-
-
Save ActuallyaDeviloper/2a1b84554b0d9e9ff8f14fa5c1b0bcce to your computer and use it in GitHub Desktop.
Branchless f32 to integer tests
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
#![feature(test)] | |
#![feature(inclusive_range_syntax)] | |
extern crate test; | |
extern crate rand; | |
use test::*; | |
fn u16_cast(x: f32) -> u16 { | |
x as u16 | |
} | |
fn u16_cast_clip(x: f32) -> u16 { | |
if u16::min_value() as f32 <= x && x < u16::max_value() as f32 { | |
x as u16 | |
} else if x.is_nan() { | |
0 | |
} else if x < 0.0 { | |
u16::min_value() | |
} else { | |
u16::max_value() | |
} | |
} | |
fn u16_cast_clip2(x: f32) -> u16 { | |
let min = u16::min_value() as f32; | |
let max = u16::max_value() as f32; | |
let y = if min < x { x } else { min }; // We are good to go with NaN, it's already handled with this. | |
let y = if max < y { max } else { y }; | |
y as u16 | |
} | |
fn u16_array_cast(arr: [f32; 128]) -> [u16; 128] { | |
let mut out: [u16; 128] = unsafe { std::mem::uninitialized() }; | |
for i in 0..128 { | |
out[i] = u16_cast(arr[i]); | |
} | |
out | |
} | |
fn u16_array_cast_clip(arr: [f32; 128]) -> [u16; 128] { | |
let mut out: [u16; 128] = unsafe { std::mem::uninitialized() }; | |
for i in 0..128 { | |
out[i] = u16_cast_clip(arr[i]); | |
} | |
out | |
} | |
fn u16_array_cast_clip2(arr: [f32; 128]) -> [u16; 128] { | |
let mut out: [u16; 128] = unsafe { std::mem::uninitialized() }; | |
for i in 0..128 { | |
out[i] = u16_cast_clip2(arr[i]); | |
} | |
out | |
} | |
#[bench] | |
fn u16_bench_cast(b: &mut Bencher) { | |
let x = 0f32; | |
b.iter(|| for _ in 0..128 { | |
black_box(u16_cast(black_box(x))); | |
}); | |
} | |
#[bench] | |
fn u16_bench_cast_clip(b: &mut Bencher) { | |
let x = 0f32; | |
b.iter(|| for _ in 0..128 { | |
black_box(u16_cast_clip(black_box(x))); | |
}); | |
} | |
#[bench] | |
fn u16_bench_cast_clip2(b: &mut Bencher) { | |
let x = 0f32; | |
b.iter(|| for _ in 0..128 { | |
black_box(u16_cast_clip2(black_box(x))); | |
}); | |
} | |
#[bench] | |
fn u16_bench_rng_cast(b: &mut Bencher) { | |
use rand::Rng; | |
let mut rng = rand::thread_rng(); | |
b.iter(|| for _ in 0..128 { | |
black_box(u16_cast(rng.next_f32() * 100000.0f32 - 50000.0f32)); | |
}); | |
} | |
#[bench] | |
fn u16_bench_rng_cast_clip(b: &mut Bencher) { | |
use rand::Rng; | |
let mut rng = rand::thread_rng(); | |
b.iter(|| for _ in 0..128 { | |
black_box(u16_cast_clip(rng.next_f32() * 100000.0f32 - 50000.0f32)); | |
}); | |
} | |
#[bench] | |
fn u16_bench_rng_cast_clip2(b: &mut Bencher) { | |
use rand::Rng; | |
let mut rng = rand::thread_rng(); | |
b.iter(|| for _ in 0..128 { | |
black_box(u16_cast_clip2(rng.next_f32() * 100000.0f32 - 50000.0f32)); | |
}); | |
} | |
#[bench] | |
fn u16_bench_array_cast(b: &mut Bencher) { | |
let x = [0f32; 128]; | |
b.iter(|| u16_array_cast(black_box(x))); | |
} | |
#[bench] | |
fn u16_bench_array_cast_clip(b: &mut Bencher) { | |
let x = [0f32; 128]; | |
b.iter(|| u16_array_cast_clip(black_box(x))); | |
} | |
#[bench] | |
fn u16_bench_array_cast_clip2(b: &mut Bencher) { | |
let x = [0f32; 128]; | |
b.iter(|| u16_array_cast_clip2(black_box(x))); | |
} | |
#[test] | |
fn u16_clip_test() { | |
for i in 0..=u32::max_value() { | |
let flt = unsafe { std::mem::transmute::<_, f32>(i) }; | |
let result = u16_cast_clip(flt); | |
let result2 = u16_cast_clip2(flt); | |
if (result != result2) { | |
panic!("Error at {}", flt); | |
