diff --git a/.github/workflows/cygwinccompiler.py b/.github/workflows/cygwinccompiler.py
deleted file mode 100755
index 72d8558ae..000000000
--- a/.github/workflows/cygwinccompiler.py
+++ /dev/null
@@ -1,408 +0,0 @@
-"""distutils.cygwinccompiler
-
-Provides the CygwinCCompiler class, a subclass of UnixCCompiler that
-handles the Cygwin port of the GNU C compiler to Windows.  It also contains
-the Mingw32CCompiler class which handles the mingw32 port of GCC (same as
-cygwin in no-cygwin mode).
-"""
-
-# problems:
-#
-# * if you use a msvc compiled python version (1.5.2)
-#   1. you have to insert a __GNUC__ section in its config.h
-#   2. you have to generate an import library for its dll
-#      - create a def-file for python??.dll
-#      - create an import library using
-#             dlltool --dllname python15.dll --def python15.def \
-#                       --output-lib libpython15.a
-#
-#   see also http://starship.python.net/crew/kernr/mingw32/Notes.html
-#
-# * We put export_symbols in a def-file, and don't use
-#   --export-all-symbols because it doesn't worked reliable in some
-#   tested configurations. And because other windows compilers also
-#   need their symbols specified this no serious problem.
-#
-# tested configurations:
-#
-# * cygwin gcc 2.91.57/ld 2.9.4/dllwrap 0.2.4 works
-#   (after patching python's config.h and for C++ some other include files)
-#   see also http://starship.python.net/crew/kernr/mingw32/Notes.html
-# * mingw32 gcc 2.95.2/ld 2.9.4/dllwrap 0.2.4 works
-#   (ld doesn't support -shared, so we use dllwrap)
-# * cygwin gcc 2.95.2/ld 2.10.90/dllwrap 2.10.90 works now
-#   - its dllwrap doesn't work, there is a bug in binutils 2.10.90
-#     see also http://sources.redhat.com/ml/cygwin/2000-06/msg01274.html
-#   - using gcc -mdll instead dllwrap doesn't work without -static because
-#     it tries to link against dlls instead their import libraries. (If
-#     it finds the dll first.)
-#     By specifying -static we force ld to link against the import libraries,
-#     this is windows standard and there are normally not the necessary symbols
-#     in the dlls.
-#   *** only the version of June 2000 shows these problems
-# * cygwin gcc 3.2/ld 2.13.90 works
-#   (ld supports -shared)
-# * mingw gcc 3.2/ld 2.13 works
-#   (ld supports -shared)
-
-import os
-import sys
-import copy
-from subprocess import Popen, PIPE, check_output
-import re
-
-from distutils.ccompiler import gen_preprocess_options, gen_lib_options
-from distutils.unixccompiler import UnixCCompiler
-from distutils.file_util import write_file
-from distutils.errors import (DistutilsExecError, CCompilerError,
-        CompileError, UnknownFileError)
-from distutils import log
-from distutils.version import LooseVersion
-from distutils.spawn import find_executable
-
-def get_msvcr():
-    """Include the appropriate MSVC runtime library if Python was built
-    with MSVC 7.0 or later.
-    """
-    msc_pos = sys.version.find('MSC v.')
-    if msc_pos != -1:
-        msc_ver = sys.version[msc_pos+6:msc_pos+10]
-        if msc_ver == '1300':
-            # MSVC 7.0
-            return ['msvcr70']
-        elif msc_ver == '1310':
-            # MSVC 7.1
-            return ['msvcr71']
-        elif msc_ver == '1400':
-            # VS2005 / MSVC 8.0
-            return ['msvcr80']
-        elif msc_ver == '1500':
-            # VS2008 / MSVC 9.0
-            return ['msvcr90']
-        elif msc_ver == '1600':
-            # VS2010 / MSVC 10.0
-            return ['msvcr100']
-        elif msc_ver == '1900':
-            return ['vcruntime140']
-        else:
-            return ['vcruntime140']
-            #raise ValueError("Unknown MS Compiler version %s " % msc_ver)
-
-
-class CygwinCCompiler(UnixCCompiler):
-    """ Handles the Cygwin port of the GNU C compiler to Windows.
-    """
-    compiler_type = 'cygwin'
-    obj_extension = ".o"
-    static_lib_extension = ".a"
-    shared_lib_extension = ".dll"
-    static_lib_format = "lib%s%s"
-    shared_lib_format = "%s%s"
-    exe_extension = ".exe"
-
-    def __init__(self, verbose=0, dry_run=0, force=0):
-
-        UnixCCompiler.__init__(self, verbose, dry_run, force)
-
-        status, details = check_config_h()
-        self.debug_print("Python's GCC status: %s (details: %s)" %
-                         (status, details))
-        if status is not CONFIG_H_OK:
-            self.warn(
-                "Python's pyconfig.h doesn't seem to support your compiler. "
-                "Reason: %s. "
-                "Compiling may fail because of undefined preprocessor macros."
-                % details)
-
-        self.gcc_version, self.ld_version, self.dllwrap_version = \
-            get_versions()
-        self.debug_print(self.compiler_type + ": gcc %s, ld %s, dllwrap %s\n" %
-                         (self.gcc_version,
-                          self.ld_version,
-                          self.dllwrap_version) )
-
-        # ld_version >= "2.10.90" and < "2.13" should also be able to use
-        # gcc -mdll instead of dllwrap
-        # Older dllwraps had own version numbers, newer ones use the
-        # same as the rest of binutils ( also ld )
-        # dllwrap 2.10.90 is buggy
-        if self.ld_version >= "2.10.90":
-            self.linker_dll = "gcc"
-        else:
-            self.linker_dll = "dllwrap"
-
-        # ld_version >= "2.13" support -shared so use it instead of
-        # -mdll -static
-        if self.ld_version >= "2.13":
-            shared_option = "-shared"
-        else:
-            shared_option = "-mdll -static"
-
-        # Hard-code GCC because that's what this is all about.
-        # XXX optimization, warnings etc. should be customizable.
-        self.set_executables(compiler='gcc -mcygwin -O -Wall',
-                             compiler_so='gcc -mcygwin -mdll -O -Wall',
-                             compiler_cxx='g++ -mcygwin -O -Wall',
-                             linker_exe='gcc -mcygwin',
-                             linker_so=('%s -mcygwin %s' %
-                                        (self.linker_dll, shared_option)))
-
-        # cygwin and mingw32 need different sets of libraries
-        if self.gcc_version == "2.91.57":
-            # cygwin shouldn't need msvcrt, but without the dlls will crash
-            # (gcc version 2.91.57) -- perhaps something about initialization
-            self.dll_libraries=["msvcrt"]
-            self.warn(
-                "Consider upgrading to a newer version of gcc")
-        else:
-            # Include the appropriate MSVC runtime library if Python was built
-            # with MSVC 7.0 or later.
-            self.dll_libraries = get_msvcr()
-
-    def _compile(self, obj, src, ext, cc_args, extra_postargs, pp_opts):
-        """Compiles the source by spawning GCC and windres if needed."""
-        if ext == '.rc' or ext == '.res':
-            # gcc needs '.res' and '.rc' compiled to object files !!!
-            try:
-                self.spawn(["windres", "-i", src, "-o", obj])
-            except DistutilsExecError as msg:
-                raise CompileError(msg)
-        else: # for other files use the C-compiler
-            try:
-                self.spawn(self.compiler_so + cc_args + [src, '-o', obj] +
-                           extra_postargs)
-            except DistutilsExecError as msg:
-                raise CompileError(msg)
-
-    def link(self, target_desc, objects, output_filename, output_dir=None,
-             libraries=None, library_dirs=None, runtime_library_dirs=None,
-             export_symbols=None, debug=0, extra_preargs=None,
-             extra_postargs=None, build_temp=None, target_lang=None):
-        """Link the objects."""
-        # use separate copies, so we can modify the lists
-        extra_preargs = copy.copy(extra_preargs or [])
-        libraries = copy.copy(libraries or [])
-        objects = copy.copy(objects or [])
-
-        # Additional libraries
-        libraries.extend(self.dll_libraries)
-
-        # handle export symbols by creating a def-file
-        # with executables this only works with gcc/ld as linker
-        if ((export_symbols is not None) and
-            (target_desc != self.EXECUTABLE or self.linker_dll == "gcc")):
-            # (The linker doesn't do anything if output is up-to-date.
-            # So it would probably better to check if we really need this,
-            # but for this we had to insert some unchanged parts of
-            # UnixCCompiler, and this is not what we want.)
