snappy/encode_arm64.s

723 строки
16 KiB
ArmAsm

// Copyright 2020 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.
// +build !appengine
// +build gc
// +build !noasm
#include "textflag.h"
// The asm code generally follows the pure Go code in encode_other.go, except
// where marked with a "!!!".
// ----------------------------------------------------------------------------
// func emitLiteral(dst, lit []byte) int
//
// All local variables fit into registers. The register allocation:
// - R3 len(lit)
// - R4 n
// - R6 return value
// - R8 &dst[i]
// - R10 &lit[0]
//
// The 32 bytes of stack space is to call runtime·memmove.
//
// The unusual register allocation of local variables, such as R10 for the
// source pointer, matches the allocation used at the call site in encodeBlock,
// which makes it easier to manually inline this function.
TEXT ·emitLiteral(SB), NOSPLIT, $40-56
MOVD dst_base+0(FP), R8
MOVD lit_base+24(FP), R10
MOVD lit_len+32(FP), R3
MOVD R3, R6
MOVW R3, R4
SUBW $1, R4, R4
CMPW $60, R4
BLT oneByte
CMPW $256, R4
BLT twoBytes
threeBytes:
MOVD $0xf4, R2
MOVB R2, 0(R8)
MOVW R4, 1(R8)
ADD $3, R8, R8
ADD $3, R6, R6
B memmove
twoBytes:
MOVD $0xf0, R2
MOVB R2, 0(R8)
MOVB R4, 1(R8)
ADD $2, R8, R8
ADD $2, R6, R6
B memmove
oneByte:
LSLW $2, R4, R4
MOVB R4, 0(R8)
ADD $1, R8, R8
ADD $1, R6, R6
memmove:
MOVD R6, ret+48(FP)
// copy(dst[i:], lit)
//
// This means calling runtime·memmove(&dst[i], &lit[0], len(lit)), so we push
// R8, R10 and R3 as arguments.
MOVD R8, 8(RSP)
MOVD R10, 16(RSP)
MOVD R3, 24(RSP)
CALL runtime·memmove(SB)
RET
// ----------------------------------------------------------------------------
// func emitCopy(dst []byte, offset, length int) int
//
// All local variables fit into registers. The register allocation:
// - R3 length
// - R7 &dst[0]
// - R8 &dst[i]
// - R11 offset
//
// The unusual register allocation of local variables, such as R11 for the
// offset, matches the allocation used at the call site in encodeBlock, which
// makes it easier to manually inline this function.
TEXT ·emitCopy(SB), NOSPLIT, $0-48
MOVD dst_base+0(FP), R8
MOVD R8, R7
MOVD offset+24(FP), R11
MOVD length+32(FP), R3
loop0:
// for length >= 68 { etc }
CMPW $68, R3
BLT step1
// Emit a length 64 copy, encoded as 3 bytes.
MOVD $0xfe, R2
MOVB R2, 0(R8)
MOVW R11, 1(R8)
ADD $3, R8, R8
SUB $64, R3, R3
B loop0
step1:
// if length > 64 { etc }
CMP $64, R3
BLE step2
// Emit a length 60 copy, encoded as 3 bytes.
MOVD $0xee, R2
MOVB R2, 0(R8)
MOVW R11, 1(R8)
ADD $3, R8, R8
SUB $60, R3, R3
step2:
// if length >= 12 || offset >= 2048 { goto step3 }
CMP $12, R3
BGE step3
CMPW $2048, R11
BGE step3
// Emit the remaining copy, encoded as 2 bytes.
MOVB R11, 1(R8)
LSRW $3, R11, R11
AND $0xe0, R11, R11
SUB $4, R3, R3
LSLW $2, R3
AND $0xff, R3, R3
ORRW R3, R11, R11
ORRW $1, R11, R11
MOVB R11, 0(R8)
ADD $2, R8, R8
// Return the number of bytes written.
SUB R7, R8, R8
MOVD R8, ret+40(FP)
RET
step3:
// Emit the remaining copy, encoded as 3 bytes.
SUB $1, R3, R3
AND $0xff, R3, R3
LSLW $2, R3, R3
ORRW $2, R3, R3
MOVB R3, 0(R8)
MOVW R11, 1(R8)
ADD $3, R8, R8
// Return the number of bytes written.
