// but not the two instructions that led to them being combined here.  For example
				// (CMPBconstload c (ADDQ x y)) -> (CMPBconstloadidx1 c x y) -> (CMPB c (MOVBloadidx1 x y))
				// The final pair of instructions turns out to be notably faster, at least in some benchmarks.
				f.Config.splitLoad(v)
			}
		}
	}
}

	// CMHI <V><d>, <V><n>, <V><m>
	{0xff20fc00, 0x7e203400, CMHI, instArgs{arg_Vd_22_2__D_3, arg_Vn_22_2__D_3, arg_Vm_22_2__D_3}, nil},
	// CMHI <Vd>.<t>, <Vn>.<t>, <Vm>.<t>

	// Clear the upper bits to get to a clean state. See issue #37174.
	// It is safe here as Go code don't use the upper bits of Y registers.
	p("#ifdef GOOS_darwin")
	p("#ifndef hasAVX")
	p("CMPB internal∕cpu·X86+const_offsetX86HasAVX(SB), $0")
	p("JE 2(PC)")
	p("#endif")
	p("VZEROUPPER")
	p("#endif")

	lSSE.save()
	p("CALL ·asyncPreempt2(SB)")
	lSSE.restore()
	l.restore()

	ORR.P	R2, R3, R4         // ERROR "invalid .P suffix"
	CMP.S	R1, R2	           // ERROR "invalid .S suffix"
	BIC.P	$124, R1, R2       // ERROR "invalid .P suffix"
	MOVW.S	$124, R1           // ERROR "invalid .S suffix"
	MVN.S	$123, g            // ERROR "invalid .S suffix"
	RSB.U	$0, R9             // ERROR "invalid .U suffix"
	CMP.S	$29, g             // ERROR "invalid .S suffix"

	MOVQ	_cgo_callers(SB), AX
	JMP	AX

sigtramp:
	JMP	runtime·sigtramp(SB)

sigtrampnog:
	// Signal arrived on a non-Go thread. If this is SIGPROF, get a
	// stack trace.
	CMPL	DI, $27 // 27 == SIGPROF
	JNZ	sigtramp

	// Lock sigprofCallersUse.
	MOVL	$0, AX
	MOVL	$1, CX
	MOVQ	$runtime·sigprofCallersUse(SB), R11
	LOCK
	CMPXCHGL	CX, 0(R11)

	CMPQ (BX), $7                           // 48833b07
	CMPQ (R11), $7                          // 49833b07
	CMPQ DX, $7                             // 4883fa07
	CMPQ R11, $7                            // 4983fb07
	CMPQ (BX), DX                           // 483913
	CMPQ (BX), R11                          // 4c391b
	CMPQ (R11), DX                          // 493913
	CMPQ (R11), R11                         // 4d391b

	SHA256RNDS2		msg, state1, state0		\
	sha256Msg1		(m,a)

TEXT ·block(SB), 0, $536-32
	CMPB	·useSHA(SB), $1
	JE	sha_ni
	CMPB	·useAVX2(SB), $1
	JE	avx2

	MOVQ p_base+8(FP), SI
	MOVQ p_len+16(FP), DX
	SHRQ $6, DX
	SHLQ $6, DX

	LEAQ (SI)(DX*1), DI
	MOVQ DI, 256(SP)
	CMPQ SI, DI
	JEQ  end

	MOVQ dig+0(FP), BP
	MOVL (0*4)(BP), R8  // a = H0
	MOVL (1*4)(BP), R9  // b = H1

// Fold boolean tests into blocks
(NE (TESTB (SETL  cmp) (SETL  cmp)) yes no) => (LT  cmp yes no)
(NE (TESTB (SETLE cmp) (SETLE cmp)) yes no) => (LE  cmp yes no)
(NE (TESTB (SETG  cmp) (SETG  cmp)) yes no) => (GT  cmp yes no)
(NE (TESTB (SETGE cmp) (SETGE cmp)) yes no) => (GE  cmp yes no)
(NE (TESTB (SETEQ cmp) (SETEQ cmp)) yes no) => (EQ  cmp yes no)
(NE (TESTB (SETNE cmp) (SETNE cmp)) yes no) => (NE  cmp yes no)

			// before its corresponding VarDef.
			// See issue 28445.
			//   v1 = LOAD ...
			//   v2 = VARDEF
			//   v3 = LEAQ
			//   v4 = CMPQ v1 v3
			// We don't want to combine the CMPQ with the load, because
			// that would force the CMPQ to schedule before the VARDEF, which
			// in turn requires the LEAQ to schedule before the VARDEF.
			return false
		}
		if v.Type.IsMemory() {

			kind = ssa.BoundsSlice3CU
		}
	}

	var cmp *ssa.Value
	if kind == ssa.BoundsIndex || kind == ssa.BoundsIndexU {
		cmp = s.newValue2(ssa.OpIsInBounds, types.Types[types.TBOOL], idx, len)
	} else {
		cmp = s.newValue2(ssa.OpIsSliceInBounds, types.Types[types.TBOOL], idx, len)
	}
	b := s.endBlock()
	b.Kind = ssa.BlockIf
	b.SetControl(cmp)
	b.Likely = ssa.BranchLikely
	b.AddEdgeTo(bNext)