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_task_thread: Replace atomic intrinsics #672

Merged
merged 2 commits into from
Jan 18, 2024
Merged

struct Rav1dFrameContext_task_thread: Replace atomic intrinsics #672

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
942dc58
Replace `Rav1dFrameContext_task_thread` atomics
rinon Jan 10, 2024
2e2a18b
Replace `pending_tasks.merge` with AtomicI32
rinon Jan 10, 2024
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
20 changes: 8 additions & 12 deletions src/decode.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4850,7 +4850,7 @@ unsafe fn rav1d_decode_frame_main(f: &mut Rav1dFrameContext) -> Rav1dResult {
pub(crate) unsafe fn rav1d_decode_frame_exit(f: &mut Rav1dFrameContext, retval: Rav1dResult) {
let c = &*f.c;
if !f.sr_cur.p.data[0].is_null() {
f.task_thread.error = 0;
f.task_thread.error = AtomicI32::new(0);
}
if c.n_fc > 1 && retval.is_err() && !f.frame_thread.cf.is_null() {
slice::from_raw_parts_mut(
Expand Down Expand Up @@ -4905,7 +4905,7 @@ pub(crate) unsafe fn rav1d_decode_frame(f: &mut Rav1dFrameContext) -> Rav1dResul
pthread_mutex_lock(&mut (*f.task_thread.ttd).lock);
pthread_cond_signal(&mut (*f.task_thread.ttd).cond);
if res.is_ok() {
while f.task_thread.done[0] == 0
while f.task_thread.done[0].load(Ordering::Relaxed) == 0
// TODO(kkysen) Make `.task_counter` an `AtomicI32`, but that requires recursively removing `impl Copy`s.
|| (*(addr_of_mut!(f.task_thread.task_counter) as *mut AtomicI32))
.load(Ordering::SeqCst)
Expand Down Expand Up @@ -4953,10 +4953,7 @@ pub unsafe fn rav1d_submit_frame(c: &mut Rav1dContext) -> Rav1dResult {
pthread_cond_wait(&mut f.task_thread.cond, &mut c.task_thread.lock);
}
let out_delayed = &mut *c.frame_thread.out_delayed.offset(next as isize);
if !out_delayed.p.data[0].is_null()
|| ::core::intrinsics::atomic_load_seqcst(&mut f.task_thread.error as *mut atomic_int)
!= 0
{
if !out_delayed.p.data[0].is_null() || f.task_thread.error.load(Ordering::SeqCst) != 0 {
let first = ::core::intrinsics::atomic_load_seqcst(&mut c.task_thread.first);
if first + 1 < c.n_fc {
::core::intrinsics::atomic_xadd_seqcst(&mut c.task_thread.first, 1);
Expand Down Expand Up @@ -5003,7 +5000,7 @@ pub unsafe fn rav1d_submit_frame(c: &mut Rav1dContext) -> Rav1dResult {
c: &mut Rav1dContext,
out_delayed: *mut Rav1dThreadPicture,
) {
f.task_thread.error = 1;
f.task_thread.error = AtomicI32::new(1);
rav1d_cdf_thread_unref(&mut f.in_cdf);
if f.frame_hdr.as_ref().unwrap().refresh_context != 0 {
rav1d_cdf_thread_unref(&mut f.out_cdf);
Expand Down Expand Up @@ -5203,14 +5200,13 @@ pub unsafe fn rav1d_submit_frame(c: &mut Rav1dContext) -> Rav1dResult {
f.sbh = f.bh + f.sb_step - 1 >> f.sb_shift;
f.b4_stride = (f.bw + 31 & !31) as ptrdiff_t;
f.bitdepth_max = (1 << f.cur.p.bpc) - 1;
*&mut f.task_thread.error = 0;
f.task_thread.error = AtomicI32::new(0);
let uses_2pass = (c.n_fc > 1) as c_int;
let cols = frame_hdr.tiling.cols;
let rows = frame_hdr.tiling.rows;
::core::intrinsics::atomic_store_seqcst(
&mut f.task_thread.task_counter,
cols * rows + f.sbh << uses_2pass,
);
f.task_thread
.task_counter
.store(cols * rows + f.sbh << uses_2pass, Ordering::SeqCst);

// ref_mvs
if frame_hdr.frame_type.is_inter_or_switch() || frame_hdr.allow_intrabc != 0 {
Expand Down
10 changes: 5 additions & 5 deletions src/internal.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -412,7 +412,6 @@ pub struct Rav1dFrameContext_lf {

