Make new mesaurements for correct bytes calculation

include/ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp

@@ -240,7 +240,7 @@ struct DeviceImageToColumnImpl
 
             const auto block_2_tile_map =
                 BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, KPerBlock, OutputGridDesc>(
-                    arg.out_grid_desc_m_k_, I1 /*M01*/);
+                    arg.out_grid_desc_m_k_);
             const index_t grid_size = block_2_tile_map.CalculateGridSize(arg.out_grid_desc_m_k_);
             const auto kernel       = kernel_image_to_column<InputGridDesc,
                                                        InputDataType,

profiler/README.md

@@ -212,14 +212,13 @@ Note: This kernel use atomic add, this will cause output buffer to be accumulate
 
  ```
 
-Result (MI250, FP32, NHWC)
+Result (MI210, FP32, NHWC)
 ```
 input: dim 5, lengths {1, 256, 512, 28, 28}, strides {102760448, 401408, 1, 14336, 512}
 output: dim 2, lengths {173056, 4608}, strides {4608, 1}
 ....
 Best configuration parameters:
-name: DeviceImageToColumn<256, 64, 64, 4>
-avg_time: 3.19792
-tflops: 0
-GB/s: 1125.99
+name: DeviceImageToColumn<128, 32, 64, 4>
+avg_time: 3.12326
+GB/s: 2042.59
 ```