LLM inference in C/C++
C++ ggml
ggml: avoid KleidiAI buffer type init on dispatch (#27891)
Aman Chadha(IVIXMMI)b81c99b479d4
diff --git a/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp b/ggml/src/ggml-cpu/kleidiai/kleidiai.cppindex 92d7fd644f..dbd1987807 100644--- a/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp+++ b/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp@@ -1823,7 +1823,7 @@ class extra_buffer_type : ggml::cpu::extra_buffer_type {const bool src0_is_kleidiai =op->src[0]->buffer &&(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;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 {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;} else {// KleidiAI only has kernels for Q4_0 and Q8_0. For a quantized weight of any