z0 := (mPi4[digit] << bitshift) | (mPi4[digit+1] >> (64 - bitshift))
	z1 := (mPi4[digit+1] << bitshift) | (mPi4[digit+2] >> (64 - bitshift))
	z2 := (mPi4[digit+2] << bitshift) | (mPi4[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)

	// add 2¹²⁸ to each message block value
	VLEIB $4, $1, M_4
	VLEIB $12, $1, M_4

multiply:
	// accumulate the incoming message
	VAG H_0, M_0, M_0
	VAG H_3, M_3, M_3
	VAG H_1, M_1, M_1
	VAG H_4, M_4, M_4
	VAG H_2, M_2, M_2

	// multiply the accumulator by the key coefficient
	MULTIPLY(M_0, M_1, M_2, M_3, M_4, R_0, R_1, R_2, R_3, R_4, R5_1, R5_2, R5_3, R5_4, H_0, H_1, H_2, H_3, H_4)

	}

	for name, f := range map[string]interface{}{
		"Negate": func(v, x Scalar) bool {
			return checkAliasingOneArg((*Scalar).Negate, v, x)
		},
		"Multiply": func(v, x, y Scalar) bool {
			return checkAliasingTwoArgs((*Scalar).Multiply, v, x, y)
		},
		"Add": func(v, x, y Scalar) bool {
			return checkAliasingTwoArgs((*Scalar).Add, v, x, y)
		},
		"Subtract": func(v, x, y Scalar) bool {

class SwiftAppWithXCTest extends MainWithXCTestSourceElement implements AppElement {
    final SwiftApp main = new SwiftApp()
    final XCTestSourceElement test = new SwiftAppTest(main, main.greeter, main.sum, main.multiply)

    String expectedOutput = main.expectedOutput

    SwiftAppWithXCTest() {
        super('app')
    }

        @Override
        List<XCTestSourceFileElement> getTestSuites() {
            return [new XCTestSourceFileElement("CombinedTests") {
                final delegate = new SwiftAppTest(main, main.greeter, main.sum, main.multiply)

                @Override
                List<XCTestCaseElement> getTestCases() {
                    return delegate.sumTest.testCases + delegate.greeterTest.testCases + delegate.multiplyTest.testCases
                }

// CHECK:         %[[SUB:.*]] = s32[] subtract(s32[] %[[CONVERT_1]],
// CHECK:         %[[CONVERT_2:.*]] = f32[] convert(s32[] %[[SUB]])
// CHECK:         %[[MUL:.*]] = f32[] multiply(f32[] %[[CONVERT_2]],

					}
					atomic.AddInt64(&ops, int64(n))
					atomic.AddInt64(&lat, latency)
				})
				spent := time.Since(t)
				if spent > 0 && n > 0 {
					// Since we are benchmarking n parallel servers we need to multiply by n.
					// This will give an estimate of the total ops/s.
					latency := float64(atomic.LoadInt64(&lat)) / float64(time.Millisecond)
					b.ReportMetric(float64(n)*float64(ops)/spent.Seconds(), "vops/s")

        super(projectName)
        main = new SwiftLibWithCppDep(projectName, cppGreeter)
        test = new SwiftLibTest(main, main.greeter, main.sum, main.multiply)
    }

//   ⎡x / c⎤ = ⎣x * m / 2^(n+s)⎦ + 1
// (TODO: derivation?)
//
// The multiply is a bit odd, as it is a signed n-bit value
// times an unsigned n-bit value.  For n smaller than the
// word size, we can extend x and m appropriately and use the
// signed multiply instruction.  For n == word size,
// we must use the signed multiply high and correct
// the result by adding x*2^n.
//

	// This does not seem worthwhile, at least for Go: not having any high
	// bits in the multiplier reduces the effect of low bits on the highest bits,
	// and it only saves 1 multiply out of 3.
	// (On 32-bit systems, it saves 1 out of 6, since Mul64 is doing 4.)
	const (
		mulHi = 2549297995355413924
		mulLo = 4865540595714422341
		incHi = 6364136223846793005
		incLo = 1442695040888963407
	)