intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 06:49:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

Pass DispatchInfo to estimation functions

Browse Source 
Related-To: NEO-5546

Signed-off-by: Maciej Dziuban <maciej.dziuban@intel.com>
This commit is contained in:
Maciej Dziuban
2021-03-03 17:29:32 +00:00
committed by Compute-Runtime-Automation
parent bfd9aba350
commit 1350aa52fb
12 changed files with 43 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
shared/source/command_container/command_encoder.h
View File

@@ -46,7 +46,7 @@ struct EncodeDispatchKernel {
static void *getInterfaceDescriptor(CommandContainer &container, uint32_t &iddOffset);
static size_t estimateEncodeDispatchKernelCmdsSize(Device *device);
static size_t estimateEncodeDispatchKernelCmdsSize(Device *device, Vec3<size_t> groupStart, Vec3<size_t> groupCount);
static bool isRuntimeLocalIdsGenerationRequired(uint32_t activeChannels,
size_t *lws,

12
shared/source/command_container/command_encoder_bdw_plus.inl
View File

@@ -43,7 +43,13 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container,
LinearStream *listCmdBufferStream = container.getCommandStream();
size_t sshOffset = 0;
size_t estimatedSizeRequired = estimateEncodeDispatchKernelCmdsSize(device);
auto threadDims = static_cast<const uint32_t *>(pThreadGroupDimensions);
const Vec3<size_t> threadStartVec{0, 0, 0};
Vec3<size_t> threadDimsVec{0, 0, 0};
if (threadDims != nullptr) {
threadDimsVec = {threadDims[0], threadDims[1], threadDims[2]};
}
size_t estimatedSizeRequired = estimateEncodeDispatchKernelCmdsSize(device, threadStartVec, threadDimsVec);
if (container.getCommandStream()->getAvailableSpace() < estimatedSizeRequired) {
auto bbEnd = listCmdBufferStream->getSpaceForCmd<MI_BATCH_BUFFER_END>();
*bbEnd = Family::cmdInitBatchBufferEnd;
@@ -191,7 +197,7 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container,
EncodeDispatchKernel<Family>::encodeThreadData(cmd,
nullptr,
static_cast<const uint32_t *>(pThreadGroupDimensions),
threadDims,
dispatchInterface->getGroupSize(),
kernelDescriptor.kernelAttributes.simdSize,
kernelDescriptor.kernelAttributes.numLocalIdChannels,
@@ -316,7 +322,7 @@ template <typename Family>
void EncodeDispatchKernel<Family>::appendAdditionalIDDFields(INTERFACE_DESCRIPTOR_DATA *pInterfaceDescriptor, const HardwareInfo &hwInfo, const uint32_t threadsPerThreadGroup, uint32_t slmTotalSize, SlmPolicy slmPolicy) {}
template <typename Family>
size_t EncodeDispatchKernel<Family>::estimateEncodeDispatchKernelCmdsSize(Device *device) {
size_t EncodeDispatchKernel<Family>::estimateEncodeDispatchKernelCmdsSize(Device *device, Vec3<size_t> groupStart, Vec3<size_t> groupCount) {
using MEDIA_STATE_FLUSH = typename Family::MEDIA_STATE_FLUSH;
using MEDIA_INTERFACE_DESCRIPTOR_LOAD = typename Family::MEDIA_INTERFACE_DESCRIPTOR_LOAD;
using MI_BATCH_BUFFER_END = typename Family::MI_BATCH_BUFFER_END;

4
shared/test/common/gen12lp/test_command_encoder_gen12lp.cpp
View File

@@ -105,9 +105,9 @@ GEN12LPTEST_F(CommandEncoderTest, givenVariousEngineTypesWhenEstimateCommandBuff
using PIPELINE_SELECT = typename FamilyType::PIPELINE_SELECT;
using STATE_COMPUTE_MODE = typename FamilyType::STATE_COMPUTE_MODE;
auto sizeWA = EncodeDispatchKernel<FamilyType>::estimateEncodeDispatchKernelCmdsSize(pDevice);
auto sizeWA = EncodeDispatchKernel<FamilyType>::estimateEncodeDispatchKernelCmdsSize(pDevice, Vec3<size_t>(0, 0, 0), Vec3<size_t>(1, 1, 1));
static_cast<MockOsContext *>(pDevice->getDefaultEngine().osContext)->engineType = aub_stream::ENGINE_CCS;
auto size = EncodeDispatchKernel<FamilyType>::estimateEncodeDispatchKernelCmdsSize(pDevice);
auto size = EncodeDispatchKernel<FamilyType>::estimateEncodeDispatchKernelCmdsSize(pDevice, Vec3<size_t>(0, 0, 0), Vec3<size_t>(1, 1, 1));
auto expectedDiff = 2 * PreambleHelper<FamilyType>::getCmdSizeForPipelineSelect(pDevice->getHardwareInfo());
auto diff = sizeWA - size;