small change

Author: JohannesGaessler

ggml-cuda.cu

@@ -1403,6 +1403,8 @@ static __device__ __forceinline__ float vec_dot_q2_K_q8_1(
 // #if __CUDA_ARCH__ >= 600 // lowest compute capability for integer intrinsics
     const block_q2_K * bq2_K = (const block_q2_K *) vbq;
 
+    const int bq8_offset = 4 * (iqs/8);
+
     float sumf = 0;
 
     const float d = bq2_K->d;
@@ -1417,7 +1419,7 @@ static __device__ __forceinline__ float vec_dot_q2_K_q8_1(
 
         const int vii = (vi >> (2*i)) & 0x03030303;
 
-        const block_q8_1 * bq8i = bq8_1 + 4 * (iqs/8) + i;
+        const block_q8_1 * bq8i = bq8_1 + bq8_offset + i;
         const float d8 = bq8i->d;
         const int qs8 = *((int*) &bq8i->qs[4*(iqs%8)]);