From 6db4fd68439319df6389d0deaac327dafc6d3c2f Mon Sep 17 00:00:00 2001
From: rzumer <rzumer@tebako.net>
Date: Wed, 1 Aug 2018 19:54:47 -0400
Subject: [PATCH] Add optional comparative benchmarks
Run with --features=comparative_bench
Applies to intra prediction tests and disables native function inlining
Also refactor benchmarks and reduce code duplication
---
Cargo.toml | 1 +
README.md | 10 +
benches/bench.rs | 364 ++-------------------------------
benches/comparative/mod.rs | 30 +++
benches/comparative/predict.rs | 124 +++++++++++
benches/predict.rs | 102 +++++++++
src/partition.rs | 6 +-
src/predict.rs | 28 ++-
8 files changed, 309 insertions(+), 356 deletions(-)
mode change 100644 => 100755 Cargo.toml
mode change 100644 => 100755 benches/bench.rs
create mode 100755 benches/comparative/mod.rs
create mode 100755 benches/comparative/predict.rs
create mode 100755 benches/predict.rs
mode change 100644 => 100755 src/partition.rs
mode change 100644 => 100755 src/predict.rs
diff --git a/Cargo.toml b/Cargo.toml
old mode 100644
new mode 100755
index 1c48de10..f65d26bd
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -8,6 +8,7 @@ include = ["/src/**", "/aom_build/**", "/Cargo.toml"]
[features]
repl = ["rustyline"]
decode_test = ["bindgen"]
+comparative_bench = []
[dependencies]
bitstream-io = "0.6"
diff --git a/README.md b/README.md
index 2f0c8ec5..eb410ed7 100755
--- a/README.md
+++ b/README.md
@@ -104,6 +104,16 @@ Run encode-decode integration tests with:
cargo test --release --features=decode_test -- --ignored
```
+Run regular benchmarks with:
+```
+cargo bench
+```
+
+Run comparative benchmarks with:
+```
+cargo bench --features=comparative_bench
+```
+
# Getting in Touch
Come chat with us on the IRC channel #daala on Freenode! If you don't have IRC set
diff --git a/benches/bench.rs b/benches/bench.rs
old mode 100644
new mode 100755
index c8f5bb71..8125e0a6
--- a/benches/bench.rs
+++ b/benches/bench.rs
@@ -9,341 +9,20 @@
#[macro_use]
extern crate bencher;
-extern crate libc;
extern crate rand;
extern crate rav1e;
-use bencher::*;
-use rand::{ChaChaRng, Rng, SeedableRng};
-use rav1e::predict::*;
-
-extern {
- fn highbd_dc_predictor(
- dst: *mut u16, stride: libc::ptrdiff_t, bw: libc::c_int, bh: libc::c_int,
- above: *const u16, left: *const u16, bd: libc::c_int
- );
-
- fn highbd_h_predictor(
- dst: *mut u16, stride: libc::ptrdiff_t, bw: libc::c_int, bh: libc::c_int,
- above: *const u16, left: *const u16, bd: libc::c_int
- );
-
- fn highbd_v_predictor(
- dst: *mut u16, stride: libc::ptrdiff_t, bw: libc::c_int, bh: libc::c_int,
- above: *const u16, left: *const u16, bd: libc::c_int
- );
-
- fn highbd_paeth_predictor(
- dst: *mut u16, stride: libc::ptrdiff_t, bw: libc::c_int, bh: libc::c_int,
- above: *const u16, left: *const u16, bd: libc::c_int
- );
-
- fn highbd_smooth_predictor(
- dst: *mut u16, stride: libc::ptrdiff_t, bw: libc::c_int, bh: libc::c_int,
- above: *const u16, left: *const u16, bd: libc::c_int
- );
-
- fn highbd_smooth_h_predictor(
