Media/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-09-15 13:01:45 +08:00
Code Issues Packages Projects Releases Wiki Activity

Refactor programming of surface states

Browse Source 
Related-To: NEO-5069

Change-Id: Id7442fcdcc8c7df57f00e8dc383c11869bf1a677
Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
This commit is contained in:
Zbigniew Zdanowicz
2020-09-15 20:27:18 +02:00
committed by sys_ocldev
parent 0de5e422fb
commit 394e626db9
14 changed files with 54 additions and 68 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
level_zero/core/source/cmdlist/cmdlist_hw_base.inl
View File

@ -75,14 +75,14 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendLaunchKernelWithParams(z
if (device->getNEODevice()->getDebugger()) {
auto *ssh = commandContainer.getIndirectHeap(NEO::HeapType::SURFACE_STATE);
auto surfaceState = device->getNEODevice()->getDebugger()->getDebugSurfaceReservedSurfaceState(*ssh);
NEO::Device *neoDevice = device->getNEODevice();
auto surfaceState = neoDevice->getDebugger()->getDebugSurfaceReservedSurfaceState(*ssh);
auto debugSurface = device->getDebugSurface();
auto mocs = device->getMOCS(true, false);
NEO::EncodeSurfaceState<GfxFamily>::encodeBuffer(surfaceState, debugSurface->getGpuAddress(),
debugSurface->getUnderlyingBufferSize(),
mocs, true);
NEO::EncodeSurfaceState<GfxFamily>::encodeExtraBufferParams(debugSurface,
device->getNEODevice()->getGmmHelper(), surfaceState, false, false);
debugSurface->getUnderlyingBufferSize(), mocs,
true, false, neoDevice->getNumAvailableDevices(),
debugSurface, neoDevice->getGmmHelper());
}
appendSignalEventPostWalker(hEvent);

8
level_zero/core/source/kernel/kernel_hw.h
View File

@ -51,10 +51,10 @@ struct KernelHw : public KernelImp {
bufferSizeForSsh = alignUp(bufferSizeForSsh, alignment);
auto mocs = this->module->getDevice()->getMOCS(true, false);
bool requiresCoherency;
NEO::EncodeSurfaceState<GfxFamily>::encodeBuffer(surfaceStateAddress,
bufferAddressForSsh, bufferSizeForSsh, mocs,
requiresCoherency = false);
NEO::Device *neoDevice = module->getDevice()->getNEODevice();
NEO::EncodeSurfaceState<GfxFamily>::encodeBuffer(surfaceStateAddress, bufferAddressForSsh, bufferSizeForSsh, mocs,
true, false, neoDevice->getNumAvailableDevices(),
alloc, neoDevice->getGmmHelper());
}
std::unique_ptr<Kernel> clone() const override {

2
opencl/source/gen11/buffer_gen11.cpp
View File

@ -7,7 +7,7 @@
#include "shared/source/gen11/hw_cmds.h"
#include "opencl/source/mem_obj/buffer_bdw_plus.inl"
#include "opencl/source/mem_obj/buffer_base.inl"
namespace NEO {

2
opencl/source/gen12lp/buffer_gen12lp.cpp
View File

@ -7,7 +7,7 @@
#include "shared/source/gen12lp/hw_cmds.h"
#include "opencl/source/mem_obj/buffer_bdw_plus.inl"
#include "opencl/source/mem_obj/buffer_base.inl"
namespace NEO {

2
opencl/source/gen8/buffer_gen8.cpp
View File

@ -7,7 +7,7 @@
#include "shared/source/gen8/hw_cmds.h"
#include "opencl/source/mem_obj/buffer_bdw_plus.inl"
#include "opencl/source/mem_obj/buffer_base.inl"
namespace NEO {

2
opencl/source/gen9/buffer_gen9.cpp
View File

@ -7,7 +7,7 @@
#include "shared/source/gen9/hw_cmds.h"
#include "opencl/source/mem_obj/buffer_bdw_plus.inl"
#include "opencl/source/mem_obj/buffer_base.inl"
namespace NEO {

