Descriptioncompress/flate: add an assembly version of forwardCopy.
Benchmarks on my amd64 desktop below.
compress/flate:
benchmark old ns/op new ns/op delta
BenchmarkDecoderBestSpeed1K 435140 445898 +2.47%
BenchmarkDecoderBestSpeed10K 1062112 1017976 -4.16%
BenchmarkDecoderBestSpeed100K 5833680 5293258 -9.26%
BenchmarkDecoderDefaultCompression1K 421301 449105 +6.60%
BenchmarkDecoderDefaultCompression10K 613890 581874 -5.22%
BenchmarkDecoderDefaultCompression100K 2466726 1839578 -25.42%
BenchmarkDecoderBestCompression1K 421634 449330 +6.57%
BenchmarkDecoderBestCompression10K 614671 581692 -5.37%
BenchmarkDecoderBestCompression100K 2464996 1871082 -24.09%
image/png:
benchmark old ns/op new ns/op delta
BenchmarkDecodeGray 2389624 2378432 -0.47%
BenchmarkDecodeNRGBAGradient 9534565 9584200 +0.52%
BenchmarkDecodeNRGBAOpaque 8163430 8177840 +0.18%
BenchmarkDecodePaletted 1325017 1319442 -0.42%
BenchmarkDecodeRGB 6794668 6761754 -0.48%
Wall time for Denis Cheremisov's PNG-decoding program given in
https://groups.google.com/group/golang-nuts/browse_thread/thread/22aa8a05040fdd49
Before: 2.32s
After: 2.35s
Delta: +1%
Patch Set 1 #Patch Set 2 : diff -r 3a8932ef3669 https://go.googlecode.com/hg/ #
MessagesTotal messages: 1
|