pytorch

9ce95ce1 - [pytorch] add Vulkan support for the `t` and `transpose` operators for 2d, 3d and 4d tensors (#101808)

[pytorch] add Vulkan support for the `t` and `transpose` operators for 2d, 3d and 4d tensors (#101808) Summary: Use the existing permute shader to implement the following two operators for Vulkan backend - `aten::transpose` The behavior of the operator is shown in https://pytorch.org/docs/stable/generated/torch.transpose.html. - `aten::t` The behavior of the operator is shown in https://pytorch.org/docs/stable/generated/torch.t.html#torch.t. 1d tensors are returned as is. When input is a 2d tensor this is equivalent to `aten::transpose(input, 0, 1)`. Test Plan: At local repo of fbsource on MacBook, run `buck run --target-platforms ovr_config//platform/macos:arm64-fbsource //xplat/caffe2:pt_vulkan_api_test_binAppleMac\#macosx-arm64 -c pt.vulkan_full_precision=1` - Full test results P739033174. - `aten::t` and `aten::tranpose` related results shown below ``` (base) luwei@luwei-mbp fbsource % buck run --target-platforms ovr_config//platform/macos:arm64-fbsource //xplat/caffe2:pt_vulkan_api_test_binAppleMac\#macosx-arm64 -c pt.vulkan_full_precision=1 [... other tests ...] [ RUN ] VulkanAPITest.transpose_t_1d [ OK ] VulkanAPITest.transpose_t_1d (0 ms) [ RUN ] VulkanAPITest.transpose_t_2d_small [ OK ] VulkanAPITest.transpose_t_2d_small (1 ms) [ RUN ] VulkanAPITest.transpose_t_2d_medium [ OK ] VulkanAPITest.transpose_t_2d_medium (0 ms) [ RUN ] VulkanAPITest.transpose_t_2d_large [ OK ] VulkanAPITest.transpose_t_2d_large (0 ms) [ RUN ] VulkanAPITest.transpose_2d_height_and_width_small [ OK ] VulkanAPITest.transpose_2d_height_and_width_small (0 ms) [ RUN ] VulkanAPITest.transpose_2d_height_and_width_medium [ OK ] VulkanAPITest.transpose_2d_height_and_width_medium (0 ms) [ RUN ] VulkanAPITest.transpose_2d_height_and_width_large [ OK ] VulkanAPITest.transpose_2d_height_and_width_large (0 ms) [ RUN ] VulkanAPITest.transpose_2d_height_and_height_large [ OK ] VulkanAPITest.transpose_2d_height_and_height_large (0 ms) [ RUN ] VulkanAPITest.transpose_2d_width_and_width_large [ OK ] VulkanAPITest.transpose_2d_width_and_width_large (0 ms) [ RUN ] VulkanAPITest.transpose_3d_height_and_width_small [ OK ] VulkanAPITest.transpose_3d_height_and_width_small (0 ms) [ RUN ] VulkanAPITest.transpose_3d_height_and_width_medium [ OK ] VulkanAPITest.transpose_3d_height_and_width_medium (1 ms) [ RUN ] VulkanAPITest.transpose_3d_height_and_width_large [ OK ] VulkanAPITest.transpose_3d_height_and_width_large (1 ms) [ RUN ] VulkanAPITest.transpose_3d_width_and_width_large [ OK ] VulkanAPITest.transpose_3d_width_and_width_large (0 ms) [ RUN ] VulkanAPITest.transpose_3d_depth_and_width_small [ OK ] VulkanAPITest.transpose_3d_depth_and_width_small (0 ms) [ RUN ] VulkanAPITest.transpose_3d_depth_and_width_medium [ OK ] VulkanAPITest.transpose_3d_depth_and_width_medium (0 ms) [ RUN ] VulkanAPITest.transpose_3d_depth_and_width_large [ OK ] VulkanAPITest.transpose_3d_depth_and_width_large (0 ms) [ RUN ] VulkanAPITest.transpose_3d_depth_and_depth_large [ OK ] VulkanAPITest.transpose_3d_depth_and_depth_large (0 ms) [ RUN ] VulkanAPITest.transpose_3d_depth_and_height_small [ OK ] VulkanAPITest.transpose_3d_depth_and_height_small (0 ms) [ RUN ] VulkanAPITest.transpose_3d_depth_and_height_medium [ OK ] VulkanAPITest.transpose_3d_depth_and_height_medium (0 ms) [ RUN ] VulkanAPITest.transpose_3d_depth_and_height_large [ OK ] VulkanAPITest.transpose_3d_depth_and_height_large (2 ms) [ RUN ] VulkanAPITest.transpose_3d_height_and_height_large [ OK ] VulkanAPITest.transpose_3d_height_and_height_large (1 ms) [ RUN ] VulkanAPITest.transpose_4d_batch_and_batch_large [ OK ] VulkanAPITest.transpose_4d_batch_and_batch_large (1 ms) [ RUN ] VulkanAPITest.transpose_4d_depth_and_depth_large [ OK ] VulkanAPITest.transpose_4d_depth_and_depth_large (1 ms) [ RUN ] VulkanAPITest.transpose_4d_height_and_height_large [ OK ] VulkanAPITest.transpose_4d_height_and_height_large (1 ms) [ RUN ] VulkanAPITest.transpose_4d_width_and_width_large [ OK ] VulkanAPITest.transpose_4d_width_and_width_large (0 ms) [ RUN ] VulkanAPITest.transpose_4d_batch_and_depth_large [ OK ] VulkanAPITest.transpose_4d_batch_and_depth_large (1 ms) [ RUN ] VulkanAPITest.transpose_4d_batch_and_height_large [ OK ] VulkanAPITest.transpose_4d_batch_and_height_large (2 ms) [ RUN ] VulkanAPITest.transpose_4d_batch_and_width_large [ OK ] VulkanAPITest.transpose_4d_batch_and_width_large (2 ms) [ RUN ] VulkanAPITest.transpose_4d_depth_and_height_large [ OK ] VulkanAPITest.transpose_4d_depth_and_height_large (2 ms) [ RUN ] VulkanAPITest.transpose_4d_depth_and_width_large [ OK ] VulkanAPITest.transpose_4d_depth_and_width_large (2 ms) [ RUN ] VulkanAPITest.transpose_4d_height_and_width_large [ OK ] VulkanAPITest.transpose_4d_height_and_width_large (1 ms) [... other tests ...] ``` Reviewed By: SS-JIA Differential Revision: D45878333 Pull Request resolved: https://github.com/pytorch/pytorch/pull/101808 Approved by: https://github.com/SS-JIA