Tree:
25b51b2c44
master
n4.4-jackoutdev
oldabi
release/0.10
release/0.11
release/0.5
release/0.6
release/0.7
release/0.8
release/0.9
release/1.0
release/1.1
release/1.2
release/2.0
release/2.1
release/2.2
release/2.3
release/2.4
release/2.5
release/2.6
release/2.7
release/2.8
release/3.0
release/3.1
release/3.2
release/3.3
release/3.4
release/4.0
release/4.1
release/4.2
release/4.3
release/4.4
v0.6.1
v0.6
v0.5.3
v0.5.2
v0.5.1
v0.5
n4.5-dev
n4.4-dev
n4.3-dev
n4.2-dev
n4.1-dev
n3.5-dev
n3.4-dev
n3.3-dev
n3.2-dev
n3.1-dev
n2.9-dev
n2.8-dev
n2.7-dev
n2.6-dev
n2.5-dev
n2.4-dev
n2.3-dev
n2.2-dev
n2.1-dev
n1.3-dev
n1.2-dev
n1.1-dev
n0.12-dev
n0.11-dev
n4.4
n4.3.2
n4.3.1
n4.3
n4.2.4
n4.2.3
n4.2.2
n4.2.1
n4.2
n4.1.6
n4.1.5
n4.1.4
n4.1.3
n4.1.2
n4.1.1
n4.1
n4.0.6
n4.0.5
n4.0.4
n4.0.3
n4.0.2
n4.0.1
n4.0
n3.4.8
n3.4.7
n3.4.6
n3.4.5
n3.4.4
n3.4.3
n3.4.2
n3.4.1
n3.4
n3.3.9
n3.3.8
n3.3.7
n3.3.6
n3.3.5
n3.3.4
n3.3.3
n3.3.2
n3.3.1
n3.3
n3.2.15
n3.2.14
n3.2.13
n3.2.12
n3.2.11
n3.2.10
n3.2.9
n3.2.8
n3.2.7
n3.2.6
n3.2.5
n3.2.4
n3.2.3
n3.2.2
n3.2.1
n3.2
n3.1.11
n3.1.10
n3.1.9
n3.1.8
n3.1.7
n3.1.6
n3.1.5
n3.1.4
n3.1.3
n3.1.2
n3.1.1
n3.1
n3.0.12
n3.0.11
n3.0.10
n3.0.9
n3.0.8
n3.0.7
n3.0.6
n3.0.5
n3.0.4
n3.0.3
n3.0.2
n3.0.1
n3.0
n2.8.17
n2.8.16
n2.8.15
n2.8.14
n2.8.13
n2.8.12
n2.8.11
n2.8.10
n2.8.9
n2.8.8
n2.8.7
n2.8.6
n2.8.5
n2.8.4
n2.8.3
n2.8.2
n2.8.1
n2.8
n2.7.7
n2.7.6
n2.7.5
n2.7.4
n2.7.3
n2.7.2
n2.7.1
n2.7
n2.6.9
n2.6.8
n2.6.7
n2.6.6
n2.6.5
n2.6.4
n2.6.3
n2.6.2
n2.6.1
n2.6
n2.5.11
n2.5.10
n2.5.9
n2.5.8
n2.5.7
n2.5.6
n2.5.5
n2.5.4
n2.5.3
n2.5.2
n2.5.1
n2.5
n2.4.14
n2.4.13
n2.4.12
n2.4.11
n2.4.10
n2.4.9
n2.4.8
n2.4.7
n2.4.6
n2.4.5
n2.4.4
n2.4.3
n2.4.2
n2.4.1
n2.4
n2.3.6
n2.3.5
n2.3.4
n2.3.3
n2.3.2
n2.3.1
n2.3
n2.2.16
n2.2.15
n2.2.14
n2.2.13
n2.2.12
n2.2.11
n2.2.10
n2.2.9
n2.2.8
n2.2.7
n2.2.6
n2.2.5
n2.2.4
n2.2.3
n2.2.2
n2.2.1
n2.2
n2.2-rc2
n2.2-rc1
n2.1.8
n2.1.7
n2.1.6
n2.1.5
n2.1.4
n2.1.3
n2.1.2
n2.1.1
n2.1
n2.0.7
n2.0.6
n2.0.5
n2.0.4
n2.0.3
n2.0.2
n2.0.1
n2.0
n1.2.12
n1.2.11
n1.2.10
n1.2.9
n1.2.8
n1.2.7
n1.2.6
n1.2.5
n1.2.4
n1.2.3
n1.2.2
n1.2.1
n1.2
n1.1.16
n1.1.15
n1.1.14
n1.1.13
n1.1.12
n1.1.11
n1.1.10
n1.1.9
n1.1.8
n1.1.7
n1.1.6
n1.1.5
n1.1.4
n1.1.3
n1.1.2
n1.1.1
n1.1
n1.0.10
n1.0.9
n1.0.8
n1.0.7
n1.0.6
n1.0.5
n1.0.4
n1.0.3
n1.0.2
n1.0.1
n1.0
n0.11.5
n0.11.4
n0.11.3
n0.11.2
n0.11.1
n0.11
n0.10.16
n0.10.15
n0.10.14
n0.10.13
n0.10.12
n0.10.11
n0.10.10
n0.10.9
n0.10.8
n0.10.7
n0.10.6
n0.10.5
n0.10.4
n0.10.3
n0.10.2
n0.10.1
n0.10
n0.9.4
n0.9.3
n0.9.2
n0.9.1
n0.9
n0.8.15
n0.8.14
n0.8.13
n0.8.12
n0.8.11
n0.8.10
n0.8.9
n0.8.8
n0.8.7
n0.8.6
n0.8.5
n0.8.4
n0.8.3
n0.8.2
n0.8.1
n0.8
n0.7.17
n0.7.16
n0.7.15
n0.7.14
n0.7.13
n0.7.12
n0.7.11
n0.7.10
n0.7.9
n0.7.8
n0.7.7
n0.7.6
n0.7.5
n0.7.4
n0.7.3
n0.7.2
n0.7.1
n0.6.7
n0.6.6
n0.6.5
n0.6.4
ffmpeg-0.6.3
n0.5.15
n0.5.14
n0.5.13
n0.5.12
n0.5.11
n0.5.10