- dst: *mut u16, stride: libc::ptrdiff_t, bw: libc::c_int, bh: libc::c_int,
- above: *const u16, left: *const u16, bd: libc::c_int
- );
-
- fn highbd_smooth_v_predictor(
- dst: *mut u16, stride: libc::ptrdiff_t, bw: libc::c_int, bh: libc::c_int,
- above: *const u16, left: *const u16, bd: libc::c_int
- );
-}
-
-#[inline(always)]
-fn pred_dc_4x4(
- output: &mut [u16], stride: usize, above: &[u16], left: &[u16]
-) {
- unsafe {
- highbd_dc_predictor(
- output.as_mut_ptr(),
- stride as libc::ptrdiff_t,
- 4,
- 4,
- above.as_ptr(),
- left.as_ptr(),
- 8
- );
- }
-}
-
-#[inline(always)]
-fn pred_h_4x4(output: &mut [u16], stride: usize, above: &[u16], left: &[u16]) {
- unsafe {
- highbd_h_predictor(
- output.as_mut_ptr(),
- stride as libc::ptrdiff_t,
- 4,
- 4,
- above.as_ptr(),
- left.as_ptr(),
- 8
- );
- }
-}
-
-#[inline(always)]
-fn pred_v_4x4(output: &mut [u16], stride: usize, above: &[u16], left: &[u16]) {
- unsafe {
- highbd_v_predictor(
- output.as_mut_ptr(),
- stride as libc::ptrdiff_t,
- 4,
- 4,
- above.as_ptr(),
- left.as_ptr(),
- 8
- );
- }
-}
-
-#[inline(always)]
-fn pred_paeth_4x4(
- output: &mut [u16], stride: usize, above: &[u16], left: &[u16]
-) {
- unsafe {
- highbd_paeth_predictor(
- output.as_mut_ptr(),
- stride as libc::ptrdiff_t,
- 4,
- 4,
- above.as_ptr(),
- left.as_ptr(),
- 8
- );
- }
-}
-
-#[inline(always)]
-fn pred_smooth_4x4(
- output: &mut [u16], stride: usize, above: &[u16], left: &[u16]
-) {
- unsafe {
- highbd_smooth_predictor(
- output.as_mut_ptr(),
- stride as libc::ptrdiff_t,
- 4,
- 4,
- above.as_ptr(),
- left.as_ptr(),
- 8
- );
- }
-}
-
-#[inline(always)]
-fn pred_smooth_h_4x4(
- output: &mut [u16], stride: usize, above: &[u16], left: &[u16]
-) {
- unsafe {
- highbd_smooth_h_predictor(
- output.as_mut_ptr(),
- stride as libc::ptrdiff_t,
- 4,
- 4,
- above.as_ptr(),
- left.as_ptr(),
- 8
- );
- }
-}
-
-#[inline(always)]
-fn pred_smooth_v_4x4(
- output: &mut [u16], stride: usize, above: &[u16], left: &[u16]
-) {
- unsafe {
- highbd_smooth_v_predictor(
- output.as_mut_ptr(),
- stride as libc::ptrdiff_t,
- 4,
- 4,
- above.as_ptr(),
- left.as_ptr(),
- 8
- );
- }
-}
-
-const MAX_ITER: usize = 50000;
-
-fn setup_pred(ra: &mut ChaChaRng) -> (Vec<u16>, Vec<u16>, Vec<u16>) {
- let output = vec![0u16; 32 * 32];
- let above: Vec<u16> = (0..32).map(|_| ra.gen()).collect();
- let left: Vec<u16> = (0..32).map(|_| ra.gen()).collect();
-
- (above, left, output)
-}
-
-fn intra_dc_pred_native(b: &mut Bencher) {
- let mut ra = ChaChaRng::from_seed([0; 32]);
- let (above, left, mut output) = setup_pred(&mut ra);
-
- b.iter(|| {
- for _ in 0..MAX_ITER {
- Block4x4::pred_dc(&mut output, 32, &above[..4], &left[..4]);
- }
- })
-}
-
-fn intra_dc_pred_aom(b: &mut Bencher) {
- let mut ra = ChaChaRng::from_seed([0; 32]);
- let (above, left, mut output) = setup_pred(&mut ra);
-
- b.iter(|| {
- for _ in 0..MAX_ITER {
- pred_dc_4x4(&mut output, 32, &above[..4], &left[..4]);
- }
- })
-}
-
-fn intra_h_pred_native(b: &mut Bencher) {
