EngineX-Ascend/enginex-ascend-910-llama.cpp
10
0
Fork 0
Code Issues Pull Requests Actions 4 Projects Releases Wiki Activity

SYCL: fix rms_norm_mul_add for tensor dim not a multiple of sg_size (#15592)

Browse Source 
The original implementation unconditionally returned true for this operation, leading to a failure when the tensor's first dimension (ne[0]) was not a multiple of WARP_SIZE. This caused an GGML_ASSERT(ncols % WARP_SIZE == 0) failure in ggml-sycl/norm.cpp.

This change updates the ggml_backend_sycl_device_supports_op check to correctly return true for GGML_OP_RMS_NORM only when the first dimension of the tensor is a multiple of WARP_SIZE, ensuring the operation can be performed without error.
This commit is contained in:
Akarshan Biswas
2025-08-27 00:27:49 +05:30
committed by GitHub
parent 8ce3ff1d91
commit 8b69686136
1 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -4364,11 +4364,12 @@ static bool ggml_backend_sycl_device_supports_op(ggml_backend_dev_t dev, const g
return (op->type == GGML_TYPE_F32 && op->src[0]->type == GGML_TYPE_F32) && (op->type == op->src[0]->type); return (op->type == GGML_TYPE_F32 && op->src[0]->type == GGML_TYPE_F32) && (op->type == op->src[0]->type);
#endif #endif
case GGML_OP_NORM: case GGML_OP_NORM:
case GGML_OP_RMS_NORM:
return true; return true;
case GGML_OP_L2_NORM: case GGML_OP_L2_NORM:
case GGML_OP_GROUP_NORM: case GGML_OP_GROUP_NORM:
return ggml_is_contiguous(op->src[0]); return ggml_is_contiguous(op->src[0]);
case GGML_OP_RMS_NORM:
return ((op->src[0]->ne[0] % WARP_SIZE) == 0);
case GGML_OP_SCALE: case GGML_OP_SCALE:
return true; return true;
case GGML_OP_CONT: case GGML_OP_CONT: