source_root = meson.project_source_root()

mod_features = import('features')

compiler_id = meson.get_compiler('c').get_id()

VSX = mod_features.new(

'VSX', 1, args: ['-mvsx', '-DHWY_COMPILE_ONLY_STATIC', '-DHWY_DISABLE_ATTR'] +

(compiler_id == 'clang' ? ['-maltivec'] : []),

test_code: files(source_root + '/numpy/_core/src/_simd/checks/cpu_vsx.c')[0],

extra_tests: {

'VSX_ASM': files(source_root + '/numpy/_core/src/_simd/checks/extra_vsx_asm.c')[0]

}

)

VSX2 = mod_features.new(

'VSX2', 2, implies: VSX, args: {'val': '-mcpu=power8', 'match': '.*vsx'},

detect: {'val': 'VSX2', 'match': 'VSX'},

test_code: files(source_root + '/numpy/_core/src/_simd/checks/cpu_vsx2.c')[0],

)

# VSX2 is hardware baseline feature on ppc64le since the first little-endian

# support was part of Power8

if host_machine.endian() == 'little'

VSX.update(implies: VSX2)

endif

VSX3 = mod_features.new(

'VSX3', 3, implies: VSX2, args: {'val': '-mcpu=power9', 'match': '.*(?:mcpu=|vsx).*'},

detect: {'val': 'VSX3', 'match': 'VSX.*'},

test_code: files(source_root + '/numpy/_core/src/_simd/checks/cpu_vsx3.c')[0],

extra_tests: {

'VSX3_HALF_DOUBLE': files(source_root + '/numpy/_core/src/_simd/checks/extra_vsx3_half_double.c')[0]

}

)

VSX4 = mod_features.new(

'VSX4', 4, implies: VSX3, args: {'val': '-mcpu=power10', 'match': '.*(?:mcpu=|vsx).*'},

detect: {'val': 'VSX4', 'match': 'VSX.*'},

test_code: files(source_root + '/numpy/_core/src/_simd/checks/cpu_vsx4.c')[0],

extra_tests: {

'VSX4_MMA': files(source_root + '/numpy/_core/src/_simd/checks/extra_vsx4_mma.c')[0]

}

)

PPC64_FEATURES = {'VSX': VSX, 'VSX2': VSX2, 'VSX3': VSX3, 'VSX4': VSX4}