} | |
} | |
} | |
fn i16_cast(x: f32) -> i16 { | |
x as i16 | |
} | |
fn i16_cast_clip(x: f32) -> i16 { | |
if i16::min_value() as f32 <= x && x < i16::max_value() as f32 { | |
x as i16 | |
} else if x.is_nan() { | |
0 | |
} else if x < 0.0 { | |
i16::min_value() | |
} else { | |
i16::max_value() | |
} | |
} | |
fn i16_cast_clip2(x: f32) -> i16 { | |
let min = i16::min_value() as f32; | |
let max = i16::max_value() as f32; | |
let y = if min > x { min } else { x }; | |
let y = if max < y { max } else { y }; | |
if x.is_nan() { 0 } else { y as i16 } // Special case for returning 0 in case of NaN. | |
} | |
fn i16_array_cast(arr: [f32; 128]) -> [i16; 128] { | |
let mut out: [i16; 128] = unsafe { std::mem::uninitialized() }; | |
for i in 0..128 { | |
out[i] = i16_cast(arr[i]); | |
} | |
out | |
} | |
fn i16_array_cast_clip(arr: [f32; 128]) -> [i16; 128] { | |
let mut out: [i16; 128] = unsafe { std::mem::uninitialized() }; | |
for i in 0..128 { | |
out[i] = i16_cast_clip(arr[i]); | |
} | |
out | |
} | |
fn i16_array_cast_clip2(arr: [f32; 128]) -> [i16; 128] { | |
let mut out: [i16; 128] = unsafe { std::mem::uninitialized() }; | |
for i in 0..128 { | |
out[i] = i16_cast_clip2(arr[i]); | |
} | |
out | |
} | |
#[bench] | |
fn i16_bench_cast(b: &mut Bencher) { | |
let x = 0f32; | |
b.iter(|| for _ in 0..128 { | |
black_box(i16_cast(black_box(x))); | |
}); | |
} | |
#[bench] | |
fn i16_bench_cast_clip(b: &mut Bencher) { | |
let x = 0f32; | |
b.iter(|| for _ in 0..128 { | |
black_box(i16_cast_clip(black_box(x))); | |
}); | |
} | |
#[bench] | |
fn i16_bench_cast_clip2(b: &mut Bencher) { | |
let x = 0f32; | |
b.iter(|| for _ in 0..128 { | |
black_box(i16_cast_clip2(black_box(x))); | |
}); | |
} | |
#[bench] | |
fn i16_bench_rng_cast(b: &mut Bencher) { | |
use rand::Rng; | |
let mut rng = rand::thread_rng(); | |
b.iter(|| for _ in 0..128 { | |
black_box(i16_cast(rng.next_f32() * 100000.0f32 - 50000.0f32)); | |
}); | |
} | |
#[bench] | |
fn i16_bench_rng_cast_clip(b: &mut Bencher) { | |
use rand::Rng; | |
let mut rng = rand::thread_rng(); | |
b.iter(|| for _ in 0..128 { | |
black_box(i16_cast_clip(rng.next_f32() * 100000.0f32 - 50000.0f32)); | |
}); | |
} | |
#[bench] | |
fn i16_bench_rng_cast_clip2(b: &mut Bencher) { | |
use rand::Rng; | |
let mut rng = rand::thread_rng(); | |
b.iter(|| for _ in 0..128 { | |
black_box(i16_cast_clip2(rng.next_f32() * 100000.0f32 - 50000.0f32)); | |
}); | |
} | |
#[bench] | |
fn i16_bench_array_cast(b: &mut Bencher) { | |
let x = [0f32; 128]; | |
b.iter(|| i16_array_cast(black_box(x))); | |
} | |
#[bench] | |
fn i16_bench_array_cast_clip(b: &mut Bencher) { | |
let x = [0f32; 128]; | |
b.iter(|| i16_array_cast_clip(black_box(x))); | |
} | |
#[bench] | |
fn i16_bench_array_cast_clip2(b: &mut Bencher) { | |
let x = [0f32; 128]; | |
b.iter(|| i16_array_cast_clip2(black_box(x))); | |
} | |
#[test] | |
fn i16_clip_test() { | |
for i in 0..=u32::max_value() { | |
let flt = unsafe { std::mem::transmute::<_, f32>(i) }; | |
let result = i16_cast_clip(flt); | |
let result2 = i16_cast_clip2(flt); | |
if (result != result2) { | |
panic!("Error at {}", flt); | |
} | |
} | |
} | |
fn i32_cast(x: f32) -> i32 { | |
x as i32 | |
} | |
fn i32_cast_clip(x: f32) -> i32 { | |
if i32::min_value() as f32 <= x && x < i32::max_value() as f32 { | |
x as i32 | |
} else if x.is_nan() { | |
0 | |
} else if x < 0.0 { | |
i32::min_value() | |
