = pl_get_staging_cached_allocator,
.get_blocks = pl_get_allocator_blocks,
.get_ranges = pl_get_allocator_ranges,
.cleanup = pl_cleanup_allocators
.get_local_dedicated_allocator = pl_get_local_dedicated_allocator,
.get_local_buddy_allocator = pl_get_local_buddy_allocator,
.get_staging_uncached_allocator = pl_get_staging_uncached_allocator,
.get_staging_cached_allocator = pl_get_staging_cached_allocator,
.get_staging_uncached_buddy_allocator = pl_get_staging_uncached_allocator_buddy,
.get_blocks = pl_get_allocator_blocks,
.get_ranges = pl_get_allocator_ranges,
.cleanup = pl_cleanup_allocators
};
pl_set_api(ptApiRegistry, plGPUAllocatorsI, &tApi);

Expand Down
11 changes: 6 additions & 5 deletions extensions/pl_gpu_allocators_ext.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,7 +41,7 @@ typedef struct _plDevice plDevice;
// [SECTION] APIs
//-----------------------------------------------------------------------------

#define plGPUAllocatorsI_version (plVersion){1, 0, 0}
#define plGPUAllocatorsI_version (plVersion){1, 1, 0}

//-----------------------------------------------------------------------------
// [SECTION] public api structs
Expand All @@ -50,10 +50,11 @@ typedef struct _plDevice plDevice;
typedef struct _plGPUAllocatorsI
{
// allocators
plDeviceMemoryAllocatorI* (*get_local_dedicated_allocator) (plDevice*);
plDeviceMemoryAllocatorI* (*get_local_buddy_allocator) (plDevice*);
plDeviceMemoryAllocatorI* (*get_staging_uncached_allocator)(plDevice*);
plDeviceMemoryAllocatorI* (*get_staging_cached_allocator) (plDevice*);
plDeviceMemoryAllocatorI* (*get_local_dedicated_allocator) (plDevice*);
plDeviceMemoryAllocatorI* (*get_local_buddy_allocator) (plDevice*);
plDeviceMemoryAllocatorI* (*get_staging_uncached_allocator) (plDevice*);
plDeviceMemoryAllocatorI* (*get_staging_uncached_buddy_allocator)(plDevice*);
plDeviceMemoryAllocatorI* (*get_staging_cached_allocator) (plDevice*);

void (*cleanup)(plDevice*);

Expand Down
11 changes: 6 additions & 5 deletions extensions/pl_renderer_ext.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -180,15 +180,16 @@ pl_refr_initialize(plWindow* ptWindow)
}

// load gpu allocators
gptData->ptLocalBuddyAllocator = gptGpuAllocators->get_local_buddy_allocator(gptData->ptDevice);
gptData->ptLocalDedicatedAllocator = gptGpuAllocators->get_local_dedicated_allocator(gptData->ptDevice);
gptData->ptStagingUnCachedAllocator = gptGpuAllocators->get_staging_uncached_allocator(gptData->ptDevice);
gptData->ptStagingCachedAllocator = gptGpuAllocators->get_staging_cached_allocator(gptData->ptDevice);
gptData->ptLocalBuddyAllocator = gptGpuAllocators->get_local_buddy_allocator(gptData->ptDevice);
gptData->ptLocalDedicatedAllocator = gptGpuAllocators->get_local_dedicated_allocator(gptData->ptDevice);
gptData->ptStagingUnCachedAllocator = gptGpuAllocators->get_staging_uncached_allocator(gptData->ptDevice);
gptData->ptStagingUnCachedBuddyAllocator = gptGpuAllocators->get_staging_uncached_buddy_allocator(gptData->ptDevice);
gptData->ptStagingCachedAllocator = gptGpuAllocators->get_staging_cached_allocator(gptData->ptDevice);

