AI/rocm-systems
2
0
Fork 0
程式碼 問題 合併請求 Actions 套件 專案 版本發布 Wiki 動態

Revert "SWDEV-301667 - Disable HostBlit copy for HIP"

瀏覽代碼 
This reverts commit 96e25898c8.

Reason for revert: SWDEV-455075, SWDEV-461507 - This change forces to
use ROCr's copy path. Reintroducing hostBlit copy path for
host-to-host copies.


Change-Id: Ic3c45b49e481c9dcdaa7611f61071778790b7e6c


[ROCm/clr commit: a9f05e22db]
此提交包含在：
Sourabh Betigeri
2024-05-16 19:10:25 -04:00
提交者 Maneesh Gupta
父節點 43aeeaf836
當前提交 da8bb9780e
共有 3 個檔案被更改，包括 19 行新增 和 29 行删除
下載 Patch 檔 下載差異檔 展開所有檔案 折疊所有檔案
projects/clr/rocclr/device/blit.hpp
-1
查看文件
@@ -53,7 +53,6 @@ class BlitManager : public amd::HeapObject {
uint disableFillImage_ : 1;
uint disableCopyBufferToImageOpt_ : 1;
uint disableHwlCopyBuffer_ : 1;
uint disableHostCopyBuffer_ : 1;
};
uint32_t value_;
Setup() : value_(0) {}
projects/clr/rocclr/device/rocm/rocblit.cpp
+19 -26
查看文件
@@ -72,8 +72,7 @@ bool DmaBlitManager::readBuffer(device::Memory& srcMemory, void* dstHost,
// Use host copy if memory has direct access
if (setup_.disableReadBuffer_ ||
(srcMemory.isHostMemDirectAccess() && !srcMemory.isCpuUncached() &&
!setup_.disableHostCopyBuffer_)) {
(srcMemory.isHostMemDirectAccess() && !srcMemory.isCpuUncached())) {
// Stall GPU before CPU access
gpu().Barriers().WaitCurrent();
return HostBlitManager::readBuffer(srcMemory, dstHost, origin, size, entire, copyMetadata);
@@ -166,12 +165,10 @@ bool DmaBlitManager::readBufferRect(device::Memory& srcMemory, void* dstHost,
// Use host copy if memory has direct access
if (setup_.disableReadBufferRect_ ||
(srcMemory.isHostMemDirectAccess() && !srcMemory.isCpuUncached() &&
!setup_.disableHostCopyBuffer_)) {
(srcMemory.isHostMemDirectAccess() && !srcMemory.isCpuUncached())) {
// Stall GPU before CPU access
gpu().Barriers().WaitCurrent();
return HostBlitManager::readBufferRect(srcMemory, dstHost, bufRect, hostRect, size,
entire, copyMetadata);
return HostBlitManager::readBufferRect(srcMemory, dstHost, bufRect, hostRect, size, entire, copyMetadata);
} else {
Memory& xferBuf = dev().xferRead().acquire();
address staging = xferBuf.getDeviceMemory();
@@ -239,8 +236,8 @@ bool DmaBlitManager::writeBuffer(const void* srcHost, device::Memory& dstMemory,
const amd::Coord3D& origin, const amd::Coord3D& size,
bool entire, amd::CopyMetadata copyMetadata) const {
// Use host copy if memory has direct access
if ((setup_.disableWriteBuffer_ || dstMemory.isHostMemDirectAccess() ||
gpuMem(dstMemory).IsPersistentDirectMap()) && !setup_.disableHostCopyBuffer_) {
if (setup_.disableWriteBuffer_ || dstMemory.isHostMemDirectAccess() ||
gpuMem(dstMemory).IsPersistentDirectMap()) {
// Stall GPU before CPU access
gpu().releaseGpuMemoryFence();
return HostBlitManager::writeBuffer(srcHost, dstMemory, origin, size, entire, copyMetadata);
@@ -336,8 +333,8 @@ bool DmaBlitManager::writeBufferRect(const void* srcHost, device::Memory& dstMem
gpu().releaseGpuMemoryFence();
// Use host copy if memory has direct access
if ((setup_.disableWriteBufferRect_ || dstMemory.isHostMemDirectAccess() ||
gpuMem(dstMemory).IsPersistentDirectMap()) && !setup_.disableHostCopyBuffer_) {
if (setup_.disableWriteBufferRect_ || dstMemory.isHostMemDirectAccess() ||
gpuMem(dstMemory).IsPersistentDirectMap()) {
return HostBlitManager::writeBufferRect(srcHost, dstMemory, hostRect, bufRect, size, entire,
copyMetadata);
} else {
@@ -393,9 +390,8 @@ bool DmaBlitManager::copyBuffer(device::Memory& srcMemory, device::Memory& dstMe
const amd::Coord3D& size, bool entire,
amd::CopyMetadata copyMetadata) const {
if (setup_.disableCopyBuffer_ ||
(!setup_.disableHostCopyBuffer_ && srcMemory.isHostMemDirectAccess() &&
!srcMemory.isCpuUncached() && (dev().agent_profile() != HSA_PROFILE_FULL) &&
dstMemory.isHostMemDirectAccess())) {
(srcMemory.isHostMemDirectAccess() && !srcMemory.isCpuUncached() &&
(dev().agent_profile() != HSA_PROFILE_FULL) && dstMemory.isHostMemDirectAccess())) {
// Stall GPU before CPU access
gpu().releaseGpuMemoryFence();
return HostBlitManager::copyBuffer(srcMemory, dstMemory, srcOrigin, dstOrigin, size, false,
@@ -413,8 +409,8 @@ bool DmaBlitManager::copyBufferRect(device::Memory& srcMemory, device::Memory& d
