aw/golang-image
1
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
280 Commits 1 Branch 2 Tags 8 MiB
a594737139
Commit Graph

7 Commits

Author SHA1 Message Date
Nigel Tao
46e4eb730a vector: use asm opcode mnemonics 
There's no change in the binary output, just less mystery in the asm.

These mnemonics were introduced in Go 1.10:
https://golang.org/doc/go1.10#asm and https://golang.org/cl/75490

Current stable release (as of 2018-11-10) is Go 1.11, and
https://golang.org/doc/devel/release.html#policy says that Go 1.9 and
below are therefore no longer supported.

Change-Id: I1f9a63521bc8d5e8f8d395605f62bf7fb6a63bc5
Reviewed-on: https://go-review.googlesource.com/c/148997
Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org>
 2018-11-15 02:24:11 +00:00
Nigel Tao
c78039e8ce vector: tweak the asm register assignment. 
This makes the fixed point and floating point code more similar.

Benchmarks don't show any significant change.

Change-Id: I723fa1605eaa248b40e784201b680c16cc3d26a2
Reviewed-on: https://go-review.googlesource.com/32134
Reviewed-by: David Crawshaw <crawshaw@golang.org>
 2016-10-26 22:33:17 +00:00
Nigel Tao
93fad3647f vector: remove a shift from the inner loop. 
PSHUFB already lets us pick certain bytes out of an XMM register.
There's no need to shift by 8 bits (1 byte) beforehand.

The code generator is simpler. There's also a small win on the
benchmarks, especially for FixedAccumulateOpOverSIMD.

name                              old time/op  new time/op  delta
FixedAccumulateOpOverSIMD16-8      183ns ± 0%   174ns ± 0%  -4.92%    (p=0.000 n=8+8)
FixedAccumulateOpSrcSIMD16-8      87.0ns ± 1%  86.3ns ± 0%  -0.77%   (p=0.000 n=10+9)
FixedAccumulateMaskSIMD16-8       80.4ns ± 1%  81.2ns ± 1%  +1.01%  (p=0.000 n=10+10)
FloatingAccumulateOpOverSIMD16-8   250ns ± 1%   244ns ± 0%  -2.39%   (p=0.000 n=10+8)
FloatingAccumulateOpSrcSIMD16-8    176ns ± 1%   176ns ± 0%    ~      (p=0.142 n=10+8)
FloatingAccumulateMaskSIMD16-8     167ns ± 0%   167ns ± 0%    ~      (p=0.137 n=8+10)
FixedAccumulateOpOverSIMD64-8     2.73µs ± 1%  2.58µs ± 0%  -5.36%   (p=0.000 n=10+7)
FixedAccumulateOpSrcSIMD64-8      1.18µs ± 1%  1.17µs ± 0%  -0.33%   (p=0.003 n=10+9)
FixedAccumulateMaskSIMD64-8       1.09µs ± 0%  1.09µs ± 0%  -0.17%    (p=0.047 n=9+9)
FloatingAccumulateOpOverSIMD64-8  3.67µs ± 0%  3.61µs ± 1%  -1.47%   (p=0.000 n=7+10)
FloatingAccumulateOpSrcSIMD64-8   2.60µs ± 0%  2.61µs ± 0%  +0.19%    (p=0.003 n=8+8)
FloatingAccumulateMaskSIMD64-8    2.47µs ± 0%  2.46µs ± 0%    ~      (p=0.162 n=10+9)
GlyphAlpha16Over-8                2.99µs ± 0%  2.98µs ± 1%  -0.50%   (p=0.021 n=9+10)
GlyphAlpha16Src-8                 2.89µs ± 1%  2.89µs ± 0%    ~     (p=0.381 n=10+10)
GlyphAlpha32Over-8                4.53µs ± 0%  4.50µs ± 0%  -0.83%  (p=0.000 n=10+10)
GlyphAlpha32Src-8                 4.14µs ± 0%  4.13µs ± 0%  -0.21%   (p=0.026 n=9+10)
GlyphAlpha64Over-8                8.97µs ± 1%  8.80µs ± 0%  -1.85%   (p=0.000 n=10+9)
GlyphAlpha64Src-8                 7.42µs ± 1%  7.39µs ± 0%  -0.45%  (p=0.011 n=10+10)
GlyphAlpha128Over-8               21.8µs ± 0%  21.2µs ± 0%  -2.91%    (p=0.000 n=9+9)
GlyphAlpha128Src-8                15.6µs ± 0%  15.6µs ± 0%    ~      (p=0.982 n=10+7)
GlyphAlpha256Over-8               66.3µs ± 1%  63.7µs ± 0%  -3.84%   (p=0.000 n=10+9)
GlyphAlpha256Src-8                41.2µs ± 1%  41.2µs ± 1%    ~     (p=1.000 n=10+10)
GlyphRGBA16Over-8                 4.75µs ± 0%  4.75µs ± 1%    ~      (p=0.735 n=9+10)
GlyphRGBA16Src-8                  4.20µs ± 0%  4.20µs ± 0%    ~       (p=0.503 n=8+8)
GlyphRGBA32Over-8                 11.4µs ± 0%  11.4µs ± 0%    ~       (p=0.119 n=9+9)
GlyphRGBA32Src-8                  9.34µs ± 1%  9.32µs ± 0%    ~       (p=0.062 n=9+8)
GlyphRGBA64Over-8                 36.0µs ± 0%  36.1µs ± 0%    ~       (p=0.209 n=8+9)
GlyphRGBA64Src-8                  27.9µs ± 1%  27.8µs ± 0%    ~     (p=0.796 n=10+10)
GlyphRGBA128Over-8                 131µs ± 0%   131µs ± 0%    ~       (p=0.931 n=9+9)
GlyphRGBA128Src-8                 97.9µs ± 0%  97.7µs ± 1%    ~      (p=0.053 n=9+10)
GlyphRGBA256Over-8                 503µs ± 0%   503µs ± 1%    ~      (p=0.274 n=8+10)
GlyphRGBA256Src-8                  370µs ± 0%   369µs ± 0%    ~      (p=0.497 n=9+10)

