bytes_between_0_and_15:
	TESTQ R15, R15
	JZ    done
	MOVQ  $1, BX
	XORQ  CX, CX
	XORQ  R13, R13
	ADDQ  R15, SI

flush_buffer:
	SHLQ $8, BX, CX
	SHLQ $8, BX
	MOVB -1(SI), R13
	XORQ R13, BX
	DECQ SI
	DECQ R15
	JNZ  flush_buffer

	ADDQ BX, R8
	ADCQ CX, R9
	ADCQ $0, R10
	MOVQ $16, R15
	JMP  multiply

done:
	MOVQ R8, 0(DI)
	MOVQ R9, 8(DI)
	MOVQ R10, 16(DI)

	// Process the first few bytes to 8-byte align the input.

	// BX = 8 - BX. We need to process this many bytes to align.
	SUBQ $1, BX
	XORQ $7, BX

	BTQ $0, BX
	JNC align_2

	CRC32B (SI), AX
	DECQ CX
	INCQ SI

align_2:
	BTQ $1, BX
	JNC align_4

	CRC32W (SI), AX

	SUBQ $2, CX
	ADDQ $2, SI

align_4:
	BTQ $2, BX
	JNC aligned

	CRC32L (SI), AX

	SUBQ $4, CX

	// Hash last < 16 byte tail
	XORQ t0, t0
	XORQ t1, t1
	XORQ t2, t2
	ADDQ itr2, adp

hashADTailLoop:
	SHLQ $8, t0, t1
	SHLQ $8, t0
	MOVB -1(adp), t2
	XORQ t2, t0
	DECQ adp
	DECQ itr2
	JNE  hashADTailLoop

hashADTailFinish:
	ADDQ t0, acc0; ADCQ t1, acc1; ADCQ $1, acc2
	polyMul

	// Finished AD
hashADDone:
	RET

	AESKEYGENASSIST $0x36, X0, X1
	CALL _expand_key_128<>(SB)
Lexp_dec:
	// dec
	SUBQ $16, BX
	MOVUPS (BX), X1
	MOVUPS X1, (DX)
	DECQ CX
Lexp_dec_loop:
	MOVUPS -16(BX), X1
	AESIMC X1, X0
	MOVUPS X0, 16(DX)
	SUBQ $16, BX
	ADDQ $16, DX
	DECQ CX
	JNZ Lexp_dec_loop
	MOVUPS -16(BX), X0
	MOVUPS X0, 16(DX)
	RET

TEXT _expand_key_128<>(SB),NOSPLIT,$0
	PSHUFD $0xff, X1, X1

TEXT ·gcmAesInit(SB),NOSPLIT,$0
#define dst DI
#define KS SI
#define NR DX

	MOVQ productTable+0(FP), dst
	MOVQ ks_base+8(FP), KS
	MOVQ ks_len+16(FP), NR

	SHRQ $2, NR
	DECQ NR

	MOVOU bswapMask<>(SB), BSWAP
	MOVOU gcmPoly<>(SB), POLY

	// Encrypt block 0, with the AES key to generate the hash key H
	MOVOU (16*0)(KS), B0
	MOVOU (16*1)(KS), T0
	AESENC T0, B0

	CMOVQCS acc5, acc1
	CMOVQCS y_ptr, acc2
	CMOVQCS t1, acc3

	MOVQ acc0, (8*0)(res_ptr)
	MOVQ acc1, (8*1)(res_ptr)
	MOVQ acc2, (8*2)(res_ptr)
	MOVQ acc3, (8*3)(res_ptr)
	MOVQ res_ptr, x_ptr
	DECQ BX
	JNE  sqrLoop

	RET
/* ---------------------------------------*/
// func p256Mul(res, in1, in2 *p256Element)
TEXT ·p256Mul(SB),NOSPLIT,$0
	MOVQ res+0(FP), res_ptr
	MOVQ in1+8(FP), x_ptr

	"CVTSS2SL",
	"CVTSS2SQ",
	"CVTTPD2PL",
	"CVTTPS2PL",
	"CVTTSD2SL",
	"CVTTSD2SQ",
	"CVTTSS2SL",
	"CVTTSS2SQ",
	"CWD",
	"CWDE",
	"DAA",
	"DAS",
	"DECB",
	"DECL",
	"DECQ",
	"DECW",
	"DIVB",
	"DIVL",
	"DIVPD",
	"DIVPS",
	"DIVQ",
	"DIVSD",
	"DIVSS",
	"DIVW",
	"DPPD",
	"DPPS",
	"EMMS",
	"ENTER",
	"EXTRACTPS",
	"F2XM1",
	"FABS",

	atomic.AddInt64(&v.nowCount, 1)
	r.qCache.bucketStats[bucket] = v
	atomic.AddInt64(&r.qCache.srQueueStats.nowBytes, sz)
	atomic.AddInt64(&r.qCache.srQueueStats.nowCount, 1)
}

func (r *ReplicationStats) decQ(bucket string, sz int64, isDelMarker bool, opType replication.Type) {
	r.qCache.Lock()
	defer r.qCache.Unlock()
	v, ok := r.qCache.bucketStats[bucket]
	if !ok {
		v = newInQueueStats(r.registry, bucket)
	}

	ef -= arr[b]
	// 386:`SUBL\s[A-Z]+,\s\([A-Z]+\)\([A-Z]+\*4\)`
	arr[c] -= b
	// 386:`ADDL\s[$]-15,\s\([A-Z]+\)\([A-Z]+\*4\)`
	arr[d] -= 15
	// 386:`DECL\s\([A-Z]+\)\([A-Z]+\*4\)`
	arr[b]--
	// amd64:`DECQ\s64\([A-Z]+\)`
	arr[8]--
	// 386:"SUBL\t4"
	// amd64:"SUBQ\t8"
	return arr[0] - arr[1]
}

func SubFromConst(a int) int {
	// ppc64x: `SUBC\tR[0-9]+,\s[$]40,\sR`
	b := 40 - a
	return b
}

				atomic.AddInt32(&p.activeMRFWorkers, 1)
				replicateObject(p.ctx, v, p.objLayer)
				atomic.AddInt32(&p.activeMRFWorkers, -1)
				globalReplicationStats.decQ(v.Bucket, v.Size, v.DeleteMarker, v.OpType)

			default:
				bugLogIf(p.ctx, fmt.Errorf("unknown mrf replication type: %T", oi), "unknown-mrf-replicate-type")
			}
		case <-p.mrfWorkerKillCh:
			return