const amd::Coord3D& size, bool entire,
amd::CopyMetadata copyMetadata) const {
if (setup_.disableCopyBufferRect_ ||
(!setup_.disableHostCopyBuffer_ && srcMemory.isHostMemDirectAccess() &&
!srcMemory.isCpuUncached() && dstMemory.isHostMemDirectAccess())) {
(srcMemory.isHostMemDirectAccess() && !srcMemory.isCpuUncached() &&
dstMemory.isHostMemDirectAccess())) {
// Stall GPU before CPU access
gpu().releaseGpuMemoryFence();
return HostBlitManager::copyBufferRect(srcMemory, dstMemory, srcRect, dstRect, size, entire,
@@ -1843,7 +1839,7 @@ bool KernelBlitManager::readBuffer(device::Memory& srcMemory, void* dstHost,
// Use host copy if memory has direct access
if (setup_.disableReadBuffer_ || (srcMemory.isHostMemDirectAccess() &&
!srcMemory.isCpuUncached() && !setup_.disableHostCopyBuffer_)) {
!srcMemory.isCpuUncached())) {
// Stall GPU before CPU access
gpu().releaseGpuMemoryFence();
result = HostBlitManager::readBuffer(srcMemory, dstHost, origin, size, entire, copyMetadata);
@@ -1895,8 +1891,7 @@ bool KernelBlitManager::readBufferRect(device::Memory& srcMemory, void* dstHost,
// Use host copy if memory has direct access
if (setup_.disableReadBufferRect_ ||
(srcMemory.isHostMemDirectAccess() && !srcMemory.isCpuUncached() &&
!setup_.disableHostCopyBuffer_)) {
(srcMemory.isHostMemDirectAccess() && !srcMemory.isCpuUncached())) {
// Stall GPU before CPU access
gpu().releaseGpuMemoryFence();
result = HostBlitManager::readBufferRect(srcMemory, dstHost, bufRect, hostRect, size, entire,
@@ -1946,8 +1941,8 @@ bool KernelBlitManager::writeBuffer(const void* srcHost, device::Memory& dstMemo
bool result = false;
// Use host copy if memory has direct access
if ((setup_.disableWriteBuffer_ || dstMemory.isHostMemDirectAccess() ||
gpuMem(dstMemory).IsPersistentDirectMap()) && !setup_.disableHostCopyBuffer_) {
if (setup_.disableWriteBuffer_ || dstMemory.isHostMemDirectAccess() ||
gpuMem(dstMemory).IsPersistentDirectMap()) {
// Stall GPU before CPU access
gpu().releaseGpuMemoryFence();
result = HostBlitManager::writeBuffer(srcHost, dstMemory, origin, size, entire, copyMetadata);
@@ -1998,8 +1993,8 @@ bool KernelBlitManager::writeBufferRect(const void* srcHost, device::Memory& dst
bool result = false;
// Use host copy if memory has direct access
if ((setup_.disableWriteBufferRect_ || dstMemory.isHostMemDirectAccess() ||
gpuMem(dstMemory).IsPersistentDirectMap()) && !setup_.disableHostCopyBuffer_) {
if (setup_.disableWriteBufferRect_ || dstMemory.isHostMemDirectAccess() ||
gpuMem(dstMemory).IsPersistentDirectMap()) {
// Stall GPU before CPU access
gpu().releaseGpuMemoryFence();
result = HostBlitManager::writeBufferRect(srcHost, dstMemory, hostRect, bufRect, size, entire,
@@ -2080,8 +2075,7 @@ bool KernelBlitManager::fillBuffer1D(device::Memory& memory, const void* pattern
bool result = false;
// Use host fill if memory has direct access
if (setup_.disableFillBuffer_ || (!forceBlit && memory.isHostMemDirectAccess() &&
!setup_.disableHostCopyBuffer_)) {
if (setup_.disableFillBuffer_ || (!forceBlit && memory.isHostMemDirectAccess())) {
// Stall GPU before CPU access
gpu().releaseGpuMemoryFence();
result = HostBlitManager::fillBuffer(memory, pattern, patternSize, size, origin, size, entire);
@@ -2161,8 +2155,7 @@ bool KernelBlitManager::fillBuffer2D(device::Memory& memory, const void* pattern
bool result = false;
// Use host fill if memory has direct access
if (setup_.disableFillBuffer_ || (!forceBlit && memory.isHostMemDirectAccess() &&
!setup_.disableHostCopyBuffer_)) {
if (setup_.disableFillBuffer_ || (!forceBlit && memory.isHostMemDirectAccess())) {
// Stall GPU before CPU access
gpu().releaseGpuMemoryFence();
result = HostBlitManager::fillBuffer(memory, pattern, patternSize, size, origin, size, entire);
projects/clr/rocclr/device/rocm/rocvirtual.cpp
-2
查看文件
@@ -1329,8 +1329,6 @@ bool VirtualGPU::create() {
}
device::BlitManager::Setup blitSetup;
// Disable HostBlit copy path for HIP
blitSetup.disableHostCopyBuffer_ = amd::IS_HIP;
blitMgr_ = new KernelBlitManager(*this, blitSetup);
if ((nullptr == blitMgr_) || !blitMgr_->create(roc_device_)) {
LogError("Could not create BlitManager!");