1
opencl/source/mem_obj/CMakeLists.txt
View File

@ -9,7 +9,6 @@ set(RUNTIME_SRCS_MEM_OBJ
${CMAKE_CURRENT_SOURCE_DIR}/buffer.cpp
${CMAKE_CURRENT_SOURCE_DIR}/buffer.h
${CMAKE_CURRENT_SOURCE_DIR}/buffer_base.inl
${CMAKE_CURRENT_SOURCE_DIR}/buffer_bdw_plus.inl
${CMAKE_CURRENT_SOURCE_DIR}/buffer_factory_init.inl
${CMAKE_CURRENT_SOURCE_DIR}/image.cpp
${CMAKE_CURRENT_SOURCE_DIR}/image.h

1
opencl/source/mem_obj/buffer.h
View File

@ -206,7 +206,6 @@ class BufferHw : public Buffer {
zeroCopy, isHostPtrSVM, isObjectRedescribed) {}
void setArgStateful(void *memory, bool forceNonAuxMode, bool disableL3, bool alignSizeForAuxTranslation, bool isReadOnlyArgument, const Device &device) override;
void appendBufferState(void *memory, const Device &device, bool isReadOnlyArgument);
void appendSurfaceStateExt(void *memory);
static Buffer *create(Context *context,

8
opencl/source/mem_obj/buffer_base.inl
View File

@ -38,11 +38,9 @@ void BufferHw<GfxFamily>::setArgStateful(void *memory, bool forceNonAuxMode, boo
auto graphicsAllocation = multiGraphicsAllocation.getGraphicsAllocation(rootDeviceIndex);
EncodeSurfaceState<GfxFamily>::encodeBuffer(memory, getBufferAddress(rootDeviceIndex),
getSurfaceSize(alignSizeForAuxTranslation, rootDeviceIndex),
getMocsValue(disableL3, isReadOnlyArgument, rootDeviceIndex), true);
EncodeSurfaceState<GfxFamily>::encodeExtraBufferParams(graphicsAllocation,
device.getGmmHelper(), memory, forceNonAuxMode, isReadOnlyArgument);
appendBufferState(memory, device, isReadOnlyArgument);
getMocsValue(disableL3, isReadOnlyArgument, rootDeviceIndex),
true, forceNonAuxMode, device.getNumAvailableDevices(),
graphicsAllocation, device.getGmmHelper());
appendSurfaceStateExt(memory);
}
} // namespace NEO

16
opencl/source/mem_obj/buffer_bdw_plus.inl
View File

@ -1,16 +0,0 @@
/*
* Copyright (C) 2017-2020 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "opencl/source/mem_obj/buffer_base.inl"
namespace NEO {
template <typename GfxFamily>
void BufferHw<GfxFamily>::appendBufferState(void *memory, const Device &device, bool isReadOnly) {
}
} // namespace NEO

6
shared/source/command_container/command_encoder.h
View File

@ -194,8 +194,10 @@ struct EncodeSurfaceState {
using AUXILIARY_SURFACE_MODE = typename R_SURFACE_STATE::AUXILIARY_SURFACE_MODE;
static void encodeBuffer(void *dst, uint64_t address, size_t size, uint32_t mocs,
bool cpuCoherent);
static void encodeExtraBufferParams(GraphicsAllocation *allocation, GmmHelper *gmmHelper, void *memory, bool forceNonAuxMode, bool isReadOnlyArgument);
bool cpuCoherent, bool forceNonAuxMode, uint32_t numAvailableDevices,
GraphicsAllocation *allocation, GmmHelper *gmmHelper);
static void encodeExtraBufferParams(R_SURFACE_STATE *surfaceState, GraphicsAllocation *allocation, GmmHelper *gmmHelper,
uint32_t numAvailableDevices);
static constexpr uintptr_t getSurfaceBaseAddressAlignmentMask() {
return ~(getSurfaceBaseAddressAlignment() - 1);

