6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2024-12-15 20:29:48 +00:00
Code Issues Projects Releases Wiki Activity
5e66eea5ba
llvm-6502/test/Analysis/BranchProbabilityInfo/loop.ll

366 lines
11 KiB
LLVM
Raw Normal View History

Fix the API usage in loop probability heuristics. It was incorrectly classifying many edges as exiting which were in fact not. These mainly formed edges into sub-loops. It was also not correctly classifying all returning edges out of loops as leaving the loop. With this match most of the loop heuristics are more rational. Several serious regressions on loop-intesive benchmarks like perlbench's loop tests when built with -enable-block-placement are fixed by these updated heuristics. Unfortunately they in turn uncover some other regressions. There are still several improvemenst that should be made to loop heuristics including trip-count, and early back-edge management. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142917 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-25 09:47:41 +00:00
; Test the static branch probability heuristics for no-return functions.
; RUN: opt < %s -analyze -branch-prob | FileCheck %s
declare void @g1()
declare void @g2()
declare void @g3()
declare void @g4()
define void @test1(i32 %a, i32 %b) {
entry:
br label %do.body
; CHECK: edge entry -> do.body probability is 16 / 16 = 100%
do.body:
%i.0 = phi i32 [ 0, %entry ], [ %inc3, %do.end ]
call void @g1()
br label %do.body1
Fix a bug in which BranchProbabilityInfo wasn't setting branch weights of basic blocks inside loops correctly. Previously, BranchProbabilityInfo::calcLoopBranchHeuristics would determine the weights of basic blocks inside loops even when it didn't have enough information to estimate the branch probabilities correctly. This patch fixes the function to exit early if it doesn't see any exit edges or back edges and let the later heuristics determine the weights. This fixes PR18705 and <rdar://problem/15991090>. Differential Revision: http://reviews.llvm.org/D3363 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206194 91177308-0d34-0410-b5e6-96231b3b80d8
2014-04-14 16:56:19 +00:00
; CHECK: edge do.body -> do.body1 probability is 16 / 16 = 100%
Fix the API usage in loop probability heuristics. It was incorrectly classifying many edges as exiting which were in fact not. These mainly formed edges into sub-loops. It was also not correctly classifying all returning edges out of loops as leaving the loop. With this match most of the loop heuristics are more rational. Several serious regressions on loop-intesive benchmarks like perlbench's loop tests when built with -enable-block-placement are fixed by these updated heuristics. Unfortunately they in turn uncover some other regressions. There are still several improvemenst that should be made to loop heuristics including trip-count, and early back-edge management. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142917 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-25 09:47:41 +00:00
do.body1:
%j.0 = phi i32 [ 0, %do.body ], [ %inc, %do.body1 ]
call void @g2()
%inc = add nsw i32 %j.0, 1
%cmp = icmp slt i32 %inc, %b
br i1 %cmp, label %do.body1, label %do.end
; CHECK: edge do.body1 -> do.body1 probability is 124 / 128
; CHECK: edge do.body1 -> do.end probability is 4 / 128
do.end:
call void @g3()
%inc3 = add nsw i32 %i.0, 1
%cmp4 = icmp slt i32 %inc3, %a
br i1 %cmp4, label %do.body, label %do.end5
; CHECK: edge do.end -> do.body probability is 124 / 128
; CHECK: edge do.end -> do.end5 probability is 4 / 128
do.end5:
call void @g4()
ret void
}
define void @test2(i32 %a, i32 %b) {
entry:
%cmp9 = icmp sgt i32 %a, 0
br i1 %cmp9, label %for.body.lr.ph, label %for.end6
; CHECK: edge entry -> for.body.lr.ph probability is 20 / 32
; CHECK: edge entry -> for.end6 probability is 12 / 32
for.body.lr.ph:
%cmp27 = icmp sgt i32 %b, 0
br label %for.body
; CHECK: edge for.body.lr.ph -> for.body probability is 16 / 16 = 100%
for.body:
%i.010 = phi i32 [ 0, %for.body.lr.ph ], [ %inc5, %for.end ]
call void @g1()
br i1 %cmp27, label %for.body3, label %for.end
