[JIT] X64/ARM64 - Fold 'x & 255' and 'x & 65535' to a cast #79630
Conversation
dotnet-issue-labeler
bot
added
the
area-CodeGen-coreclr
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch Issue DetailsDescription Fixes #8833 Acceptance Criteria
|
TIHan
changed the title
|
@dotnet/jit-contrib @EgorBo This is ready |
|
@TIHan the diffs seem to be expired - do you have the most recent ones maybe? |
|
@EgorBo I will re-run CI to get new diffs. |
| assert(varTypeIsUnsigned(castToType)); | ||
|
|
||
| GenTreeCast* cast = gtNewCastNode(TYP_INT, op1, false, castToType); | ||
| if (fgGlobalMorph) |
There was a problem hiding this comment.
can you educate me on why fgMorphTreeDone is needed here (I don't see it in other morph cases) and does it decrease jit-diffs if you just do this opt in GlobalMorph case only?
There was a problem hiding this comment.
I'm following the precedent that was set. Right above NewZeroExtendNode, you see:
// Helper function that creates a new IntCon node and morphs it, if required
auto NewMorphedIntConNode = [&](int value) -> GenTreeIntCon* {
GenTreeIntCon* icon = gtNewIconNode(value);
if (fgGlobalMorph)
{
fgMorphTreeDone(icon);
}
return icon;
};Also, gtFoldExprSpecial has a comment that says:
// Return value:
// Tree (possibly modified at root or below), or a new tree
// Any new tree is fully morphed, if necessary
|
Merging, the failed CI for Mono_LLVMFullAot is unrelated to this change. |
ghost
locked as resolved and limited conversation to collaborators
Description
Fixes #8833
Acceptance Criteria