intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-18 22:08:53 +08:00
Code Issues Packages Projects Releases Wiki Activity

Zebin: Add test for default execEnv values

Browse Source 
Added test checking population of Kernel Descriptor with default
execution environment values.
Sorting.

Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
This commit is contained in:
Krystian Chmielewski
2021-11-10 11:56:49 +00:00
committed by Compute-Runtime-Automation
parent c20ec6c148
commit 620484010d
3 changed files with 74 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
shared/source/device_binary_format/elf/zebin_elf.h
View File

@@ -293,6 +293,9 @@ using HasDpasT = bool;
using HasFenceForImageAccessT = bool;
using HasGlobalAtomicsT = bool;
using HasMultiScratchSpacesT = bool;
using HasNonKernelArgAtomicT = int32_t;
using HasNonKernelArgLoadT = int32_t;
using HasNonKernelArgStoreT = int32_t;
using HasNoStatelessWriteT = bool;
using HwPreemptionModeT = int32_t;
using OffsetToSkipPerThreadDataLoadT = int32_t;
@@ -303,9 +306,6 @@ using SimdSizeT = int32_t;
using SlmSizeT = int32_t;
using SubgroupIndependentForwardProgressT = bool;
using WorkgroupWalkOrderDimensionsT = int32_t[3];
using HasNonKernelArgLoad = int32_t;
using HasNonKernelArgStore = int32_t;
using HasNonKernelArgAtomic = int32_t;
namespace Defaults {
static constexpr BarrierCountT barrierCount = 0;
@@ -316,6 +316,9 @@ static constexpr HasDpasT hasDpas = false;
static constexpr HasFenceForImageAccessT hasFenceForImageAccess = false;
static constexpr HasGlobalAtomicsT hasGlobalAtomics = false;
static constexpr HasMultiScratchSpacesT hasMultiScratchSpaces = false;
static constexpr HasNonKernelArgAtomicT hasNonKernelArgAtomic = false;
static constexpr HasNonKernelArgLoadT hasNonKernelArgLoad = false;
static constexpr HasNonKernelArgStoreT hasNonKernelArgStore = false;
static constexpr HasNoStatelessWriteT hasNoStatelessWrite = false;
static constexpr HwPreemptionModeT hwPreemptionMode = -1;
static constexpr OffsetToSkipPerThreadDataLoadT offsetToSkipPerThreadDataLoad = 0;
@@ -325,9 +328,6 @@ static constexpr RequiredWorkGroupSizeT requiredWorkGroupSize = {0, 0, 0};
static constexpr SlmSizeT slmSize = 0;
static constexpr SubgroupIndependentForwardProgressT subgroupIndependentForwardProgress = false;
static constexpr WorkgroupWalkOrderDimensionsT workgroupWalkOrderDimensions = {0, 1, 2};
static constexpr HasNonKernelArgLoad hasNonKernelArgLoad = false;
static constexpr HasNonKernelArgStore hasNonKernelArgStore = false;
static constexpr HasNonKernelArgAtomic hasNonKernelArgAtomic = false;
} // namespace Defaults
static constexpr ConstStringRef required[] = {
@@ -359,9 +359,9 @@ struct ExecutionEnvBaseT {
};
struct ExperimentalPropertiesBaseT {
HasNonKernelArgLoad hasNonKernelArgLoad = Defaults::hasNonKernelArgLoad;
HasNonKernelArgStore hasNonKernelArgStore = Defaults::hasNonKernelArgStore;
HasNonKernelArgAtomic hasNonKernelArgAtomic = Defaults::hasNonKernelArgAtomic;
HasNonKernelArgLoadT hasNonKernelArgLoad = Defaults::hasNonKernelArgLoad;
HasNonKernelArgStoreT hasNonKernelArgStore = Defaults::hasNonKernelArgStore;
HasNonKernelArgAtomicT hasNonKernelArgAtomic = Defaults::hasNonKernelArgAtomic;
};
} // namespace ExecutionEnv

