SCCP support

[dstogov]

Do you plan to implement Sparse Conditional Constants Propagation?
It took me a while to understand that the iterative optimization pass doesn't do it at all.
The example from "Combining Analyses, Combining Optimizations"?

        Parser parser = new Parser(
                """
int x = 1;
int i = 1;
while(i < 10) {
    if (x != 1) {
    	x = 2;
    }
    i = i + 1;
}
return x;
                """);
        StopNode stop = parser.parse().iterate();
        assertEquals("return 1", stop.toString());

[cliffclick]

Yes, its in the plans. Honestly should be really tiny amount of work at this point; the compiler is geared to support it.

[dstogov]

Did you see "A sparse algorithm for predicated global value numbering" by Karthik Gargi (paper and slides). The algorithm may perform combination of SCCP and GVN, and it's significantly simpler than your algorithm from "Combining Analyses, Combining Optimizations". I plan to try it in my project.

[cliffclick]

Yes, good luck with your implementation.
I will point out that some of the examples at the bottom are flawed, in that normal dominating test - as has been done by many compilers including C2 for many decades - cleans out the examples, such that they never reach a later analysis. C2 will clean such things out while parsing bytecodes and building SSA, i.e. before any other passes. This makes reading improvement numbers a little gray: no optimization runs in a vacuum, and easy/fast/cheap earlier phases may "eat the heart" out of any gains made in later optimizations.