furyhawk/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-07-21 10:15:35 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix bf16 graph reduce type (#1938)

Browse Source
This commit is contained in:
Kawrakow
2026-06-08 16:51:05 +02:00
committed by GitHub
parent 1660459db5
commit eea6a82b25
1 changed files with 12 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cuda/binbcast.cu
+12
View File
@@ -476,6 +476,18 @@ void ggml_cuda_op_add(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
else if (dst->src[0]->type == GGML_TYPE_F32 && dst->src[1]->type == GGML_TYPE_F32) {
k_fast_add_2<<<nblocks, kBlockSize, 0, ctx.stream()>>>(dst->ne[0], nelem,
(const float *)dst->src[0]->data, (const float *)dst->src[1]->data, (float *)dst->data);
}
else if (dst->src[0]->type == GGML_TYPE_BF16 && dst->src[1]->type == GGML_TYPE_BF16) {
k_fast_add_2<<<nblocks, kBlockSize, 0, ctx.stream()>>>(dst->ne[0], nelem,
(const nv_bfloat16 *)dst->src[0]->data, (const nv_bfloat16 *)dst->src[1]->data, (float *)dst->data);
}
else if (dst->src[0]->type == GGML_TYPE_BF16 && dst->src[1]->type == GGML_TYPE_F32) {
k_fast_add_2<<<nblocks, kBlockSize, 0, ctx.stream()>>>(dst->ne[0], nelem,
(const nv_bfloat16 *)dst->src[0]->data, (const float *)dst->src[1]->data, (float *)dst->data);
}
else if (dst->src[0]->type == GGML_TYPE_F32 && dst->src[1]->type == GGML_TYPE_BF16) {
k_fast_add_2<<<nblocks, kBlockSize, 0, ctx.stream()>>>(dst->ne[0], nelem,
(const float *)dst->src[0]->data, (const nv_bfloat16 *)dst->src[1]->data, (float *)dst->data);
} else {
k_fast_add_2<<<nblocks, kBlockSize, 0, ctx.stream()>>>(dst->ne[0], nelem,
(const float *)dst->src[0]->data, (const half *)dst->src[1]->data, (float *)dst->data);