Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[Bindless][Exp] Add support for device to device copies between CuArrays #1265

Merged
merged 4 commits into from
Apr 24, 2024
Merged

[Bindless][Exp] Add support for device to device copies between CuArrays #1265

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
f489829
[Bindless][CUDA] Add initial support for device to device copies
cppchedy Jan 11, 2024
624bd83
move the scope of sync to only device to device copies instead of all.
cppchedy Apr 11, 2024
25578d6
add comment to indicate that the device to device copy calls are sync…
cppchedy Apr 11, 2024
f9fb116
fix odd formatting
cppchedy Apr 11, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
83 changes: 76 additions & 7 deletions source/adapters/cuda/image.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -690,12 +690,12 @@ UR_APIEXPORT ur_result_t UR_APICALL urBindlessImagesImageCopyExp(
}
} else if (pImageDesc->type == UR_MEM_TYPE_IMAGE2D) {
CUDA_MEMCPY2D cpy_desc = {};
cpy_desc.srcMemoryType = CUmemorytype_enum::CU_MEMORYTYPE_HOST;
cpy_desc.srcHost = pSrc;
cpy_desc.srcXInBytes = srcOffset.x * PixelSizeBytes;
cpy_desc.srcY = srcOffset.y;
cpy_desc.dstXInBytes = dstOffset.x * PixelSizeBytes;
cpy_desc.dstY = dstOffset.y;
cpy_desc.srcMemoryType = CUmemorytype_enum::CU_MEMORYTYPE_HOST;
cpy_desc.srcHost = pSrc;
isaacault marked this conversation as resolved.
Show resolved Hide resolved
cpy_desc.srcPitch = hostExtent.width * PixelSizeBytes;
if (pImageDesc->rowPitch == 0) {
cpy_desc.dstMemoryType = CUmemorytype_enum::CU_MEMORYTYPE_ARRAY;
Expand Down Expand Up @@ -781,8 +781,6 @@ UR_APIEXPORT ur_result_t UR_APICALL urBindlessImagesImageCopyExp(
cpy_desc.srcY = srcOffset.y;
cpy_desc.dstXInBytes = dstOffset.x;
cpy_desc.dstY = dstOffset.y;
cpy_desc.dstMemoryType = CUmemorytype_enum::CU_MEMORYTYPE_HOST;
cpy_desc.dstHost = pDst;
if (pImageDesc->rowPitch == 0) {
cpy_desc.srcMemoryType = CUmemorytype_enum::CU_MEMORYTYPE_ARRAY;
cpy_desc.srcArray = (CUarray)pSrc;
Expand All @@ -792,6 +790,8 @@ UR_APIEXPORT ur_result_t UR_APICALL urBindlessImagesImageCopyExp(
cpy_desc.srcPitch = pImageDesc->rowPitch;
cpy_desc.srcDevice = (CUdeviceptr)pSrc;
}
cpy_desc.dstMemoryType = CUmemorytype_enum::CU_MEMORYTYPE_HOST;
cpy_desc.dstHost = pDst;
cpy_desc.WidthInBytes = PixelSizeBytes * copyExtent.width;
cpy_desc.Height = copyExtent.height;
UR_CHECK_ERROR(cuMemcpy2DAsync(&cpy_desc, Stream));
Expand Down Expand Up @@ -831,10 +831,79 @@ UR_APIEXPORT ur_result_t UR_APICALL urBindlessImagesImageCopyExp(
UR_CHECK_ERROR(cuMemcpy3DAsync(&cpy_desc, Stream));
}
} else {
/// imageCopyFlags == UR_EXP_IMAGE_COPY_FLAG_DEVICE_TO_DEVICE
cppchedy marked this conversation as resolved.
Show resolved Hide resolved
/// TODO: implemet device to device copy
return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
// imageCopyFlags == UR_EXP_IMAGE_COPY_FLAG_DEVICE_TO_DEVICE