26
shared/source/device_binary_format/zebin_decoder.cpp
View File

@@ -1000,29 +1000,27 @@ NEO::DecodeError populateKernelDescriptor(NEO::ProgramInfo &dst, NEO::Elf::Elf<N
kernelDescriptor.kernelAttributes.hasNonKernelArgAtomic = outExperimentalProperties.hasNonKernelArgAtomic;
}
kernelDescriptor.kernelAttributes.barrierCount = execEnv.barrierCount;
kernelDescriptor.kernelAttributes.flags.requiresDisabledMidThreadPreemption = execEnv.disableMidThreadPreemption;
kernelDescriptor.kernelAttributes.numGrfRequired = execEnv.grfCount;
if (execEnv.has4GBBuffers) {
kernelDescriptor.kernelAttributes.bufferAddressingMode = KernelDescriptor::Stateless;
}
kernelDescriptor.kernelAttributes.flags.usesDeviceSideEnqueue = execEnv.hasDeviceEnqueue;
kernelDescriptor.kernelAttributes.flags.usesFencesForReadWriteImages = execEnv.hasFenceForImageAccess;
kernelDescriptor.kernelAttributes.flags.useGlobalAtomics = execEnv.hasGlobalAtomics;
kernelDescriptor.kernelAttributes.flags.usesStatelessWrites = (false == execEnv.hasNoStatelessWrite);
kernelDescriptor.kernelAttributes.flags.usesSpecialPipelineSelectMode = execEnv.hasDpas;
kernelDescriptor.entryPoints.skipPerThreadDataLoad = execEnv.offsetToSkipPerThreadDataLoad;
kernelDescriptor.entryPoints.skipSetFFIDGP = execEnv.offsetToSkipSetFfidGp;
kernelDescriptor.kernelMetadata.requiredSubGroupSize = execEnv.requiredSubGroupSize;
kernelDescriptor.kernelAttributes.simdSize = execEnv.simdSize;
kernelDescriptor.kernelAttributes.slmInlineSize = execEnv.slmSize;
kernelDescriptor.kernelAttributes.flags.requiresDisabledMidThreadPreemption = execEnv.disableMidThreadPreemption;
kernelDescriptor.kernelAttributes.flags.requiresSubgroupIndependentForwardProgress = execEnv.subgroupIndependentForwardProgress;
kernelDescriptor.kernelAttributes.flags.useGlobalAtomics = execEnv.hasGlobalAtomics;
kernelDescriptor.kernelAttributes.flags.usesDeviceSideEnqueue = execEnv.hasDeviceEnqueue;
kernelDescriptor.kernelAttributes.flags.usesFencesForReadWriteImages = execEnv.hasFenceForImageAccess;
kernelDescriptor.kernelAttributes.flags.usesSpecialPipelineSelectMode = execEnv.hasDpas;
kernelDescriptor.kernelAttributes.flags.usesStatelessWrites = (false == execEnv.hasNoStatelessWrite);
kernelDescriptor.kernelAttributes.barrierCount = execEnv.barrierCount;
kernelDescriptor.kernelAttributes.bufferAddressingMode = (execEnv.has4GBBuffers) ? KernelDescriptor::Stateless : KernelDescriptor::BindfulAndStateless;
kernelDescriptor.kernelAttributes.numGrfRequired = execEnv.grfCount;
kernelDescriptor.kernelAttributes.requiredWorkgroupSize[0] = static_cast<uint16_t>(execEnv.requiredWorkGroupSize[0]);
kernelDescriptor.kernelAttributes.requiredWorkgroupSize[1] = static_cast<uint16_t>(execEnv.requiredWorkGroupSize[1]);
kernelDescriptor.kernelAttributes.requiredWorkgroupSize[2] = static_cast<uint16_t>(execEnv.requiredWorkGroupSize[2]);
kernelDescriptor.kernelAttributes.simdSize = execEnv.simdSize;
kernelDescriptor.kernelAttributes.slmInlineSize = execEnv.slmSize;
kernelDescriptor.kernelAttributes.workgroupWalkOrder[0] = static_cast<uint8_t>(execEnv.workgroupWalkOrderDimensions[0]);
kernelDescriptor.kernelAttributes.workgroupWalkOrder[1] = static_cast<uint8_t>(execEnv.workgroupWalkOrderDimensions[1]);
kernelDescriptor.kernelAttributes.workgroupWalkOrder[2] = static_cast<uint8_t>(execEnv.workgroupWalkOrderDimensions[2]);
kernelDescriptor.kernelMetadata.requiredSubGroupSize = execEnv.requiredSubGroupSize;
if ((kernelDescriptor.kernelAttributes.simdSize != 1) && (kernelDescriptor.kernelAttributes.simdSize != 8) && (kernelDescriptor.kernelAttributes.simdSize != 16) && (kernelDescriptor.kernelAttributes.simdSize != 32)) {
outErrReason.append("DeviceBinaryFormat::Zebin : Invalid simd size : " + std::to_string(kernelDescriptor.kernelAttributes.simdSize) + " in context of : " + kernelDescriptor.kernelMetadata.kernelName + ". Expected 1, 8, 16 or 32. Got : " + std::to_string(kernelDescriptor.kernelAttributes.simdSize) + "\n");

