intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-22 10:17:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

OCL: optimize creating printf buffer

Browse Source 
Dont create printf buffer when kernel doesnt require it

Related-To: HSD-18023825570
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
This commit is contained in:
Mateusz Jablonski
2022-09-29 16:36:48 +02:00
committed by Compute-Runtime-Automation
parent 48824acab2
commit 23eff82d0a
11 changed files with 66 additions and 89 deletions
Download Patch File Download Diff File Expand all files Collapse all files

42
opencl/source/program/printf_handler.cpp
View File

@@ -7,6 +7,7 @@
#include "printf_handler.h"
#include "shared/source/device/device.h"
#include "shared/source/helpers/aligned_memory.h"
#include "shared/source/helpers/blit_commands_helper.h"
#include "shared/source/helpers/hw_helper.h"
@@ -15,15 +16,13 @@
#include "shared/source/os_interface/hw_info_config.h"
#include "shared/source/program/print_formatter.h"
#include "opencl/source/cl_device/cl_device.h"
#include "opencl/source/context/context.h"
#include "opencl/source/helpers/dispatch_info.h"
#include "opencl/source/kernel/kernel.h"
#include "opencl/source/mem_obj/buffer.h"
namespace NEO {
PrintfHandler::PrintfHandler(ClDevice &deviceArg) : device(deviceArg) {
PrintfHandler::PrintfHandler(Device &deviceArg) : device(deviceArg) {
printfSurfaceInitialDataSizePtr = std::make_unique<uint32_t>();
*printfSurfaceInitialDataSizePtr = sizeof(uint32_t);
}
@@ -32,21 +31,15 @@ PrintfHandler::~PrintfHandler() {
device.getMemoryManager()->freeGraphicsMemory(printfSurface);
}
PrintfHandler *PrintfHandler::create(const MultiDispatchInfo &multiDispatchInfo, ClDevice &device) {
PrintfHandler *PrintfHandler::create(const MultiDispatchInfo &multiDispatchInfo, Device &device) {
if (multiDispatchInfo.usesStatelessPrintfSurface()) {
return new PrintfHandler(device);
}
auto mainKernel = multiDispatchInfo.peekMainKernel();
if (mainKernel != nullptr) {
if (mainKernel->getImplicitArgs()) {
return new PrintfHandler(device);
}
}
return nullptr;
}
void PrintfHandler::prepareDispatch(const MultiDispatchInfo &multiDispatchInfo) {
auto printfSurfaceSize = device.getSharedDeviceInfo().printfBufferSize;
auto printfSurfaceSize = device.getDeviceInfo().printfBufferSize;
if (printfSurfaceSize == 0) {
return;
}
@@ -58,22 +51,19 @@ void PrintfHandler::prepareDispatch(const MultiDispatchInfo &multiDispatchInfo)
const auto &hwInfoConfig = *HwInfoConfig::get(hwInfo.platform.eProductFamily);
MemoryTransferHelper::transferMemoryToAllocation(hwInfoConfig.isBlitCopyRequiredForLocalMemory(hwInfo, *printfSurface),
device.getDevice(), printfSurface, 0, printfSurfaceInitialDataSizePtr.get(),
device, printfSurface, 0, printfSurfaceInitialDataSizePtr.get(),
sizeof(*printfSurfaceInitialDataSizePtr.get()));
if (kernel->getKernelInfo().kernelDescriptor.kernelAttributes.flags.usesPrintf) {
const auto &printfSurfaceArg = kernel->getKernelInfo().kernelDescriptor.payloadMappings.implicitArgs.printfSurfaceAddress;
auto printfPatchAddress = ptrOffset(reinterpret_cast<uintptr_t *>(kernel->getCrossThreadData()), printfSurfaceArg.stateless);
patchWithRequiredSize(printfPatchAddress, printfSurfaceArg.pointerSize, (uintptr_t)printfSurface->getGpuAddressToPatch());
if (isValidOffset(printfSurfaceArg.bindful)) {
auto surfaceState = ptrOffset(reinterpret_cast<uintptr_t *>(kernel->getSurfaceStateHeap()), printfSurfaceArg.bindful);
void *addressToPatch = printfSurface->getUnderlyingBuffer();
size_t sizeToPatch = printfSurface->getUnderlyingBufferSize();
Buffer::setSurfaceState(&device.getDevice(), surfaceState, false, false, sizeToPatch, addressToPatch, 0, printfSurface, 0, 0,
kernel->getKernelInfo().kernelDescriptor.kernelAttributes.flags.useGlobalAtomics,
kernel->areMultipleSubDevicesInContext());
}
const auto &printfSurfaceArg = kernel->getKernelInfo().kernelDescriptor.payloadMappings.implicitArgs.printfSurfaceAddress;
auto printfPatchAddress = ptrOffset(reinterpret_cast<uintptr_t *>(kernel->getCrossThreadData()), printfSurfaceArg.stateless);
patchWithRequiredSize(printfPatchAddress, printfSurfaceArg.pointerSize, (uintptr_t)printfSurface->getGpuAddressToPatch());
if (isValidOffset(printfSurfaceArg.bindful)) {
auto surfaceState = ptrOffset(reinterpret_cast<uintptr_t *>(kernel->getSurfaceStateHeap()), printfSurfaceArg.bindful);
void *addressToPatch = printfSurface->getUnderlyingBuffer();
size_t sizeToPatch = printfSurface->getUnderlyingBufferSize();
Buffer::setSurfaceState(&device, surfaceState, false, false, sizeToPatch, addressToPatch, 0, printfSurface, 0, 0,
kernel->getKernelInfo().kernelDescriptor.kernelAttributes.flags.useGlobalAtomics,
kernel->areMultipleSubDevicesInContext());
}
auto pImplicitArgs = kernel->getImplicitArgs();
if (pImplicitArgs) {
@@ -107,7 +97,7 @@ bool PrintfHandler::printEnqueueOutput() {
printfSurface->getGpuAddress(),
0, 0, 0, Vec3<size_t>(printfOutputSize, 0, 0), 0, 0, 0, 0));
const auto newTaskCount = bcsEngine.commandStreamReceiver->flushBcsTask(blitPropertiesContainer, true, false, device.getDevice());
const auto newTaskCount = bcsEngine.commandStreamReceiver->flushBcsTask(blitPropertiesContainer, true, false, device);
if (!newTaskCount) {
return false;
}