intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 06:49:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

Enable blit operation on enqueue copy buffer

Browse Source 
Change-Id: Ie9571344a9b1e2a783c59955a68094f2868f36ee
Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
This commit is contained in:
Dunajski, Bartosz
2019-11-19 11:54:11 +01:00
committed by sys_ocldev
parent 61e91a376b
commit 08ceaf432a
7 changed files with 97 additions and 45 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
unit_tests/command_queue/enqueue_command_without_kernel_tests.cpp
View File

@@ -155,7 +155,7 @@ HWTEST_F(DispatchFlagsTests, givenBlitEnqueueWhenDispatchingCommandsWithoutKerne
mockCmdQ->obtainNewTimestampPacketNodes(1, timestampPacketDependencies.previousEnqueueNodes, true);
BlitProperties blitProperties = mockCmdQ->processDispatchForBlitEnqueue(multiDispatchInfo, timestampPacketDependencies,
eventsRequest, mockCmdQ->getCS(0), 0, false);
eventsRequest, mockCmdQ->getCS(0), CL_COMMAND_READ_BUFFER, false);
BlitPropertiesContainer blitPropertiesContainer;
blitPropertiesContainer.push_back(blitProperties);
@@ -194,7 +194,7 @@ HWTEST_F(DispatchFlagsTests, givenN1EnabledWhenDispatchingWithoutKernelTheAllowO
mockCmdQ->obtainNewTimestampPacketNodes(1, timestampPacketDependencies.previousEnqueueNodes, true);
BlitProperties blitProperties = mockCmdQ->processDispatchForBlitEnqueue(multiDispatchInfo, timestampPacketDependencies,
eventsRequest, mockCmdQ->getCS(0), 0, false);
eventsRequest, mockCmdQ->getCS(0), CL_COMMAND_READ_BUFFER, false);
BlitPropertiesContainer blitPropertiesContainer;
blitPropertiesContainer.push_back(blitProperties);
EnqueueProperties enqueueProperties(true, false, false, false, &blitPropertiesContainer);

16
unit_tests/command_stream/command_stream_receiver_hw_tests.cpp
View File

@@ -855,8 +855,8 @@ HWTEST_F(BcsTests, givenMapAllocationInBuiltinOpParamsWhenConstructingThenUseItA
builtinOpParams.size.x = 1;
builtinOpParams.transferAllocation = mapAllocation;
auto blitProperties = BlitProperties::constructPropertiesForReadWriteBuffer(BlitterConstants::BlitDirection::HostPtrToBuffer,
csr, builtinOpParams);
auto blitProperties = BlitProperties::constructProperties(BlitterConstants::BlitDirection::HostPtrToBuffer,
csr, builtinOpParams);
EXPECT_EQ(mapAllocation, blitProperties.srcAllocation);
}
{
@@ -867,8 +867,8 @@ HWTEST_F(BcsTests, givenMapAllocationInBuiltinOpParamsWhenConstructingThenUseItA
builtinOpParams.size.x = 1;
builtinOpParams.transferAllocation = mapAllocation;
auto blitProperties = BlitProperties::constructPropertiesForReadWriteBuffer(BlitterConstants::BlitDirection::BufferToHostPtr,
csr, builtinOpParams);
auto blitProperties = BlitProperties::constructProperties(BlitterConstants::BlitDirection::BufferToHostPtr,
csr, builtinOpParams);
EXPECT_EQ(mapAllocation, blitProperties.dstAllocation);
}
@@ -896,8 +896,8 @@ HWTEST_F(BcsTests, givenNonZeroCopySvmAllocationWhenConstructingBlitPropertiesFo
builtinOpParams.srcPtr = reinterpret_cast<void *>(svmData->cpuAllocation->getGpuAddress());
builtinOpParams.size.x = 1;
auto blitProperties = BlitProperties::constructPropertiesForReadWriteBuffer(BlitterConstants::BlitDirection::HostPtrToBuffer,
csr, builtinOpParams);
auto blitProperties = BlitProperties::constructProperties(BlitterConstants::BlitDirection::HostPtrToBuffer,
csr, builtinOpParams);
EXPECT_EQ(svmData->cpuAllocation, blitProperties.srcAllocation);
EXPECT_EQ(svmData->gpuAllocation, blitProperties.dstAllocation);
}
@@ -909,8 +909,8 @@ HWTEST_F(BcsTests, givenNonZeroCopySvmAllocationWhenConstructingBlitPropertiesFo
builtinOpParams.dstPtr = reinterpret_cast<void *>(svmData->cpuAllocation->getGpuAddress());
builtinOpParams.size.x = 1;
auto blitProperties = BlitProperties::constructPropertiesForReadWriteBuffer(BlitterConstants::BlitDirection::BufferToHostPtr,
csr, builtinOpParams);
auto blitProperties = BlitProperties::constructProperties(BlitterConstants::BlitDirection::BufferToHostPtr,
csr, builtinOpParams);
EXPECT_EQ(svmData->cpuAllocation, blitProperties.dstAllocation);
EXPECT_EQ(svmData->gpuAllocation, blitProperties.srcAllocation);
}

