Switch QBLAS dependency to 1.5.0

.github/workflows/build_wheels.yml

@@ -114,7 +114,9 @@ jobs:
           path: ./wheelhouse/*.whl
           name: wheels-${{ matrix.os }}
 
-  # disabling QBLAS optimization for windows due to incompatibility with MSVC
+  # QBLAS is auto-disabled on Windows by meson.build (it uses GCC/POSIX-only
+  # constructs that MSVC does not support); the wheel falls back to the
+  # naive matmul kernel. No CFLAGS hack needed.
   build_wheels_windows:
     name: Build wheels on Windows
     runs-on: windows-latest
@@ -153,9 +155,6 @@ jobs:
           CIBW_BUILD_VERBOSITY: "3"
           DISTUTILS_USE_SDK: "1"
           MSSdk: "1"
-          CIBW_ENVIRONMENT: >
-            CFLAGS="/DDISABLE_QUADBLAS $CFLAGS"
-            CXXFLAGS="/DDISABLE_QUADBLAS $CXXFLAGS"
           CIBW_REPAIR_WHEEL_COMMAND: 'delvewheel repair -w {dest_dir} {wheel} --add-path C:\sleef\bin'
           CIBW_TEST_COMMAND_WINDOWS: pip install numpy && pip install --no-deps {wheel} && pip install pytest pytest-run-parallel && pytest -s {project}/tests
           CIBW_TEST_EXTRAS: test

.gitignore

@@ -146,3 +146,4 @@ compile_commands.json
 
 # docs
 /docs/_build/
+build_log.txt

meson.build

@@ -15,10 +15,7 @@ if is_windows
   add_project_arguments('-DWIN32', '-D_WINDOWS', language : ['c', 'cpp'])
 endif
 
-qblas_dep = dependency('qblas', fallback: ['qblas', 'qblas_dep'])
 
-# Try to find SLEEF system-wide first, fall back to subproject if not found
-# Required SLEEF version (must match sleef.wrap revision)
 required_sleef_version = '3.9.0'
 # Don't use fallback here - we need to call subproject() explicitly later with disable_fma option
 sleef_dep = dependency('sleef', version: '>=' + required_sleef_version, required: false)
@@ -30,7 +27,7 @@ if sleef_dep.found() and sleef_dep.version().startswith(required_sleef_version)
   # SLEEF found system-wide - verify quad-precision support
   cpp = meson.get_compiler('cpp')
   sleefquad_lib = cpp.find_library('sleefquad', required: false)
-  
+
   if sleefquad_lib.found()
     sleefquad_test_code = '''
     #include <sleefquad.h>
@@ -48,7 +45,7 @@ if sleef_dep.found() and sleef_dep.version().startswith(required_sleef_version)
       dependencies: [sleef_dep, sleefquad_lib],
       name: 'SLEEF quad-precision support'
     )
-    
+
     if quad_works
       sleefquad_dep = declare_dependency(
         dependencies: [sleef_dep, sleefquad_lib]
@@ -80,6 +77,23 @@ else
   message('Proceeding with vendored SLEEF subproject instead')
 endif
 
+# QBLAS does not build under MSVC (GCC-only flags, POSIX-only APIs, GCC
+# built-ins for CPUID); force-disable it on Windows. Users on other
+# platforms can opt out via -Ddisable_quadblas=true to fall back to the
+# naive matmul kernel.
+disable_quadblas = is_windows or get_option('disable_quadblas')
+if disable_quadblas
+  if is_windows
+    message('QBLAS disabled (Windows / MSVC) - using naive matmul kernel')
+  else
+    message('QBLAS disabled by user option - using naive matmul kernel')
+  endif
+  add_project_arguments('-DDISABLE_QUADBLAS', language: ['c', 'cpp'])
+  qblas_dep = declare_dependency()
+else
+  qblas_dep = dependency('qblas', fallback: ['qblas', 'qblas_dep'])
+endif
+
 incdir_numpy = run_command(py,
   ['-c', 'import numpy; print(numpy.get_include())'],
   check : true
@@ -101,11 +115,14 @@ npymath_lib = c.find_library('npymath', dirs: npymath_path)
 
 dependencies = [py_dep, qblas_dep, sleef_dep, sleefquad_dep, npymath_lib]
 
-# Add OpenMP dependency (optional, for threading)
-openmp_dep = dependency('openmp', required: false, static: false)
-if openmp_dep.found()
-    dependencies += openmp_dep
-endif
+# OpenMP is not used directly by any source in this package (#pragma omp
+# is absent); it only matters because qblas's static lib has OpenMP
+# objects baked in. qblas_dep already propagates the OpenMP requirement
+# transitively, so adding a second dependency('openmp') here would put
+# OpenMP into the compile-args closure twice. On Apple's clang++ that
+# duplication leaves an orphan '-Xpreprocessor' in $ARGS which then
+# pairs with '-MD' from the dep-gen flags ('-Xpreprocessor -MD'), and
+# the preprocessor rejects -MD as unknown - failing all C++ compiles.
 
 # compiler flags for QBLAS compatibility
 if not is_windows

meson.options

@@ -1,3 +1,9 @@
 option('disable_fma', type: 'boolean', value: false,
        description: 'Disable FMA (Fused Multiply-Add) code paths' +
-                    'Set to true when building for older CPUs like Sandy Bridge that lack FMA support.')
+                    'Set to true when building for older CPUs like Sandy Bridge that lack FMA support.')
+
+option('disable_quadblas', type: 'boolean', value: false,
+       description: 'Skip the QBLAS subproject and fall back to naive ' +
+                    'matmul kernels. Auto-enabled on Windows because ' +
+                    'QBLAS uses GCC/POSIX-only constructs that do not ' +
+                    'build under MSVC.')