Revert r101471. For tight recursive functions which have multiple

recursive callsites, inlining can reduce the number of calls by
exponential factors, as it does in
MultiSource/Benchmarks/Olden/treeadd. More involved heuristics
will be needed.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101969 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Dan Gohman 2010-04-21 00:43:30 +00:00
parent a040fffefb
commit d217cfcf46
2 changed files with 0 additions and 36 deletions

View File

@ -263,13 +263,6 @@ InlineCost InlineCostAnalyzer::getInlineCost(CallSite CS,
CS.isNoInline())
return llvm::InlineCost::getNever();
// Don't inline directly recursive calls, for now. Inlining a directly
// recursive call is effectively unrolling a loop, so it calls for different
// heuristics, which aren't implemented yet. Until then, err on the
// conservative side.
if (Callee == Caller)
return llvm::InlineCost::getNever();
// InlineCost - This value measures how good of an inline candidate this call
// site is to inline. A lower inline cost make is more likely for the call to
// be inlined. This value may go negative.

View File

@ -1,29 +0,0 @@
; RUN: opt -inline -tailcallelim -indvars -loop-deletion -S < %s | FileCheck %s
; Inline shouldn't inline foo into itself because it's a tailcallelim
; candidate. Tailcallelim should convert the call into a loop. Indvars
; should calculate the exit value, making the loop dead. Loop deletion
; should delete the loop.
; PR6842
; CHECK: define i32 @bar() nounwind {
; CHECK-NEXT: ret i32 10000
; CHECK-NEXT: }
define internal i32 @foo(i32 %x) nounwind {
%i = add i32 %x, 1 ; <i32> [#uses=3]
%a = icmp slt i32 %i, 10000 ; <i1> [#uses=1]
br i1 %a, label %more, label %done
done: ; preds = %0
ret i32 %i
more: ; preds = %0
%z = tail call i32 @foo(i32 %i) ; <i32> [#uses=1]
ret i32 %z
}
define i32 @bar() nounwind {
%z = call i32 @foo(i32 0) ; <i32> [#uses=1]
ret i32 %z
}