From: Vedant Kumar <vsk@apple.com>
Date: Thu, 17 Jan 2019 22:36:05 +0000 (+0000)
Subject: [HotColdSplit] Allow outlining with live outputs
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=7064dbae35e2b566b0587b231b2c413f032571c0;p=llvm

[HotColdSplit] Allow outlining with live outputs

Prior to r348205, extracting code regions with live output values was
disabled because of a miscompilation (PR39433). Lift the restriction as
PR39433 has been addressed.

Tested on LNT+externals, on a run of check-llvm in a stage2 build, and
with a full build of iOS (with hot/cold splitting enabled).

As a drive-by, remove an errant TODO.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@351492 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Transforms/IPO/HotColdSplitting.cpp b/lib/Transforms/IPO/HotColdSplitting.cpp
index 577f270863c..6876ae60259 100644
--- a/lib/Transforms/IPO/HotColdSplitting.cpp
+++ b/lib/Transforms/IPO/HotColdSplitting.cpp
@@ -252,16 +252,6 @@ Function *HotColdSplitting::extractColdRegion(const BlockSequence &Region,
                    /* AllowAlloca */ false,
                    /* Suffix */ "cold." + std::to_string(Count));
 
-  SetVector<Value *> Inputs, Outputs, Sinks;
-  CE.findInputsOutputs(Inputs, Outputs, Sinks);
-
-  // Do not extract regions that have live exit variables.
-  if (Outputs.size() > 0) {
-    LLVM_DEBUG(llvm::dbgs() << "Not outlining; live outputs\n");
-    return nullptr;
-  }
-
-  // TODO: Run MergeBasicBlockIntoOnlyPred on the outlined function.
   Function *OrigF = Region[0]->getParent();
   if (Function *OutF = CE.extractCodeRegion()) {
     User *U = *OutF->user_begin();
diff --git a/test/Transforms/HotColdSplit/split-phis-in-exit-blocks.ll b/test/Transforms/HotColdSplit/split-phis-in-exit-blocks.ll
new file mode 100644
index 00000000000..6b907ee3cc5
--- /dev/null
+++ b/test/Transforms/HotColdSplit/split-phis-in-exit-blocks.ll
@@ -0,0 +1,68 @@
+; RUN: opt -hotcoldsplit-threshold=0 -hotcoldsplit -S < %s | FileCheck %s
+
+target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx10.14.0"
+
+; CHECK-LABEL: define {{.*}}@pluto(
+; CHECK-NEXT: bb:
+; CHECK-NEXT:  %tmp8.ce.loc = alloca i1
+; CHECK-NEXT:  switch i8 undef, label %codeRepl [
+; CHECK-NEXT:    i8 0, label %bb7
+; CHECK-NEXT:    i8 1, label %bb7
+; CHECK-NEXT:  ]
+;
+; CHECK:  codeRepl:
+; CHECK-NEXT:    call void @pluto.cold.1(i1* %tmp8.ce.loc)
+; CHECK-NEXT:    %tmp8.ce.reload = load i1, i1* %tmp8.ce.loc
+; CHECK-NEXT:    br label %bb7
+;
+; CHECK:  bb7:
+; CHECK:    %tmp8 = phi i1 [ true, %bb ], [ true, %bb ], [ %tmp8.ce.reload, %codeRepl ]
+; CHECK:    ret void
+
+; CHECK-LABEL: define {{.*}}@pluto.cold.1(
+; CHECK: call {{.*}}@sideeffect(i32 1)
+; CHECK: call {{.*}}@sink(
+; CHECK: call {{.*}}@sideeffect(i32 3)
+; CHECK: call {{.*}}@sideeffect(i32 4)
+; CHECK: call {{.*}}@sideeffect(i32 5)
+define void @pluto() {
+bb:
+  switch i8 undef, label %bb1 [
+    i8 0, label %bb7
+    i8 1, label %bb7
+  ]
+
+bb1:                                              ; preds = %bb
+  call void @sideeffect(i32 1)
+  br label %bb2
+
+bb2:                                              ; preds = %bb1
+  call void @sink()
+  br i1 undef, label %bb7, label %bb3
+
+bb3:                                              ; preds = %bb2
+  call void @sideeffect(i32 3)
+  br label %bb4
+
+bb4:                                              ; preds = %bb3
+  call void @sideeffect(i32 4)
+  br i1 undef, label %bb5, label %bb6
+
+bb5:                                              ; preds = %bb4
+  call void @sideeffect(i32 5)
+  br label %bb6
+
+bb6:                                              ; preds = %bb5, %bb4
+  %tmp = phi i1 [ true, %bb5 ], [ false, %bb4 ]
+  call void @sideeffect(i32 6)
+  br label %bb7
+
+bb7:                                              ; preds = %bb6, %bb2, %bb, %bb
+  %tmp8 = phi i1 [ true, %bb ], [ true, %bb ], [ true, %bb2 ], [ %tmp, %bb6 ]
+  ret void
+}
+
+declare void @sink() cold
+
+declare void @sideeffect(i32)