Cambricon · PetrelYy · Jan 6, 2025 · Jan 2, 2025 · Jan 2, 2025 · Jan 3, 2025
diff --git a/kernels/utils/cnnl_helper.cpp → core/cnnl_helper.cpp b/kernels/utils/cnnl_helper.cpp → core/cnnl_helper.cpp
diff --git a/kernels/utils/cnnl_helper.h → core/cnnl_helper.h b/kernels/utils/cnnl_helper.h → core/cnnl_helper.h
diff --git a/docs/MLU-OPS-Accuracy-Acceptance-Standard.md b/docs/MLU-OPS-Accuracy-Acceptance-Standard.md
@@ -28,7 +28,7 @@ $$
 **$diff2$：标准相对误差**
 
 $$
-diff2 = \sqrt{\frac{ \sum (evaluated \\_ data - baseline \\_ data)^2}{\sum  baseline \\_ data^2 }}
+diff2 = \sqrt{\left(\frac{\sum (evaluated \\_ data - baseline \\_ data)^2}{\sum baseline \\_ data^2}\right)}
 $$
 
 **$diff3_1$：最大单点相对误差**

diff --git a/docs/MLU-OPS-How-To-Use-CNNL-API.md b/docs/MLU-OPS-How-To-Use-CNNL-API.md
@@ -5,7 +5,7 @@
 在算子中调用CNNL tranpose 算子。
 ```c++
 // 需要包含该文件
-#include "utils/cnnl_helper.h"
+#include "core/cnnl_helper.h"
 
 // https://github.com/Cambricon/mlu-ops/blob/master/kernels/three_nn_forward/three_nn_forward.cpp
 static mluOpStatus_t transposeTensor(
@@ -67,7 +67,7 @@ static mluOpStatus_t transposeTensor(
 在算子中调用CNNL addN 算子的GetWorkspaceSize, 与示例1不同的点在于，当输入的tensor是数组时的处理方法。
 ```c++
 // 需要包含该文件
-#include "utils/cnnl_helper.h"
+#include "core/cnnl_helper.h"
 
 mluOpStatus_t GetAddNWorkspaceSize(
     mluOpHandle_t handle, const mluOpTensorDescriptor_t input_descs[],
@@ -109,7 +109,7 @@ mluOpStatus_t GetAddNWorkspaceSize(
 ```c++
 // https://github.com/Cambricon/mlu-ops/blob/master/kernels/indice_convolution_backward_data/indice_convolution_backward_data.cpp
 // 需要包含该文件
-#include "utils/cnnl_helper.h"
+#include "core/cnnl_helper.h"
 
 mluOpStatus_t MLUOP_WIN_API mluOpGetIndiceConvolutionBackwardDataWorkspaceSize(
     mluOpHandle_t handle, const mluOpTensorDescriptor_t output_grad_desc,

diff --git a/kernels/border_align/border_align_backward/border_align_backward.cpp b/kernels/border_align/border_align_backward/border_align_backward.cpp
@@ -24,6 +24,7 @@
 
 #include <string>
 
+#include "core/cnnl_helper.h"
 #include "core/context.h"
 #include "core/gen_case.h"
 #include "core/logging.h"
@@ -32,7 +33,6 @@
 #include "core/type.h"
 #include "core/tool.h"
 #include "kernels/kernel.h"
-#include "kernels/utils/cnnl_helper.h"
 
 static void policyFunc(mluOpHandle_t handle, cnrtDim3_t *k_dim,
                        cnrtFunctionType_t *k_type) {

diff --git a/kernels/carafe/carafe.cpp b/kernels/carafe/carafe.cpp
@@ -25,6 +25,7 @@
 #include <algorithm>
 #include <vector>
 
+#include "core/cnnl_helper.h"
 #include "core/context.h"
 #include "core/gen_case.h"
 #include "core/logging.h"
@@ -34,7 +35,6 @@
 #include "kernels/kernel.h"
 #include "kernels/tensor_stride_process/tensor_stride_process_host.h"
 #include "kernels/tensor_stride_process/tensor_stride_process_mlu.h"
-#include "kernels/utils/cnnl_helper.h"
 
 // 1.creat set destroy
 mluOpStatus_t MLUOP_WIN_API

diff --git a/kernels/deform_conv/dcn_backward_data/dcn_backward_data.cpp b/kernels/deform_conv/dcn_backward_data/dcn_backward_data.cpp
@@ -24,7 +24,7 @@
 #include <math.h>
 #include <vector>
 
-#include "kernels/utils/cnnl_helper.h"