Fix a bug in which BranchProbabilityInfo wasn't setting branch weights of basic blocks inside loops correctly. Previously, BranchProbabilityInfo::calcLoopBranchHeuristics would determine the weights of basic blocks inside loops even when it didn't have enough information to estimate the branch probabilities correctly. This patch fixes the function to exit early if it doesn't see any exit edges or back edges and let the later heuristics determine the weights. This fixes PR18705 and <rdar://problem/15991090>. Differential Revision: http://reviews.llvm.org/D3363 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206194 91177308-0d34-0410-b5e6-96231b3b80d8
2014-04-14 16:56:19 +00:00
; CHECK: edge for.body -> for.body3 probability is 20 / 32 = 62.5%
; CHECK: edge for.body -> for.end probability is 12 / 32 = 37.5%
Fix the API usage in loop probability heuristics. It was incorrectly classifying many edges as exiting which were in fact not. These mainly formed edges into sub-loops. It was also not correctly classifying all returning edges out of loops as leaving the loop. With this match most of the loop heuristics are more rational. Several serious regressions on loop-intesive benchmarks like perlbench's loop tests when built with -enable-block-placement are fixed by these updated heuristics. Unfortunately they in turn uncover some other regressions. There are still several improvemenst that should be made to loop heuristics including trip-count, and early back-edge management. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142917 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-25 09:47:41 +00:00
for.body3:
%j.08 = phi i32 [ %inc, %for.body3 ], [ 0, %for.body ]
call void @g2()
%inc = add nsw i32 %j.08, 1
%exitcond = icmp eq i32 %inc, %b
br i1 %exitcond, label %for.end, label %for.body3
; CHECK: edge for.body3 -> for.end probability is 4 / 128
; CHECK: edge for.body3 -> for.body3 probability is 124 / 128
for.end:
call void @g3()
%inc5 = add nsw i32 %i.010, 1
%exitcond11 = icmp eq i32 %inc5, %a
br i1 %exitcond11, label %for.end6, label %for.body
; CHECK: edge for.end -> for.end6 probability is 4 / 128
; CHECK: edge for.end -> for.body probability is 124 / 128
for.end6:
call void @g4()
ret void
}
define void @test3(i32 %a, i32 %b, i32* %c) {
entry:
br label %do.body
; CHECK: edge entry -> do.body probability is 16 / 16 = 100%
do.body:
%i.0 = phi i32 [ 0, %entry ], [ %inc4, %if.end ]
call void @g1()
%0 = load i32* %c, align 4
%cmp = icmp slt i32 %0, 42
br i1 %cmp, label %do.body1, label %if.end
Fix a bug in which BranchProbabilityInfo wasn't setting branch weights of basic blocks inside loops correctly. Previously, BranchProbabilityInfo::calcLoopBranchHeuristics would determine the weights of basic blocks inside loops even when it didn't have enough information to estimate the branch probabilities correctly. This patch fixes the function to exit early if it doesn't see any exit edges or back edges and let the later heuristics determine the weights. This fixes PR18705 and <rdar://problem/15991090>. Differential Revision: http://reviews.llvm.org/D3363 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206194 91177308-0d34-0410-b5e6-96231b3b80d8
2014-04-14 16:56:19 +00:00
; CHECK: edge do.body -> do.body1 probability is 16 / 32 = 50%
; CHECK: edge do.body -> if.end probability is 16 / 32 = 50%
Fix the API usage in loop probability heuristics. It was incorrectly classifying many edges as exiting which were in fact not. These mainly formed edges into sub-loops. It was also not correctly classifying all returning edges out of loops as leaving the loop. With this match most of the loop heuristics are more rational. Several serious regressions on loop-intesive benchmarks like perlbench's loop tests when built with -enable-block-placement are fixed by these updated heuristics. Unfortunately they in turn uncover some other regressions. There are still several improvemenst that should be made to loop heuristics including trip-count, and early back-edge management. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142917 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-25 09:47:41 +00:00
do.body1:
%j.0 = phi i32 [ %inc, %do.body1 ], [ 0, %do.body ]
call void @g2()
%inc = add nsw i32 %j.0, 1
%cmp2 = icmp slt i32 %inc, %b
br i1 %cmp2, label %do.body1, label %if.end
; CHECK: edge do.body1 -> do.body1 probability is 124 / 128
; CHECK: edge do.body1 -> if.end probability is 4 / 128
if.end:
call void @g3()
%inc4 = add nsw i32 %i.0, 1
%cmp5 = icmp slt i32 %inc4, %a
br i1 %cmp5, label %do.body, label %do.end6
; CHECK: edge if.end -> do.body probability is 124 / 128
; CHECK: edge if.end -> do.end6 probability is 4 / 128
do.end6:
call void @g4()
ret void
}
define void @test4(i32 %a, i32 %b, i32* %c) {
