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

ggml-backend : reduce alignment to 32 to match gguf and fix mmap

Browse source
This commit is contained in:
slaren 2024-02-11 21:04:03 +01:00
parent 7d404b383c
commit 7cba240bcf
1 changed files with 2 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

3
ggml-backend.c
View file

@ -475,7 +475,7 @@ ggml_backend_buffer_t ggml_backend_reg_alloc_buffer(size_t i, size_t size) {
// backend CPU
static const size_t TENSOR_ALIGNMENT = 64; // should be enough for AVX 512
static const size_t TENSOR_ALIGNMENT = 32; // required for mmap as gguf only guarantees 32-byte alignment
GGML_CALL static const char * ggml_backend_cpu_buffer_name(ggml_backend_buffer_t buffer) {
return "CPU";
@ -818,6 +818,7 @@ void ggml_backend_cpu_set_abort_callback(ggml_backend_t backend_cpu, ggml_abort_
}
GGML_CALL ggml_backend_buffer_t ggml_backend_cpu_buffer_from_ptr(void * ptr, size_t size) {
GGML_ASSERT((uintptr_t)ptr % TENSOR_ALIGNMENT == 0 && "buffer pointer must be aligned");
return ggml_backend_buffer_init(ggml_backend_cpu_buffer_type(), cpu_backend_buffer_i_from_ptr, ptr, size);
}