Files
compute-runtime/runtime/command_stream/submissions_aggregator.cpp
Pawel Wilma c4c35955d1 Debug option to flatten main and chained batch buffer in AUB dump.
New debug option FlattenBatchBufferForAUBDump has been added. When set it
modifies AUB dump in such way that commands from main and chained batch
buffer are dumped as single allocation. Commands from chained batch buffer are
dumped directly after commands from main batch buffer without
MI_BATCH_BUFFER_START. This feature also requires ImmediateDispatch mode which
can be forced using debug option CsrDispatchMode = 1.

Change-Id: I730760791693a748e7f4e1463ce8e7af94287b93
2018-03-06 12:26:50 +01:00

115 lines
5.2 KiB
C++

/*
* Copyright (c) 2017 - 2018, Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included
* in all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*/
#include "submissions_aggregator.h"
#include "runtime/helpers/flush_stamp.h"
void OCLRT::SubmissionAggregator::recordCommandBuffer(CommandBuffer *commandBuffer) {
this->cmdBuffers.pushTailOne(*commandBuffer);
}
void OCLRT::SubmissionAggregator::aggregateCommandBuffers(ResourcePackage &resourcePackage, size_t &totalUsedSize, size_t totalMemoryBudget) {
auto primaryCommandBuffer = this->cmdBuffers.peekHead();
auto currentInspection = this->inspectionId;
if (!primaryCommandBuffer) {
return;
}
auto primaryBatchGraphicsAllocation = primaryCommandBuffer->batchBuffer.commandBufferAllocation;
this->inspectionId++;
primaryCommandBuffer->inspectionId = currentInspection;
//primary command buffers must fix to budget
for (auto &graphicsAllocation : primaryCommandBuffer->surfaces) {
if (graphicsAllocation->inspectionId < currentInspection) {
graphicsAllocation->inspectionId = currentInspection;
resourcePackage.push_back(graphicsAllocation);
totalUsedSize += graphicsAllocation->getUnderlyingBufferSize();
}
}
//check if we have anything for merge
if (!primaryCommandBuffer->next) {
return;
}
//check if next cmd buffer is compatible
if (primaryCommandBuffer->next->batchBuffer.requiresCoherency != primaryCommandBuffer->batchBuffer.requiresCoherency) {
return;
}
if (primaryCommandBuffer->next->batchBuffer.low_priority != primaryCommandBuffer->batchBuffer.low_priority) {
return;
}
if (primaryCommandBuffer->next->batchBuffer.throttle != primaryCommandBuffer->batchBuffer.throttle) {
return;
}
auto nextCommandBuffer = primaryCommandBuffer->next;
ResourcePackage newResources;
while (nextCommandBuffer) {
size_t nextCommandBufferNewResourcesSize = 0;
//evaluate if buffer fits
for (auto &graphicsAllocation : nextCommandBuffer->surfaces) {
if (graphicsAllocation == primaryBatchGraphicsAllocation) {
continue;
}
if (graphicsAllocation->inspectionId < currentInspection) {
graphicsAllocation->inspectionId = currentInspection;
newResources.push_back(graphicsAllocation);
nextCommandBufferNewResourcesSize += graphicsAllocation->getUnderlyingBufferSize();
}
}
if (nextCommandBuffer->batchBuffer.commandBufferAllocation && (nextCommandBuffer->batchBuffer.commandBufferAllocation != primaryBatchGraphicsAllocation)) {
if (nextCommandBuffer->batchBuffer.commandBufferAllocation->inspectionId < currentInspection) {
nextCommandBuffer->batchBuffer.commandBufferAllocation->inspectionId = currentInspection;
newResources.push_back(nextCommandBuffer->batchBuffer.commandBufferAllocation);
nextCommandBufferNewResourcesSize += nextCommandBuffer->batchBuffer.commandBufferAllocation->getUnderlyingBufferSize();
}
}
if (nextCommandBufferNewResourcesSize + totalUsedSize <= totalMemoryBudget) {
auto currentNode = nextCommandBuffer;
nextCommandBuffer = nextCommandBuffer->next;
totalUsedSize += nextCommandBufferNewResourcesSize;
currentNode->inspectionId = currentInspection;
for (auto &newResource : newResources) {
resourcePackage.push_back(newResource);
}
newResources.clear();
} else {
break;
}
}
}
OCLRT::BatchBuffer::BatchBuffer(GraphicsAllocation *commandBufferAllocation, size_t startOffset, size_t chainedBatchBufferStartOffset, GraphicsAllocation *chainedBatchBuffer, bool requiresCoherency, bool lowPriority, QueueThrottle throttle, size_t usedSize, LinearStream *stream) : commandBufferAllocation(commandBufferAllocation), startOffset(startOffset), chainedBatchBufferStartOffset(chainedBatchBufferStartOffset), chainedBatchBuffer(chainedBatchBuffer), requiresCoherency(requiresCoherency), low_priority(lowPriority), throttle(throttle), usedSize(usedSize), stream(stream) {
}
OCLRT::CommandBuffer::CommandBuffer() {
flushStamp.reset(new FlushStampTracker(false));
}