compute-runtime/shared/source/command_stream/preemption.inl

157 lines
6.1 KiB
C++

/*
* Copyright (C) 2018-2024 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "shared/source/built_ins/sip.h"
#include "shared/source/command_stream/linear_stream.h"
#include "shared/source/command_stream/preemption.h"
#include "shared/source/command_stream/preemption_mode.h"
#include "shared/source/device/device.h"
#include "shared/source/execution_environment/execution_environment.h"
#include "shared/source/helpers/compiler_product_helper.h"
#include "shared/source/helpers/gfx_core_helper.h"
#include "shared/source/helpers/pipe_control_args.h"
#include "shared/source/helpers/preamble.h"
#include "shared/source/memory_manager/graphics_allocation.h"
namespace NEO {
template <typename GfxFamily>
void PreemptionHelper::programCsrBaseAddress(LinearStream &preambleCmdStream, Device &device, const GraphicsAllocation *preemptionCsr) {
if (device.getPreemptionMode() == PreemptionMode::MidThread) {
UNRECOVERABLE_IF(nullptr == preemptionCsr);
programCsrBaseAddressCmd<GfxFamily>(preambleCmdStream, preemptionCsr);
}
}
template <typename GfxFamily>
void PreemptionHelper::programCsrBaseAddressCmd(LinearStream &preambleCmdStream, const GraphicsAllocation *preemptionCsr) {
using GPGPU_CSR_BASE_ADDRESS = typename GfxFamily::GPGPU_CSR_BASE_ADDRESS;
auto csr = reinterpret_cast<GPGPU_CSR_BASE_ADDRESS *>(preambleCmdStream.getSpace(sizeof(GPGPU_CSR_BASE_ADDRESS)));
GPGPU_CSR_BASE_ADDRESS cmd = GfxFamily::cmdInitGpgpuCsrBaseAddress;
cmd.setGpgpuCsrBaseAddress(preemptionCsr->getGpuAddressToPatch());
*csr = cmd;
}
template <typename GfxFamily>
void PreemptionHelper::programStateSip(LinearStream &preambleCmdStream, Device &device, OsContext *context) {
using STATE_SIP = typename GfxFamily::STATE_SIP;
bool debuggingEnabled = device.getDebugger() != nullptr;
bool isMidThreadPreemption = device.getPreemptionMode() == PreemptionMode::MidThread;
auto &compilerProductHelper = device.getCompilerProductHelper();
bool useFullAddress = compilerProductHelper.isHeaplessModeEnabled();
if (isMidThreadPreemption || debuggingEnabled) {
GraphicsAllocation *sipAllocation = SipKernel::getSipKernel(device, context).getSipAllocation();
programStateSipCmd<GfxFamily>(preambleCmdStream, sipAllocation, useFullAddress);
}
}
template <typename GfxFamily>
void PreemptionHelper::programStateSipCmd(LinearStream &preambleCmdStream, GraphicsAllocation *sipAllocation, bool useFullAddress) {
using STATE_SIP = typename GfxFamily::STATE_SIP;
auto sip = reinterpret_cast<STATE_SIP *>(preambleCmdStream.getSpace(sizeof(STATE_SIP)));
STATE_SIP cmd = GfxFamily::cmdInitStateSip;
if (useFullAddress) {
cmd.setSystemInstructionPointer(sipAllocation->getGpuAddress());
} else {
cmd.setSystemInstructionPointer(sipAllocation->getGpuAddressToPatch());
}
*sip = cmd;
}
template <typename GfxFamily>
void PreemptionHelper::programCmdStream(LinearStream &cmdStream, PreemptionMode newPreemptionMode,
PreemptionMode oldPreemptionMode, GraphicsAllocation *preemptionCsr) {
if (oldPreemptionMode == newPreemptionMode) {
return;
}
uint32_t regVal = 0;
if (newPreemptionMode == PreemptionMode::MidThread) {
regVal = PreemptionConfig<GfxFamily>::midThreadVal | PreemptionConfig<GfxFamily>::mask;
} else if (newPreemptionMode == PreemptionMode::ThreadGroup) {
regVal = PreemptionConfig<GfxFamily>::threadGroupVal | PreemptionConfig<GfxFamily>::mask;
} else {
regVal = PreemptionConfig<GfxFamily>::cmdLevelVal | PreemptionConfig<GfxFamily>::mask;
}
LriHelper<GfxFamily>::program(&cmdStream, PreemptionConfig<GfxFamily>::mmioAddress, regVal, true, false);
}
template <typename GfxFamily>
size_t PreemptionHelper::getRequiredCmdStreamSize(PreemptionMode newPreemptionMode, PreemptionMode oldPreemptionMode) {
if (newPreemptionMode == oldPreemptionMode) {
return 0;
}
return sizeof(typename GfxFamily::MI_LOAD_REGISTER_IMM);
}
template <typename GfxFamily>
size_t PreemptionHelper::getRequiredPreambleSize(const Device &device) {
if (device.getPreemptionMode() == PreemptionMode::MidThread) {
return sizeof(typename GfxFamily::GPGPU_CSR_BASE_ADDRESS);
}
return 0;
}
template <typename GfxFamily>
size_t PreemptionHelper::getRequiredStateSipCmdSize(Device &device, bool isRcs) {
size_t size = 0;
bool isMidThreadPreemption = device.getPreemptionMode() == PreemptionMode::MidThread;
bool debuggingEnabled = device.getDebugger() != nullptr;
if (isMidThreadPreemption || debuggingEnabled) {
size += sizeof(typename GfxFamily::STATE_SIP);
}
return size;
}
template <typename GfxFamily>
size_t PreemptionHelper::getPreemptionWaCsSize(const Device &device) {
return 0u;
}
template <typename GfxFamily>
void PreemptionHelper::applyPreemptionWaCmdsBegin(LinearStream *pCommandStream, const Device &device) {
}
template <typename GfxFamily>
void PreemptionHelper::applyPreemptionWaCmdsEnd(LinearStream *pCommandStream, const Device &device) {
}
template <typename GfxFamily, typename InterfaceDescriptorType>
void PreemptionHelper::programInterfaceDescriptorDataPreemption(InterfaceDescriptorType *idd, PreemptionMode preemptionMode) {
using INTERFACE_DESCRIPTOR_DATA = typename GfxFamily::INTERFACE_DESCRIPTOR_DATA;
if (preemptionMode == PreemptionMode::MidThread) {
idd->setThreadPreemptionDisable(INTERFACE_DESCRIPTOR_DATA::THREAD_PREEMPTION_DISABLE_DISABLE);
} else {
idd->setThreadPreemptionDisable(INTERFACE_DESCRIPTOR_DATA::THREAD_PREEMPTION_DISABLE_ENABLE);
}
}
template <typename GfxFamily>
constexpr uint32_t PreemptionConfig<GfxFamily>::mmioAddress = 0x2580;
template <typename GfxFamily>
constexpr uint32_t PreemptionConfig<GfxFamily>::mask = ((1 << 1) | (1 << 2)) << 16;
template <typename GfxFamily>
constexpr uint32_t PreemptionConfig<GfxFamily>::threadGroupVal = (1 << 1);
template <typename GfxFamily>
constexpr uint32_t PreemptionConfig<GfxFamily>::cmdLevelVal = (1 << 2);
template <typename GfxFamily>
constexpr uint32_t PreemptionConfig<GfxFamily>::midThreadVal = 0;
} // namespace NEO