CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-30 08:42:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

sync : ggml (fix im2col) (#4591)

Browse Source 
* cuda : fix im2col_f32_f16 (ggml/#658)

ggml-ci

* ggml-alloc : fix ggml_tallocr_is_own

---------

Co-authored-by: leejet <leejet714@gmail.com>
This commit is contained in:
Georgi Gerganov
2023-12-22 17:53:43 +02:00
committed by GitHub
parent a55876955b
commit ba66175132
2 changed files with 5 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
ggml-alloc.c
View File

@@ -72,7 +72,7 @@ static void remove_allocated_tensor(ggml_tallocr_t alloc, struct ggml_tensor * t
// check if a tensor is allocated by this buffer
static bool ggml_tallocr_is_own(ggml_tallocr_t alloc, const struct ggml_tensor * tensor) {
return tensor->buffer == alloc->buffer;
return tensor->buffer == alloc->buffer && (!tensor->view_src || tensor->view_src->buffer == alloc->buffer);
}
static bool ggml_is_view(struct ggml_tensor * t) {

8
ggml-cuda.cu
View File

@@ -5273,17 +5273,17 @@ static __global__ void im2col_f32_f16(
const int ky = (i - kd) / OW;
const int ix = i % OW;
const int iiw = ix * s0 + kx * d0 - p0;
const int iih = blockIdx.y * s1 + ky * d1 - p1;
const int64_t iiw = ix * s0 + kx * d0 - p0;
const int64_t iih = blockIdx.y * s1 + ky * d1 - p1;
const int offset_dst =
const int64_t offset_dst =
(blockIdx.y * OW + ix) * CHW +
(blockIdx.z * (KW * KH) + ky * KW + kx);
if (iih < 0 || iih >= IH || iiw < 0 || iiw >= IW) {
dst[offset_dst] = __float2half(0.0f);
} else {
const int offset_src = blockIdx.z * offset_delta;
const int64_t offset_src = blockIdx.z * offset_delta;
dst[offset_dst] = __float2half(x[offset_src + iih * IW + iiw]);
}
}