85
unit_tests/mem_obj/buffer_tests.cpp
View File

@@ -784,64 +784,98 @@ HWTEST_TEMPLATED_F(BcsBufferTests, givenBufferWithInitializationDataAndBcsCsrWhe
EXPECT_EQ(1u, bcsCsr->blitBufferCalled);
}
HWTEST_TEMPLATED_F(BcsBufferTests, givenBcsSupportedWhenEnqueueReadWriteBufferIsCalledThenUseBcsCsr) {
HWTEST_TEMPLATED_F(BcsBufferTests, givenBcsSupportedWhenEnqueueBufferOperationIsCalledThenUseBcsCsr) {
DebugManager.flags.EnableBlitterOperationsForReadWriteBuffers.set(0);
auto bcsCsr = static_cast<UltCommandStreamReceiver<FamilyType> *>(commandQueue->getBcsCommandStreamReceiver());
auto bufferForBlt = clUniquePtr(Buffer::create(bcsMockContext.get(), CL_MEM_READ_WRITE, 1, nullptr, retVal));
bufferForBlt->forceDisallowCPUCopy = true;
auto bufferForBlt0 = clUniquePtr(Buffer::create(bcsMockContext.get(), CL_MEM_READ_WRITE, 1, nullptr, retVal));
auto bufferForBlt1 = clUniquePtr(Buffer::create(bcsMockContext.get(), CL_MEM_READ_WRITE, 1, nullptr, retVal));
bufferForBlt0->forceDisallowCPUCopy = true;
bufferForBlt1->forceDisallowCPUCopy = true;
auto *hwInfo = device->getExecutionEnvironment()->getMutableHardwareInfo();
DebugManager.flags.EnableBlitterOperationsForReadWriteBuffers.set(0);
hwInfo->capabilityTable.blitterOperationsSupported = false;
commandQueue->enqueueWriteBuffer(bufferForBlt.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueReadBuffer(bufferForBlt.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueWriteBuffer(bufferForBlt0.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueReadBuffer(bufferForBlt0.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueCopyBuffer(bufferForBlt0.get(), bufferForBlt1.get(), 0, 1, 1, 0, nullptr, nullptr);
DebugManager.flags.EnableBlitterOperationsForReadWriteBuffers.set(1);
hwInfo->capabilityTable.blitterOperationsSupported = false;
commandQueue->enqueueWriteBuffer(bufferForBlt.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueReadBuffer(bufferForBlt.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueWriteBuffer(bufferForBlt0.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueReadBuffer(bufferForBlt0.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueCopyBuffer(bufferForBlt0.get(), bufferForBlt1.get(), 0, 1, 1, 0, nullptr, nullptr);
DebugManager.flags.EnableBlitterOperationsForReadWriteBuffers.set(0);
hwInfo->capabilityTable.blitterOperationsSupported = true;
commandQueue->enqueueWriteBuffer(bufferForBlt.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueReadBuffer(bufferForBlt.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueWriteBuffer(bufferForBlt0.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueReadBuffer(bufferForBlt0.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueCopyBuffer(bufferForBlt0.get(), bufferForBlt1.get(), 0, 1, 1, 0, nullptr, nullptr);
DebugManager.flags.EnableBlitterOperationsForReadWriteBuffers.set(-1);
hwInfo->capabilityTable.blitterOperationsSupported = true;
