From 5da5a1626e7b34d1dfdb82ee1f9936fdc177b553 Mon Sep 17 00:00:00 2001 From: Benson Ma Date: Thu, 16 Nov 2023 03:40:04 -0800 Subject: [PATCH] Re-organize layout_transform_ops (#2133) Summary: Pull Request resolved: https://github.com/pytorch/FBGEMM/pull/2133 - Migrate layout_transform_ops into its own directory Reviewed By: spcyppt Differential Revision: D51290598 fbshipit-source-id: 99ff3f170b29b999a761f327f909053b3bf9f2a6 --- fbgemm_gpu/CMakeLists.txt | 16 ++++++++-------- .../src/{ => input_combine_ops}/input_combine.cu | 0 .../input_combine_cpu.cpp | 0 .../input_combine_gpu.cpp | 0 .../layout_transform_ops.cu | 9 +++------ .../layout_transform_ops_cpu.cpp | 0 .../layout_transform_ops_gpu.cpp | 5 ++--- fbgemm_gpu/src/{ => metric_ops}/metric_ops.cu | 0 fbgemm_gpu/src/{ => metric_ops}/metric_ops.h | 0 .../src/{ => metric_ops}/metric_ops_host.cpp | 0 10 files changed, 13 insertions(+), 17 deletions(-) rename fbgemm_gpu/src/{ => input_combine_ops}/input_combine.cu (100%) rename fbgemm_gpu/src/{ => input_combine_ops}/input_combine_cpu.cpp (100%) rename fbgemm_gpu/src/{ => input_combine_ops}/input_combine_gpu.cpp (100%) rename fbgemm_gpu/src/{ => layout_transform_ops}/layout_transform_ops.cu (99%) rename fbgemm_gpu/src/{ => layout_transform_ops}/layout_transform_ops_cpu.cpp (100%) rename fbgemm_gpu/src/{ => layout_transform_ops}/layout_transform_ops_gpu.cpp (99%) rename fbgemm_gpu/src/{ => metric_ops}/metric_ops.cu (100%) rename fbgemm_gpu/src/{ => metric_ops}/metric_ops.h (100%) rename fbgemm_gpu/src/{ => metric_ops}/metric_ops_host.cpp (100%) diff --git a/fbgemm_gpu/CMakeLists.txt b/fbgemm_gpu/CMakeLists.txt index 9b1a11883..95ae26361 100644 --- a/fbgemm_gpu/CMakeLists.txt +++ b/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 diff --git a/fbgemm_gpu/src/input_combine.cu b/fbgemm_gpu/src/input_combine_ops/input_combine.cu similarity index 100% rename from fbgemm_gpu/src/input_combine.cu rename to fbgemm_gpu/src/input_combine_ops/input_combine.cu diff --git a/fbgemm_gpu/src/input_combine_cpu.cpp b/fbgemm_gpu/src/input_combine_ops/input_combine_cpu.cpp similarity index 100% rename from fbgemm_gpu/src/input_combine_cpu.cpp rename to fbgemm_gpu/src/input_combine_ops/input_combine_cpu.cpp diff --git a/fbgemm_gpu/src/input_combine_gpu.cpp b/fbgemm_gpu/src/input_combine_ops/input_combine_gpu.cpp similarity index 100% rename from fbgemm_gpu/src/input_combine_gpu.cpp rename to fbgemm_gpu/src/input_combine_ops/input_combine_gpu.cpp diff --git a/fbgemm_gpu/src/layout_transform_ops.cu b/fbgemm_gpu/src/layout_transform_ops/layout_transform_ops.cu similarity index 99% rename from fbgemm_gpu/src/layout_transform_ops.cu rename to fbgemm_gpu/src/layout_transform_ops/layout_transform_ops.cu index 3b76e5b32..40b796d23 100644 --- a/fbgemm_gpu/src/layout_transform_ops.cu +++ b/fbgemm_gpu/src/layout_transform_ops/layout_transform_ops.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 #include #include #include #include - #include - #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; diff --git a/fbgemm_gpu/src/layout_transform_ops_cpu.cpp b/fbgemm_gpu/src/layout_transform_ops/layout_transform_ops_cpu.cpp similarity index 100% rename from fbgemm_gpu/src/layout_transform_ops_cpu.cpp rename to fbgemm_gpu/src/layout_transform_ops/layout_transform_ops_cpu.cpp diff --git a/fbgemm_gpu/src/layout_transform_ops_gpu.cpp b/fbgemm_gpu/src/layout_transform_ops/layout_transform_ops_gpu.cpp similarity index 99% rename from fbgemm_gpu/src/layout_transform_ops_gpu.cpp rename to fbgemm_gpu/src/layout_transform_ops/layout_transform_ops_gpu.cpp index 986b479bd..19e757875 100644 --- a/fbgemm_gpu/src/layout_transform_ops_gpu.cpp +++ b/fbgemm_gpu/src/layout_transform_ops/layout_transform_ops_gpu.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 #include #include +#include "fbgemm_gpu/sparse_ops.h" +#include "fbgemm_gpu/sparse_ops_utils.h" TORCH_LIBRARY_IMPL(fbgemm, CUDA, m) { DISPATCH_TO_CUDA( diff --git a/fbgemm_gpu/src/metric_ops.cu b/fbgemm_gpu/src/metric_ops/metric_ops.cu similarity index 100% rename from fbgemm_gpu/src/metric_ops.cu rename to fbgemm_gpu/src/metric_ops/metric_ops.cu diff --git a/fbgemm_gpu/src/metric_ops.h b/fbgemm_gpu/src/metric_ops/metric_ops.h similarity index 100% rename from fbgemm_gpu/src/metric_ops.h rename to fbgemm_gpu/src/metric_ops/metric_ops.h diff --git a/fbgemm_gpu/src/metric_ops_host.cpp b/fbgemm_gpu/src/metric_ops/metric_ops_host.cpp similarity index 100% rename from fbgemm_gpu/src/metric_ops_host.cpp rename to fbgemm_gpu/src/metric_ops/metric_ops_host.cpp