intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-24 20:39:56 +08:00
Code Issues Packages Projects Releases Wiki Activity

refactor: add SetMaxBVHLevels debug flag

Browse Source 
Signed-off-by: Kamil Kopryk <kamil.kopryk@intel.com>
This commit is contained in:
Kamil Kopryk
2025-03-13 13:34:22 +00:00
committed by Compute-Runtime-Automation
parent 251ced0e4b
commit 2c77d2c698
10 changed files with 56 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
shared/source/debug_settings/debug_variables_base.inl
View File

@@ -304,6 +304,7 @@ DECLARE_DEBUG_VARIABLE(int32_t, DeferStateInitSubmissionToFirstRegularUsage, -1,
DECLARE_DEBUG_VARIABLE(int32_t, ForceNonWalkerSplitMemoryCopy, -1, "-1: default, 0: disabled, 1: enabled. If set, memory copy will be executed as single byte copy Walker without performance optimizations")
DECLARE_DEBUG_VARIABLE(int32_t, OverrideTimestampWidth, -1, "-1: default from KMD, > 0: Override timestamp width used for profiling. Requires XeKMD kernel.")
DECLARE_DEBUG_VARIABLE(int32_t, DebugUmdFifoPollInterval, -1, "-1: default , > 0: Fifo will be polled based on input in milliseconds.")
DECLARE_DEBUG_VARIABLE(int32_t, SetMaxBVHLevels, -1, "-1: default , > 0: Set maxBVHLevel in RTDispatchGlobal.")
DECLARE_DEBUG_VARIABLE(int32_t, DebugUmdInterruptTimeout, -1, "-1: default , > 0: interruptTimeout based on input in milliseconds. Default is 2000 milliseconds")
DECLARE_DEBUG_VARIABLE(int32_t, DebugUmdMaxReadWriteRetry, -1, "-1: default , > 0: max pread/pwrite retry attempts in read/writeGpuMemory calls based on input in milliseconds. Default is 3")
DECLARE_DEBUG_VARIABLE(int32_t, ForceIndirectDetectionForCMKernels, -1, "-1: default , 0 : disable indirect detection for CM kernels, 1 : enable indirect detection for CM kernels")

12
shared/source/device/device.cpp
View File

@@ -1134,18 +1134,24 @@ void Device::allocateRTDispatchGlobals(uint32_t maxBvhLevels) {
break;
}
struct RTDispatchGlobals dispatchGlobals = {0};
RTDispatchGlobals dispatchGlobals = {0};
dispatchGlobals.rtMemBasePtr = rtStackAllocation->getGpuAddress() + rtStackSize;
dispatchGlobals.callStackHandlerKSP = reinterpret_cast<uint64_t>(nullptr);
auto releaseHelper = getReleaseHelper();
dispatchGlobals.stackSizePerRay = releaseHelper ? releaseHelper->getStackSizePerRay() : 0;
dispatchGlobals.numDSSRTStacks = RayTracingHelper::getNumRtStacksPerDss(*this);
dispatchGlobals.maxBVHLevels = maxBvhLevels;
auto rtStacksPerDss = RayTracingHelper::getNumRtStacksPerDss(*this);
dispatchGlobals.numDSSRTStacks = rtStacksPerDss;
dispatchGlobals.maxBVHLevels = maxBvhLevels;
uint32_t *dispatchGlobalsAsArray = reinterpret_cast<uint32_t *>(&dispatchGlobals);
dispatchGlobalsAsArray[7] = 1;
if (releaseHelper) {
bool heaplessEnabled = this->getCompilerProductHelper().isHeaplessModeEnabled();
releaseHelper->adjustRTDispatchGlobals(static_cast<void *>(&dispatchGlobals), rtStacksPerDss, heaplessEnabled, maxBvhLevels);
}
MemoryTransferHelper::transferMemoryToAllocation(productHelper.isBlitCopyRequiredForLocalMemory(this->getRootDeviceEnvironment(), *dispatchGlobalsArrayAllocation),
*this,
dispatchGlobalsArrayAllocation,

2
shared/source/release_helper/release_helper.h
View File

@@ -57,6 +57,7 @@ class ReleaseHelper {
virtual uint32_t getAdditionalFp16Caps() const = 0;
virtual uint32_t getAdditionalExtraCaps() const = 0;
virtual uint32_t getStackSizePerRay() const = 0;
virtual void adjustRTDispatchGlobals(void *rtDispatchGlobals, uint32_t rtStacksPerDss, bool heaplessEnabled, uint32_t maxBvhLevels) const = 0;
virtual bool isLocalOnlyAllowed() const = 0;
virtual bool isDummyBlitWaRequired() const = 0;
virtual bool isDirectSubmissionLightSupported() const = 0;
@@ -100,6 +101,7 @@ class ReleaseHelperHw : public ReleaseHelper {
uint32_t getAdditionalFp16Caps() const override;
uint32_t getAdditionalExtraCaps() const override;
uint32_t getStackSizePerRay() const override;
void adjustRTDispatchGlobals(void *rtDispatchGlobals, uint32_t rtStacksPerDss, bool heaplessEnabled, uint32_t maxBvhLevels) const override;
bool isLocalOnlyAllowed() const override;
bool isDummyBlitWaRequired() const override;
bool isDirectSubmissionLightSupported() const override;

4
shared/source/release_helper/release_helper_base.inl
View File

@@ -137,6 +137,10 @@ uint32_t ReleaseHelperHw<releaseType>::getStackSizePerRay() const {
return 0u;
}
template <ReleaseType releaseType>
void ReleaseHelperHw<releaseType>::adjustRTDispatchGlobals(void *rtDispatchGlobals, uint32_t rtStacksPerDss, bool heaplessEnabled, uint32_t maxBvhLevels) const {
}
template <ReleaseType releaseType>
bool ReleaseHelperHw<releaseType>::isLocalOnlyAllowed() const {
return true;