DarthAffe/stable-diffusion.cpp
1
0
Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-09-25 12:40:41 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat: use backend-native FP8 matmul when supported (#1916)

Browse Source
This commit is contained in:
leejet 2026-08-30 20:19:56 +08:00 committed by GitHub
parent dc4000d9f8
commit 2540a4fc25
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 11 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
src/core/ggml_extend.hpp
View File

@ -3470,8 +3470,18 @@ public:
ggml_tensor* w = params["weight"]; ggml_tensor* w = params["weight"];
ggml_tensor* weight_scale = has_weight_scale ? params["weight_scale"] : nullptr; ggml_tensor* weight_scale = has_weight_scale ? params["weight_scale"] : nullptr;
if (w->type == GGML_TYPE_F8_E4M3 || w->type == GGML_TYPE_F8_E5M2) { if (w->type == GGML_TYPE_F8_E4M3 || w->type == GGML_TYPE_F8_E5M2) {
bool supports_fp8_matmul = false;
if (ctx->backend != nullptr) {
ggml_tensor* fp8_matmul = ggml_mul_mat(ctx->ggml_ctx, w, x);
if (force_prec_f32) {
ggml_mul_mat_set_prec(fp8_matmul, GGML_PREC_F32);
}
supports_fp8_matmul = ggml_backend_supports_op(ctx->backend, fp8_matmul);
}
if (!supports_fp8_matmul) {
w = ggml_cast(ctx->ggml_ctx, w, GGML_TYPE_BF16); w = ggml_cast(ctx->ggml_ctx, w, GGML_TYPE_BF16);
} }
}
ggml_tensor* b = nullptr; ggml_tensor* b = nullptr;
if (bias) { if (bias) {
b = params["bias"]; b = params["bias"];