SUB R7, R8, R8
MOVD R8, ret+40(FP)
RET
// ----------------------------------------------------------------------------
// func extendMatch(src []byte, i, j int) int
//
// All local variables fit into registers. The register allocation:
// - R6 &src[0]
// - R7 &src[j]
// - R13 &src[len(src) - 8]
// - R14 &src[len(src)]
// - R15 &src[i]
//
// The unusual register allocation of local variables, such as R15 for a source
// pointer, matches the allocation used at the call site in encodeBlock, which
// makes it easier to manually inline this function.
TEXT ·extendMatch(SB), NOSPLIT, $0-48
MOVD src_base+0(FP), R6
MOVD src_len+8(FP), R14
MOVD i+24(FP), R15
MOVD j+32(FP), R7
ADD R6, R14, R14
ADD R6, R15, R15
ADD R6, R7, R7
MOVD R14, R13
SUB $8, R13, R13
cmp8:
// As long as we are 8 or more bytes before the end of src, we can load and
// compare 8 bytes at a time. If those 8 bytes are equal, repeat.
CMP R13, R7
BHI cmp1
MOVD (R15), R3
MOVD (R7), R4
CMP R4, R3
BNE bsf
ADD $8, R15, R15
ADD $8, R7, R7
B cmp8
bsf:
// If those 8 bytes were not equal, XOR the two 8 byte values, and return
// the index of the first byte that differs.
// RBIT reverses the bit order, then CLZ counts the leading zeros, the
// combination of which finds the least significant bit which is set.
// The arm64 architecture is little-endian, and the shift by 3 converts
// a bit index to a byte index.
EOR R3, R4, R4
RBIT R4, R4
CLZ R4, R4
ADD R4>>3, R7, R7
// Convert from &src[ret] to ret.
SUB R6, R7, R7
MOVD R7, ret+40(FP)
RET
cmp1:
// In src's tail, compare 1 byte at a time.
CMP R7, R14
BLS extendMatchEnd
MOVB (R15), R3
MOVB (R7), R4
CMP R4, R3
BNE extendMatchEnd
ADD $1, R15, R15
ADD $1, R7, R7
B cmp1
extendMatchEnd:
// Convert from &src[ret] to ret.
SUB R6, R7, R7
MOVD R7, ret+40(FP)
RET
// ----------------------------------------------------------------------------
// func encodeBlock(dst, src []byte) (d int)
//
// All local variables fit into registers, other than "var table". The register
// allocation:
// - R3 . .
// - R4 . .
// - R5 64 shift
// - R6 72 &src[0], tableSize
// - R7 80 &src[s]
// - R8 88 &dst[d]
// - R9 96 sLimit
// - R10 . &src[nextEmit]
// - R11 104 prevHash, currHash, nextHash, offset
// - R12 112 &src[base], skip
// - R13 . &src[nextS], &src[len(src) - 8]
// - R14 . len(src), bytesBetweenHashLookups, &src[len(src)], x
// - R15 120 candidate
// - R16 . hash constant, 0x1e35a7bd
// - R17 . &table
// - . 128 table
//
// The second column (64, 72, etc) is the stack offset to spill the registers
// when calling other functions. We could pack this slightly tighter, but it's
// simpler to have a dedicated spill map independent of the function called.
//
// "var table [maxTableSize]uint16" takes up 32768 bytes of stack space. An
// extra 64 bytes, to call other functions, and an extra 64 bytes, to spill
// local variables (registers) during calls gives 32768 + 64 + 64 = 32896.
TEXT ·encodeBlock(SB), 0, $32904-56
MOVD dst_base+0(FP), R8
MOVD src_base+24(FP), R7
MOVD src_len+32(FP), R14
// shift, tableSize := uint32(32-8), 1<<8
MOVD $24, R5
MOVD $256, R6
MOVW $0xa7bd, R16
MOVKW $(0x1e35<<16), R16
calcShift:
// for ; tableSize < maxTableSize && tableSize < len(src); tableSize *= 2 {
// shift--
// }
MOVD $16384, R2
CMP R2, R6
BGE varTable
CMP R14, R6
BGE varTable
SUB $1, R5, R5
LSL $1, R6, R6
B calcShift
varTable:
// var table [maxTableSize]uint16
//
// In the asm code, unlike the Go code, we can zero-initialize only the
// first tableSize elements. Each uint16 element is 2 bytes and each
// iterations writes 64 bytes, so we can do only tableSize/32 writes
// instead of the 2048 writes that would zero-initialize all of table's
// 32768 bytes. This clear could overrun the first tableSize elements, but
// it won't overrun the allocated stack size.
