[Feature](mlu-ops): Use __bang_div (#1292)

Author: WhiteNightTSMD

kernels/diff_iou_rotated_sort_vertices_forward/diff_iou_rotated_sort_vertices_forward_block.mlu

@@ -54,11 +54,7 @@ static __mlu_func__ void load(const T *addr_vertices, const bool *addr_mask,
 template <typename T>
 static __mlu_func__ void computeDiv(T *nram_dst, T *nram_src0, T *nram_src1,
                                     int deal_num) {
-#if (__BANG_ARCH__ != 372)
   __bang_div(nram_dst, nram_src0, nram_src1, deal_num);
-#else
-  __cn_vector_div_f32_rn(deal_num, nram_dst, nram_src0, nram_src1);
-#endif
 }
 
 template <typename T>

kernels/fft/common/chirpz_aux_block.mlu

@@ -73,7 +73,7 @@ __mlu_global__ void MLUKernelChirpz(const int length, int n, int pad_n,
     __bang_square(nram_index, nram_index, num_deal);
     __bang_mul_scalar(nram_index, nram_index, M_PI, num_deal);
     __bang_write_value(nram_temp, num_deal, (float)n);
-    __cn_vector_div_f32_rn(num_deal, nram_temp, nram_index, nram_temp);
+    __bang_div(nram_temp, nram_index, nram_temp, num_deal);
     __bang_move(nram_index, nram_temp, num_deal * sizeof(float));
     __cn_vector_cos_f32(num_deal, nram_index, nram_index);
     __cn_vector_sin_f32(num_deal, nram_temp, nram_temp);
@@ -137,7 +137,7 @@ __mlu_global__ void MLUKernelSignal(const int length, int n, int pad_n,
     __bang_square(nram_index, nram_index, 2 * num_deal);
     __bang_mul_scalar(nram_index, nram_index, M_PI, 2 * num_deal);
     __bang_write_value(nram_temp, 2 * num_deal, (float)n);
-    __cn_vector_div_f32_rn(2 * num_deal, nram_temp, nram_index, nram_temp);
+    __bang_div(nram_temp, nram_index, nram_temp, 2 * num_deal);
     __bang_move(nram_index, nram_temp, 2 * num_deal * sizeof(float));
     __cn_vector_cos_f32(2 * num_deal, nram_index, nram_index);
     __cn_vector_sin_f32(2 * num_deal, nram_temp, nram_temp);

kernels/sparse_conv/get_indice_pairs/get_indice_pairs_utils.h

@@ -294,25 +294,23 @@ __mlu_func__ void genIndiceOutLast(int32_t *nram_output, int32_t *nram_input,
           o_w = output_space.o_w;
   int32_t o_hw = o_h * o_w, o_dhw = o_d * o_h * o_w;
   __bang_write_value((int32_t *)nram_aux + 4 * deal_num, deal_num, int(o_dhw));
-  __cn_vector_div_s32(deal_num, (int32_t *)nram_aux, (int32_t *)nram_input,
-                      (int32_t *)nram_aux + 4 * deal_num);
+  __bang_div((int32_t *)nram_aux, (int32_t *)nram_input,
+             (int32_t *)nram_aux + 4 * deal_num, deal_num);
   __bang_mul_scalar((int32_t *)nram_output, (int32_t *)nram_aux, (int)o_dhw,
                     deal_num);
   __bang_sub((int32_t *)nram_input, (int32_t *)nram_input, (int *)nram_output,
              deal_num);
   __bang_write_value((int32_t *)nram_aux + 4 * deal_num, deal_num, int(o_hw));
-  __cn_vector_div_s32(deal_num, (int32_t *)nram_aux + deal_num,
-                      (int32_t *)nram_input,
-                      (int32_t *)nram_aux + 4 * deal_num);
+  __bang_div((int32_t *)nram_aux + deal_num, (int32_t *)nram_input,
+             (int32_t *)nram_aux + 4 * deal_num, deal_num);
   __bang_mul_scalar((int32_t *)nram_output, (int32_t *)nram_aux + deal_num,
                     (int)o_hw, deal_num);
   __bang_sub((int32_t *)nram_input, (int32_t *)nram_input,
              (int32_t *)nram_output, deal_num);
 
   __bang_write_value((int32_t *)nram_aux + 4 * deal_num, deal_num, int(o_w));
-  __cn_vector_div_s32(deal_num, (int32_t *)nram_aux + 2 * deal_num,
-                      (int32_t *)nram_input,
-                      (int32_t *)nram_aux + 4 * deal_num);
+  __bang_div((int32_t *)nram_aux + 2 * deal_num, (int32_t *)nram_input,
+             (int32_t *)nram_aux + 4 * deal_num, deal_num);
   __bang_mul_scalar((int32_t *)nram_output, (int32_t *)nram_aux + 2 * deal_num,
                     (int)o_w, deal_num);
   __bang_sub((int32_t *)nram_aux + 3 * deal_num, (int32_t *)nram_input,