Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

sync : ggml (#28379)

Browse Source 
* ggml : rename and make private ggml_op_alloc_size_may_expand() (ggml/0)

cont https://github.com/ggml-org/llama.cpp/pull/27960

* ggml : bump version to 0.23.0 (ggml/1618)

* sync : ggml
This commit is contained in:
Georgi Gerganov
2026-09-04 14:39:19 +03:00
committed by GitHub
parent 163a40796f
commit 64a155d242
6 changed files with 10 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-backend-impl.h
+5
View File
@@ -34,6 +34,11 @@ extern "C" {
void * context;
};
// [TAG_ALLOC_SIZE_EXPAND]
// returns true for ops that may require additional memory for fleeting data on some backends,
// i.e. the backend buffer type's get_alloc_size may return more than ggml_nbytes for the output tensor
GGML_API bool ggml_op_alloc_size_may_expand(enum ggml_op op);
//
// Backend buffer
//
ggml/src/ggml-backend.cpp
+2 -5
View File
@@ -71,7 +71,7 @@ size_t ggml_backend_buft_get_alloc_size(ggml_backend_buffer_type_t buft, const s
GGML_ASSERT(size <= ggml_nbytes(tensor) ||
ggml_op_is_empty(tensor->op) ||
ggml_is_quantized(tensor->type) || // [TAG_ALLOC_SIZE_EXPAND]
ggml_backend_op_alloc_size_may_expand(tensor->op));
ggml_op_alloc_size_may_expand(tensor->op));
return size;
}
@@ -2109,10 +2109,7 @@ ggml_backend_t ggml_backend_sched_get_tensor_backend(ggml_backend_sched_t sched,
// utils
// [TAG_ALLOC_SIZE_EXPAND]
// returns true for ops that may require additional memory for fleeting data on some backends,
// i.e. the backend's get_alloc_size may return more than ggml_nbytes for the output tensor
bool ggml_backend_op_alloc_size_may_expand(enum ggml_op op) {
bool ggml_op_alloc_size_may_expand(enum ggml_op op) {
switch (op) {
case GGML_OP_FLASH_ATTN_EXT:
case GGML_OP_MUL_MAT:
ggml/src/ggml-rpc/ggml-rpc.cpp
+1 -1
View File
@@ -835,7 +835,7 @@ static size_t ggml_backend_rpc_buffer_type_get_alloc_size(ggml_backend_buffer_ty
// [TAG_ALLOC_SIZE_EXPAND]
// ops that may require additional memory for fleeting data on certain backends
// ref: https://github.com/ggml-org/llama.cpp/pull/15966
rpc_get |= ggml_backend_op_alloc_size_may_expand(tensor->op);
rpc_get |= ggml_op_alloc_size_may_expand(tensor->op);
if (rpc_get) {
ggml_backend_rpc_buffer_type_context * buft_ctx = (ggml_backend_rpc_buffer_type_context *)buft->context;