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

Jump to bottom

struct Rav1dFrameContext_lf::cdef_line_buf: Make into AlignedVec32 #776

Merged
merged 1 commit into from
Mar 4, 2024
Merged

struct Rav1dFrameContext_lf::cdef_line_buf: Make into AlignedVec32 #776

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
1 change: 1 addition & 0 deletions src/align.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -205,4 +205,5 @@ impl<T: Copy, C: AlignedByteChunk> Default for AlignedVec<T, C> {
}
}

pub type AlignedVec32<T> = AlignedVec<T, Align32<[u8; 32]>>;
randomPoison marked this conversation as resolved.
Show resolved Hide resolved
pub type AlignedVec64<T> = AlignedVec<T, Align64<[u8; 64]>>;
94 changes: 36 additions & 58 deletions src/decode.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4387,74 +4387,52 @@ pub(crate) unsafe fn rav1d_decode_frame_init(
let mut uv_stride = f.cur.stride[1];
let has_resize = (frame_hdr.size.width[0] != frame_hdr.size.width[1]) as c_int;
let need_cdef_lpf_copy = (c.tc.len() > 1 && has_resize != 0) as c_int;
if y_stride * f.sbh as isize * 4 != f.lf.cdef_buf_plane_sz[0] as isize
|| uv_stride * f.sbh as isize * 8 != f.lf.cdef_buf_plane_sz[1] as isize
|| need_cdef_lpf_copy != f.lf.need_cdef_lpf_copy
|| f.sbh != f.lf.cdef_buf_sbh
{
rav1d_free_aligned(f.lf.cdef_line_buf as *mut c_void);
let mut alloc_sz: usize = 64;
alloc_sz += (y_stride.unsigned_abs() * 4 * f.sbh as usize) << need_cdef_lpf_copy;
alloc_sz += (uv_stride.unsigned_abs() * 8 * f.sbh as usize) << need_cdef_lpf_copy;
f.lf.cdef_line_buf = rav1d_alloc_aligned(alloc_sz, 32) as *mut u8;
let mut ptr = f.lf.cdef_line_buf;
if ptr.is_null() {
f.lf.cdef_buf_plane_sz[1] = 0;
f.lf.cdef_buf_plane_sz[0] = f.lf.cdef_buf_plane_sz[1];
return Err(ENOMEM);
}
let mut alloc_sz: usize = 64;
alloc_sz += (y_stride.unsigned_abs() * 4 * f.sbh as usize) << need_cdef_lpf_copy;
alloc_sz += (uv_stride.unsigned_abs() * 8 * f.sbh as usize) << need_cdef_lpf_copy;
// TODO: Fallible allocation.
f.lf.cdef_line_buf.resize(alloc_sz, 0);
let mut ptr = f.lf.cdef_line_buf.as_mut_ptr();

ptr = ptr.offset(32);
if y_stride < 0 {
f.lf.cdef_line[0][0] = ptr.offset(-(y_stride * (f.sbh as isize * 4 - 1))) as *mut DynPixel;
f.lf.cdef_line[1][0] = ptr.offset(-(y_stride * (f.sbh as isize * 4 - 3))) as *mut DynPixel;
} else {
f.lf.cdef_line[0][0] = ptr.offset(y_stride * 0) as *mut DynPixel;
f.lf.cdef_line[1][0] = ptr.offset(y_stride * 2) as *mut DynPixel;
}
ptr = ptr.offset(y_stride.abs() * f.sbh as isize * 4);
if uv_stride < 0 {
f.lf.cdef_line[0][1] = ptr.offset(-(uv_stride * (f.sbh as isize * 8 - 1))) as *mut DynPixel;
f.lf.cdef_line[0][2] = ptr.offset(-(uv_stride * (f.sbh as isize * 8 - 3))) as *mut DynPixel;
f.lf.cdef_line[1][1] = ptr.offset(-(uv_stride * (f.sbh as isize * 8 - 5))) as *mut DynPixel;
f.lf.cdef_line[1][2] = ptr.offset(-(uv_stride * (f.sbh as isize * 8 - 7))) as *mut DynPixel;
} else {
f.lf.cdef_line[0][1] = ptr.offset(uv_stride * 0) as *mut DynPixel;
f.lf.cdef_line[0][2] = ptr.offset(uv_stride * 2) as *mut DynPixel;
f.lf.cdef_line[1][1] = ptr.offset(uv_stride * 4) as *mut DynPixel;
f.lf.cdef_line[1][2] = ptr.offset(uv_stride * 6) as *mut DynPixel;
}

