tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-28 12:25:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

vulkan: use smaller combined allocations to avoid fragmentation (#11551)

Browse Source
This commit is contained in:
Jeff Bolz
2025-02-06 00:02:18 -06:00
committed by GitHub
parent 902368a06b
commit 1b598b3058
2 changed files with 18 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
ggml/src/ggml-alloc.c
View File

@ -989,19 +989,7 @@ ggml_backend_buffer_t ggml_backend_alloc_ctx_tensors_from_buft(struct ggml_conte
this_size = GGML_PAD(ggml_backend_buft_get_alloc_size(buft, t), alignment);
}
if (this_size > max_size) {
GGML_LOG_ERROR("%s: tensor %s is too large to fit in a %s buffer (tensor size: %zu, max buffer size: %zu)\n",
__func__, t->name,
ggml_backend_buft_name(buft),
this_size, max_size);
for (size_t i = 0; i < n_buffers; i++) {
ggml_backend_buffer_free(buffers[i]);
}
free(buffers);
return NULL;
}
if ((cur_buf_size + this_size) > max_size) {
if (cur_buf_size > 0 && (cur_buf_size + this_size) > max_size) {
// allocate tensors in the current buffer
if (!alloc_tensor_range(ctx, first, t, buft, cur_buf_size, &buffers, &n_buffers)) {
return NULL;