[ARM] fix fp16 transpose (#9272)

mjp9527 · web-flow · commit fb86945e9954 · 2022-08-18T14:18:42.000+08:00
diff --git a/lite/kernels/arm/transpose_compute.cc b/lite/kernels/arm/transpose_compute.cc
@@ -414,6 +414,7 @@ void transpose_mat(const lite_api::float16_t* din,
 #endif
         ptr_din_row += 4;
       }
+      data_out_ptr0 = ptr_out + size_wh + 4 * height * remain_ww;
       for (int w = 0; w < remain_ww_rem; w++) {
         INIT_PTR_4(lite_api::float16_t, data_out_ptr0, height)
         INIT_PTR_A4(lite_api::float16_t)

Original file line number	Diff line number	Diff line change
`@@ -414,6 +414,7 @@ void transpose_mat(const lite_api::float16_t* din,`
`414`	`414`	`#endif`
`415`	`415`	`ptr_din_row += 4;`
`416`	`416`	`}`
	`417`	`+ data_out_ptr0 = ptr_out + size_wh + 4 * height * remain_ww;`
`417`	`418`	`for (int w = 0; w < remain_ww_rem; w++) {`
`418`	`419`	`INIT_PTR_4(lite_api::float16_t, data_out_ptr0, height)`
`419`	`420`	`INIT_PTR_A4(lite_api::float16_t)`