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

P4 to Git Change 1364923 by gandryey@gera-lnx-rcf-lc on 2017/01/23 11:48:45

Browse Source 
SWDEV-110996 - OCL to use the blit manager instead ROCr implementing copyRect API

	- Implement the blit manager functionality in ROCm backened. This checki-in also fixes SWDEV-95079, SWDEV-95068, SWDEV-95069, SWDEV-95071

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocblit.cpp#6 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocblit.hpp#4 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdefs.hpp#9 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#35 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.hpp#13 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmemory.cpp#6 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmemory.hpp#5 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocsettings.cpp#13 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocsettings.hpp#6 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.cpp#27 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.hpp#6 edit
This commit is contained in:
foreman
2017-01-23 11:59:51 -05:00
parent 77f4cda6b8
commit 454621b7f1
11 changed files with 2820 additions and 1527 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rocclr/runtime/device/rocm/rocsettings.cpp
+46 -15
View File
@@ -53,14 +53,38 @@ Settings::Settings()
enablePartialDispatch_ = (partialDispatch) ? false : true;
partialDispatch_ = (partialDispatch) ? false : true;
commandQueues_ = 100; //!< Field value set to maximum number
//!< concurrent Virtual GPUs for ROCm backend
//!< concurrent Virtual GPUs for ROCm backend
// Disable image DMA by default (ROCM runtime doesn't support it)
imageDMA_ = false;
stagedXferRead_ = true;
stagedXferWrite_ = true;
stagedXferSize_ = GPU_STAGING_BUFFER_SIZE * Ki;
// Initialize transfer buffer size to 1MB by default
xferBufSize_ = 1024 * Ki;
const static size_t MaxPinnedXferSize = 32;
pinnedXferSize_ = std::min(GPU_PINNED_XFER_SIZE, MaxPinnedXferSize) * Mi;
pinnedMinXferSize_ = std::min(GPU_PINNED_MIN_XFER_SIZE * Ki, pinnedXferSize_);
}
bool
Settings::create(bool doublePrecision)
Settings::create(bool fullProfile)
{
customHostAllocator_ = true;
if (fullProfile) {
pinnedXferSize_ = 0;
stagedXferSize_ = 0;
xferBufSize_ = 0;
}
else {
pinnedXferSize_ = std::max(pinnedXferSize_, pinnedMinXferSize_);
stagedXferSize_ = std::max(stagedXferSize_, pinnedMinXferSize_ + 4 * Ki);
}
// Enable extensions
enableExtension(ClKhrByteAddressableStore);
enableExtension(ClKhrGlobalInt32BaseAtomics);
@@ -72,21 +96,16 @@ Settings::create(bool doublePrecision)
enableExtension(ClKhr3DImageWrites);
enableExtension(ClAmdMediaOps);
enableExtension(ClAmdMediaOps2);
if(MesaInterop::Supported())
enableExtension(ClKhrGlSharing);
// Make sure device supports doubles
doublePrecision_ &= doublePrecision;
if (doublePrecision_) {
// Enable KHR double precision extension
enableExtension(ClKhrFp64);
#if !defined(WITH_LIGHTNING_COMPILER)
// Also enable AMD double precision extension?
enableExtension(ClAmdFp64);
#endif // !defined(WITH_LIGHTNING_COMPILER)
if(MesaInterop::Supported()) {
enableExtension(ClKhrGlSharing);
}
// Enable KHR double precision extension
enableExtension(ClKhrFp64);
#if !defined(WITH_LIGHTNING_COMPILER)
// Also enable AMD double precision extension?
enableExtension(ClAmdFp64);
#endif // !defined(WITH_LIGHTNING_COMPILER)
enableExtension(ClKhrSubGroups);
enableExtension(ClKhrDepthImages);
@@ -109,6 +128,18 @@ Settings::override()
if (!flagIsDefault(GPU_MAX_COMMAND_QUEUES)) {
commandQueues_ = GPU_MAX_COMMAND_QUEUES;
}
if (!flagIsDefault(GPU_IMAGE_DMA)) {
commandQueues_ = GPU_IMAGE_DMA;
}
if (!flagIsDefault(GPU_XFER_BUFFER_SIZE)) {
xferBufSize_ = GPU_XFER_BUFFER_SIZE * Ki;
}
if (!flagIsDefault(GPU_PINNED_MIN_XFER_SIZE)) {
pinnedMinXferSize_ = std::min(GPU_PINNED_MIN_XFER_SIZE * Ki, pinnedXferSize_);
}
}
} // namespace roc