%10 = mhlo.add %arg0, %5 : tensor<2xf32>
  %11 = mhlo.clamp %7, %10, %3 : tensor<2xf32>
  %12 = mhlo.clamp %9, %11, %1 : tensor<2xf32>
  %13 = mhlo.multiply %arg0, %12 : tensor<2xf32>
  %14 = mhlo.divide %13, %1 : tensor<2xf32>
  return %14 : tensor<2xf32>
}

// CHECK-LABEL:   func.func @hardswish(

// It also runs ssa/check and gccheck to be sure that those are checked at least a
// little in each run.bash.  It does not check or run the generated code.
// The test file is however a useful example of fused-vs-cascaded multiply-add.
func TestFmaHash(t *testing.T) {
	switch runtime.GOOS {
	case "linux", "darwin":
	default:
		t.Skipf("Slow test, usually avoid it, os=%s not linux or darwin", runtime.GOOS)
	}
	switch runtime.GOARCH {

  // CHECK: %[[CONSTANT2:.*]] = stablehlo.constant dense<1.000000e+03> : tensor<1024x3xf32>
  // CHECK: %[[ADD:.*]] = stablehlo.add %[[INPUT1]], %[[CONSTANT2]] : tensor<1024x3xf32>
  // CHECK: %[[MUL:.*]] = stablehlo.multiply %[[INPUT1]], %[[INPUT2]] : tensor<1024x3xf32>
  // CHECK: return %[[ADD]], %[[MUL]]
  // CHECK: }

  // CHECK: @serving_default

	err := syscall.Sysinfo(&info)
	if err != nil {
		errorList = append(errorList, errors.Wrapf(err, "failed to get system info"))
	}

	// Totalram holds the total usable memory. Unit holds the size of a memory unit in bytes. Multiply them and convert to MB
	actual := uint64(info.Totalram) * uint64(info.Unit) / 1024 / 1024
	if actual < mc.Mem {
		errorList = append(errorList, errors.Errorf("the system RAM (%d MB) is less than the minimum %d MB", actual, mc.Mem))

				sz = listSize.Multiply(elemSize)
			}

			if len(args) > 0 {
				sepSize := l.sizeEstimate(args[0])
				minSeparators := uint64(0)
				maxSeparators := uint64(0)
				if listSize.Min > 0 {
					minSeparators = listSize.Min - 1
				}
				if listSize.Max > 0 {
					maxSeparators = listSize.Max - 1
				}

// pages into objects of the given size wastes at most 12.5% (1.125x)
// of the memory. It is not necessary that the cutoff here be
// the same as above.
//
// The two sources of waste multiply, so the worst possible case
// for the above constraints would be that allocations of some
// size might have a 26.6% (1.266x) overhead.
// In practice, only one of the wastes comes into play for a

        try {
          BigInteger quotient = BigIntegerMath.divide(p, q, UNNECESSARY);
          BigInteger undone = quotient.multiply(q);
          if (!p.equals(undone)) {
            failFormat("expected %s.multiply(%s) = %s; got %s", quotient, q, p, undone);
          }
          assertTrue(dividesEvenly);
        } catch (ArithmeticException e) {
          assertFalse(dividesEvenly);

    module_str = self._extract_first_xla_call_module_op(
        self._output_saved_model_path
    )

    # Tests that the output graph contains multiply for symmetric
    # dequantization.
    self.assertTrue(re.search('stablehlo.multiply', module_str))
    # Tests that the output graph contains float dot_general.
    self.assertTrue(
        re.search('stablehlo.dot_general.*xf32>.*xf32>.*xf32>', module_str)

		runtime.HandleError(fmt.Errorf("feature %q is not implemented in DefaultFeatureSupportChecker", feature))
	}
}

func (f *defaultFeatureSupportChecker) checkClient(ctx context.Context, c client) {
	// start with 10 ms, multiply by 2 each step, until 15 s and stays on 15 seconds.
	delayFunc := wait.Backoff{
		Duration: 10 * time.Millisecond,
		Cap:      15 * time.Second,
		Factor:   2.0,
		Steps:    11}.DelayFunc()
	f.lock.Lock()

            matmul_op.getAdjY() ? shape_y.back() : *(shape_y.rbegin() + 1);
        const int y_col =
            !matmul_op.getAdjY() ? shape_y.back() : *(shape_y.rbegin() + 1);

        // Checks that matrix multiply can perform a valid contraction.
        if (x_col != y_row) {
          result_shape.clear();
          return false;
        }

        result_shape.push_back(x_row);
        result_shape.push_back(y_col);