Skip to content

Commit

Permalink
fn dav1d_sgr_filter1_neon: Deduplicate w/ generics
Browse files Browse the repository at this point in the history
  • Loading branch information
@randomPoison
randomPoison committed Jul 25, 2023
1 parent c207227 commit 045d4ba
Show file tree
Hide file tree
Showing 3 changed files with 227 additions and 185 deletions.
211 changes: 211 additions & 0 deletions src/looprestoration.rs
View file
Original file line number Diff line number Diff line change
@@ -1,5 +1,7 @@
use crate::include::common::bitdepth::AsPrimitive;
use crate::include::common::bitdepth::BitDepth;
#[cfg(all(feature = "asm", any(target_arch = "arm", target_arch = "aarch64")))]
use crate::include::common::bitdepth::{BitDepth16, BitDepth8};
use crate::include::common::intops::iclip;
use crate::include::common::intops::imax;
use crate::include::common::intops::umin;
Expand Down Expand Up @@ -934,3 +936,212 @@ unsafe fn sgr_mix_rust<BD: BitDepth>(
j += 1;
}
}

// TODO(randomPoison): Temporarily pub until all usages can be made private.
#[cfg(all(feature = "asm", any(target_arch = "arm", target_arch = "aarch64")))]
pub(crate) trait BitDepthLooprestorationArm: BitDepth {
fn dav1d_sgr_box3_h_neon(
sumsq: *mut int32_t,
sum: *mut int16_t,
left: *const [Self::Pixel; 4],
src: *const Self::Pixel,
stride: ptrdiff_t,
w: libc::c_int,
h: libc::c_int,
edges: LrEdgeFlags,
);

fn dav1d_sgr_finish_filter1_neon(
tmp: *mut int16_t,
src: *const Self::Pixel,
stride: ptrdiff_t,
a: *const int32_t,
b: *const int16_t,
w: libc::c_int,
h: libc::c_int,
);
}

#[cfg(all(feature = "asm", any(target_arch = "arm", target_arch = "aarch64")))]
impl BitDepthLooprestorationArm for BitDepth8 {
#[inline(always)]
fn dav1d_sgr_box3_h_neon(
sumsq: *mut int32_t,
sum: *mut int16_t,
left: *const [Self::Pixel; 4],
src: *const Self::Pixel,
stride: ptrdiff_t,
w: libc::c_int,
h: libc::c_int,
edges: LrEdgeFlags,
) {
extern "C" {
fn dav1d_sgr_box3_h_8bpc_neon(
sumsq: *mut int32_t,
sum: *mut int16_t,
left: *const [<BitDepth8 as BitDepth>::Pixel; 4],
src: *const <BitDepth8 as BitDepth>::Pixel,
stride: ptrdiff_t,
w: libc::c_int,
h: libc::c_int,
edges: LrEdgeFlags,
);
}

unsafe { dav1d_sgr_box3_h_8bpc_neon(sumsq, sum, left, src, stride, w, h, edges) }
}

#[inline(always)]
fn dav1d_sgr_finish_filter1_neon(
tmp: *mut int16_t,
src: *const Self::Pixel,
stride: ptrdiff_t,
a: *const int32_t,
b: *const int16_t,
w: libc::c_int,
h: libc::c_int,
) {
extern "C" {
fn dav1d_sgr_finish_filter1_8bpc_neon(
tmp: *mut int16_t,
src: *const <BitDepth8 as BitDepth>::Pixel,
stride: ptrdiff_t,
a: *const int32_t,
b: *const int16_t,
w: libc::c_int,
h: libc::c_int,
);
}

unsafe { dav1d_sgr_finish_filter1_8bpc_neon(tmp, src, stride, a, b, w, h) }
}
}

#[cfg(all(feature = "asm", any(target_arch = "arm", target_arch = "aarch64")))]
impl BitDepthLooprestorationArm for BitDepth16 {
#[inline(always)]
fn dav1d_sgr_box3_h_neon(
sumsq: *mut int32_t,
sum: *mut int16_t,
left: *const [Self::Pixel; 4],
src: *const Self::Pixel,
stride: ptrdiff_t,
w: libc::c_int,
h: libc::c_int,
edges: LrEdgeFlags,
) {
extern "C" {
fn dav1d_sgr_box3_h_16bpc_neon(
sumsq: *mut int32_t,
sum: *mut int16_t,
left: *const [<BitDepth16 as BitDepth>::Pixel; 4],
src: *const <BitDepth16 as BitDepth>::Pixel,
stride: ptrdiff_t,
w: libc::c_int,
h: libc::c_int,
edges: LrEdgeFlags,
);
}

unsafe { dav1d_sgr_box3_h_16bpc_neon(sumsq, sum, left, src, stride, w, h, edges) }
}