commandQueue->enqueueWriteBuffer(bufferForBlt.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueReadBuffer(bufferForBlt.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueWriteBuffer(bufferForBlt0.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueReadBuffer(bufferForBlt0.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueCopyBuffer(bufferForBlt0.get(), bufferForBlt1.get(), 0, 1, 1, 0, nullptr, nullptr);
EXPECT_EQ(2u, bcsCsr->blitBufferCalled);
EXPECT_EQ(3u, bcsCsr->blitBufferCalled);
DebugManager.flags.EnableBlitterOperationsForReadWriteBuffers.set(1);
hwInfo->capabilityTable.blitterOperationsSupported = true;
commandQueue->enqueueWriteBuffer(bufferForBlt.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
EXPECT_EQ(3u, bcsCsr->blitBufferCalled);
commandQueue->enqueueReadBuffer(bufferForBlt.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueWriteBuffer(bufferForBlt0.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
EXPECT_EQ(4u, bcsCsr->blitBufferCalled);
commandQueue->enqueueReadBuffer(bufferForBlt0.get(), CL_TRUE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
EXPECT_EQ(5u, bcsCsr->blitBufferCalled);
commandQueue->enqueueCopyBuffer(bufferForBlt0.get(), bufferForBlt1.get(), 0, 1, 1, 0, nullptr, nullptr);
EXPECT_EQ(6u, bcsCsr->blitBufferCalled);
}
HWTEST_TEMPLATED_F(BcsBufferTests, givenBcsSupportedWhenQueueIsBlockedThenDispatchBlitWhenUnblocked) {
auto bcsCsr = static_cast<UltCommandStreamReceiver<FamilyType> *>(commandQueue->getBcsCommandStreamReceiver());
auto bufferForBlt = clUniquePtr(Buffer::create(bcsMockContext.get(), CL_MEM_READ_WRITE, 1, nullptr, retVal));
bufferForBlt->forceDisallowCPUCopy = true;
auto bufferForBlt0 = clUniquePtr(Buffer::create(bcsMockContext.get(), CL_MEM_READ_WRITE, 1, nullptr, retVal));
auto bufferForBlt1 = clUniquePtr(Buffer::create(bcsMockContext.get(), CL_MEM_READ_WRITE, 1, nullptr, retVal));
bufferForBlt0->forceDisallowCPUCopy = true;
bufferForBlt1->forceDisallowCPUCopy = true;
UserEvent userEvent(bcsMockContext.get());
cl_event waitlist = &userEvent;
commandQueue->enqueueWriteBuffer(bufferForBlt.get(), CL_FALSE, 0, 1, &hostPtr, nullptr, 1, &waitlist, nullptr);
commandQueue->enqueueReadBuffer(bufferForBlt.get(), CL_FALSE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueWriteBuffer(bufferForBlt0.get(), CL_FALSE, 0, 1, &hostPtr, nullptr, 1, &waitlist, nullptr);
commandQueue->enqueueReadBuffer(bufferForBlt1.get(), CL_FALSE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueCopyBuffer(bufferForBlt0.get(), bufferForBlt1.get(), 0, 1, 1, 0, nullptr, nullptr);
EXPECT_EQ(0u, bcsCsr->blitBufferCalled);
userEvent.setStatus(CL_COMPLETE);
EXPECT_EQ(2u, bcsCsr->blitBufferCalled);
commandQueue->enqueueWriteBuffer(bufferForBlt.get(), CL_FALSE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
EXPECT_EQ(3u, bcsCsr->blitBufferCalled);
commandQueue->enqueueReadBuffer(bufferForBlt.get(), CL_FALSE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
commandQueue->enqueueWriteBuffer(bufferForBlt0.get(), CL_FALSE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
