ggml-cpu: add AVX2 optimization for nvfp4 dot product and use UE4M3 LUT (#23961)

2026-07-01 15:31:20 +08:00
parent 9d88e7cedd
commit 6dbc1174b8
5 changed files with 164 additions and 7 deletions
@@ -1111,11 +1111,12 @@ GGML_TABLE_BEGIN(int8_t, kvalues_iq4nl, 16)
    -127, -104, -83, -65, -49, -35, -22, -10, 1, 13, 25, 38, 53, 69, 89, 113,
 GGML_TABLE_END()

-// e2m1 values (doubled)
+// e2m1 values (doubled), shared by MXFP4 and NVFP4
 // ref: https://www.opencompute.org/documents/ocp-microscaling-formats-mx-v1-0-spec-final-pdf
-GGML_TABLE_BEGIN(int8_t, kvalues_mxfp4, 16)
+GGML_TABLE_BEGIN(int8_t, kvalues_fp4, 16)
    0, 1, 2, 3, 4, 6, 8, 12, 0, -1, -2, -3, -4, -6, -8, -12,
 GGML_TABLE_END()
+#define kvalues_mxfp4 kvalues_fp4

 #define NGRID_IQ1S 2048
 #define IQ1S_DELTA 0.125f