Rewrite AVX2 loop in assembly
Helps to get rid of MOV and generating constants for each iteration. ``` name old time/op new time/op delta Sum/AVX2Inline_digest-8 1.57ms ± 2% 1.41ms ± 0% -10.52% (p=0.000 n=9+9) name old speed new speed delta Sum/AVX2Inline_digest-8 63.6MB/s ± 1% 71.1MB/s ± 0% +11.76% (p=0.000 n=9+9) ``` Signed-off-by: Evgenii Stratonikov <evgeniy@nspcc.ru>
This commit is contained in:
parent
d7c96f5d2e
commit
73d978c31e
2 changed files with 18 additions and 6 deletions
|
@ -18,8 +18,8 @@
|
||||||
VPAND out_1, Y2, Y4 \
|
VPAND out_1, Y2, Y4 \
|
||||||
VPXOR Y4, in_1, out_2 \
|
VPXOR Y4, in_1, out_2 \
|
||||||
|
|
||||||
// func mulByteRightx2(c00c10, c01c11 *[4]uint64, b byte)
|
// func mulByteSliceRightx2(c00c10, c01c11 *[4]uint64, n int, data *byte)
|
||||||
TEXT ·mulByteRightx2(SB),NOSPLIT,$0
|
TEXT ·mulByteSliceRightx2(SB),NOSPLIT,$0
|
||||||
MOVQ c00c10+0(FP), AX
|
MOVQ c00c10+0(FP), AX
|
||||||
VMOVDQU (AX), Y0
|
VMOVDQU (AX), Y0
|
||||||
MOVQ c01c11+8(FP), BX
|
MOVQ c01c11+8(FP), BX
|
||||||
|
@ -31,8 +31,17 @@ TEXT ·mulByteRightx2(SB),NOSPLIT,$0
|
||||||
VPSUBW Y14, Y13, Y12 // Y12 = 0x00010001... (packed words of 1)
|
VPSUBW Y14, Y13, Y12 // Y12 = 0x00010001... (packed words of 1)
|
||||||
VPSLLQ $63, Y10, Y14 // Y14 = 0x10000000... (packed quad-words with HSB set)
|
VPSLLQ $63, Y10, Y14 // Y14 = 0x10000000... (packed quad-words with HSB set)
|
||||||
|
|
||||||
VPBROADCASTB b+16(FP), X10 // X10 = packed bytes of b.
|
MOVQ n+16(FP), CX
|
||||||
|
MOVQ data+24(FP), DX
|
||||||
|
|
||||||
|
loop:
|
||||||
|
CMPQ CX, $0
|
||||||
|
JEQ finish
|
||||||
|
SUBQ $1, CX
|
||||||
|
|
||||||
|
VPBROADCASTB (DX), X10 // X10 = packed bytes of b.
|
||||||
VPMOVZXBW X10, Y10 // Extend with zeroes to packed words.
|
VPMOVZXBW X10, Y10 // Extend with zeroes to packed words.
|
||||||
|
ADDQ $1, DX
|
||||||
|
|
||||||
mulBit($7, Y0, Y8, Y5, Y6)
|
mulBit($7, Y0, Y8, Y5, Y6)
|
||||||
mulBit($6, Y5, Y6, Y0, Y8)
|
mulBit($6, Y5, Y6, Y0, Y8)
|
||||||
|
@ -43,6 +52,9 @@ TEXT ·mulByteRightx2(SB),NOSPLIT,$0
|
||||||
mulBit($1, Y0, Y8, Y5, Y6)
|
mulBit($1, Y0, Y8, Y5, Y6)
|
||||||
mulBit($0, Y5, Y6, Y0, Y8)
|
mulBit($0, Y5, Y6, Y0, Y8)
|
||||||
|
|
||||||
|
JMP loop
|
||||||
|
|
||||||
|
finish:
|
||||||
VMOVDQU Y8, (BX)
|
VMOVDQU Y8, (BX)
|
||||||
VMOVDQU Y0, (AX)
|
VMOVDQU Y0, (AX)
|
||||||
|
|
||||||
|
|
|
@ -25,8 +25,8 @@ func newAVX2Inline() *digest3 {
|
||||||
|
|
||||||
func (d *digest3) Write(data []byte) (n int, err error) {
|
func (d *digest3) Write(data []byte) (n int, err error) {
|
||||||
n = len(data)
|
n = len(data)
|
||||||
for _, b := range data {
|
if len(data) != 0 {
|
||||||
mulByteRightx2(&d.x[0], &d.x[1], b)
|
mulByteSliceRightx2(&d.x[0], &d.x[1], n, &data[0])
|
||||||
}
|
}
|
||||||
return
|
return
|
||||||
}
|
}
|
||||||
|
@ -56,4 +56,4 @@ func (d *digest3) checkSum() (b [hashSize]byte) {
|
||||||
return
|
return
|
||||||
}
|
}
|
||||||
|
|
||||||
func mulByteRightx2(c00c10 *avx2.GF127x2, c01c11 *avx2.GF127x2, b byte)
|
func mulByteSliceRightx2(c00c10 *avx2.GF127x2, c01c11 *avx2.GF127x2, n int, data *byte)
|
||||||
|
|
Loading…
Reference in a new issue