EXPECT_EQ(4u, bcsCsr->blitBufferCalled);
commandQueue->enqueueReadBuffer(bufferForBlt0.get(), CL_FALSE, 0, 1, &hostPtr, nullptr, 0, nullptr, nullptr);
EXPECT_EQ(5u, bcsCsr->blitBufferCalled);
commandQueue->enqueueCopyBuffer(bufferForBlt0.get(), bufferForBlt1.get(), 0, 1, 1, 0, nullptr, nullptr);
}
HWTEST_TEMPLATED_F(BcsBufferTests, givenBuffersWhenCopyBufferCalledThenUseBcs) {
using XY_COPY_BLT = typename FamilyType::XY_COPY_BLT;
auto cmdQ = clUniquePtr(new MockCommandQueueHw<FamilyType>(bcsMockContext.get(), device.get(), nullptr));
auto bufferForBlt0 = clUniquePtr(Buffer::create(bcsMockContext.get(), CL_MEM_READ_WRITE, 1, nullptr, retVal));
auto bufferForBlt1 = clUniquePtr(Buffer::create(bcsMockContext.get(), CL_MEM_READ_WRITE, 1, nullptr, retVal));
bufferForBlt0->forceDisallowCPUCopy = true;
bufferForBlt1->forceDisallowCPUCopy = true;
cmdQ->enqueueCopyBuffer(bufferForBlt0.get(), bufferForBlt1.get(), 0, 0, 1, 0, nullptr, nullptr);
HardwareParse hwParser;
hwParser.parseCommands<FamilyType>(commandQueue->getBcsCommandStreamReceiver()->getCS(0));
auto commandItor = find<XY_COPY_BLT *>(hwParser.cmdList.begin(), hwParser.cmdList.end());
EXPECT_NE(hwParser.cmdList.end(), commandItor);
auto copyBltCmd = genCmdCast<XY_COPY_BLT *>(*commandItor);
EXPECT_EQ(bufferForBlt0->getGraphicsAllocation()->getGpuAddress(), copyBltCmd->getSourceBaseAddress());
EXPECT_EQ(bufferForBlt1->getGraphicsAllocation()->getGpuAddress(), copyBltCmd->getDestinationBaseAddress());
}
HWTEST_TEMPLATED_F(BcsBufferTests, givenBlockedBlitEnqueueWhenUnblockingThenMakeResidentAllTimestampPackets) {
@@ -1110,7 +1144,7 @@ HWTEST_TEMPLATED_F(BcsBufferTests, givenPipeControlRequestWhenDispatchingBlocked
cmdQ->isQueueBlocked();
}
HWTEST_TEMPLATED_F(BcsBufferTests, givenReadOrWriteBufferOperationWithoutKernelWhenEstimatingCommandsSizeThenReturnCorrectValue) {
HWTEST_TEMPLATED_F(BcsBufferTests, givenBufferOperationWithoutKernelWhenEstimatingCommandsSizeThenReturnCorrectValue) {
auto cmdQ = clUniquePtr(new MockCommandQueueHw<FamilyType>(bcsMockContext.get(), device.get(), nullptr));
CsrDependencies csrDependencies;
MultiDispatchInfo multiDispatchInfo;
@@ -1119,10 +1153,13 @@ HWTEST_TEMPLATED_F(BcsBufferTests, givenReadOrWriteBufferOperationWithoutKernelW
true, *cmdQ, multiDispatchInfo);
auto writeBufferCmdsSize = EnqueueOperation<FamilyType>::getTotalSizeRequiredCS(CL_COMMAND_WRITE_BUFFER, csrDependencies, false, false,
true, *cmdQ, multiDispatchInfo);
auto copyBufferCmdsSize = EnqueueOperation<FamilyType>::getTotalSizeRequiredCS(CL_COMMAND_COPY_BUFFER, csrDependencies, false, false,
true, *cmdQ, multiDispatchInfo);
auto expectedSize = TimestampPacketHelper::getRequiredCmdStreamSizeForNodeDependency<FamilyType>();
EXPECT_EQ(expectedSize, readBufferCmdsSize);
EXPECT_EQ(expectedSize, writeBufferCmdsSize);
EXPECT_EQ(expectedSize, copyBufferCmdsSize);
}
HWTEST_TEMPLATED_F(BcsBufferTests, givenOutputTimestampPacketWhenBlitCalledThenProgramMiFlushDwWithDataWrite) {