(Round(32|64)F ...) => (Copy ...)

// comparisons
(Eq8 x y)  => (SGTU (MOVVconst [1]) (XOR (ZeroExt8to64 x) (ZeroExt8to64 y)))
(Eq16 x y) => (SGTU (MOVVconst [1]) (XOR (ZeroExt16to64 x) (ZeroExt16to64 y)))
(Eq32 x y) => (SGTU (MOVVconst [1]) (XOR (ZeroExt32to64 x) (ZeroExt32to64 y)))
(Eq64 x y) => (SGTU (MOVVconst [1]) (XOR x y))
(EqPtr x y) => (SGTU (MOVVconst [1]) (XOR x y))
(Eq(32|64)F x y) => (FPFlagTrue (CMPEQ(F|D) x y))

	// amd64:"NEGL","ADCQ","SBBQ","NEGQ"
	// loong64: "ADDV", "SGTU"
	// ppc64x: "ADDC", "ADDE", "ADDZE"
	// s390x:"ADDE","ADDC\t[$]-1,"
	// mips64:"ADDV","SGTU"
	// riscv64: "ADD","SLTU"
	return bits.Add(x, 7, ci)
}

func AddZ(x, y uint) (r, co uint) {
	// arm64:"ADDS","ADC",-"ADCS",-"ADD\t",-"CMP"
	// amd64:"ADDQ","SBBQ","NEGQ",-"NEGL",-"ADCQ"
	// loong64: "ADDV", "SGTU"
	// ppc64x: "ADDC", -"ADDE", "ADDZE"

TEXT runtime·memmove<ABIInternal>(SB), NOSPLIT|NOFRAME, $0-24
	BNE	R6, check
	RET

check:
	SGTU	R4, R5, R7
	BNE	R7, backward

	ADDV	R4, R6, R9 // end pointer

	// if the two pointers are not of same alignments, do byte copying
	SUBVU	R5, R4, R7
	AND	$7, R7
	BNE	R7, out

	// if less than 8 bytes, do byte copying
	SGTU	$8, R6, R7
	BNE	R7, out

	// do one byte at a time until 8-aligned
	AND	$7, R4, R8

byte_loop:
	BEQ	R4, R12, samebytes
	MOVBU	(R4), R8
	ADDVU	$1, R4
	MOVBU	(R6), R9
	ADDVU	$1, R6
	BEQ	R8, R9, byte_loop

byte_cmp:
	SGTU	R8, R9, R4 // R12 = 1 if (R8 > R9)
	BNE	R0, R4, ret
	MOVV	$-1, R4
	JMP	ret

samebytes:
	SGTU	R5, R7, R8
	SGTU	R7, R5, R9
	SUBV	R9, R8, R4

ret:

	ADDV	R4, R5, R6

	// if less than 8 bytes, do one byte at a time
	SGTU	$8, R5, R8
	BNE	R8, out

	// do one byte at a time until 8-aligned
	AND	$7, R4, R8
	BEQ	R8, words
	MOVB	R0, (R4)
	ADDV	$1, R4
	JMP	-4(PC)

words:
	// do 8 bytes at a time if there is room
	ADDV	$-7, R6, R5

	PCALIGN	$16
	SGTU	R5, R4, R8
	BEQ	R8, out
	MOVV	R0, (R4)
	ADDV	$8, R4
	JMP	-4(PC)

out:

	// match: (IsInBounds idx len)
	// result: (SGTU len idx)
	for {
		idx := v_0
		len := v_1
		v.reset(OpLOONG64SGTU)
		v.AddArg2(len, idx)
		return true
	}
}
func rewriteValueLOONG64_OpIsNonNil(v *Value) bool {
	v_0 := v.Args[0]
	b := v.Block
	typ := &b.Func.Config.Types
	// match: (IsNonNil ptr)
	// result: (SGTU ptr (MOVVconst [0]))
	for {
		ptr := v_0

	.globl	toc_func
	.type	toc_func, @function
toc_func:
	addis	2,12,.TOC.-toc_func@ha
	addi	2,2,.TOC.-toc_func@l
	.localentry toc_func, .-toc_func
	mflr	0
	std	0,16(1)
	stdu	1,-32(1)

	// Call a NOTOC function which clobbers R2.
	bl	notoc_nor2_func
	nop

	// Call libc random. This should generate a TOC relative plt stub.
	bl	random
	nop

	addi	1,1,32
	ld 	0,16(1)

	"MULD",
	"MULF",
	"MULU",
	"MULH",
	"MULHU",
	"MULW",
	"NEGD",
	"NEGF",
	"NEGW",
	"NEGV",
	"NOOP",
	"NOR",
	"OR",
	"REM",
	"REMU",
	"RFE",
	"SC",
	"SCV",
	"SGT",
	"SGTU",
	"SLL",
	"SQRTD",
	"SQRTF",
	"SRA",
	"SRL",
	"ROTR",
	"SUB",
	"SUBD",
	"SUBF",
	"SUBU",
	"SUBW",
	"DBAR",
	"SYSCALL",
	"TEQ",
	"TNE",
	"WORD",
	"XOR",
	"MASKEQZ",

    def call(self, x):
      return x + 1

  model = AddOneModel()

  x_train = tf.constant([1, 2, 3, 4, 5], dtype=tf.float32)
  y_train = tf.constant([2, 3, 4, 5, 6], dtype=tf.float32)

  model.compile(optimizer='sgd', loss='mse')
  model.fit(x_train, y_train, epochs=1)

  path = tempdir + '/add_one_model'
  model.save(path)
  return path


class TensorflowToStableHLOTest(test.TestCase):

  <mime-type type="application/vnd.oma.bcast.provisioningtrigger"/>
  <mime-type type="application/vnd.oma.bcast.sgboot"/>
  <mime-type type="application/vnd.oma.bcast.sgdd+xml"/>
  <mime-type type="application/vnd.oma.bcast.sgdu"/>
  <mime-type type="application/vnd.oma.bcast.simple-symbol-container"/>
  <mime-type type="application/vnd.oma.bcast.smartcard-trigger+xml"/>