Home Explore Help
Sign In
cturan
/
llama.cpp
mirror of https://github.com/cturan/llama.cpp
1
0
Fork 0
Files Issues 0 Wiki
Browse Source

Fix scalar version of Q5_K when QK_K = 64 (#2362)

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Kawrakow 2 years ago
parent
84e09a7d8b
commit
42f70cb2f6
1 changed files with 2 additions and 3 deletions
Unified View Show Diff Stats
  1. 2 3
      k_quants.c

+ 2 - 3
k_quants.c
View File 

@@ -3297,8 +3297,7 @@ void ggml_vec_dot_q5_K_q8_K(const int n, float * restrict s, const void * restri
 
 
 
 #else
 
 #else
 
 
 
-
 
-    uint8_t aux8[QK_K];
 
+    int8_t aux8[QK_K];
 
     int16_t aux16[16];
 
     int16_t aux16[16];
 
     float   sums [8];
 
     float   sums [8];
 
     memset(sums, 0, 8*sizeof(float));
 
     memset(sums, 0, 8*sizeof(float));
 
@@ -3308,7 +3307,7 @@ void ggml_vec_dot_q5_K_q8_K(const int n, float * restrict s, const void * restri
 
         const uint8_t * restrict q4 = x[i].qs;
 
         const uint8_t * restrict q4 = x[i].qs;
 
         const uint8_t * restrict hm = x[i].qh;
 
         const uint8_t * restrict hm = x[i].qh;
 
         const  int8_t * restrict q8 = y[i].qs;
 
         const  int8_t * restrict q8 = y[i].qs;
 
-        uint8_t * restrict a = aux8;
 
+        int8_t * restrict a = aux8;
 
         for (int l = 0; l < 32; ++l) {
 
         for (int l = 0; l < 32; ++l) {
 
             a[l+ 0] = q4[l] & 0xF;
 
             a[l+ 0] = q4[l] & 0xF;
 
             a[l+32] = q4[l]  >> 4;
 
             a[l+32] = q4[l]  >> 4;