+#include "core/cnnl_helper.h"
 
 #define DCNBPDATA_API "mluOpDCNBackwardData"
 

diff --git a/kernels/deform_conv/dcn_backward_weight/dcn_backward_weight.cpp b/kernels/deform_conv/dcn_backward_weight/dcn_backward_weight.cpp
@@ -24,7 +24,7 @@
 #include <math.h>
 #include <vector>
 
-#include "kernels/utils/cnnl_helper.h"
+#include "core/cnnl_helper.h"
 
 #define DCNBACKWARDWEIGHT_API "mluOpDCNBackwardWeight"
 

diff --git a/kernels/deform_conv/dcn_common/dcn_common.cpp b/kernels/deform_conv/dcn_common/dcn_common.cpp
@@ -26,7 +26,7 @@
 #include <math.h>
 #include <vector>
 
-#include "kernels/utils/cnnl_helper.h"
+#include "core/cnnl_helper.h"
 
 #define DCN_API "mluOpDCN"
 

diff --git a/kernels/deform_conv/dcn_forward/dcn_forward.cpp b/kernels/deform_conv/dcn_forward/dcn_forward.cpp
@@ -20,7 +20,7 @@
  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  *************************************************************************/
-#include "kernels/utils/cnnl_helper.h"
+#include "core/cnnl_helper.h"
 
 #define DCNFORWARD_API "mluOpDCNForward"
 

diff --git a/kernels/deform_roi_pool/deform_roi_pool.cpp b/kernels/deform_roi_pool/deform_roi_pool.cpp
@@ -29,7 +29,7 @@
 #include "core/tensor.h"
 #include "core/type.h"
 #include "kernels/kernel.h"
-#include "kernels/utils/cnnl_helper.h"
+#include "core/cnnl_helper.h"
 
 // policy function
 void policyFunc(const mluOpHandle_t handle,

diff --git a/...ynamic_point_to_voxel/dynamic_point_to_voxel_backward/dynamic_point_to_voxel_backward.cpp b/...ynamic_point_to_voxel/dynamic_point_to_voxel_backward/dynamic_point_to_voxel_backward.cpp
@@ -25,13 +25,13 @@
 #include <algorithm>  // std::min
 #include <string>
 
+#include "core/cnnl_helper.h"
 #include "core/gen_case.h"
 #include "core/logging.h"
 #include "core/runtime/device.h"
 #include "core/tensor.h"
 #include "core/type.h"  // mluop::getSizeOfDataType
 #include "kernels/kernel.h"
-#include "kernels/utils/cnnl_helper.h"
 
 static mluOpStatus_t DynamicPointToVoxelBackwardParamCheck(
     const char *interface_name, const mluOpHandle_t handle,

diff --git a/.../dynamic_point_to_voxel/dynamic_point_to_voxel_forward/dynamic_point_to_voxel_forward.cpp b/.../dynamic_point_to_voxel/dynamic_point_to_voxel_forward/dynamic_point_to_voxel_forward.cpp
@@ -24,13 +24,13 @@
 
 #include <string>
 
+#include "core/cnnl_helper.h"
 #include "core/gen_case.h"
 #include "core/logging.h"
 #include "core/runtime/device.h"
 #include "core/tensor.h"
 #include "core/type.h"
 #include "kernels/kernel.h"
-#include "kernels/utils/cnnl_helper.h"
 
 // policy function
 static void policyFuncDynamicPointToVoxelForward(const mluOpHandle_t handle,

diff --git a/kernels/fft/common/fft_basic_ops.h b/kernels/fft/common/fft_basic_ops.h
@@ -26,11 +26,12 @@
 
 #include <algorithm>
 #include <string>
-#include "core/tensor.h"
+
+#include "core/cnnl_helper.h"
 #include "core/context.h"
+#include "core/tensor.h"
 #include "core/tool.h"
 #include "kernels/kernel.h"
-#include "kernels/utils/cnnl_helper.h"
 #include "mlu_op.h"
 
 bool fftIsIntDtype(const mluOpDataType_t dtype);

diff --git a/kernels/fft/common/fft_common_kernels.mlu b/kernels/fft/common/fft_common_kernels.mlu