ptr = ptr.offset(32);
if need_cdef_lpf_copy != 0 {
ptr = ptr.offset(uv_stride.abs() * f.sbh as isize * 8);
if y_stride < 0 {
f.lf.cdef_line[0][0] =
f.lf.cdef_lpf_line[0] =
ptr.offset(-(y_stride * (f.sbh as isize * 4 - 1))) as *mut DynPixel;
f.lf.cdef_line[1][0] =
ptr.offset(-(y_stride * (f.sbh as isize * 4 - 3))) as *mut DynPixel;
} else {
f.lf.cdef_line[0][0] = ptr.offset(y_stride * 0) as *mut DynPixel;
f.lf.cdef_line[1][0] = ptr.offset(y_stride * 2) as *mut DynPixel;
f.lf.cdef_lpf_line[0] = ptr as *mut DynPixel;
}
ptr = ptr.offset(y_stride.abs() * f.sbh as isize * 4);
if uv_stride < 0 {
f.lf.cdef_line[0][1] =
f.lf.cdef_lpf_line[1] =
ptr.offset(-(uv_stride * (f.sbh as isize * 4 - 1))) as *mut DynPixel;
f.lf.cdef_lpf_line[2] =
ptr.offset(-(uv_stride * (f.sbh as isize * 8 - 1))) as *mut DynPixel;
f.lf.cdef_line[0][2] =
ptr.offset(-(uv_stride * (f.sbh as isize * 8 - 3))) as *mut DynPixel;
f.lf.cdef_line[1][1] =
ptr.offset(-(uv_stride * (f.sbh as isize * 8 - 5))) as *mut DynPixel;
f.lf.cdef_line[1][2] =
ptr.offset(-(uv_stride * (f.sbh as isize * 8 - 7))) as *mut DynPixel;
} else {
f.lf.cdef_line[0][1] = ptr.offset(uv_stride * 0) as *mut DynPixel;
f.lf.cdef_line[0][2] = ptr.offset(uv_stride * 2) as *mut DynPixel;
f.lf.cdef_line[1][1] = ptr.offset(uv_stride * 4) as *mut DynPixel;
f.lf.cdef_line[1][2] = ptr.offset(uv_stride * 6) as *mut DynPixel;
}

if need_cdef_lpf_copy != 0 {
ptr = ptr.offset(uv_stride.abs() * f.sbh as isize * 8);
if y_stride < 0 {
f.lf.cdef_lpf_line[0] =
ptr.offset(-(y_stride * (f.sbh as isize * 4 - 1))) as *mut DynPixel;
} else {
f.lf.cdef_lpf_line[0] = ptr as *mut DynPixel;
}
ptr = ptr.offset(y_stride.abs() * f.sbh as isize * 4);
if uv_stride < 0 {
f.lf.cdef_lpf_line[1] =
ptr.offset(-(uv_stride * (f.sbh as isize * 4 - 1))) as *mut DynPixel;
f.lf.cdef_lpf_line[2] =
ptr.offset(-(uv_stride * (f.sbh as isize * 8 - 1))) as *mut DynPixel;
} else {
f.lf.cdef_lpf_line[1] = ptr as *mut DynPixel;
f.lf.cdef_lpf_line[2] = ptr.offset(uv_stride * f.sbh as isize * 4) as *mut DynPixel;
}
f.lf.cdef_lpf_line[1] = ptr as *mut DynPixel;
f.lf.cdef_lpf_line[2] = ptr.offset(uv_stride * f.sbh as isize * 4) as *mut DynPixel;
}

