ggml : fix scalar implementation of Q4_1 dot

ggerganov · ggerganov · commit 4fd3e29297e3 · 2023-05-20T10:13:19.000+03:00
diff --git a/ggml.c b/ggml.c
@@ -2481,7 +2481,7 @@ static void ggml_vec_dot_q4_1_q8_1(const int n, float * restrict s, const void *
             sumi += (v0 * y[i].qs[j]) + (v1 * y[i].qs[j + qk/2]);
         }
 
-        sumf += (GGML_FP16_TO_FP32(x[i]).d*y[i].d)*sumi + GGML_FP16_TO_FP32(x[i].m)*y[i].s;
+        sumf += (GGML_FP16_TO_FP32(x[i].d)*y[i].d)*sumi + GGML_FP16_TO_FP32(x[i].m)*y[i].s;
     }
 
     *s = sumf;

Original file line number	Diff line number	Diff line change
`@@ -2481,7 +2481,7 @@ static void ggml_vec_dot_q4_1_q8_1(const int n, float * restrict s, const void *`
`2481`	`2481`	`sumi += (v0 * y[i].qs[j]) + (v1 * y[i].qs[j + qk/2]);`
`2482`	`2482`	`}`
`2483`	`2483`
`2484`		`- sumf += (GGML_FP16_TO_FP32(x[i]).dy[i].d)sumi + GGML_FP16_TO_FP32(x[i].m)*y[i].s;`
	`2484`	`+ sumf += (GGML_FP16_TO_FP32(x[i].d)y[i].d)sumi + GGML_FP16_TO_FP32(x[i].m)*y[i].s;`
`2485`	`2485`	`}`
`2486`	`2486`
`2487`	`2487`	`*s = sumf;`