Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Improvements to sgrproj HBD performance #3116

Merged
merged 3 commits into from
Jan 28, 2023
Merged

Improvements to sgrproj HBD performance #3116

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
974f55b
Move bounds checks out of hot loop in sgrproj
shssoichiro Jan 27, 2023
7b2da57
Use saturating_sub in sgrproj_sum_finish
shssoichiro Jan 27, 2023
afe02e3
Use generics for BD-8 on sgrproj functions
shssoichiro Jan 27, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
61 changes: 25 additions & 36 deletions src/asm/x86/lrf.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,16 +19,15 @@ use std::mem;

// computes an intermediate (ab) row for stripe_w + 2 columns at row y
#[inline]
pub fn sgrproj_box_ab_r1(
pub fn sgrproj_box_ab_r1<const BD: usize>(
af: &mut [u32], bf: &mut [u32], iimg: &[u32], iimg_sq: &[u32],
iimg_stride: usize, y: usize, stripe_w: usize, s: u32, bdm8: usize,
cpu: CpuFeatureLevel,
iimg_stride: usize, y: usize, stripe_w: usize, s: u32, cpu: CpuFeatureLevel,
) {
// only use 8-bit AVX2 assembly when bitdepth minus 8 equals 0
if cpu >= CpuFeatureLevel::AVX2 && bdm8 == 0 {
// only use 8-bit AVX2 assembly when bitdepth equals 8
if cpu >= CpuFeatureLevel::AVX2 && BD == 8 {
// SAFETY: Calls Assembly code.
return unsafe {
sgrproj_box_ab_r1_avx2(
sgrproj_box_ab_r1_avx2::<BD>(
af,
bf,
iimg,
Expand All @@ -37,12 +36,11 @@ pub fn sgrproj_box_ab_r1(
y,
stripe_w,
s,
bdm8,
);
};
}

rust::sgrproj_box_ab_r1(
rust::sgrproj_box_ab_r1::<BD>(
af,
bf,
iimg,
Expand All @@ -51,23 +49,21 @@ pub fn sgrproj_box_ab_r1(
y,
stripe_w,
s,
bdm8,
cpu,
);
}

// computes an intermediate (ab) row for stripe_w + 2 columns at row y
#[inline]
pub fn sgrproj_box_ab_r2(
pub fn sgrproj_box_ab_r2<const BD: usize>(
af: &mut [u32], bf: &mut [u32], iimg: &[u32], iimg_sq: &[u32],
iimg_stride: usize, y: usize, stripe_w: usize, s: u32, bdm8: usize,
cpu: CpuFeatureLevel,
iimg_stride: usize, y: usize, stripe_w: usize, s: u32, cpu: CpuFeatureLevel,
) {
// only use 8-bit AVX2 assembly when bitdepth minus 8 equals 0
if cpu >= CpuFeatureLevel::AVX2 && bdm8 == 0 {
// only use 8-bit AVX2 assembly when bitdepth equals 8
if cpu >= CpuFeatureLevel::AVX2 && BD == 8 {
// SAFETY: Calls Assembly code.
return unsafe {
sgrproj_box_ab_r2_avx2(
sgrproj_box_ab_r2_avx2::<BD>(
af,
bf,
iimg,
Expand All @@ -76,12 +72,11 @@ pub fn sgrproj_box_ab_r2(
y,
stripe_w,
s,
bdm8,
);
};
}

rust::sgrproj_box_ab_r2(
rust::sgrproj_box_ab_r2::<BD>(
af,
bf,
iimg,
Expand All @@ -90,7 +85,6 @@ pub fn sgrproj_box_ab_r2(
y,
stripe_w,
s,
bdm8,
cpu,
);
}
Expand Down Expand Up @@ -164,10 +158,11 @@ static X_BY_XPLUS1: [u32; 256] = [

#[inline]
#[target_feature(enable = "avx2")]
unsafe fn sgrproj_box_ab_8_avx2(
unsafe fn sgrproj_box_ab_8_avx2<const BD: usize>(
r: usize, af: &mut [u32], bf: &mut [u32], iimg: &[u32], iimg_sq: &[u32],
iimg_stride: usize, x: usize, y: usize, s: u32, bdm8: usize,
iimg_stride: usize, x: usize, y: usize, s: u32,
) {
let bdm8 = BD - 8;
let d: usize = r * 2 + 1;
let n: i32 = (d * d) as i32;
let one_over_n = if r == 1 { 455 } else { 164 };
Expand Down Expand Up @@ -240,13 +235,13 @@ unsafe fn sgrproj_box_ab_8_avx2(
}