Change-Id: I56651e70b258792b83ea2a74904756243c88bef4
Reviewed-on: https://go-review.googlesource.com/31537
Reviewed-by: David Crawshaw <crawshaw@golang.org>
 2016-10-22 21:39:22 +00:00
Nigel Tao
8874bef159 vector: change ϕ from 10 to 9. 
This slight loss in quality allows us to use int32 math exclusively
throughout raster_fixed.go, instead of occasionally dropping into int64
math. The change in ϕ doesn't affect the benchmarks noticably, but
staying in int32 does. The net effect:

name                              old time/op  new time/op  delta
GlyphAlpha16Over-8                3.36µs ± 0%  2.99µs ± 0%  -10.89%         (p=0.000 n=10+9)
GlyphAlpha16Src-8                 3.26µs ± 0%  2.89µs ± 1%  -11.34%         (p=0.000 n=9+10)
GlyphAlpha32Over-8                5.20µs ± 0%  4.53µs ± 0%  -12.76%         (p=0.000 n=8+10)
GlyphAlpha32Src-8                 4.81µs ± 1%  4.14µs ± 0%  -13.91%          (p=0.000 n=9+9)
GlyphAlpha64Over-8                10.2µs ± 0%   9.0µs ± 1%  -11.99%         (p=0.000 n=9+10)
GlyphAlpha64Src-8                 8.62µs ± 0%  7.42µs ± 1%  -13.89%         (p=0.000 n=9+10)
GlyphAlpha128Over-8               24.1µs ± 0%  21.8µs ± 0%   -9.32%          (p=0.000 n=9+9)
GlyphAlpha128Src-8                17.9µs ± 0%  15.6µs ± 0%  -12.68%         (p=0.000 n=9+10)
GlyphAlpha256Over-8               70.1µs ± 0%  66.3µs ± 1%   -5.44%        (p=0.000 n=10+10)
GlyphAlpha256Src-8                45.2µs ± 1%  41.2µs ± 1%   -8.92%        (p=0.000 n=10+10)
GlyphRGBA16Over-8                 5.12µs ± 0%  4.75µs ± 0%   -7.15%         (p=0.000 n=10+9)
GlyphRGBA16Src-8                  4.57µs ± 1%  4.20µs ± 0%   -8.18%          (p=0.000 n=9+8)
GlyphRGBA32Over-8                 12.1µs ± 0%  11.4µs ± 0%   -5.50%         (p=0.000 n=10+9)
GlyphRGBA32Src-8                  10.0µs ± 0%   9.3µs ± 1%   -6.80%         (p=0.000 n=10+9)
GlyphRGBA64Over-8                 37.2µs ± 0%  36.0µs ± 0%   -3.17%          (p=0.000 n=9+8)
GlyphRGBA64Src-8                  29.0µs ± 1%  27.9µs ± 1%   -4.05%         (p=0.000 n=9+10)
GlyphRGBA128Over-8                 134µs ± 1%   131µs ± 0%   -1.85%          (p=0.000 n=9+9)
GlyphRGBA128Src-8                  100µs ± 1%    98µs ± 0%   -2.27%         (p=0.000 n=10+9)
GlyphRGBA256Over-8                 506µs ± 0%   503µs ± 0%   -0.56%         (p=0.000 n=10+8)
GlyphRGBA256Src-8                  373µs ± 0%   370µs ± 0%   -1.01%         (p=0.000 n=10+9)