#[inline(always)]
fn dav1d_sgr_finish_filter1_neon(
tmp: *mut int16_t,
src: *const Self::Pixel,
stride: ptrdiff_t,
a: *const int32_t,
b: *const int16_t,
w: libc::c_int,
h: libc::c_int,
) {
extern "C" {
fn dav1d_sgr_finish_filter1_16bpc_neon(
tmp: *mut int16_t,
src: *const <BitDepth16 as BitDepth>::Pixel,
stride: ptrdiff_t,
a: *const int32_t,
b: *const int16_t,
w: libc::c_int,
h: libc::c_int,
);
}

unsafe { dav1d_sgr_finish_filter1_16bpc_neon(tmp, src, stride, a, b, w, h) }
}
}

extern "C" {
fn dav1d_sgr_box3_v_neon(
sumsq: *mut int32_t,
sum: *mut int16_t,
w: libc::c_int,
h: libc::c_int,
edges: LrEdgeFlags,
);
fn dav1d_sgr_calc_ab1_neon(
a: *mut int32_t,
b: *mut int16_t,
w: libc::c_int,
h: libc::c_int,
strength: libc::c_int,
bitdepth_max: libc::c_int,
);
}

// TODO(randomPoison): Temporarily pub until callers are deduplicated.
#[cfg(all(feature = "asm", any(target_arch = "arm", target_arch = "aarch64")))]
pub(crate) unsafe fn dav1d_sgr_filter1_neon<BD: BitDepthLooprestorationArm>(
mut tmp: *mut int16_t,
mut src: *const BD::Pixel,
stride: ptrdiff_t,
mut left: *const [BD::Pixel; 4],
mut lpf: *const BD::Pixel,
w: libc::c_int,
h: libc::c_int,
strength: libc::c_int,
edges: LrEdgeFlags,
bd: BD,
) {
let mut sumsq_mem: Align16<[int32_t; 27208]> = Align16([0; 27208]);
let sumsq: *mut int32_t = &mut *sumsq_mem
.0
.as_mut_ptr()
.offset(((384 + 16) * 2 + 8) as isize) as *mut int32_t;
let a: *mut int32_t = sumsq;
let mut sum_mem: Align16<[int16_t; 27216]> = Align16([0; 27216]);
let sum: *mut int16_t = &mut *sum_mem
.0
.as_mut_ptr()
.offset(((384 + 16) * 2 + 16) as isize) as *mut int16_t;
let b: *mut int16_t = sum;
BD::dav1d_sgr_box3_h_neon(sumsq, sum, left, src, stride, w, h, edges);
if edges as libc::c_uint & LR_HAVE_TOP as libc::c_int as libc::c_uint != 0 {
BD::dav1d_sgr_box3_h_neon(
&mut *sumsq.offset((-(2 as libc::c_int) * (384 + 16)) as isize),
&mut *sum.offset((-(2 as libc::c_int) * (384 + 16)) as isize),
0 as *const [BD::Pixel; 4],
lpf,
stride,
w,
2 as libc::c_int,
edges,
);
}
if edges as libc::c_uint & LR_HAVE_BOTTOM as libc::c_int as libc::c_uint != 0 {
BD::dav1d_sgr_box3_h_neon(
&mut *sumsq.offset((h * (384 + 16)) as isize),
&mut *sum.offset((h * (384 + 16)) as isize),
0 as *const [BD::Pixel; 4],
lpf.offset((6 * BD::pxstride(stride as usize)) as isize),
stride,
w,
2 as libc::c_int,
edges,
);
}
dav1d_sgr_box3_v_neon(sumsq, sum, w, h, edges);
dav1d_sgr_calc_ab1_neon(a, b, w, h, strength, bd.bitdepth_max().as_());
BD::dav1d_sgr_finish_filter1_neon(tmp, src, stride, a, b, w, h);
}
102 changes: 8 additions & 94 deletions src/looprestoration_tmpl_16.rs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -33,40 +33,6 @@ extern "C" {
w: libc::c_int,
h: libc::c_int,
);
fn dav1d_sgr_box3_h_16bpc_neon(
sumsq: *mut int32_t,
sum: *mut int16_t,
left: *const [pixel; 4],
src: *const pixel,
stride: ptrdiff_t,
w: libc::c_int,
h: libc::c_int,
edges: LrEdgeFlags,
);
fn dav1d_sgr_box3_v_neon(
sumsq: *mut int32_t,
sum: *mut int16_t,
w: libc::c_int,
h: libc::c_int,
edges: LrEdgeFlags,
);
fn dav1d_sgr_calc_ab1_neon(
a: *mut int32_t,
b: *mut int16_t,
w: libc::c_int,
h: libc::c_int,
strength: libc::c_int,
bitdepth_max: libc::c_int,
);
fn dav1d_sgr_finish_filter1_16bpc_neon(
tmp: *mut int16_t,
src: *const pixel,
stride: ptrdiff_t,
a: *const int32_t,
b: *const int16_t,
w: libc::c_int,
h: libc::c_int,
);
fn dav1d_sgr_weighted2_16bpc_neon(
dst: *mut pixel,
dst_stride: ptrdiff_t,
Expand Down Expand Up @@ -383,6 +349,9 @@ unsafe extern "C" fn sgr_filter_3x3_neon(
edges: LrEdgeFlags,
bitdepth_max: libc::c_int,
) {
use crate::include::common::bitdepth::BitDepth;
use crate::src::looprestoration::dav1d_sgr_filter1_neon;

