[flang] [llvm] [compiler-rt] [clang-tools-extra] [libc] [lldb] [clang] [GlobalIsel] Combine select of binops (PR #76763)

[Matt Arsenault via cfe-commits]

Thorsten =?utf-8?q?Schütt?= <schuett at gmail.com>,
Thorsten =?utf-8?q?Schütt?= <schuett at gmail.com>,
Thorsten =?utf-8?q?Schütt?= <schuett at gmail.com>
Message-ID:
In-Reply-To: <llvm.org/llvm/llvm-project/pull/76763 at github.com>


================
@@ -6548,6 +6534,54 @@ bool CombinerHelper::tryFoldBoolSelectToLogic(GSelect *Select,
   return false;
 }
 
+bool CombinerHelper::tryFoldSelectOfBinOps(GSelect *Select,
+                                           BuildFnTy &MatchInfo) {
+  Register DstReg = Select->getReg(0);
+  Register Cond = Select->getCondReg();
+  Register False = Select->getFalseReg();
+  Register True = Select->getTrueReg();
+  LLT DstTy = MRI.getType(DstReg);
+
+  GBinOp *LHS = getOpcodeDef<GBinOp>(True, MRI);
+  GBinOp *RHS = getOpcodeDef<GBinOp>(False, MRI);
+
+  // We need two binops of the same kind on the true/false registers.
+  if (!LHS || !RHS || LHS->getOpcode() != RHS->getOpcode())
+    return false;
+
+  // Note that there are no constraints on CondTy.
+  unsigned Flags = (LHS->getFlags() & RHS->getFlags()) | Select->getFlags();
----------------
arsenm wrote:

I would have split this by matching the DAG behavior in the initial commit but doesn't really matter. I *think* this is OK but would be nice to have alive verify 

https://github.com/llvm/llvm-project/pull/76763