Skip to content

Commit ce170f7

Browse files
jan-service-accountjan-service-account
authored
Merge pull request #119 from menloresearch/update-dev-from-master-2025-06-09-00-09
Sync master with upstream release b5606
2 parents 71ad2d5 + 247e5c6 commit ce170f7

File tree

2 files changed

+114
-4
lines changed

2 files changed

+114
-4
lines changed

‎.github/workflows/build-linux-cross.yml‎

Lines changed: 113 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -231,3 +231,116 @@ jobs:
231231
-DCMAKE_FIND_ROOT_PATH_MODE_INCLUDE=BOTH
232232
233233
cmake --build build --config Release -j $(nproc)
234+
235+
debian-13-loongarch64-cpu-cross:
236+
runs-on: ubuntu-24.04
237+
container: debian@sha256:653dfb9f86c3782e8369d5f7d29bb8faba1f4bff9025db46e807fa4c22903671
238+
239+
steps:
240+
- uses: actions/checkout@v4
241+
- name: Setup LoongArch
242+
run: |
243+
rm -f /etc/apt/sources.list.d/*
244+
cat << EOF | tee /etc/apt/sources.list.d/debian-ports.list
245+
deb http://snapshot.debian.org/archive/debian/20250515T202920Z/ trixie main
246+
EOF
247+
( echo 'quiet "true";'; \
248+
echo 'APT::Get::Assume-Yes "true";'; \
249+
echo 'APT::Install-Recommends "false";'; \
250+
echo 'Acquire::Check-Valid-Until "false";'; \
251+
echo 'Acquire::Retries "5";'; \
252+
) > /etc/apt/apt.conf.d/99snapshot-repos
253+
254+
apt-get update
255+
apt-get install -y ca-certificates debian-ports-archive-keyring cmake git zip
256+
dpkg --add-architecture loong64
257+
258+
# Add arch-specific repositories for non-amd64 architectures
259+
cat << EOF | tee /etc/apt/sources.list.d/loong64-ports.list
260+
deb [arch=loong64] http://snapshot.debian.org/archive/debian-ports/20250515T194251Z/ sid main
261+
EOF
262+
263+
apt-get update || true ;# Prevent failure due to missing URLs.
264+
265+
apt-get install -y --no-install-recommends \
266+
build-essential \
267+
gcc-14-loongarch64-linux-gnu \
268+
g++-14-loongarch64-linux-gnu
269+
270+
- name: Build
271+
run: |
272+
cmake -B build -DLLAMA_CURL=OFF \
273+
-DCMAKE_BUILD_TYPE=Release \
274+
-DGGML_OPENMP=OFF \
275+
-DLLAMA_BUILD_EXAMPLES=ON \
276+
-DLLAMA_BUILD_TOOLS=ON \
277+
-DLLAMA_BUILD_TESTS=OFF \
278+
-DCMAKE_SYSTEM_NAME=Linux \
279+
-DCMAKE_SYSTEM_PROCESSOR=loongarch64 \
280+
-DCMAKE_C_COMPILER=loongarch64-linux-gnu-gcc-14 \
281+
-DCMAKE_CXX_COMPILER=loongarch64-linux-gnu-g++-14 \
282+
-DCMAKE_POSITION_INDEPENDENT_CODE=ON \
283+
-DCMAKE_FIND_ROOT_PATH=/usr/lib/loongarch64-linux-gnu \
284+
-DCMAKE_FIND_ROOT_PATH_MODE_PROGRAM=NEVER \
285+
-DCMAKE_FIND_ROOT_PATH_MODE_LIBRARY=ONLY \
286+
-DCMAKE_FIND_ROOT_PATH_MODE_INCLUDE=BOTH
287+
288+
cmake --build build --config Release -j $(nproc)
289+
290+
debian-13-loongarch64-vulkan-cross:
291+
runs-on: ubuntu-24.04
292+
container: debian@sha256:653dfb9f86c3782e8369d5f7d29bb8faba1f4bff9025db46e807fa4c22903671
293+
294+
steps:
295+
- uses: actions/checkout@v4
296+
- name: Setup LoongArch
297+
run: |
298+
rm -f /etc/apt/sources.list.d/*
299+
cat << EOF | tee /etc/apt/sources.list.d/debian-ports.list
300+
deb http://snapshot.debian.org/archive/debian/20250515T202920Z/ trixie main
301+
EOF
302+
( echo 'quiet "true";'; \
303+
echo 'APT::Get::Assume-Yes "true";'; \
304+
echo 'APT::Install-Recommends "false";'; \
305+
echo 'Acquire::Check-Valid-Until "false";'; \
306+
echo 'Acquire::Retries "5";'; \
307+
) > /etc/apt/apt.conf.d/99snapshot-repos
308+
309+
apt-get update
310+
apt-get install -y ca-certificates debian-ports-archive-keyring cmake git zip
311+
dpkg --add-architecture loong64
312+
313+
# Add arch-specific repositories for non-amd64 architectures
314+
cat << EOF | tee /etc/apt/sources.list.d/loong64-ports.list
315+
deb [arch=loong64] http://snapshot.debian.org/archive/debian-ports/20250515T194251Z/ sid main
316+
EOF
317+
318+
apt-get update || true ;# Prevent failure due to missing URLs.
319+
320+
apt-get install -y --no-install-recommends \
321+
build-essential \
322+
glslc \
323+
gcc-14-loongarch64-linux-gnu \
324+
g++-14-loongarch64-linux-gnu \
325+
libvulkan-dev:loong64
326+
327+
- name: Build
328+
run: |
329+
cmake -B build -DLLAMA_CURL=OFF \
330+
-DCMAKE_BUILD_TYPE=Release \
331+
-DGGML_VULKAN=ON \
332+
-DGGML_OPENMP=OFF \
333+
-DLLAMA_BUILD_EXAMPLES=ON \
334+
-DLLAMA_BUILD_TOOLS=ON \
335+
-DLLAMA_BUILD_TESTS=OFF \
336+
-DCMAKE_SYSTEM_NAME=Linux \
337+
-DCMAKE_SYSTEM_PROCESSOR=loongarch64 \
338+
-DCMAKE_C_COMPILER=loongarch64-linux-gnu-gcc-14 \
339+
-DCMAKE_CXX_COMPILER=loongarch64-linux-gnu-g++-14 \
340+
-DCMAKE_POSITION_INDEPENDENT_CODE=ON \
341+
-DCMAKE_FIND_ROOT_PATH=/usr/lib/loongarch64-linux-gnu \
342+
-DCMAKE_FIND_ROOT_PATH_MODE_PROGRAM=NEVER \
343+
-DCMAKE_FIND_ROOT_PATH_MODE_LIBRARY=ONLY \
344+
-DCMAKE_FIND_ROOT_PATH_MODE_INCLUDE=BOTH
345+
346+
cmake --build build --config Release -j $(nproc)

‎ggml/src/ggml-cuda/ggml-cuda.cu‎

Lines changed: 1 addition & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -1144,7 +1144,6 @@ typedef void (*ggml_cuda_op_mul_mat_t)(
11441144
static cudaError_t ggml_cuda_cpy_tensor_2d(
11451145
void * dst, const struct ggml_tensor * src, int64_t i3, int64_t i2, int64_t i1_low, int64_t i1_high, cudaStream_t stream) {
11461146

1147-
GGML_ASSERT(ggml_backend_buffer_is_cuda(src->buffer));
11481147
const char * src_ptr = (const char *) src->data;
11491148
char * dst_ptr = (char *) dst;
11501149

@@ -1427,8 +1426,6 @@ static void ggml_cuda_op_mul_mat(
14271426
const int64_t nb2 = dst->nb[2];
14281427
const int64_t nb3 = dst->nb[3];
14291428

1430-
GGML_ASSERT(ggml_backend_buffer_is_cuda(dst->buffer));
1431-
GGML_ASSERT(ggml_backend_buffer_is_cuda(src1->buffer));
14321429
ggml_backend_cuda_buffer_context * src1_ctx = (ggml_backend_cuda_buffer_context *) src1->buffer->context;
14331430
ggml_backend_cuda_buffer_context * dst_ctx = (ggml_backend_cuda_buffer_context *) dst->buffer->context;
14341431

@@ -1750,7 +1747,7 @@ static void ggml_cuda_mul_mat_batched_cublas(ggml_backend_cuda_context & ctx, co
17501747
GGML_ASSERT(!ggml_is_transposed(src0));
17511748
GGML_ASSERT(!ggml_is_transposed(src1));
17521749

1753-
GGML_ASSERT(ggml_backend_buffer_is_cuda(src0->buffer));
1750+
GGML_ASSERT(!ggml_backend_buft_is_cuda_split(src0->buffer->buft));
17541751
GGML_ASSERT(src0->type == GGML_TYPE_F16);
17551752

17561753
// Byte offsets and tensor dimensions are currently used in an inconsistent way for dst.

0 commit comments

Comments
 (0)