// create staging buffers
const plBufferDesc tStagingBufferDesc = {
.tUsage = PL_BUFFER_USAGE_STAGING,
.szByteSize = 268435456 / 2
.szByteSize = 268435456
};
for(uint32_t i = 0; i < gptGfx->get_frames_in_flight(); i++)
gptData->tStagingBufferHandle[i] = pl__refr_create_staging_buffer(&tStagingBufferDesc, "staging", i);
Expand Down
11 changes: 10 additions & 1 deletion extensions/pl_renderer_internal.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -300,8 +300,17 @@ pl__refr_create_staging_buffer(const plBufferDesc* ptDesc, const char* pcName, u
plBuffer* ptBuffer = NULL;
const plBufferHandle tHandle = gptGfx->create_buffer(ptDevice, ptDesc, &ptBuffer);

// PL_DEVICE_BUDDY_BLOCK_SIZE

plDeviceMemoryAllocatorI* ptAllocator = NULL;

if(ptBuffer->tMemoryRequirements.ulSize > PL_DEVICE_BUDDY_BLOCK_SIZE)
ptAllocator = gptData->ptStagingUnCachedAllocator;
else
ptAllocator = gptData->ptStagingUnCachedBuddyAllocator;

// allocate memory
const plDeviceMemoryAllocation tAllocation = gptData->ptStagingUnCachedAllocator->allocate(gptData->ptStagingUnCachedAllocator->ptInst,
const plDeviceMemoryAllocation tAllocation = ptAllocator->allocate(ptAllocator->ptInst,
ptBuffer->tMemoryRequirements.uMemoryTypeBits,
ptBuffer->tMemoryRequirements.ulSize,
ptBuffer->tMemoryRequirements.ulAlignment,
Expand Down
1 change: 1 addition & 0 deletions extensions/pl_renderer_internal.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -515,6 +515,7 @@ typedef struct _plRefRendererData
plDeviceMemoryAllocatorI* ptLocalDedicatedAllocator;
plDeviceMemoryAllocatorI* ptLocalBuddyAllocator;
plDeviceMemoryAllocatorI* ptStagingUnCachedAllocator;
plDeviceMemoryAllocatorI* ptStagingUnCachedBuddyAllocator;
plDeviceMemoryAllocatorI* ptStagingCachedAllocator;

// default textures & samplers & bindgroups
Expand Down
9 changes: 8 additions & 1 deletion sandbox/pl_ecs_tools.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -98,6 +98,8 @@ pl_show_ecs_window(plEntity* ptSelectedEntity, plComponentLibrary* ptLibrary, bo
plInverseKinematicsComponent* ptIKComp = gptEcs->get_component(ptLibrary, PL_COMPONENT_TYPE_INVERSE_KINEMATICS, *ptSelectedEntity);
plLightComponent* ptLightComp = gptEcs->get_component(ptLibrary, PL_COMPONENT_TYPE_LIGHT, *ptSelectedEntity);
plEnvironmentProbeComponent* ptProbeComp = gptEcs->get_component(ptLibrary, PL_COMPONENT_TYPE_ENVIRONMENT_PROBE, *ptSelectedEntity);
plHumanoidComponent* ptHumanComp = gptEcs->get_component(ptLibrary, PL_COMPONENT_TYPE_HUMANOID, *ptSelectedEntity);


gptUi->text("Entity: %u, %u", ptSelectedEntity->uIndex, ptSelectedEntity->uGeneration);

Expand All @@ -107,6 +109,11 @@ pl_show_ecs_window(plEntity* ptSelectedEntity, plComponentLibrary* ptLibrary, bo
gptUi->end_collapsing_header();
}

if(ptHumanComp && gptUi->begin_collapsing_header("Humanoid", 0))
{
gptUi->end_collapsing_header();
}

if(ptProbeComp && gptUi->begin_collapsing_header("Environment Probe", 0))
{
gptUi->input_float3("Position", ptProbeComp->tPosition.d, NULL, 0);
Expand Down Expand Up @@ -331,7 +338,7 @@ pl_show_ecs_window(plEntity* ptSelectedEntity, plComponentLibrary* ptLibrary, bo
// int iResolution = (int)ptLight->uShadowResolution;
gptUi->input_float4("Cascade Splits", ptLightComp->afCascadeSplits, NULL, 0);
}

gptUi->end_collapsing_header();
}

if(ptMaterialComp && gptUi->begin_collapsing_header("Material", 0))
Expand Down
Loading