ADDITIONAL SYSTEM INFORMATION :
- Operating system: Linux
- Compiler/toolchain: GCC 9.5
- Source commit: 24fe61ea13610d2f5d22f1b6696b871eb0505379 (tag: 17.0.19-internal+0, branch: master)
- Build type: release
- Build environment: Linux, compiled from source with GCC 9.5
- Also failed in 17.0.17+10, release

A DESCRIPTION OF THE PROBLEM :
The following small Java program returns different results depending on the VM execution mode. When run with -Xcomp (C2 compiled only), the method Func returns 10L. When run with -Xmixed or -Xint, it returns 20L. The logic hinges on comparisons of Float.floatToRawIntBits of expressions that include:
- NaN propagation through Math.max/Math.min
- Use of -0.0f
- Division and modulo
- Reassociation-sensitive floating-point operations


STEPS TO FOLLOW TO REPRODUCE THE PROBLEM :
1. Save the following code as Test.java.
2. Run the program with different VM modes:
   - java -Xint Test.java
   - java -Xmixed Test.java
   - java -Xcomp Test.java

---------- BEGIN SOURCE ----------

public class Test {

    public static long Func(int param) {
        float varFloat1 = param + 0.5f;
        float varFloat2 = -0.0f;
        if (
            !(
                    Float.floatToRawIntBits(
                            100.1f / Math.max(varFloat1, 1) / Math.max(varFloat1, Float.NaN)
                        +
                            Math.min(Float.NaN, varFloat1)
                    )
                ==
                    Float.floatToRawIntBits(
                            100.2f - varFloat2 % varFloat2 / Math.max(varFloat1, 1) * 0.1f
                        +
                            Math.max(varFloat1, Float.NaN) + Math.min(4.2f, varFloat1)
                    )
            )
        ) {
            return 10L;
        }
        return 20L;
    }

    public static void main(String[] strArr) {
        System.out.println("FuncReturn: " + Func(10));
    }
}

---------- END SOURCE ----------

FREQUENCY :
ALWAYS