intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-20 08:53:55 +08:00
Code Issues Packages Projects Releases Wiki Activity

Enqueue write buffer instead of memcpy for renderCompressed allocation

Browse Source 
Change-Id: I5505add129e08537fc0009b12228a446aa732645
This commit is contained in:
Dunajski, Bartosz
2018-07-24 09:06:33 +02:00
committed by sys_ocldev
parent c61364bd3f
commit 579367fe4a
6 changed files with 175 additions and 92 deletions
Download Patch File Download Diff File Expand all files Collapse all files

141
runtime/mem_obj/buffer.cpp
View File

@@ -103,6 +103,10 @@ Buffer *Buffer::create(Context *context,
checkMemory(flags, size, hostPtr, errcodeRet, zeroCopy, allocateMemory, copyMemoryFromHostPtr, allocationType, memoryManager);
if (errcodeRet != CL_SUCCESS) {
return nullptr;
}
if (hostPtr && context->isProvidingPerformanceHints()) {
if (zeroCopy) {
context->providePerformanceHint(CL_CONTEXT_DIAGNOSTICS_LEVEL_GOOD_INTEL, CL_BUFFER_MEETS_ALIGNMENT_RESTRICTIONS, hostPtr, size);
@@ -117,75 +121,82 @@ Buffer *Buffer::create(Context *context,
copyMemoryFromHostPtr = false;
allocateMemory = false;
}
if (errcodeRet == CL_SUCCESS) {
while (true) {
if (flags & CL_MEM_USE_HOST_PTR) {
memory = context->getSVMAllocsManager()->getSVMAlloc(hostPtr);
if (memory) {
allocationType = GraphicsAllocation::AllocationType::BUFFER;
zeroCopy = true;
isHostPtrSVM = true;
copyMemoryFromHostPtr = false;
allocateMemory = false;
}
}
if (!memory) {
memory = memoryManager->allocateGraphicsMemoryInPreferredPool(zeroCopy, allocateMemory, true, false, hostPtr, static_cast<size_t>(size), allocationType);
}
if (allocateMemory) {
if (memory) {
memoryManager->addAllocationToHostPtrManager(memory);
}
if (context->isProvidingPerformanceHints()) {
context->providePerformanceHint(CL_CONTEXT_DIAGNOSTICS_LEVEL_GOOD_INTEL, CL_BUFFER_NEEDS_ALLOCATE_MEMORY);
}
} else {
if (!memory && Buffer::isReadOnlyMemoryPermittedByFlags(flags)) {
zeroCopy = false;
copyMemoryFromHostPtr = true;
allocateMemory = true;
memory = memoryManager->allocateGraphicsMemoryInPreferredPool(zeroCopy, allocateMemory, true, false, nullptr, static_cast<size_t>(size), allocationType);
}
}
if (!memory) {
errcodeRet = CL_OUT_OF_HOST_MEMORY;
break;
}
memory->setAllocationType(allocationType);
memory->setMemObjectsAllocationWithWritableFlags(!(flags & (CL_MEM_READ_ONLY | CL_MEM_HOST_READ_ONLY | CL_MEM_HOST_NO_ACCESS)));
DBG_LOG(LogMemoryObject, __FUNCTION__, "hostPtr:", hostPtr, "size:", size, "memoryStorage:", memory->getUnderlyingBuffer(), "GPU address:", std::hex, memory->getGpuAddress());
if (copyMemoryFromHostPtr) {
memcpy_s(memory->getUnderlyingBuffer(), size, hostPtr, size);
}
pBuffer = createBufferHw(context,
flags,
size,
memory->getUnderlyingBuffer(),
const_cast<void *>(hostPtr),
memory,
zeroCopy,
isHostPtrSVM,
false);
if (!pBuffer && allocateMemory) {
memoryManager->removeAllocationFromHostPtrManager(memory);
memoryManager->freeGraphicsMemory(memory);
memory = nullptr;
}
if (pBuffer) {
pBuffer->setHostPtrMinSize(size);
}
break;
if (flags & CL_MEM_USE_HOST_PTR) {
memory = context->getSVMAllocsManager()->getSVMAlloc(hostPtr);
if (memory) {
allocationType = GraphicsAllocation::AllocationType::BUFFER;
zeroCopy = true;
isHostPtrSVM = true;
copyMemoryFromHostPtr = false;
allocateMemory = false;
}
}
if (!memory) {
memory = memoryManager->allocateGraphicsMemoryInPreferredPool(zeroCopy, allocateMemory, true, false, hostPtr, static_cast<size_t>(size), allocationType);
}
if (allocateMemory) {
if (memory) {
memoryManager->addAllocationToHostPtrManager(memory);
}
if (context->isProvidingPerformanceHints()) {
context->providePerformanceHint(CL_CONTEXT_DIAGNOSTICS_LEVEL_GOOD_INTEL, CL_BUFFER_NEEDS_ALLOCATE_MEMORY);
}
} else {
if (!memory && Buffer::isReadOnlyMemoryPermittedByFlags(flags)) {
zeroCopy = false;
copyMemoryFromHostPtr = true;
allocateMemory = true;
memory = memoryManager->allocateGraphicsMemoryInPreferredPool(zeroCopy, allocateMemory, true, false, nullptr, static_cast<size_t>(size), allocationType);
}
}
if (!memory) {
errcodeRet = CL_OUT_OF_HOST_MEMORY;
return nullptr;
}
memory->setAllocationType(allocationType);
memory->setMemObjectsAllocationWithWritableFlags(!(flags & (CL_MEM_READ_ONLY | CL_MEM_HOST_READ_ONLY | CL_MEM_HOST_NO_ACCESS)));
DBG_LOG(LogMemoryObject, __FUNCTION__, "hostPtr:", hostPtr, "size:", size, "memoryStorage:", memory->getUnderlyingBuffer(), "GPU address:", std::hex, memory->getGpuAddress());
pBuffer = createBufferHw(context,
flags,
size,
memory->getUnderlyingBuffer(),
const_cast<void *>(hostPtr),
memory,
zeroCopy,
isHostPtrSVM,
false);
if (!pBuffer) {
errcodeRet = CL_OUT_OF_HOST_MEMORY;
memoryManager->removeAllocationFromHostPtrManager(memory);
memoryManager->freeGraphicsMemory(memory);
return nullptr;
}
pBuffer->setHostPtrMinSize(size);
if (copyMemoryFromHostPtr) {
if (memory->gmm && memory->gmm->isRenderCompressed) {
auto cmdQ = context->getSpecialQueue();
if (CL_SUCCESS != cmdQ->enqueueWriteBuffer(pBuffer, CL_TRUE, 0, size, hostPtr, 0, nullptr, nullptr)) {
errcodeRet = CL_OUT_OF_RESOURCES;
}
} else {
memcpy_s(memory->getUnderlyingBuffer(), size, hostPtr, size);
}
}
if (errcodeRet != CL_SUCCESS) {
pBuffer->release();
return nullptr;
}
return pBuffer;
}