arm: add single instruction fma

vfma is available in the vfpv4 fpu and above, the ACLE standard feature
test for double precision hardware fma support is
  __ARM_FEATURE_FMA && __ARM_FP&8
we need further checks to work around clang bugs (fixed in clang >=7.0)
  && !__SOFTFP__
because __ARM_FP is defined even with -mfloat-abi=soft
  && !BROKEN_VFP_ASM
to disable the single precision code when inline asm handling is broken.

For runtime selection the HWCAP_ARM_VFPv4 hwcap flag can be used, but
that requires further work.

2 files changed, 30 insertions, 0 deletions

diff --git a/src/math/arm/fma.c b/src/math/arm/fma.c
new file mode 100644
index 00000000..2a9b8efa
--- /dev/null
+++ b/src/math/arm/fma.c
@@ -0,0 +1,15 @@
+#include <math.h>
+
+#if __ARM_FEATURE_FMA && __ARM_FP&8 && !__SOFTFP__
+
+double fma(double x, double y, double z)
+{
+	__asm__ ("vfma.f64 %P0, %P1, %P2" : "+w"(z) : "w"(x), "w"(y));
+	return z;
+}
+
+#else
+
+#include "../fma.c"
+
+#endif
diff --git a/src/math/arm/fmaf.c b/src/math/arm/fmaf.c
new file mode 100644
index 00000000..a1793d27
--- /dev/null
+++ b/src/math/arm/fmaf.c
@@ -0,0 +1,15 @@
+#include <math.h>
+
+#if __ARM_FEATURE_FMA && __ARM_FP&4 && !__SOFTFP__ && !BROKEN_VFP_ASM
+
+float fmaf(float x, float y, float z)
+{
+	__asm__ ("vfma.f32 %0, %1, %2" : "+t"(z) : "t"(x), "t"(y));
+	return z;
+}
+
+#else
+
+#include "../fmaf.c"
+
+#endif