Change-Id: Ie02afac6fd6fa95f090bf3fe0a5c744799ea9dc5
Reviewed-on: https://go-review.googlesource.com/31532
Reviewed-by: David Crawshaw <crawshaw@golang.org>
 2016-10-20 04:12:57 +00:00
Nigel Tao
e78c45720c vector: add SIMD versions of xxxAccumulateMask. 
name                              old time/op  new time/op  delta
GlyphAlphaLoose16Over-8           3.96µs ± 0%  3.64µs ± 1%   -8.08%   (p=0.000 n=8+10)
GlyphAlphaLoose16Src-8            3.64µs ± 0%  3.35µs ± 0%   -7.88%    (p=0.000 n=8+9)
GlyphAlphaLoose32Over-8           8.45µs ± 0%  6.74µs ± 0%  -20.22%    (p=0.000 n=8+9)
GlyphAlphaLoose32Src-8            7.24µs ± 0%  5.54µs ± 1%  -23.48%   (p=0.000 n=8+10)
GlyphAlphaLoose64Over-8           22.2µs ± 0%  17.4µs ± 0%  -21.67%    (p=0.000 n=9+9)
GlyphAlphaLoose64Src-8            17.6µs ± 1%  12.2µs ± 1%  -30.32%  (p=0.000 n=10+10)
GlyphAlphaLoose128Over-8          67.9µs ± 0%  53.3µs ± 1%  -21.53%  (p=0.000 n=10+10)
GlyphAlphaLoose128Src-8           48.2µs ± 0%  32.6µs ± 2%  -32.41%   (p=0.000 n=9+10)
GlyphAlphaLoose256Over-8           242µs ± 1%   187µs ± 1%  -22.96%    (p=0.000 n=9+9)
GlyphAlphaLoose256Src-8            163µs ± 0%   105µs ± 1%  -35.83%    (p=0.000 n=9+9)
GlyphRGBA16Over-8                 5.25µs ± 1%  4.95µs ± 0%   -5.78%    (p=0.000 n=9+9)
GlyphRGBA16Src-8                  4.72µs ± 0%  4.43µs ± 1%   -6.22%   (p=0.000 n=9+10)
GlyphRGBA32Over-8                 13.5µs ± 0%  11.9µs ± 1%  -12.19%   (p=0.000 n=9+10)
GlyphRGBA32Src-8                  11.5µs ± 1%   9.8µs ± 0%  -14.72%    (p=0.000 n=9+9)
GlyphRGBA64Over-8                 42.0µs ± 2%  36.9µs ± 1%  -12.19%  (p=0.000 n=10+10)
GlyphRGBA64Src-8                  34.1µs ± 1%  28.5µs ± 0%  -16.25%    (p=0.000 n=9+7)
GlyphRGBA128Over-8                 149µs ± 2%   133µs ± 1%  -10.24%   (p=0.000 n=10+9)
GlyphRGBA128Src-8                  115µs ± 1%    99µs ± 1%  -13.57%   (p=0.000 n=9+10)
GlyphRGBA256Over-8                 566µs ± 0%   511µs ± 1%   -9.85%   (p=0.000 n=9+10)
GlyphRGBA256Src-8                  435µs ± 0%   372µs ± 0%  -14.64%    (p=0.000 n=9+8)
GlyphNRGBA16Over-8                26.9µs ± 3%  26.0µs ± 3%   -3.55%   (p=0.000 n=10+9)
GlyphNRGBA16Src-8                 18.8µs ± 2%  18.4µs ± 2%   -2.21%   (p=0.000 n=9+10)
GlyphNRGBA32Over-8                99.1µs ± 2%  95.9µs ± 3%   -3.23%  (p=0.000 n=10+10)
GlyphNRGBA32Src-8                 65.6µs ± 3%  62.8µs ± 2%   -4.36%  (p=0.000 n=10+10)
GlyphNRGBA64Over-8                 376µs ± 4%   370µs ± 2%     ~     (p=0.063 n=10+10)
GlyphNRGBA64Src-8                  238µs ± 3%   233µs ± 1%   -2.21%   (p=0.000 n=9+10)
GlyphNRGBA128Over-8               1.52ms ± 2%  1.48ms ± 0%   -2.11%   (p=0.000 n=10+8)
GlyphNRGBA128Src-8                 951µs ± 3%   935µs ± 1%   -1.69%   (p=0.013 n=10+9)
GlyphNRGBA256Over-8               6.00ms ± 1%  5.87ms ± 3%   -2.12%  (p=0.002 n=10+10)
GlyphNRGBA256Src-8                3.94ms ± 2%  3.80ms ± 2%   -3.64%  (p=0.000 n=10+10)