-
-            # we want to put some files in the same directory as the
-            # object files are, build_temp doesn't help much
-            # where are the object files
-            temp_dir = os.path.dirname(objects[0])
-            # name of dll to give the helper files the same base name
-            (dll_name, dll_extension) = os.path.splitext(
-                os.path.basename(output_filename))
-
-            # generate the filenames for these files
-            def_file = os.path.join(temp_dir, dll_name + ".def")
-            lib_file = os.path.join(temp_dir, 'lib' + dll_name + ".a")
-
-            # Generate .def file
-            contents = [
-                "LIBRARY %s" % os.path.basename(output_filename),
-                "EXPORTS"]
-            for sym in export_symbols:
-                contents.append(sym)
-            self.execute(write_file, (def_file, contents),
-                         "writing %s" % def_file)
-
-            # next add options for def-file and to creating import libraries
-
-            # dllwrap uses different options than gcc/ld
-            if self.linker_dll == "dllwrap":
-                extra_preargs.extend(["--output-lib", lib_file])
-                # for dllwrap we have to use a special option
-                extra_preargs.extend(["--def", def_file])
-            # we use gcc/ld here and can be sure ld is >= 2.9.10
-            else:
-                # doesn't work: bfd_close build\...\libfoo.a: Invalid operation
-                #extra_preargs.extend(["-Wl,--out-implib,%s" % lib_file])
-                # for gcc/ld the def-file is specified as any object files
-                objects.append(def_file)
-
-        #end: if ((export_symbols is not None) and
-        #        (target_desc != self.EXECUTABLE or self.linker_dll == "gcc")):
-
-        # who wants symbols and a many times larger output file
-        # should explicitly switch the debug mode on
-        # otherwise we let dllwrap/ld strip the output file
-        # (On my machine: 10KB < stripped_file < ??100KB
-        #   unstripped_file = stripped_file + XXX KB
-        #  ( XXX=254 for a typical python extension))
-        if not debug:
-            extra_preargs.append("-s")
-
-        UnixCCompiler.link(self, target_desc, objects, output_filename,
-                           output_dir, libraries, library_dirs,
-                           runtime_library_dirs,
-                           None, # export_symbols, we do this in our def-file
-                           debug, extra_preargs, extra_postargs, build_temp,
-                           target_lang)
-
-    # -- Miscellaneous methods -----------------------------------------
-
-    def object_filenames(self, source_filenames, strip_dir=0, output_dir=''):
-        """Adds supports for rc and res files."""
-        if output_dir is None:
-            output_dir = ''
-        obj_names = []
-        for src_name in source_filenames:
-            # use normcase to make sure '.rc' is really '.rc' and not '.RC'
-            base, ext = os.path.splitext(os.path.normcase(src_name))
-            if ext not in (self.src_extensions + ['.rc','.res']):
-                raise UnknownFileError("unknown file type '%s' (from '%s')" % \
-                      (ext, src_name))
-            if strip_dir:
-                base = os.path.basename (base)
-            if ext in ('.res', '.rc'):
-                # these need to be compiled to object files
-                obj_names.append (os.path.join(output_dir,
-                                              base + ext + self.obj_extension))
-            else:
-                obj_names.append (os.path.join(output_dir,
-                                               base + self.obj_extension))
-        return obj_names
-
-# the same as cygwin plus some additional parameters
-class Mingw32CCompiler(CygwinCCompiler):
-    """ Handles the Mingw32 port of the GNU C compiler to Windows.
-    """
-    compiler_type = 'mingw32'
-
-    def __init__(self, verbose=0, dry_run=0, force=0):
-
-        CygwinCCompiler.__init__ (self, verbose, dry_run, force)
-
-        # ld_version >= "2.13" support -shared so use it instead of
-        # -mdll -static
-        if self.ld_version >= "2.13":
-            shared_option = "-shared"
-        else:
-            shared_option = "-mdll -static"
-
-        # A real mingw32 doesn't need to specify a different entry point,
-        # but cygwin 2.91.57 in no-cygwin-mode needs it.
-        if self.gcc_version <= "2.91.57":
-            entry_point = '--entry _DllMain@12'
-        else:
-            entry_point = ''
-
-        if is_cygwingcc():
-            raise CCompilerError(
-                'Cygwin gcc cannot be used with --compiler=mingw32')
-
-        self.set_executables(compiler='gcc -O -Wall',
-                             compiler_so='gcc -mdll -O -Wall',
-                             compiler_cxx='g++ -O -Wall',
-                             linker_exe='gcc',
-                             linker_so='%s %s %s'
-                                        % (self.linker_dll, shared_option,
-                                           entry_point))
-        # Maybe we should also append -mthreads, but then the finished
-        # dlls need another dll (mingwm10.dll see Mingw32 docs)
-        # (-mthreads: Support thread-safe exception handling on `Mingw32')
-
-        # no additional libraries needed
-        self.dll_libraries=[]
-
-        # Include the appropriate MSVC runtime library if Python was built
-        # with MSVC 7.0 or later.
-        self.dll_libraries = get_msvcr()
-
-# Because these compilers aren't configured in Python's pyconfig.h file by
-# default, we should at least warn the user if he is using an unmodified
-# version.
-
-CONFIG_H_OK = "ok"
-CONFIG_H_NOTOK = "not ok"
-CONFIG_H_UNCERTAIN = "uncertain"
-
-def check_config_h():
-    """Check if the current Python installation appears amenable to building
-    extensions with GCC.
-
-    Returns a tuple (status, details), where 'status' is one of the following
-    constants:
-
-    - CONFIG_H_OK: all is well, go ahead and compile
-    - CONFIG_H_NOTOK: doesn't look good
-    - CONFIG_H_UNCERTAIN: not sure -- unable to read pyconfig.h
-
-    'details' is a human-readable string explaining the situation.
-
-    Note there are two ways to conclude "OK": either 'sys.version' contains
-    the string "GCC" (implying that this Python was built with GCC), or the
-    installed "pyconfig.h" contains the string "__GNUC__".
-    """
-
-    # XXX since this function also checks sys.version, it's not strictly a
-    # "pyconfig.h" check -- should probably be renamed...
-
-    from distutils import sysconfig
-
-    # if sys.version contains GCC then python was compiled with GCC, and the
-    # pyconfig.h file should be OK
-    if "GCC" in sys.version:
-        return CONFIG_H_OK, "sys.version mentions 'GCC'"
-
-    # let's see if __GNUC__ is mentioned in python.h
-    fn = sysconfig.get_config_h_filename()
-    try:
-        config_h = open(fn)
-        try:
-            if "__GNUC__" in config_h.read():
-                return CONFIG_H_OK, "'%s' mentions '__GNUC__'" % fn
-            else:
-                return CONFIG_H_NOTOK, "'%s' does not mention '__GNUC__'" % fn
-        finally:
-            config_h.close()
-    except OSError as exc:
-        return (CONFIG_H_UNCERTAIN,
-                "couldn't read '%s': %s" % (fn, exc.strerror))
-
-RE_VERSION = re.compile(br'(\d+\.\d+(\.\d+)*)')
-
-def _find_exe_version(cmd):
-    """Find the version of an executable by running `cmd` in the shell.
-
-    If the command is not found, or the output does not match
-    `RE_VERSION`, returns None.
-    """
-    executable = cmd.split()[0]
-    if find_executable(executable) is None:
-        return None
-    out = Popen(cmd, shell=True, stdout=PIPE).stdout
-    try:
-        out_string = out.read()
-    finally:
-        out.close()
-    result = RE_VERSION.search(out_string)
-    if result is None:
-        return None
-    # LooseVersion works with strings
-    # so we need to decode our bytes
-    return LooseVersion(result.group(1).decode())
-
-def get_versions():
-    """ Try to find out the versions of gcc, ld and dllwrap.
-
-    If not possible it returns None for it.
-    """
-    commands = ['gcc -dumpversion', 'ld -v', 'dllwrap --version']
-    return tuple([_find_exe_version(cmd) for cmd in commands])
-
-def is_cygwingcc():
-    '''Try to determine if the gcc that would be used is from cygwin.'''
-    out_string = check_output(['gcc', '-dumpmachine'])
-    return out_string.strip().endswith(b'cygwin')
diff --git a/.github/workflows/generate_matrix.py b/.github/workflows/generate_matrix.py
new file mode 100644
index 000000000..29049e864
--- /dev/null
+++ b/.github/workflows/generate_matrix.py
@@ -0,0 +1,36 @@
+import json
+
+matrix = {
+    "include": []
+}
+
+python_versions = ["3.8", "3.11"]
+
+combinations = {
+    "ubuntu-22.04": {
+        "compiler": ["llvm", "gcc"],
+        "arch_flags": ["-march=native", "-march=x86-64"]
+    },
+    "windows-2022": {
+        "compiler": ["msvc", "llvm"],
+        "arch_flags": ["/arch:AVX2", "/arch:SSE2"]
+    },
+    "macos-13": {
+        "compiler": ["llvm", "gcc-14"],
+        "arch_flags": ["-march=native", "-march=x86-64"]
+    }
+}
+
+for platform in combinations.keys():
+    for python_version in python_versions:
+        for compiler in combinations[platform]["compiler"]:
+            for arch_flag in combinations[platform]["arch_flags"]:
+                matrix["include"].append({
+                    "os": platform,
+                    "python-version": python_version,
+                    "compiler": compiler,
+                    "arch_flags": arch_flag
+                })
+
+json_str = json.dumps(matrix, ensure_ascii=False)
+print(json_str)
diff --git a/.github/workflows/libvcruntime140.a b/.github/workflows/libvcruntime140.a
deleted file mode 100755
index 3075d72b0..000000000
Binary files a/.github/workflows/libvcruntime140.a and /dev/null differ
diff --git a/.github/workflows/python_build_wheels.yml b/.github/workflows/python_build_wheels.yml
new file mode 100644
index 000000000..4d8a2f549
--- /dev/null
+++ b/.github/workflows/python_build_wheels.yml
@@ -0,0 +1,97 @@
+name: Build and test Python wheels
+
+on: [push, pull_request]
+
+jobs:
+  build_wheels_unix:
+    name: Build wheels on ${{ matrix.os }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      matrix:
+        os: [ubuntu-latest, macos-13]
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Build wheels
+        uses: pypa/cibuildwheel@v2.19.2
+        with:
+          package-dir: 'python/finufft'
+        env:
+          CIBW_BEFORE_ALL_MACOS: brew install gcc@14 fftw
+          CIBW_ARCHS_MACOS: "x86_64"
+          # Need following versions of GCC for compatibility with fftw
+          # installed by homebrew. Similarly, we set the macOS version
+          # for compatibility with those libraries.