ADD $128, RSP, R17
MOVD R17, R4
// !!! R6 = &src[tableSize]
ADD R6<<1, R17, R6
memclr:
STP.P (ZR, ZR), 64(R4)
STP (ZR, ZR), -48(R4)
STP (ZR, ZR), -32(R4)
STP (ZR, ZR), -16(R4)
CMP R4, R6
BHI memclr
// !!! R6 = &src[0]
MOVD R7, R6
// sLimit := len(src) - inputMargin
MOVD R14, R9
SUB $15, R9, R9
// !!! Pre-emptively spill R5, R6 and R9 to the stack. Their values don't
// change for the rest of the function.
MOVD R5, 64(RSP)
MOVD R6, 72(RSP)
MOVD R9, 96(RSP)
// nextEmit := 0
MOVD R6, R10
// s := 1
ADD $1, R7, R7
// nextHash := hash(load32(src, s), shift)
MOVW 0(R7), R11
MULW R16, R11, R11
LSRW R5, R11, R11
outer:
// for { etc }
// skip := 32
MOVD $32, R12
// nextS := s
MOVD R7, R13
// candidate := 0
MOVD $0, R15
inner0:
// for { etc }
// s := nextS
MOVD R13, R7
// bytesBetweenHashLookups := skip >> 5
MOVD R12, R14
LSR $5, R14, R14
// nextS = s + bytesBetweenHashLookups
ADD R14, R13, R13
// skip += bytesBetweenHashLookups
ADD R14, R12, R12
// if nextS > sLimit { goto emitRemainder }
MOVD R13, R3
SUB R6, R3, R3
CMP R9, R3
BHI emitRemainder
// candidate = int(table[nextHash])
MOVHU 0(R17)(R11<<1), R15
// table[nextHash] = uint16(s)
MOVD R7, R3
SUB R6, R3, R3
MOVH R3, 0(R17)(R11<<1)
// nextHash = hash(load32(src, nextS), shift)
MOVW 0(R13), R11
MULW R16, R11
LSRW R5, R11, R11
// if load32(src, s) != load32(src, candidate) { continue } break
MOVW 0(R7), R3
MOVW (R6)(R15), R4
CMPW R4, R3
BNE inner0
fourByteMatch:
// As per the encode_other.go code:
//
// A 4-byte match has been found. We'll later see etc.
// !!! Jump to a fast path for short (<= 16 byte) literals. See the comment
// on inputMargin in encode.go.
MOVD R7, R3
SUB R10, R3, R3
CMP $16, R3
BLE emitLiteralFastPath
// ----------------------------------------
// Begin inline of the emitLiteral call.
//
// d += emitLiteral(dst[d:], src[nextEmit:s])
MOVW R3, R4
SUBW $1, R4, R4
MOVW $60, R2
CMPW R2, R4
BLT inlineEmitLiteralOneByte
MOVW $256, R2
CMPW R2, R4
BLT inlineEmitLiteralTwoBytes
inlineEmitLiteralThreeBytes:
MOVD $0xf4, R1
MOVB R1, 0(R8)
MOVW R4, 1(R8)
ADD $3, R8, R8
B inlineEmitLiteralMemmove
inlineEmitLiteralTwoBytes:
MOVD $0xf0, R1
MOVB R1, 0(R8)
MOVB R4, 1(R8)
ADD $2, R8, R8
B inlineEmitLiteralMemmove
inlineEmitLiteralOneByte:
LSLW $2, R4, R4
MOVB R4, 0(R8)
ADD $1, R8, R8
inlineEmitLiteralMemmove:
// Spill local variables (registers) onto the stack; call; unspill.
//
// copy(dst[i:], lit)
//
// This means calling runtime·memmove(&dst[i], &lit[0], len(lit)), so we push
// R8, R10 and R3 as arguments.
