tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-28 12:25:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

musa: bump MUSA SDK version to rc3.1.1 (#11822)

Browse Source 
* musa: Update MUSA SDK version to rc3.1.1

Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>

* musa: Remove workaround in PR #10042

Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>

---------

Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
This commit is contained in:
R0CKSTAR
2025-02-13 20:28:18 +08:00
committed by GitHub
parent c7f460ab88
commit bd6e55bfd3
4 changed files with 3 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
.devops/musa.Dockerfile
View File

@ -1,6 +1,6 @@
ARG UBUNTU_VERSION=22.04
# This needs to generally match the container host's environment.
ARG MUSA_VERSION=rc3.1.0
ARG MUSA_VERSION=rc3.1.1
# Target the MUSA build image
ARG BASE_MUSA_DEV_CONTAINER=mthreads/musa:${MUSA_VERSION}-devel-ubuntu${UBUNTU_VERSION}

2
.github/workflows/build.yml vendored
View File

@ -443,7 +443,7 @@ jobs:
ubuntu-22-cmake-musa:
runs-on: ubuntu-22.04
container: mthreads/musa:rc3.1.0-devel-ubuntu22.04
container: mthreads/musa:rc3.1.1-devel-ubuntu22.04
steps:
- name: Clone

2
docs/docker.md
View File

@ -104,7 +104,7 @@ You may want to pass in some different `ARGS`, depending on the MUSA environment
The defaults are:
- `MUSA_VERSION` set to `rc3.1.0`
- `MUSA_VERSION` set to `rc3.1.1`
The resulting images, are essentially the same as the non-MUSA images:

5
ggml/src/ggml-cuda/ggml-cuda.cu
View File

@ -1480,12 +1480,7 @@ static void ggml_cuda_op_mul_mat(
const size_t nbytes_data = ggml_nbytes(src0);
const size_t nbytes_padding = ggml_row_size(src0->type, MATRIX_ROW_PADDING - ne00 % MATRIX_ROW_PADDING);
dev[id].src0_dd = dev[id].src0_dd_alloc.alloc(ctx.pool(id), nbytes_data + nbytes_padding);
// TODO: remove this for MUSA once the Guilty Lockup issue is resolved
#ifndef GGML_USE_MUSA
CUDA_CHECK(cudaMemsetAsync(dev[id].src0_dd, 0, nbytes_data + nbytes_padding, stream));
#else // GGML_USE_MUSA
CUDA_CHECK(cudaMemsetAsync(dev[id].src0_dd + nbytes_data, 0, nbytes_padding, stream));
#endif // !GGML_USE_MUSA
}
// If src0 is on a temporary compute buffer (partial offloading) there may be some padding that needs to be cleared: