Use the workbuf instead of lzw.decoder.suffixes

This commit will soon be followed by a rollback, but it is committed
anyway so that we can refer to these numbers in the git log.

name                                     old speed      new speed      delta

wuffs_lzw_decode_20k/clang5               255MB/s ± 0%   241MB/s ± 0%   -5.63%  (p=0.008 n=5+5)
wuffs_lzw_decode_100k/clang5              418MB/s ± 1%   401MB/s ± 1%   -4.02%  (p=0.008 n=5+5)

wuffs_lzw_decode_20k/gcc7                 255MB/s ± 1%   235MB/s ± 0%   -7.88%  (p=0.008 n=5+5)
wuffs_lzw_decode_100k/gcc7                444MB/s ± 1%   382MB/s ± 0%  -13.90%  (p=0.008 n=5+5)

wuffs_gif_decode_1k_bw/clang5             239MB/s ± 2%   235MB/s ± 1%   -1.89%  (p=0.016 n=5+5)
wuffs_gif_decode_1k_color/clang5          135MB/s ± 0%   132MB/s ± 0%   -1.86%  (p=0.008 n=5+5)
wuffs_gif_decode_10k_bgra/clang5          692MB/s ± 1%   678MB/s ± 0%   -2.08%  (p=0.008 n=5+5)
wuffs_gif_decode_10k_indexed/clang5       186MB/s ± 0%   181MB/s ± 0%   -2.58%  (p=0.008 n=5+5)
wuffs_gif_decode_20k/clang5               237MB/s ± 1%   226MB/s ± 0%   -4.46%  (p=0.008 n=5+5)
wuffs_gif_decode_100k_artificial/clang5   536MB/s ± 0%   506MB/s ± 0%   -5.53%  (p=0.008 n=5+5)
wuffs_gif_decode_100k_realistic/clang5    219MB/s ± 0%   203MB/s ± 1%   -7.11%  (p=0.008 n=5+5)
wuffs_gif_decode_1000k/clang5             222MB/s ± 0%   206MB/s ± 0%   -7.25%  (p=0.008 n=5+5)
wuffs_gif_decode_anim_screencap/clang5   1.07GB/s ± 0%  1.04GB/s ± 1%   -3.11%  (p=0.008 n=5+5)

wuffs_gif_decode_1k_bw/gcc7               262MB/s ± 0%   255MB/s ± 1%   -2.63%  (p=0.008 n=5+5)
wuffs_gif_decode_1k_color/gcc7            142MB/s ± 1%   137MB/s ± 1%   -3.38%  (p=0.008 n=5+5)
wuffs_gif_decode_10k_bgra/gcc7            596MB/s ± 1%   593MB/s ± 0%     ~     (p=0.151 n=5+5)
wuffs_gif_decode_10k_indexed/gcc7         185MB/s ± 0%   182MB/s ± 2%   -1.63%  (p=0.008 n=5+5)
wuffs_gif_decode_20k/gcc7                 230MB/s ± 0%   226MB/s ± 0%   -1.78%  (p=0.008 n=5+5)
wuffs_gif_decode_100k_artificial/gcc7     490MB/s ± 1%   488MB/s ± 0%     ~     (p=0.095 n=5+5)
wuffs_gif_decode_100k_realistic/gcc7      203MB/s ± 1%   203MB/s ± 0%     ~     (p=0.421 n=5+5)
wuffs_gif_decode_1000k/gcc7               207MB/s ± 0%   207MB/s ± 0%     ~     (p=1.000 n=5+5)
wuffs_gif_decode_anim_screencap/gcc7     1.04GB/s ± 0%  1.03GB/s ± 0%   -0.83%  (p=0.008 n=5+5)
This commit is contained in:
Nigel Tao committed 2019-02-03 14:57:34 +11:00
1 parent d1f1340d47
commit f877fb2c59
7 files changed
+109 -59

