VictoriaMetrics/vendor/github.com/klauspost/compress/s2/decode_amd64.s
Aliaksandr Valialkin 34db3fdd3f
app/vmagent/remotewrite: add benchmarks for comparing the performance of standard Snappy encoder with github.com/klauspost/compress/s2 encoder
The standard Snappy encoder from github.com/golang/snappy shows quite good performance number
for compressing the Prometheus remote_write proto messages according to the added benchmarks,
so there is no need in switching to github.com/klauspost/compress/s2 yet.
2022-09-19 14:27:56 +03:00

569 lines
14 KiB
ArmAsm

// Copyright 2016 The Go Authors. All rights reserved.
// Copyright (c) 2019 Klaus Post. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
// +build !appengine
// +build gc
// +build !noasm
#include "textflag.h"
#define R_TMP0 AX
#define R_TMP1 BX
#define R_LEN CX
#define R_OFF DX
#define R_SRC SI
#define R_DST DI
#define R_DBASE R8
#define R_DLEN R9
#define R_DEND R10
#define R_SBASE R11
#define R_SLEN R12
#define R_SEND R13
#define R_TMP2 R14
#define R_TMP3 R15
// The asm code generally follows the pure Go code in decode_other.go, except
// where marked with a "!!!".
// func decode(dst, src []byte) int
//
// All local variables fit into registers. The non-zero stack size is only to
// spill registers and push args when issuing a CALL. The register allocation:
// - R_TMP0 scratch
// - R_TMP1 scratch
// - R_LEN length or x (shared)
// - R_OFF offset
// - R_SRC &src[s]
// - R_DST &dst[d]
// + R_DBASE dst_base
// + R_DLEN dst_len
// + R_DEND dst_base + dst_len
// + R_SBASE src_base
// + R_SLEN src_len
// + R_SEND src_base + src_len
// - R_TMP2 used by doCopy
// - R_TMP3 used by doCopy
//
// The registers R_DBASE-R_SEND (marked with a "+") are set at the start of the
// function, and after a CALL returns, and are not otherwise modified.
//
// The d variable is implicitly R_DST - R_DBASE, and len(dst)-d is R_DEND - R_DST.
// The s variable is implicitly R_SRC - R_SBASE, and len(src)-s is R_SEND - R_SRC.
TEXT ·s2Decode(SB), NOSPLIT, $48-56
// Initialize R_SRC, R_DST and R_DBASE-R_SEND.
MOVQ dst_base+0(FP), R_DBASE
MOVQ dst_len+8(FP), R_DLEN
MOVQ R_DBASE, R_DST
MOVQ R_DBASE, R_DEND
ADDQ R_DLEN, R_DEND
MOVQ src_base+24(FP), R_SBASE
MOVQ src_len+32(FP), R_SLEN
MOVQ R_SBASE, R_SRC
MOVQ R_SBASE, R_SEND
ADDQ R_SLEN, R_SEND
XORQ R_OFF, R_OFF
loop:
// for s < len(src)
CMPQ R_SRC, R_SEND
JEQ end
// R_LEN = uint32(src[s])
//
// switch src[s] & 0x03
MOVBLZX (R_SRC), R_LEN
MOVL R_LEN, R_TMP1
ANDL $3, R_TMP1
CMPL R_TMP1, $1
JAE tagCopy
// ----------------------------------------
// The code below handles literal tags.
// case tagLiteral:
// x := uint32(src[s] >> 2)
// switch
SHRL $2, R_LEN
CMPL R_LEN, $60
JAE tagLit60Plus
// case x < 60:
// s++
INCQ R_SRC
doLit:
// This is the end of the inner "switch", when we have a literal tag.
//
// We assume that R_LEN == x and x fits in a uint32, where x is the variable
// used in the pure Go decode_other.go code.
// length = int(x) + 1
//
// Unlike the pure Go code, we don't need to check if length <= 0 because
// R_LEN can hold 64 bits, so the increment cannot overflow.
INCQ R_LEN
// Prepare to check if copying length bytes will run past the end of dst or
// src.
//
// R_TMP0 = len(dst) - d
// R_TMP1 = len(src) - s
MOVQ R_DEND, R_TMP0
SUBQ R_DST, R_TMP0
MOVQ R_SEND, R_TMP1
SUBQ R_SRC, R_TMP1
// !!! Try a faster technique for short (16 or fewer bytes) copies.
//
// if length > 16 || len(dst)-d < 16 || len(src)-s < 16 {
// goto callMemmove // Fall back on calling runtime·memmove.
// }
//
// The C++ snappy code calls this TryFastAppend. It also checks len(src)-s
// against 21 instead of 16, because it cannot assume that all of its input
// is contiguous in memory and so it needs to leave enough source bytes to
// read the next tag without refilling buffers, but Go's Decode assumes
// contiguousness (the src argument is a []byte).
