tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-07-14 22:58:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: add bf16 and i32 to getrows (#14529)

Browse Source
This commit is contained in:
Aman Gupta
2025-07-07 21:45:43 +08:00
committed by GitHub
parent 6491d6e4f1
commit b9c3eefde1
2 changed files with 10 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
ggml/src/ggml-cuda/getrows.cu
View File

@ -168,6 +168,10 @@ static void ggml_cuda_get_rows_switch_src0_type(
get_rows_cuda_float((const float *) src0_d, src1_d, dst_d,
ne00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb1, nb2, nb3, stream);
break;
case GGML_TYPE_I32:
get_rows_cuda_float((const int32_t *) src0_d, src1_d, dst_d,
ne00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb1, nb2, nb3, stream);
break;
case GGML_TYPE_BF16:
get_rows_cuda_float((const nv_bfloat16 *) src0_d, src1_d, dst_d,
ne00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb1, nb2, nb3, stream);
@ -210,6 +214,10 @@ void get_rows_cuda(
ggml_cuda_get_rows_switch_src0_type(src0_d, src0_type, src1_d, (float *) dst_d,
ne00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb1, nb2, nb3, stream);
break;
case GGML_TYPE_I32:
ggml_cuda_get_rows_switch_src0_type(src0_d, src0_type, src1_d, (int32_t *) dst_d,
ne00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb1, nb2, nb3, stream);
break;
case GGML_TYPE_F16:
ggml_cuda_get_rows_switch_src0_type(src0_d, src0_type, src1_d, (half *) dst_d,
ne00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb1, nb2, nb3, stream);

2
ggml/src/ggml-cuda/ggml-cuda.cu
View File

@ -3200,6 +3200,8 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
switch (op->src[0]->type) {
case GGML_TYPE_F16:
case GGML_TYPE_F32:
case GGML_TYPE_BF16:
case GGML_TYPE_I32:
case GGML_TYPE_Q4_0:
case GGML_TYPE_Q4_1:
case GGML_TYPE_Q5_0: