intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-08 14:02:58 +08:00
Code Issues Packages Projects Releases Wiki Activity

feature: Add Initial Support for Get Vector Width Properties

Browse Source 
- Added support for L0 to query Vector Width Properties which allows for
querying all the supported vector widths and their prefferred and
native vector widths per data type.
- Updated shared code to support querying the same values thru
getPreferredVectorWidth* and getNativeVectorWidth* helper functions
for OpenCL and Level Zero.
- Initial support for reporting base values for the vector
widths with future support for reporting different values per platform and
based on the SIMD/SIMT size used.

Related-To: NEO-13745

Signed-off-by: Neil R. Spruit <neil.r.spruit@intel.com>
This commit is contained in:
Neil R. Spruit
2025-06-24 22:48:43 +00:00
committed by Compute-Runtime-Automation
parent 39b28474ae
commit 532c663da6
15 changed files with 233 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
opencl/source/cl_device/cl_device_caps.cpp
View File

@@ -149,11 +149,11 @@ void ClDevice::initializeCaps() {
deviceInfo.independentForwardProgress = hwInfo.capabilityTable.supportsIndependentForwardProgress;
deviceInfo.maxNumOfSubGroups = 0;
if (ocl21FeaturesEnabled) {
auto simdSizeUsed = debugManager.flags.UseMaxSimdSizeToDeduceMaxWorkgroupSize.get()
? CommonConstants::maximalSimdSize
: gfxCoreHelper.getMinimalSIMDSize();
auto simdSizeUsed = debugManager.flags.UseMaxSimdSizeToDeduceMaxWorkgroupSize.get()
? CommonConstants::maximalSimdSize
: gfxCoreHelper.getMinimalSIMDSize();
if (ocl21FeaturesEnabled) {
// calculate a maximum number of subgroups in a workgroup (for the required SIMD size)
deviceInfo.maxNumOfSubGroups = static_cast<uint32_t>(sharedDeviceInfo.maxWorkGroupSize / simdSizeUsed);
@@ -233,18 +233,18 @@ void ClDevice::initializeCaps() {
deviceInfo.partitionAffinityDomain = 0;
}
deviceInfo.partitionType[0] = 0;
deviceInfo.preferredVectorWidthChar = 16;
deviceInfo.preferredVectorWidthShort = 8;
deviceInfo.preferredVectorWidthInt = 4;
deviceInfo.preferredVectorWidthLong = 1;
deviceInfo.preferredVectorWidthFloat = 1;
deviceInfo.preferredVectorWidthHalf = 8;
deviceInfo.nativeVectorWidthChar = 16;
deviceInfo.nativeVectorWidthShort = 8;
deviceInfo.nativeVectorWidthInt = 4;
deviceInfo.nativeVectorWidthLong = 1;
deviceInfo.nativeVectorWidthFloat = 1;
deviceInfo.nativeVectorWidthHalf = 8;
deviceInfo.preferredVectorWidthChar = gfxCoreHelper.getPreferredVectorWidthChar(simdSizeUsed);
deviceInfo.preferredVectorWidthShort = gfxCoreHelper.getPreferredVectorWidthShort(simdSizeUsed);
deviceInfo.preferredVectorWidthInt = gfxCoreHelper.getPreferredVectorWidthInt(simdSizeUsed);
deviceInfo.preferredVectorWidthLong = gfxCoreHelper.getPreferredVectorWidthLong(simdSizeUsed);
deviceInfo.preferredVectorWidthFloat = gfxCoreHelper.getPreferredVectorWidthFloat(simdSizeUsed);
deviceInfo.preferredVectorWidthHalf = gfxCoreHelper.getPreferredVectorWidthHalf(simdSizeUsed);
deviceInfo.nativeVectorWidthChar = gfxCoreHelper.getNativeVectorWidthChar(simdSizeUsed);
deviceInfo.nativeVectorWidthShort = gfxCoreHelper.getNativeVectorWidthShort(simdSizeUsed);
deviceInfo.nativeVectorWidthInt = gfxCoreHelper.getNativeVectorWidthInt(simdSizeUsed);
deviceInfo.nativeVectorWidthLong = gfxCoreHelper.getNativeVectorWidthLong(simdSizeUsed);
deviceInfo.nativeVectorWidthFloat = gfxCoreHelper.getNativeVectorWidthFloat(simdSizeUsed);
deviceInfo.nativeVectorWidthHalf = gfxCoreHelper.getNativeVectorWidthHalf(simdSizeUsed);
deviceInfo.maxReadWriteImageArgs = hwInfo.capabilityTable.supportsImages ? 128 : 0;
deviceInfo.executionCapabilities = CL_EXEC_KERNEL;