build

Tencent · Sep 30, 2024 · 1e50e88 · 1e50e88
1 parent 87d5cd4
commit 1e50e88
Showing 1 changed file with 3 additions and 3 deletions.
diff --git a/src/layer/arm/gemm_int8_fp16s.h b/src/layer/arm/gemm_int8_fp16s.h
@@ -102,8 +102,8 @@ static void compute_A_tile_fp16_int8_scales(const Mat& A, Mat& scales, float B_s
                 _amax0 = vmaxq_f16(_amax0, vabsq_f16(_p));
                 p0 += 8;
             }
-            float32x4_t _absmax0 = vcvt_f32_f16(vget_low_f16(_absmax));
-            float32x4_t _absmax1 = vcvt_f32_f16(vget_high_f16(_absmax));
+            float32x4_t _absmax0 = vcvt_f32_f16(vget_low_f16(_amax0));
+            float32x4_t _absmax1 = vcvt_f32_f16(vget_high_f16(_amax0));
 #else  // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
             const unsigned short* p0 = (const unsigned short*)A + (i + ii) * A_hstep;
 
@@ -237,7 +237,7 @@ static void compute_A_tile_fp16_int8_scales(const Mat& A, Mat& scales, float B_s
                 _amax = vmax_f16(_amax, vabs_f16(_p));
                 p0 += 4;
             }
-            float32x4_t _absmax0 = vcvt_f32_f16(_absmax);
+            float32x4_t _absmax0 = vcvt_f32_f16(_amax);
 #else  // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
             const unsigned short* p0 = (const unsigned short*)A + (i + ii) * A_hstep;