} else { | |
i32::max_value() | |
} | |
} | |
fn i32_cast_clip2(x: f32) -> i32 { | |
// i32::max_value() rounds upwards to an out of range value. | |
// By doing an i32 decrement, it's value is reduced so that it fits inside the range of i32 again. | |
let min = i32::min_value() as f32; | |
let max = unsafe { std::mem::transmute::<_, f32>( | |
std::mem::transmute::<_, i32>(i32::max_value() as f32) - 1) }; | |
let y = if min > x { min } else { x }; | |
let y = if max < y { max } else { y }; | |
let z = if x.is_nan() { 0 } else { y as i32 }; // Special case for returning 0 in case of NaN. | |
// f32 can't represent i32::max_value(). | |
// So to get a proper clamp that returns i32::max_value(), this fixup step is needed | |
if max < x { i32::max_value() } else { z } | |
} | |
fn i32_array_cast(arr: [f32; 128]) -> [i32; 128] { | |
let mut out: [i32; 128] = unsafe { std::mem::uninitialized() }; | |
for i in 0..128 { | |
out[i] = i32_cast(arr[i]); | |
} | |
out | |
} | |
fn i32_array_cast_clip(arr: [f32; 128]) -> [i32; 128] { | |
let mut out: [i32; 128] = unsafe { std::mem::uninitialized() }; | |
for i in 0..128 { | |
out[i] = i32_cast_clip(arr[i]); | |
} | |
out | |
} | |
fn i32_array_cast_clip2(arr: [f32; 128]) -> [i32; 128] { | |
let mut out: [i32; 128] = unsafe { std::mem::uninitialized() }; | |
for i in 0..128 { | |
out[i] = i32_cast_clip2(arr[i]); | |
} | |
out | |
} | |
#[bench] | |
fn i32_bench_cast(b: &mut Bencher) { | |
let x = 0f32; | |
b.iter(|| for _ in 0..128 { | |
black_box(i32_cast(black_box(x))); | |
}); | |
} | |
#[bench] | |
fn i32_bench_cast_clip(b: &mut Bencher) { | |
let x = 0f32; | |
b.iter(|| for _ in 0..128 { | |
black_box(i32_cast_clip(black_box(x))); | |
}); | |
} | |
#[bench] | |
fn i32_bench_cast_clip2(b: &mut Bencher) { | |
let x = 0f32; | |
b.iter(|| for _ in 0..128 { | |
black_box(i32_cast_clip2(black_box(x))); | |
}); | |
} | |
#[bench] | |
fn i32_bench_rng_cast(b: &mut Bencher) { | |
use rand::Rng; | |
let mut rng = rand::thread_rng(); | |
b.iter(|| for _ in 0..128 { | |
black_box(i32_cast(rng.next_f32() * 10000000000.0f32 - 5000000000.0f32)); | |
}); | |
} | |
#[bench] | |
fn i32_bench_rng_cast_clip(b: &mut Bencher) { | |
use rand::Rng; | |
let mut rng = rand::thread_rng(); | |
b.iter(|| for _ in 0..128 { | |
black_box(i32_cast_clip(rng.next_f32() * 10000000000.0f32 - 5000000000.0f32)); | |
}); | |
} | |
#[bench] | |
fn i32_bench_rng_cast_clip2(b: &mut Bencher) { | |
use rand::Rng; | |
let mut rng = rand::thread_rng(); | |
b.iter(|| for _ in 0..128 { | |
black_box(i32_cast_clip2(rng.next_f32() * 10000000000.0f32 - 5000000000.0f32)); | |
}); | |
} | |
#[bench] | |
fn i32_bench_array_cast(b: &mut Bencher) { | |
let x = [0f32; 128]; | |
b.iter(|| i32_array_cast(black_box(x))); | |
} | |
#[bench] | |
fn i32_bench_array_cast_clip(b: &mut Bencher) { | |
let x = [0f32; 128]; | |
b.iter(|| i32_array_cast_clip(black_box(x))); | |
} | |
#[bench] | |
fn i32_bench_array_cast_clip2(b: &mut Bencher) { | |
let x = [0f32; 128]; | |
b.iter(|| i32_array_cast_clip2(black_box(x))); | |
} | |
#[test] | |
fn i32_clip_test() { | |
for i in 0..=u32::max_value() { | |
let flt = unsafe { std::mem::transmute::<_, f32>(i) }; | |
let result = i32_cast_clip(flt); | |
let result2 = i32_cast_clip2(flt); | |
if (result != result2) { | |
panic!("Error at {}", flt); | |
} | |
} | |
} |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment