CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-31 08:51:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

sycl: fix mul_mat selection (#15092)

Browse Source
This commit is contained in:
Romain Biessy
2025-08-05 18:39:55 +02:00
committed by GitHub
parent c81de6e107
commit 3306ceabf0
1 changed files with 3 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
ggml/src/ggml-sycl/ggml-sycl.cpp
View File

@@ -2609,6 +2609,8 @@ static void ggml_sycl_mul_mat_vec_nc(ggml_backend_sycl_context & ctx, const ggml
GGML_ASSERT(!ggml_backend_buffer_is_sycl_split(src0->buffer)); GGML_ASSERT(!ggml_backend_buffer_is_sycl_split(src0->buffer));
GGML_ASSERT(src0->type == GGML_TYPE_F16); GGML_ASSERT(src0->type == GGML_TYPE_F16);
GGML_ASSERT(src1->type == GGML_TYPE_F32); GGML_ASSERT(src1->type == GGML_TYPE_F32);
GGML_ASSERT(src1->ne[1] == 1);
GGML_ASSERT(src1->ne[3] == 1);
const int64_t ne00 = src0->ne[0]; const int64_t ne00 = src0->ne[0];
const int64_t ne01 = src0->ne[1]; const int64_t ne01 = src0->ne[1];
@@ -3196,7 +3198,7 @@ static void ggml_sycl_mul_mat(ggml_backend_sycl_context & ctx, const ggml_tensor
// The kernel from the if path is faster for that specific case, but does not support all mul mats. // The kernel from the if path is faster for that specific case, but does not support all mul mats.
ggml_sycl_mul_mat_batched_sycl(ctx, src0, src1, dst); ggml_sycl_mul_mat_batched_sycl(ctx, src0, src1, dst);
} }
} else if (!split && src0->type == GGML_TYPE_F16 && !ggml_is_contiguous(src0) && !ggml_is_transposed(src1) && src1->ne[1] == 1) { } else if (!split && src0->type == GGML_TYPE_F16 && !ggml_is_contiguous(src0) && !ggml_is_transposed(src1) && src1->ne[1] == 1 && src1->ne[3] == 1) {
// KQV single-batch // KQV single-batch
ggml_sycl_mul_mat_vec_nc(ctx, src0, src1, dst); ggml_sycl_mul_mat_vec_nc(ctx, src0, src1, dst);
} else if (!split && src0->type == GGML_TYPE_F16 && !ggml_is_transposed(src0) && !ggml_is_transposed(src1) && src1->ne[2] * src1->ne[3] > 1) { } else if (!split && src0->type == GGML_TYPE_F16 && !ggml_is_transposed(src0) && !ggml_is_transposed(src1) && src1->ne[2] * src1->ne[3] > 1) {