n0.5.9
n0.5.8
n0.5.7
n0.5.6
n0.5.5
N
5 Commits (25b51b2c44f16f46d6bd5db4f3421ae57909924e)
5 Commits (25b51b2c44f16f46d6bd5db4f3421ae57909924e)
| Author | SHA1 | Message | Date |
|---|---|---|---|
 Michael Kostylev
|
6797d1948b | x86: rv40: Mark rv40_weight functions as MMX2; they use MMX2 instructions. | 13 years ago |
 Christophe Gisquet
|
110d0cdc9d |
rv40dsp x86: MMX/MMX2/3DNow/SSE2/SSSE3 implementations of MC
Code mostly inspired by vp8's MC, however:
- its MMX2 horizontal filter is worse because it can't take advantage of
the coefficient redundancy
- that same coefficient redundancy allows better code for non-SSSE3 versions
Benchmark (rounded to tens of unit):
V8x8 H8x8 2D8x8 V16x16 H16x16 2D16x16
C 445 358 985 1785 1559 3280
MMX* 219 271 478 714 929 1443
SSE2 131 158 294 425 515 892
SSSE3 120 122 248 387 390 763
End result is overall around a 15% speedup for SSSE3 version (on 6 sequences);
all loop filter functions now take around 55% of decoding time, while luma MC
dsp functions are around 6%, chroma ones are 1.3% and biweight around 2.3%.
Signed-off-by: Diego Biurrun <diego@biurrun.de>
|
13 years ago |
|
Christophe GISQUET
|
2130bd8f5b |
rv40dsp x86: use only one register, for both increment and loop counter
Around 10 cycles faster for luma. Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com> |
13 years ago |
|
Christophe GISQUET
|
272b252c01 |
rv40dsp: implement prescaled versions for biweight.
Quite often, the original weights are multiple of 512. By prescaling them by 1/512 when they are computed (once per frame), no intermediate shifting is needed, and no prescaling on each call either. The x86 code already used that trick. Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com> |
13 years ago |
|
Christophe Gisquet
|
e5c9de2ab7 |
rv40: x86 SIMD for biweight
Provide MMX, SSE2 and SSSE3 versions, with a fast-path when the weights are multiples of 512 (which is often the case when the values round up nicely). *_TIMER report for the 16x16 and 8x8 cases: C: 9015 decicycles in 16, 524257 runs, 31 skips 2656 decicycles in 8, 524271 runs, 17 skips MMX: 4156 decicycles in 16, 262090 runs, 54 skips 1206 decicycles in 8, 262131 runs, 13 skips MMX on fast-path: 2760 decicycles in 16, 524222 runs, 66 skips 995 decicycles in 8, 524252 runs, 36 skips SSE2: 2163 decicycles in 16, 262131 runs, 13 skips 832 decicycles in 8, 262137 runs, 7 skips SSE2 with fast path: 1783 decicycles in 16, 524276 runs, 12 skips 711 decicycles in 8, 524283 runs, 5 skips SSSE3: 2117 decicycles in 16, 262136 runs, 8 skips 814 decicycles in 8, 262143 runs, 1 skips SSSE3 with fast path: 1315 decicycles in 16, 524285 runs, 3 skips 578 decicycles in 8, 524286 runs, 2 skips This means around a 4% speedup for some sequences. Signed-off-by: Diego Biurrun <diego@biurrun.de> |
13 years ago |