CMPQ R_LEN, $16
JGT callMemmove
CMPQ R_TMP0, $16
JLT callMemmove
CMPQ R_TMP1, $16
JLT callMemmove
// !!! Implement the copy from src to dst as a 16-byte load and store.
// (Decode's documentation says that dst and src must not overlap.)
//
// This always copies 16 bytes, instead of only length bytes, but that's
// OK. If the input is a valid Snappy encoding then subsequent iterations
// will fix up the overrun. Otherwise, Decode returns a nil []byte (and a
// non-nil error), so the overrun will be ignored.
//
// Note that on amd64, it is legal and cheap to issue unaligned 8-byte or
// 16-byte loads and stores. This technique probably wouldn't be as
// effective on architectures that are fussier about alignment.
MOVOU 0(R_SRC), X0
MOVOU X0, 0(R_DST)
// d += length
// s += length
ADDQ R_LEN, R_DST
ADDQ R_LEN, R_SRC
JMP loop
callMemmove:
// if length > len(dst)-d || length > len(src)-s { etc }
CMPQ R_LEN, R_TMP0
JGT errCorrupt
CMPQ R_LEN, R_TMP1
JGT errCorrupt
// copy(dst[d:], src[s:s+length])
//
// This means calling runtime·memmove(&dst[d], &src[s], length), so we push
// R_DST, R_SRC and R_LEN as arguments. Coincidentally, we also need to spill those
// three registers to the stack, to save local variables across the CALL.
MOVQ R_DST, 0(SP)
MOVQ R_SRC, 8(SP)
MOVQ R_LEN, 16(SP)
MOVQ R_DST, 24(SP)
MOVQ R_SRC, 32(SP)
MOVQ R_LEN, 40(SP)
MOVQ R_OFF, 48(SP)
CALL runtime·memmove(SB)
// Restore local variables: unspill registers from the stack and
// re-calculate R_DBASE-R_SEND.
MOVQ 24(SP), R_DST
MOVQ 32(SP), R_SRC
MOVQ 40(SP), R_LEN
MOVQ 48(SP), R_OFF
MOVQ dst_base+0(FP), R_DBASE
MOVQ dst_len+8(FP), R_DLEN
MOVQ R_DBASE, R_DEND
ADDQ R_DLEN, R_DEND
MOVQ src_base+24(FP), R_SBASE
MOVQ src_len+32(FP), R_SLEN
MOVQ R_SBASE, R_SEND
ADDQ R_SLEN, R_SEND
// d += length
// s += length
ADDQ R_LEN, R_DST
ADDQ R_LEN, R_SRC
JMP loop
tagLit60Plus:
// !!! This fragment does the
//
// s += x - 58; if uint(s) > uint(len(src)) { etc }
//
// checks. In the asm version, we code it once instead of once per switch case.
ADDQ R_LEN, R_SRC
SUBQ $58, R_SRC
CMPQ R_SRC, R_SEND
JA errCorrupt
// case x == 60:
CMPL R_LEN, $61
JEQ tagLit61
JA tagLit62Plus
// x = uint32(src[s-1])
MOVBLZX -1(R_SRC), R_LEN
JMP doLit
tagLit61:
// case x == 61:
// x = uint32(src[s-2]) | uint32(src[s-1])<<8
MOVWLZX -2(R_SRC), R_LEN
JMP doLit
tagLit62Plus:
CMPL R_LEN, $62
JA tagLit63
// case x == 62:
// x = uint32(src[s-3]) | uint32(src[s-2])<<8 | uint32(src[s-1])<<16
// We read one byte, safe to read one back, since we are just reading tag.
// x = binary.LittleEndian.Uint32(src[s-1:]) >> 8
MOVL -4(R_SRC), R_LEN
SHRL $8, R_LEN
JMP doLit
tagLit63:
// case x == 63:
// x = uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24
MOVL -4(R_SRC), R_LEN
JMP doLit
// The code above handles literal tags.
// ----------------------------------------
// The code below handles copy tags.
