RFR: 8277850: C2: optimize mask checks in counted loops

[Tobias Hartmann]

On Fri, 3 Dec 2021 10:04:58 GMT, Roland Westrelin <roland at openjdk.org> wrote:

> This is another fix that addresses a performance issue with panama and that was brought up by Maurizio. The pattern to optimize is:
> if ((base + (offset << 2)) & 3) != 0) {
> }
> 
> where base is loop independent but offset depends on a loop variable. This can be transformed to:
> 
> if ((base & 3) != 0) {
> 
> That check becomes loop independent and be optimized by loop predication (or I suppose loop unswitching but that wasn't the case of the micro benchmark I worked on).
> 
> This change also optimizes the pattern:
> 
> (offset << 2) & 3
> 
> to return 0.

Looks good to me, I submitted testing.

src/hotspot/share/opto/mulnode.cpp line 702:

> 700: 
> 701:   Node* in1 = in(1);
> 702:   int op = in1->Opcode();

I got curious and checked the code, we have similar patterns everywhere. Filed [JDK-8278328](https://bugs.openjdk.java.net/browse/JDK-8278328) to clean up this mess.

src/hotspot/share/opto/mulnode.cpp line 1753:

> 1751: }
> 1752: 
> 1753: Node* MulNode::AndIL_add_shift_and_mask(PhaseGVN* phase, BasicType bt) {

Please add a comment describing the pattern.

test/hotspot/jtreg/compiler/c2/irTests/TestShiftAndMask.java line 30:

> 28: /*
> 29:  * @test
> 30:  * @bug JDK-8277850

Should be `@bug 8277850` (not sure if that is a requirement but all other tests use that format).

test/hotspot/jtreg/compiler/c2/irTests/TestShiftAndMask.java line 36:

> 34:  */
> 35: 
> 36: public class TestShiftAndMask {

Nice tests!

-------------

PR: https://git.openjdk.java.net/jdk/pull/6697