// All the following async copy function calls should be treated as
// synchronous because of the explicit call to cuStreamSynchronize at
// the end
if (pImageDesc->type == UR_MEM_TYPE_IMAGE1D) {
CUDA_MEMCPY2D cpy_desc = {};
cpy_desc.srcXInBytes = srcOffset.x;
cpy_desc.srcY = 0;
cpy_desc.dstXInBytes = dstOffset.x;
cpy_desc.dstY = 0;
cpy_desc.srcMemoryType = CUmemorytype_enum::CU_MEMORYTYPE_ARRAY;
cpy_desc.srcArray = (CUarray)pSrc;
cpy_desc.dstMemoryType = CUmemorytype_enum::CU_MEMORYTYPE_ARRAY;
cpy_desc.dstArray = (CUarray)pDst;
cpy_desc.WidthInBytes = PixelSizeBytes * copyExtent.width;
cpy_desc.Height = 1;
UR_CHECK_ERROR(cuMemcpy2DAsync(&cpy_desc, Stream));
przemektmalon marked this conversation as resolved.
Show resolved Hide resolved
} else if (pImageDesc->type == UR_MEM_TYPE_IMAGE2D) {
CUDA_MEMCPY2D cpy_desc = {};
cpy_desc.srcXInBytes = srcOffset.x;
cpy_desc.srcY = srcOffset.y;
cpy_desc.dstXInBytes = dstOffset.x;
cpy_desc.dstY = dstOffset.y;
cpy_desc.srcMemoryType = CUmemorytype_enum::CU_MEMORYTYPE_ARRAY;
cpy_desc.srcArray = (CUarray)pSrc;
cpy_desc.dstMemoryType = CUmemorytype_enum::CU_MEMORYTYPE_ARRAY;
cpy_desc.dstArray = (CUarray)pDst;
cpy_desc.WidthInBytes = PixelSizeBytes * copyExtent.width;
cpy_desc.Height = copyExtent.height;
UR_CHECK_ERROR(cuMemcpy2DAsync(&cpy_desc, Stream));
} else if (pImageDesc->type == UR_MEM_TYPE_IMAGE3D) {
CUDA_MEMCPY3D cpy_desc = {};
cpy_desc.srcXInBytes = srcOffset.x;
cpy_desc.srcY = srcOffset.y;
cpy_desc.srcZ = srcOffset.z;
cpy_desc.dstXInBytes = dstOffset.x;
cpy_desc.dstY = dstOffset.y;
cpy_desc.dstZ = dstOffset.z;
cpy_desc.srcMemoryType = CUmemorytype_enum::CU_MEMORYTYPE_ARRAY;
cpy_desc.srcArray = (CUarray)pSrc;
cpy_desc.dstMemoryType = CUmemorytype_enum::CU_MEMORYTYPE_ARRAY;
cpy_desc.dstArray = (CUarray)pDst;
cpy_desc.WidthInBytes = PixelSizeBytes * copyExtent.width;
cpy_desc.Height = copyExtent.height;
cpy_desc.Depth = copyExtent.depth;
UR_CHECK_ERROR(cuMemcpy3DAsync(&cpy_desc, Stream));
} else if (pImageDesc->type == UR_MEM_TYPE_IMAGE1D_ARRAY ||
pImageDesc->type == UR_MEM_TYPE_IMAGE2D_ARRAY ||
pImageDesc->type == UR_MEM_TYPE_IMAGE_CUBEMAP_EXP) {
CUDA_MEMCPY3D cpy_desc = {};
cpy_desc.srcXInBytes = srcOffset.x;
cpy_desc.srcY = srcOffset.y;
cpy_desc.srcZ = srcOffset.z;
cpy_desc.dstXInBytes = dstOffset.x;
cpy_desc.dstY = dstOffset.y;
cpy_desc.dstZ = dstOffset.z;
cpy_desc.srcMemoryType = CUmemorytype_enum::CU_MEMORYTYPE_ARRAY;
cpy_desc.srcArray = (CUarray)pSrc;
cpy_desc.dstMemoryType = CUmemorytype_enum::CU_MEMORYTYPE_ARRAY;
cpy_desc.dstArray = (CUarray)pDst;
cpy_desc.WidthInBytes = PixelSizeBytes * copyExtent.width;
cpy_desc.Height = std::max(uint64_t{1}, copyExtent.height);
przemektmalon marked this conversation as resolved.
Show resolved Hide resolved
cpy_desc.Depth = pImageDesc->arraySize;
UR_CHECK_ERROR(cuMemcpy3DAsync(&cpy_desc, Stream));
}
isaacault marked this conversation as resolved.
Show resolved Hide resolved
// Synchronization is required here to handle the case of copying data
// from host to device, then device to device and finally device to host.
// Without it, there is a risk of the copies not being executed in the
// intended order.
cuStreamSynchronize(Stream);
}

if (phEvent) {
auto NewEvent = ur_event_handle_t_::makeNative(UR_COMMAND_MEM_IMAGE_COPY,
hQueue, Stream);
Expand Down