[PartialInliner] Skip call sites where inlining fails.

Summary:
InlineFunction can fail, for example when trying to inline vararg
fuctions. In those cases, we do not want to bump partial inlining
counters or set AnyInlined to true, because this could leave an unused
function hanging around.

Reviewers: davidxl, davide, gyiu

Reviewed By: davide

Subscribers: llvm-commits, eraman

Differential Revision: https://reviews.llvm.org/D39581

llvm-svn: 317314

2 files changed, 54 insertions, 7 deletions

diff --git a/llvm/lib/Transforms/IPO/PartialInlining.cpp b/llvm/lib/Transforms/IPO/PartialInlining.cpp
index b5267f75e41..c47d8b78df3 100644
--- a/llvm/lib/Transforms/IPO/PartialInlining.cpp
+++ b/llvm/lib/Transforms/IPO/PartialInlining.cpp
@@ -931,15 +931,17 @@ bool PartialInlinerImpl::tryPartialInline(FunctionCloner &Cloner) {
     if (!shouldPartialInline(CS, Cloner, WeightedRcost, ORE))
       continue;
 
-    ORE.emit([&]() {
-      return OptimizationRemark(DEBUG_TYPE, "PartiallyInlined",
-                                CS.getInstruction())
-             << ore::NV("Callee", Cloner.OrigFunc) << " partially inlined into "
-             << ore::NV("Caller", CS.getCaller());
-    });
+    // Construct remark before doing the inlining, as after successful inlining
+    // the callsite is removed.
+    OptimizationRemark OR(DEBUG_TYPE, "PartiallyInlined", CS.getInstruction());
+    OR << ore::NV("Callee", Cloner.OrigFunc) << " partially inlined into "
+       << ore::NV("Caller", CS.getCaller());
 
     InlineFunctionInfo IFI(nullptr, GetAssumptionCache, PSI);
-    InlineFunction(CS, IFI);
+    if (!InlineFunction(CS, IFI))
+      continue;
+
+    ORE.emit(OR);
 
     // Now update the entry count:
     if (CalleeEntryCountV && CallSiteToProfCountMap.count(User)) {
diff --git a/llvm/test/Transforms/CodeExtractor/PartialInlineNoInline.ll b/llvm/test/Transforms/CodeExtractor/PartialInlineNoInline.ll
new file mode 100644
index 00000000000..6c0b83298d2
--- /dev/null
+++ b/llvm/test/Transforms/CodeExtractor/PartialInlineNoInline.ll
@@ -0,0 +1,45 @@
+; RUN: opt < %s -partial-inliner -S -stats -pass-remarks=partial-inlining 2>&1 | FileCheck %s
+; RUN: opt < %s -passes=partial-inliner -S -stats -pass-remarks=partial-inlining 2>&1 | FileCheck %s
+
+@stat = external global i32, align 4
+
+define i32 @inline_fail(i32 %count, ...) {
+entry:
+  %vargs = alloca i8*, align 8
+  %vargs1 = bitcast i8** %vargs to i8*
+  call void @llvm.va_start(i8* %vargs1)
+  %stat1 = load i32, i32* @stat, align 4
+  %cmp = icmp slt i32 %stat1, 0
+  br i1 %cmp, label %bb2, label %bb1
+
+bb1:                                              ; preds = %entry
+  %vg1 = add nsw i32 %stat1, 1
+  store i32 %vg1, i32* @stat, align 4
+  %va1 = va_arg i8** %vargs, i32
+  call void @foo(i32 %count, i32 %va1) #2
+  br label %bb2
+
+bb2:                                              ; preds = %bb1, %entry
+  %res = phi i32 [ 1, %bb1 ], [ 0, %entry ]
+  call void @llvm.va_end(i8* %vargs1)
+  ret i32 %res
+}
+
+define i32 @caller(i32 %arg) {
+bb:
+  %res = tail call i32 (i32, ...) @inline_fail(i32 %arg, i32 %arg)
+  ret i32 %res
+}
+
+declare void @foo(i32, i32)
+declare void @llvm.va_start(i8*)
+declare void @llvm.va_end(i8*)
+
+; Check that no remarks have been emitted, inline_fail has not been partial
+; inlined, no code has been extracted and the partial-inlining counter
+; has not been incremented.
+
+; CHECK-NOT: remark
+; CHECK: tail call i32 (i32, ...) @inline_fail(i32 %arg, i32 %arg)
+; CHECK-NOT: inline_fail.1_bb1
+; CHECK-NOT: partial-inlining