MOVD R8, 8(RSP)
MOVD R10, 16(RSP)
MOVD R3, 24(RSP)
// Finish the "d +=" part of "d += emitLiteral(etc)".
ADD R3, R8, R8
MOVD R7, 80(RSP)
MOVD R8, 88(RSP)
MOVD R15, 120(RSP)
CALL runtime·memmove(SB)
MOVD 64(RSP), R5
MOVD 72(RSP), R6
MOVD 80(RSP), R7
MOVD 88(RSP), R8
MOVD 96(RSP), R9
MOVD 120(RSP), R15
ADD $128, RSP, R17
MOVW $0xa7bd, R16
MOVKW $(0x1e35<<16), R16
B inner1
inlineEmitLiteralEnd:
// End inline of the emitLiteral call.
// ----------------------------------------
emitLiteralFastPath:
// !!! Emit the 1-byte encoding "uint8(len(lit)-1)<<2".
MOVB R3, R4
SUBW $1, R4, R4
AND $0xff, R4, R4
LSLW $2, R4, R4
MOVB R4, (R8)
ADD $1, R8, R8
// !!! Implement the copy from lit to dst as a 16-byte load and store.
// (Encode's documentation says that dst and src must not overlap.)
//
// This always copies 16 bytes, instead of only len(lit) bytes, but that's
// OK. Subsequent iterations will fix up the overrun.
//
// Note that on arm64, 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.
LDP 0(R10), (R0, R1)
STP (R0, R1), 0(R8)
ADD R3, R8, R8
inner1:
// for { etc }
// base := s
MOVD R7, R12
// !!! offset := base - candidate
MOVD R12, R11
SUB R15, R11, R11
SUB R6, R11, R11
// ----------------------------------------
// Begin inline of the extendMatch call.
//
// s = extendMatch(src, candidate+4, s+4)
// !!! R14 = &src[len(src)]
MOVD src_len+32(FP), R14
ADD R6, R14, R14
// !!! R13 = &src[len(src) - 8]
MOVD R14, R13
SUB $8, R13, R13
// !!! R15 = &src[candidate + 4]
ADD $4, R15, R15
ADD R6, R15, R15
// !!! s += 4
ADD $4, R7, R7
inlineExtendMatchCmp8:
// As long as we are 8 or more bytes before the end of src, we can load and
// compare 8 bytes at a time. If those 8 bytes are equal, repeat.
CMP R13, R7
BHI inlineExtendMatchCmp1
MOVD (R15), R3
MOVD (R7), R4
CMP R4, R3
BNE inlineExtendMatchBSF
ADD $8, R15, R15
ADD $8, R7, R7
B inlineExtendMatchCmp8
inlineExtendMatchBSF:
// If those 8 bytes were not equal, XOR the two 8 byte values, and return
// the index of the first byte that differs.
// RBIT reverses the bit order, then CLZ counts the leading zeros, the
// combination of which finds the least significant bit which is set.
// The arm64 architecture is little-endian, and the shift by 3 converts
// a bit index to a byte index.
EOR R3, R4, R4
RBIT R4, R4
CLZ R4, R4
ADD R4>>3, R7, R7
B inlineExtendMatchEnd
inlineExtendMatchCmp1:
// In src's tail, compare 1 byte at a time.
CMP R7, R14
BLS inlineExtendMatchEnd
MOVB (R15), R3
MOVB (R7), R4
CMP R4, R3
BNE inlineExtendMatchEnd
ADD $1, R15, R15
ADD $1, R7, R7
B inlineExtendMatchCmp1
inlineExtendMatchEnd:
// End inline of the extendMatch call.
// ----------------------------------------
// ----------------------------------------
// Begin inline of the emitCopy call.
//
// d += emitCopy(dst[d:], base-candidate, s-base)
// !!! length := s - base
MOVD R7, R3
SUB R12, R3, R3
inlineEmitCopyLoop0:
// for length >= 68 { etc }
MOVW $68, R2
CMPW R2, R3
BLT inlineEmitCopyStep1
// Emit a length 64 copy, encoded as 3 bytes.
