mirror of
https://github.com/mozilla/gecko-dev.git
synced 2024-11-24 13:21:05 +00:00
servo: Merge #7527 - Use SIMD in fast path for advance_for_char_range (from bjwbell:simd_advance_for_char_range); r=metajack
In advance_for_char_range add a fast SIMD code path for the the common case where there are no detailed glyphs. r? @mbrubeck Source-Repo: https://github.com/servo/servo Source-Revision: c83825938346f794790bcd1343b7375e9e34f443
This commit is contained in:
parent
a7948af624
commit
6a96bf0eb9
@ -103,3 +103,18 @@ git = "https://github.com/servo/rust-freetype"
|
||||
|
||||
[target.x86_64-apple-darwin.dependencies.core-text]
|
||||
git = "https://github.com/servo/core-text-rs"
|
||||
|
||||
[target.x86_64-unknown-linux-gnu.dependencies.simd]
|
||||
git = "https://github.com/huonw/simd"
|
||||
|
||||
[target.x86_64-apple-darwin.dependencies.simd]
|
||||
git = "https://github.com/huonw/simd"
|
||||
|
||||
[target.aarch64-unknown-linux-gnu.dependencies.simd]
|
||||
git = "https://github.com/huonw/simd"
|
||||
|
||||
[target.x86_64-pc-windows-gnu.dependencies.simd]
|
||||
git = "https://github.com/huonw/simd"
|
||||
|
||||
[target.x86_64-pc-windows-msvc.dependencies.simd]
|
||||
git = "https://github.com/huonw/simd"
|
||||
|
@ -5,6 +5,10 @@
|
||||
#![feature(arc_weak)]
|
||||
#![feature(box_raw)]
|
||||
#![feature(box_syntax)]
|
||||
|
||||
// For simd (currently x86_64/aarch64)
|
||||
#![cfg_attr(any(target_arch = "x86_64", target_arch = "aarch64"), feature(convert))]
|
||||
|
||||
#![feature(custom_attribute)]
|
||||
#![feature(custom_derive)]
|
||||
#![feature(hashmap_hasher)]
|
||||
@ -40,6 +44,10 @@ extern crate net_traits;
|
||||
extern crate util;
|
||||
extern crate msg;
|
||||
extern crate rand;
|
||||
|
||||
#[cfg(any(target_arch = "x86_64", target_arch = "aarch64"))]
|
||||
extern crate simd;
|
||||
|
||||
extern crate smallvec;
|
||||
extern crate string_cache;
|
||||
extern crate style;
|
||||
|
@ -3,6 +3,10 @@
|
||||
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
|
||||
|
||||
use euclid::point::Point2D;
|
||||
|
||||
#[cfg(any(target_arch = "x86_64", target_arch = "aarch64"))]
|
||||
use simd::u32x4;
|
||||
|
||||
use std::cmp::{Ordering, PartialOrd};
|
||||
use std::mem;
|
||||
use std::u16;
|
||||
@ -406,12 +410,16 @@ pub struct GlyphStore {
|
||||
// TODO(pcwalton): Allocation of this buffer is expensive. Consider a small-vector
|
||||
// optimization.
|
||||
/// A buffer of glyphs within the text run, in the order in which they
|
||||
/// appear in the input text
|
||||
/// appear in the input text.
|
||||
/// Any changes will also need to be reflected in
|
||||
/// transmute_entry_buffer_to_u32_buffer().
|
||||
entry_buffer: Vec<GlyphEntry>,
|
||||
/// A store of the detailed glyph data. Detailed glyphs contained in the
|
||||
/// `entry_buffer` point to locations in this data structure.
|
||||
detail_store: DetailedGlyphStore,
|
||||
|
||||
/// Used to check if fast path should be used in glyph iteration.