- let mut ra = ChaChaRng::from_seed([0; 32]);
- let (_above, left, mut output) = setup_pred(&mut ra);
-
- b.iter(|| {
- for _ in 0..MAX_ITER {
- Block4x4::pred_h(&mut output, 32, &left[..4]);
- }
- })
-}
-
-fn intra_h_pred_aom(b: &mut Bencher) {
- let mut ra = ChaChaRng::from_seed([0; 32]);
- let (above, left, mut output) = setup_pred(&mut ra);
-
- b.iter(|| {
- for _ in 0..MAX_ITER {
- pred_h_4x4(&mut output, 32, &above[..4], &left[..4]);
- }
- })
-}
-
-fn intra_v_pred_native(b: &mut Bencher) {
- let mut ra = ChaChaRng::from_seed([0; 32]);
- let (above, _left, mut output) = setup_pred(&mut ra);
-
- b.iter(|| {
- for _ in 0..MAX_ITER {
- Block4x4::pred_v(&mut output, 32, &above[..4]);
- }
- })
-}
-
-fn intra_v_pred_aom(b: &mut Bencher) {
- let mut ra = ChaChaRng::from_seed([0; 32]);
- let (above, left, mut output) = setup_pred(&mut ra);
-
- b.iter(|| {
- for _ in 0..MAX_ITER {
- pred_v_4x4(&mut output, 32, &above[..4], &left[..4]);
- }
- })
-}
-
-fn intra_paeth_pred_native(b: &mut Bencher) {
- let mut ra = ChaChaRng::from_seed([0; 32]);
- let (above, left, mut output) = setup_pred(&mut ra);
- let above_left = unsafe { *above.as_ptr().offset(-1) };
-
- b.iter(|| {
- for _ in 0..MAX_ITER {
- Block4x4::pred_paeth(
- &mut output,
- 32,
- &above[..4],
- &left[..4],
- above_left
- );
- }
- })
-}
-
-fn intra_paeth_pred_aom(b: &mut Bencher) {
- let mut ra = ChaChaRng::from_seed([0; 32]);
- let (above, left, mut output) = setup_pred(&mut ra);
-
- b.iter(|| {
- for _ in 0..MAX_ITER {
- pred_paeth_4x4(&mut output, 32, &above[..4], &left[..4]);
- }
- })
-}
-
-fn intra_smooth_pred_native(b: &mut Bencher) {
- let mut ra = ChaChaRng::from_seed([0; 32]);
- let (above, left, mut output) = setup_pred(&mut ra);
-
- b.iter(|| {
- for _ in 0..MAX_ITER {
- Block4x4::pred_smooth(&mut output, 32, &above[..4], &left[..4], 8);
- }
- })
-}
-
-fn intra_smooth_pred_aom(b: &mut Bencher) {
- let mut ra = ChaChaRng::from_seed([0; 32]);
- let (above, left, mut output) = setup_pred(&mut ra);
-
- b.iter(|| {
- for _ in 0..MAX_ITER {
- pred_smooth_4x4(&mut output, 32, &above[..4], &left[..4]);
- }
- })
-}
-
-fn intra_smooth_h_pred_native(b: &mut Bencher) {
- let mut ra = ChaChaRng::from_seed([0; 32]);
- let (above, left, mut output) = setup_pred(&mut ra);
-
- b.iter(|| {
- for _ in 0..MAX_ITER {
- Block4x4::pred_smooth_h(&mut output, 32, &above[..4], &left[..4], 8);
- }
- })
-}
-
-fn intra_smooth_h_pred_aom(b: &mut Bencher) {
- let mut ra = ChaChaRng::from_seed([0; 32]);
- let (above, left, mut output) = setup_pred(&mut ra);
-
- b.iter(|| {
- for _ in 0..MAX_ITER {
- pred_smooth_h_4x4(&mut output, 32, &above[..4], &left[..4]);
- }
- })
-}
-
-fn intra_smooth_v_pred_native(b: &mut Bencher) {
- let mut ra = ChaChaRng::from_seed([0; 32]);
- let (above, left, mut output) = setup_pred(&mut ra);
-
- b.iter(|| {
- for _ in 0..MAX_ITER {
- Block4x4::pred_smooth_v(&mut output, 32, &above[..4], &left[..4], 8);
- }
- })
-}
-
-fn intra_smooth_v_pred_aom(b: &mut Bencher) {
