Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

ggml : fix conv 2d dw (#25490)

Browse Source
This commit is contained in:
Georgi Gerganov
2026-07-09 17:56:32 +03:00
committed by GitHub
parent c15c5c77a4
commit 5c3a586860
2 changed files with 6 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cpu/ops.cpp
+4 -3
View File
@@ -7324,11 +7324,12 @@ static void ggml_compute_forward_conv_2d_dw_cwhn(
#ifdef GGML_SIMD #ifdef GGML_SIMD
int64_t c_pkg_end = 0; int64_t c_pkg_end = 0;
int64_t pkg_size = GGML_F32_EPR;
if (knl_type == GGML_TYPE_F32) { if (knl_type == GGML_TYPE_F32) {
#if defined(__ARM_FEATURE_SVE) #if defined(__ARM_FEATURE_SVE)
const int64_t pkg_size = svcntw(); pkg_size = svcntw();
#else #else
const int64_t pkg_size = GGML_F32_EPR; pkg_size = GGML_F32_EPR;
#endif #endif
c_pkg_end = (c / pkg_size) * pkg_size; c_pkg_end = (c / pkg_size) * pkg_size;
} }
@@ -7345,7 +7346,7 @@ static void ggml_compute_forward_conv_2d_dw_cwhn(
const int64_t src_x_base = dst_x * p.stride_x - p.pad_x; const int64_t src_x_base = dst_x * p.stride_x - p.pad_x;
#ifdef GGML_SIMD #ifdef GGML_SIMD
for (int64_t c_i = 0; c_i < c_pkg_end; c_i += GGML_F32_EPR) { for (int64_t c_i = 0; c_i < c_pkg_end; c_i += pkg_size) {
GGML_F32_VEC sum = GGML_F32_VEC_ZERO; GGML_F32_VEC sum = GGML_F32_VEC_ZERO;
for (int64_t knl_y = 0; knl_y < p.knl_h; ++knl_y) { for (int64_t knl_y = 0; knl_y < p.knl_h; ++knl_y) {
const int64_t src_y = src_y_base + knl_y * p.dilation_y; const int64_t src_y = src_y_base + knl_y * p.dilation_y;
ggml/src/ggml-cuda/ggml-cuda.cu
+2
View File
@@ -4917,7 +4917,9 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
case GGML_OP_IM2COL: case GGML_OP_IM2COL:
case GGML_OP_IM2COL_3D: case GGML_OP_IM2COL_3D:
case GGML_OP_CONV_2D: case GGML_OP_CONV_2D:
return true;
case GGML_OP_CONV_2D_DW: case GGML_OP_CONV_2D_DW:
return op->src[0]->type == GGML_TYPE_F32;
case GGML_OP_CONV_TRANSPOSE_2D: case GGML_OP_CONV_TRANSPOSE_2D:
case GGML_OP_POOL_2D: case GGML_OP_POOL_2D:
return true; return true;