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

[SYCL] support op col2im_1d (#25264)

Browse Source 
* support op col2im_1d

* update ops.md

* rm unused words

* update for bf16

* optimize 1%-11% as the review comments

* fix the format issue

* update as the review comments
This commit is contained in:
Neo Zhang
2026-07-07 11:07:46 +03:00
committed by GitHub
parent 55edb2de44
commit 47e1de77aa
7 changed files with 162 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-sycl/ggml-sycl.cpp
+16 -1
View File
@@ -4771,6 +4771,11 @@ static void ggml_sycl_im2col_3d(ggml_backend_sycl_context & ctx, ggml_tensor * d
ggml_sycl_op_im2col_3d(ctx, dst);
}
static void ggml_sycl_col2im_1d(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
scope_op_debug_print scope_dbg_print(__func__, dst, /*num_src=*/1);
ggml_sycl_op_col2im_1d(ctx, dst);
}
static void ggml_sycl_conv_3d(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
scope_op_debug_print scope_dbg_print(__func__, dst, /*num_src=*/2);
ggml_sycl_op_conv_3d(ctx, dst);
@@ -5097,6 +5102,9 @@ static bool ggml_sycl_compute_forward(ggml_backend_sycl_context & ctx, struct gg
case GGML_OP_IM2COL_3D:
ggml_sycl_im2col_3d(ctx, dst);
break;
case GGML_OP_COL2IM_1D:
ggml_sycl_col2im_1d(ctx, dst);
break;
case GGML_OP_POOL_2D:
ggml_sycl_pool2d(ctx, dst);
break;
@@ -5646,7 +5654,6 @@ static bool do_ggml_backend_sycl_device_supports_op(ggml_backend_dev_t dev, cons
// TODO: This specific configuration can fail with oneDNN and needs more debugging
if (!ggml_is_permuted(a) && ggml_is_permuted(b) && b->ne[2] > 1 && b->ne[3] > 1 &&
a->ne[0] > 128 && a->ne[2] == 1 && src0_type == GGML_TYPE_F16) {
printf("zjy 2\n");
return false;
}
return true;
@@ -5842,6 +5849,14 @@ static bool do_ggml_backend_sycl_device_supports_op(ggml_backend_dev_t dev, cons
case GGML_OP_IM2COL_3D:
case GGML_OP_UPSCALE:
return true;
case GGML_OP_COL2IM_1D:
return ggml_is_contiguous(op->src[0]) &&
(op->type == GGML_TYPE_F32 || op->type == GGML_TYPE_F16
#ifdef GGML_SYCL_HAS_BF16
|| op->type == GGML_TYPE_BF16
#endif
) &&
op->src[0]->type == op->type;
case GGML_OP_CONV_3D:
return op->type == GGML_TYPE_F32 &&
(op->src[0]->type == GGML_TYPE_F32 || op->src[0]->type == GGML_TYPE_F16) &&