[llvm] r207403 - [inliner] Significantly improve the compile time in cases like PR19499

[Chandler Carruth]

Author: chandlerc
Date: Mon Apr 28 03:52:44 2014
New Revision: 207403

URL: http://llvm.org/viewvc/llvm-project?rev=207403&view=rev
Log:
[inliner] Significantly improve the compile time in cases like PR19499
by avoiding inlining massive switches merely because they have no
instructions in them. These switches still show up where we fail to form
lookup tables, and in those cases they are actually going to cause
a very significant code size hit anyways, so inlining them is not the
right call. The right way to fix any performance regressions stemming
from this is to enhance the switch-to-lookup-table logic to fire in more
places.

This makes PR19499 about 5x less bad. It uncovers a second compile time
problem in that test case that is unrelated (surprisingly!).

Added:
    llvm/trunk/test/Transforms/Inline/switch.ll
Modified:
    llvm/trunk/lib/Analysis/IPA/InlineCost.cpp

Modified: llvm/trunk/lib/Analysis/IPA/InlineCost.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/IPA/InlineCost.cpp?rev=207403&r1=207402&r2=207403&view=diff
==============================================================================
--- llvm/trunk/lib/Analysis/IPA/InlineCost.cpp (original)
+++ llvm/trunk/lib/Analysis/IPA/InlineCost.cpp Mon Apr 28 03:52:44 2014
@@ -808,9 +808,29 @@ bool CallAnalyzer::visitBranchInst(Branc
 bool CallAnalyzer::visitSwitchInst(SwitchInst &SI) {
   // We model unconditional switches as free, see the comments on handling
   // branches.
-  return isa<ConstantInt>(SI.getCondition()) ||
-         dyn_cast_or_null<ConstantInt>(
-             SimplifiedValues.lookup(SI.getCondition()));
+  if (isa<ConstantInt>(SI.getCondition()))
+    return true;
+  if (Value *V = SimplifiedValues.lookup(SI.getCondition()))
+    if (isa<ConstantInt>(V))
+      return true;
+
+  // Otherwise, we need to accumulate a cost proportional to the number of
+  // distinct successor blocks. This fan-out in the CFG cannot be represented
+  // for free even if we can represent the core switch as a jumptable that
+  // takes a single instruction.
+  //
+  // NB: We convert large switches which are just used to initialize large phi
+  // nodes to lookup tables instead in simplify-cfg, so this shouldn't prevent
+  // inlining those. It will prevent inlining in cases where the optimization
+  // does not (yet) fire.
+  SmallPtrSet<BasicBlock *, 8> SuccessorBlocks;
+  SuccessorBlocks.insert(SI.getDefaultDest());
+  for (auto I = SI.case_begin(), E = SI.case_end(); I != E; ++I)
+    SuccessorBlocks.insert(I.getCaseSuccessor());
+  // Add cost corresponding to the number of distinct destinations. The first
+  // we model as free because of fallthrough.
+  Cost += (SuccessorBlocks.size() - 1) * InlineConstants::InstrCost;
+  return false;
 }
 
 bool CallAnalyzer::visitIndirectBrInst(IndirectBrInst &IBI) {

Added: llvm/trunk/test/Transforms/Inline/switch.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/Inline/switch.ll?rev=207403&view=auto
==============================================================================
--- llvm/trunk/test/Transforms/Inline/switch.ll (added)
+++ llvm/trunk/test/Transforms/Inline/switch.ll Mon Apr 28 03:52:44 2014
@@ -0,0 +1,60 @@
+; RUN: opt < %s -inline -inline-threshold=20 -S | FileCheck %s
+
+define i32 @callee(i32 %a) {
+  switch i32 %a, label %sw.default [
+    i32 0, label %sw.bb0
+    i32 1, label %sw.bb1
+    i32 2, label %sw.bb2
+    i32 3, label %sw.bb3
+    i32 4, label %sw.bb4
+    i32 5, label %sw.bb5
+    i32 6, label %sw.bb6
+    i32 7, label %sw.bb7
+    i32 8, label %sw.bb8
+    i32 9, label %sw.bb9
+  ]
+
+sw.default:
+  br label %return
+
+sw.bb0:
+  br label %return
+
+sw.bb1:
+  br label %return
+
+sw.bb2:
+  br label %return
+
+sw.bb3:
+  br label %return
+
+sw.bb4:
+  br label %return
+
+sw.bb5:
+  br label %return
+
+sw.bb6:
+  br label %return
+
+sw.bb7:
+  br label %return
+
+sw.bb8:
+  br label %return
+
+sw.bb9:
+  br label %return
+
+return:
+  ret i32 42
+}
+
+define i32 @caller(i32 %a) {
+; CHECK-LABEL: @caller(
+; CHECK: call i32 @callee(
+
+  %result = call i32 @callee(i32 %a)
+  ret i32 %result
+}