[SimplifyCFG] When merging conditional stores, don't count the store we're merging against the PHINodeFoldingThreshold

[craig.topper]

Summary

Merging conditional stores tries to check to see if the code is if convertible after the store is moved. But the store hasn't been moved yet so its being counted against the threshold.

The patch adds 1 to the threshold comparison to make sure we don't count the store. I've adjusted a test to use a lower threshold to ensure we still do that conversion with the lower threshold.

[spatel]

Adding Eli and Alexey as potential reviewers based on D36841 because I've never looked at this function before.

[efriedma]

What effect does this have on performance?

[craig.topper]

This had no affect on the binary output for any of the benchmarks we ran on trunk.

I only noticed this because one of our internal branches has diverged from community and changed the PHINodeFoldingThreshold from 2 to 1 (I wasn't involved in why that change was made in our branch). This lower threshold causes us to miss cases like merge-cond-store-2.ll despite the fact that it is if convertible with the threshold of 1 after the store is removed.

[efriedma]

LGTM

llvm/trunk/lib/Transforms/Utils/SimplifyCFG.cpp

for (auto &I : *BB) {
         ++N;
       // Free instructions.
       else if (isa<TerminatorInst>(I) || isa<DbgInfoIntrinsic>(I) ||
                IsaBitcastOfPointerType(I))
         continue;
       else
         return false;
     }
-    return N <= PHINodeFoldingThreshold;
+    // The store we want to merge is counted in N, so add 1 to make sure
+    // we're counting the instructions that would be left.
+    return N <= (PHINodeFoldingThreshold + 1);
   };
 
   if (!MergeCondStoresAggressively &&
       (!IsWorthwhile(PTB) || !IsWorthwhile(PFB) || !IsWorthwhile(QTB) ||
        !IsWorthwhile(QFB)))
     return false;
 
   // For every pointer, there must be exactly two stores, one coming from

llvm/trunk/test/Transforms/SimplifyCFG/merge-cond-stores-2.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
-; RUN: opt -S < %s -simplifycfg -simplifycfg-merge-cond-stores=true -simplifycfg-merge-cond-stores-aggressively=false -phi-node-folding-threshold=2 | FileCheck %s
+; RUN: opt -S < %s -simplifycfg -simplifycfg-merge-cond-stores=true -simplifycfg-merge-cond-stores-aggressively=false -phi-node-folding-threshold=1 | FileCheck %s
 
 target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
 target triple = "armv7--linux-gnueabihf"
 
 ; This is a bit reversal that has been run through the early optimizer (-mem2reg -gvn -instcombine).
 ; There should be no additional PHIs created at all. The store should be on its own in a predicated
 ; block and there should be no PHIs.