crypto/sha1: native ARM assembler version

crypto/sha1: native ARM assembler version

An ARM version of sha1block.go with a big improvement in
throughput (up to 3x).

This is a fully unrolled version for maximum performance but also
maximum code size.

Code size

  Before 1636 bytes
  After 5508 bytes
  3.4x larger

Benchmarks on Samsung Exynos 5 ARMv7 Chromebook

benchmark              old ns/op    new ns/op    delta
BenchmarkHash8Bytes         2178         1365  -37.33%
BenchmarkHash1K            19977         7046  -64.73%
BenchmarkHash8K           146436        48919  -66.59%

benchmark               old MB/s     new MB/s  speedup
BenchmarkHash8Bytes         3.67         5.86    1.60x
BenchmarkHash1K            51.26       145.33    2.84x
BenchmarkHash8K            55.94       167.46    2.99x

[dave_cheney.net, 2014-01-28 10:57:02]

1Ghz pandaboard, dual Cortex-A9

panda(~/go/src/pkg/crypto/sha1) % ~/go/misc/benchcmp {old,new}.txt
benchmark              old ns/op    new ns/op    delta
BenchmarkHash8Bytes         8687         4182  -51.86%
BenchmarkHash1K           106347        34841  -67.24%
BenchmarkHash8K           785537       250411  -68.12%

benchmark               old MB/s     new MB/s  speedup
BenchmarkHash8Bytes         0.92         1.91    2.08x
BenchmarkHash1K             9.63        29.39    3.05x
BenchmarkHash8K            10.43        32.71    3.14x

[dave_cheney.net, 2014-01-28 11:10:17]

On 2014/01/28 10:57:02, dfc wrote:
> 1Ghz pandaboard, dual Cortex-A9
> 
> panda(~/go/src/pkg/crypto/sha1) % ~/go/misc/benchcmp {old,new}.txt
> benchmark              old ns/op    new ns/op    delta
> BenchmarkHash8Bytes         8687         4182  -51.86%
> BenchmarkHash1K           106347        34841  -67.24%
> BenchmarkHash8K           785537       250411  -68.12%
> 
> benchmark               old MB/s     new MB/s  speedup
> BenchmarkHash8Bytes         0.92         1.91    2.08x
> BenchmarkHash1K             9.63        29.39    3.05x
> BenchmarkHash8K            10.43        32.71    3.14x

dfc@qnap:~/go/src/pkg/crypto/sha1$ cat /proc/cpuinfo 
processor       : 0
model name      : Feroceon 88FR131 rev 1 (v5l)
BogoMIPS        : 1980.82
Features        : swp half thumb fastmult edsp 
CPU implementer : 0x56
CPU architecture: 5TE
CPU variant     : 0x2
CPU part        : 0x131
CPU revision    : 1

Hardware        : QNAP TS-119/TS-219
Revision        : 0000
Serial          : 0000000000000000
dfc@qnap:~/go/src/pkg/crypto/sha1$ ~/go/misc/benchcmp {old,new}.txt
benchmark              old ns/op    new ns/op    delta
BenchmarkHash8Bytes         4539         2019  -55.52%
BenchmarkHash1K            56114        14468  -74.22%
BenchmarkHash8K           417030       102616  -75.39%

benchmark               old MB/s     new MB/s  speedup
BenchmarkHash8Bytes         1.76         3.96    2.25x
BenchmarkHash1K            18.25        70.77    3.88x
BenchmarkHash8K            19.64        79.83    4.06x