P4 to Git Change 1102328 by gandryey@gera-dev-w7 on 2014/12/03 18:56:06
EPR #410736 - [CQE OCL][ISV][QR][G] FFMPEG app generating corrupted video output; Faulty CL:1101352
- Add detection for AHP allocation.
FFmpeg uses AHP allocations with CL_MAP_READ flag, but actually performs CPU write into the buffer. With indirect map runtime executes useless transfer on map and doesn't write updated memory on unmap, because a wrong flag sent by the app.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpumemory.cpp#113 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpumemory.hpp#44 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#341 edit
... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/perf/TestList.cpp#40 edit
[ROCm/clr commit: f9f5df731e]
This commit is contained in:
@@ -923,7 +923,7 @@ Memory::allocMapTarget(
|
||||
|
||||
// If host memory exists, use it
|
||||
if ((owner()->getHostMem() != NULL) &&
|
||||
(isCacheable() || !isHostMemDirectAccess() || !(mapFlags & CL_MAP_READ))) {
|
||||
isDirectMap(mapFlags)) {
|
||||
mapAddress = reinterpret_cast<address>(owner()->getHostMem());
|
||||
}
|
||||
// If resource is a persistent allocation, we can use it directly
|
||||
@@ -1249,7 +1249,7 @@ Image::allocMapTarget(
|
||||
|
||||
// If host memory exists, use it
|
||||
if ((owner()->getHostMem() != NULL) &&
|
||||
(isCacheable() || !isHostMemDirectAccess() || !(mapFlags & CL_MAP_READ))) {
|
||||
isDirectMap(mapFlags)) {
|
||||
useRemoteResource = false;
|
||||
mapAddress = reinterpret_cast<address>(owner()->getHostMem());
|
||||
amd::Image* amdImage = owner()->asImage();
|
||||
|
||||
@@ -200,6 +200,15 @@ public:
|
||||
//! Returns the interop resource for this memory object
|
||||
const Memory* parent() const { return parent_; }
|
||||
|
||||
//! Returns TRUE if direct map is acceaptable
|
||||
//! The method detects forced USWC memory on APU and
|
||||
//! will cause a switch to indirect map for MAP_READ operations
|
||||
bool isDirectMap(uint mapFlags)
|
||||
{
|
||||
return (isCacheable() || (owner()->getMemFlags() & CL_MEM_ALLOC_HOST_PTR) ||
|
||||
!isHostMemDirectAccess() || !(mapFlags & CL_MAP_READ));
|
||||
}
|
||||
|
||||
protected:
|
||||
//! Decrement map count
|
||||
void decIndMapCount();
|
||||
|
||||
@@ -1091,9 +1091,7 @@ VirtualGPU::submitMapMemory(amd::MapMemoryCommand& vcmd)
|
||||
|
||||
// If we have host memory, use it
|
||||
if ((memory->owner()->getHostMem() != NULL) &&
|
||||
(memory->isCacheable() ||
|
||||
!memory->isHostMemDirectAccess() ||
|
||||
!(vcmd.mapFlags() & CL_MAP_READ))) {
|
||||
memory->isDirectMap(vcmd.mapFlags())) {
|
||||
if (!memory->isHostMemDirectAccess()) {
|
||||
// Make sure GPU finished operation before
|
||||
// synchronization with the backing store
|
||||
@@ -1179,9 +1177,7 @@ VirtualGPU::submitUnmapMemory(amd::UnmapMemoryCommand& vcmd)
|
||||
|
||||
// We used host memory
|
||||
if ((owner->getHostMem() != NULL) &&
|
||||
(memory->isCacheable() ||
|
||||
!memory->isHostMemDirectAccess() ||
|
||||
!memory->isUnmapRead())) {
|
||||
memory->isDirectMap(memory->isUnmapRead() ? CL_MAP_READ : 0)) {
|
||||
if (memory->isUnmapWrite()) {
|
||||
// Target is the backing store, so sync
|
||||
owner->signalWrite(NULL);
|
||||
|
||||
Reference in New Issue
Block a user