Klaus Post
7daa20bf74
Generate AVX2 code ( #141 )
...
Replaces AVX2 up to 10x8 configurations with specific generated functions.
If code size is a concern `-tags=nogen` can be used.
Biggest speedup when not memory constrained.
```
benchmark old MB/s new MB/s speedup
BenchmarkEncode_8x5x8M 5895.75 9648.18 1.64x
BenchmarkEncode_8x5x8M-4 16773.41 17220.67 1.03x
BenchmarkEncode_8x5x8M-16 18263.12 17176.28 0.94x
BenchmarkEncode_8x6x8M 5075.89 8548.39 1.68x
BenchmarkEncode_8x6x8M-4 14559.83 15370.95 1.06x
BenchmarkEncode_8x6x8M-16 16183.37 15291.98 0.94x
BenchmarkEncode_8x7x8M 4481.18 7015.60 1.57x
BenchmarkEncode_8x7x8M-4 12835.35 13695.90 1.07x
BenchmarkEncode_8x7x8M-16 14246.94 13737.36 0.96x
BenchmarkEncode_8x8x05M 5569.95 7947.70 1.43x
BenchmarkEncode_8x8x05M-4 17334.91 25271.37 1.46x
BenchmarkEncode_8x8x05M-16 29349.42 35043.36 1.19x
BenchmarkEncode_8x8x1M 4830.58 7891.32 1.63x
BenchmarkEncode_8x8x1M-4 17531.36 27371.42 1.56x
BenchmarkEncode_8x8x1M-16 29593.98 39241.09 1.33x
BenchmarkEncode_8x8x8M 3953.66 6584.26 1.67x
BenchmarkEncode_8x8x8M-4 11527.34 12331.23 1.07x
BenchmarkEncode_8x8x8M-16 12718.89 12173.08 0.96x
BenchmarkEncode_8x8x32M 3927.51 6195.91 1.58x
BenchmarkEncode_8x8x32M-4 11490.85 11424.39 0.99x
BenchmarkEncode_8x8x32M-16 12506.09 11888.55 0.95x
benchmark old MB/s new MB/s speedup
BenchmarkParallel_8x8x64K 5490.24 6959.57 1.27x
BenchmarkParallel_8x8x64K-4 21078.94 29557.51 1.40x
BenchmarkParallel_8x8x64K-16 57508.45 73672.54 1.28x
BenchmarkParallel_8x8x1M 4755.49 7667.84 1.61x
BenchmarkParallel_8x8x1M-4 11818.66 12013.49 1.02x
BenchmarkParallel_8x8x1M-16 12923.12 12109.42 0.94x
BenchmarkParallel_8x8x8M 3973.94 6525.85 1.64x
BenchmarkParallel_8x8x8M-4 11725.68 11312.46 0.96x
BenchmarkParallel_8x8x8M-16 12608.20 11484.98 0.91x
BenchmarkParallel_8x3x1M 14139.71 17993.04 1.27x
BenchmarkParallel_8x3x1M-4 21805.97 23053.92 1.06x
BenchmarkParallel_8x3x1M-16 24673.05 23596.71 0.96x
BenchmarkParallel_8x4x1M 10617.88 14474.54 1.36x
BenchmarkParallel_8x4x1M-4 18635.82 18965.65 1.02x
BenchmarkParallel_8x4x1M-16 21518.12 20171.47 0.94x
BenchmarkParallel_8x5x1M 8669.88 11833.96 1.36x
BenchmarkParallel_8x5x1M-4 16321.00 17500.30 1.07x
BenchmarkParallel_8x5x1M-16 17267.16 17191.04 1.00x
```
2020-05-20 12:48:34 +02:00
Frank Wessels
1b9e129671
Avx512 parallel81 ( #131 )
...
* AVX512 routine for 8x1 parallel processing (WIP)
* Testing and integration of Parallel81 assembly routine
2020-05-06 12:32:31 +02:00
Klaus Post
dccac354fe
Add cross compilation ( #127 )
...
* Add cross compilation
Add 386 as 32 bit test, arm64 and ppc64le since they have assembly.
2020-05-04 21:19:49 +02:00
Klaus Post
de70cc155f
AVX512 parallel processing ( #120 )
...
Do concurrent processing in AVX512 mode and split jobs by cache size.
2020-05-04 09:17:40 +02:00
Klaus Post
d2cfcb8065
Add commandline arg to disable asm for tests. ( #116 )
...
* Add commandline test args
2020-04-22 15:38:21 +02:00
Klaus Post
0883d2f011
Only enable AVX512 on AMD64
...
Fixes #102
2019-05-26 12:12:55 +02:00
Klaus Post
a9588190c0
Optimize pure Go version. ( #96 )
...
* Optimize pure Go version.
* Update docs. Add Go 1.12 CI
* Avoid dst bounds check when using noasm ~ 40-50% faster.
* Convert multiply table to a slice whenever used.
* Split on 32 byte boundaries instead of 16 byte.
2019-03-08 10:49:27 +01:00
Frank Wessels
79aee05119
AVX512 accelerated version resulting in a 4x speed improvement over AVX2 ( #91 )
...
The performance on AVX512 has been accelerated for Intel CPUs. This gives speedups on a per-core basis of up to 4x compared to AVX2 as can be seen in the following table:
```
$ benchcmp avx2.txt avx512.txt
benchmark AVX2 MB/s AVX512 MB/s speedup
BenchmarkEncode8x8x1M-72 1681.35 4125.64 2.45x
BenchmarkEncode8x4x8M-72 1529.36 5507.97 3.60x
BenchmarkEncode8x8x8M-72 791.16 2952.29 3.73x
BenchmarkEncode8x8x32M-72 573.26 2168.61 3.78x
BenchmarkEncode12x4x12M-72 1234.41 4912.37 3.98x
BenchmarkEncode16x4x16M-72 1189.59 5138.01 4.32x
BenchmarkEncode24x8x24M-72 690.68 2583.70 3.74x
BenchmarkEncode24x8x48M-72 674.20 2643.31 3.92x
```
2019-02-10 11:17:23 +01:00