lhs_contracting_dimensions = [4, 1, 6],
      rhs_contracting_dimensions = [6, 0, 4]
  >} : (tensor<1x9x2x3x8x4x10xf32>, tensor<9x2x1x5x10x5x8x7xf32>) -> tensor<1x2x3x4x5x5x7xf32>
  %1 = "mhlo.dot_general"(%arg1, %arg2) {
    dot_dimension_numbers = #mhlo.dot<
      lhs_batching_dimensions = [0, 2],
      rhs_batching_dimensions = [2, 1],
      lhs_contracting_dimensions = [4, 1, 6],
      rhs_contracting_dimensions = [6, 0, 4]

	LDP	3*16(a_ptr), (R4, R5)
	LDP	4*16(a_ptr), (R6, R7)
	LDP	5*16(a_ptr), (R8, R9)
	LDP	3*16(b_ptr), (R16, R17)
	LDP	4*16(b_ptr), (R19, R20)
	LDP	5*16(b_ptr), (R21, R22)
	CSEL	EQ, R16, R4, R4
	CSEL	EQ, R17, R5, R5
	CSEL	EQ, R19, R6, R6
	CSEL	EQ, R20, R7, R7
	CSEL	EQ, R21, R8, R8
	CSEL	EQ, R22, R9, R9
	STP	(R4, R5), 3*16(res_ptr)
	STP	(R6, R7), 4*16(res_ptr)
	STP	(R8, R9), 5*16(res_ptr)

	RET

	SYS_TIMERFD_SETTIME64            = 411
	SYS_UTIMENSAT_TIME64             = 412
	SYS_PSELECT6_TIME64              = 413
	SYS_PPOLL_TIME64                 = 414
	SYS_IO_PGETEVENTS_TIME64         = 416
	SYS_RECVMMSG_TIME64              = 417
	SYS_MQ_TIMEDSEND_TIME64          = 418
	SYS_MQ_TIMEDRECEIVE_TIME64       = 419
	SYS_SEMTIMEDOP_TIME64            = 420
	SYS_RT_SIGTIMEDWAIT_TIME64       = 421

	SYS_TIMERFD_SETTIME64            = 411
	SYS_UTIMENSAT_TIME64             = 412
	SYS_PSELECT6_TIME64              = 413
	SYS_PPOLL_TIME64                 = 414
	SYS_IO_PGETEVENTS_TIME64         = 416
	SYS_RECVMMSG_TIME64              = 417
	SYS_MQ_TIMEDSEND_TIME64          = 418
	SYS_MQ_TIMEDRECEIVE_TIME64       = 419
	SYS_SEMTIMEDOP_TIME64            = 420
	SYS_RT_SIGTIMEDWAIT_TIME64       = 421

	SYS_TIMERFD_SETTIME64            = 411
	SYS_UTIMENSAT_TIME64             = 412
	SYS_PSELECT6_TIME64              = 413
	SYS_PPOLL_TIME64                 = 414
	SYS_IO_PGETEVENTS_TIME64         = 416
	SYS_RECVMMSG_TIME64              = 417
	SYS_MQ_TIMEDSEND_TIME64          = 418
	SYS_MQ_TIMEDRECEIVE_TIME64       = 419
	SYS_SEMTIMEDOP_TIME64            = 420
	SYS_RT_SIGTIMEDWAIT_TIME64       = 421

		return &inst{0x53, 0x0, 0x0, 96, 0x3}
	case AFADDS:
		return &inst{0x53, 0x0, 0x0, 0, 0x0}
	case AFCLASSD:
		return &inst{0x53, 0x1, 0x0, -480, 0x71}
	case AFCLASSQ:
		return &inst{0x53, 0x1, 0x0, -416, 0x73}
	case AFCLASSS:
		return &inst{0x53, 0x1, 0x0, -512, 0x70}
	case AFCVTDL:
		return &inst{0x53, 0x0, 0x2, -734, 0x69}
	case AFCVTDLU:
		return &inst{0x53, 0x0, 0x3, -733, 0x69}
	case AFCVTDQ:

                          ShapeUtil::MakeScalarShape(xla::F32), "y");
    Add(arg0, arg1);
    TF_ASSERT_OK_AND_ASSIGN(to_apply, sub_builder->Build());
  }
  auto x = Parameter(&builder, 0, ShapeUtil::MakeShape(xla::F32, {4, 16}), "x");
  ReplicaGroup group;
  group.add_replica_ids(0);
  group.add_replica_ids(1);
  XlaOp reduce_scatter =
      ReduceScatter(x, to_apply, /*scatter_dimension=*/1, /*shard_count=*/2,

	github.com/subosito/gotenv v1.4.1 // indirect
	github.com/t-yuki/gocover-cobertura v0.0.0-20180217150009-aaee18c8195c // indirect
	github.com/tdakkota/asciicheck v0.2.0 // indirect
	github.com/tetafro/godot v1.4.16 // indirect
	github.com/timakin/bodyclose v0.0.0-20230421092635-574207250966 // indirect
	github.com/timonwong/loggercheck v0.9.4 // indirect
	github.com/tomarrell/wrapcheck/v2 v2.8.1 // indirect

	increment(0)

	CMPQ ptxLen, $128
	JB gcmAesDecSingles

	MOVOU T0, (1*16)(SP)
	increment(1)
	MOVOU T0, (2*16)(SP)
	increment(2)
	MOVOU T0, (3*16)(SP)
	increment(3)
	MOVOU T0, (4*16)(SP)
	increment(4)
	MOVOU T0, (5*16)(SP)
	increment(5)
	MOVOU T0, (6*16)(SP)
	increment(6)
	MOVOU T0, (7*16)(SP)
	increment(7)

gcmAesDecOctetsLoop:

		CMPQ ptxLen, $128

	serverHelloEmptySCT = append(serverHelloEmptySCT, serverHelloBytes[i+4:]...)

	// Update the handshake message length.
	serverHelloEmptySCT[1] = byte((len(serverHelloEmptySCT) - 4) >> 16)
	serverHelloEmptySCT[2] = byte((len(serverHelloEmptySCT) - 4) >> 8)
	serverHelloEmptySCT[3] = byte(len(serverHelloEmptySCT) - 4)

	// Update the extensions length