- let mut ra = ChaChaRng::from_seed([0; 32]);
- let (above, left, mut output) = setup_pred(&mut ra);
-
- b.iter(|| {
- for _ in 0..MAX_ITER {
- pred_smooth_v_4x4(&mut output, 32, &above[..4], &left[..4]);
- }
- })
-}
+mod predict;
+use bencher::*;
+use rav1e::*;
use rav1e::context::*;
use rav1e::ec;
use rav1e::partition::*;
-use rav1e::*;
+use rav1e::predict::*;
+
+#[cfg(feature = "comparative_bench")]
+mod comparative;
struct WriteB {
tx_size: TxSize,
@@ -423,21 +102,18 @@ fn write_b_bench(b: &mut Bencher, tx_size: TxSize, qindex: usize) {
}
benchmark_group!(
- intra,
- intra_dc_pred_native,
- intra_dc_pred_aom,
- intra_h_pred_native,
- intra_h_pred_aom,
- intra_v_pred_native,
- intra_v_pred_aom,
- intra_paeth_pred_native,
- intra_paeth_pred_aom,
- intra_smooth_pred_native,
- intra_smooth_pred_aom,
- intra_smooth_h_pred_native,
- intra_smooth_h_pred_aom,
- intra_smooth_v_pred_native,
- intra_smooth_v_pred_aom
+ intra_prediction,
+ predict::intra_dc_4x4,
+ predict::intra_h_4x4,
+ predict::intra_v_4x4,
+ predict::intra_paeth_4x4,
+ predict::intra_smooth_4x4,
+ predict::intra_smooth_h_4x4,
+ predict::intra_smooth_v_4x4
);
-benchmark_main!(intra, write_b);
+#[cfg(feature = "comparative_bench")]
+benchmark_main!(comparative::intra_prediction);
+
+#[cfg(not(feature = "comparative_bench"))]
+benchmark_main!(write_b, intra_prediction);
diff --git a/benches/comparative/mod.rs b/benches/comparative/mod.rs
new file mode 100755
index 00000000..691e4ddb
--- /dev/null
+++ b/benches/comparative/mod.rs
@@ -0,0 +1,30 @@
+// Copyright (c) 2017-2018, The rav1e contributors. All rights reserved
+//
+// This source code is subject to the terms of the BSD 2 Clause License and
+// the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+// was not distributed with this source code in the LICENSE file, you can
+// obtain it at www.aomedia.org/license/software. If the Alliance for Open
+// Media Patent License 1.0 was not distributed with this source code in the
+// PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+
+extern crate libc;
+
+mod predict;
+
+benchmark_group!(
+ intra_prediction,
+ predict::intra_dc_4x4_native,
+ predict::intra_dc_4x4_aom,
+ predict::intra_h_4x4_native,
+ predict::intra_h_4x4_aom,
+ predict::intra_v_4x4_native,
+ predict::intra_v_4x4_aom,
+ predict::intra_paeth_4x4_native,
+ predict::intra_paeth_4x4_aom,
+ predict::intra_smooth_4x4_native,
+ predict::intra_smooth_4x4_aom,
+ predict::intra_smooth_h_4x4_native,
+ predict::intra_smooth_h_4x4_aom,
+ predict::intra_smooth_v_4x4_native,
+ predict::intra_smooth_v_4x4_aom
+);
diff --git a/benches/comparative/predict.rs b/benches/comparative/predict.rs
new file mode 100755
index 00000000..f2d76589
--- /dev/null
+++ b/benches/comparative/predict.rs
@@ -0,0 +1,124 @@
+// Copyright (c) 2017-2018, The rav1e contributors. All rights reserved
+//
+// This source code is subject to the terms of the BSD 2 Clause License and
+// the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+// was not distributed with this source code in the LICENSE file, you can
