Skip to content

ggml-cuda: use cudaMemcpyDefault in the ggml_cuda_cpy 2D fast path #157

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

Open
danielhanchen wants to merge 1 commit into
base: master
Choose a base branch
from
+10 −1
Open

ggml-cuda: use cudaMemcpyDefault in the ggml_cuda_cpy 2D fast path #157

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
9 changes: 8 additions & 1 deletion ggml/src/ggml-cuda/cpy.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -474,8 +474,15 @@ void ggml_cuda_cpy(ggml_backend_cuda_context & ctx, const ggml_tensor * src0, gg
CUDA_CHECK(cudaMemcpyAsync(src1_ddc, src0_ddc, ggml_nbytes(src0), cudaMemcpyDeviceToDevice, main_stream));
}
} else if (ggml_cuda_cpy_as_memcpy_2d(src0, src1, mc_width, mc_height, mc_spitch, mc_dpitch)) {
// cudaMemcpyDefault, not DeviceToDevice: under
// GGML_CUDA_ENABLE_UNIFIED_MEMORY these buffers come from
// hipMallocManaged, and managed pages are not guaranteed device
Comment on lines +478 to +479

@chatgpt-codex-connector chatgpt-codex-connector Bot Aug 31, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P1 Badge Keep the shared-path comment backend-neutral and concise

In CUDA and MUSA builds, these buffers come from cudaMallocManaged and musaMallocManaged, respectively, so the backend-shared comment incorrectly states that they come from hipMallocManaged. Generalize the explanation to managed allocations and reduce the seven-line hard-wrapped comment to the repository's expected one or two concise lines.

AGENTS.md reference: AGENTS.md:L74-L78

Useful? React with 👍 / 👎.

// resident. DeviceToDevice asserts that they are, so the runtime skips
// residency resolution and a migrated page is read as garbage. Default
// lets it infer per pointer, and is a no-op when both really are on the
// device.
CUDA_CHECK(cudaMemcpy2DAsync(src1_ddc, mc_dpitch, src0_ddc, mc_spitch,
mc_width, mc_height, cudaMemcpyDeviceToDevice, main_stream));
mc_width, mc_height, cudaMemcpyDefault, main_stream));
} else if (src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F32) {
if (can_be_transposed) {
ggml_cpy_scalar_cuda<float, float, true>
Expand Down
1 change: 1 addition & 0 deletions ggml/src/ggml-cuda/vendors/hip.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -87,6 +87,7 @@
#define cudaMemcpyAsync hipMemcpyAsync
#define cudaMemcpyPeerAsync hipMemcpyPeerAsync
#define cudaMemcpy2DAsync hipMemcpy2DAsync
#define cudaMemcpyDefault hipMemcpyDefault
#define cudaMemcpyDeviceToDevice hipMemcpyDeviceToDevice
#define cudaMemcpyDeviceToHost hipMemcpyDeviceToHost
#define cudaMemcpyHostToDevice hipMemcpyHostToDevice
Expand Down
1 change: 1 addition & 0 deletions ggml/src/ggml-cuda/vendors/musa.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -74,6 +74,7 @@
#define cudaMemcpyAsync musaMemcpyAsync
#define cudaMemcpyPeerAsync musaMemcpyPeerAsync
#define cudaMemcpy2DAsync musaMemcpy2DAsync
#define cudaMemcpyDefault musaMemcpyDefault
#define cudaMemcpyDeviceToDevice musaMemcpyDeviceToDevice
#define cudaMemcpyDeviceToHost musaMemcpyDeviceToHost
#define cudaMemcpyHostToDevice musaMemcpyHostToDevice
Expand Down