crypto/sha256, crypto/sha512: 1.3x speedup

SHA-256:

benchmark          old ns/op    new ns/op    delta
BenchmarkHash1K        21686        16912  -22.01%
BenchmarkHash8K       173216       135020  -22.05%

benchmark           old MB/s     new MB/s  speedup
BenchmarkHash1K        47.22        60.55    1.28x
BenchmarkHash8K        47.29        60.67    1.28x

SHA-512:

benchmark          old ns/op    new ns/op    delta
BenchmarkHash1K        14323        11163  -22.06%
BenchmarkHash8K       114120        88693  -22.28%

benchmark           old MB/s     new MB/s  speedup
BenchmarkHash1K        71.49        91.73    1.28x
BenchmarkHash8K        71.78        92.36    1.29x

R=golang-dev, agl
CC=golang-dev
https://golang.org/cl/6584071

src/pkg/crypto/sha256/sha256block.go

@@ -86,10 +86,10 @@ func block(dig *digest, p []byte) {
 			w[i] = uint32(p[j])<<24 | uint32(p[j+1])<<16 | uint32(p[j+2])<<8 | uint32(p[j+3])
 		}
 		for i := 16; i < 64; i++ {
-			t1 := (w[i-2]>>17 | w[i-2]<<(32-17)) ^ (w[i-2]>>19 | w[i-2]<<(32-19)) ^ (w[i-2] >> 10)
-
-			t2 := (w[i-15]>>7 | w[i-15]<<(32-7)) ^ (w[i-15]>>18 | w[i-15]<<(32-18)) ^ (w[i-15] >> 3)
-
+			v1 := w[i-2]
+			t1 := (v1>>17 | v1<<(32-17)) ^ (v1>>19 | v1<<(32-19)) ^ (v1 >> 10)
+			v2 := w[i-15]
+			t2 := (v2>>7 | v2<<(32-7)) ^ (v2>>18 | v2<<(32-18)) ^ (v2 >> 3)
 			w[i] = t1 + w[i-7] + t2 + w[i-16]
 		}
 

src/pkg/crypto/sha512/sha512block.go

@@ -101,9 +101,10 @@ func block(dig *digest, p []byte) {
 				uint64(p[j+4])<<24 | uint64(p[j+5])<<16 | uint64(p[j+6])<<8 | uint64(p[j+7])
 		}
 		for i := 16; i < 80; i++ {
-			t1 := (w[i-2]>>19 | w[i-2]<<(64-19)) ^ (w[i-2]>>61 | w[i-2]<<(64-61)) ^ (w[i-2] >> 6)
-
-			t2 := (w[i-15]>>1 | w[i-15]<<(64-1)) ^ (w[i-15]>>8 | w[i-15]<<(64-8)) ^ (w[i-15] >> 7)
+			v1 := w[i-2]
+			t1 := (v1>>19 | v1<<(64-19)) ^ (v1>>61 | v1<<(64-61)) ^ (v1 >> 6)
+			v2 := w[i-15]
+			t2 := (v2>>1 | v2<<(64-1)) ^ (v2>>8 | v2<<(64-8)) ^ (v2 >> 7)
 
 			w[i] = t1 + w[i-7] + t2 + w[i-16]
 		}