Skip to content

Commit

Permalink
mod looprestoration::neon_erased: Move fn *_neon_eraseds to their…
Browse files Browse the repository at this point in the history 
… own `mod`.
  • Loading branch information
@kkysen
kkysen committed Jul 4, 2024
1 parent 2b52795 commit 4ca42c9
Showing 1 changed file with 92 additions and 87 deletions.
179 changes: 92 additions & 87 deletions src/looprestoration.rs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1581,35 +1581,6 @@ unsafe fn rav1d_sgr_weighted2_neon<BD: BitDepth>(
)
}

/// # Safety
///
/// Must be called by [`loop_restoration_filter::Fn::call`].
#[cfg(all(feature = "asm", any(target_arch = "arm", target_arch = "aarch64")))]
unsafe extern "C" fn sgr_filter_5x5_neon_erased<BD: BitDepth>(
_p_ptr: *mut DynPixel,
_stride: ptrdiff_t,
left: *const LeftPixelRow<DynPixel>,
lpf: *const DynPixel,
w: c_int,
h: c_int,
params: &LooprestorationParams,
edges: LrEdgeFlags,
bitdepth_max: c_int,
p: *const FFISafe<Rav1dPictureDataComponentOffset>,
_lpf: *const FFISafe<DisjointMut<AlignedVec64<u8>>>,
) {
// SAFETY: Was passed as `FFISafe::new(_)` in `loop_restoration_filter::Fn::call`.
let p = *unsafe { FFISafe::get(p) };
let left = left.cast();
let lpf = lpf.cast();
let bd = BD::from_c(bitdepth_max);
let w = w as usize;
let h = h as usize;
// SAFETY: Length sliced in `loop_restoration_filter::Fn::call`.
let left = unsafe { slice::from_raw_parts(left, h) };
sgr_filter_5x5_neon(p, left, lpf, w, h, params, edges, bd)
}

#[cfg(all(feature = "asm", target_arch = "arm"))]
unsafe fn sgr_filter_5x5_neon<BD: BitDepth>(
dst: Rav1dPictureDataComponentOffset,
Expand All @@ -1629,35 +1600,6 @@ unsafe fn sgr_filter_5x5_neon<BD: BitDepth>(
rav1d_sgr_weighted1_neon(dst, dst, &mut tmp.0, w, h, sgr.w0, bd);
}

/// # Safety
///
/// Must be called by [`loop_restoration_filter::Fn::call`].
#[cfg(all(feature = "asm", any(target_arch = "arm", target_arch = "aarch64")))]
unsafe extern "C" fn sgr_filter_3x3_neon_erased<BD: BitDepth>(
_p_ptr: *mut DynPixel,
_stride: ptrdiff_t,
left: *const LeftPixelRow<DynPixel>,
lpf: *const DynPixel,
w: c_int,
h: c_int,
params: &LooprestorationParams,
edges: LrEdgeFlags,
bitdepth_max: c_int,
p: *const FFISafe<Rav1dPictureDataComponentOffset>,
_lpf: *const FFISafe<DisjointMut<AlignedVec64<u8>>>,
) {
// SAFETY: Was passed as `FFISafe::new(_)` in `loop_restoration_filter::Fn::call`.
let p = *unsafe { FFISafe::get(p) };
let left = left.cast();
let lpf = lpf.cast();
let w = w as usize;
let h = h as usize;
let bd = BD::from_c(bitdepth_max);
// SAFETY: Length sliced in `loop_restoration_filter::Fn::call`.
let left = unsafe { slice::from_raw_parts(left, h) };
sgr_filter_3x3_neon(p, left, lpf, w, h, params, edges, bd)
}

#[cfg(all(feature = "asm", target_arch = "arm"))]
unsafe fn sgr_filter_3x3_neon<BD: BitDepth>(
dst: Rav1dPictureDataComponentOffset,
Expand Down Expand Up @@ -3444,35 +3386,6 @@ unsafe fn sgr_filter_mix_neon<BD: BitDepth>(
}
}

/// # Safety
///
/// Must be called by [`loop_restoration_filter::Fn::call`].
#[cfg(all(feature = "asm", any(target_arch = "arm", target_arch = "aarch64")))]
unsafe extern "C" fn sgr_filter_mix_neon_erased<BD: BitDepth>(
_p_ptr: *mut DynPixel,
_stride: ptrdiff_t,
left: *const LeftPixelRow<DynPixel>,
lpf: *const DynPixel,
w: c_int,
h: c_int,
params: &LooprestorationParams,
edges: LrEdgeFlags,
bitdepth_max: c_int,
p: *const FFISafe<Rav1dPictureDataComponentOffset>,
_lpf: *const FFISafe<DisjointMut<AlignedVec64<u8>>>,
) {
// SAFETY: Was passed as `FFISafe::new(_)` in `loop_restoration_filter::Fn::call`.
let p = *unsafe { FFISafe::get(p) };
let left = left.cast();
let lpf = lpf.cast();
let bd = BD::from_c(bitdepth_max);
let w = w as usize;
let h = h as usize;
// SAFETY: Length sliced in `loop_restoration_filter::Fn::call`.
let left = unsafe { slice::from_raw_parts(left, h) };
sgr_filter_mix_neon(p, left, lpf, w, h, params, edges, bd)
}