46
shared/source/command_container/command_encoder.inl
View File

@ -267,52 +267,48 @@ void EncodeStoreMMIO<Family>::encode(LinearStream &csr, uint32_t offset, uint64_
template <typename Family>
void EncodeSurfaceState<Family>::encodeBuffer(void *dst, uint64_t address, size_t size, uint32_t mocs,
bool cpuCoherent) {
auto ss = reinterpret_cast<R_SURFACE_STATE *>(dst);
bool cpuCoherent, bool forceNonAuxMode, uint32_t numAvailableDevices,
GraphicsAllocation *allocation, GmmHelper *gmmHelper) {
auto surfaceState = reinterpret_cast<R_SURFACE_STATE *>(dst);
UNRECOVERABLE_IF(!isAligned<getSurfaceBaseAddressAlignment()>(size));
SURFACE_STATE_BUFFER_LENGTH Length = {0};
Length.Length = static_cast<uint32_t>(size - 1);
ss->setWidth(Length.SurfaceState.Width + 1);
ss->setHeight(Length.SurfaceState.Height + 1);
ss->setDepth(Length.SurfaceState.Depth + 1);
surfaceState->setWidth(Length.SurfaceState.Width + 1);
surfaceState->setHeight(Length.SurfaceState.Height + 1);
surfaceState->setDepth(Length.SurfaceState.Depth + 1);
ss->setSurfaceType((address != 0) ? R_SURFACE_STATE::SURFACE_TYPE_SURFTYPE_BUFFER
surfaceState->setSurfaceType((address != 0) ? R_SURFACE_STATE::SURFACE_TYPE_SURFTYPE_BUFFER
: R_SURFACE_STATE::SURFACE_TYPE_SURFTYPE_NULL);
ss->setSurfaceFormat(SURFACE_FORMAT::SURFACE_FORMAT_RAW);
ss->setSurfaceVerticalAlignment(R_SURFACE_STATE::SURFACE_VERTICAL_ALIGNMENT_VALIGN_4);
ss->setSurfaceHorizontalAlignment(R_SURFACE_STATE::SURFACE_HORIZONTAL_ALIGNMENT_HALIGN_4);
surfaceState->setSurfaceFormat(SURFACE_FORMAT::SURFACE_FORMAT_RAW);
surfaceState->setSurfaceVerticalAlignment(R_SURFACE_STATE::SURFACE_VERTICAL_ALIGNMENT_VALIGN_4);
surfaceState->setSurfaceHorizontalAlignment(R_SURFACE_STATE::SURFACE_HORIZONTAL_ALIGNMENT_HALIGN_4);
ss->setTileMode(R_SURFACE_STATE::TILE_MODE_LINEAR);
ss->setVerticalLineStride(0);
ss->setVerticalLineStrideOffset(0);
ss->setMemoryObjectControlState(mocs);
ss->setSurfaceBaseAddress(address);
surfaceState->setTileMode(R_SURFACE_STATE::TILE_MODE_LINEAR);
surfaceState->setVerticalLineStride(0);
surfaceState->setVerticalLineStrideOffset(0);
surfaceState->setMemoryObjectControlState(mocs);
surfaceState->setSurfaceBaseAddress(address);
ss->setCoherencyType(cpuCoherent ? R_SURFACE_STATE::COHERENCY_TYPE_IA_COHERENT
surfaceState->setCoherencyType(cpuCoherent ? R_SURFACE_STATE::COHERENCY_TYPE_IA_COHERENT
: R_SURFACE_STATE::COHERENCY_TYPE_GPU_COHERENT);
ss->setAuxiliarySurfaceMode(AUXILIARY_SURFACE_MODE::AUXILIARY_SURFACE_MODE_AUX_NONE);
}
surfaceState->setAuxiliarySurfaceMode(AUXILIARY_SURFACE_MODE::AUXILIARY_SURFACE_MODE_AUX_NONE);
template <typename Family>
void EncodeSurfaceState<Family>::encodeExtraBufferParams(GraphicsAllocation *allocation, GmmHelper *gmmHelper, void *memory, bool forceNonAuxMode, bool isReadOnlyArgument) {
using RENDER_SURFACE_STATE = typename Family::RENDER_SURFACE_STATE;
using AUXILIARY_SURFACE_MODE = typename RENDER_SURFACE_STATE::AUXILIARY_SURFACE_MODE;
auto surfaceState = reinterpret_cast<RENDER_SURFACE_STATE *>(memory);
Gmm *gmm = allocation ? allocation->getDefaultGmm() : nullptr;
if (gmm && gmm->isRenderCompressed && !forceNonAuxMode) {
// Its expected to not program pitch/qpitch/baseAddress for Aux surface in CCS scenarios
surfaceState->setCoherencyType(RENDER_SURFACE_STATE::COHERENCY_TYPE_GPU_COHERENT);
surfaceState->setCoherencyType(R_SURFACE_STATE::COHERENCY_TYPE_GPU_COHERENT);
surfaceState->setAuxiliarySurfaceMode(AUXILIARY_SURFACE_MODE::AUXILIARY_SURFACE_MODE_AUX_CCS_E);
}
if (DebugManager.flags.DisableCachingForStatefulBufferAccess.get()) {
surfaceState->setMemoryObjectControlState(gmmHelper->getMOCS(GMM_RESOURCE_USAGE_OCL_BUFFER_CACHELINE_MISALIGNED));
}
EncodeSurfaceState<Family>::encodeExtraBufferParams(surfaceState, allocation, gmmHelper, numAvailableDevices);
}
template <typename Family>
void *EncodeDispatchKernel<Family>::getInterfaceDescriptor(CommandContainer &container, uint32_t &iddOffset) {

