[PATCH] D48586: [AMDGPU] Early expansion of 32 bit udiv/urem

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

Should we enable BypassSlowDivision or possibly merge this expansion with it?

The DAG expansion also probably needs to remain for all targets. DAGCombiner could still potentially introduce new div nodes that would need to be handled



================
Comment at: lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp:573-574
+  // fr = fabs(fr);
+  Function *FAbs = Intrinsic::getDeclaration(Mod, Intrinsic::fabs, { F32Ty });
+  FR = Builder.CreateCall(FAbs, { FR });
+
----------------
CreateIntrinsic should work for all of these


https://reviews.llvm.org/D48586