solve apl dependency issue (#145215)

alinpahontu2912 · Ozan Aydin · pytorchmergebot · commit 0674ab7e33c3 · 2025-01-27T13:02:16.000Z
According to the [APL documentation](https://developer.arm.com/documentation/101004/2404/General-information/Arm-Performance-Libraries-example-programs), libraries ending with _mp are OpenMP multi-threaded libraries. When a project is compiled with MSVC and the -openmp flag, the vcomp library (Visual C++ implementation of OpenMP) is used for runtime calls. However, the current APL implementation uses the libomp.dll (LLVM) variant. As a result, there are unexpected behaviors at runtime. --- For Example: ```python import torch # Create a sparse tensor # Input (Sparse Tensor): # [[0, 1], # [1, 0]] indices = torch.tensor([[0, 1], [1, 0]]) values = torch.tensor([1, 1], dtype=torch.float32) size = torch.Size([2, 2]) sparse_tensor = torch.sparse_coo_tensor(indices, values, size) # Convert sparse tensor to dense tensor dense_tensor = sparse_tensor.to_dense() # Expected Output (Dense Tensor): # [[0, 1], # [1, 0]] print("\nDense Tensor:") print(dense_tensor) ``` However, it prints unexpected outputs such as: ```python # [[0, 11], # [10, 0]] ``` The issue arises because the following code does not function as expected at runtime: https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/ParallelOpenMP.h#L30 ```c++ // returns 1 , however since OpenMP is enabled it should return total number of threads int64_t num_threads = omp_get_num_threads(); ``` --- In the runtime, loading multiple OpenMP libraries (in this case `libomp` and `vcomp`) is causing unexpected behaviours. So, we've changed libraries from `_mp` to non `_mp` versions and we used `vcomp` for OpenMP calls. Pull Request resolved: #145215 Approved by: https://github.com/ozanMSFT, https://github.com/malfet Co-authored-by: Ozan Aydin <148207261+ozanMSFT@users.noreply.github.com>
diff --git a/cmake/Modules/FindAPL.cmake b/cmake/Modules/FindAPL.cmake
@@ -10,6 +10,7 @@
 
 SET(APL_INCLUDE_SEARCH_PATHS $ENV{ARMPL_DIR}/include)
 SET(APL_LIB_SEARCH_PATHS $ENV{ARMPL_DIR}/lib)
+SET(APL_BIN_SEARCH_PATHS $ENV{ARMPL_DIR}/bin)
 
 SET(APL_FOUND ON)
 
@@ -21,21 +22,36 @@ IF(NOT APL_INCLUDE_DIR)
 ENDIF()
 
 # Check lib file
-FIND_PATH(APL_LIB_DIR NAMES libarmpl_lp64_mp.dll.lib libomp.dll.lib libarmpl_lp64_mp.a PATHS ${APL_LIB_SEARCH_PATHS})
+FIND_PATH(APL_LIB_DIR NAMES armpl_lp64.dll.lib libarmpl_lp64.a PATHS ${APL_LIB_SEARCH_PATHS})
 IF(NOT APL_LIB_DIR)
     SET(APL_FOUND OFF)
     MESSAGE(STATUS "Could not verify APL lib directory. Turning APL_FOUND off")
 ENDIF()
 
+# Check bin file
+FIND_PATH(APL_BIN_DIR NAMES armpl_lp64.dll libarmpl_lp64.a PATHS ${APL_BIN_SEARCH_PATHS})
+IF(NOT APL_BIN_DIR)
+    SET(APL_FOUND OFF)
+    MESSAGE(STATUS "Could not verify APL bin directory. Turning APL_FOUND off")
+ENDIF()
+
 IF (APL_FOUND)
   IF(WIN32)
     set(APL_LIBRARIES
-      "${APL_LIB_DIR}/libarmpl_lp64_mp.dll.lib"
-      "${APL_LIB_DIR}/libomp.dll.lib"
+      "${APL_LIB_DIR}/armpl_lp64.dll.lib"
+    )
+    set(APL_DLLS
+      "${CMAKE_INSTALL_PREFIX}/lib/armpl_lp64.dll"
+    )
+    add_custom_command(
+      OUTPUT ${APL_DLLS}
+      COMMAND ${CMAKE_COMMAND} -E make_directory "${CMAKE_INSTALL_PREFIX}/lib"
+      COMMAND ${CMAKE_COMMAND} -E copy_if_different "${APL_BIN_DIR}/armpl_lp64.dll" "${CMAKE_INSTALL_PREFIX}/lib/armpl_lp64.dll"
     )
+    add_custom_target(copy_apl_dlls ALL DEPENDS ${APL_DLLS})
   ELSEIF(UNIX)
     set(APL_LIBRARIES
-      "${APL_LIB_DIR}/libarmpl_lp64_mp.a"
+      "${APL_LIB_DIR}/libarmpl_lp64.a"
     )
   ENDIF()
   MESSAGE(STATUS "Found APL header: ${APL_INCLUDE_DIR}")
diff --git a/cmake/Modules/FindOpenMP.cmake b/cmake/Modules/FindOpenMP.cmake
@@ -254,6 +254,16 @@ function(_OPENMP_GET_FLAGS LANG FLAG_MODE OPENMP_FLAG_VAR OPENMP_LIB_NAMES_VAR)
       set(OpenMP_libomp_LIBRARY "${MKL_OPENMP_LIBRARY}" CACHE STRING "libomp location for OpenMP")
     endif()
 
+    if ((NOT OpenMP_libomp_LIBRARY) AND MSVC AND CMAKE_SYSTEM_PROCESSOR STREQUAL "ARM64")
+      # On MSVC ARM64, OpenMP is provided by vcomp, which is a part of the Visual Studio installation.
+      find_library(OpenMP_libomp_LIBRARY
+      NAMES vcomp
+      HINTS ${CMAKE_${LANG}_IMPLICIT_LINK_DIRECTORIES}
+      DOC "vcomp location for OpenMP on MSVC ARM64"
+      )
+      mark_as_advanced(OpenMP_libomp_LIBRARY)
+    endif()
+
     if (NOT OpenMP_libomp_LIBRARY)
       find_library(OpenMP_libomp_LIBRARY
         NAMES omp gomp iomp5
