compute-runtime/shared/source/direct_submission/linux/drm_direct_submission.inl

259 lines
10 KiB
C++

/*
* Copyright (C) 2020-2022 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "shared/source/command_container/implicit_scaling.h"
#include "shared/source/command_stream/linear_stream.h"
#include "shared/source/device/device.h"
#include "shared/source/direct_submission/linux/drm_direct_submission.h"
#include "shared/source/os_interface/linux/drm_allocation.h"
#include "shared/source/os_interface/linux/drm_buffer_object.h"
#include "shared/source/os_interface/linux/drm_neo.h"
#include "shared/source/os_interface/linux/drm_wrappers.h"
#include "shared/source/os_interface/linux/os_context_linux.h"
#include "shared/source/utilities/wait_util.h"
#include <memory>
namespace NEO {
template <typename GfxFamily, typename Dispatcher>
DrmDirectSubmission<GfxFamily, Dispatcher>::DrmDirectSubmission(const DirectSubmissionInputParams &inputParams)
: DirectSubmissionHw<GfxFamily, Dispatcher>(inputParams) {
this->disableMonitorFence = true;
if (DebugManager.flags.DirectSubmissionDisableMonitorFence.get() != -1) {
this->disableMonitorFence = DebugManager.flags.DirectSubmissionDisableMonitorFence.get();
}
auto osContextLinux = static_cast<OsContextLinux *>(&this->osContext);
auto subDevices = osContextLinux->getDeviceBitfield();
bool dispatcherSupport = Dispatcher::isMultiTileSynchronizationSupported();
if (ImplicitScalingHelper::isImplicitScalingEnabled(subDevices, true) && dispatcherSupport) {
this->activeTiles = static_cast<uint32_t>(subDevices.count());
}
this->partitionedMode = this->activeTiles > 1u;
this->partitionConfigSet = !this->partitionedMode;
auto &drm = osContextLinux->getDrm();
drm.setDirectSubmissionActive(true);
if (this->partitionedMode) {
this->workPartitionAllocation = inputParams.workPartitionAllocation;
UNRECOVERABLE_IF(this->workPartitionAllocation == nullptr);
}
if (this->miMemFenceRequired || drm.completionFenceSupport()) {
this->completionFenceAllocation = inputParams.completionFenceAllocation;
if (this->completionFenceAllocation) {
this->gpuVaForAdditionalSynchronizationWA = this->completionFenceAllocation->getGpuAddress() + 8u;
if (drm.completionFenceSupport()) {
this->completionFenceSupported = true;
}
if (DebugManager.flags.PrintCompletionFenceUsage.get()) {
std::cout << "Completion fence for DirectSubmission:"
<< " GPU address: " << std::hex << (this->completionFenceAllocation->getGpuAddress() + Drm::completionFenceOffset)
<< ", CPU address: " << (castToUint64(this->completionFenceAllocation->getUnderlyingBuffer()) + Drm::completionFenceOffset)
<< std::dec << std::endl;
}
}
}
}
template <typename GfxFamily, typename Dispatcher>
inline DrmDirectSubmission<GfxFamily, Dispatcher>::~DrmDirectSubmission() {
if (this->ringStart) {
this->stopRingBuffer();
this->wait(static_cast<uint32_t>(this->currentTagData.tagValue));
}
if (this->isCompletionFenceSupported()) {
auto osContextLinux = static_cast<OsContextLinux *>(&this->osContext);
auto &drm = osContextLinux->getDrm();
auto completionFenceCpuAddress = reinterpret_cast<uint64_t>(this->completionFenceAllocation->getUnderlyingBuffer()) + Drm::completionFenceOffset;
drm.waitOnUserFences(*osContextLinux, completionFenceCpuAddress, this->completionFenceValue, this->activeTiles, this->postSyncOffset);
}
this->deallocateResources();
}
template <typename GfxFamily, typename Dispatcher>
uint32_t *DrmDirectSubmission<GfxFamily, Dispatcher>::getCompletionValuePointer() {
if (this->isCompletionFenceSupported()) {
return &this->completionFenceValue;
}
return DirectSubmissionHw<GfxFamily, Dispatcher>::getCompletionValuePointer();
}
template <typename GfxFamily, typename Dispatcher>
bool DrmDirectSubmission<GfxFamily, Dispatcher>::allocateOsResources() {
this->currentTagData.tagAddress = this->semaphoreGpuVa + offsetof(RingSemaphoreData, tagAllocation);
this->currentTagData.tagValue = 0u;
this->tagAddress = reinterpret_cast<volatile uint32_t *>(reinterpret_cast<uint8_t *>(this->semaphorePtr) + offsetof(RingSemaphoreData, tagAllocation));
return true;
}
template <typename GfxFamily, typename Dispatcher>
bool DrmDirectSubmission<GfxFamily, Dispatcher>::submit(uint64_t gpuAddress, size_t size) {
auto bb = static_cast<DrmAllocation *>(this->ringCommandStream.getGraphicsAllocation())->getBO();
auto osContextLinux = static_cast<OsContextLinux *>(&this->osContext);
auto &drm = osContextLinux->getDrm();
auto execFlags = osContextLinux->getEngineFlag() | drm.getIoctlHelper()->getDrmParamValue(DrmParam::ExecNoReloc);
auto &drmContextIds = osContextLinux->getDrmContextIds();
ExecObject execObject{};
this->handleResidency();