+// obtain it at www.aomedia.org/license/software. If the Alliance for Open
+// Media Patent License 1.0 was not distributed with this source code in the
+// PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+
+use bencher::*;
+use comparative::libc;
+use predict as predict_native;
+use predict::*;
+use rand::{ChaChaRng, SeedableRng};
+
+extern {
+ fn highbd_dc_predictor(
+ dst: *mut u16, stride: libc::ptrdiff_t, bw: libc::c_int, bh: libc::c_int,
+ above: *const u16, left: *const u16, bd: libc::c_int
+ );
+
+ fn highbd_h_predictor(
+ dst: *mut u16, stride: libc::ptrdiff_t, bw: libc::c_int, bh: libc::c_int,
+ above: *const u16, left: *const u16, bd: libc::c_int
+ );
+
+ fn highbd_v_predictor(
+ dst: *mut u16, stride: libc::ptrdiff_t, bw: libc::c_int, bh: libc::c_int,
+ above: *const u16, left: *const u16, bd: libc::c_int
+ );
+
+ fn highbd_paeth_predictor(
+ dst: *mut u16, stride: libc::ptrdiff_t, bw: libc::c_int, bh: libc::c_int,
+ above: *const u16, left: *const u16, bd: libc::c_int
+ );
+
+ fn highbd_smooth_predictor(
+ dst: *mut u16, stride: libc::ptrdiff_t, bw: libc::c_int, bh: libc::c_int,
+ above: *const u16, left: *const u16, bd: libc::c_int
+ );
+
+ fn highbd_smooth_h_predictor(
+ dst: *mut u16, stride: libc::ptrdiff_t, bw: libc::c_int, bh: libc::c_int,
+ above: *const u16, left: *const u16, bd: libc::c_int
+ );
+
+ fn highbd_smooth_v_predictor(
+ dst: *mut u16, stride: libc::ptrdiff_t, bw: libc::c_int, bh: libc::c_int,
+ above: *const u16, left: *const u16, bd: libc::c_int
+ );
+}
+
+fn predict_intra_4x4_aom(
+ b: &mut Bencher,
+ predictor: unsafe extern "C" fn(*mut u16, libc::ptrdiff_t, libc::c_int, libc::c_int, *const u16, *const u16, libc::c_int))
+{
+ let mut rng = ChaChaRng::from_seed([0; 32]);
+ let (mut block, above_context, left_context) = generate_block(&mut rng);
+
+ b.iter(|| {
+ for _ in 0..MAX_ITER {
+ unsafe {
+ predictor(block.as_mut_ptr(), BLOCK_SIZE.width() as libc::ptrdiff_t,
+ 4, 4, above_context.as_ptr(), left_context.as_ptr(), 8);
+ }
+ }
+ })
+}
+
+pub fn intra_dc_4x4_native(b: &mut Bencher) {
+ predict_native::intra_dc_4x4(b);
+}
+
+pub fn intra_dc_4x4_aom(b: &mut Bencher) {
+ predict_intra_4x4_aom(b, highbd_dc_predictor);
+}
+
+pub fn intra_h_4x4_native(b: &mut Bencher) {
+ predict_native::intra_h_4x4(b);
+}
+
+pub fn intra_h_4x4_aom(b: &mut Bencher) {
+ predict_intra_4x4_aom(b, highbd_h_predictor);
+}
+
+pub fn intra_v_4x4_native(b: &mut Bencher) {
+ predict_native::intra_v_4x4(b);
+}
+
+pub fn intra_v_4x4_aom(b: &mut Bencher) {
+ predict_intra_4x4_aom(b, highbd_v_predictor);
+}
+
+pub fn intra_paeth_4x4_native(b: &mut Bencher) {
+ predict_native::intra_paeth_4x4(b);
+}
+
+pub fn intra_paeth_4x4_aom(b: &mut Bencher) {
+ predict_intra_4x4_aom(b, highbd_paeth_predictor);
+}
+
+pub fn intra_smooth_4x4_native(b: &mut Bencher) {
+ predict_native::intra_smooth_4x4(b);
+}
+
+pub fn intra_smooth_4x4_aom(b: &mut Bencher) {
+ predict_intra_4x4_aom(b, highbd_smooth_predictor);
