[llvm-branch-commits] [llvm] [CGP][PAC] Flip PHI and blends when all immediate modifiers are the same (PR #150226)

[Oliver Hunt via llvm-branch-commits]

================
@@ -8309,6 +8316,74 @@ bool CodeGenPrepare::optimizeExtractElementInst(Instruction *Inst) {
   return false;
 }
 
+// Given the instruction Inst, rewrite its discriminator operand Disc if it is
+// a PHI node with all incoming values being @llvm.ptrauth.blend(addr, imm)
+// with the same immediate modifier.
+bool CodeGenPrepare::optimizePtrauthInst(Instruction *Inst, Value *Disc) {
----------------
ojhunt wrote:

This matching is fine for now, but the llvm intrinsic does not require the parameter to be a pointer. I think the best path forward would be to simply make the intrinsic match the expected interface? In principle no one should be emitting `@llvm.ptrauth.blend(i64, (i64)pointer)` (reversing the normal blend parameters), but maybe we should simply make it not an option to construct the intrinsic in that order?

That will require clang fixes to remove casts in the codegen, but also would presumably simplify things as it's removing those casts?

Maybe @ahmedbougacha or @rjmccall know why it is set up like this?


https://github.com/llvm/llvm-project/pull/150226