+          CIBW_ENVIRONMENT_MACOS: >
+            CC=gcc-14
+            CXX=g++-14
+            MACOSX_DEPLOYMENT_TARGET=13
+
+      - uses: actions/upload-artifact@v4
+        with:
+          name: cibw-wheels-${{ matrix.os }}
+          path: ./wheelhouse/*.whl
+
+  build_wheels_macos_arm64:
+    name: Build wheels on macos-14
+    runs-on: macos-14
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Build wheels
+        uses: pypa/cibuildwheel@v2.19.2
+        with:
+          package-dir: 'python/finufft'
+        env:
+          CIBW_ARCHS_MACOS: "arm64"
+          # Make sure to install the ARM64-specific versions of FFTW and GCC.
+          # Perhaps this is done automatically on the macos-14 image. We should
+          # look into this further.
+          CIBW_BEFORE_ALL_MACOS: |
+            pkg=$(brew fetch --force --bottle-tag=arm64_ventura fftw | grep 'Downloaded to' | cut -d' ' -f3)
+            brew install $pkg
+            pkg=$(brew fetch --force --bottle-tag=arm64_ventura gcc | grep 'Downloaded to' | cut -d' ' -f3)
+            brew install $pkg
+          CIBW_ENVIRONMENT_MACOS: >
+            CC=gcc-14
+            CXX=g++-14
+            MACOSX_DEPLOYMENT_TARGET=14
+
+      - uses: actions/upload-artifact@v4
+        with:
+          name: cibw-wheels-macos-arm64
+          path: ./wheelhouse/*.whl
+
+  build_wheels_win:
+    name: Build wheels on windows
+    runs-on: windows-latest
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install dependencies
+        run: |
+          # Here we install the mingw64 versions of gcc and FFTW that we will
+          # use to compile the library. We also need pkg-config so that cmake
+          # can easily find FFTW when configurating the build.
+          c:\msys64\usr\bin\pacman.exe -Sy --noconfirm mingw-w64-x86_64-gcc mingw-w64-x86_64-fftw mingw-w64-x86_64-pkgconf
+          # This particular install of mingw64 *is not* in the path by default
+          # (another one at c:\mingw64 is, however), so we add it to the path.
+          echo "c:\msys64\mingw64\bin;" >> $env:GITHUB_PATH
+
+      - name: Build wheels
+        uses: pypa/cibuildwheel@v2.19.2
+        with:
+          package-dir: 'python/finufft'
+        env:
+          # This is required to force cmake to avoid using MSVC (the default).
+          # By setting the generator to Ninja, cmake will pick gcc (mingw64)
+          # as the compiler.
+          CIBW_CONFIG_SETTINGS: "cmake.args='-G Ninja'"
+
+      - uses: actions/upload-artifact@v4
+        with:
+          name: cibw-wheels-windows
+          path: ./wheelhouse/*.whl
diff --git a/.github/workflows/python_build_win.ps1 b/.github/workflows/python_build_win.ps1
deleted file mode 100644
index 9e59f342b..000000000
--- a/.github/workflows/python_build_win.ps1
+++ /dev/null
@@ -1,54 +0,0 @@
-$ErrorActionPreference = "Stop"
-Set-Variable -Name PYTHON -Value (Get-Command python).definition
-Set-Variable -Name MSYSTEM -Value MINGW64
-
-# setup setup.cfg
-New-Item -Force -Path .\python\finufft -Name "setup.cfg" -ItemType "file" -Value "[build]`r`ncompiler=mingw32`r`n[build_ext]`r`ncompiler=mingw32"
-
-# Setup the make.inc file
-Copy-Item -Path make.inc.windows_mingw -Destination make.inc
-Add-Content -Path make.inc -Value "PYTHON=""$PYTHON"""
-Add-Content -Path make.inc -Value "FFLAGS+= -fallow-argument-mismatch -march=x86-64"
-Add-Content -Path make.inc -Value "CFLAGS+= -march=x86-64"
-Add-Content -Path make.inc -Value "CXXFLAGS+= -march=x86-64"
-
-Set-Variable libvcruntime140_a -Value ([IO.Path]::Combine((Split-Path -Path $PYTHON), "libs", 'libvcruntime140.a'))
-Copy-Item -Path .\.github\workflows\libvcruntime140.a -Destination $libvcruntime140_a
-
-python -m pip install --upgrade setuptools==70.1.1 wheel numpy pip
-if (-not $?) {throw "Failed pip install"}
-
-# mingw gcc compiler pacth to work with python
-Set-Variable cygwinccompiler_py -Value ([IO.Path]::Combine((Split-Path -Path $PYTHON), "lib", "site-packages", "setuptools", "_distutils", "cygwinccompiler.py"))
-Remove-Item -Path $cygwinccompiler_py -Force
-Copy-Item -Path .\.github\workflows\cygwinccompiler.py -Destination $cygwinccompiler_py
-
-# Setup the distutils.cfg file
-Set-Variable distutils_cfg -Value ([IO.Path]::Combine((Split-Path -Path $PYTHON), "lib", "site-packages", "setuptools", "_distutils", "distutils.cfg"))
-Set-Content -Path $distutils_cfg -Value "[build]`r`ncompiler=mingw32`r`n[build_ext]`r`ncompiler=mingw32"
-
-# call make
-Set-Variable repo_root -Value ([IO.Path]::Combine($PSScriptRoot, '..', '..'))
-c:\msys64\usr\bin\env MSYSTEM=MINGW64 c:\msys64\usr\bin\bash.exe -lc "cd '$repo_root' && make python-dist"
-if (-not $?) {throw "Failed make python-dist"}
-
-# Move the required DLLs inside the wheel
-wheel.exe unpack (get-item .\python\finufft\wheelhouse\finufft*.whl).FullName -d .\tpm
-if (-not $?) {throw "Failed unpack wheel"}
-Set-Variable unpacked_wheel -Value (get-item .\tpm\finufft-*).FullName
-Copy-Item -Path .\lib\libfinufft.dll -Destination ([IO.Path]::Combine($unpacked_wheel, 'finufft'))
-Copy-Item -Path C:\msys64\mingw64\bin\libstdc++-*.dll -Destination ([IO.Path]::Combine($unpacked_wheel, 'finufft'))
-Copy-Item -Path C:\msys64\mingw64\bin\libgcc_s_seh-*.dll -Destination ([IO.Path]::Combine($unpacked_wheel, 'finufft'))
-Copy-Item -Path C:\msys64\mingw64\bin\libgomp-*.dll -Destination ([IO.Path]::Combine($unpacked_wheel, 'finufft'))
-Copy-Item -Path C:\msys64\mingw64\bin\libwinpthread-*.dll -Destination ([IO.Path]::Combine($unpacked_wheel, 'finufft'))
-Copy-Item -Path C:\msys64\mingw64\bin\libfftw3-*.dll -Destination ([IO.Path]::Combine($unpacked_wheel, 'finufft'))
-Copy-Item -Path C:\msys64\mingw64\bin\libfftw3f-*.dll -Destination ([IO.Path]::Combine($unpacked_wheel, 'finufft'))
-Copy-Item -Path C:\msys64\mingw64\bin\libfftw3_omp-*.dll -Destination ([IO.Path]::Combine($unpacked_wheel, 'finufft'))
-Copy-Item -Path C:\msys64\mingw64\bin\libfftw3f_omp-*.dll -Destination ([IO.Path]::Combine($unpacked_wheel, 'finufft'))
-New-Item -Path .\wheelhouse -ItemType Directory -Force
-wheel.exe pack $unpacked_wheel -d .\wheelhouse
-if (-not $?) {throw "Failed pack wheel"}
-
-# Cleanup
-Remove-Item -Path .\python\finufft\wheelhouse -Force -Recurse
-Remove-Item -Path $unpacked_wheel -Force -Recurse
diff --git a/.github/workflows/python_cmake.yml b/.github/workflows/python_cmake.yml
new file mode 100644
index 000000000..3079f852a
--- /dev/null
+++ b/.github/workflows/python_cmake.yml
@@ -0,0 +1,63 @@
+name: Test python skbuild with CMake
+
+on: [push, pull_request]
+
+jobs:
+  prepare:
+    runs-on: ubuntu-latest
+    outputs:
+      matrix: ${{ steps.generate_matrix.outputs.matrix }}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+      - name: Generate matrix
+        id: generate_matrix
+        run: |
+          echo "MACOSX_DEPLOYMENT_TARGET=11.0" >> $GITHUB_ENV 
+          MATRIX=$(python3 ${{ github.workspace }}/.github/workflows/generate_matrix.py)
+          echo "matrix=$MATRIX" >> $GITHUB_OUTPUT
+  build:
+    name: Build with Pip
+    runs-on: ${{ matrix.os }}
+    needs: prepare
+    strategy:
+      fail-fast: false
+      matrix: ${{ fromJSON(needs.prepare.outputs.matrix) }}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+      - name: Setup Cpp
+        uses: aminya/setup-cpp@v1
+        with:
+          compiler: ${{ matrix.compiler }}
+          vcvarsall: ${{ contains(matrix.os, 'windows') }}
+          cmake: false
+          ninja: false
+          vcpkg: false
+          cppcheck: false
+          clangtidy: false
+      - name: Set min macOS version and install fftw
+        if: runner.os == 'macOS'
+        run: |
+          brew install fftw