@@ -150,7 +150,7 @@ __mlu_func__ void generateRFFTHalfDFTMatrixImpl(int n, void *output) {
   float *row_addr = temp_addr;
 
   // generate 0 to n indices
-  __mlu_op_gen_stage_index(inc_addr, deal_size, 0.0f, 1.0f);
+  __mluop_gen_stage_index(inc_addr, deal_size, 0.0f, 1.0f);
 
   // generate sin and cos vectors
   const float scale = -2.0 * M_PI / n;
@@ -234,7 +234,7 @@ __mlu_func__ void generateRFFTFullDFTMatrixImpl(int row, int n, void *output) {
   float *row_addr = temp_addr;
 
   // generate 0 to n indices
-  __mlu_op_gen_stage_index(inc_addr, deal_size, 0.0f, 1.0f);
+  __mluop_gen_stage_index(inc_addr, deal_size, 0.0f, 1.0f);
 
   // generate sin and cos vectors
   const float scale = -2.0 * M_PI / n;
@@ -323,7 +323,7 @@ __mlu_func__ void generateIRFFTHalfDFTMatrixImpl(int n, void *output) {
   float *row_addr = temp_addr;
 
   // generate 0 to n indices
-  __mlu_op_gen_stage_index(inc_addr, deal_size, 0.0f, 1.0f);
+  __mluop_gen_stage_index(inc_addr, deal_size, 0.0f, 1.0f);
 
   // generate sin and cos coefficient vectors
   __bang_write_value((float *)cos_coeff_addr, deal_size, (float)2.0);
@@ -418,7 +418,7 @@ __mlu_func__ void generateIRFFTFullDFTMatrixImpl(int n, void *output) {
   float *row_addr = temp_addr;
 
   // generate 0 to n indices
-  __mlu_op_gen_stage_index(inc_addr, deal_size, 0.0f, 1.0f);
+  __mluop_gen_stage_index(inc_addr, deal_size, 0.0f, 1.0f);
 
   // generate sin and cos vectors
   const float scale = 2.0 * M_PI / n;
@@ -514,7 +514,7 @@ __mlu_func__ void generateC2CFFTDFTMatrixImpl(int n, void *output) {
   float *row_addr = temp_addr;
 
   // generate 0 to n indices
-  __mlu_op_gen_stage_index(inc_addr, deal_size, 0.0f, 1.0f);
+  __mluop_gen_stage_index(inc_addr, deal_size, 0.0f, 1.0f);
 
   // generate sin and cos vectors
   const float forward_scale = -2.0 * M_PI / n;

diff --git a/kernels/generate_proposals_v2/generate_proposals_v2.cpp b/kernels/generate_proposals_v2/generate_proposals_v2.cpp
@@ -25,14 +25,14 @@
 #include <algorithm>
 #include <string>
 
+#include "core/cnnl_helper.h"
 #include "core/context.h"
 #include "core/gen_case.h"
 #include "core/logging.h"
 #include "core/runtime/device.h"
 #include "core/tensor.h"
 #include "core/type.h"
 #include "kernels/kernel.h"
-#include "kernels/utils/cnnl_helper.h"
 
 #define GDRAM_ALIGN_SIZE 128
 static void policyFunc(mluOpHandle_t handle, cnrtDim3_t *k_dim,

diff --git a/kernels/masked_im2col/masked_col2im_forward/masked_col2im_forward.cpp b/kernels/masked_im2col/masked_col2im_forward/masked_col2im_forward.cpp
@@ -22,14 +22,14 @@
  *************************************************************************/
 #include "masked_col2im_forward.h"
 
+#include "core/cnnl_helper.h"
 #include "core/context.h"
 #include "core/gen_case.h"
 #include "core/logging.h"
 #include "core/runtime/device.h"
 #include "core/tensor.h"
 #include "core/type.h"
 #include "kernels/kernel.h"
-#include "kernels/utils/cnnl_helper.h"
 
 static void policyFunc(const mluOpHandle_t handle, const int mask_cnt,
                        cnrtDim3_t *k_dim, cnrtFunctionType_t *k_type) {

