CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-21 12:16:57 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggml-zdnn: activate fp16 and bf16

Browse Source 
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
This commit is contained in:
Aaron Teo
2025-09-06 21:57:21 +08:00
parent 0a08b1df0a
commit 9ed5947a12
1 changed files with 18 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
ggml/src/ggml-zdnn/ggml-zdnn.cpp
View File

@@ -258,22 +258,30 @@ static bool ggml_zdnn_supports_op(const ggml_backend_zdnn_device_context * ctx_d
case GGML_OP_MUL_MAT: case GGML_OP_MUL_MAT:
{ {
const ggml_tensor * src0 = op->src[0]; const ggml_tensor * weights = op->src[0];
const ggml_tensor * src1 = op->src[1]; const ggml_tensor * inputs = op->src[1];
const int64_t ne10 = src1->ne[0]; const int64_t ne10 = inputs->ne[0];
const int64_t ne0 = op->ne[0]; const int64_t ne0 = op->ne[0];
const int64_t ne1 = op->ne[1]; const int64_t ne1 = op->ne[1];
const int64_t max_batch = ctx_dev->max_size; const int64_t max_batch = ctx_dev->max_size;
return ggml_is_matrix(src0) && if (!ggml_is_matrix(weights) || !ggml_is_matrix(inputs) ||
ggml_is_matrix(src1) && !ggml_is_contiguous(weights) || !ggml_is_contiguous(inputs) ||
ggml_is_contiguous(src0) && weights->view_src != nullptr || inputs->view_src != nullptr ||
ggml_is_contiguous(src1) && ne0 > max_batch || ne1 > max_batch || ne10 > max_batch) {
src0->view_src == nullptr && src1->view_src == nullptr && return false;
src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F32 && }
(ne0 <= max_batch && ne1 <= max_batch && ne10 <= max_batch);
switch (weights->type) {
case GGML_TYPE_F32:
case GGML_TYPE_F16:
case GGML_TYPE_BF16:
return true;
default:
return false;
}
} break; } break;
default: default: