tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-07-21 18:28:31 +00:00
Code Issues Packages Projects Releases Wiki Activity

CPU/CUDA: fix (GQA) mul mat back, add CUDA support (#11380)

Browse Source
This commit is contained in:
Johannes Gäßler
2025-01-24 12:38:31 +01:00
committed by GitHub
parent 1af6945eb0
commit 8137b4bb2b
7 changed files with 156 additions and 61 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
ggml/src/ggml-cpu/ggml-cpu.c
View File

@ -7883,7 +7883,7 @@ static void ggml_compute_forward_out_prod_f32(
float * s0 = (float *) ((char *) src0->data + ( i01*nb01 + i02*nb02 + i03*nb03));
float * s1 = (float *) ((char *) src1->data + (i1*nb10 + i11*nb11 + i12*nb12 + i13*nb13));
float * d = (float *) ((char *) dst->data + ( i1*nb1 + i2*nb2 + i3*nb3));
float * d = (float *) ((char *) dst->data + ( i1*nb1 + i2*nb2 + i3*nb3));
ggml_vec_mad_f32_unroll(ne0, nb01, nb11, d, s0, s1);
}
@ -7892,7 +7892,7 @@ static void ggml_compute_forward_out_prod_f32(
float * s0 = (float *) ((char *) src0->data + ( i01*nb01 + i02*nb02 + i03*nb03));
float * s1 = (float *) ((char *) src1->data + (i1*nb10 + i11*nb11 + i12*nb12 + i13*nb13));
float * d = (float *) ((char *) dst->data + ( i1*nb1 + i2*nb2 + i3*nb3));
float * d = (float *) ((char *) dst->data + ( i1*nb1 + i2*nb2 + i3*nb3));
ggml_vec_mad_f32(ne0, d, s0, *s1);
}

3
ggml/src/ggml-cpu/ggml-cpu.cpp
View File

@ -416,7 +416,8 @@ static bool ggml_backend_cpu_device_supports_op(ggml_backend_dev_t dev, const st
case GGML_OP_IM2COL_BACK:
return src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F32;
case GGML_OP_OUT_PROD:
return (src0->type == GGML_TYPE_F32 || ggml_is_quantized(src0->type)) && src1->type == GGML_TYPE_F32;
return (src0->type == GGML_TYPE_F32 || (ggml_is_quantized(src0->type) && src0->ne[2] == src1->ne[2] && src0->ne[3] == src1->ne[3])) &&
src1->type == GGML_TYPE_F32 && op->type == GGML_TYPE_F32;
default:
return true;
}