No files matched your search

+7 -7
View File
@@ -25,7 +25,7 @@ pub status "?too much pixel data"
pri status "?internal error: inconsistent ri/wi"
// TODO: reference lzw.decoder_workbuf_len_max_incl.
pub const decoder_workbuf_len_max_incl base.u64 = 0xFFFF
pub const decoder_workbuf_len_max_incl base.u64 = 32768 + 0xFFFF
// See the spec appendix E "Interlaced Images" on page 29. The first element
// represents either that the frame was non-interlaced, or that all interlace
@@ -184,8 +184,8 @@ pub func decoder.frame_dirty_rect() base.rect_ie_u32 {
pub func decoder.workbuf_len() base.range_ii_u64 {
return this.util.make_range_ii_u64(
min_incl:this.width as base.u64,
max_incl:this.width as base.u64)
min_incl:32768 + (this.width as base.u64),
max_incl:32768 + (this.width as base.u64))
}
pub func decoder.restart_frame!(index base.u64, io_position base.u64) base.status {
@@ -707,9 +707,9 @@ pri func decoder.decode_id_part2?(dst ptr base.pixel_buffer, src base.io_reader,
args.src.skip_fast!(actual:1, worst_case:1)
}
if (this.width as base.u64) < args.workbuf.length() {
args.workbuf = args.workbuf[:this.width as base.u64]
} else if (this.width as base.u64) > args.workbuf.length() {
if (32768 + (this.width as base.u64)) < args.workbuf.length() {
args.workbuf = args.workbuf[:32768 + (this.width as base.u64)]
} else if (32768 + (this.width as base.u64)) > args.workbuf.length() {
return base."?bad workbuf length"
}
@@ -719,7 +719,7 @@ pri func decoder.decode_id_part2?(dst ptr base.pixel_buffer, src base.io_reader,
}
io_bind (io:r, data:this.compressed[this.compressed_ri:this.compressed_wi]) {
lzw_status =? this.lzw.decode_io_writer?(
dst:this.util.null_io_writer(), src:r, workbuf:this.util.null_slice_u8())
dst:this.util.null_io_writer(), src:r, workbuf:args.workbuf)
this.compressed_ri ~sat+= r.since_mark().length()
}
+36 -19
View File
@@ -16,9 +16,8 @@ pub status "?bad code"
pri status "?internal error: inconsistent I/O"
// TODO: move bulk data buffers like decoder.suffixes or decoder.output into
// the workbuf?
pub const decoder_workbuf_len_max_incl base.u64 = 0
// The suffixes table is 8 bytes per code, so 4096 * 8 = 32768 bytes.
pub const decoder_workbuf_len_max_incl base.u64 = 32768
pub struct decoder?(
// The value passed to set_literal_width. This lw field is copied to the
@@ -85,16 +84,25 @@ pub func decoder.decode_io_writer?(dst base.io_writer, src base.io_reader, workb
this.n_bits = 0
this.output_ri = 0
this.output_wi = 0
if args.workbuf.length() < 32768 {
return base."?bad workbuf length"
}
i = 0
while i < this.clear_code {
while i < this.clear_code,
inv args.workbuf.length() >= 32768,
{
assert i < 256 via "a < b: a < c; c <= b"(c:this.clear_code)
this.lm1s[i] = 0
this.suffixes[(8 * i) + 0] = i as base.u8
args.workbuf[((8 * i) + 0) as base.u64] = i as base.u8
i += 1
}
while true {
this.read_from!(src:args.src)
if args.workbuf.length() < 32768 {
return base."?bad workbuf length"
}
this.read_from!(src:args.src, workbuf:args.workbuf)
if this.output_wi > 0 {
this.write_to?(dst:args.dst)
@@ -114,7 +122,7 @@ pub func decoder.decode_io_writer?(dst base.io_writer, src base.io_reader, workb
}
}
pri func decoder.read_from!(src base.io_reader) {
pri func decoder.read_from!(src base.io_reader, workbuf slice base.u8) {
var clear_code base.u32[..256]
