vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

fix: use vm_allocate to allocate CPU backend buffer on macOS (#9875)

Browse source 
* fix: use `vm_allocate` to allocate CPU backend buffer on macOS

* fix: switch to `posix_memalign` to keep existing `free()` usages work

* feat: move `GGML_ALIGNED_MALLOC` to `ggml-backend-impl.h`, add support for `vm_allocate` on macOS

* style: formatting

* fix: move const outside of `#ifndef`

* style: formatting

* fix: unused var

* fix: transform `GGML_ALIGNED_MALLOC` and `GGML_ALIGNED_FREE` into functions and add them to `ggml-impl.h`

* fix: unused var

* fix: page align to `GGUF_DEFAULT_ALIGNMENT`

* fix: page align to `TENSOR_ALIGNMENT`

* fix: convert `TENSOR_ALIGNMENT` to a macro

* fix: increase page size to `32` on iOS

* fix: iOS page size

* fix: `hbw_posix_memalign` alignment
This commit is contained in:
Gilad S. 2024-10-17 01:36:51 +03:00 • committed by GitHub
parent 9e04102448
commit 73afe681aa
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 63 additions and 27 deletions
Download patch file Download diff file Expand all files Collapse all files

8
ggml/src/ggml-impl.h
View file

@ -19,6 +19,9 @@ extern "C" {
#define MIN(a, b) ((a) < (b) ? (a) : (b))
#define MAX(a, b) ((a) > (b) ? (a) : (b))
// required for mmap as gguf only guarantees 32-byte alignment
#define TENSOR_ALIGNMENT 32
// static_assert should be a #define, but if it's not,
// fall back to the _Static_assert C11 keyword.
// if C99 - static_assert is noop
@ -196,6 +199,11 @@ struct ggml_cgraph {
struct ggml_cgraph ggml_graph_view(struct ggml_cgraph * cgraph, int i0, int i1);
// Memory allocation
void * ggml_aligned_malloc(size_t size);
void ggml_aligned_free(void * ptr, size_t size);
#ifdef __cplusplus
}
#endif