ADDU	$(1024*8), R29

	// Ask signaller to setgid
	MOVW	$1, R1
	SYNC
	MOVW	R1, ·Baton(SB)
	SYNC

	// Wait for setgid completion
loop:
	SYNC
	MOVW	·Baton(SB), R1
	OR	R2, R2, R2	// hint that we're in a spin loop
	BNE	R1, loop
	SYNC

	// Restore stack
	ADDU	$(-1024*8), R29

	ADD	$(1024 * 8), R1

	// Ask signaller to setgid
	MOVW	$1, R3
	SYNC
	MOVW	R3, ·Baton(SB)

	// Wait for setgid completion
loop:
	SYNC
	MOVW	·Baton(SB), R3
	CMP	R3, $0
	// Hint that we're in a spin loop
	OR	R1, R1, R1
	BNE	loop
	ISYNC

	// Restore stack
	SUB	$(1024 * 8), R1

TEXT ·cryptBlocks(SB),NOSPLIT,$0-40
	MOVD	key+8(FP), R1
	MOVD	dst+16(FP), R2
	MOVD	src+24(FP), R4
	MOVD	length+32(FP), R5
	MOVD	c+0(FP), R0
loop:
	KM	R2, R4      // cipher message (KM)
	BVS	loop        // branch back if interrupted
	XOR	R0, R0
	RET

// func cryptBlocksChain(c code, iv, key, dst, src *byte, length int)
TEXT ·cryptBlocksChain(SB),NOSPLIT,$48-48
	LA	params-48(SP), R1

	i := 0
	if false {
		goto gogoloop
	}
	if false {
		goto gogoloop
	}
	if false {
		goto gogoloop
	}
	goto gogoloop

	// backward declared
loop:
	i = i + 1
	if i < 100 {
		goto loop
	}
	return

gogoloop:
	goto loop

	MOVOU	48(DI), X1
	PCMPEQB X0, X1
	PMOVMSKB X1, AX
	XORQ	$0xffff, AX
	JNE	diff64

	ADDQ	$64, SI
	ADDQ	$64, DI
	SUBQ	$64, R8
	CMPQ	R8, $64
	JBE	loop
	JMP	big_loop
#endif

	// Compare 64-bytes per loop iteration.
	// Loop is unrolled and uses AVX2.
big_loop_avx2:
	VMOVDQU	(SI), Y2
	VMOVDQU	(DI), Y3
	VMOVDQU	32(SI), Y4
	VMOVDQU	32(DI), Y5
	VPCMPEQB Y2, Y3, Y0
	VPMOVMSKB Y0, AX

TEXT runtime·memequal(SB),NOSPLIT,$0-13
	MOVW	a+0(FP), R1
	MOVW	b+4(FP), R2
	BEQ	R1, R2, eq
	MOVW	size+8(FP), R3
	ADDU	R1, R3, R4
loop:
	BNE	R1, R4, test
	MOVW	$1, R1
	MOVB	R1, ret+12(FP)
	RET
test:
	MOVBU	(R1), R6
	ADDU	$1, R1
	MOVBU	(R2), R7
	ADDU	$1, R2
	BEQ	R6, R7, loop

	MOVB	R0, ret+12(FP)
	RET
eq:
	MOVW	$1, R1
	MOVB	R1, ret+12(FP)
	RET

// memequal_varlen(a, b unsafe.Pointer) bool

	CMP   R5, $0
	MOVD  R5, CTR         // Set up loop counter
	BEQ   tail            // If R5 = 0, we can't use the loop

	// Process 4 elements per iteration. Unrolling this loop
	// means a performance trade-off: we will lose performance
	// for small values of z_len (0.90x in the worst case), but
	// gain significant performance as z_len increases (up to
	// 1.45x).

	PCALIGN $16
loop:
	MOVD  8(R8), R11      // R11 = x[i]

	}

	golden := make(map[string]image.Image)
loop:
	for _, it := range imageTests {
		g := golden[it.goldenFilename]
		if g == nil {
			var err error
			g, _, err = decode(it.goldenFilename)
			if err != nil {
				t.Errorf("%s: %v", it.goldenFilename, err)
				continue loop
			}
			golden[it.goldenFilename] = g
		}
		m, imageFormat, err := decode(it.filename)

	CMPPD	X0, X1, 4
	CMPPD	foo+4(SB), X1, 4

// LTYPEX spec10	{ outcode(int($1), &$2); }
	PINSRD	$1, (AX), X0
	PINSRD	$2, foo+4(FP), X0

// Was bug: LOOP is a branch instruction.
	JCS	2(PC)
loop:
	LOOP	loop // LOOP

// Tests for TLS reference.
	MOVL    (TLS), AX
	MOVL    8(TLS), DX

// LTYPE0 nonnon	{ outcode(int($1), &$2); }
	RET

	ADDV	$(1024*8), R29

	// Ask signaller to setgid
	MOVW	$1, R1
	SYNC
	MOVW	R1, ·Baton(SB)
	SYNC

	// Wait for setgid completion
loop:
	SYNC
	MOVW	·Baton(SB), R1
	OR	R2, R2, R2	// hint that we're in a spin loop
	BNE	R1, loop
	SYNC

	// Restore stack
	ADDV	$(-1024*8), R29