RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-22 10:42:39 +01:00
Code Issues Projects Releases Wiki Activity

[inliner] Skip debug intrinsics even earlier in computing the inline

Browse Source 
cost so that they don't impact the vector bonus. Fundamentally, counting
unsimplified instructions is just *wrong*; it will continue to introduce
instability as things which do not generate code bizarrely impact
inlining. For example, sufficiently nested inlined functions could turn
off the vector bonus with lifetime markers just like the debug
intrinsics do. =/

This is a short-term tactical fix. Long term, I think we need to remove
the vector bonus entirely. That's a separate patch and discussion
though.

The patch to fix this provided by Dario Domizioli. I've added some
comments about the planned direction and used a heavily pruned form of
debug info intrinsics for the test case. While this debug info doesn't
work or "do" anything useful, it lets us easily test all manner of
interference easily, and I suspect this will not be the last time we
want to craft a pattern where debug info interferes with the inliner in
a problematic way.

llvm-svn: 200609
This commit is contained in:
Chandler Carruth 2014-02-01 10:38:17 +00:00
parent f97f309449
commit e29471d99d
2 changed files with 65 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
lib/Analysis/IPA/InlineCost.cpp
View File

@ -872,6 +872,16 @@ bool CallAnalyzer::visitInstruction(Instruction &I) {
/// viable, and true if inlining remains viable.
bool CallAnalyzer::analyzeBlock(BasicBlock *BB) {
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
// FIXME: Currently, the number of instructions in a function regardless of
// our ability to simplify them during inline to constants or dead code,
// are actually used by the vector bonus heuristic. As long as that's true,
// we have to special case debug intrinsics here to prevent differences in
// inlining due to debug symbols. Eventually, the number of unsimplified
// instructions shouldn't factor into the cost computation, but until then,
// hack around it here.
if (isa<DbgInfoIntrinsic>(I))
continue;
++NumInstructions;
if (isa<ExtractElementInst>(I) || I->getType()->isVectorTy())
++NumVectorInstructions;

55
test/Transforms/Inline/ignore-debug-info.ll Normal file
View File

@ -0,0 +1,55 @@
; RUN: opt < %s -S -inline -inline-threshold=2 | FileCheck %s
; RUN: opt < %s -S -strip-debug -inline -inline-threshold=2 | FileCheck %s
;
; The purpose of this test is to check that debug info doesn't influence
; inlining decisions.
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"
declare void @llvm.dbg.declare(metadata, metadata) #1
declare void @llvm.dbg.value(metadata, i64, metadata) #1
define <4 x float> @inner_vectors(<4 x float> %a, <4 x float> %b) {
entry:
call void @llvm.dbg.value(metadata !{}, i64 0, metadata !{})
%mul = fmul <4 x float> %a, <float 3.000000e+00, float 3.000000e+00, float 3.000000e+00, float 3.000000e+00>
call void @llvm.dbg.value(metadata !{}, i64 0, metadata !{})
%mul1 = fmul <4 x float> %b, <float 5.000000e+00, float 5.000000e+00, float 5.000000e+00, float 5.000000e+00>
call void @llvm.dbg.value(metadata !{}, i64 0, metadata !{})
%add = fadd <4 x float> %mul, %mul1
ret <4 x float> %add
}
define float @outer_vectors(<4 x float> %a, <4 x float> %b) {
; CHECK-LABEL: @outer_vectors(
; CHECK-NOT: call <4 x float> @inner_vectors(
; CHECK: ret float
entry:
call void @llvm.dbg.value(metadata !{}, i64 0, metadata !{})
call void @llvm.dbg.value(metadata !{}, i64 0, metadata !{})
%call = call <4 x float> @inner_vectors(<4 x float> %a, <4 x float> %b)
call void @llvm.dbg.value(metadata !{}, i64 0, metadata !{})
%vecext = extractelement <4 x float> %call, i32 0
%vecext1 = extractelement <4 x float> %call, i32 1
%add = fadd float %vecext, %vecext1
%vecext2 = extractelement <4 x float> %call, i32 2
%add3 = fadd float %add, %vecext2
%vecext4 = extractelement <4 x float> %call, i32 3
%add5 = fadd float %add3, %vecext4
ret float %add5
}
attributes #0 = { nounwind readnone }
!llvm.dbg.cu = !{!0}
!llvm.module.flags = !{!3, !4}
!llvm.ident = !{!5}
!0 = metadata !{i32 786449, metadata !1, i32 4, metadata !"", i1 false, metadata !"", i32 0, metadata !2, metadata !2, metadata !{}, metadata !2, metadata !2, metadata !""}
!1 = metadata !{metadata !"", metadata !""}
!2 = metadata !{i32 0}
!3 = metadata !{i32 2, metadata !"Dwarf Version", i32 4}
!4 = metadata !{i32 1, metadata !"Debug Info Version", i32 1}
!5 = metadata !{metadata !""}