diff --git a/kernels/masked_im2col/masked_im2col_forward/masked_im2col_forward.cpp b/kernels/masked_im2col/masked_im2col_forward/masked_im2col_forward.cpp
@@ -22,11 +22,11 @@
  *************************************************************************/
 #include "kernels/masked_im2col/masked_im2col_forward/masked_im2col_forward.h"
 
+#include "core/cnnl_helper.h"
 #include "core/gen_case.h"
 #include "core/logging.h"
 #include "core/runtime/device.h"
 #include "kernels/kernel.h"
-#include "kernels/utils/cnnl_helper.h"
 
 // policy function
 static void policyFunc(const mluOpHandle_t handle, const int mask_cnt,

diff --git a/kernels/moe_dispatch/moe_dispatch_backward_data/moe_dispatch_backward_data.cpp b/kernels/moe_dispatch/moe_dispatch_backward_data/moe_dispatch_backward_data.cpp
@@ -24,13 +24,13 @@
 
 #include <string>
 
+#include "core/cnnl_helper.h"
 #include "core/context.h"
 #include "core/gen_case.h"
 #include "core/logging.h"
 #include "core/runtime/device.h"
 #include "core/tensor.h"
 #include "core/type.h"
-#include "kernels/utils/cnnl_helper.h"
 
 // policy function
 static void PolicyFunc(const mluOpHandle_t handle, cnrtDim3_t *k_dim,

diff --git a/kernels/moe_dispatch/moe_dispatch_backward_gate/moe_dispatch_backward_gate.cpp b/kernels/moe_dispatch/moe_dispatch_backward_gate/moe_dispatch_backward_gate.cpp
@@ -24,14 +24,14 @@
 
 #include <string>
 
+#include "core/cnnl_helper.h"
 #include "core/context.h"
 #include "core/gen_case.h"
 #include "core/logging.h"
 #include "core/runtime/device.h"
 #include "core/tensor.h"
 #include "core/type.h"
 #include "kernels/kernel.h"
-#include "kernels/utils/cnnl_helper.h"
 
 static void policyFunc(const mluOpHandle_t handle, const int samples,
                        cnrtDim3_t *k_dim, cnrtFunctionType_t *k_type) {

diff --git a/kernels/ms_deform_attn/ms_deform_attn_backward/ms_deform_attn_backward.cpp b/kernels/ms_deform_attn/ms_deform_attn_backward/ms_deform_attn_backward.cpp
@@ -24,6 +24,7 @@
 
 #include <string>
 
+#include "core/cnnl_helper.h"
 #include "core/context.h"
 #include "core/gen_case.h"
 #include "core/logging.h"
@@ -33,7 +34,6 @@
 #include "core/type.h"
 #include "kernels/debug.h"
 #include "kernels/kernel.h"
-#include "kernels/utils/cnnl_helper.h"
 
 char API[] = "[mluOpMsDeformAttnBackward]";
 

diff --git a/kernels/ms_deform_attn/ms_deform_attn_forward/ms_deform_attn_forward.mlu b/kernels/ms_deform_attn/ms_deform_attn_forward/ms_deform_attn_forward.mlu
@@ -22,6 +22,7 @@
  *************************************************************************/
 #include "kernels/ms_deform_attn/ms_deform_attn_forward/ms_deform_attn_forward.h"
 
+#include "core/cnnl_helper.h"
 #include "core/context.h"
 #include "core/logging.h"
 #include "core/gen_case.h"
@@ -31,7 +32,6 @@
 #include "core/type.h"
 #include "kernels/debug.h"
 #include "kernels/kernel.h"
-#include "kernels/utils/cnnl_helper.h"
 
 typedef enum {
   /*!< Index is invalid. */

diff --git a/kernels/mutual_information/mutual_information_backward/mutual_information_backward.cpp b/kernels/mutual_information/mutual_information_backward/mutual_information_backward.cpp
@@ -25,13 +25,13 @@
