diff options
Diffstat (limited to '')
-rw-r--r-- | libc-bottom-half/sources/math/fmin-fmax.c | 34 |
1 files changed, 34 insertions, 0 deletions
diff --git a/libc-bottom-half/sources/math/fmin-fmax.c b/libc-bottom-half/sources/math/fmin-fmax.c new file mode 100644 index 0000000..9293798 --- /dev/null +++ b/libc-bottom-half/sources/math/fmin-fmax.c @@ -0,0 +1,34 @@ +// Wasm's `min` and `max` operators implement the IEEE 754-2019 +// `minimum` and `maximum` operations, meaning that given a choice +// between NaN and a number, they return NaN. This differs from +// the C standard library's `fmin` and `fmax` functions, which +// return the number. However, we can still use wasm's builtins +// by handling the NaN cases explicitly, and it still turns out +// to be faster than doing the whole operation in +// target-independent C. And, it's smaller. + +#include <math.h> + +float fminf(float x, float y) { + if (isnan(x)) return y; + if (isnan(y)) return x; + return __builtin_wasm_min_f32(x, y); +} + +float fmaxf(float x, float y) { + if (isnan(x)) return y; + if (isnan(y)) return x; + return __builtin_wasm_max_f32(x, y); +} + +double fmin(double x, double y) { + if (isnan(x)) return y; + if (isnan(y)) return x; + return __builtin_wasm_min_f64(x, y); +} + +double fmax(double x, double y) { + if (isnan(x)) return y; + if (isnan(y)) return x; + return __builtin_wasm_max_f64(x, y); +} |