+      - name: Install fftw
+        if: runner.os == 'linux'
+        run: |
+          sudo apt update
+          sudo apt install -y libfftw3-dev
+      - name: Setup Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+      - name: Install pytest
+        run: |
+          python3 -m pip install --upgrade pip
+          python3 -m pip install pytest
+      - name: Set compiler flags
+        run: |
+          echo CMAKE_ARGS="-DFINUFFT_ARCH_FLAGS=${{ matrix.arch_flags }}" >> $GITHUB_ENV
+        shell: bash
+      - name: Build
+        run: python3 -m pip install ${{ github.workspace }}/python/finufft --verbose
+      - name: Test
+        run: python3 -m pytest ${{ github.workspace }}/python/finufft/test
diff --git a/.github/workflows/python_test_win.ps1 b/.github/workflows/python_test_win.ps1
deleted file mode 100644
index 949a6906d..000000000
--- a/.github/workflows/python_test_win.ps1
+++ /dev/null
@@ -1,9 +0,0 @@
-python -m pip install --pre finufft -f .\wheelhouse\
-if (-not $?) {throw "Failed to pip install finufft"}
-python python/finufft/test/run_accuracy_tests.py
-if (-not $?) {throw "Tests failed"}
-python python/finufft/examples/simple1d1.py
-if (-not $?) {throw "Simple1d1 test failed"}
-python -m pip install pytest
-python -m pytest python/finufft/test
-if (-not $?) {throw "Pytest suite failed"}
diff --git a/.github/workflows/python_wheel.yml b/.github/workflows/python_wheel.yml
deleted file mode 100644
index 5f25feca5..000000000
--- a/.github/workflows/python_wheel.yml
+++ /dev/null
@@ -1,242 +0,0 @@
-name: Python Wheel Build
-
-on:
-  push:
-    branches:
-      - master
-    tags:
-      - v*
-  pull_request:
-    branches:
-      - master
-
-jobs:
-  Linux:
-    runs-on: ubuntu-latest
-    container: quay.io/pypa/manylinux_2_28_x86_64:latest
-
-    steps:
-    - uses: actions/checkout@v4
-
-    - name: Install fftw
-      run: |
-        yum install -y fftw3-devel
-
-    - name: Install ffi
-      run: |
-        yum install -y libffi-devel
-
-    - name: Compile python bindings
-      run: |
-        tools/finufft/build-wheels-linux.sh
-
-    - name: Upload wheels
-      uses: actions/upload-artifact@v4
-      with:
-        name: linux-wheels
-        path: python/finufft/wheelhouse/finufft*manylinux*.whl
-
-  MacOS:
-    runs-on: macos-13
-    env:
-      MACOSX_DEPLOYMENT_TARGET: 10.15
-
-    steps:
-    - uses: actions/checkout@v4
-
-    - name: Unlink gcc
-      run: |
-        brew unlink gcc
-      continue-on-error: true
-
-    - name: Install gcc@13 and fftw
-      run: |
-        brew install gcc@13 fftw
-        cp make.inc.macosx_gcc-12 make.inc
-        echo "FC=gfortran-13" >> make.inc
-        echo "CC=gcc-13" >> make.inc
-        echo "CXX=g++-13" >> make.inc
-        echo "FFLAGS += -march=x86-64" >> make.inc
-        echo "CFLAGS += -march=x86-64" >> make.inc
-        echo "CXXFLAGS += -march=x86-64" >> make.inc
-        # link statically to libgcc, libgfortran and libquadmath
-        # otherwise binaries are incompatible with older systems
-        echo "LIBS += -static-libgfortran -static-libgcc -static-libstdc++" >> make.inc
-        # hack to make libquadmath link statically
-        sudo rm /usr/local/opt/gcc@13/lib/gcc/13/libquadmath.*dylib
-
-    # Download and install Python instead of using the setup_python
-    # as the python interpreters in the Github machines
-    # were compiled in 10.14, the wheels built with them
-    # are incompatible with older MacOS versions
-    - if: steps.cache-python.outputs.cache-hit != 'true'
-      name: Download and install Python
-      run: |
-        curl \
-          https://www.python.org/ftp/python/3.6.8/python-3.6.8-macosx10.9.pkg \
-          --output python_installer.pkg
-        sudo installer -pkg python_installer.pkg -target /
-
-        curl \
-          https://www.python.org/ftp/python/3.7.9/python-3.7.9-macosx10.9.pkg \
-          --output python_installer.pkg
-        sudo installer -pkg python_installer.pkg -target /
-
-        curl \
-          https://www.python.org/ftp/python/3.8.10/python-3.8.10-macosx10.9.pkg \
-          --output python_installer.pkg
-        sudo installer -pkg python_installer.pkg -target /
-
-        curl \
-          https://www.python.org/ftp/python/3.9.13/python-3.9.13-macos11.pkg \
-          --output python_installer.pkg
-        sudo installer -pkg python_installer.pkg -target /
-
-        curl \
-          https://www.python.org/ftp/python/3.10.11/python-3.10.11-macos11.pkg \
-          --output python_installer.pkg
-        sudo installer -pkg python_installer.pkg -target /
-
-        curl \
-          https://www.python.org/ftp/python/3.11.7/python-3.11.7-macos11.pkg \
-          --output python_installer.pkg
-        sudo installer -pkg python_installer.pkg -target /
-
-        curl \
-          https://www.python.org/ftp/python/3.12.1/python-3.12.1-macos11.pkg \
-          --output python_installer.pkg
-        sudo installer -pkg python_installer.pkg -target /
-
-    - name: Compile python bindings
-      run: |
-        make lib
-        export FINUFFT_DIR=`pwd`
-        export CC=gcc-13
-        export CXX=g++-13
-        /Library/Frameworks/Python.framework/Versions/3.6/bin/python3 -m pip install --upgrade setuptools wheel numpy pip
-        /Library/Frameworks/Python.framework/Versions/3.6/bin/python3 -m pip install -U wheel --user
-        /Library/Frameworks/Python.framework/Versions/3.6/bin/python3 -m pip wheel python/finufft -w wheelhouse
-        /Library/Frameworks/Python.framework/Versions/3.7/bin/python3 -m pip install --upgrade setuptools wheel numpy pip
-        /Library/Frameworks/Python.framework/Versions/3.7/bin/python3 -m pip install -U wheel --user
-        /Library/Frameworks/Python.framework/Versions/3.7/bin/python3 -m pip wheel python/finufft -w wheelhouse
-        /Library/Frameworks/Python.framework/Versions/3.8/bin/python3 -m pip install --upgrade setuptools wheel numpy pip
-        /Library/Frameworks/Python.framework/Versions/3.8/bin/python3 -m pip install -U wheel --user
-        /Library/Frameworks/Python.framework/Versions/3.8/bin/python3 -m pip wheel python/finufft -w wheelhouse
-        /Library/Frameworks/Python.framework/Versions/3.9/bin/python3 -m pip install --upgrade setuptools wheel numpy pip
-        /Library/Frameworks/Python.framework/Versions/3.9/bin/python3 -m pip install -U wheel --user
-        /Library/Frameworks/Python.framework/Versions/3.9/bin/python3 -m pip wheel python/finufft -w wheelhouse
-        /Library/Frameworks/Python.framework/Versions/3.10/bin/python3 -m pip install --upgrade setuptools wheel numpy pip
-        /Library/Frameworks/Python.framework/Versions/3.10/bin/python3 -m pip install -U wheel --user
-        /Library/Frameworks/Python.framework/Versions/3.10/bin/python3 -m pip wheel python/finufft -w wheelhouse
-        /Library/Frameworks/Python.framework/Versions/3.11/bin/python3 -m pip install --upgrade setuptools wheel numpy pip
-        /Library/Frameworks/Python.framework/Versions/3.11/bin/python3 -m pip install -U wheel --user
-        /Library/Frameworks/Python.framework/Versions/3.11/bin/python3 -m pip wheel python/finufft -w wheelhouse
-        /Library/Frameworks/Python.framework/Versions/3.12/bin/python3 -m pip install --upgrade setuptools wheel numpy pip
-        /Library/Frameworks/Python.framework/Versions/3.12/bin/python3 -m pip install -U wheel --user
-        /Library/Frameworks/Python.framework/Versions/3.12/bin/python3 -m pip wheel python/finufft -w wheelhouse
-
-        PYTHON_BIN=/Library/Frameworks/Python.framework/Versions/3.12/bin/
-        $PYTHON_BIN/python3 -m pip install delocate==0.10.7