+}
+
+pub fn intra_smooth_h_4x4_native(b: &mut Bencher) {
+ predict_native::intra_smooth_h_4x4(b);
+}
+
+pub fn intra_smooth_h_4x4_aom(b: &mut Bencher) {
+ predict_intra_4x4_aom(b, highbd_smooth_h_predictor);
+}
+
+pub fn intra_smooth_v_4x4_native(b: &mut Bencher) {
+ predict_native::intra_smooth_v_4x4(b);
+}
+
+pub fn intra_smooth_v_4x4_aom(b: &mut Bencher) {
+ predict_intra_4x4_aom(b, highbd_smooth_v_predictor);
+}
diff --git a/benches/predict.rs b/benches/predict.rs
new file mode 100755
index 00000000..74c7c95f
--- /dev/null
+++ b/benches/predict.rs
@@ -0,0 +1,102 @@
+// Copyright (c) 2017-2018, The rav1e contributors. All rights reserved
+//
+// This source code is subject to the terms of the BSD 2 Clause License and
+// the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+// was not distributed with this source code in the LICENSE file, you can
+// obtain it at www.aomedia.org/license/software. If the Alliance for Open
+// Media Patent License 1.0 was not distributed with this source code in the
+// PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+
+use bencher::*;
+use rand::{ChaChaRng, Rng, SeedableRng};
+use rav1e::partition::BlockSize;
+use rav1e::predict::{Block4x4, Intra};
+
+pub const MAX_ITER: usize = 50000;
+pub const BLOCK_SIZE: BlockSize = BlockSize::BLOCK_32X32;
+
+pub fn generate_block(rng: &mut ChaChaRng) -> (Vec<u16>, Vec<u16>, Vec<u16>) {
+ let block = vec![0u16; BLOCK_SIZE.width() * BLOCK_SIZE.height()];
+ let above_context: Vec<u16> = (0..BLOCK_SIZE.height()).map(|_| rng.gen()).collect();
+ let left_context: Vec<u16> = (0..BLOCK_SIZE.width()).map(|_| rng.gen()).collect();
+
+ (block, above_context, left_context)
+}
+
+pub fn intra_dc_4x4(b: &mut Bencher) {
+ let mut ra = ChaChaRng::from_seed([0; 32]);
+ let (mut block, above, left) = generate_block(&mut ra);
+
+ b.iter(|| {
+ for _ in 0..MAX_ITER {
+ Block4x4::pred_dc(&mut block, BLOCK_SIZE.width(), &above[..4], &left[..4]);
+ }
+ })
+}
+
+pub fn intra_h_4x4(b: &mut Bencher) {
+ let mut rng = ChaChaRng::from_seed([0; 32]);
+ let (mut block, _above, left) = generate_block(&mut rng);
+
+ b.iter(|| {
+ for _ in 0..MAX_ITER {
+ Block4x4::pred_h(&mut block, BLOCK_SIZE.width(), &left[..4]);
+ }
+ })
+}
+
+pub fn intra_v_4x4(b: &mut Bencher) {
+ let mut rng = ChaChaRng::from_seed([0; 32]);
+ let (mut block, above, _left) = generate_block(&mut rng);
+
+ b.iter(|| {
+ for _ in 0..MAX_ITER {
+ Block4x4::pred_v(&mut block, BLOCK_SIZE.width(), &above[..4]);
+ }
+ })
+}
+
+pub fn intra_paeth_4x4(b: &mut Bencher) {
+ let mut rng = ChaChaRng::from_seed([0; 32]);
+ let (mut block, above, left) = generate_block(&mut rng);
+ let above_left = unsafe { *above.as_ptr().offset(-1) };
+
+ b.iter(|| {
+ for _ in 0..MAX_ITER {
+ Block4x4::pred_paeth(&mut block, BLOCK_SIZE.width(), &above[..4], &left[..4], above_left);
+ }
+ })
+}
+
+pub fn intra_smooth_4x4(b: &mut Bencher) {
+ let mut rng = ChaChaRng::from_seed([0; 32]);
+ let (mut block, above, left) = generate_block(&mut rng);
