compute-runtime/shared/test/unit_test/cmd_parse/hw_parse.h

232 lines
8.6 KiB
C++

/*
* Copyright (C) 2017-2020 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#pragma once
#include "shared/source/command_stream/command_stream_receiver.h"
#include "shared/source/command_stream/linear_stream.h"
#include "shared/source/helpers/pipeline_select_helper.h"
#include "shared/source/helpers/ptr_math.h"
#include "shared/test/unit_test/cmd_parse/gen_cmd_parse.h"
#include "shared/test/unit_test/helpers/default_hw_info.h"
#include "opencl/source/command_queue/command_queue.h"
#include "opencl/source/kernel/kernel.h"
#include "gtest/gtest.h"
namespace NEO {
struct HardwareParse {
HardwareParse() {
itorMediaInterfaceDescriptorLoad = cmdList.end();
itorMediaVfeState = cmdList.end();
itorPipelineSelect = cmdList.end();
itorStateBaseAddress = cmdList.end();
itorWalker = cmdList.end();
itorGpgpuCsrBaseAddress = cmdList.end();
}
void SetUp() {
}
void TearDown() {
cmdList.clear();
lriList.clear();
pipeControlList.clear();
}
template <typename CmdType>
GenCmdList getCommandsList() {
GenCmdList list;
for (auto it = cmdList.begin(); it != cmdList.end(); it++) {
auto cmd = genCmdCast<CmdType *>(*it);
if (cmd) {
list.push_back(*it);
}
}
return list;
}
template <typename FamilyType>
void findCsrBaseAddress();
template <typename FamilyType>
void findHardwareCommands();
template <typename FamilyType>
void findHardwareCommands(IndirectHeap *dsh);
template <typename FamilyType>
void parseCommands(NEO::LinearStream &commandStream, size_t startOffset = 0) {
ASSERT_LE(startOffset, commandStream.getUsed());
auto sizeToParse = commandStream.getUsed() - startOffset;
ASSERT_TRUE(FamilyType::PARSE::parseCommandBuffer(
cmdList,
ptrOffset(commandStream.getCpuBase(), startOffset),
sizeToParse));
}
template <typename FamilyType>
void parseCommands(NEO::CommandQueue &commandQueue) {
auto &commandStreamReceiver = commandQueue.getGpgpuCommandStreamReceiver();
auto &commandStreamCSR = commandStreamReceiver.getCS();
parseCommands<FamilyType>(commandStreamCSR, startCSRCS);
startCSRCS = commandStreamCSR.getUsed();
auto &commandStream = commandQueue.getCS(1024);
if (previousCS != &commandStream) {
startCS = 0;
}
parseCommands<FamilyType>(commandStream, startCS);
startCS = commandStream.getUsed();
previousCS = &commandStream;
sizeUsed = commandStream.getUsed();
findHardwareCommands<FamilyType>(&commandStreamReceiver.getIndirectHeap(IndirectHeap::DYNAMIC_STATE, 0));
}
template <typename FamilyType>
const typename FamilyType::RENDER_SURFACE_STATE &getSurfaceState(IndirectHeap *ssh, uint32_t index) {
typedef typename FamilyType::BINDING_TABLE_STATE BINDING_TABLE_STATE;
typedef typename FamilyType::INTERFACE_DESCRIPTOR_DATA INTERFACE_DESCRIPTOR_DATA;
typedef typename FamilyType::RENDER_SURFACE_STATE RENDER_SURFACE_STATE;
typedef typename FamilyType::STATE_BASE_ADDRESS STATE_BASE_ADDRESS;
const auto &interfaceDescriptorData = *(INTERFACE_DESCRIPTOR_DATA *)cmdInterfaceDescriptorData;
auto cmdSBA = (STATE_BASE_ADDRESS *)cmdStateBaseAddress;
auto surfaceStateHeap = cmdSBA->getSurfaceStateBaseAddress();
if (ssh && (ssh->getHeapGpuBase() == surfaceStateHeap)) {
surfaceStateHeap = reinterpret_cast<uint64_t>(ssh->getCpuBase());
}
EXPECT_NE(0u, surfaceStateHeap);
auto bindingTablePointer = interfaceDescriptorData.getBindingTablePointer();
const auto &bindingTableState = reinterpret_cast<BINDING_TABLE_STATE *>(surfaceStateHeap + bindingTablePointer)[index];
auto surfaceStatePointer = bindingTableState.getSurfaceStatePointer();
return *(RENDER_SURFACE_STATE *)(surfaceStateHeap + surfaceStatePointer);
}
template <typename FamilyType>
const typename FamilyType::SAMPLER_STATE &getSamplerState(uint32_t index) {
typedef typename FamilyType::INTERFACE_DESCRIPTOR_DATA INTERFACE_DESCRIPTOR_DATA;