#[target_feature(enable = "avx2")]
pub(crate) unsafe fn sgrproj_box_ab_r1_avx2(
pub(crate) unsafe fn sgrproj_box_ab_r1_avx2<const BD: usize>(
af: &mut [u32], bf: &mut [u32], iimg: &[u32], iimg_sq: &[u32],
iimg_stride: usize, y: usize, stripe_w: usize, s: u32, bdm8: usize,
iimg_stride: usize, y: usize, stripe_w: usize, s: u32,
) {
for x in (0..stripe_w + 2).step_by(8) {
if x + 8 <= stripe_w + 2 {
sgrproj_box_ab_8_avx2(
sgrproj_box_ab_8_avx2::<BD>(
1,
af,
bf,
Expand All @@ -256,11 +251,10 @@ pub(crate) unsafe fn sgrproj_box_ab_r1_avx2(
x,
y,
s,
bdm8,
);
} else {
// finish using scalar
rust::sgrproj_box_ab_internal(
rust::sgrproj_box_ab_internal::<BD>(
1,
af,
bf,
Expand All @@ -271,7 +265,6 @@ pub(crate) unsafe fn sgrproj_box_ab_r1_avx2(
y,
stripe_w,
s,
bdm8,
);
}
}
Expand All @@ -280,7 +273,7 @@ pub(crate) unsafe fn sgrproj_box_ab_r1_avx2(
{
let mut af_ref: Vec<u32> = vec![0; stripe_w + 2];
let mut bf_ref: Vec<u32> = vec![0; stripe_w + 2];
rust::sgrproj_box_ab_internal(
rust::sgrproj_box_ab_internal::<BD>(
1,
&mut af_ref,
&mut bf_ref,
Expand All @@ -291,21 +284,20 @@ pub(crate) unsafe fn sgrproj_box_ab_r1_avx2(
y,
stripe_w,
s,
bdm8,
);
assert_eq!(&af[..stripe_w + 2], &af_ref[..]);
assert_eq!(&bf[..stripe_w + 2], &bf_ref[..]);
}
}

#[target_feature(enable = "avx2")]
pub(crate) unsafe fn sgrproj_box_ab_r2_avx2(
pub(crate) unsafe fn sgrproj_box_ab_r2_avx2<const BD: usize>(
af: &mut [u32], bf: &mut [u32], iimg: &[u32], iimg_sq: &[u32],
iimg_stride: usize, y: usize, stripe_w: usize, s: u32, bdm8: usize,
iimg_stride: usize, y: usize, stripe_w: usize, s: u32,
) {
for x in (0..stripe_w + 2).step_by(8) {
if x + 8 <= stripe_w + 2 {
sgrproj_box_ab_8_avx2(
sgrproj_box_ab_8_avx2::<BD>(
2,
af,
bf,
Expand All @@ -315,11 +307,10 @@ pub(crate) unsafe fn sgrproj_box_ab_r2_avx2(
x,
y,
s,
bdm8,
);
} else {
// finish using scalar
rust::sgrproj_box_ab_internal(
rust::sgrproj_box_ab_internal::<BD>(
2,
af,
bf,
Expand All @@ -330,7 +321,6 @@ pub(crate) unsafe fn sgrproj_box_ab_r2_avx2(
y,
stripe_w,
s,
bdm8,
);
}
}
Expand All @@ -339,7 +329,7 @@ pub(crate) unsafe fn sgrproj_box_ab_r2_avx2(
{
let mut af_ref: Vec<u32> = vec![0; stripe_w + 2];
let mut bf_ref: Vec<u32> = vec![0; stripe_w + 2];
rust::sgrproj_box_ab_internal(
rust::sgrproj_box_ab_internal::<BD>(
2,
&mut af_ref,
&mut bf_ref,
Expand All @@ -350,7 +340,6 @@ pub(crate) unsafe fn sgrproj_box_ab_r2_avx2(
y,
stripe_w,
s,
bdm8,
);
assert_eq!(&af[..stripe_w + 2], &af_ref[..]);
assert_eq!(&bf[..stripe_w + 2], &bf_ref[..]);
Expand Down
Loading