Make AVX/FMA support optional

Older compilers do not yet have fmaintrin.h. This fixes compile errors:
```
x86_functions_fma.cc:51:39: error: '_mm_fmadd_ss' was not declared in this scope
x86_functions_fma.cc:85:35: error: '_mm256_fmadd_ps' was not declared in this scope
x86_functions_fma.cc:106:35: error: '_mm256_fmadd_ps' was not declared in this scope
x86_functions_fma.cc:129:39: error: '_mm_fmadd_ss' was not declared in this scope
```

libs/ardour/ardour/mix.h

@@ -46,7 +46,9 @@ extern "C" {
 }
 
 /* FMA functions */
+#ifdef FPU_AVX_FMA_SUPPORT
 LIBARDOUR_API void  x86_fma_mix_buffers_with_gain (float * dst, const float * src, uint32_t nframes, float gain);
+#endif
 
 LIBARDOUR_API void  x86_sse_find_peaks     (const float * buf, uint32_t nsamples, float *min, float *max);
 #ifdef PLATFORM_WINDOWS

libs/ardour/globals.cc

@@ -186,8 +186,9 @@ setup_hardware_optimization (bool try_optimization)
 		FPU* fpu = FPU::instance ();
 
 #if defined(ARCH_X86) && defined(BUILD_SSE_OPTIMIZATIONS)
-
 		/* We have AVX-optimized code for Windows and Linux */
+
+#ifdef FPU_AVX_FMA_SUPPORT
 		if (fpu->has_fma ()) {
 			info << "Using AVX and FMA optimized routines" << endmsg;
 
@@ -201,7 +202,9 @@ setup_hardware_optimization (bool try_optimization)
 
 			generic_mix_functions = false;
 
-		} else if (fpu->has_avx ()) {
+		} else
+#endif
+		if (fpu->has_avx ()) {
 			info << "Using AVX optimized routines" << endmsg;
 
 			// AVX SET

libs/ardour/wscript

@@ -471,14 +471,17 @@ def build(bld):
         obj.source += [ 'audio_unit.cc' ]
 
     avx_sources = []
+    fma_sources = []
 
     if Options.options.fpu_optimization:
         if (bld.env['build_target'] == 'i386' or bld.env['build_target'] == 'i686'):
             obj.source += [ 'sse_functions_xmm.cc', 'sse_functions.s', ]
-            avx_sources = [ 'sse_functions_avx_linux.cc', 'x86_functions_fma.cc' ]
+            avx_sources = [ 'sse_functions_avx_linux.cc' ]
+            fma_sources = [ 'x86_functions_fma.cc' ]
         elif bld.env['build_target'] == 'x86_64':
             obj.source += [ 'sse_functions_xmm.cc', 'sse_functions_64bit.s', ]
-            avx_sources = [ 'sse_functions_avx_linux.cc', 'x86_functions_fma.cc' ]
+            avx_sources = [ 'sse_functions_avx_linux.cc' ]
+            fma_sources = [ 'x86_functions_fma.cc' ]
         elif bld.env['build_target'] == 'mingw':
                 # usability of the 64 bit windows assembler depends on the compiler target,
                 # not the build host, which in turn can only be inferred from the name
@@ -486,14 +489,14 @@ def build(bld):
                 if re.search ('x86_64-w64', str(bld.env['CC'])):
                         obj.source += [ 'sse_functions_xmm.cc' ]
                         obj.source += [ 'sse_functions_64bit_win.s',  'sse_avx_functions_64bit_win.s' ]
-                        avx_sources = [ 'sse_functions_avx.cc', 'x86_functions_fma.cc' ]
+                        avx_sources = [ 'sse_functions_avx.cc' ]
+                        fma_sources = [ 'x86_functions_fma.cc' ]
         elif bld.env['build_target'] == 'aarch64':
             obj.source += ['arm_neon_functions.cc']
             obj.defines += [ 'ARM_NEON_SUPPORT' ]
 
         elif bld.env['build_target'] == 'armhf':
             # 32bit ARM needs -mfpu=neon
-            obj.defines += [ 'ARM_NEON_SUPPORT' ]
             arm_neon_cxxflags = list(bld.env['CXXFLAGS'])
             arm_neon_cxxflags.append (bld.env['compiler_flags_dict']['neon'])
             bld(features = 'cxx cxxstlib asm',
@@ -506,13 +509,13 @@ def build(bld):
                 target   = 'arm_neon_functions')
 
