Skip to content

LLM inference in C/C++

C++ ggml

ggml-blas: default hadamard mul_mat to cpu routine (#25710)

Aaron Teo11fd0a6fb756
diff --git a/ggml/src/ggml-blas/ggml-blas.cpp b/ggml/src/ggml-blas/ggml-blas.cpp
index b4c735267e..9745fa29f5 100644
--- a/ggml/src/ggml-blas/ggml-blas.cpp
+++ b/ggml/src/ggml-blas/ggml-blas.cpp
@@ -1,3 +1,4 @@
+#include "ggml.h"
#include "ggml-impl.h"
#include "ggml-blas.h"
#include "ggml-backend-impl.h"
@@ -415,6 +416,12 @@ static bool ggml_backend_blas_device_supports_op(ggml_backend_dev_t dev, const s
// TODO: find the optimal value
const int64_t min_batch = 32;
+ // default back to CPU fast path
+ // see: https://github.com/ggml-org/llama.cpp/issues/25565
+ if (ggml_get_op_params_i32(op, 1) == GGML_HINT_SRC0_IS_HADAMARD) {
+ return false;
+ }
+
return ggml_is_contiguous(src0) &&
ggml_is_contiguous(src1) &&
src1->type == GGML_TYPE_F32 &&