entry:
br label %do.body
; CHECK: edge entry -> do.body probability is 16 / 16 = 100%
do.body:
%i.0 = phi i32 [ 0, %entry ], [ %inc4, %do.end ]
call void @g1()
%0 = load i32* %c, align 4
%cmp = icmp slt i32 %0, 42
br i1 %cmp, label %return, label %do.body1
; CHECK: edge do.body -> return probability is 4 / 128
; CHECK: edge do.body -> do.body1 probability is 124 / 128
do.body1:
%j.0 = phi i32 [ %inc, %do.body1 ], [ 0, %do.body ]
call void @g2()
%inc = add nsw i32 %j.0, 1
%cmp2 = icmp slt i32 %inc, %b
br i1 %cmp2, label %do.body1, label %do.end
; CHECK: edge do.body1 -> do.body1 probability is 124 / 128
; CHECK: edge do.body1 -> do.end probability is 4 / 128
do.end:
call void @g3()
%inc4 = add nsw i32 %i.0, 1
%cmp5 = icmp slt i32 %inc4, %a
br i1 %cmp5, label %do.body, label %do.end6
; CHECK: edge do.end -> do.body probability is 124 / 128
; CHECK: edge do.end -> do.end6 probability is 4 / 128
do.end6:
call void @g4()
br label %return
; CHECK: edge do.end6 -> return probability is 16 / 16 = 100%
return:
ret void
}
define void @test5(i32 %a, i32 %b, i32* %c) {
entry:
br label %do.body
; CHECK: edge entry -> do.body probability is 16 / 16 = 100%
do.body:
%i.0 = phi i32 [ 0, %entry ], [ %inc4, %do.end ]
call void @g1()
br label %do.body1
Fix a bug in which BranchProbabilityInfo wasn't setting branch weights of basic blocks inside loops correctly. Previously, BranchProbabilityInfo::calcLoopBranchHeuristics would determine the weights of basic blocks inside loops even when it didn't have enough information to estimate the branch probabilities correctly. This patch fixes the function to exit early if it doesn't see any exit edges or back edges and let the later heuristics determine the weights. This fixes PR18705 and <rdar://problem/15991090>. Differential Revision: http://reviews.llvm.org/D3363 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206194 91177308-0d34-0410-b5e6-96231b3b80d8
2014-04-14 16:56:19 +00:00
; CHECK: edge do.body -> do.body1 probability is 16 / 16 = 100%
Fix the API usage in loop probability heuristics. It was incorrectly classifying many edges as exiting which were in fact not. These mainly formed edges into sub-loops. It was also not correctly classifying all returning edges out of loops as leaving the loop. With this match most of the loop heuristics are more rational. Several serious regressions on loop-intesive benchmarks like perlbench's loop tests when built with -enable-block-placement are fixed by these updated heuristics. Unfortunately they in turn uncover some other regressions. There are still several improvemenst that should be made to loop heuristics including trip-count, and early back-edge management. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142917 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-25 09:47:41 +00:00
do.body1:
%j.0 = phi i32 [ 0, %do.body ], [ %inc, %if.end ]
%0 = load i32* %c, align 4
%cmp = icmp slt i32 %0, 42
br i1 %cmp, label %return, label %if.end
; CHECK: edge do.body1 -> return probability is 4 / 128
; CHECK: edge do.body1 -> if.end probability is 124 / 128
if.end:
call void @g2()
%inc = add nsw i32 %j.0, 1
%cmp2 = icmp slt i32 %inc, %b
br i1 %cmp2, label %do.body1, label %do.end
; CHECK: edge if.end -> do.body1 probability is 124 / 128
; CHECK: edge if.end -> do.end probability is 4 / 128
do.end:
call void @g3()
%inc4 = add nsw i32 %i.0, 1
%cmp5 = icmp slt i32 %inc4, %a
br i1 %cmp5, label %do.body, label %do.end6
; CHECK: edge do.end -> do.body probability is 124 / 128
; CHECK: edge do.end -> do.end6 probability is 4 / 128
do.end6:
call void @g4()
br label %return
; CHECK: edge do.end6 -> return probability is 16 / 16 = 100%
return:
ret void
}
define void @test6(i32 %a, i32 %b, i32* %c) {
entry:
br label %do.body
; CHECK: edge entry -> do.body probability is 16 / 16 = 100%
do.body:
%i.0 = phi i32 [ 0, %entry ], [ %inc4, %do.end ]
call void @g1()
br label %do.body1
Fix a bug in which BranchProbabilityInfo wasn't setting branch weights of basic blocks inside loops correctly. Previously, BranchProbabilityInfo::calcLoopBranchHeuristics would determine the weights of basic blocks inside loops even when it didn't have enough information to estimate the branch probabilities correctly. This patch fixes the function to exit early if it doesn't see any exit edges or back edges and let the later heuristics determine the weights. This fixes PR18705 and <rdar://problem/15991090>. Differential Revision: http://reviews.llvm.org/D3363 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206194 91177308-0d34-0410-b5e6-96231b3b80d8
