encounter
/
dawn-cmake
mirror of https://github.com/encounter/dawn-cmake.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Reland "Populate some D3D12 limits from the backend" 

This is a reland of aa8fcfc64b
It changes MaxLimitTests to use at most a 512MB buffer on 32-bit
platforms.

Original change's description:
> Populate some D3D12 limits from the backend
>
> Also bumps the required D3D feature level to 11.1
>
> Bug: dawn:685
> Change-Id: I40bc3a162e0aee596d61118ba0dfe0bf9cb60d93
> Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/65120
> Commit-Queue: Austin Eng <enga@chromium.org>
> Reviewed-by: Rafael Cintron <rafael.cintron@microsoft.com>
> Reviewed-by: Corentin Wallez <cwallez@chromium.org>

Bug: dawn:685
Change-Id: I2e1df5f7ac0c9bbb6476ca2e1964a9af4afd89b6
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/67145
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Rafael Cintron <rafael.cintron@microsoft.com>
Commit-Queue: Austin Eng <enga@chromium.org>
This commit is contained in:
Austin Eng 2021-10-26 16:48:36 +00:00 committed by Dawn LUCI CQ
parent a6ddc8f0a2
commit 167d299916
2 changed files with 143 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

138
src/dawn_native/d3d12/AdapterD3D12.cpp
View File

@ -66,7 +66,7 @@ namespace dawn_native { namespace d3d12 {
// Create the device to populate the adapter properties then reuse it when needed for actual
// rendering.
const PlatformFunctions* functions = GetBackend()->GetFunctions();
if (FAILED(functions->d3d12CreateDevice(GetHardwareAdapter(), D3D_FEATURE_LEVEL_11_0,
if (FAILED(functions->d3d12CreateDevice(GetHardwareAdapter(), D3D_FEATURE_LEVEL_11_1,
_uuidof(ID3D12Device), &mD3d12Device))) {
return DAWN_INTERNAL_ERROR("D3D12CreateDevice failed");
}
@ -139,7 +139,143 @@ namespace dawn_native { namespace d3d12 {
}
MaybeError Adapter::InitializeSupportedLimitsImpl(CombinedLimits* limits) {
D3D12_FEATURE_DATA_D3D12_OPTIONS featureData = {};
DAWN_TRY(CheckHRESULT(mD3d12Device->CheckFeatureSupport(D3D12_FEATURE_D3D12_OPTIONS,
&featureData, sizeof(featureData)),
"CheckFeatureSupport"));
GetDefaultLimits(&limits->v1);
// https://docs.microsoft.com/en-us/windows/win32/direct3d12/hardware-feature-levels
// Limits that are the same across D3D feature levels
limits->v1.maxTextureDimension1D = D3D12_REQ_TEXTURE1D_U_DIMENSION;
limits->v1.maxTextureDimension2D = D3D12_REQ_TEXTURE2D_U_OR_V_DIMENSION;
limits->v1.maxTextureDimension3D = D3D12_REQ_TEXTURE3D_U_V_OR_W_DIMENSION;
limits->v1.maxTextureArrayLayers = D3D12_REQ_TEXTURE2D_ARRAY_AXIS_DIMENSION;
// Slot values can be 0-15, inclusive:
// https://docs.microsoft.com/en-ca/windows/win32/api/d3d12/ns-d3d12-d3d12_input_element_desc
limits->v1.maxVertexBuffers = 16;
limits->v1.maxVertexAttributes = D3D12_IA_VERTEX_INPUT_RESOURCE_SLOT_COUNT;
// Note: WebGPU requires FL11.1+
// https://docs.microsoft.com/en-us/windows/win32/direct3d12/hardware-support
// Resource Binding Tier: 1 2 3
// Max(CBV+UAV+SRV) 1M 1M 1M+
// Max CBV per stage 14 14 full
// Max SRV per stage 128 full full
// Max UAV in all stages 64 64 full
// Max Samplers per stage 16 2048 2048
// https://docs.microsoft.com/en-us/windows-hardware/test/hlk/testref/efad06e8-51d1-40ce-ad5c-573a134b4bb6
// "full" means the full heap can be used. This is tested
// to work for 1 million descriptors, and 1.1M for tier 3.
uint32_t maxCBVsPerStage;
uint32_t maxSRVsPerStage;
uint32_t maxUAVsAllStages;
uint32_t maxSamplersPerStage;
switch (featureData.ResourceBindingTier) {
case D3D12_RESOURCE_BINDING_TIER_1:
maxCBVsPerStage = 14;
maxSRVsPerStage = 128;
maxUAVsAllStages = 64;
maxSamplersPerStage = 16;
break;
case D3D12_RESOURCE_BINDING_TIER_2:
maxCBVsPerStage = 14;
maxSRVsPerStage = 1'000'000;
maxUAVsAllStages = 64;
maxSamplersPerStage = 2048;
break;
case D3D12_RESOURCE_BINDING_TIER_3:
maxCBVsPerStage = 1'100'000;
maxSRVsPerStage = 1'100'000;
maxUAVsAllStages = 1'100'000;
maxSamplersPerStage = 2048;
break;
default:
UNREACHABLE();
}
ASSERT(maxUAVsAllStages / 2 > limits->v1.maxStorageTexturesPerShaderStage);
ASSERT(maxUAVsAllStages / 2 > limits->v1.maxStorageBuffersPerShaderStage);
limits->v1.maxUniformBuffersPerShaderStage = maxCBVsPerStage;