-        ls wheelhouse/finufft*.whl | xargs -n1 $PYTHON_BIN/delocate-wheel -w fixed_wheel/
-        /Library/Frameworks/Python.framework/Versions/3.6/bin/python3 -m pip install --pre finufft -f fixed_wheel/
-        /Library/Frameworks/Python.framework/Versions/3.6/bin/python3 python/finufft/test/run_accuracy_tests.py
-        /Library/Frameworks/Python.framework/Versions/3.6/bin/python3 python/finufft/examples/simple1d1.py
-        /Library/Frameworks/Python.framework/Versions/3.6/bin/python3 -m pip install pytest
-        /Library/Frameworks/Python.framework/Versions/3.6/bin/python3 -m pytest python/finufft/test
-        /Library/Frameworks/Python.framework/Versions/3.7/bin/python3 -m pip install --pre finufft -f fixed_wheel/
-        /Library/Frameworks/Python.framework/Versions/3.7/bin/python3 python/finufft/test/run_accuracy_tests.py
-        /Library/Frameworks/Python.framework/Versions/3.7/bin/python3 python/finufft/examples/simple1d1.py
-        /Library/Frameworks/Python.framework/Versions/3.7/bin/python3 -m pip install pytest
-        /Library/Frameworks/Python.framework/Versions/3.7/bin/python3 -m pytest python/finufft/test
-        /Library/Frameworks/Python.framework/Versions/3.8/bin/python3 -m pip install --pre finufft -f fixed_wheel/
-        /Library/Frameworks/Python.framework/Versions/3.8/bin/python3 python/finufft/test/run_accuracy_tests.py
-        /Library/Frameworks/Python.framework/Versions/3.8/bin/python3 python/finufft/examples/simple1d1.py
-        /Library/Frameworks/Python.framework/Versions/3.8/bin/python3 -m pip install pytest
-        /Library/Frameworks/Python.framework/Versions/3.8/bin/python3 -m pytest python/finufft/test
-        /Library/Frameworks/Python.framework/Versions/3.9/bin/python3 -m pip install --pre finufft -f fixed_wheel/
-        /Library/Frameworks/Python.framework/Versions/3.9/bin/python3 python/finufft/test/run_accuracy_tests.py
-        /Library/Frameworks/Python.framework/Versions/3.9/bin/python3 python/finufft/examples/simple1d1.py
-        /Library/Frameworks/Python.framework/Versions/3.9/bin/python3 -m pip install pytest
-        /Library/Frameworks/Python.framework/Versions/3.9/bin/python3 -m pytest python/finufft/test
-        /Library/Frameworks/Python.framework/Versions/3.10/bin/python3 -m pip install --pre finufft -f fixed_wheel/
-        /Library/Frameworks/Python.framework/Versions/3.10/bin/python3 python/finufft/test/run_accuracy_tests.py
-        /Library/Frameworks/Python.framework/Versions/3.10/bin/python3 python/finufft/examples/simple1d1.py
-        /Library/Frameworks/Python.framework/Versions/3.10/bin/python3 -m pip install pytest
-        /Library/Frameworks/Python.framework/Versions/3.10/bin/python3 -m pytest python/finufft/test
-        /Library/Frameworks/Python.framework/Versions/3.11/bin/python3 -m pip install --pre finufft -f fixed_wheel/
-        /Library/Frameworks/Python.framework/Versions/3.11/bin/python3 python/finufft/test/run_accuracy_tests.py
-        /Library/Frameworks/Python.framework/Versions/3.11/bin/python3 python/finufft/examples/simple1d1.py
-        /Library/Frameworks/Python.framework/Versions/3.11/bin/python3 -m pip install pytest
-        /Library/Frameworks/Python.framework/Versions/3.11/bin/python3 -m pytest python/finufft/test
-        /Library/Frameworks/Python.framework/Versions/3.12/bin/python3 -m pip install --pre finufft -f fixed_wheel/
-        /Library/Frameworks/Python.framework/Versions/3.12/bin/python3 python/finufft/test/run_accuracy_tests.py
-        /Library/Frameworks/Python.framework/Versions/3.12/bin/python3 python/finufft/examples/simple1d1.py
-        /Library/Frameworks/Python.framework/Versions/3.12/bin/python3 -m pip install pytest
-        /Library/Frameworks/Python.framework/Versions/3.12/bin/python3 -m pytest python/finufft/test
-
-
-    - name: Upload wheels
-      uses: actions/upload-artifact@v4
-      with:
-        name: macos-wheels
-        path: fixed_wheel/*.whl
-
-  Windows:
-    runs-on: windows-latest
-
-    steps:
-    - uses: actions/checkout@v4
-
-    - name: Install GCC and make
-      run: C:\msys64\usr\bin\bash.exe -lc "pacman -Sy --noconfirm make mingw-w64-x86_64-toolchain mingw-w64-x86_64-fftw git"
-
-    - name: Build and Test Python 3.8
-      uses: actions/setup-python@v5
-      with:
-        python-version: '3.8'
-        architecture: 'x64'
-    - run: |
-        .\.github\workflows\python_build_win.ps1
-        .\.github\workflows\python_test_win.ps1
-
-    - name: Build and Test Python 3.9
-      uses: actions/setup-python@v5
-      with:
-        python-version: '3.9'
-        architecture: 'x64'
-    - run: |
-        .\.github\workflows\python_build_win.ps1
-        .\.github\workflows\python_test_win.ps1
-
-    - name: Build and Test Python 3.10
-      uses: actions/setup-python@v5
-      with:
-        python-version: '3.10'
-        architecture: 'x64'
-    - run: |
-        .\.github\workflows\python_build_win.ps1
-        .\.github\workflows\python_test_win.ps1
-
-    - name: Build and Test Python 3.11
-      uses: actions/setup-python@v5
-      with:
-        python-version: '3.11'
-        architecture: 'x64'
-    - run: |
-        .\.github\workflows\python_build_win.ps1
-        .\.github\workflows\python_test_win.ps1
-
-    - name: Build and Test Python 3.12
-      uses: actions/setup-python@v5
-      with:
-        python-version: '3.12'
-        architecture: 'x64'
-    - run: |
-        .\.github\workflows\python_build_win.ps1
-        .\.github\workflows\python_test_win.ps1
-
-    - name: Upload wheels
-      uses: actions/upload-artifact@v4
-      with:
-        name: windows-wheels
-        path: wheelhouse\*.whl
diff --git a/.github/workflows/python_wheel_macos_arm64.yml b/.github/workflows/python_wheel_macos_arm64.yml
deleted file mode 100644
index 4bde22f3e..000000000
--- a/.github/workflows/python_wheel_macos_arm64.yml
+++ /dev/null
@@ -1,54 +0,0 @@
-name: Python Wheel Build MacOS Arm64
-
-on:
-  push:
-    branches:
-      - master
-    tags:
-      - v*
-  pull_request:
-    branches:
-      - master
-
-jobs:
-  MacOS:
-    runs-on: macos-13
-    env:
-      MACOSX_DEPLOYMENT_TARGET: 11.0
-
-    steps:
-    - uses: actions/checkout@v4
-
-    - name: Install libomp and fftw
-      run: |
-        pkg=$(brew fetch --force --bottle-tag=arm64_ventura fftw | grep 'Downloaded to' | cut -d' ' -f3)
-        brew install $pkg
-        pkg=$(brew fetch --force --bottle-tag=arm64_ventura libomp | grep 'Downloaded to' | cut -d' ' -f3)
-        brew install $pkg
-
-    - name: Compile libfinufft
-      run: |
-        cp make.inc.macosx_arm64 make.inc
-        make lib
-
-    - name: Build wheels
-      uses: pypa/cibuildwheel@v2.17.0
-      env:
-        FINUFFT_DIR: ${{ github.workspace }}
-        CC: Clang
-        CXX: Clang++
-        CIBW_ARCHS_MACOS: arm64
-        CIBW_BUILD: cp38-* cp39-* cp310-* cp311-* cp312-*
-      with:
-        package-dir: ./python/finufft
-        output-dir: wheelhouse
-
-    - name: Upload wheels
-      uses: actions/upload-artifact@v4
-      with:
-        name: macos-arm64-wheels
-        path: wheelhouse/*.whl
-
-    - name: Setup tmate session
-      if: ${{ failure() }}
-      uses: mxschmitt/action-tmate@v3
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 6d600fe3b..8196ee662 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1,6 +1,8 @@
 cmake_minimum_required(VERSION 3.23)
 
-project(finufft VERSION 2.2.0 LANGUAGES C CXX)
+project(FINUFFT VERSION 2.2.0 LANGUAGES C CXX)
+
+set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 include(CheckCXXCompilerFlag)
@@ -100,6 +102,7 @@ option(FINUFFT_BUILD_EXAMPLES "Whether to build the FINUFFT examples" OFF)