#[cfg(all(feature = "asm", target_arch = "arm"))]
unsafe fn sgr_filter_mix_neon<BD: BitDepth>(
dst: Rav1dPictureDataComponentOffset,
Expand All @@ -3495,6 +3408,96 @@ unsafe fn sgr_filter_mix_neon<BD: BitDepth>(
rav1d_sgr_weighted2_neon(dst, dst, &mut tmp1.0, &mut tmp2.0, w, h, &wt, bd);
}

#[cfg(all(feature = "asm", any(target_arch = "arm", target_arch = "aarch64")))]
mod neon_erased {
use super::*;

/// # Safety
///
/// Must be called by [`loop_restoration_filter::Fn::call`].
pub unsafe extern "C" fn sgr_filter_5x5_neon_erased<BD: BitDepth>(
_p_ptr: *mut DynPixel,
_stride: ptrdiff_t,
left: *const LeftPixelRow<DynPixel>,
lpf: *const DynPixel,
w: c_int,
h: c_int,
params: &LooprestorationParams,
edges: LrEdgeFlags,
bitdepth_max: c_int,
p: *const FFISafe<Rav1dPictureDataComponentOffset>,
_lpf: *const FFISafe<DisjointMut<AlignedVec64<u8>>>,
) {
// SAFETY: Was passed as `FFISafe::new(_)` in `loop_restoration_filter::Fn::call`.
let p = *unsafe { FFISafe::get(p) };
let left = left.cast();
let lpf = lpf.cast();
let bd = BD::from_c(bitdepth_max);
let w = w as usize;
let h = h as usize;
// SAFETY: Length sliced in `loop_restoration_filter::Fn::call`.
let left = unsafe { slice::from_raw_parts(left, h) };
sgr_filter_5x5_neon(p, left, lpf, w, h, params, edges, bd)
}

/// # Safety
///
/// Must be called by [`loop_restoration_filter::Fn::call`].
#[cfg(all(feature = "asm", any(target_arch = "arm", target_arch = "aarch64")))]
pub unsafe extern "C" fn sgr_filter_3x3_neon_erased<BD: BitDepth>(
_p_ptr: *mut DynPixel,
_stride: ptrdiff_t,
left: *const LeftPixelRow<DynPixel>,
lpf: *const DynPixel,
w: c_int,
h: c_int,
params: &LooprestorationParams,
edges: LrEdgeFlags,
bitdepth_max: c_int,
p: *const FFISafe<Rav1dPictureDataComponentOffset>,
_lpf: *const FFISafe<DisjointMut<AlignedVec64<u8>>>,
) {
// SAFETY: Was passed as `FFISafe::new(_)` in `loop_restoration_filter::Fn::call`.
let p = *unsafe { FFISafe::get(p) };
let left = left.cast();
let lpf = lpf.cast();
let w = w as usize;
let h = h as usize;
let bd = BD::from_c(bitdepth_max);
// SAFETY: Length sliced in `loop_restoration_filter::Fn::call`.
let left = unsafe { slice::from_raw_parts(left, h) };
sgr_filter_3x3_neon(p, left, lpf, w, h, params, edges, bd)
}

/// # Safety
///
/// Must be called by [`loop_restoration_filter::Fn::call`].
pub unsafe extern "C" fn sgr_filter_mix_neon_erased<BD: BitDepth>(
_p_ptr: *mut DynPixel,
_stride: ptrdiff_t,
left: *const LeftPixelRow<DynPixel>,
lpf: *const DynPixel,
w: c_int,
h: c_int,
params: &LooprestorationParams,
edges: LrEdgeFlags,
bitdepth_max: c_int,
p: *const FFISafe<Rav1dPictureDataComponentOffset>,
_lpf: *const FFISafe<DisjointMut<AlignedVec64<u8>>>,
) {
// SAFETY: Was passed as `FFISafe::new(_)` in `loop_restoration_filter::Fn::call`.
let p = *unsafe { FFISafe::get(p) };
let left = left.cast();
let lpf = lpf.cast();
let bd = BD::from_c(bitdepth_max);
let w = w as usize;
let h = h as usize;
// SAFETY: Length sliced in `loop_restoration_filter::Fn::call`.
let left = unsafe { slice::from_raw_parts(left, h) };
sgr_filter_mix_neon(p, left, lpf, w, h, params, edges, bd)
}
}

impl Rav1dLoopRestorationDSPContext {
pub const fn default<BD: BitDepth>() -> Self {
Self {
Expand Down Expand Up @@ -3610,6 +3613,8 @@ impl Rav1dLoopRestorationDSPContext {
}

if matches!(BD::BPC, BPC::BPC8) || bpc == 10 {
use neon_erased::*;

self.sgr[0] = loop_restoration_filter::Fn::new(sgr_filter_5x5_neon_erased::<BD>);
self.sgr[1] = loop_restoration_filter::Fn::new(sgr_filter_3x3_neon_erased::<BD>);
self.sgr[2] = loop_restoration_filter::Fn::new(sgr_filter_mix_neon_erased::<BD>);
Expand Down

0 comments on commit 4ca42c9

Please sign in to comment.