MOVD $0xfe, R1
MOVB R1, 0(R8)
MOVW R11, 1(R8)
ADD $3, R8, R8
SUBW $64, R3, R3
B inlineEmitCopyLoop0
inlineEmitCopyStep1:
// if length > 64 { etc }
MOVW $64, R2
CMPW R2, R3
BLE inlineEmitCopyStep2
// Emit a length 60 copy, encoded as 3 bytes.
MOVD $0xee, R1
MOVB R1, 0(R8)
MOVW R11, 1(R8)
ADD $3, R8, R8
SUBW $60, R3, R3
inlineEmitCopyStep2:
// if length >= 12 || offset >= 2048 { goto inlineEmitCopyStep3 }
MOVW $12, R2
CMPW R2, R3
BGE inlineEmitCopyStep3
MOVW $2048, R2
CMPW R2, R11
BGE inlineEmitCopyStep3
// Emit the remaining copy, encoded as 2 bytes.
MOVB R11, 1(R8)
LSRW $8, R11, R11
LSLW $5, R11, R11
SUBW $4, R3, R3
AND $0xff, R3, R3
LSLW $2, R3, R3
ORRW R3, R11, R11
ORRW $1, R11, R11
MOVB R11, 0(R8)
ADD $2, R8, R8
B inlineEmitCopyEnd
inlineEmitCopyStep3:
// Emit the remaining copy, encoded as 3 bytes.
SUBW $1, R3, R3
LSLW $2, R3, R3
ORRW $2, R3, R3
MOVB R3, 0(R8)
MOVW R11, 1(R8)
ADD $3, R8, R8
inlineEmitCopyEnd:
// End inline of the emitCopy call.
// ----------------------------------------
// nextEmit = s
MOVD R7, R10
// if s >= sLimit { goto emitRemainder }
MOVD R7, R3
SUB R6, R3, R3
CMP R3, R9
BLS emitRemainder
// As per the encode_other.go code:
//
// We could immediately etc.
// x := load64(src, s-1)
MOVD -1(R7), R14
// prevHash := hash(uint32(x>>0), shift)
MOVW R14, R11
MULW R16, R11, R11
LSRW R5, R11, R11
// table[prevHash] = uint16(s-1)
MOVD R7, R3
SUB R6, R3, R3
SUB $1, R3, R3
MOVHU R3, 0(R17)(R11<<1)
// currHash := hash(uint32(x>>8), shift)
LSR $8, R14, R14
MOVW R14, R11
MULW R16, R11, R11
LSRW R5, R11, R11
// candidate = int(table[currHash])
MOVHU 0(R17)(R11<<1), R15
// table[currHash] = uint16(s)
ADD $1, R3, R3
MOVHU R3, 0(R17)(R11<<1)
// if uint32(x>>8) == load32(src, candidate) { continue }
MOVW (R6)(R15), R4
CMPW R4, R14
BEQ inner1
// nextHash = hash(uint32(x>>16), shift)
LSR $8, R14, R14
MOVW R14, R11
MULW R16, R11, R11
LSRW R5, R11, R11
// s++
ADD $1, R7, R7
// break out of the inner1 for loop, i.e. continue the outer loop.
B outer
emitRemainder:
// if nextEmit < len(src) { etc }
MOVD src_len+32(FP), R3
ADD R6, R3, R3
CMP R3, R10
BEQ encodeBlockEnd
// d += emitLiteral(dst[d:], src[nextEmit:])
//
// Push args.
MOVD R8, 8(RSP)
MOVD $0, 16(RSP) // Unnecessary, as the callee ignores it, but conservative.
MOVD $0, 24(RSP) // Unnecessary, as the callee ignores it, but conservative.
MOVD R10, 32(RSP)
SUB R10, R3, R3
MOVD R3, 40(RSP)
MOVD R3, 48(RSP) // Unnecessary, as the callee ignores it, but conservative.
// Spill local variables (registers) onto the stack; call; unspill.
MOVD R8, 88(RSP)
CALL ·emitLiteral(SB)
MOVD 88(RSP), R8
// Finish the "d +=" part of "d += emitLiteral(etc)".
MOVD 56(RSP), R1
ADD R1, R8, R8
encodeBlockEnd:
MOVD dst_base+0(FP), R3
SUB R3, R8, R8
MOVD R8, d+48(FP)
RET