 option(FINUFFT_BUILD_TESTS "Whether to build the FINUFFT tests" OFF)
 option(FINUFFT_BUILD_FORTRAN "Whether to build the FINUFFT Fortran examples" OFF)
 option(FINUFFT_BUILD_MATLAB "Whether to build the FINUFFT Matlab interface" OFF)
+option(FINUFFT_BUILD_PYTHON "Whether the Python wrapper should be built." OFF)
 option(FINUFFT_ENABLE_SANITIZERS "Whether to enable sanitizers, only effective for Debug configuration." ON)
 option(FINUFFT_USE_OPENMP "Whether to use OpenMP for parallelization. If disabled, the finufft library will be single threaded. This does not affect the choice of FFTW library." ON)
 option(FINUFFT_USE_CUDA "Whether to build CUDA accelerated FINUFFT library (libcufinufft). This is completely independent of the main FINUFFT library" OFF)
@@ -107,17 +110,10 @@ option(FINUFFT_USE_CPU "Whether to build the ordinary FINUFFT library (libfinuff
 option(FINUFFT_STATIC_LINKING "Whether to link the static FINUFFT library (libfinufft_static)." ON)
 option(FINUFFT_BUILD_DEVEL "Whether to build development executables" OFF)
 option(FINUFFT_USE_DUCC0 "Whether to use DUCC0 (instead of FFTW) for CPU FFTs" OFF)
+option(FINUFFT_BUILD_DEVEL "Whether to build developement executables" OFF)
+option(FINUFFT_ENABLE_INSTALL "Whether to enable installation of FINUFFT library" ON)
 # sphinx tag (don't remove): @cmake_opts_end
 
-# Check if FINUFFT_STATIC_LINKING is true
-if(NOT FINUFFT_STATIC_LINKING)
-    # If true, set FINUFFT_SHARED_LINKING to false
-    set(FINUFFT_SHARED_LINKING TRUE)
-else()
-    # If not true, set FINUFFT_SHARED_LINKING to true
-    set(FINUFFT_SHARED_LINKING FALSE)
-endif()
-
 if (FINUFFT_USE_CPU)
     # suppress Windows warnings about "unsafe" functions
     if (WIN32)
@@ -170,7 +166,7 @@ if (FINUFFT_BUILD_MATLAB)
 else ()
     # For non-matlab builds, find system OpenMP
     if (FINUFFT_USE_OPENMP)
-        find_package(OpenMP REQUIRED)
+        find_package(OpenMP COMPONENTS CXX REQUIRED)
     endif ()
 endif ()
 
@@ -192,6 +188,7 @@ endfunction()
 
 # Utility function to link static/dynamic lib
 function(finufft_link_test target)
+
     if (FINUFFT_USE_DUCC0)
         target_compile_definitions(${target} PRIVATE FINUFFT_USE_DUCC0)
     endif ()
@@ -308,30 +305,62 @@ if (FINUFFT_BUILD_DEVEL)
     add_subdirectory(devel)
 endif ()
 
-include(GNUInstallDirs)
-install(TARGETS ${INSTALL_TARGETS} PUBLIC_HEADER)
-install(FILES ${PROJECT_SOURCE_DIR}/LICENSE
-        DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/licenses/finufft)
+if (FINUFFT_BUILD_PYTHON)
+    add_subdirectory(python)
+endif ()
+
+message(STATUS " CMAKE_BUILD_TYPE: ${CMAKE_BUILD_TYPE}")
+message(STATUS "FINUFFT configuration summary:")
+message(STATUS "  FINUFFT_USE_CPU: ${FINUFFT_USE_CPU}")
+message(STATUS "  FINUFFT_USE_CUDA: ${FINUFFT_USE_CUDA}")
+message(STATUS "  FINUFFT_USE_OPENMP: ${FINUFFT_USE_OPENMP}")
+message(STATUS "  FINUFFT_STATIC_LINKING: ${FINUFFT_STATIC_LINKING}")
+message(STATUS "  FINUFFT_ENABLE_INSTALL: ${FINUFFT_ENABLE_INSTALL}")
+message(STATUS "  FINUFFT_BUILD_EXAMPLES: ${FINUFFT_BUILD_EXAMPLES}")
+message(STATUS "  FINUFFT_BUILD_TESTS: ${FINUFFT_BUILD_TESTS}")
+message(STATUS "  FINUFFT_BUILD_FORTRAN: ${FINUFFT_BUILD_FORTRAN}")
+message(STATUS "  FINUFFT_BUILD_MATLAB: ${FINUFFT_BUILD_MATLAB}")
+message(STATUS "  FINUFFT_BUILD_PYTHON: ${FINUFFT_BUILD_PYTHON}")
+message(STATUS "  FINUFFT_ENABLE_SANITIZERS: ${FINUFFT_ENABLE_SANITIZERS}")
+message(STATUS "  FINUFFT_FFTW_SUFFIX: ${FINUFFT_FFTW_SUFFIX}")
+message(STATUS "  FINUFFT_FFTW_LIBRARIES: ${FINUFFT_FFTW_LIBRARIES}")
+message(STATUS "  FINUFFT_ARCH_FLAGS: ${FINUFFT_ARCH_FLAGS}")
+
 if (FINUFFT_USE_CPU)
-    install(DIRECTORY ${PROJECT_SOURCE_DIR}/examples
-            DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/finufft
-            PATTERN "CMakeLists.txt" EXCLUDE
-            PATTERN "README" EXCLUDE
-            PATTERN "examples/cuda" EXCLUDE
-    )
-    if (FINUFFT_BUILD_FORTRAN)
-        install(DIRECTORY ${PROJECT_SOURCE_DIR}/fortran/examples
-                DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/finufft/fortran
-        )
-        install(FILES ${PROJECT_SOURCE_DIR}/include/finufft.fh
-                DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}
-        )
+    if (FINUFFT_STATIC_LINKING)
+        get_target_property(FINUFFT_COMPILE_OPTIONS finufft_static COMPILE_OPTIONS)
+    else ()
+        get_target_property(FINUFFT_COMPILE_OPTIONS finufft COMPILE_OPTIONS)
     endif ()
+    message("Compile options for finufft: ${MY_COMPILE_OPTIONS}")
 endif ()
-if (FINUFFT_USE_CUDA)
-    install(DIRECTORY ${PROJECT_SOURCE_DIR}/examples/cuda
-            DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/finufft/examples
-            PATTERN "README" EXCLUDE
-            PATTERN "CMakeLists.txt" EXCLUDE
-    )
+
+if (FINUFFT_ENABLE_INSTALL)
+    include(GNUInstallDirs)
+    install(TARGETS ${INSTALL_TARGETS} PUBLIC_HEADER)
+    install(FILES ${PROJECT_SOURCE_DIR}/LICENSE
+            DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/licenses/finufft)
+    if (FINUFFT_USE_CPU)
+        install(DIRECTORY ${PROJECT_SOURCE_DIR}/examples
+                DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/finufft
+                PATTERN "CMakeLists.txt" EXCLUDE
+                PATTERN "README" EXCLUDE
+                PATTERN "examples/cuda" EXCLUDE
+        )
+        if (FINUFFT_BUILD_FORTRAN)
+            install(DIRECTORY ${PROJECT_SOURCE_DIR}/fortran/examples
+                    DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/finufft/fortran
+            )
+            install(FILES ${PROJECT_SOURCE_DIR}/include/finufft.fh
+                    DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}
+            )
+        endif()
+    endif ()
+    if (FINUFFT_USE_CUDA)
+        install(DIRECTORY ${PROJECT_SOURCE_DIR}/examples/cuda
+                DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/finufft/examples
+                PATTERN "README" EXCLUDE
+                PATTERN "CMakeLists.txt" EXCLUDE
+        )
+    endif()
 endif ()
diff --git a/CMakePresets.json b/CMakePresets.json
index b04204500..32229d1c5 100644
--- a/CMakePresets.json
+++ b/CMakePresets.json
@@ -110,6 +110,18 @@
         "FINUFFT_BUILD_MATLAB": "ON",
         "FINUFFT_ENABLE_SANITIZERS": "OFF"
       }
+    },
+    {
+      "name": "python",
+      "binaryDir": "build/python",
+      "displayName": "python",
+      "description": "Build with the python interface",
+      "generator": "Ninja Multi-Config",
+      "cacheVariables": {
+        "FINUFFT_BUILD_PYTHON": "ON",
+        "FINUFFT_ENABLE_SANITIZERS": "OFF",
+        "CMAKE_INSTALL_DATAROOTDIR": "."
+      }
     }
   ],
   "buildPresets": [
diff --git a/docs/install.rst b/docs/install.rst
index 05d4f6f65..bb0bd22cc 100644
--- a/docs/install.rst
+++ b/docs/install.rst
@@ -59,14 +59,14 @@ The basic quick download, building, and test is then:
   cd finufft
   mkdir build
   cd build
-  cmake .. -D FINUFFT_BUILD_TESTS=ON -D CMAKE_BUILD_TYPE=Release --install-prefix /path/to/install
+  cmake .. -D FINUFFT_BUILD_TESTS=ON --install-prefix /path/to/install
   cmake --build . -j
   ctest
   cmake --install .
 
+.. note::
 
-Note: `Release` here is essential to get the best performing version. Also, if you don't supply `--install-prefix`, it will default to ``/usr/local`` on most systems. If you don't
-have root access, you must supply a prefix you can write to such as ``$HOME/local``. Now...
+   If you don't supply `--install-prefix`, it will default to ``/usr/local`` on most systems. If you don't have root access, you must supply a prefix you can write to such as ``$HOME/local``.
