// process particular chunks of the input data stream in parallel.
//
// Note that the constant definitions below are extended in order to compute
// intermediate results with a single VECTOR GALOIS FIELD MULTIPLY instruction.
// The rightmost doubleword can be 0 to prevent contribution to the result or
// can be multiplied by 1 to perform an XOR without the need for a separate
// VECTOR EXCLUSIVE OR instruction.
//

  // CHECK-NEXT: %[[mul2:.*]] = mhlo.multiply %arg2, %[[scr1]] : tensor<8xf32>
  // CHECK:      %[[bcast_mul2:.+]] = "mhlo.dynamic_broadcast_in_dim"(%[[mul2]], {{.*}}) <{broadcast_dimensions = dense<3> : tensor<1xi64>}> : (tensor<8xf32>, tensor<4xindex>) -> tensor<8x8x8x8xf32>
  // CHECK-NEXT: %[[mul3:.*]] = mhlo.multiply %[[grad]], %[[bcast_mul2]] : tensor<8x8x8x8xf32>

		// each.
		//
		// If they were not mutually exclusive (for example, two Add
		// calls), then we could not definitively select the correct
		// callee.
		val += m.Multiply(42, selectA(i).Add(one(i), 2))
	}
	return val
}

type AddFunc func(int, int) int

func AddFn(a, b int) int {
	for i := 0; i < 1000; i++ {
		sink++
	}
	return a + b
}

  // CHECK: return %[[CONV]] : tensor<1x28x23x2xf32>
}

// CHECK-LABEL: @notfuseMulIntoConv2d
// filter and multiply are not broadcastable
func.func @notfuseMulIntoConv2d(%arg0: tensor<1x112x112x3xf32>) -> tensor<1x28x23x2xf32> {

	MOVW	$sysnsec_lo-8(SP), R0	// destination addr
	MOVW	R0,res-12(SP)
	MOVW	$SYS_NSEC, R0
	SWI	$0
	MOVW	sysnsec_lo-8(SP), R1	// R1:R2 = nsec
	MOVW	sysnsec_hi-4(SP), R2

	// multiply nanoseconds by reciprocal of 10**9 (scaled by 2**61)
	// to get seconds (96 bit scaled result)
	MOVW	$0x89705f41, R3		// 2**61 * 10**-9
	MULLU	R1,R3,(R6,R5)		// R5:R6:R7 = R1:R2 * R3
	MOVW	$0,R7
	MULALU	R2,R3,(R7,R6)

  // will overcount the divisors.
  //    Following from (LEMMA 1) the only windows which contain overcounted
  // divisors are the ones on the outside right and bottom edge. We can iterate
  // over these windows and multiply the corresponding out element by
  // `kernel_size / X` where `X` is the number of elements in the padded input
  // tensor not in the newly padded zone. This corrects the overcounting of

        sharedLibrary("build/lib/main/release/${lib.moduleName}").assertExists()
        sharedLibrary("build/lib/main/release/${lib.moduleName}").assertHasStrippedDebugSymbolsFor(["greeter.o", "sum.o", "multiply.o"])
    }

    def "can use link file as task dependency"() {
        given:
        def lib = new SwiftLib()
        settingsFile << "rootProject.name = '${lib.projectName}'"

        """.stripIndent()

        when:
        succeeds 'components'

        then:
        output.contains """
            $componentType 'myComponent'
            ${'-'.multiply("$componentType 'myComponent'".length())}

            Source sets
                $languageName source 'myComponent:myComponentSource'

        4      | Fruit.apples  | 1.2    | Fruit.oranges    | 1.111111
        125    | Fruit.oranges | 23.4   | Fruit.grapefruit | 3.205128
    }

    def "can multiply amount by a unitless value"() {
        expect:
        Amount.valueOf(a, Fruit.apples) * b == Amount.valueOf(c, Fruit.apples)

        where:
        a     | b   | c
        0     | 200 | 0

	z0 := (mPi[digit] << bitshift) | (mPi[digit+1] >> (64 - bitshift))
	z1 := (mPi[digit+1] << bitshift) | (mPi[digit+2] >> (64 - bitshift))
	z2 := (mPi[digit+2] << bitshift) | (mPi[digit+3] >> (64 - bitshift))
	// Multiply mantissa by the digits and extract the upper two digits (hi, lo).
	z2hi, _ := bits.Mul64(z2, ix)
	z1hi, z1lo := bits.Mul64(z1, ix)
	z0lo := z0 * ix
	lo, c := bits.Add64(z1lo, z2hi, 0)
	hi, _ := bits.Add64(z0lo, z1hi, c)