CUDA: fix partial offloading for ne0 % 256 != 0 (#8572)

2024-07-18 23:48:47 +02:00 · 2024-07-18 23:48:47 +02:00 · a15ef8f8a0
commit a15ef8f8a0
parent 705b7ecf60
4 changed files with 29 additions and 15 deletions
--- a/ggml/src/ggml-alloc.c
+++ b/ggml/src/ggml-alloc.c
@ -776,6 +776,7 @@ bool ggml_gallocr_reserve_n(ggml_gallocr_t galloc, struct ggml_cgraph * graph, c
                fprintf(stderr, "%s: failed to allocate %s buffer of size %zu\n", __func__, ggml_backend_buft_name(galloc->bufts[i]), new_size);
                return false;
            }
+            ggml_backend_buffer_set_usage(galloc->buffers[i], GGML_BACKEND_BUFFER_USAGE_COMPUTE);
        }
    }