 
 In ``build``, this creates ``libfinufft_static.a`` and ``libfinufft.so``, and runs a test that should take a
 few seconds and report ``100% tests passed, 0 tests failed out of 17``.  To use the library, link against
diff --git a/python/CMakeLists.txt b/python/CMakeLists.txt
new file mode 100644
index 000000000..e1eb698fb
--- /dev/null
+++ b/python/CMakeLists.txt
@@ -0,0 +1,11 @@
+if(FINUFFT_USE_CPU)
+    if (WIN32)
+        install(TARGETS finufft LIBRARY DESTINATION finufft RUNTIME DESTINATION finufft)
+    else ()
+        install(TARGETS finufft LIBRARY DESTINATION finufft)
+    endif ()
+endif()
+
+if(FINUFFT_USE_GPU)
+    install(TARGETS finufft_gpu LIBRARY DESTINATION cufinufft)
+endif()
diff --git a/python/finufft/finufft/_finufft.py b/python/finufft/finufft/_finufft.py
index 96308db11..fe2748d70 100644
--- a/python/finufft/finufft/_finufft.py
+++ b/python/finufft/finufft/_finufft.py
@@ -5,20 +5,18 @@
 Seperate bindings are provided for single and double precision libraries,
 differentiated by 'f' suffix.
 """
-
 import ctypes
-import os
-import warnings
-import platform
-import importlib.util
-
-import numpy as np
-
+import pathlib
+from ctypes.util import find_library
 from ctypes import c_double
-from ctypes import c_int
 from ctypes import c_float
-from ctypes import c_void_p
+from ctypes import c_int
 from ctypes import c_longlong
+from ctypes import c_void_p
+
+import numpy as np
+import os
+import platform
 from numpy.ctypeslib import ndpointer
 
 c_int_p = ctypes.POINTER(c_int)
@@ -28,32 +26,31 @@
 
 # TODO: See if there is a way to improve this so it is less hacky.
 lib = None
-# Try to load a local library directly.
-try:
-    lib = ctypes.cdll.LoadLibrary('libfinufft.so')
-except OSError:
-    pass
-
-# Should that not work, try to find the full path of a packaged lib.
-#   The packaged lib should have a py/platform decorated name,
-#   and be rpath'ed the true FINUFFT library through the Extension and wheel
-#   systems.
-try:
-    if lib is None:
-        # Find the library.
-        lib_path = importlib.util.find_spec('finufft.finufftc').origin
-        # Get the full path for the ctypes loader.
-        if platform.system() == 'Windows':
-            os.environ["PATH"] += os.pathsep + os.path.join(os.path.dirname(os.path.dirname(os.path.realpath(lib_path))),'finufft')
-            full_lib_path = os.path.join(os.path.dirname(os.path.dirname(os.path.realpath(lib_path))),'finufft','libfinufft.dll')
-        else:
-            full_lib_path = os.path.realpath(lib_path)
-
-        # Load the library,
-        #    which rpaths the libraries we care about.
-        lib = ctypes.cdll.LoadLibrary(full_lib_path)
-except Exception:
-    raise ImportError('Failed to find a suitable finufft library')
+# Try to load finufft installed from the python package.
+path = pathlib.Path(__file__).parent.resolve()
+# Ignoring the exceptions to avoid the print
+# exception, during the process of an exception another exception occurred
+# unix systems have lib prefix, non unix systems do not
+library_names = ['libfinufft', 'finufft']
+for lib_name in library_names:
+    try:
+        lib = np.ctypeslib.load_library(lib_name, path)
+        break
+    except OSError:
+        # Paranoid, in case lib is set to something and then an exception is thrown
+        lib = None
+
+if lib is None:
+    # If that fails, try to load the library from the system path.
+    libname = find_library('finufft')
+    if libname is not None:
+        lib = ctypes.cdll.LoadLibrary(libname)
+        # we probably should add a version check and trow a warning if the version is different
+    else:
+        # if that does not work, finufft is not installed correctly.
+        raise ImportError('Failed to find a suitable finufft library. '
+                          'Please check your installation, '
+                          'finufft does not seem to be installed correctly.')
 
 
 class FinufftOpts(ctypes.Structure):
diff --git a/python/finufft/pyproject.toml b/python/finufft/pyproject.toml
new file mode 100644
index 000000000..f9c74fc75
--- /dev/null
+++ b/python/finufft/pyproject.toml
@@ -0,0 +1,71 @@
+[build-system]
+requires = [
+    "scikit-build-core >= 0.4.3",
+    "cmake >= 3.19",
+    "ninja >= 1.9.0",
+]
+
+build-backend = "scikit_build_core.build"
+
+[project]
+name = "finufft"
+readme = "README.md"
+requires-python = ">=3.8"
+dependencies = ["numpy >= 1.12.0"]
+authors = [
+    {name = "Jeremy Magland"},
+    {name = "Daniel Foreman-Mackey"},
+    {name = "Joakim Anden"},
+    {name = "Libin Lu"},
+    {name = "Alex Barnett"}]
+maintainers = [{name = "Alex Barnett", email = "abarnett@flatironinstitute.org"}]
+description = "Python interface to FINUFFT"
+classifiers = [
+    "License :: OSI Approved :: Apache Software License",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: C++",
+    "Operating System :: POSIX :: Linux",
+    "Operating System :: MacOS :: MacOS X",
+    "Operating System :: Microsoft :: Windows"
+    ]
+dynamic = ["version"]
+
+[tool.scikit-build]
+# Protect the configuration against future changes in scikit-build-core
+minimum-version = "0.4"
+# Setuptools-style build caching in a local directory
+build-dir = "build/{wheel_tag}"
+
+# Tell skbuild to look for the CMakeLists.txt file two directories up.
+cmake.source-dir = "../../"
+cmake.targets = ["finufft"]
+cmake.define = {"FINUFFT_BUILD_PYTHON" = "ON", "FINUFFT_ENABLE_INSTALL" = "OFF"}
+
+wheel.packages = ["finufft"]
+
+# Indicate that we don't depend on the CPython API
+wheel.py-api = "py3"
+
+[tool.scikit-build.metadata.version]
+# Instead of hardcoding the version here, extract it from the source files.
+provider = "scikit_build_core.metadata.regex"
+input = "finufft/__init__.py"
+
+[tool.cibuildwheel]
+# Necessary to see build output from the actual compilation
+build-verbosity = 1
+# Not building for PyPy and musllinux for now.
+skip = "pp* *musllinux*"
+test-requires = "pytest"
+test-command = "pytest {project}/python/finufft/test"
+
+[tool.cibuildwheel.linux]
+archs = "x86_64"
+before-all = "yum install -y fftw3-devel"
+
+[tool.cibuildwheel.windows]
+archs = "AMD64"
+before-build = "pip install delvewheel"
+# CIBW doesn't do vendoring of DLLs on Windows by default, so we have to
+# install delvewheel and run it.
+repair-wheel-command = "delvewheel repair -v --analyze-existing -w {dest_dir} {wheel}"
diff --git a/python/finufft/setup.py b/python/finufft/setup.py
deleted file mode 100644
index c2bf12f8d..000000000
--- a/python/finufft/setup.py
+++ /dev/null
@@ -1,99 +0,0 @@
-# This defines the Python module installation.
-
-# Barnett 3/1/18. Updates by Yu-Hsuan Shih, June 2018.
-# win32 mingw patch by Vineet Bansal, Feb 2019.
-# attempt ../make.inc reading (failed) and default finufftdir. 2/25/20
-# Barnett trying to get sphinx.ext.autodoc to work w/ this, 10/5/20
-
-__version__ = '2.2.0'
-
-from setuptools import setup, Extension
-import os
-import platform
-from pathlib import Path
-
-from tempfile import mkstemp
-
-finufft_dir = os.environ.get('FINUFFT_DIR')
-
-# Note: This will not work if run through pip install since setup.py is copied
-# to a different location.
-if finufft_dir == None or finufft_dir == '':
-    finufft_dir = Path(__file__).resolve().parents[2]
-
-# Set include and library paths relative to FINUFFT root directory.
-inc_dir = os.path.join(finufft_dir, 'include')
-lib_dir = os.path.join(finufft_dir, 'lib')
-lib_dir_cmake = os.path.join(finufft_dir, 'build')   # lib may be only here
-
-# Read in long description from README.md.
-with open(os.path.join(finufft_dir, 'python', 'finufft', 'README.md'), 'r') as f:
-        long_description = f.read()
-
-finufft_dlib = 'finufft'
-
-# Windows does not have the concept of rpath and as a result, MSVC crashes if
-# supplied with one.
-if platform.system() != "Windows":
-    runtime_library_dirs = [lib_dir, lib_dir_cmake]
-else:
-    runtime_library_dirs = []
-
-# For certain platforms (e.g. Ubuntu 20.04), we need to create a dummy source
-# that calls one of the functions in the FINUFFT dynamic library. The reason
-# is that these platforms override the default --no-as-needed flag for ld,
-# which means that the linker will only link to those dynamic libraries for
-# which there are unresolved symbols in the object files. Since we do not have
-# a real source, the result is that no dynamic libraries are linked. To
-# prevent this, we create a dummy source so that the library will link as
-# expected.
