Media/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-11-10 05:49:51 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add resolve capability for compressed USM device allocations

Browse Source 
Related-To: NEO-5107

Signed-off-by: Slawomir Milczarek <slawomir.milczarek@intel.com>
This commit is contained in:
Slawomir Milczarek
2020-12-22 00:03:25 +00:00
committed by Compute-Runtime-Automation
parent d7ff26cc5a
commit 55f3c8f134
27 changed files with 636 additions and 215 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
opencl/source/built_ins/aux_translation_builtin.h
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2018-2020 Intel Corporation
* Copyright (C) 2018-2021 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -8,9 +8,12 @@
#pragma once
#include "shared/source/built_ins/built_ins.h"
#include "shared/source/helpers/hw_helper.h"
#include "shared/source/memory_manager/graphics_allocation.h"
#include "opencl/source/built_ins/builtins_dispatch_builder.h"
#include "opencl/source/helpers/dispatch_info_builder.h"
#include "opencl/source/kernel/kernel_objects_for_aux_translation.h"
#include "opencl/source/mem_obj/buffer.h"
#include "pipe_control_args.h"
@@ -24,13 +27,12 @@ class BuiltInOp<EBuiltInOps::AuxTranslation> : public BuiltinDispatchInfoBuilder
template <typename GfxFamily>
bool buildDispatchInfosForAuxTranslation(MultiDispatchInfo &multiDispatchInfo, const BuiltinOpParams &operationParams) const {
size_t kernelInstanceNumber = 0;
size_t numMemObjectsToTranslate = multiDispatchInfo.getMemObjsForAuxTranslation()->size();
resizeKernelInstances(numMemObjectsToTranslate);
size_t numKernelObjectsToTranslate = multiDispatchInfo.getKernelObjsForAuxTranslation()->size();
resizeKernelInstances(numKernelObjectsToTranslate);
multiDispatchInfo.setBuiltinOpParams(operationParams);
for (auto &memObj : *multiDispatchInfo.getMemObjsForAuxTranslation()) {
for (auto &kernelObj : *multiDispatchInfo.getKernelObjsForAuxTranslation()) {
DispatchInfoBuilder<SplitDispatch::Dim::d1D, SplitDispatch::SplitMode::NoSplit> builder(clDevice);
size_t allocationSize = alignUp(memObj->getSize(), 512);
UNRECOVERABLE_IF(builder.getMaxNumDispatches() != 1);
@@ -38,7 +40,7 @@ class BuiltInOp<EBuiltInOps::AuxTranslation> : public BuiltinDispatchInfoBuilder
// Before Kernel
registerPipeControlProgramming<GfxFamily>(builder.getDispatchInfo(0).dispatchInitCommands, true);
}
if (kernelInstanceNumber == numMemObjectsToTranslate - 1) {
if (kernelInstanceNumber == numKernelObjectsToTranslate - 1) {
// After Kernel
registerPipeControlProgramming<GfxFamily>(builder.getDispatchInfo(0).dispatchEpilogueCommands, false);
}
@@ -50,8 +52,20 @@ class BuiltInOp<EBuiltInOps::AuxTranslation> : public BuiltinDispatchInfoBuilder
builder.setKernel(convertToAuxKernel[kernelInstanceNumber++].get());
}
builder.setArg(0, memObj);
builder.setArg(1, memObj);
size_t allocationSize = 0;
if (kernelObj.type == KernelObjForAuxTranslation::Type::MEM_OBJ) {
auto buffer = static_cast<Buffer *>(kernelObj.object);
builder.setArg(0, buffer);
builder.setArg(1, buffer);
allocationSize = alignUp(buffer->getSize(), 512);
} else {
DEBUG_BREAK_IF(kernelObj.type != KernelObjForAuxTranslation::Type::GFX_ALLOC);
auto svmAlloc = static_cast<GraphicsAllocation *>(kernelObj.object);
auto svmPtr = reinterpret_cast<void *>(svmAlloc->getGpuAddressToPatch());
builder.setArgSvmAlloc(0, svmPtr, svmAlloc);
builder.setArgSvmAlloc(1, svmPtr, svmAlloc);
allocationSize = alignUp(svmAlloc->getUnderlyingBufferSize(), 512);
}
size_t xGws = allocationSize / 16;