 #include <algorithm>
 #include <string>
 
+#include "core/cnnl_helper.h"
 #include "core/context.h"
 #include "core/gen_case.h"
 #include "core/logging.h"
 #include "core/runtime/device.h"
 #include "core/tensor.h"
 #include "core/type.h"
-#include "kernels/utils/cnnl_helper.h"
 
 #define API_NAME "[mluOpMutualInformationBackward]"
 

diff --git a/kernels/nms/nms.cpp b/kernels/nms/nms.cpp
@@ -20,7 +20,7 @@
  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  *************************************************************************/
-#include "kernels/utils/cnnl_helper.h"
+#include "core/cnnl_helper.h"
 
 mluOpStatus_t MLUOP_WIN_API
 mluOpCreateNmsDescriptor(mluOpNmsDescriptor_t *desc) {

diff --git a/kernels/psroipool/psroipool.cpp b/kernels/psroipool/psroipool.cpp
@@ -24,14 +24,14 @@
 
 #include <string>
 
+#include "core/cnnl_helper.h"
 #include "core/context.h"
 #include "core/gen_case.h"
 #include "core/logging.h"
 #include "core/runtime/device.h"
 #include "core/tensor.h"
 #include "core/type.h"
 #include "kernels/kernel.h"
-#include "kernels/utils/cnnl_helper.h"
 
 // policy function
 static void policyFuncPsRoiPool(const mluOpHandle_t handle, cnrtDim3_t *k_dim,

diff --git a/kernels/roi_align/roi_align_backward/roi_align_backward.cpp b/kernels/roi_align/roi_align_backward/roi_align_backward.cpp
@@ -20,7 +20,7 @@
  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  *************************************************************************/
-#include "kernels/utils/cnnl_helper.h"
+#include "core/cnnl_helper.h"
 
 mluOpStatus_t MLUOP_WIN_API mluOpRoiAlignBackward(
     mluOpHandle_t handle, const float spatial_scale, const int sampling_ratio,

diff --git a/kernels/roi_align/roialign_forward/roialign_forward.cpp b/kernels/roi_align/roialign_forward/roialign_forward.cpp
@@ -20,7 +20,7 @@
  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  *************************************************************************/
-#include "kernels/utils/cnnl_helper.h"
+#include "core/cnnl_helper.h"
 
 mluOpStatus_t MLUOP_WIN_API
 mluOpCreateRoiAlignForwardDescriptor(mluOpRoiAlignForwardDescriptor_t *desc) {

diff --git a/kernels/roi_align_rotated/roi_align_rotated.cpp b/kernels/roi_align_rotated/roi_align_rotated.cpp
@@ -24,13 +24,13 @@
 
 #include <string>
 
+#include "core/cnnl_helper.h"
 #include "core/context.h"
 #include "core/gen_case.h"
 #include "core/logging.h"
 #include "core/runtime/device.h"
 #include "core/tensor.h"
 #include "core/type.h"
-#include "kernels/utils/cnnl_helper.h"
 
 static void policyFunc(const mluOpHandle_t handle, const int bin_num,
                        cnrtDim3_t *k_dim, cnrtFunctionType_t *k_type) {

diff --git a/kernels/roi_align_rotated/roi_align_rotated_forward_vector.mlu b/kernels/roi_align_rotated/roi_align_rotated_forward_vector.mlu
@@ -348,7 +348,7 @@ __mlu_global__ void roiAlignRotatedForward(
     bin_order_num = 16;
   }
 
-  __mlu_op_gen_stage_index(order, bin_order_num, (T)0.5);
+  __mluop_gen_stage_index(order, bin_order_num, (T)0.5);
   const uint32_t bin_hw_order_num = bin_order_num * bin_order_num;
 
   const uint32_t cache_channels_num = 1024;