Re-organize layout_transform_ops (pytorch#2133)

Summary:

- Migrate layout_transform_ops into its own directory

Reviewed By: spcyppt

Differential Revision: D51290598

Author: Benson Ma

fbgemm_gpu/CMakeLists.txt

@@ -568,8 +568,8 @@ set(fbgemm_gpu_sources_static_cpu
     src/jagged_tensor_ops/jagged_tensor_ops_autograd.cpp
     src/jagged_tensor_ops/jagged_tensor_ops_meta.cpp
     src/jagged_tensor_ops/jagged_tensor_ops_cpu.cpp
-    src/input_combine_cpu.cpp
-    src/layout_transform_ops_cpu.cpp
+    src/input_combine_ops/input_combine_cpu.cpp
+    src/layout_transform_ops/layout_transform_ops_cpu.cpp
     src/quantize_ops/quantize_ops_cpu.cpp
     src/quantize_ops/quantize_ops_meta.cpp
     src/sparse_ops/sparse_ops_cpu.cpp
@@ -589,16 +589,16 @@ if(NOT FBGEMM_CPU_ONLY)
     codegen/embedding_bounds_check_host.cpp
     src/memory_utils/memory_utils.cpp
     src/memory_utils/memory_utils_ops.cpp
-    src/layout_transform_ops_gpu.cpp
+    src/layout_transform_ops/layout_transform_ops_gpu.cpp
     src/permute_pooled_embedding_ops/permute_pooled_embedding_ops_gpu.cpp
     src/permute_pooled_embedding_ops/permute_pooled_embedding_ops_split_gpu.cpp
     src/quantize_ops/quantize_ops_gpu.cpp
     src/sparse_ops/sparse_ops_gpu.cpp
     src/split_embeddings_utils.cpp
     src/split_embeddings_cache/split_embeddings_cache_ops.cu
-    src/metric_ops_host.cpp
+    src/metric_ops/metric_ops_host.cpp
     src/embedding_inplace_ops/embedding_inplace_update_gpu.cpp
-    src/input_combine_gpu.cpp
+    src/input_combine_ops/input_combine_gpu.cpp
     codegen/batch_index_select_dim0_host.cpp)
 
   if(NVML_LIB_PATH)
@@ -633,7 +633,7 @@ if(NOT FBGEMM_CPU_ONLY)
       src/memory_utils/memory_utils_ops.cu
       src/embedding_inplace_ops/embedding_inplace_update.cu
       src/histogram_binning_calibration_ops.cu
-      src/input_combine.cu
+      src/input_combine_ops/input_combine.cu
       src/jagged_tensor_ops/batched_dense_vec_jagged_2d_mul_backward.cu
       src/jagged_tensor_ops/batched_dense_vec_jagged_2d_mul_forward.cu
       src/jagged_tensor_ops/dense_to_jagged_forward.cu
@@ -651,8 +651,8 @@ if(NOT FBGEMM_CPU_ONLY)
       src/jagged_tensor_ops/jagged_to_padded_dense_forward.cu
       src/jagged_tensor_ops/jagged_unique_indices.cu
       src/jagged_tensor_ops/keyed_jagged_index_select_dim1.cu
-      src/layout_transform_ops.cu
-      src/metric_ops.cu
+      src/layout_transform_ops/layout_transform_ops.cu
+      src/metric_ops/metric_ops.cu
       src/permute_pooled_embedding_ops/permute_pooled_embedding_ops_split.cu
       src/permute_pooled_embedding_ops/permute_pooled_embedding_ops.cu
       src/quantize_ops/quantize_bfloat16.cu

fbgemm_gpu/src/input_combine.cu renamed to fbgemm_gpu/src/input_combine_ops/input_combine.cu

@@ -13,19 +13,16 @@
 #include "fbgemm_gpu/cub_namespace_postfix.cuh"
 // clang-format on
 
-#include "fbgemm_gpu/layout_transform_ops.cuh"
-#include "fbgemm_gpu/sparse_ops.h"
-#include "fbgemm_gpu/sparse_ops_utils.h"
-
 #include <ATen/ATen.h>
 #include <ATen/core/op_registration/op_registration.h>
 #include <ATen/cuda/CUDAContext.h>
 #include <ATen/cuda/Exceptions.h>
 #include <c10/cuda/CUDAGuard.h>
-
 #include <torch/library.h>
-
 #include "ATen/Parallel.h"
+#include "fbgemm_gpu/layout_transform_ops.cuh"
+#include "fbgemm_gpu/sparse_ops.h"
+#include "fbgemm_gpu/sparse_ops_utils.h"
 
 using Tensor = at::Tensor;
 

fbgemm_gpu/src/input_combine_cpu.cpp renamed to fbgemm_gpu/src/input_combine_ops/input_combine_cpu.cpp

@@ -6,12 +6,11 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-#include "fbgemm_gpu/sparse_ops.h"
-#include "fbgemm_gpu/sparse_ops_utils.h"
-
 #include <ATen/ATen.h>
 #include <ATen/core/op_registration/op_registration.h>
 #include <torch/library.h>
+#include "fbgemm_gpu/sparse_ops.h"
+#include "fbgemm_gpu/sparse_ops_utils.h"
 
 TORCH_LIBRARY_IMPL(fbgemm, CUDA, m) {
   DISPATCH_TO_CUDA(