#[repr(C)]
pub struct Rav1dFrameContext_task_thread_pending_tasks {
pub merge: atomic_int,
pub lock: pthread_mutex_t,
pub head: *mut Rav1dTask,
pub tail: *mut Rav1dTask,
Expand All @@ -428,12 +427,12 @@ pub(crate) struct Rav1dFrameContext_task_thread {
pub init_task: Rav1dTask,
pub num_tasks: c_int,
pub num_tile_tasks: c_int,
pub init_done: atomic_int,
pub done: [atomic_int; 2],
pub init_done: AtomicI32,
pub done: [AtomicI32; 2],
pub retval: Rav1dResult,
pub update_set: bool, // whether we need to update CDF reference
pub error: atomic_int,
pub task_counter: atomic_int,
pub error: AtomicI32,
rinon marked this conversation as resolved.
Show resolved Hide resolved
pub task_counter: AtomicI32,
pub task_head: *mut Rav1dTask,
pub task_tail: *mut Rav1dTask,
// Points to the task directly before the cur pointer in the queue.
Expand All @@ -442,6 +441,7 @@ pub(crate) struct Rav1dFrameContext_task_thread {
// [head;cur-1] when picking one for execution.
pub task_cur_prev: *mut Rav1dTask,
// async task insertion
pub pending_tasks_merge: AtomicI32,
pub pending_tasks: Rav1dFrameContext_task_thread_pending_tasks,
}

Expand Down
9 changes: 3 additions & 6 deletions src/lib.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,7 +20,6 @@ use crate::include::dav1d::headers::Rav1dFilmGrainData;
use crate::include::dav1d::headers::Rav1dSequenceHeader;
use crate::include::dav1d::picture::Dav1dPicture;
use crate::include::dav1d::picture::Rav1dPicture;
use crate::include::stdatomic::atomic_int;
use crate::src::align::Align64;
use crate::src::cdf::rav1d_cdf_thread_unref;
use crate::src::cpu::rav1d_init_cpu;
Expand Down Expand Up @@ -102,6 +101,7 @@ use std::process::abort;
use std::ptr;
use std::ptr::NonNull;
use std::slice;
use std::sync::atomic::AtomicI32;
use std::sync::atomic::Ordering;
use std::sync::Arc;
use std::sync::Once;
Expand Down Expand Up @@ -619,9 +619,7 @@ unsafe fn drain_picture(c: &mut Rav1dContext, out: &mut Rav1dPicture) -> Rav1dRe
let out_delayed: *mut Rav1dThreadPicture =
&mut *(c.frame_thread.out_delayed).offset(next as isize) as *mut Rav1dThreadPicture;
if !((*out_delayed).p.data[0]).is_null()
|| ::core::intrinsics::atomic_load_seqcst(
&mut (*f).task_thread.error as *mut atomic_int,
) != 0
|| (*f).task_thread.error.load(Ordering::SeqCst) != 0
{
let mut first: c_uint =
::core::intrinsics::atomic_load_seqcst(&mut c.task_thread.first);
Expand Down Expand Up @@ -909,8 +907,7 @@ pub(crate) unsafe fn rav1d_flush(c: *mut Rav1dContext) {
*fresh5 = 0 as *mut Rav1dTask;
*&mut (*((*c).fc).offset(i_1 as isize))
.task_thread
.pending_tasks
.merge = 0 as c_int;
.pending_tasks_merge = AtomicI32::new(0);
i_1 = i_1.wrapping_add(1);
}
*&mut (*c).task_thread.first = 0 as c_int as c_uint;
Expand Down
5 changes: 1 addition & 4 deletions src/obu.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -73,7 +73,6 @@ use crate::include::dav1d::headers::RAV1D_WM_TYPE_AFFINE;
use crate::include::dav1d::headers::RAV1D_WM_TYPE_IDENTITY;
use crate::include::dav1d::headers::RAV1D_WM_TYPE_ROT_ZOOM;
use crate::include::dav1d::headers::RAV1D_WM_TYPE_TRANSLATION;
use crate::include::stdatomic::atomic_int;
use crate::src::c_arc::CArc;
use crate::src::cdf::rav1d_cdf_thread_ref;
use crate::src::cdf::rav1d_cdf_thread_unref;
Expand Down Expand Up @@ -2541,9 +2540,7 @@ unsafe fn parse_obus(
}
let out_delayed = &mut *c.frame_thread.out_delayed.offset(next as isize);
if !(*out_delayed).p.data[0].is_null()
|| ::core::intrinsics::atomic_load_seqcst(
&mut (*f).task_thread.error as *mut atomic_int,
) != 0
|| (*f).task_thread.error.load(Ordering::SeqCst) != 0
{
let first = ::core::intrinsics::atomic_load_seqcst(&mut c.task_thread.first);
if first + 1 < c.n_fc {
Expand Down
Loading