fix

Tencent · Sep 26, 2024 · 09af847 · 09af847
1 parent dcd0636
commit 09af847
Show file tree

Hide file tree

Showing 2 changed files with 8 additions and 19 deletions.
diff --git a/src/layer/arm/gemm_arm.cpp b/src/layer/arm/gemm_arm.cpp
@@ -58,8 +58,6 @@ void pack_A_tile(const Mat& A, Mat& AT, int i, int max_ii, int k, int max_kk)
     const int elempack = A.elempack;
     const int A_hstep = A.dims == 3 ? (int)A.cstep : A.w;
 
-    NCNN_LOGE("pack_A_tile %d %d %d %d  %d %d", i, max_ii, k, max_kk, elempack, A_hstep);
-
     float* pp = AT;
 
     int ii = 0;
@@ -6037,15 +6035,6 @@ int Gemm_arm::create_pipeline_int8(const Option& opt)
         }
 #endif
 
-#if __ARM_NEON
-        if (constant_broadcast_type_C == 3 && opt.use_packing_layout && CT_data.h % 4 == 0)
-        {
-            Mat C2;
-            ncnn::convert_packing(CT_data, C2, 4, opt);
-            CT_data = C2;
-        }
-#endif
-
         if (opt.lightmode)
             C_data.release();
     }

diff --git a/src/layer/arm/gemm_int8_bf16s.h b/src/layer/arm/gemm_int8_bf16s.h
@@ -4923,15 +4923,15 @@ static void unpack_output_tile_int32_to_bf16(const Mat& topT, const Mat& C, Mat&
                             _c01 = vsetq_lane_u16(pC[c_hstep], _c01, 1);
                             _c01 = vsetq_lane_u16(pC[c_hstep * 2], _c01, 2);
                             _c01 = vsetq_lane_u16(pC[c_hstep * 3], _c01, 3);
-                            _c01 = vsetq_lane_u16(pC[c_hstep * 4], _c01, 4);
-                            _c01 = vsetq_lane_u16(pC[c_hstep * 5], _c01, 5);
-                            _c01 = vsetq_lane_u16(pC[c_hstep * 6], _c01, 6);
-                            _c01 = vsetq_lane_u16(pC[c_hstep * 7], _c01, 7);
+                            _c01 = vsetq_lane_u16(pC[1], _c01, 4);
+                            _c01 = vsetq_lane_u16(pC[c_hstep + 1], _c01, 5);
+                            _c01 = vsetq_lane_u16(pC[c_hstep * 2 + 1], _c01, 6);
+                            _c01 = vsetq_lane_u16(pC[c_hstep * 3 + 1], _c01, 7);
                             _c23 = uint16x8_t();
-                            _c23 = vsetq_lane_u16(pC[1], _c23, 0);
-                            _c23 = vsetq_lane_u16(pC[c_hstep + 1], _c23, 1);
-                            _c23 = vsetq_lane_u16(pC[c_hstep * 2 + 1], _c23, 2);
-                            _c23 = vsetq_lane_u16(pC[c_hstep * 3 + 1], _c23, 3);
+                            _c23 = vsetq_lane_u16(pC[c_hstep * 4], _c23, 0);
+                            _c23 = vsetq_lane_u16(pC[c_hstep * 5], _c23, 1);
+                            _c23 = vsetq_lane_u16(pC[c_hstep * 6], _c23, 2);
+                            _c23 = vsetq_lane_u16(pC[c_hstep * 7], _c23, 3);
                             _c23 = vsetq_lane_u16(pC[c_hstep * 4 + 1], _c23, 4);
                             _c23 = vsetq_lane_u16(pC[c_hstep * 5 + 1], _c23, 5);
                             _c23 = vsetq_lane_u16(pC[c_hstep * 6 + 1], _c23, 6);