[llvm] r290738 - [LICM] When promoting scalars, allow inserting stores to thread-local allocas.

[Michael Kuperstein via llvm-commits]

Author: mkuper
Date: Thu Dec 29 19:03:17 2016
New Revision: 290738

URL: http://llvm.org/viewvc/llvm-project?rev=290738&view=rev
Log:
[LICM] When promoting scalars, allow inserting stores to thread-local allocas.

This is similar to the allocfn case - if an alloca is not captured, then it's
necessarily thread-local.

Differential Revision: https://reviews.llvm.org/D28170

Modified:
    llvm/trunk/lib/Transforms/Scalar/LICM.cpp
    llvm/trunk/test/Transforms/LICM/promote-tls.ll

Modified: llvm/trunk/lib/Transforms/Scalar/LICM.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LICM.cpp?rev=290738&r1=290737&r2=290738&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/LICM.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/LICM.cpp Thu Dec 29 19:03:17 2016
@@ -1034,7 +1034,8 @@ bool llvm::promoteLoopAccessesToScalars(
   if (!SafeToInsertStore) {
     Value *Object = GetUnderlyingObject(SomePtr, MDL);
     SafeToInsertStore =
-        isAllocLikeFn(Object, TLI) && !PointerMayBeCaptured(Object, true, true);
+        (isAllocLikeFn(Object, TLI) || isa<AllocaInst>(Object)) &&
+        !PointerMayBeCaptured(Object, true, true);
   }
 
   // If we've still failed to prove we can sink the store, give up.

Modified: llvm/trunk/test/Transforms/LICM/promote-tls.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LICM/promote-tls.ll?rev=290738&r1=290737&r2=290738&view=diff
==============================================================================
--- llvm/trunk/test/Transforms/LICM/promote-tls.ll (original)
+++ llvm/trunk/test/Transforms/LICM/promote-tls.ll Thu Dec 29 19:03:17 2016
@@ -12,6 +12,7 @@ target triple = "x86_64-linux-generic"
 declare i8* @malloc(i64)
 
 ; Exercise the TLS case
+; CHECK-LABEL: @test
 define i32* @test(i32 %n) {
 entry:
   ;; ignore the required null check for simplicity
@@ -49,9 +50,48 @@ for.cond.for.end_crit_edge:
   ret i32* null
 }
 
+; Stack allocations can also be thread-local
+; CHECK-LABEL: @test2
+define i32* @test2(i32 %n) {
+entry:
+  %mem = alloca i8, i32 16
+  %addr = bitcast i8* %mem to i32*
+  br label %for.body.lr.ph
+
+for.body.lr.ph:                                   ; preds = %entry
+  br label %for.header
+
+for.header:
+  %i.02 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.body ]
+  %old = load i32, i32* %addr, align 4
+  ; deliberate impossible to analyze branch
+  %guard = load atomic i8*, i8** @p monotonic, align 8
+  %exitcmp = icmp eq i8* %guard, null
+  br i1 %exitcmp, label %for.body, label %early-exit
+
+early-exit:
+; CHECK-LABEL: early-exit:
+; CHECK: store i32 %new1.lcssa, i32* %addr, align 1
+  ret i32* null
+
+for.body:
+  %new = add i32 %old, 1
+  store i32 %new, i32* %addr, align 4
+  %inc = add nsw i32 %i.02, 1
+  %cmp = icmp slt i32 %inc, %n
+  br i1 %cmp, label %for.header, label %for.cond.for.end_crit_edge
+
+for.cond.for.end_crit_edge:                       ; preds = %for.body
+; CHECK-LABEL: for.cond.for.end_crit_edge:
+; CHECK: store i32 %new.lcssa, i32* %addr, align 1
+  %split = phi i32* [ %addr, %for.body ]
+  ret i32* null
+}
+
 declare i8* @not_malloc(i64)
 
 ; Negative test - not TLS
+; CHECK-LABEL: @test_neg
 define i32* @test_neg(i32 %n) {
 entry:
   ;; ignore the required null check for simplicity
@@ -93,6 +133,7 @@ for.cond.for.end_crit_edge:
 
 ; Negative test - can't speculate load since branch
 ; may control alignment
+; CHECK-LABEL: @test_neg2
 define i32* @test_neg2(i32 %n) {
 entry:
   ;; ignore the required null check for simplicity
@@ -131,4 +172,3 @@ for.cond.for.end_crit_edge:
   %split = phi i32* [ %addr, %for.body ]
   ret i32* null
 }
-