[llvm] [SimplifyCFG] Treat umul + extract pattern as cheap single instruction. (PR #124933)

[via llvm-commits]

================
@@ -3329,20 +3331,29 @@ bool SimplifyCFGOpt::speculativelyExecuteBB(BranchInst *BI,
     else
       ++SpeculatedInstructions;
 
-    if (SpeculatedInstructions > 1)
-      return false;
-
     // Don't hoist the instruction if it's unsafe or expensive.
     if (!IsSafeCheapLoadStore &&
         !isSafeToSpeculativelyExecute(&I, BI, Options.AC) &&
         !(HoistCondStores && !SpeculatedStoreValue &&
           (SpeculatedStoreValue =
                isSafeToSpeculateStore(&I, BB, ThenBB, EndBB))))
       return false;
-    if (!IsSafeCheapLoadStore && !SpeculatedStoreValue &&
-        computeSpeculationCost(&I, TTI) >
+
+    if (match(&I,
+              m_ExtractValue<1>(m_OneUse(
+                  m_Intrinsic<Intrinsic::umul_with_overflow>(m_Value())))) &&
+        ThenBB->size() <= 3)
+      PatternFound = true;
+
+    BlockCostSoFar += computeSpeculationCost(&I, TTI);
+    if (!PatternFound && !IsSafeCheapLoadStore && !SpeculatedStoreValue &&
+        BlockCostSoFar >
             PHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic)
       return false;
+    // If we don't find any pattern, that must be cheap, then only speculatively
+    // execute a single instruction (not counting the terminator).
+    if (!PatternFound && SpeculatedInstructions > 1)
+      return false;
----------------
goldsteinn wrote:

I think it may be easier to just add a `MaxSpeculatedInstructionsToHoist` variable and increment it if you find an overflow intrinsic whose only user is an extract.

https://github.com/llvm/llvm-project/pull/124933