metal : improve F32, F16 and BF16 mat-vec multiplication (#16057)

* metal : improve F32, F16 and BF16 mat-vec multiplication ggml-ci * metal : make the NSG a function constant in mul_mv kernels ggml-ci
2025-09-18 12:33:45 +03:00
parent e00f3fd8ff
commit b213fce89b
6 changed files with 355 additions and 288 deletions
--- a/ggml/src/ggml-metal/ggml-metal-impl.h
+++ b/ggml/src/ggml-metal/ggml-metal-impl.h
@@ -8,6 +8,9 @@
 //
 // TODO: for optimal performance, become function of the device and work size

+#define N_R0_F 2
+#define N_SG_F 4
+
 #define N_R0_Q4_0 4
 #define N_SG_Q4_0 2

@@ -72,6 +75,7 @@
 #define FC_FLASH_ATTN_EXT              100
 #define FC_FLASH_ATTN_EXT_VEC          200
 #define FC_FLASH_ATTN_EXT_VEC_REDUCE   300
+#define FC_MUL_MV                      400

 // kernel argument structs
 //