intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-11 16:45:25 +08:00
Code Issues Packages Projects Releases Wiki Activity

feature: bindless addressing - store bindlessInfo in allocation

Browse Source 
- store surface state info for bindless addressing in graphics
allocation
- remove map in BindlessHeapsHelper - bindlessInfo is constant for
the lifetime of an allocation
- program bindless offsets and surface states for images when used in
bindless kernel
- handle ouf of memory on surface state heap - return error

Related-To: NEO-7063

Signed-off-by: Mateusz Hoppe <mateusz.hoppe@intel.com>
This commit is contained in:
Mateusz Hoppe
2023-07-21 11:55:22 +00:00
committed by Compute-Runtime-Automation
parent e52712b800
commit 67d39f88e6
10 changed files with 240 additions and 61 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
level_zero/core/source/kernel/kernel_imp.cpp
View File

@@ -561,6 +561,13 @@ ze_result_t KernelImp::setArgBufferWithAlloc(uint32_t argIndex, uintptr_t argVal
NEO::patchPointer(ArrayRef<uint8_t>(crossThreadData.get(), crossThreadDataSize), arg, val);
if (NEO::isValidOffset(arg.bindful) || NEO::isValidOffset(arg.bindless)) {
if (NEO::isValidOffset(arg.bindless)) {
if (!this->module->getDevice()->getNEODevice()->getMemoryManager()->allocateBindlessSlot(allocation)) {
return ZE_RESULT_ERROR_OUT_OF_HOST_MEMORY;
}
}
setBufferSurfaceState(argIndex, reinterpret_cast<void *>(val), allocation);
}
NEO::SvmAllocationData *allocData = nullptr;
@@ -710,7 +717,22 @@ ze_result_t KernelImp::setArgImage(uint32_t argIndex, size_t argSize, const void
const auto image = Image::fromHandle(*static_cast<const ze_image_handle_t *>(argVal));
if (kernelImmData->getDescriptor().kernelAttributes.imageAddressingMode == NEO::KernelDescriptor::Bindless) {
image->copySurfaceStateToSSH(patchBindlessSurfaceState(image->getAllocation(), arg.bindless), 0u, isMediaBlockImage);
NEO::BindlessHeapsHelper *bindlessHeapsHelper = this->module->getDevice()->getNEODevice()->getBindlessHeapsHelper();
if (bindlessHeapsHelper) {
if (!this->module->getDevice()->getNEODevice()->getMemoryManager()->allocateBindlessSlot(image->getAllocation())) {
return ZE_RESULT_ERROR_OUT_OF_HOST_MEMORY;
}
auto ssPtr = patchBindlessSurfaceState(image->getAllocation(), arg.bindless);
image->copySurfaceStateToSSH(ssPtr, 0u, isMediaBlockImage);
} else {
auto &gfxCoreHelper = this->module->getDevice()->getNEODevice()->getRootDeviceEnvironmentRef().getHelper<NEO::GfxCoreHelper>();
auto surfaceStateSize = gfxCoreHelper.getRenderSurfaceStateSize();
auto ssPtr = ptrOffset(surfaceStateHeapData.get(), getSurfaceStateIndexForBindlessOffset(arg.bindless) * surfaceStateSize);
image->copySurfaceStateToSSH(ssPtr, 0u, isMediaBlockImage);
}
} else {
image->copySurfaceStateToSSH(surfaceStateHeapData.get(), arg.bindful, isMediaBlockImage);
}
@@ -1069,19 +1091,18 @@ void KernelImp::setDebugSurface() {
*device->getNEODevice(), getKernelDescriptor().kernelAttributes.flags.useGlobalAtomics, device->isImplicitScalingCapable());
}
}
void *KernelImp::patchBindlessSurfaceState(NEO::GraphicsAllocation *alloc, uint32_t bindless) {
auto &gfxCoreHelper = this->module->getDevice()->getGfxCoreHelper();
auto surfaceStateSize = gfxCoreHelper.getRenderSurfaceStateSize();
NEO::BindlessHeapsHelper *bindlessHeapsHelper = this->module->getDevice()->getNEODevice()->getBindlessHeapsHelper();
UNRECOVERABLE_IF(bindlessHeapsHelper == nullptr);
auto ssInHeap = alloc->getBindlessInfo();
auto ssInHeap = bindlessHeapsHelper->allocateSSInHeap(surfaceStateSize, alloc, NEO::BindlessHeapsHelper::GLOBAL_SSH);
this->residencyContainer.push_back(ssInHeap.heapAllocation);
auto patchLocation = ptrOffset(getCrossThreadData(), bindless);
auto patchValue = gfxCoreHelper.getBindlessSurfaceExtendedMessageDescriptorValue(static_cast<uint32_t>(ssInHeap.surfaceStateOffset));
patchWithRequiredSize(const_cast<uint8_t *>(patchLocation), sizeof(patchValue), patchValue);
return ssInHeap.ssPtr;
}
void KernelImp::patchWorkgroupSizeInCrossThreadData(uint32_t x, uint32_t y, uint32_t z) {
const NEO::KernelDescriptor &desc = kernelImmData->getDescriptor();
auto dst = ArrayRef<uint8_t>(crossThreadData.get(), crossThreadDataSize);