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

P4 to Git Change 1288063 by bsumner@bsumner-lin-opencl on 2016/07/06 17:07:15

Browse Source 
SWDEV-97663 - optimize cbrt for AMD devices

Affected files ...

... //depot/stg/opencl/drivers/opencl/library/ocml/src/cbrtF.cl#5 edit
... //depot/stg/opencl/drivers/opencl/library/ocml/src/rcbrtF.cl#4 edit
This commit is contained in:
foreman
2016-07-07 00:56:14 -04:00
parent 7e285f4a16
commit cd9d1dfd35
7 changed files with 82 additions and 90 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rocclr/runtime/device/pal/palresource.cpp
+7 -5
View File
@@ -1072,10 +1072,9 @@ Resource::free()
void
Resource::writeRawData(
VirtualGPU& gpu,
size_t offset,
size_t size,
size_t size,
const void* data,
bool waitForEvent) const
bool waitForEvent) const
{
GpuEvent event;
@@ -1083,8 +1082,11 @@ Resource::writeRawData(
// size needs to be DWORD aligned
assert((size & 3) == 0);
gpu.eventBegin(MainEngine);
//! @todo Remove cache flush
//! It's a workaround for a PAL crash with embedded data, allocated before any command
gpu.flushCUCaches();
gpu.queue(MainEngine).addCmdMemRef(iMem());
gpu.iCmd()->CmdUpdateMemory(*iMem(), offset, size, reinterpret_cast<const uint32_t*>(data));
gpu.iCmd()->CmdUpdateMemory(*iMem(), 0, size, reinterpret_cast<const uint32_t*>(data));
gpu.eventEnd(MainEngine, event);
setBusy(gpu, event);
@@ -1936,7 +1938,7 @@ Resource::warmUpRenames(VirtualGPU& gpu)
uint dummy = 0;
const bool NoWait = false;
// Write 0 for the buffer paging by VidMM
writeRawData(gpu, 0, sizeof(dummy), &dummy, NoWait);
writeRawData(gpu, sizeof(dummy), &dummy, NoWait);
const bool Force = true;
rename(gpu, Force);
}