tagCopy4:
// case tagCopy4:
// s += 5
ADDQ $5, R_SRC
// if uint(s) > uint(len(src)) { etc }
CMPQ R_SRC, R_SEND
JA errCorrupt
// length = 1 + int(src[s-5])>>2
SHRQ $2, R_LEN
INCQ R_LEN
// offset = int(uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24)
MOVLQZX -4(R_SRC), R_OFF
JMP doCopy
tagCopy2:
// case tagCopy2:
// s += 3
ADDQ $3, R_SRC
// if uint(s) > uint(len(src)) { etc }
CMPQ R_SRC, R_SEND
JA errCorrupt
// length = 1 + int(src[s-3])>>2
SHRQ $2, R_LEN
INCQ R_LEN
// offset = int(uint32(src[s-2]) | uint32(src[s-1])<<8)
MOVWQZX -2(R_SRC), R_OFF
JMP doCopy
tagCopy:
// We have a copy tag. We assume that:
// - R_TMP1 == src[s] & 0x03
// - R_LEN == src[s]
CMPQ R_TMP1, $2
JEQ tagCopy2
JA tagCopy4
// case tagCopy1:
// s += 2
ADDQ $2, R_SRC
// if uint(s) > uint(len(src)) { etc }
CMPQ R_SRC, R_SEND
JA errCorrupt
// offset = int(uint32(src[s-2])&0xe0<<3 | uint32(src[s-1]))
// length = 4 + int(src[s-2])>>2&0x7
MOVBQZX -1(R_SRC), R_TMP1
MOVQ R_LEN, R_TMP0
SHRQ $2, R_LEN
ANDQ $0xe0, R_TMP0
ANDQ $7, R_LEN
SHLQ $3, R_TMP0
ADDQ $4, R_LEN
ORQ R_TMP1, R_TMP0
// check if repeat code, ZF set by ORQ.
JZ repeatCode
// This is a regular copy, transfer our temporary value to R_OFF (length)
MOVQ R_TMP0, R_OFF
JMP doCopy
// This is a repeat code.
repeatCode:
// If length < 9, reuse last offset, with the length already calculated.
CMPQ R_LEN, $9
JL doCopyRepeat
// Read additional bytes for length.
JE repeatLen1
// Rare, so the extra branch shouldn't hurt too much.
CMPQ R_LEN, $10
JE repeatLen2
JMP repeatLen3
// Read repeat lengths.
repeatLen1:
// s ++
ADDQ $1, R_SRC
// if uint(s) > uint(len(src)) { etc }
CMPQ R_SRC, R_SEND
JA errCorrupt
// length = src[s-1] + 8
MOVBQZX -1(R_SRC), R_LEN
ADDL $8, R_LEN
JMP doCopyRepeat
repeatLen2:
// s +=2
ADDQ $2, R_SRC
// if uint(s) > uint(len(src)) { etc }
CMPQ R_SRC, R_SEND
JA errCorrupt
// length = uint32(src[s-2]) | (uint32(src[s-1])<<8) + (1 << 8)
MOVWQZX -2(R_SRC), R_LEN
ADDL $260, R_LEN
JMP doCopyRepeat
repeatLen3:
// s +=3
ADDQ $3, R_SRC
// if uint(s) > uint(len(src)) { etc }
CMPQ R_SRC, R_SEND
JA errCorrupt
// length = uint32(src[s-3]) | (uint32(src[s-2])<<8) | (uint32(src[s-1])<<16) + (1 << 16)
// Read one byte further back (just part of the tag, shifted out)
MOVL -4(R_SRC), R_LEN
SHRL $8, R_LEN
ADDL $65540, R_LEN
JMP doCopyRepeat
doCopy:
// This is the end of the outer "switch", when we have a copy tag.
//
// We assume that:
// - R_LEN == length && R_LEN > 0
// - R_OFF == offset
// if d < offset { etc }
MOVQ R_DST, R_TMP1
SUBQ R_DBASE, R_TMP1
CMPQ R_TMP1, R_OFF
JLT errCorrupt
// Repeat values can skip the test above, since any offset > 0 will be in dst.
doCopyRepeat:
// if offset <= 0 { etc }
CMPQ R_OFF, $0
JLE errCorrupt
// if length > len(dst)-d { etc }
MOVQ R_DEND, R_TMP1
SUBQ R_DST, R_TMP1
CMPQ R_LEN, R_TMP1
JGT errCorrupt
// forwardCopy(dst[d:d+length], dst[d-offset:]); d += length
//
// Set:
// - R_TMP2 = len(dst)-d
// - R_TMP3 = &dst[d-offset]
MOVQ R_DEND, R_TMP2
SUBQ R_DST, R_TMP2
MOVQ R_DST, R_TMP3
SUBQ R_OFF, R_TMP3
// !!! Try a faster technique for short (16 or fewer bytes) forward copies.
//
// First, try using two 8-byte load/stores, similar to the doLit technique
// above. Even if dst[d:d+length] and dst[d-offset:] can overlap, this is
// still OK if offset >= 8. Note that this has to be two 8-byte load/stores
// and not one 16-byte load/store, and the first store has to be before the
// second load, due to the overlap if offset is in the range [8, 16).