// Allocate half of the UAVs to storage buffers, and half to storage textures.
limits->v1.maxStorageTexturesPerShaderStage = maxUAVsAllStages / 2;
limits->v1.maxStorageBuffersPerShaderStage = maxUAVsAllStages - maxUAVsAllStages / 2;
limits->v1.maxSampledTexturesPerShaderStage = maxSRVsPerStage;
limits->v1.maxSamplersPerShaderStage = maxSamplersPerStage;
// https://docs.microsoft.com/en-us/windows/win32/direct3d12/root-signature-limits
// In DWORDS. Descriptor tables cost 1, Root constants cost 1, Root descriptors cost 2.
static constexpr uint32_t kMaxRootSignatureSize = 64u;
// Dawn maps WebGPU's binding model by:
// - (maxBindGroups)
// CBVs/UAVs/SRVs for bind group are a root descriptor table
// - (maxBindGroups)
// Samplers for each bind group are a root descriptor table
// - (2 * maxDynamicBuffers)
// Each dynamic buffer is a root descriptor
// RESERVED:
// - 2 root constants for the baseVertex/baseInstance constants.
// - 3 root constants for num workgroups X, Y, Z
// - (1)
// TODO(crbug.com/dawn/429): Dynamic storage buffers need bounds checks.
// This will probably be 1 CBV (root descriptor table) to store all the lengths.
static constexpr uint32_t kReservedSlots = 6;
// Available slots after base limits considered.
uint32_t availableRootSignatureSlots =
kMaxRootSignatureSize - kReservedSlots -
2 * (limits->v1.maxBindGroups + limits->v1.maxDynamicUniformBuffersPerPipelineLayout +
limits->v1.maxDynamicStorageBuffersPerPipelineLayout);
// Because we need either:
// - 1 cbv/uav/srv table + 1 sampler table
// - 2 slots for a root descriptor
uint32_t availableDynamicBufferOrBindGroup = availableRootSignatureSlots / 2;
// We can either have a bind group, a dyn uniform buffer or a dyn storage buffer.
// Distribute evenly.
limits->v1.maxBindGroups += availableDynamicBufferOrBindGroup / 3;
limits->v1.maxDynamicUniformBuffersPerPipelineLayout +=
availableDynamicBufferOrBindGroup / 3;
limits->v1.maxDynamicStorageBuffersPerPipelineLayout +=
(availableDynamicBufferOrBindGroup - 2 * (availableDynamicBufferOrBindGroup / 3));
ASSERT(2 * (limits->v1.maxBindGroups +
limits->v1.maxDynamicUniformBuffersPerPipelineLayout +
limits->v1.maxDynamicStorageBuffersPerPipelineLayout) <=
kMaxRootSignatureSize - kReservedSlots);
// https://docs.microsoft.com/en-us/windows/win32/direct3dhlsl/sm5-attributes-numthreads
limits->v1.maxComputeWorkgroupSizeX = D3D12_CS_THREAD_GROUP_MAX_X;
limits->v1.maxComputeWorkgroupSizeY = D3D12_CS_THREAD_GROUP_MAX_Y;
limits->v1.maxComputeWorkgroupSizeZ = D3D12_CS_THREAD_GROUP_MAX_Z;
limits->v1.maxComputeInvocationsPerWorkgroup = D3D12_CS_THREAD_GROUP_MAX_THREADS_PER_GROUP;
// https://docs.maxComputeWorkgroupSizeXmicrosoft.com/en-us/windows/win32/api/d3d12/ns-d3d12-d3d12_dispatch_arguments
limits->v1.maxComputeWorkgroupsPerDimension =
D3D12_CS_DISPATCH_MAX_THREAD_GROUPS_PER_DIMENSION;
// https://docs.microsoft.com/en-us/windows/win32/direct3d11/overviews-direct3d-11-devices-downlevel-compute-shaders
// Thread Group Shared Memory is limited to 16Kb on downlevel hardware. This is less than
// the 32Kb that is available to Direct3D 11 hardware. D3D12 is also 32kb.
limits->v1.maxComputeWorkgroupStorageSize = 32768;
// Max number of "constants" where each constant is a 16-byte float4
limits->v1.maxUniformBufferBindingSize = D3D12_REQ_CONSTANT_BUFFER_ELEMENT_COUNT * 16;
// D3D12 has no documented limit on the size of a storage buffer binding.
limits->v1.maxStorageBufferBindingSize = 4294967295;
// TODO(crbug.com/dawn/685):
// LIMITS NOT SET:
// - maxInterStageShaderComponents
// - maxVertexBufferArrayStride
return {};
}

8
src/tests/end2end/MaxLimitTests.cpp
View File

@ -111,9 +111,13 @@ TEST_P(MaxLimitTests, MaxBufferBindingSize) {
// because allocating the buffer for zero-initialization fails.
maxBufferBindingSize =
std::min(maxBufferBindingSize, uint64_t(2) * 1024 * 1024 * 1024);
if (IsWARP()) {
// With WARP or on 32-bit platforms, such large buffer allocations often fail.
#ifndef DAWN_PLATFORM_32BIT
if (IsWARP())
#endif
{
maxBufferBindingSize =
std::min(maxBufferBindingSize, uint64_t(1) * 1024 * 1024 * 1024);
std::min(maxBufferBindingSize, uint64_t(512) * 1024 * 1024);
}
shader = R"(
[[block]] struct Buf {