2014-04-14 16:56:19 +00:00
; CHECK: edge do.body -> do.body1 probability is 16 / 16 = 100%
Fix the API usage in loop probability heuristics. It was incorrectly classifying many edges as exiting which were in fact not. These mainly formed edges into sub-loops. It was also not correctly classifying all returning edges out of loops as leaving the loop. With this match most of the loop heuristics are more rational. Several serious regressions on loop-intesive benchmarks like perlbench's loop tests when built with -enable-block-placement are fixed by these updated heuristics. Unfortunately they in turn uncover some other regressions. There are still several improvemenst that should be made to loop heuristics including trip-count, and early back-edge management. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142917 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-25 09:47:41 +00:00
do.body1:
%j.0 = phi i32 [ 0, %do.body ], [ %inc, %do.cond ]
call void @g2()
%0 = load i32* %c, align 4
%cmp = icmp slt i32 %0, 42
br i1 %cmp, label %return, label %do.cond
; CHECK: edge do.body1 -> return probability is 4 / 128
; CHECK: edge do.body1 -> do.cond probability is 124 / 128
do.cond:
%inc = add nsw i32 %j.0, 1
%cmp2 = icmp slt i32 %inc, %b
br i1 %cmp2, label %do.body1, label %do.end
; CHECK: edge do.cond -> do.body1 probability is 124 / 128
; CHECK: edge do.cond -> do.end probability is 4 / 128
do.end:
call void @g3()
%inc4 = add nsw i32 %i.0, 1
%cmp5 = icmp slt i32 %inc4, %a
br i1 %cmp5, label %do.body, label %do.end6
; CHECK: edge do.end -> do.body probability is 124 / 128
; CHECK: edge do.end -> do.end6 probability is 4 / 128
do.end6:
call void @g4()
br label %return
; CHECK: edge do.end6 -> return probability is 16 / 16 = 100%
return:
ret void
}
define void @test7(i32 %a, i32 %b, i32* %c) {
entry:
%cmp10 = icmp sgt i32 %a, 0
br i1 %cmp10, label %for.body.lr.ph, label %for.end7
; CHECK: edge entry -> for.body.lr.ph probability is 20 / 32
; CHECK: edge entry -> for.end7 probability is 12 / 32
for.body.lr.ph:
%cmp38 = icmp sgt i32 %b, 0
br label %for.body
; CHECK: edge for.body.lr.ph -> for.body probability is 16 / 16 = 100%
for.body:
%i.011 = phi i32 [ 0, %for.body.lr.ph ], [ %inc6, %for.inc5 ]
%0 = load i32* %c, align 4
%cmp1 = icmp eq i32 %0, %i.011
br i1 %cmp1, label %for.inc5, label %if.end
Fix a bug in which BranchProbabilityInfo wasn't setting branch weights of basic blocks inside loops correctly. Previously, BranchProbabilityInfo::calcLoopBranchHeuristics would determine the weights of basic blocks inside loops even when it didn't have enough information to estimate the branch probabilities correctly. This patch fixes the function to exit early if it doesn't see any exit edges or back edges and let the later heuristics determine the weights. This fixes PR18705 and <rdar://problem/15991090>. Differential Revision: http://reviews.llvm.org/D3363 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206194 91177308-0d34-0410-b5e6-96231b3b80d8
2014-04-14 16:56:19 +00:00
; CHECK: edge for.body -> for.inc5 probability is 16 / 32 = 50%
; CHECK: edge for.body -> if.end probability is 16 / 32 = 50%
Fix the API usage in loop probability heuristics. It was incorrectly classifying many edges as exiting which were in fact not. These mainly formed edges into sub-loops. It was also not correctly classifying all returning edges out of loops as leaving the loop. With this match most of the loop heuristics are more rational. Several serious regressions on loop-intesive benchmarks like perlbench's loop tests when built with -enable-block-placement are fixed by these updated heuristics. Unfortunately they in turn uncover some other regressions. There are still several improvemenst that should be made to loop heuristics including trip-count, and early back-edge management. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142917 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-25 09:47:41 +00:00
if.end:
call void @g1()
br i1 %cmp38, label %for.body4, label %for.end