auto currentBase = this->ringCommandStream.getGraphicsAllocation()->getGpuAddress();
auto offset = ptrDiff(gpuAddress, currentBase);
bool ret = false;
uint32_t drmContextId = 0u;
uint32_t completionValue = 0u;
uint64_t completionFenceGpuAddress = 0u;
if (this->isCompletionFenceSupported()) {
completionValue = ++completionFenceValue;
completionFenceGpuAddress = this->completionFenceAllocation->getGpuAddress() + Drm::completionFenceOffset;
}
for (auto drmIterator = 0u; drmIterator < osContextLinux->getDeviceBitfield().size(); drmIterator++) {
if (osContextLinux->getDeviceBitfield().test(drmIterator)) {
ret |= !!bb->exec(static_cast<uint32_t>(size),
offset,
execFlags,
false,
&this->osContext,
drmIterator,
drmContextIds[drmContextId],
nullptr,
0,
&execObject,
completionFenceGpuAddress,
completionValue);
drmContextId++;
if (completionFenceGpuAddress) {
completionFenceGpuAddress += this->postSyncOffset;
}
}
}
return !ret;
}
template <typename GfxFamily, typename Dispatcher>
bool DrmDirectSubmission<GfxFamily, Dispatcher>::handleResidency() {
auto osContextLinux = static_cast<OsContextLinux *>(&this->osContext);
osContextLinux->waitForPagingFence();
return true;
}
template <typename GfxFamily, typename Dispatcher>
bool DrmDirectSubmission<GfxFamily, Dispatcher>::isNewResourceHandleNeeded() {
auto osContextLinux = static_cast<OsContextLinux *>(&this->osContext);
auto newResourcesBound = osContextLinux->isTlbFlushRequired();
if (DebugManager.flags.DirectSubmissionNewResourceTlbFlush.get() != -1) {
newResourcesBound = DebugManager.flags.DirectSubmissionNewResourceTlbFlush.get();
}
return newResourcesBound;
}
template <typename GfxFamily, typename Dispatcher>
void DrmDirectSubmission<GfxFamily, Dispatcher>::handleNewResourcesSubmission() {
if (isNewResourceHandleNeeded()) {
auto osContextLinux = static_cast<OsContextLinux *>(&this->osContext);
auto tlbFlushCounter = osContextLinux->peekTlbFlushCounter();
Dispatcher::dispatchTlbFlush(this->ringCommandStream, this->gpuVaForMiFlush, *this->hwInfo);
osContextLinux->setTlbFlushed(tlbFlushCounter);
}
}
template <typename GfxFamily, typename Dispatcher>
size_t DrmDirectSubmission<GfxFamily, Dispatcher>::getSizeNewResourceHandler() {
// Overestimate to avoid race
return Dispatcher::getSizeTlbFlush();
}
template <typename GfxFamily, typename Dispatcher>
void DrmDirectSubmission<GfxFamily, Dispatcher>::handleStopRingBuffer() {
if (this->disableMonitorFence) {
this->currentTagData.tagValue++;
}
}
template <typename GfxFamily, typename Dispatcher>
void DrmDirectSubmission<GfxFamily, Dispatcher>::handleSwitchRingBuffers() {
if (this->disableMonitorFence) {
this->currentTagData.tagValue++;
bool updateCompletionFences = this->ringStart;
if (DebugManager.flags.EnableRingSwitchTagUpdateWa.get() == 0) {
updateCompletionFences = true;
}
if (updateCompletionFences) {
this->ringBuffers[this->previousRingBuffer].completionFence = this->currentTagData.tagValue;
}
}
if (this->ringStart) {
if (this->ringBuffers[this->currentRingBuffer].completionFence != 0) {
this->wait(static_cast<uint32_t>(this->ringBuffers[this->currentRingBuffer].completionFence));
}
}
}
template <typename GfxFamily, typename Dispatcher>
uint64_t DrmDirectSubmission<GfxFamily, Dispatcher>::updateTagValue() {
if (!this->disableMonitorFence) {
this->currentTagData.tagValue++;
this->ringBuffers[this->currentRingBuffer].completionFence = this->currentTagData.tagValue;
}
return 0ull;
}
template <typename GfxFamily, typename Dispatcher>
void DrmDirectSubmission<GfxFamily, Dispatcher>::getTagAddressValue(TagData &tagData) {
tagData.tagAddress = this->currentTagData.tagAddress;
tagData.tagValue = this->currentTagData.tagValue + 1;
}
template <typename GfxFamily, typename Dispatcher>
inline bool DrmDirectSubmission<GfxFamily, Dispatcher>::isCompleted(uint32_t ringBufferIndex) {
auto taskCount = this->ringBuffers[ringBufferIndex].completionFence;
auto pollAddress = this->tagAddress;
for (uint32_t i = 0; i < this->activeTiles; i++) {
if (*pollAddress < taskCount) {
return false;
}
pollAddress = ptrOffset(pollAddress, this->postSyncOffset);
}
return true;
}
template <typename GfxFamily, typename Dispatcher>
bool DrmDirectSubmission<GfxFamily, Dispatcher>::isCompletionFenceSupported() {
return this->completionFenceSupported;
}
template <typename GfxFamily, typename Dispatcher>
void DrmDirectSubmission<GfxFamily, Dispatcher>::wait(uint32_t taskCountToWait) {
auto pollAddress = this->tagAddress;
for (uint32_t i = 0; i < this->activeTiles; i++) {
while (!WaitUtils::waitFunction(pollAddress, taskCountToWait)) {
}
pollAddress = ptrOffset(pollAddress, this->postSyncOffset);
}
}
} // namespace NEO