Superminaren/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-09-24 13:37:01 +02:00
Code Issues Packages Projects Releases Wiki Activity

Move macros and ggml_cpu_iqp_mul_mat_id_min_batch definition

Browse Source
This commit is contained in:
Colin Kealty
2026-08-30 08:40:26 -04:00
parent 152473eb9a
commit 31868838dc
2 changed files with 12 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cpu/iqp.cpp
+10
View File
@@ -16,6 +16,16 @@
#define UNUSED GGML_UNUSED
// smallest src1 batch for which the decode pays for itself
#define GGML_IQP_MIN_BATCH 8
// same, per expert, for MUL_MAT_ID
#define GGML_IQP_MIN_BATCH_ID 8
bool ggml_cpu_iqp_mul_mat_id_min_batch(int64_t cne1) {
return cne1 >= GGML_IQP_MIN_BATCH_ID;
}
// src0 rows interleaved per panel
#define IQP_NB_ROWS 8
ggml/src/ggml-cpu/iqp.h
+2 -9
View File
@@ -12,15 +12,8 @@
extern "C" {
#endif
// smallest src1 batch for which the decode pays for itself
#define GGML_IQP_MIN_BATCH 8
// same, per expert, for MUL_MAT_ID
#define GGML_IQP_MIN_BATCH_ID 8
static inline bool ggml_cpu_iqp_mul_mat_id_min_batch(int64_t cne1) {
return cne1 >= GGML_IQP_MIN_BATCH_ID;
}
// whether cne1 rows of src1 are enough for the decode to pay for itself, per expert, for MUL_MAT_ID
bool ggml_cpu_iqp_mul_mat_id_min_batch(int64_t cne1);
bool ggml_cpu_iqp_supports_mul_mat(const struct ggml_tensor * dst);