JPEG decoder with SIMD optimizations.
Compared to the standard library:
BenchmarkDecodeBaseline420-8 3124 742291 ns/op 426321 B/op 6 allocs/op
BenchmarkDecodeBaseline420StdLib-8 1318 1819781 ns/op 407089 B/op 5 allocs/op
BenchmarkDecodeProgressive420-8 1156 2128195 ns/op 1999883 B/op 10 allocs/op
BenchmarkDecodeProgressive420StdLib-8 662 3602531 ns/op 1980106 B/op 17 allocs/op
BenchmarkDecodeConfig-8 5923922 410.4 ns/op 48 B/op 1 allocs/op
BenchmarkDecodeConfigStdLib-8 1473702 1474 ns/op 13616 B/op 2 allocs/op
BenchmarkDecodeToRGBANearestNeighbor-8 2053 1157808 ns/op 2000039 B/op 10 allocs/op
BenchmarkDecodeToRGBACatmullRom-8 1915 1263899 ns/op 2262274 B/op 13 allocs/op
BenchmarkDecodeToRGBAStdLib-8 849 2839869 ns/op 1455735 B/op 7 allocs/op
Difference with assembly optimizations (noasm vs asm):
benchmark old ns/op new ns/op delta
BenchmarkDecodeBaseline420-8 1584411 742291 -53.15%
BenchmarkDecodeProgressive420-8 2882890 2128195 -26.18%
BenchmarkDecodeConfig-8 453.7 410.4 -9.54%
BenchmarkDecodeToRGBANearestNeighbor-8 2942280 1157808 -60.65%
BenchmarkDecodeToRGBACatmullRom-8 4016591 1263899 -68.53%
BenchmarkIdct-8 144.1 36.68 -74.55%
BenchmarkUpsampleNearestNeighbor-8 359914 116759 -67.56%
BenchmarkUpsampleCatmullRom-8 2703968 319568 -88.18%
noasm- do not use SIMD optimizations