f.lf.cdef_buf_plane_sz[0] = y_stride as c_int * f.sbh * 4;
f.lf.cdef_buf_plane_sz[1] = uv_stride as c_int * f.sbh * 8;
f.lf.need_cdef_lpf_copy = need_cdef_lpf_copy;
f.lf.cdef_buf_sbh = f.sbh;
}

let sb128 = seq_hdr.sb128;
Expand Down
5 changes: 1 addition & 4 deletions src/internal.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -428,14 +428,12 @@ pub struct Rav1dFrameContext_lf {
pub level: Vec<[u8; 4]>,
pub mask: Vec<Av1Filter>, /* len = w*h */
pub lr_mask: Vec<Av1Restoration>,
pub cdef_buf_plane_sz: [c_int; 2], /* stride*sbh*4 */
pub cdef_buf_sbh: c_int,
pub lr_buf_plane_sz: [c_int; 2], /* (stride*sbh*4) << sb128 if n_tc > 1, else stride*4 */
pub lim_lut: Align16<Av1FilterLUT>,
pub last_sharpness: c_int,
pub lvl: [[[[u8; 2]; 8]; 4]; 8], /* [8 seg_id][4 dir][8 ref][2 is_gmv] */
pub tx_lpf_right_edge: Vec<u8>, /* len = h*2 */
pub cdef_line_buf: *mut u8,
pub cdef_line_buf: AlignedVec32<u8>, /* AlignedVec32<DynPixel> */
pub lr_line_buf: *mut u8,
randomPoison marked this conversation as resolved.
Show resolved Hide resolved
pub cdef_line: [[*mut DynPixel; 3]; 2], /* [2 pre/post][3 plane] */
pub cdef_lpf_line: [*mut DynPixel; 3], /* plane */
Expand All @@ -444,7 +442,6 @@ pub struct Rav1dFrameContext_lf {
// in-loop filter per-frame state keeping
pub start_of_tile_row: *mut u8,
pub start_of_tile_row_sz: c_int,
pub need_cdef_lpf_copy: c_int,
pub p: [*mut DynPixel; 3],
pub sr_p: [*mut DynPixel; 3],
pub restore_planes: c_int, // enum LrRestorePlanes
Expand Down
7 changes: 6 additions & 1 deletion src/lib.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -325,6 +325,11 @@ pub(crate) unsafe fn rav1d_open(c_out: &mut *mut Rav1dContext, s: &Rav1dSettings
f.task_thread.pending_tasks = Default::default();
}
(&mut f.task_thread.ttd as *mut Arc<TaskThreadData>).write(Arc::clone(&(*c).task_thread));
addr_of_mut!(f.lf.level).write(Default::default());
addr_of_mut!(f.lf.mask).write(Default::default());
addr_of_mut!(f.lf.lr_mask).write(Default::default());
addr_of_mut!(f.lf.tx_lpf_right_edge).write(Default::default());
addr_of_mut!(f.lf.cdef_line_buf).write(Default::default());
f.lf.last_sharpness = -(1 as c_int);
rav1d_refmvs_init(&mut f.rf);
n = n.wrapping_add(1);
Expand Down Expand Up @@ -903,7 +908,7 @@ impl Drop for Rav1dContext {
let _ = mem::take(&mut f.lf.tx_lpf_right_edge); // TODO: remove when context is owned
free(f.lf.start_of_tile_row as *mut c_void);
rav1d_refmvs_clear(&mut f.rf);
rav1d_free_aligned(f.lf.cdef_line_buf as *mut c_void);
let _ = mem::take(&mut f.lf.cdef_line_buf); // TODO: remove when context is owned
rav1d_free_aligned(f.lf.lr_line_buf as *mut c_void);
n_1 = n_1.wrapping_add(1);
}
Expand Down
Loading