53
shared/test/unit_test/device_binary_format/zebin_decoder_tests.cpp
View File

@@ -2593,6 +2593,59 @@ kernels:
EXPECT_EQ(32, programInfo.kernelInfos[1]->kernelDescriptor.kernelAttributes.simdSize);
}
TEST(PopulateKernelDescriptor, GivenMinimalExecutionEnvThenPopulateKernelDescriptorWithDefaults) {
std::string zeinfo = R"===(
kernels:
- name : some_kernel
execution_env:
simd_size: 32
grf_count: 128
)===";
NEO::ProgramInfo programInfo;
ZebinTestData::ValidEmptyProgram zebin;
zebin.appendSection(NEO::Elf::SHT_PROGBITS, NEO::Elf::SectionsNamesZebin::textPrefix.str() + "some_kernel", {});
std::string errors, warnings;
auto elf = NEO::Elf::decodeElf(zebin.storage, errors, warnings);
ASSERT_NE(nullptr, elf.elfFileHeader) << errors << " " << warnings;
NEO::Yaml::YamlParser parser;
bool parseSuccess = parser.parse(zeinfo, errors, warnings);
ASSERT_TRUE(parseSuccess) << errors << " " << warnings;
NEO::ZebinSections zebinSections;
auto extractErr = NEO::extractZebinSections(elf, zebinSections, errors, warnings);
ASSERT_EQ(NEO::DecodeError::Success, extractErr) << errors << " " << warnings;
auto &kernelNode = *parser.createChildrenRange(*parser.findNodeWithKeyDfs("kernels")).begin();
auto err = NEO::populateKernelDescriptor(programInfo, elf, zebinSections, parser, kernelNode, errors, warnings);
EXPECT_EQ(NEO::DecodeError::Success, err);
EXPECT_TRUE(errors.empty()) << errors;
EXPECT_TRUE(warnings.empty()) << warnings;
ASSERT_EQ(1U, programInfo.kernelInfos.size());
namespace Defaults = NEO::Elf::ZebinKernelMetadata::Types::Kernel::ExecutionEnv::Defaults;
const auto &kernelDescriptor = programInfo.kernelInfos[0]->kernelDescriptor;
EXPECT_EQ(kernelDescriptor.entryPoints.skipPerThreadDataLoad, static_cast<NEO::InstructionsSegmentOffset>(Defaults::offsetToSkipPerThreadDataLoad));
EXPECT_EQ(kernelDescriptor.entryPoints.skipSetFFIDGP, static_cast<NEO::InstructionsSegmentOffset>(Defaults::offsetToSkipSetFfidGp));
EXPECT_EQ(kernelDescriptor.kernelAttributes.flags.requiresDisabledMidThreadPreemption, Defaults::disableMidThreadPreemption);
EXPECT_EQ(kernelDescriptor.kernelAttributes.flags.requiresSubgroupIndependentForwardProgress, Defaults::subgroupIndependentForwardProgress);
EXPECT_EQ(kernelDescriptor.kernelAttributes.flags.useGlobalAtomics, Defaults::hasGlobalAtomics);
EXPECT_EQ(kernelDescriptor.kernelAttributes.flags.usesDeviceSideEnqueue, Defaults::hasDeviceEnqueue);
EXPECT_EQ(kernelDescriptor.kernelAttributes.flags.usesFencesForReadWriteImages, Defaults::hasFenceForImageAccess);
EXPECT_EQ(kernelDescriptor.kernelAttributes.flags.usesSpecialPipelineSelectMode, Defaults::hasDpas);
EXPECT_EQ(kernelDescriptor.kernelAttributes.flags.usesStatelessWrites, (false == Defaults::hasNoStatelessWrite));
EXPECT_EQ(kernelDescriptor.kernelAttributes.barrierCount, static_cast<uint8_t>(Defaults::barrierCount));
EXPECT_EQ(kernelDescriptor.kernelAttributes.bufferAddressingMode, (Defaults::has4GBBuffers) ? KernelDescriptor::Stateless : KernelDescriptor::BindfulAndStateless);
EXPECT_EQ(kernelDescriptor.kernelAttributes.requiredWorkgroupSize[0], static_cast<uint16_t>(Defaults::requiredWorkGroupSize[0]));
EXPECT_EQ(kernelDescriptor.kernelAttributes.requiredWorkgroupSize[1], static_cast<uint16_t>(Defaults::requiredWorkGroupSize[1]));
EXPECT_EQ(kernelDescriptor.kernelAttributes.requiredWorkgroupSize[2], static_cast<uint16_t>(Defaults::requiredWorkGroupSize[2]));
EXPECT_EQ(kernelDescriptor.kernelAttributes.slmInlineSize, static_cast<uint32_t>(Defaults::slmSize));
EXPECT_EQ(kernelDescriptor.kernelAttributes.workgroupWalkOrder[0], static_cast<uint8_t>(Defaults::workgroupWalkOrderDimensions[0]));
EXPECT_EQ(kernelDescriptor.kernelAttributes.workgroupWalkOrder[1], static_cast<uint8_t>(Defaults::workgroupWalkOrderDimensions[1]));
EXPECT_EQ(kernelDescriptor.kernelAttributes.workgroupWalkOrder[2], static_cast<uint8_t>(Defaults::workgroupWalkOrderDimensions[2]));
EXPECT_EQ(kernelDescriptor.kernelMetadata.requiredSubGroupSize, static_cast<uint8_t>(Defaults::requiredSubGroupSize));
}
TEST(PopulateKernelDescriptor, WhenValidationOfZeinfoSectionsCountFailsThenDecodingFails) {
NEO::ConstStringRef zeinfo = R"===(
kernels: