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-backend.cpp
View file

@ -682,8 +682,6 @@ ggml_backend_t ggml_backend_init_best(void) {
// backend CPU
static const size_t TENSOR_ALIGNMENT = 32; // required for mmap as gguf only guarantees 32-byte alignment
static const char * ggml_backend_cpu_buffer_get_name(ggml_backend_buffer_t buffer) {
return "CPU";
@ -702,7 +700,7 @@ static void * ggml_backend_cpu_buffer_get_base(ggml_backend_buffer_t buffer) {
}
static void ggml_backend_cpu_buffer_free_buffer(ggml_backend_buffer_t buffer) {
free(buffer->context);
ggml_aligned_free(buffer->context, buffer->size);
}
static void ggml_backend_cpu_buffer_memset_tensor(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor, uint8_t value, size_t offset, size_t size) {
@ -770,8 +768,8 @@ static const char * ggml_backend_cpu_buffer_type_get_name(ggml_backend_buffer_ty
}
static ggml_backend_buffer_t ggml_backend_cpu_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft, size_t size) {
size += TENSOR_ALIGNMENT; // malloc may return an address that is not aligned
void * data = malloc(size); // TODO: use GGML_ALIGNED_MALLOC (move to ggml-impl.h)
void * data = ggml_aligned_malloc(size);
if (data == NULL) {
GGML_LOG_ERROR("%s: failed to allocate buffer of size %zu\n", __func__, size);
return NULL;