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

sycl: Add more debug prints (#13640)

Browse Source
This commit is contained in:
Romain Biessy
2025-05-26 10:28:53 +02:00
committed by GitHub
parent fef693dc6b
commit 9012eb9b45
16 changed files with 245 additions and 159 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
ggml/src/ggml-sycl/cpy.cpp
View File

@@ -616,6 +616,9 @@ static void ggml_cpy_i32_i32_sycl(const char * cx, char * cdst, const int ne, co
}
void ggml_sycl_cpy(ggml_backend_sycl_context & ctx, const ggml_tensor * src0, const ggml_tensor * src1) try {
// Unlike other operators ggml_sycl_cpy takes 2 distinct tensors instead of a dst ggml_tensor and rely on its src field
scope_op_debug_print scope_dbg_print(__func__, src1, /*num_src=*/0,
std::string(" src0 type=") + ggml_type_name(src0->type));
const int64_t ne = ggml_nelements(src0);
GGML_ASSERT(ne == ggml_nelements(src1));
@@ -629,8 +632,6 @@ void ggml_sycl_cpy(ggml_backend_sycl_context & ctx, const ggml_tensor * src0, co
char * src0_ddc = (char *) src0->data;
char * src1_ddc = (char *) src1->data;
GGML_SYCL_DEBUG("[SYCL] %s: Tensor supplied: %s to %s\n", __func__, ggml_type_name(src0->type),
ggml_type_name(src1->type));
if (src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F32) {
ggml_cpy_f32_f32_sycl(src0_ddc, src1_ddc, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10,
@@ -694,8 +695,6 @@ void ggml_sycl_cpy(ggml_backend_sycl_context & ctx, const ggml_tensor * src0, co
}
void ggml_sycl_dup(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
// TODO: why do we pass dst as src1 here?
GGML_SYCL_DEBUG("[SYCL] call %s\n", __func__);
scope_op_debug_print scope_dbg_print(__func__, dst, /*num_src=*/1);
ggml_sycl_cpy(ctx, dst->src[0], dst);
GGML_SYCL_DEBUG("[SYCL] call %s done\n", __func__);
}