let mut tmp: Align16<[int16_t; 24576]> = Align16([0; 24576]);
dav1d_sgr_filter1_neon(
tmp.0.as_mut_ptr(),
Expand All @@ -394,7 +363,7 @@ unsafe extern "C" fn sgr_filter_3x3_neon(
h,
(*params).sgr.s1 as libc::c_int,
edges,
bitdepth_max,
BitDepth16::from_c(bitdepth_max),
);
dav1d_sgr_weighted1_16bpc_neon(
dst,
Expand All @@ -409,64 +378,6 @@ unsafe extern "C" fn sgr_filter_3x3_neon(
);
}

#[cfg(all(feature = "asm", any(target_arch = "arm", target_arch = "aarch64")))]
unsafe extern "C" fn dav1d_sgr_filter1_neon(
mut tmp: *mut int16_t,
mut src: *const pixel,
stride: ptrdiff_t,
mut left: *const [pixel; 4],
mut lpf: *const pixel,
w: libc::c_int,
h: libc::c_int,
strength: libc::c_int,
edges: LrEdgeFlags,
bitdepth_max: libc::c_int,
) {
use crate::src::looprestoration::LR_HAVE_BOTTOM;
use crate::src::looprestoration::LR_HAVE_TOP;

let mut sumsq_mem: Align16<[int32_t; 27208]> = Align16([0; 27208]);
let sumsq: *mut int32_t = &mut *sumsq_mem
.0
.as_mut_ptr()
.offset(((384 + 16) * 2 + 8) as isize) as *mut int32_t;
let a: *mut int32_t = sumsq;
let mut sum_mem: Align16<[int16_t; 27216]> = Align16([0; 27216]);
let sum: *mut int16_t = &mut *sum_mem
.0
.as_mut_ptr()
.offset(((384 + 16) * 2 + 16) as isize) as *mut int16_t;
let b: *mut int16_t = sum;
dav1d_sgr_box3_h_16bpc_neon(sumsq, sum, left, src, stride, w, h, edges);
if edges as libc::c_uint & LR_HAVE_TOP as libc::c_int as libc::c_uint != 0 {
dav1d_sgr_box3_h_16bpc_neon(
&mut *sumsq.offset((-(2 as libc::c_int) * (384 + 16)) as isize),
&mut *sum.offset((-(2 as libc::c_int) * (384 + 16)) as isize),
0 as *const [pixel; 4],
lpf,
stride,
w,
2 as libc::c_int,
edges,
);
}
if edges as libc::c_uint & LR_HAVE_BOTTOM as libc::c_int as libc::c_uint != 0 {
dav1d_sgr_box3_h_16bpc_neon(
&mut *sumsq.offset((h * (384 + 16)) as isize),
&mut *sum.offset((h * (384 + 16)) as isize),
0 as *const [pixel; 4],
lpf.offset((6 * PXSTRIDE(stride)) as isize),
stride,
w,
2 as libc::c_int,
edges,
);
}
dav1d_sgr_box3_v_neon(sumsq, sum, w, h, edges);
dav1d_sgr_calc_ab1_neon(a, b, w, h, strength, bitdepth_max);
dav1d_sgr_finish_filter1_16bpc_neon(tmp, src, stride, a, b, w, h);
}

#[cfg(all(feature = "asm", any(target_arch = "arm", target_arch = "aarch64")))]
unsafe extern "C" fn dav1d_sgr_filter2_neon(
mut tmp: *mut int16_t,
Expand Down Expand Up @@ -625,6 +536,9 @@ unsafe extern "C" fn sgr_filter_mix_neon(
edges: LrEdgeFlags,
bitdepth_max: libc::c_int,
) {
use crate::include::common::bitdepth::BitDepth;
use crate::src::looprestoration::dav1d_sgr_filter1_neon;

let mut tmp1: Align16<[int16_t; 24576]> = Align16([0; 24576]);
let mut tmp2: Align16<[int16_t; 24576]> = Align16([0; 24576]);
dav1d_sgr_filter2_neon(
Expand All @@ -649,7 +563,7 @@ unsafe extern "C" fn sgr_filter_mix_neon(
h,
(*params).sgr.s1 as libc::c_int,
edges,
bitdepth_max,
BitDepth16::from_c(bitdepth_max),
);
let wt: [int16_t; 2] = [(*params).sgr.w0, (*params).sgr.w1];
dav1d_sgr_weighted2_16bpc_neon(
Expand Down
Loading

0 comments on commit 045d4ba

Please sign in to comment.