var end_code base.u32[..257]
@@ -133,6 +141,10 @@ pri func decoder.read_from!(src base.io_reader) {
var lm1_b base.u16[..4095]
var lm1_a base.u16[..4095]
if args.workbuf.length() < 32768 {
return nothing
}
clear_code = this.clear_code
end_code = this.end_code
@@ -143,7 +155,9 @@ pri func decoder.read_from!(src base.io_reader) {
n_bits = this.n_bits
output_wi = this.output_wi
while true {
while true,
inv args.workbuf.length() >= 32768,
{
if n_bits < width {
assert n_bits < 12 via "a < b: a < c; c <= b"(c:width)
if args.src.available() >= 4 {
@@ -199,12 +213,12 @@ pri func decoder.read_from!(src base.io_reader) {
assert (8 * save_code) <= ((8 * save_code) + 8) via "a <= (a + b): 0 <= b"(b:8)
assert (8 * prev_code) <= ((8 * prev_code) + 8) via "a <= (a + b): 0 <= b"(b:8)
this.prefixes[save_code] = this.prefixes[prev_code]
this.suffixes[(8 * save_code):(8 * save_code) + 8].copy_from_slice!(
s:this.suffixes[(8 * prev_code):(8 * prev_code) + 8])
this.suffixes[(8 * save_code) + ((lm1_a % 8) as base.u32)] = code as base.u8
args.workbuf[(8 * save_code):(8 * save_code) + 8].copy_from_slice!(
s:args.workbuf[(8 * prev_code):(8 * prev_code) + 8])
args.workbuf[(8 * save_code) + ((lm1_a % 8) as base.u32)] = code as base.u8
} else {
this.prefixes[save_code] = prev_code as base.u16
this.suffixes[(8 * save_code) + 0] = code as base.u8
args.workbuf[(8 * save_code) + 0] = code as base.u8
}
save_code += 1
@@ -242,14 +256,17 @@ pri func decoder.read_from!(src base.io_reader) {
output_wi = (output_wi + 1 + (this.lm1s[c] as base.u32)) & 8191
steps = (this.lm1s[c] as base.u32) >> 3
while true {
while true,
inv args.workbuf.length() >= 32768,
{
assert o <= (o + 8) via "a <= (a + b): 0 <= b"(b:8)
assert (8 * c) <= ((8 * c) + 8) via "a <= (a + b): 0 <= b"(b:8)
assert ((8 * c) + 8) <= 32768
// The final "8" is redundant semantically, but helps the
// wuffs-c code generator recognize that both slices have the
// same constant length, and hence produce efficient C code.
this.output[o:o + 8].copy_from_slice!(s:this.suffixes[(8 * c):(8 * c) + 8])
this.output[o:o + 8].copy_from_slice!(s:args.workbuf[(8 * c):(8 * c) + 8])
if steps <= 0 {
break
@@ -261,7 +278,7 @@ pri func decoder.read_from!(src base.io_reader) {
o = (o ~mod- 8) & 8191
c = this.prefixes[c] as base.u32
}
first_byte = this.suffixes[(8 * c) + 0]
first_byte = args.workbuf[(8 * c) + 0]
if code == save_code {
this.output[output_wi] = first_byte
@@ -276,12 +293,12 @@ pri func decoder.read_from!(src base.io_reader) {
assert (8 * save_code) <= ((8 * save_code) + 8) via "a <= (a + b): 0 <= b"(b:8)
assert (8 * prev_code) <= ((8 * prev_code) + 8) via "a <= (a + b): 0 <= b"(b:8)
this.prefixes[save_code] = this.prefixes[prev_code]
this.suffixes[(8 * save_code):(8 * save_code) + 8].copy_from_slice!(
s:this.suffixes[(8 * prev_code):(8 * prev_code) + 8])
this.suffixes[(8 * save_code) + ((lm1_b % 8) as base.u32)] = first_byte
args.workbuf[(8 * save_code):(8 * save_code) + 8].copy_from_slice!(
s:args.workbuf[(8 * prev_code):(8 * prev_code) + 8])
args.workbuf[(8 * save_code) + ((lm1_b % 8) as base.u32)] = first_byte
} else {
this.prefixes[save_code] = prev_code as base.u16
this.suffixes[(8 * save_code) + 0] = first_byte as base.u8
args.workbuf[(8 * save_code) + 0] = first_byte as base.u8
}
save_code += 1