[fftw3] add neon support (#45222)

Co-authored-by: Michael Schollerer <michael.schollerer@ifm.com> Co-authored-by: Billy Robert O'Neal III <bion@microsoft.com> Co-authored-by: Kai Pastor <dg0yt@darc.de>
2025-05-14 23:07:16 +02:00 · 2025-05-14 23:07:16 +02:00 · 64810c6bf2
parent 115846a5b2
commit 64810c6bf2
6 changed files with 93 additions and 3 deletions
--- a/ports/fftw3/neon.patch
+++ b/ports/fftw3/neon.patch
@ -0,0 +1,73 @@
+diff --git a/CMakeLists.txt b/CMakeLists.txt
+index db20caa7..48e724f3 100644
+--- a/CMakeLists.txt
+++ b/CMakeLists.txt
+@@ -25,6 +25,7 @@ option (ENABLE_SSE "Compile with SSE instruction set support" OFF)
+ option (ENABLE_SSE2 "Compile with SSE2 instruction set support" OFF)
+ option (ENABLE_AVX "Compile with AVX instruction set support" OFF)
+ option (ENABLE_AVX2 "Compile with AVX2 instruction set support" OFF)
+option (ENABLE_NEON "Compile with NEON instruction set support" OFF)
+ 
+ option (DISABLE_FORTRAN "Disable Fortran wrapper routines" OFF)
+ 
+@@ -203,9 +204,20 @@ if (ENABLE_AVX2)
+   endforeach ()
+ endif ()
+ 
+if (ENABLE_NEON)
+  if (ENABLE_LONG_DOUBLE)
+    message (FATAL_ERROR "NEON only works in single or double precision, please disable long double support")
+  endif ()
+  if (ENABLE_QUAD_PRECISION)
+    message (FATAL_ERROR "NEON only works in single or double precision, please disable quad precision support")
+  endif ()
+  set (HAVE_NEON TRUE)
+endif ()
+
+ if (HAVE_SSE2 OR HAVE_AVX)
+   set (HAVE_SIMD TRUE)
+ endif ()
+
+ file(GLOB           fftw_api_SOURCE                 api/*.c             api/*.h)
+ file(GLOB           fftw_dft_SOURCE                 dft/*.c             dft/*.h)
+ file(GLOB           fftw_dft_scalar_SOURCE          dft/scalar/*.c      dft/scalar/*.h)
+@@ -215,6 +227,7 @@ file(GLOB           fftw_dft_simd_SOURCE            dft/simd/*.c        dft/simd
+ file(GLOB           fftw_dft_simd_sse2_SOURCE       dft/simd/sse2/*.c   dft/simd/sse2/*.h)
+ file(GLOB           fftw_dft_simd_avx_SOURCE        dft/simd/avx/*.c    dft/simd/avx/*.h)
+ file(GLOB           fftw_dft_simd_avx2_SOURCE       dft/simd/avx2/*.c   dft/simd/avx2/*.h dft/simd/avx2-128/*.c   dft/simd/avx2-128/*.h)
+file(GLOB           fftw_dft_simd_neon_SOURCE       dft/simd/neon/*.c   dft/simd/neon/*.h)
+ file(GLOB           fftw_kernel_SOURCE              kernel/*.c          kernel/*.h)
+ file(GLOB           fftw_rdft_SOURCE                rdft/*.c            rdft/*.h)
+ file(GLOB           fftw_rdft_scalar_SOURCE         rdft/scalar/*.c     rdft/scalar/*.h)
+@@ -230,6 +243,7 @@ file(GLOB           fftw_rdft_simd_SOURCE           rdft/simd/*.c       rdft/sim
+ file(GLOB           fftw_rdft_simd_sse2_SOURCE      rdft/simd/sse2/*.c  rdft/simd/sse2/*.h)
+ file(GLOB           fftw_rdft_simd_avx_SOURCE       rdft/simd/avx/*.c   rdft/simd/avx/*.h)
+ file(GLOB           fftw_rdft_simd_avx2_SOURCE      rdft/simd/avx2/*.c  rdft/simd/avx2/*.h rdft/simd/avx2-128/*.c  rdft/simd/avx2-128/*.h)
+file(GLOB           fftw_rdft_simd_neon_SOURCE      rdft/simd/neon/*.c  rdft/simd/neon/*.h)
+ 
+ file(GLOB           fftw_reodft_SOURCE              reodft/*.c          reodft/*.h)
+ file(GLOB           fftw_simd_support_SOURCE        simd-support/*.c    simd-support/*.h)
+@@ -283,6 +297,10 @@ if (HAVE_AVX2)
+   list (APPEND SOURCEFILES ${fftw_dft_simd_avx2_SOURCE} ${fftw_rdft_simd_avx2_SOURCE})
+ endif ()
+ 
+if (HAVE_NEON)
+  list (APPEND SOURCEFILES ${fftw_dft_simd_neon_SOURCE} ${fftw_rdft_simd_neon_SOURCE})
+endif ()
+
+ set (FFTW_VERSION 3.3.10)
+ 
+ set (PREC_SUFFIX)
+ diff --git a/cmake.config.h.in b/cmake.config.h.in
+index 1f4c5055..8c61b38f 100644
+--- a/cmake.config.h.in
+++ b/cmake.config.h.in
+@@ -211,7 +211,7 @@
+ /* #undef HAVE_MPI */
+ 
+ /* Define to enable ARM NEON optimizations. */
+-/* #undef HAVE_NEON */
+#cmakedefine HAVE_NEON 1
+ 
+ /* Define if OpenMP is enabled */
+ #cmakedefine HAVE_OPENMP
--- a/ports/fftw3/portfile.cmake
+++ b/ports/fftw3/portfile.cmake
@ -14,6 +14,7 @@ vcpkg_extract_source_archive(
        fix-openmp.patch
        install-subtargets.patch
        fix-wrong-version.patch # https://github.com/FFTW/fftw3/commit/0842f00ae6b6e1f3aade155bc0edd17a7313fa6a
+        neon.patch # https://github.com/FFTW/fftw3/pull/275/commits/262f5cfe23af54930b119bd3653bc25bf2d881da
 )