A comparison of the non-SIMD and SIMD versions:

name                              time/op
FixedAccumulateMask16-8            237ns ± 0%
FixedAccumulateMaskSIMD16-8       80.0ns ± 1%
FloatingAccumulateMask16-8         413ns ± 2%
FloatingAccumulateMaskSIMD16-8     166ns ± 0%
FixedAccumulateMask64-8           3.42µs ± 0%
FixedAccumulateMaskSIMD64-8       1.09µs ± 0%
FloatingAccumulateMask64-8        6.92µs ± 0%
FloatingAccumulateMaskSIMD64-8    2.47µs ± 1%

Change-Id: Ib6980e5975ed2842ff2a372f76dd5f2e95c5526c
Reviewed-on: https://go-review.googlesource.com/30898
Reviewed-by: David Crawshaw <crawshaw@golang.org>
 2016-10-14 04:48:47 +00:00
Nigel Tao
8edbaf3f9e vector: add SIMD versions of xxxAccumulateOpOver. 
name                             old time/op  new time/op  delta
GlyphAlpha16Over-8               3.55µs ± 0%  3.17µs ± 0%  -10.58%  (p=0.000 n=10+10)
GlyphAlpha32Over-8               6.73µs ± 1%  4.94µs ± 0%  -26.55%   (p=0.000 n=10+9)
GlyphAlpha64Over-8               16.4µs ± 0%   9.6µs ± 0%  -41.30%   (p=0.000 n=9+10)
GlyphAlpha128Over-8              47.3µs ± 0%  23.1µs ± 1%  -51.09%    (p=0.000 n=9+9)
GlyphAlpha256Over-8               159µs ± 0%    69µs ± 0%  -56.82%    (p=0.000 n=9+8)

A comparison of the non-SIMD and SIMD versions:

name                              time/op
FixedAccumulateOpOver16-8          579ns ± 0%
FixedAccumulateOpOverSIMD16-8      183ns ± 0%
FloatingAccumulateOpOver16-8       670ns ± 1%
FloatingAccumulateOpOverSIMD16-8   242ns ± 0%
FixedAccumulateOpOver64-8         9.61µs ± 0%
FixedAccumulateOpOverSIMD64-8     2.72µs ± 0%
FloatingAccumulateOpOver64-8      11.1µs ± 0%
FloatingAccumulateOpOverSIMD64-8  3.65µs ± 0%

Change-Id: I08273c40ac5445f39b77a88fb8b6b07fd3e5f84b
Reviewed-on: https://go-review.googlesource.com/30831
Reviewed-by: David Crawshaw <crawshaw@golang.org>
 2016-10-13 03:04:36 +00:00
Nigel Tao
ae7296db94 vector: generate acc_amd64.s from a template. 
There are currently 2 instantiations: fixed and floating math.

There will eventually be 6: 2 math types times 3 operators (OpOver,
OpSrc and Mask).

Change-Id: I16db8f3693abe30027ea2f822909980403a6142f
Reviewed-on: https://go-review.googlesource.com/30735
Reviewed-by: David Crawshaw <crawshaw@golang.org>
 2016-10-11 07:55:43 +00:00