Fix a bug in which BranchProbabilityInfo wasn't setting branch weights of basic blocks inside loops correctly. Previously, BranchProbabilityInfo::calcLoopBranchHeuristics would determine the weights of basic blocks inside loops even when it didn't have enough information to estimate the branch probabilities correctly. This patch fixes the function to exit early if it doesn't see any exit edges or back edges and let the later heuristics determine the weights. This fixes PR18705 and <rdar://problem/15991090>. Differential Revision: http://reviews.llvm.org/D3363 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206194 91177308-0d34-0410-b5e6-96231b3b80d8
2014-04-14 16:56:19 +00:00
; CHECK: edge if.end -> for.body4 probability is 20 / 32 = 62.5%
; CHECK: edge if.end -> for.end probability is 12 / 32 = 37.5%
Fix the API usage in loop probability heuristics. It was incorrectly classifying many edges as exiting which were in fact not. These mainly formed edges into sub-loops. It was also not correctly classifying all returning edges out of loops as leaving the loop. With this match most of the loop heuristics are more rational. Several serious regressions on loop-intesive benchmarks like perlbench's loop tests when built with -enable-block-placement are fixed by these updated heuristics. Unfortunately they in turn uncover some other regressions. There are still several improvemenst that should be made to loop heuristics including trip-count, and early back-edge management. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142917 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-25 09:47:41 +00:00
for.body4:
%j.09 = phi i32 [ %inc, %for.body4 ], [ 0, %if.end ]
call void @g2()
%inc = add nsw i32 %j.09, 1
%exitcond = icmp eq i32 %inc, %b
br i1 %exitcond, label %for.end, label %for.body4
; CHECK: edge for.body4 -> for.end probability is 4 / 128
; CHECK: edge for.body4 -> for.body4 probability is 124 / 128
for.end:
call void @g3()
br label %for.inc5
Fix a bug in which BranchProbabilityInfo wasn't setting branch weights of basic blocks inside loops correctly. Previously, BranchProbabilityInfo::calcLoopBranchHeuristics would determine the weights of basic blocks inside loops even when it didn't have enough information to estimate the branch probabilities correctly. This patch fixes the function to exit early if it doesn't see any exit edges or back edges and let the later heuristics determine the weights. This fixes PR18705 and <rdar://problem/15991090>. Differential Revision: http://reviews.llvm.org/D3363 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206194 91177308-0d34-0410-b5e6-96231b3b80d8
2014-04-14 16:56:19 +00:00
; CHECK: edge for.end -> for.inc5 probability is 16 / 16 = 100%
Fix the API usage in loop probability heuristics. It was incorrectly classifying many edges as exiting which were in fact not. These mainly formed edges into sub-loops. It was also not correctly classifying all returning edges out of loops as leaving the loop. With this match most of the loop heuristics are more rational. Several serious regressions on loop-intesive benchmarks like perlbench's loop tests when built with -enable-block-placement are fixed by these updated heuristics. Unfortunately they in turn uncover some other regressions. There are still several improvemenst that should be made to loop heuristics including trip-count, and early back-edge management. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142917 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-25 09:47:41 +00:00
for.inc5:
%inc6 = add nsw i32 %i.011, 1
%exitcond12 = icmp eq i32 %inc6, %a
br i1 %exitcond12, label %for.end7, label %for.body
; CHECK: edge for.inc5 -> for.end7 probability is 4 / 128
; CHECK: edge for.inc5 -> for.body probability is 124 / 128
for.end7:
call void @g4()
ret void
}
define void @test8(i32 %a, i32 %b, i32* %c) {
entry:
%cmp18 = icmp sgt i32 %a, 0
br i1 %cmp18, label %for.body.lr.ph, label %for.end15
; CHECK: edge entry -> for.body.lr.ph probability is 20 / 32
; CHECK: edge entry -> for.end15 probability is 12 / 32
for.body.lr.ph:
%cmp216 = icmp sgt i32 %b, 0
%arrayidx5 = getelementptr inbounds i32* %c, i64 1
%arrayidx9 = getelementptr inbounds i32* %c, i64 2
br label %for.body
; CHECK: edge for.body.lr.ph -> for.body probability is 16 / 16 = 100%
for.body:
%i.019 = phi i32 [ 0, %for.body.lr.ph ], [ %inc14, %for.end ]
call void @g1()
br i1 %cmp216, label %for.body3, label %for.end