5
shared/source/command_container/command_encoder_base.inl
View File

@ -388,4 +388,9 @@ inline size_t EncodeWA<GfxFamily>::getAdditionalPipelineSelectSize(Device &devic
return 0;
}
template <typename GfxFamily>
void EncodeSurfaceState<GfxFamily>::encodeExtraBufferParams(R_SURFACE_STATE *surfaceState, GraphicsAllocation *allocation, GmmHelper *gmmHelper,
uint32_t numAvailableDevices) {
}
} // namespace NEO

9
shared/test/unit_test/encoders/test_encode_states.cpp
View File

@ -47,7 +47,8 @@ HWTEST_F(CommandEncodeStatesTest, givenCreatedSurfaceStateBufferWhenAllocationPr
length.Length = static_cast<uint32_t>(allocSize - 1);
GraphicsAllocation allocation(0, GraphicsAllocation::AllocationType::UNKNOWN, cpuAddr, gpuAddr, 0u, allocSize, MemoryPool::MemoryNull, 1);
EncodeSurfaceState<FamilyType>::encodeBuffer(stateBuffer, gpuAddr, allocSize, 1,
RENDER_SURFACE_STATE::COHERENCY_TYPE_IA_COHERENT);
true, false, 1u,
&allocation, pDevice->getGmmHelper());
EXPECT_EQ(length.SurfaceState.Depth + 1u, state->getDepth());
EXPECT_EQ(length.SurfaceState.Width + 1u, state->getWidth());
EXPECT_EQ(length.SurfaceState.Height + 1u, state->getHeight());
@ -73,7 +74,8 @@ HWTEST_F(CommandEncodeStatesTest, givenCreatedSurfaceStateBufferWhenAllocationNo
length.Length = static_cast<uint32_t>(allocSize - 1);
EncodeSurfaceState<FamilyType>::encodeBuffer(stateBuffer, gpuAddr, allocSize, 1,
RENDER_SURFACE_STATE::COHERENCY_TYPE_IA_COHERENT);
true, false, 1u,
nullptr, pDevice->getGmmHelper());
EXPECT_EQ(RENDER_SURFACE_STATE::SURFACE_TYPE_SURFTYPE_NULL, state->getSurfaceType());
@ -97,7 +99,8 @@ HWTEST_F(CommandEncodeStatesTest, givenCreatedSurfaceStateBufferWhenGpuCoherency
length.Length = static_cast<uint32_t>(allocSize - 1);
EncodeSurfaceState<FamilyType>::encodeBuffer(stateBuffer, gpuAddr, allocSize, 1,
RENDER_SURFACE_STATE::COHERENCY_TYPE_GPU_COHERENT);
false, false, 1u,
nullptr, pDevice->getGmmHelper());
EXPECT_EQ(RENDER_SURFACE_STATE::COHERENCY_TYPE_GPU_COHERENT, state->getCoherencyType());