 vcpkg_check_features(
@ -34,6 +35,13 @@ set(fftw3_options  "")
 set(fftw3f_options -DENABLE_FLOAT=ON)
 set(fftw3l_options -DENABLE_LONG_DOUBLE=ON -DENABLE_AVX2=OFF -DENABLE_AVX=OFF -DENABLE_SSE2=OFF)

+if("neon" IN_LIST FEATURES)
+    list(APPEND fftw3f_options -DENABLE_NEON=ON)
+    if(VCPKG_TARGET_ARCHITECTURE STREQUAL "arm64")
+        list(APPEND fftw3_options -DENABLE_NEON=ON)
+    endif()
+endif()
+
 foreach(package_name IN LISTS package_names)
    message(STATUS "${package_name}...")
    vcpkg_cmake_configure(
--- a/ports/fftw3/vcpkg.json
+++ b/ports/fftw3/vcpkg.json
@ -1,7 +1,7 @@
 {
  "name": "fftw3",
  "version": "3.3.10",
-  "port-version": 9,
+  "port-version": 10,
  "description": "FFTW is a C subroutine library for computing the discrete Fourier transform (DFT) in one or more dimensions, of arbitrary input size, and of both real and complex data (as well as of even/odd data, i.e. the discrete cosine/sine transforms or DCT/DST).",
  "homepage": "https://www.fftw.org/",
  "license": "GPL-2.0-or-later",
@ -24,6 +24,10 @@
      "description": "Builds part of the library with avx2, fma, avx, sse2, sse",
      "supports": "!arm"
    },
+    "neon": {
+      "description": "Builds part of the library with neon",
+      "supports": "(linux | android) & arm"
+    },
    "openmp": {
      "description": "Builds openmp enabled lib"
    },
--- a/scripts/ci.feature.baseline.txt
+++ b/scripts/ci.feature.baseline.txt
@ -2164,7 +2164,7 @@ embree3[core,avx,avx2,avx512,default-features,sse2,sse42](osx) = combination-fai
 ffmpeg[all,all-gpl](x64 & android) = feature-fails
 ffmpeg[nvcodec,ffplay,opengl](android) = feature-fails
 fftw3[openmp](osx) = feature-fails # waits for https://github.com/microsoft/vcpkg/pull/30833
-fftw3[openmp](windows) = feature-fails # Linker missing symbols. See https://github.com/microsoft/vcpkg/issues/33322
+fftw3[openmp](windows & !static) = feature-fails # Linker missing symbols. See https://github.com/microsoft/vcpkg/issues/33322
 flashlight-cpu = fail # CMake Error: INSTALL(EXPORT) given unknown export "flashlightTargets". See https://github.com/microsoft/vcpkg/issues/32098
 flashlight-sequence[openmp](osx) = feature-fails # No openmp on osx
 freerdp[server](osx) = skip # Uses api removed with macOS 15. Don't get fixed. See https://github.com/FreeRDP/FreeRDP/issues/10558
--- a/versions/baseline.json
+++ b/versions/baseline.json
@ -2810,7 +2810,7 @@
    },
    "fftw3": {
      "baseline": "3.3.10",
-      "port-version": 9
+      "port-version": 10
    },
    "fftwpp": {
      "baseline": "2019-12-19",
--- a/versions/f-/fftw3.json
+++ b/versions/f-/fftw3.json
@ -1,5 +1,10 @@
 {
  "versions": [
+    {
+      "git-tree": "98a1c4076fefe0c03872693b7305204b2db65f2c",
+      "version": "3.3.10",
+      "port-version": 10
+    },
    {
      "git-tree": "d64246f2b42f53756787bd4f886488846a789030",
      "version": "3.3.10",