AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

SWDEV-305527 - Changes to handle memset blit kernel that takes width, height and depth. This also fixes SWDEV-317261.

Browse Source 
Change-Id: Ic85f63a95d9d8f48884fc8c7fd95cbb496dfbbca


[ROCm/clr commit: 7fb80a027a]
This commit is contained in:
kjayapra-amd
2021-10-19 13:00:51 -04:00
committed by Karthik Jayaprakash
parent 8fb4e3478a
commit 31c0525344
11 changed files with 224 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/clr/rocclr/device/rocm/rocvirtual.cpp
+12 -30
View File
@@ -2211,8 +2211,9 @@ void VirtualGPU::submitUnmapMemory(amd::UnmapMemoryCommand& cmd) {
}
bool VirtualGPU::fillMemory(cl_command_type type, amd::Memory* amdMemory, const void* pattern,
size_t patternSize, const amd::Coord3D& origin,
const amd::Coord3D& size, bool forceBlit) {
size_t patternSize, const amd::Coord3D& surface,
const amd::Coord3D& origin, const amd::Coord3D& size,
bool forceBlit) {
// Make sure VirtualGPU has an exclusive access to the resources
amd::ScopedLock lock(execution());
@@ -2238,8 +2239,9 @@ bool VirtualGPU::fillMemory(cl_command_type type, amd::Memory* amdMemory, const
switch (type) {
case CL_COMMAND_SVM_MEMFILL:
case CL_COMMAND_FILL_BUFFER: {
amd::Coord3D realOrigin(origin[0]);
amd::Coord3D realSize(size[0]);
amd::Coord3D realSurf(surface[0], surface[1], surface[2]);
amd::Coord3D realOrigin(origin[0], origin[1], origin[2]);
amd::Coord3D realSize(size[0], size[1], size[2]);
// Reprogram fill parameters if it's an IMAGE1D_BUFFER object
if (imageBuffer) {
size_t elemSize = amdMemory->asImage()->getImageFormat().getElementSize();
@@ -2250,7 +2252,8 @@ bool VirtualGPU::fillMemory(cl_command_type type, amd::Memory* amdMemory, const
pattern = fillValue;
patternSize = elemSize;
}
result = blitMgr().fillBuffer(*memory, pattern, patternSize, realOrigin, realSize, entire, forceBlit);
result = blitMgr().fillBuffer(*memory, pattern, patternSize, realSurf, realOrigin,
realSize, entire, forceBlit);
break;
}
case CL_COMMAND_FILL_IMAGE: {
@@ -2275,30 +2278,9 @@ void VirtualGPU::submitFillMemory(amd::FillMemoryCommand& cmd) {
amd::ScopedLock lock(execution());
profilingBegin(cmd);
if (cmd.type() == CL_COMMAND_FILL_IMAGE) {
if (!fillMemory(cmd.type(), &cmd.memory(), cmd.pattern(), cmd.patternSize(),
cmd.origin(), cmd.size())) {
cmd.setStatus(CL_INVALID_OPERATION);
}
} else {
size_t width = cmd.size().c[0];
size_t height = cmd.size().c[1];
size_t depth = cmd.size().c[2];
size_t pitch = cmd.surface().c[0];
amd::Coord3D origin = cmd.origin();
amd::Coord3D region{cmd.surface().c[1], cmd.surface().c[2], depth};
amd::BufferRect rect;
rect.create(static_cast<size_t*>(origin), static_cast<size_t*>(region),
pitch, 0);
for (size_t slice = 0; slice < depth; slice++) {
for (size_t row = 0; row < height; row++) {
const size_t rowOffset = rect.offset(0, row, slice);
if (!fillMemory(cmd.type(), &cmd.memory(), cmd.pattern(), cmd.patternSize(),
amd::Coord3D{rowOffset, 0, 0}, amd::Coord3D{width, 1, 1})) {
cmd.setStatus(CL_INVALID_OPERATION);
}
}
}
if (!fillMemory(cmd.type(), &cmd.memory(), cmd.pattern(), cmd.patternSize(),
cmd.surface(), cmd.origin(), cmd.size())) {
cmd.setStatus(CL_INVALID_OPERATION);
}
profilingEnd(cmd);
}
@@ -2636,7 +2618,7 @@ bool VirtualGPU::createVirtualQueue(uint deviceQueueSize)
amd::Coord3D origin(0, 0, 0);
amd::Coord3D region(virtualQueue_->getSize(), 0, 0);
if (!dev().xferMgr().fillBuffer(*vqMem, &pattern, sizeof(pattern), origin, region)) {
if (!dev().xferMgr().fillBuffer(*vqMem, &pattern, sizeof(pattern), region, origin, region)) {
return false;
}