diff options
| author | Tianon Gravi <admwiggin@gmail.com> | 2015-01-15 12:50:40 -0700 |
|---|---|---|
| committer | Tianon Gravi <admwiggin@gmail.com> | 2015-01-15 12:50:40 -0700 |
| commit | 2a0db60599fdd75b1bc3e297180fbe1282763759 (patch) | |
| tree | 68d43c3e30d9ab961ddf6b7365201ca6b675b253 /src/hash/crc32/crc32_amd64p32.s | |
| parent | ef33cba3c8de6c431df56503df51fcd3a473c89e (diff) | |
| parent | f154da9e12608589e8d5f0508f908a0c3e88a1bb (diff) | |
| download | golang-2a0db60599fdd75b1bc3e297180fbe1282763759.tar.gz | |
Merge tag 'upstream/1.4' into debian-experimental
* tag 'upstream/1.4':
Imported Upstream version 1.4
Diffstat (limited to 'src/hash/crc32/crc32_amd64p32.s')
| -rw-r--r-- | src/hash/crc32/crc32_amd64p32.s | 64 |
1 files changed, 64 insertions, 0 deletions
diff --git a/src/hash/crc32/crc32_amd64p32.s b/src/hash/crc32/crc32_amd64p32.s new file mode 100644 index 000000000..b6770eba3 --- /dev/null +++ b/src/hash/crc32/crc32_amd64p32.s @@ -0,0 +1,64 @@ +// Copyright 2011 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +#include "textflag.h" + +// func castagnoliSSE42(crc uint32, p []byte) uint32 +TEXT ·castagnoliSSE42(SB),NOSPLIT,$0 + MOVL crc+0(FP), AX // CRC value + MOVL p+4(FP), SI // data pointer + MOVL p_len+8(FP), CX // len(p) + + NOTL AX + + /* If there's less than 8 bytes to process, we do it byte-by-byte. */ + CMPQ CX, $8 + JL cleanup + + /* Process individual bytes until the input is 8-byte aligned. */ +startup: + MOVQ SI, BX + ANDQ $7, BX + JZ aligned + + CRC32B (SI), AX + DECQ CX + INCQ SI + JMP startup + +aligned: + /* The input is now 8-byte aligned and we can process 8-byte chunks. */ + CMPQ CX, $8 + JL cleanup + + CRC32Q (SI), AX + ADDQ $8, SI + SUBQ $8, CX + JMP aligned + +cleanup: + /* We may have some bytes left over that we process one at a time. */ + CMPQ CX, $0 + JE done + + CRC32B (SI), AX + INCQ SI + DECQ CX + JMP cleanup + +done: + NOTL AX + MOVL AX, ret+16(FP) + RET + +// func haveSSE42() bool +TEXT ·haveSSE42(SB),NOSPLIT,$0 + XORQ AX, AX + INCL AX + CPUID + SHRQ $20, CX + ANDQ $1, CX + MOVB CX, ret+0(FP) + RET + |