P4 to Git Change 1083967 by gandryey@gera-dev-w7 on 2014/10/03 11:20:24

ECR #304775 - Fix for BUG#10330.
	- Add an optimized version for unaligned buffer copy

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/blitcl.cpp#7 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpublit.cpp#111 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/hsa/hsablit.cpp#9 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/hsa_foundation/hsablit.cpp#5 edit
This commit is contained in:
foreman
2014-10-03 12:04:15 -04:00
parent 7cd16dddbb
commit bfc41a18dd
3 changed files with 60 additions and 8 deletions
+12 -1
View File
@@ -1055,7 +1055,15 @@ KernelBlitManager::copyBuffer(
}
}
size.c[0] /= CopyBuffAlignment[i];
cl_uint remain;
if (blitType == BlitCopyBufferAligned) {
size.c[0] /= CopyBuffAlignment[i];
}
else {
remain = size[0] % 4;
size.c[0] /= 4;
size.c[0] += 1;
}
// Program the dispatch dimensions
localWorkSize = 256;
@@ -1081,6 +1089,9 @@ KernelBlitManager::copyBuffer(
cl_int alignment = CopyBuffAlignment[i];
kernels_[blitType]->parameters().set(5, sizeof(alignment), &alignment);
}
else {
kernels_[blitType]->parameters().set(5, sizeof(remain), &remain);
}
// Create ND range object for the kernel's execution
amd::NDRangeContainer ndrange(