Fix a bug in which BranchProbabilityInfo wasn't setting branch weights of basic blocks inside loops correctly. Previously, BranchProbabilityInfo::calcLoopBranchHeuristics would determine the weights of basic blocks inside loops even when it didn't have enough information to estimate the branch probabilities correctly. This patch fixes the function to exit early if it doesn't see any exit edges or back edges and let the later heuristics determine the weights. This fixes PR18705 and <rdar://problem/15991090>. Differential Revision: http://reviews.llvm.org/D3363 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206194 91177308-0d34-0410-b5e6-96231b3b80d8
2014-04-14 16:56:19 +00:00
; CHECK: edge for.body -> for.body3 probability is 20 / 32 = 62.5%
; CHECK: edge for.body -> for.end probability is 12 / 32 = 37.5%
Fix the API usage in loop probability heuristics. It was incorrectly classifying many edges as exiting which were in fact not. These mainly formed edges into sub-loops. It was also not correctly classifying all returning edges out of loops as leaving the loop. With this match most of the loop heuristics are more rational. Several serious regressions on loop-intesive benchmarks like perlbench's loop tests when built with -enable-block-placement are fixed by these updated heuristics. Unfortunately they in turn uncover some other regressions. There are still several improvemenst that should be made to loop heuristics including trip-count, and early back-edge management. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142917 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-25 09:47:41 +00:00
for.body3:
%j.017 = phi i32 [ 0, %for.body ], [ %inc, %for.inc ]
%0 = load i32* %c, align 4
%cmp4 = icmp eq i32 %0, %j.017
br i1 %cmp4, label %for.inc, label %if.end
Fix a bug in which BranchProbabilityInfo wasn't setting branch weights of basic blocks inside loops correctly. Previously, BranchProbabilityInfo::calcLoopBranchHeuristics would determine the weights of basic blocks inside loops even when it didn't have enough information to estimate the branch probabilities correctly. This patch fixes the function to exit early if it doesn't see any exit edges or back edges and let the later heuristics determine the weights. This fixes PR18705 and <rdar://problem/15991090>. Differential Revision: http://reviews.llvm.org/D3363 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206194 91177308-0d34-0410-b5e6-96231b3b80d8
2014-04-14 16:56:19 +00:00
; CHECK: edge for.body3 -> for.inc probability is 16 / 32 = 50%
; CHECK: edge for.body3 -> if.end probability is 16 / 32 = 50%
Fix the API usage in loop probability heuristics. It was incorrectly classifying many edges as exiting which were in fact not. These mainly formed edges into sub-loops. It was also not correctly classifying all returning edges out of loops as leaving the loop. With this match most of the loop heuristics are more rational. Several serious regressions on loop-intesive benchmarks like perlbench's loop tests when built with -enable-block-placement are fixed by these updated heuristics. Unfortunately they in turn uncover some other regressions. There are still several improvemenst that should be made to loop heuristics including trip-count, and early back-edge management. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142917 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-25 09:47:41 +00:00
if.end:
%1 = load i32* %arrayidx5, align 4
%cmp6 = icmp eq i32 %1, %j.017
br i1 %cmp6, label %for.inc, label %if.end8
Fix a bug in which BranchProbabilityInfo wasn't setting branch weights of basic blocks inside loops correctly. Previously, BranchProbabilityInfo::calcLoopBranchHeuristics would determine the weights of basic blocks inside loops even when it didn't have enough information to estimate the branch probabilities correctly. This patch fixes the function to exit early if it doesn't see any exit edges or back edges and let the later heuristics determine the weights. This fixes PR18705 and <rdar://problem/15991090>. Differential Revision: http://reviews.llvm.org/D3363 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206194 91177308-0d34-0410-b5e6-96231b3b80d8
2014-04-14 16:56:19 +00:00
; CHECK: edge if.end -> for.inc probability is 16 / 32 = 50%
; CHECK: edge if.end -> if.end8 probability is 16 / 32 = 50%