-fd, source_filename = mkstemp(suffix='.c', text=True)
-
-with open(fd, 'w') as f:
-    f.write( \
-"""
-#include <finufft.h>
-
-void PyInit_finufftc(void) {
-    finufft_opts opt;
-
-    finufft_default_opts(&opt);
-}
-""")
-
-
-########## SETUP ###########
-setup(
-    name='finufft',
-    version=__version__,
-    author='Python interfaces by: Jeremy Magland, Daniel Foreman-Mackey, Joakim Anden, Libin Lu, and Alex Barnett',
-    author_email='abarnett@flatironinstitute.org',
-    url='https://github.com/flatironinstitute/finufft',
-    description='Python interface to FINUFFT',
-    long_description=long_description,
-    long_description_content_type='text/markdown',
-    license="Apache 2",
-    packages=['finufft'],
-    classifiers=[
-        'License :: OSI Approved :: Apache Software License',
-        'Programming Language :: Python :: 3',
-        'Programming Language :: C++',
-        'Operating System :: POSIX :: Linux',
-        'Operating System :: MacOS :: MacOS X',
-        'Operating System :: Microsoft :: Windows',
-    ],
-    install_requires=['numpy>=1.12.0'],
-    python_requires='>=3.6',
-    zip_safe=False,
-    py_modules=['finufft.finufftc'],
-    ext_modules=[
-        Extension(name='finufft.finufftc',
-                  sources=[source_filename],
-                  include_dirs=[inc_dir, '/usr/local/include'],
-                  library_dirs=[lib_dir, lib_dir_cmake, '/usr/local/lib'],
-                  libraries=[finufft_dlib],
-                  runtime_library_dirs=runtime_library_dirs)
-        ]
-)
-
-os.unlink(source_filename)
diff --git a/src/spreadinterp.cpp b/src/spreadinterp.cpp
index 4374c6e9d..3f03c79bb 100644
--- a/src/spreadinterp.cpp
+++ b/src/spreadinterp.cpp
@@ -24,6 +24,8 @@ namespace { // anonymous namespace for internal structs equivalent to declaring
             // static
 struct zip_low;
 struct zip_hi;
+template<unsigned cap> struct reverse_index;
+template<unsigned cap> struct shuffle_index;
 struct select_even;
 struct select_odd;
 // forward declaration to clean up the code and be able to use this everywhere in the file
@@ -777,23 +779,80 @@ Two upsampfacs implemented. Params must match ref formula. Barnett 4/24/18 */
   const FLT z = std::fma(FLT(2.0), x, FLT(w - 1)); // scale so local grid offset z in
                                                    // [-1,1]
   if (opts.upsampfac == 2.0) {                     // floating point equality is fine here
-    static constexpr auto alignment     = simd_type::arch_type::alignment();
+    using arch_t                        = typename simd_type::arch_type;
+    static constexpr auto alignment     = arch_t::alignment();
     static constexpr auto simd_size     = simd_type::size;
     static constexpr auto padded_ns     = (w + simd_size - 1) & ~(simd_size - 1);
     static constexpr auto nc            = nc200<w>();
     static constexpr auto horner_coeffs = get_horner_coeffs_200<FLT, w>();
+    static constexpr auto use_ker_sym   = (simd_size < w);
 
     alignas(alignment) static constexpr auto padded_coeffs =
         pad_2D_array_with_zeros<FLT, nc, w, padded_ns>(horner_coeffs);
 
-    const simd_type zv(z);
-    for (uint8_t i = 0; i < w; i += simd_size) {
-      auto k = simd_type::load_aligned(padded_coeffs[0].data() + i);
-      for (uint8_t j = 1; j < nc; ++j) {
-        const auto cji = simd_type::load_aligned(padded_coeffs[j].data() + i);
-        k              = xsimd::fma(k, zv, cji);
+    // use kernel symmetry trick if w > simd_size
+    if constexpr (use_ker_sym) {
+      static constexpr uint8_t tail          = w % simd_size;
+      static constexpr uint8_t if_odd_degree = ((nc + 1) % 2);
+      static constexpr uint8_t offset_start  = tail ? w - tail : w - simd_size;
+      static constexpr uint8_t end_idx       = (w + (tail > 0)) / 2;
+      const simd_type zv{z};
+      const auto z2v = zv * zv;
+
+      // some xsimd constant for shuffle or inverse
+      static constexpr auto shuffle_batch = []() constexpr noexcept {
+        if constexpr (tail) {
+          return xsimd::make_batch_constant<xsimd::as_unsigned_integer_t<FLT>, arch_t,
+                                            shuffle_index<tail>>();
+        } else {
+          return xsimd::make_batch_constant<xsimd::as_unsigned_integer_t<FLT>, arch_t,
+                                            reverse_index<simd_size>>();
+        }
+      }();
+
+      // process simd vecs
+      simd_type k_prev, k_sym{0};
+      for (uint8_t i{0}, offset = offset_start; i < end_idx;
+           i += simd_size, offset -= simd_size) {
+        auto k_odd = [i]() constexpr noexcept {
+          if constexpr (if_odd_degree) {
+            return simd_type::load_aligned(padded_coeffs[0].data() + i);
+          } else {
+            return simd_type{0};
+          }
+        }();
+        auto k_even = simd_type::load_aligned(padded_coeffs[if_odd_degree].data() + i);
+        for (uint8_t j{1 + if_odd_degree}; j < nc; j += 2) {
+          const auto cji_odd  = simd_type::load_aligned(padded_coeffs[j].data() + i);
+          const auto cji_even = simd_type::load_aligned(padded_coeffs[j + 1].data() + i);
+          k_odd               = xsimd::fma(k_odd, z2v, cji_odd);
+          k_even              = xsimd::fma(k_even, z2v, cji_even);
+        }
+        // left part
+        xsimd::fma(k_odd, zv, k_even).store_aligned(ker + i);
+        // right part symmetric to the left part
+        if (offset >= end_idx) {
+          if constexpr (tail) {
+            // to use aligned store, we need shuffle the previous k_sym and current k_sym
+            k_prev = k_sym;
+            k_sym  = xsimd::fnma(k_odd, zv, k_even);
+            xsimd::shuffle(k_sym, k_prev, shuffle_batch).store_aligned(ker + offset);
+          } else {
+            xsimd::swizzle(xsimd::fnma(k_odd, zv, k_even), shuffle_batch)
+                .store_aligned(ker + offset);
+          }
+        }
+      }
+    } else {
+      const simd_type zv(z);
+      for (uint8_t i = 0; i < w; i += simd_size) {
+        auto k = simd_type::load_aligned(padded_coeffs[0].data() + i);
+        for (uint8_t j = 1; j < nc; ++j) {
+          const auto cji = simd_type::load_aligned(padded_coeffs[j].data() + i);
+          k              = xsimd::fma(k, zv, cji);
+        }
+        k.store_aligned(ker + i);
       }
-      k.store_aligned(ker + i);
     }
     return;
   }
@@ -2168,6 +2227,16 @@ struct zip_hi {
     return (size + index) / 2;
   }
 };
+template<unsigned cap> struct reverse_index {
+  static constexpr unsigned get(unsigned index, const unsigned size) {
+    return index < cap ? (cap - 1 - index) : index;
+  }
+};
+template<unsigned cap> struct shuffle_index {
+  static constexpr unsigned get(unsigned index, const unsigned size) {
+    return index < cap ? (cap - 1 - index) : size + size + cap - 1 - index;
+  }
+};
 
 struct select_even {
   static constexpr unsigned get(unsigned index, unsigned /*size*/) { return index * 2; }
diff --git a/tools/finufft/build-wheels-linux.sh b/tools/finufft/build-wheels-linux.sh
deleted file mode 100755
index 5ea8d9303..000000000
--- a/tools/finufft/build-wheels-linux.sh
+++ /dev/null
@@ -1,54 +0,0 @@
-#!/bin/bash
-
-set -e -x
-
-# Replace native compilation flags with more generic ones.
-cp make.inc.manylinux make.inc
-
-# Clean up the build and make the library.
-make clean
-make lib
-
-# Test to make sure everything is ok.
-make test
-
-# Needed for pip install to work
-export FINUFFT_DIR=$(pwd)
-# Needed for auditwheel to find the dynamic libraries
-export LD_LIBRARY_PATH=${FINUFFT_DIR}/lib:${LD_LIBRARY_PATH}
-
-# Explicitly list Python versions to build
-versions=("cp36-cp36m"
-          "cp37-cp37m"
-          "cp38-cp38"
-          "cp39-cp39"
-          "cp310-cp310"
-          "cp311-cp311"
-          "cp312-cp312"
-          "pp39-pypy39_pp73")
-
-pys=()
-for version in "${versions[@]}"; do
-    pys+=("/opt/python/${version}/bin")
-done
-
-# build wheel
-for pybin in "${pys[@]}"; do
-    "${pybin}/pip" install --upgrade pip
-    "${pybin}/pip" install auditwheel wheel numpy
-    "${pybin}/pip" wheel ./python/finufft -w python/finufft/wheelhouse
-done
-
-# fix wheel
-for whl in python/finufft/wheelhouse/finufft-*.whl; do
-    auditwheel repair "$whl" -w python/finufft/wheelhouse/
-done
-
-# test wheel
-for pybin in "${pys[@]}"; do
-    "${pybin}/pip" install --pre finufft -f ./python/finufft/wheelhouse/
-    "${pybin}/python" ./python/finufft/test/run_accuracy_tests.py
-    "${pybin}/python" ./python/finufft/examples/simple1d1.py
-    "${pybin}/pip" install pytest
-    "${pybin}/pytest" python/finufft/test
-done