ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

CANN: Add 310P operator support check (#12962)

Browse source
This commit is contained in:
Chenguang Li 2025-04-16 16:21:05 +08:00 • committed by GitHub
parent 80f19b4186
commit b43d89e311
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 18 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

10
ggml/src/ggml-cann/ggml-cann.cpp
View file

@ -2022,6 +2022,10 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
return true;
case GGML_TYPE_Q8_0:
case GGML_TYPE_Q4_0:
#ifdef ASCEND_310P
// Q4 && Q8 per group is not suppor on 310p device
return false;
#endif
// only support contiguous for quantized types.
return ggml_is_contiguous(op->src[0]) &&
ggml_is_contiguous(op->src[1]);
@ -2107,6 +2111,12 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
}
case GGML_OP_POOL_2D: {
const int32_t * opts = (const int32_t *) op->op_params;
#ifdef ASCEND_310P
enum ggml_op_pool opt = static_cast<ggml_op_pool>(opts[0]);
if(opt == GGML_OP_POOL_MAX){
return false;
}
#endif
const int k0 = opts[1];
const int k1 = opts[2];
const int p0 = opts[5];