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

ggml: avoid KleidiAI buffer type init on dispatch (#27891)

Browse Source 
Co-authored-by: Acmmi <acmmi@Acmmis-MacBook-Air.local>
This commit is contained in:
Aman Chadha(IVIXMMI)
2026-09-02 09:16:15 +03:00
committed by GitHub
co-authored by Acmmi
parent 960dffab05
commit b81c99b479
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cpu/kleidiai/kleidiai.cpp
+2 -2
View File
@@ -1823,7 +1823,7 @@ class extra_buffer_type : ggml::cpu::extra_buffer_type {
const bool src0_is_kleidiai = const bool src0_is_kleidiai =
op->src[0]->buffer && op->src[0]->buffer &&
(ggml_n_dims(op->src[0]) == 2) && (ggml_n_dims(op->src[0]) == 2) &&
op->src[0]->buffer->buft == ggml_backend_cpu_kleidiai_buffer_type() && op->src[0]->buffer->buft->context == this &&
slot_total > 0; slot_total > 0;
if ((op->op == GGML_OP_MUL_MAT || op->op == GGML_OP_GET_ROWS) && if ((op->op == GGML_OP_MUL_MAT || op->op == GGML_OP_GET_ROWS) &&
@@ -1862,7 +1862,7 @@ class extra_buffer_type : ggml::cpu::extra_buffer_type {
ggml::cpu::tensor_traits * get_tensor_traits(const struct ggml_tensor * op) override { ggml::cpu::tensor_traits * get_tensor_traits(const struct ggml_tensor * op) override {
if (op->op == GGML_OP_MUL_MAT || op->op == GGML_OP_GET_ROWS) { if (op->op == GGML_OP_MUL_MAT || op->op == GGML_OP_GET_ROWS) {
if (op->src[0]->buffer && op->src[0]->buffer->buft == ggml_backend_cpu_kleidiai_buffer_type()) { if (op->src[0]->buffer && op->src[0]->buffer->buft->context == this) {
return (ggml::cpu::tensor_traits *) op->src[0]->extra; return (ggml::cpu::tensor_traits *) op->src[0]->extra;
} else { } else {
// KleidiAI only has kernels for Q4_0 and Q8_0. For a quantized weight of any // KleidiAI only has kernels for Q4_0 and Q8_0. For a quantized weight of any