+
+ b.iter(|| {
+ for _ in 0..MAX_ITER {
+ Block4x4::pred_smooth(&mut block, BLOCK_SIZE.width(), &above[..4], &left[..4]);
+ }
+ })
+}
+
+pub fn intra_smooth_h_4x4(b: &mut Bencher) {
+ let mut rng = ChaChaRng::from_seed([0; 32]);
+ let (mut block, above, left) = generate_block(&mut rng);
+
+ b.iter(|| {
+ for _ in 0..MAX_ITER {
+ Block4x4::pred_smooth_h(&mut block, BLOCK_SIZE.width(), &above[..4], &left[..4]);
+ }
+ })
+}
+
+pub fn intra_smooth_v_4x4(b: &mut Bencher) {
+ let mut rng = ChaChaRng::from_seed([0; 32]);
+ let (mut block, above, left) = generate_block(&mut rng);
+
+ b.iter(|| {
+ for _ in 0..MAX_ITER {
+ Block4x4::pred_smooth_v(&mut block, BLOCK_SIZE.width(), &above[..4], &left[..4]);
+ }
+ })
+}
diff --git a/src/partition.rs b/src/partition.rs
old mode 100644
new mode 100755
index fbb88853..57016c15
--- a/src/partition.rs
+++ b/src/partition.rs
@@ -435,11 +435,11 @@ impl PredictionMode {
PredictionMode::PAETH_PRED =>
B::pred_paeth(slice, stride, above_slice, left_slice, above[0]),
PredictionMode::SMOOTH_PRED =>
- B::pred_smooth(slice, stride, above_slice, left_slice, 8),
+ B::pred_smooth(slice, stride, above_slice, left_slice),
PredictionMode::SMOOTH_H_PRED =>
- B::pred_smooth_h(slice, stride, above_slice, left_slice, 8),
+ B::pred_smooth_h(slice, stride, above_slice, left_slice),
PredictionMode::SMOOTH_V_PRED =>
- B::pred_smooth_v(slice, stride, above_slice, left_slice, 8),
+ B::pred_smooth_v(slice, stride, above_slice, left_slice),
_ => unimplemented!()
}
}
diff --git a/src/predict.rs b/src/predict.rs
old mode 100644
new mode 100755
index ea6ac12e..6fc452e9
--- a/src/predict.rs
+++ b/src/predict.rs
@@ -178,6 +178,7 @@ impl Dim for Block32x32 {
}
pub trait Intra: Dim {
+ #[cfg_attr(feature = "comparative_bench", inline(never))]
fn pred_dc(output: &mut [u16], stride: usize, above: &[u16], left: &[u16]) {
let edges = left[..Self::H].iter().chain(above[..Self::W].iter());
let len = (Self::W + Self::H) as u32;
@@ -191,6 +192,7 @@ pub trait Intra: Dim {
}
}
+ #[cfg_attr(feature = "comparative_bench", inline(never))]
fn pred_dc_128(output: &mut [u16], stride: usize) {
for y in 0..Self::H {
for x in 0..Self::W {
@@ -199,6 +201,7 @@ pub trait Intra: Dim {
}
}
+ #[cfg_attr(feature = "comparative_bench", inline(never))]
fn pred_dc_left(
output: &mut [u16], stride: usize, above: &[u16], left: &[u16]
) {
@@ -215,6 +218,7 @@ pub trait Intra: Dim {
}
}
+ #[cfg_attr(feature = "comparative_bench", inline(never))]
fn pred_dc_top(
output: &mut [u16], stride: usize, above: &[u16], left: &[u16]
) {
@@ -231,6 +235,7 @@ pub trait Intra: Dim {
}
}
+ #[cfg_attr(feature = "comparative_bench", inline(never))]
fn pred_h(output: &mut [u16], stride: usize, left: &[u16]) {
for (line, l) in output.chunks_mut(stride).zip(left[..Self::H].iter()) {
for v in &mut line[..Self::W] {
@@ -239,12 +244,14 @@ pub trait Intra: Dim {
}
}
+ #[cfg_attr(feature = "comparative_bench", inline(never))]
fn pred_v(output: &mut [u16], stride: usize, above: &[u16]) {
