fast_roundf implementations in asm for ARM64. (#511)

2020-12-25 06:48:57 +00:00 · 2020-12-25 06:48:57 +00:00 · 8c4c14ea73
commit 8c4c14ea73
parent 6ae0115bf7
1 changed files with 6 additions and 0 deletions
--- a/common/alnumeric.h
+++ b/common/alnumeric.h
@ -329,6 +329,12 @@ inline float fast_roundf(float f) noexcept
    __asm__ __volatile__("frndint" : "=t"(out) : "0"(f));
    return out;

+#elif (defined(__GNUC__) || defined(__clang__)) && defined(__aarch64__)
+
+    float out;
+    __asm__ volatile("frintx %s0, %s1" : "=w"(out) : "w"(f));
+    return out;
+
 #else

    /* Integral limit, where sub-integral precision is not available for