typedef typename FamilyType::SAMPLER_STATE SAMPLER_STATE;
typedef typename FamilyType::STATE_BASE_ADDRESS STATE_BASE_ADDRESS;
const auto &interfaceDescriptorData = *(INTERFACE_DESCRIPTOR_DATA *)cmdInterfaceDescriptorData;
auto cmdSBA = (STATE_BASE_ADDRESS *)cmdStateBaseAddress;
auto dynamicStateHeap = cmdSBA->getDynamicStateBaseAddress();
EXPECT_NE(0, dynamicStateHeap);
const auto samplerState = reinterpret_cast<SAMPLER_STATE *>(dynamicStateHeap + interfaceDescriptorData.getSamplerStatePointer());
return samplerState[index];
}
template <typename FamilyType>
const void *getStatelessArgumentPointer(const Kernel &kernel, uint32_t indexArg, IndirectHeap &ioh);
template <typename CmdType>
CmdType *getCommand(GenCmdList::iterator itorStart, GenCmdList::iterator itorEnd) {
auto itorCmd = find<CmdType *>(itorStart, itorEnd);
return itorCmd != cmdList.end()
? genCmdCast<CmdType *>(*itorCmd)
: nullptr;
}
template <typename CmdType>
CmdType *getCommand() {
return getCommand<CmdType>(cmdList.begin(), cmdList.end());
}
template <typename FamilyType>
int getNumberOfPipelineSelectsThatEnablePipelineSelect() {
typedef typename FamilyType::PIPELINE_SELECT PIPELINE_SELECT;
int numberOfGpgpuSelects = 0;
int numberOf3dSelects = 0;
auto itorCmd = find<PIPELINE_SELECT *>(cmdList.begin(), cmdList.end());
while (itorCmd != cmdList.end()) {
auto cmd = getCommand<PIPELINE_SELECT>(itorCmd, cmdList.end());
if (cmd->getPipelineSelection() == PIPELINE_SELECT::PIPELINE_SELECTION_GPGPU &&
pipelineSelectEnablePipelineSelectMaskBits == (pipelineSelectEnablePipelineSelectMaskBits & cmd->getMaskBits())) {
numberOfGpgpuSelects++;
}
if (cmd->getPipelineSelection() == PIPELINE_SELECT::PIPELINE_SELECTION_3D &&
pipelineSelectEnablePipelineSelectMaskBits == (pipelineSelectEnablePipelineSelectMaskBits & cmd->getMaskBits())) {
numberOf3dSelects++;
}
itorCmd = find<PIPELINE_SELECT *>(++itorCmd, cmdList.end());
}
auto &hwHelper = HwHelper::get(defaultHwInfo->platform.eRenderCoreFamily);
if (hwHelper.is3DPipelineSelectWARequired(*defaultHwInfo)) {
auto maximalNumberOf3dSelectsRequired = 2;
EXPECT_LE(numberOf3dSelects, maximalNumberOf3dSelectsRequired);
EXPECT_EQ(numberOf3dSelects, numberOfGpgpuSelects);
auto numberOfGpgpuSelectsAddedByWa = numberOf3dSelects - 1;
numberOfGpgpuSelects -= numberOfGpgpuSelectsAddedByWa;
} else {
EXPECT_EQ(0, numberOf3dSelects);
}
return numberOfGpgpuSelects;
}
template <typename CmdType>
uint32_t getCommandCount() {
GenCmdList::iterator cmdItor = cmdList.begin();
uint32_t cmdCount = 0;
do {
cmdItor = find<CmdType *>(cmdItor, cmdList.end());
if (cmdItor != cmdList.end()) {
++cmdCount;
++cmdItor;
}
} while (cmdItor != cmdList.end());
return cmdCount;
}
template <typename FamilyType>
static const char *getCommandName(void *cmd) {
return FamilyType::PARSE::getCommandName(cmd);
}
// The starting point of parsing commandBuffers. This is important
// because as buffers get reused, we only want to parse the deltas.
LinearStream *previousCS = nullptr;
size_t startCS = 0u;
size_t startCSRCS = 0u;
size_t sizeUsed = 0u;
GenCmdList cmdList;
GenCmdList lriList;
GenCmdList pipeControlList;
GenCmdList::iterator itorMediaInterfaceDescriptorLoad;
GenCmdList::iterator itorMediaVfeState;
GenCmdList::iterator itorPipelineSelect;
GenCmdList::iterator itorStateBaseAddress;
GenCmdList::iterator itorWalker;
GenCmdList::iterator itorBBStartAfterWalker;
GenCmdList::iterator itorGpgpuCsrBaseAddress;
void *cmdInterfaceDescriptorData = nullptr;
void *cmdMediaInterfaceDescriptorLoad = nullptr;
void *cmdMediaVfeState = nullptr;
void *cmdPipelineSelect = nullptr;
void *cmdStateBaseAddress = nullptr;
void *cmdWalker = nullptr;
void *cmdBBStartAfterWalker = nullptr;
void *cmdGpgpuCsrBaseAddress = nullptr;
bool parsePipeControl = false;
};
} // namespace NEO