tensor<20x28x!quant.uniform<i8:f32, 0.006:-34>>, tensor<20x28x!quant.uniform<i8:f32, 0.006:-34>>, tensor<20x28x!quant.uniform<i8:f32, 0.006:-34>>, tensor<20x28x!quant.uniform<i8:f32, 0.006:-34>>,
         tensor<20x20x!quant.uniform<i8:f32, 0.006:-34>>, tensor<20x20x!quant.uniform<i8:f32, 0.006:-34>>, tensor<20x20x!quant.uniform<i8:f32, 0.006:-34>>, tensor<20x20x!quant.uniform<i8:f32, 0.006:-34>>,
         none, none, none,

            """.trimIndent()
        )

        val result = runtimeInstanceFromResult(schema, resolution, kotlinFunctionAsConfigureLambda, RuntimeCustomAccessors.none, ::TopLevel)

        assertEquals(listOf(0, 0, 2, 3, 4), result.added.map { it.x })
    }


    private
    val schema = schemaFromTypes(TopLevel::class, this::class.nestedClasses)


    class TopLevel {
        val configuredLazy = lazy { Inner() }

		{
			name: "LocalModeClusterCIDR, dual-stack IPv6-primary cluster",
			config: &proxyconfigapi.KubeProxyConfiguration{
				DetectLocalMode: proxyconfigapi.LocalModeClusterCIDR,
				ClusterCIDR:     "2002:0:0:1234::/64,10.0.0.0/14",
			},
			primaryIPFamily: v1.IPv6Protocol,
			expected: map[v1.IPFamily]proxyutil.LocalTrafficDetector{
				v1.IPv4Protocol: proxyutil.NewDetectLocalByCIDR("10.0.0.0/14"),

  cniConfDir: "/var/run/multus/cni/net.d"
`,
		},

		{
			desc: "BadIPRange",
			yamlStr: `
global:
  proxy:
    includeIPRanges: "1.1.0.256/16,2.2.0.257/16"
    excludeIPRanges: "3.3.0.0/33,4.4.0.0/34"
`,
			wantErrs: makeErrors([]string{
				`global.proxy.excludeIPRanges netip.ParsePrefix("3.3.0.0/33"): prefix length out of range`,

// The error from the true ticks-per-second in a linux/amd64 VM seems to be:
// -   1 ms -> ~0.02% error
// -   5 ms -> ~0.004% error
// -  10 ms -> ~0.002% error
// -  50 ms -> ~0.0003% error
// - 100 ms -> ~0.0001% error
//
// We're willing to take 0.004% error here, because ticksPerSecond is intended to be used for
// converting durations, not timestamps. Durations are usually going to be much larger, and so

  %1 = "tfl.dequantize"(%0) : (tensor<1x32x1x3x!quant.uniform<i8:f32, 1.0>>) -> tensor<1x32x1x3xf32>
  %cst = arith.constant dense<1> : tensor<4xi32>
  %cst_0 = arith.constant dense<[0, 0, 0, 3]> : tensor<4xi32>
  %cst_1 = arith.constant dense<[0, -2, 0, 0]> : tensor<4xi32>
  %2 = "tfl.var_handle"() {container = "", shared_name = "read_assign2/states"} : () -> tensor<!tf_type.resource>

// CHECK-DAG:           %[[VAL_3:.*]] = "tfl.pseudo_const"{{.*}}dense<[-1, -1, -1, 2]> : tensor<4xi32>
// CHECK-DAG:           %[[VAL_4:.*]] = "tfl.pseudo_const"{{.*}}dense<[0, 0, 0, 4]> : tensor<4xi32>
// CHECK-DAG:           %[[VAL_5:.*]] = "tfl.pseudo_const"{{.*}}dense<[-1, -1, -1, 81]> : tensor<4xi32>

00000300  97 b1 e9 31 37 ea a7 5d  a9 00 ce 85 08 5a b3 ac  |...17..].....Z..|
00000310  ef a6 8d c3 9f a5 ba 97  e8 bc 9f 81 63 77 7b 2e  |............cw{.|
00000320  92 5f 36 a1 00 04 c1 3f  e9 76 bb 58 da aa 03 ba  |._6....?.v.X....|
00000330  cd 11 9b 9a 56 2e 64 53  74 b4 62 00 89 4a 57 0f  |....V.dSt.b..JW.|
00000340  a2 a9 f2 16 17 ab a3 72  3d 16 4d 41 4a df 54 b1  |.......r=.MAJ.T.|

	case AFCVTZSS:
		return FPCVTI(1, 0, 0, 3, 0)

	case AFCVTZSSW:
		return FPCVTI(0, 0, 0, 3, 0)

	case AFCVTZUD:
		return FPCVTI(1, 0, 1, 3, 1)

	case AFCVTZUDW:
		return FPCVTI(0, 0, 1, 3, 1)

	case AFCVTZUS:
		return FPCVTI(1, 0, 0, 3, 1)

	case AFCVTZUSW:
		return FPCVTI(0, 0, 0, 3, 1)

	case ASCVTFD:
		return FPCVTI(1, 0, 1, 0, 2)

	case ASCVTFS:

  func.return %0#0, %0#3, %0#4 : tensor<8x8x8x8xf32>, tensor<0xf32>, tensor<*xf32>
}

// -----

// CHECK-LABEL: fusedBatchNormGradV3_noTraining_mixed_precision