             obj.use += ['arm_neon_functions' ]
+            obj.defines += [ 'ARM_NEON_SUPPORT' ]
 
         if avx_sources:
             # as long as we want to use AVX intrinsics in this file,
             # compile it with -mavx flag - append avx flag to the existing
             avx_cxxflags = list(bld.env['CXXFLAGS'])
             avx_cxxflags.append (bld.env['compiler_flags_dict']['avx'])
-            avx_cxxflags.append (bld.env['compiler_flags_dict']['fma'])
             avx_cxxflags.append (bld.env['compiler_flags_dict']['pic'])
             bld(features = 'cxx cxxstlib asm',
                 source   = avx_sources,
@@ -524,6 +527,23 @@ def build(bld):
 
             obj.use += ['sse_avx_functions' ]
 
+        if bld.is_defined('FPU_AVX_FMA_SUPPORT') and fma_sources:
+            fma_cxxflags = list(bld.env['CXXFLAGS'])
+            fma_cxxflags.append (bld.env['compiler_flags_dict']['avx'])
+            fma_cxxflags.append (bld.env['compiler_flags_dict']['pic'])
+            fma_cxxflags.append (bld.env['compiler_flags_dict']['fma'])
+
+            bld(features = 'cxx cxxstlib asm',
+                source   = fma_sources,
+                cxxflags = fma_cxxflags,
+                includes = [ '.' ],
+                use = [ 'libtemporal', 'libpbd', 'libevoral', 'liblua' ],
+                uselib = [ 'GLIBMM', 'XML' ],
+                target   = 'sse_fma_functions')
+
+            obj.use += ['sse_fma_functions' ]
+            obj.defines += [ 'FPU_AVX_FMA_SUPPORT' ]
+
     # i18n
     if bld.is_defined('ENABLE_NLS'):
         mo_files = bld.path.ant_glob('po/*.mo')

wscript

@@ -493,6 +493,18 @@ int main() { return 0; }''',
     if opt.fpu_optimization:
         if conf.env['build_target'] == 'armhf' or conf.env['build_target'] == 'aarch64':
             conf.define('ARM_NEON_SUPPORT', 1)
+        elif conf.env['build_target'] == 'aarch64':
+            conf.define ('FPU_AVX_FMA_SUPPORT', 1)
+        elif conf.env['build_target'] == 'i386' or conf.env['build_target'] == 'i686' or conf.env['build_target'] == 'x86_64':
+            conf.check_cxx(fragment = "#include <immintrin.h>\nint main(void) { __m128 a; _mm_fmadd_ss(a, a, a); return 0; }\n",
+                           features  = ['cxx'],
+                           cxxflags  = [ conf.env['compiler_flags_dict']['fma'], conf.env['compiler_flags_dict']['avx'] ],
+                           mandatory = False,
+                           execute   = False,
+                           msg       = 'Checking compiler for AVX/FMA intrinsics',
+                           okmsg     = 'Found',
+                           errmsg    = 'Not supported',
+                           define_name = 'FPU_AVX_FMA_SUPPORT')
 
     if opt.use_libcpp or conf.env['build_host'] in [ 'el_capitan', 'sierra', 'high_sierra', 'mojave', 'catalina' ]:
        cxx_flags.append('--stdlib=libc++')
@@ -1445,6 +1457,7 @@ const char* const ardour_config_info = "\\n\\
     write_config_text('Dr. Mingw',             conf.is_defined('HAVE_DRMINGW'))
     write_config_text('FLAC',                  conf.is_defined('HAVE_FLAC'))
     write_config_text('FPU optimization',      opts.fpu_optimization)
+    write_config_text('FPU AVX/FMA support',   conf.is_defined('FPU_AVX_FMA_SUPPORT'))
     write_config_text('Freedesktop files',     opts.freedesktop)
     write_config_text('Libjack linking',       conf.env['libjack_link'])
     write_config_text('Libjack metadata',      conf.is_defined ('HAVE_JACK_METADATA'))