Fix the API usage in loop probability heuristics. It was incorrectly classifying many edges as exiting which were in fact not. These mainly formed edges into sub-loops. It was also not correctly classifying all returning edges out of loops as leaving the loop. With this match most of the loop heuristics are more rational. Several serious regressions on loop-intesive benchmarks like perlbench's loop tests when built with -enable-block-placement are fixed by these updated heuristics. Unfortunately they in turn uncover some other regressions. There are still several improvemenst that should be made to loop heuristics including trip-count, and early back-edge management. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142917 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-25 09:47:41 +00:00
if.end8:
%2 = load i32* %arrayidx9, align 4
%cmp10 = icmp eq i32 %2, %j.017
br i1 %cmp10, label %for.inc, label %if.end12
Fix a bug in which BranchProbabilityInfo wasn't setting branch weights of basic blocks inside loops correctly. Previously, BranchProbabilityInfo::calcLoopBranchHeuristics would determine the weights of basic blocks inside loops even when it didn't have enough information to estimate the branch probabilities correctly. This patch fixes the function to exit early if it doesn't see any exit edges or back edges and let the later heuristics determine the weights. This fixes PR18705 and <rdar://problem/15991090>. Differential Revision: http://reviews.llvm.org/D3363 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206194 91177308-0d34-0410-b5e6-96231b3b80d8
2014-04-14 16:56:19 +00:00
; CHECK: edge if.end8 -> for.inc probability is 16 / 32 = 50%
; CHECK: edge if.end8 -> if.end12 probability is 16 / 32 = 50%
Fix the API usage in loop probability heuristics. It was incorrectly classifying many edges as exiting which were in fact not. These mainly formed edges into sub-loops. It was also not correctly classifying all returning edges out of loops as leaving the loop. With this match most of the loop heuristics are more rational. Several serious regressions on loop-intesive benchmarks like perlbench's loop tests when built with -enable-block-placement are fixed by these updated heuristics. Unfortunately they in turn uncover some other regressions. There are still several improvemenst that should be made to loop heuristics including trip-count, and early back-edge management. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142917 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-25 09:47:41 +00:00
if.end12:
call void @g2()
br label %for.inc
Fix a bug in which BranchProbabilityInfo wasn't setting branch weights of basic blocks inside loops correctly. Previously, BranchProbabilityInfo::calcLoopBranchHeuristics would determine the weights of basic blocks inside loops even when it didn't have enough information to estimate the branch probabilities correctly. This patch fixes the function to exit early if it doesn't see any exit edges or back edges and let the later heuristics determine the weights. This fixes PR18705 and <rdar://problem/15991090>. Differential Revision: http://reviews.llvm.org/D3363 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206194 91177308-0d34-0410-b5e6-96231b3b80d8
2014-04-14 16:56:19 +00:00
; CHECK: edge if.end12 -> for.inc probability is 16 / 16 = 100%
Fix the API usage in loop probability heuristics. It was incorrectly classifying many edges as exiting which were in fact not. These mainly formed edges into sub-loops. It was also not correctly classifying all returning edges out of loops as leaving the loop. With this match most of the loop heuristics are more rational. Several serious regressions on loop-intesive benchmarks like perlbench's loop tests when built with -enable-block-placement are fixed by these updated heuristics. Unfortunately they in turn uncover some other regressions. There are still several improvemenst that should be made to loop heuristics including trip-count, and early back-edge management. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142917 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-25 09:47:41 +00:00
for.inc:
%inc = add nsw i32 %j.017, 1
%exitcond = icmp eq i32 %inc, %b
br i1 %exitcond, label %for.end, label %for.body3
; CHECK: edge for.inc -> for.end probability is 4 / 128
; CHECK: edge for.inc -> for.body3 probability is 124 / 128
for.end:
call void @g3()
%inc14 = add nsw i32 %i.019, 1
%exitcond20 = icmp eq i32 %inc14, %a
br i1 %exitcond20, label %for.end15, label %for.body
; CHECK: edge for.end -> for.end15 probability is 4 / 128
; CHECK: edge for.end -> for.body probability is 124 / 128
for.end15:
call void @g4()
ret void
}
Reference in New Issue Copy Permalink