//
// if length > 16 || offset < 8 || len(dst)-d < 16 {
// goto slowForwardCopy
// }
// copy 16 bytes
// d += length
CMPQ R_LEN, $16
JGT slowForwardCopy
CMPQ R_OFF, $8
JLT slowForwardCopy
CMPQ R_TMP2, $16
JLT slowForwardCopy
MOVQ 0(R_TMP3), R_TMP0
MOVQ R_TMP0, 0(R_DST)
MOVQ 8(R_TMP3), R_TMP1
MOVQ R_TMP1, 8(R_DST)
ADDQ R_LEN, R_DST
JMP loop
slowForwardCopy:
// !!! If the forward copy is longer than 16 bytes, or if offset < 8, we
// can still try 8-byte load stores, provided we can overrun up to 10 extra
// bytes. As above, the overrun will be fixed up by subsequent iterations
// of the outermost loop.
//
// The C++ snappy code calls this technique IncrementalCopyFastPath. Its
// commentary says:
//
// ----
//
// The main part of this loop is a simple copy of eight bytes at a time
// until we've copied (at least) the requested amount of bytes. However,
// if d and d-offset are less than eight bytes apart (indicating a
// repeating pattern of length < 8), we first need to expand the pattern in
// order to get the correct results. For instance, if the buffer looks like
// this, with the eight-byte <d-offset> and <d> patterns marked as
// intervals:
//
// abxxxxxxxxxxxx
// [------] d-offset
// [------] d
//
// a single eight-byte copy from <d-offset> to <d> will repeat the pattern
// once, after which we can move <d> two bytes without moving <d-offset>:
//
// ababxxxxxxxxxx
// [------] d-offset
// [------] d
//
// and repeat the exercise until the two no longer overlap.
//
// This allows us to do very well in the special case of one single byte
// repeated many times, without taking a big hit for more general cases.
//
// The worst case of extra writing past the end of the match occurs when
// offset == 1 and length == 1; the last copy will read from byte positions
// [0..7] and write to [4..11], whereas it was only supposed to write to
// position 1. Thus, ten excess bytes.
//
// ----
//
// That "10 byte overrun" worst case is confirmed by Go's
// TestSlowForwardCopyOverrun, which also tests the fixUpSlowForwardCopy
// and finishSlowForwardCopy algorithm.
//
// if length > len(dst)-d-10 {
// goto verySlowForwardCopy
// }
SUBQ $10, R_TMP2
CMPQ R_LEN, R_TMP2
JGT verySlowForwardCopy
// We want to keep the offset, so we use R_TMP2 from here.
MOVQ R_OFF, R_TMP2
makeOffsetAtLeast8:
// !!! As above, expand the pattern so that offset >= 8 and we can use
// 8-byte load/stores.
//
// for offset < 8 {
// copy 8 bytes from dst[d-offset:] to dst[d:]
// length -= offset
// d += offset
// offset += offset
// // The two previous lines together means that d-offset, and therefore
// // R_TMP3, is unchanged.
// }
CMPQ R_TMP2, $8
JGE fixUpSlowForwardCopy
MOVQ (R_TMP3), R_TMP1
MOVQ R_TMP1, (R_DST)
SUBQ R_TMP2, R_LEN
ADDQ R_TMP2, R_DST
ADDQ R_TMP2, R_TMP2
JMP makeOffsetAtLeast8
fixUpSlowForwardCopy:
// !!! Add length (which might be negative now) to d (implied by R_DST being
// &dst[d]) so that d ends up at the right place when we jump back to the
// top of the loop. Before we do that, though, we save R_DST to R_TMP0 so that, if
// length is positive, copying the remaining length bytes will write to the
// right place.
MOVQ R_DST, R_TMP0
ADDQ R_LEN, R_DST
finishSlowForwardCopy:
// !!! Repeat 8-byte load/stores until length <= 0. Ending with a negative
// length means that we overrun, but as above, that will be fixed up by
// subsequent iterations of the outermost loop.
CMPQ R_LEN, $0
JLE loop
MOVQ (R_TMP3), R_TMP1
MOVQ R_TMP1, (R_TMP0)
ADDQ $8, R_TMP3
ADDQ $8, R_TMP0
SUBQ $8, R_LEN
JMP finishSlowForwardCopy
verySlowForwardCopy:
// verySlowForwardCopy is a simple implementation of forward copy. In C
// parlance, this is a do/while loop instead of a while loop, since we know
// that length > 0. In Go syntax:
//
// for {
// dst[d] = dst[d - offset]
// d++
// length--
// if length == 0 {
// break
// }
// }
MOVB (R_TMP3), R_TMP1
MOVB R_TMP1, (R_DST)
INCQ R_TMP3
INCQ R_DST
DECQ R_LEN
JNZ verySlowForwardCopy
JMP loop
// The code above handles copy tags.
// ----------------------------------------
end:
// This is the end of the "for s < len(src)".
//
// if d != len(dst) { etc }
CMPQ R_DST, R_DEND
JNE errCorrupt
// return 0
MOVQ $0, ret+48(FP)
RET
errCorrupt:
// return decodeErrCodeCorrupt
MOVQ $1, ret+48(FP)
RET