for line in output.chunks_mut(stride).take(Self::H) {
line[..Self::W].clone_from_slice(&above[..Self::W])
}
}
+ #[cfg_attr(feature = "comparative_bench", inline(never))]
fn pred_paeth(
output: &mut [u16], stride: usize, above: &[u16], left: &[u16],
above_left: u16
@@ -275,8 +282,9 @@ pub trait Intra: Dim {
}
}
+ #[cfg_attr(feature = "comparative_bench", inline(never))]
fn pred_smooth(
- output: &mut [u16], stride: usize, above: &[u16], left: &[u16], _bd: u8
+ output: &mut [u16], stride: usize, above: &[u16], left: &[u16]
) {
let below_pred = left[Self::H - 1]; // estimated by bottom-left pixel
let right_pred = above[Self::W - 1]; // estimated by top-right pixel
@@ -325,8 +333,9 @@ pub trait Intra: Dim {
}
}
+ #[cfg_attr(feature = "comparative_bench", inline(never))]
fn pred_smooth_h(
- output: &mut [u16], stride: usize, above: &[u16], left: &[u16], _bd: u8
+ output: &mut [u16], stride: usize, above: &[u16], left: &[u16]
) {
let right_pred = above[Self::W - 1]; // estimated by top-right pixel
let sm_weights = &sm_weight_arrays[Self::W..];
@@ -361,8 +370,9 @@ pub trait Intra: Dim {
}
}
+ #[cfg_attr(feature = "comparative_bench", inline(never))]
fn pred_smooth_v(
- output: &mut [u16], stride: usize, above: &[u16], left: &[u16], _bd: u8
+ output: &mut [u16], stride: usize, above: &[u16], left: &[u16]
) {
let below_pred = left[Self::H - 1]; // estimated by bottom-left pixel
let sm_weights = &sm_weight_arrays[Self::H..];
@@ -576,7 +586,7 @@ pub mod test {
let (above, left, mut o1, mut o2) = setup_pred(ra);
pred_smooth_4x4(&mut o1, 32, &above[..4], &left[..4]);
- Block4x4::pred_smooth(&mut o2, 32, &above[..4], &left[..4], 8);
+ Block4x4::pred_smooth(&mut o2, 32, &above[..4], &left[..4]);
(o1, o2)
}
@@ -585,7 +595,7 @@ pub mod test {
let (above, left, mut o1, mut o2) = setup_pred(ra);
pred_smooth_h_4x4(&mut o1, 32, &above[..4], &left[..4]);
- Block4x4::pred_smooth_h(&mut o2, 32, &above[..4], &left[..4], 8);
+ Block4x4::pred_smooth_h(&mut o2, 32, &above[..4], &left[..4]);
(o1, o2)
}
@@ -594,7 +604,7 @@ pub mod test {
let (above, left, mut o1, mut o2) = setup_pred(ra);
pred_smooth_v_4x4(&mut o1, 32, &above[..4], &left[..4]);
- Block4x4::pred_smooth_v(&mut o2, 32, &above[..4], &left[..4], 8);
+ Block4x4::pred_smooth_v(&mut o2, 32, &above[..4], &left[..4]);
(o1, o2)
}
@@ -686,7 +696,7 @@ pub mod test {
}
}
- Block4x4::pred_smooth(&mut o, 32, &above[..4], &left[..4], 12);
+ Block4x4::pred_smooth(&mut o, 32, &above[..4], &left[..4]);
for l in o.chunks(32).take(4) {
for v in l[..4].iter() {
@@ -694,7 +704,7 @@ pub mod test {
}
}
- Block4x4::pred_smooth_h(&mut o, 32, &above[..4], &left[..4], 12);
+ Block4x4::pred_smooth_h(&mut o, 32, &above[..4], &left[..4]);
for l in o.chunks(32).take(4) {
for v in l[..4].iter() {
@@ -702,7 +712,7 @@ pub mod test {
}
}
- Block4x4::pred_smooth_v(&mut o, 32, &above[..4], &left[..4], 12);
+ Block4x4::pred_smooth_v(&mut o, 32, &above[..4], &left[..4]);
for l in o.chunks(32).take(4) {
for v in l[..4].iter() {
--
GitLab