|
||||
has_detailed_glyphs: bool,
|
||||
is_whitespace: bool,
|
||||
is_rtl: bool,
|
||||
}
|
||||
@ -434,6 +442,7 @@ impl<'a> GlyphStore {
|
||||
GlyphStore {
|
||||
entry_buffer: vec![GlyphEntry::initial(); length],
|
||||
detail_store: DetailedGlyphStore::new(),
|
||||
has_detailed_glyphs: false,
|
||||
is_whitespace: is_whitespace,
|
||||
is_rtl: is_rtl,
|
||||
}
|
||||
@ -472,6 +481,7 @@ impl<'a> GlyphStore {
|
||||
(false, true) => GlyphEntry::simple(data.id, data.advance),
|
||||
(false, false) => {
|
||||
let glyph = &[DetailedGlyph::new(data.id, data.advance, data.offset)];
|
||||
self.has_detailed_glyphs = true;
|
||||
self.detail_store.add_detailed_glyphs_for_entry(i, glyph);
|
||||
GlyphEntry::complex(data.cluster_start, data.ligature_start, 1)
|
||||
}
|
||||
@ -500,6 +510,7 @@ impl<'a> GlyphStore {
|
||||
data_for_glyphs[i].offset)
|
||||
}).collect();
|
||||
|
||||
self.has_detailed_glyphs = true;
|
||||
self.detail_store.add_detailed_glyphs_for_entry(i, &glyphs_vec);
|
||||
GlyphEntry::complex(first_glyph_data.cluster_start,
|
||||
first_glyph_data.ligature_start,
|
||||
@ -541,10 +552,63 @@ impl<'a> GlyphStore {
|
||||
|
||||
#[inline]
|
||||
pub fn advance_for_char_range(&self, rang: &Range<CharIndex>) -> Au {
|
||||
if !self.has_detailed_glyphs {
|
||||
self.advance_for_char_range_simple_glyphs(rang)
|
||||
} else {
|
||||
self.advance_for_char_range_slow_path(rang)
|
||||
}
|
||||
}
|
||||
|
||||
#[inline]
|
||||
pub fn advance_for_char_range_slow_path(&self, rang: &Range<CharIndex>) -> Au {
|
||||
self.iter_glyphs_for_char_range(rang)
|
||||
.fold(Au(0), |advance, (_, glyph)| advance + glyph.advance())
|
||||
}
|
||||
|
||||
#[inline]
|
||||
#[cfg(any(target_arch = "x86_64", target_arch = "aarch64"))]
|
||||
fn advance_for_char_range_simple_glyphs(&self, rang: &Range<CharIndex>) -> Au {
|
||||
let mask = u32x4::splat(GLYPH_ADVANCE_MASK);
|
||||
let mut simd_advance = u32x4::splat(0);
|
||||
let begin = rang.begin().to_usize();
|
||||
let len = rang.length().to_usize();
|
||||
let num_simd_iterations = len / 4;
|
||||
let leftover_entries = rang.end().to_usize() - (len - num_simd_iterations * 4);
|
||||
let buf = self.transmute_entry_buffer_to_u32_buffer();
|
||||
|
||||
for i in 0..num_simd_iterations {
|
||||
let mut v = u32x4::load(buf, begin + i * 4);
|
||||
v = v & mask;
|
||||
v = v >> GLYPH_ADVANCE_SHIFT;
|
||||
simd_advance = simd_advance + v;
|
||||
}
|
||||
|
||||
let advance =
|
||||
(simd_advance.extract(0) +
|
||||
simd_advance.extract(1) +
|
||||
simd_advance.extract(2) +
|
||||
simd_advance.extract(3)) as i32;
|
||||
let mut leftover = Au(0);
|
||||
for i in leftover_entries..rang.end().to_usize() {
|
||||
leftover = leftover + self.entry_buffer[i].advance();
|
||||
}
|
||||
Au(advance) + leftover
|
||||
}
|
||||
|
||||
/// When SIMD isn't available (non-x86_x64/aarch64), fallback to the slow path.
|
||||
#[inline]
|
||||
#[cfg(not(any(target_arch = "x86_64", target_arch = "aarch64")))]
|
||||
fn advance_for_char_range_simple_glyphs(&self, rang: &Range<CharIndex>) -> Au {
|
||||
self.advance_for_char_range_slow_path(rang)
|
||||
}
|
||||
|
||||
/// Used for SIMD.
|
||||
#[inline]
|
||||
#[cfg(any(target_arch = "x86_64", target_arch = "aarch64"))]
|
||||
fn transmute_entry_buffer_to_u32_buffer(&self) -> &[u32] {
|
||||
unsafe { mem::transmute(self.entry_buffer.as_slice()) }
|
||||
}
|
||||
|
||||
pub fn char_is_space(&self, i: CharIndex) -> bool {
|
||||
assert!(i < self.char_len());
|
||||
self.entry_buffer[i.to_usize()].char_is_space()
|
||||
|
6
servo/components/servo/Cargo.lock
generated
6
servo/components/servo/Cargo.lock
generated
@ -605,6 +605,7 @@ dependencies = [
|
||||
"script_traits 0.0.1",
|
||||
"serde 0.5.3 (registry+https://github.com/rust-lang/crates.io-index)",
|
||||
"serde_macros 0.5.1 (registry+https://github.com/rust-lang/crates.io-index)",
|
||||
"simd 0.1.0 (git+https://github.com/huonw/simd)",
|
||||
"skia 0.0.20130412 (git+https://github.com/servo/skia)",
|
||||
"smallvec 0.1.5 (registry+https://github.com/rust-lang/crates.io-index)",
|
||||
"string_cache 0.1.11 (registry+https://github.com/rust-lang/crates.io-index)",
|
||||
@ -1534,6 +1535,11 @@ dependencies = [
|
||||
"winapi-build 0.1.1 (registry+https://github.com/rust-lang/crates.io-index)",
|
||||
]
|
||||
|
||||
[[package]]
|
||||
name = "simd"
|
||||
version = "0.1.0"
|
||||
source = "git+https://github.com/huonw/simd#d9ad79d86eab50a8f36d45fe17aa9e3a533389ee"
|
||||
|
||||
[[package]]
|
||||
name = "skia"
|
||||
version = "0.0.20130412"
|
||||
|
Loading…
Reference in New Issue
Block a user