Dehao Chen 3ecb35bc13 Use profile info to adjust loop unroll threshold. ...

Summary:
For flat loop, even if it is hot, it is not a good idea to unroll in runtime, thus we set a lower partial unroll threshold.
For hot loop, we set a higher unroll threshold and allows expensive tripcount computation to allow more aggressive unrolling.

Reviewers: davidxl, mzolotukhin

Subscribers: sanjoy, mehdi_amini, llvm-commits

Differential Revision: https://reviews.llvm.org/D26527

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287186 91177308-0d34-0410-b5e6-96231b3b80d8

2016-11-17 01:17:02 +00:00

..

AArch64

Reapply "[LoopUnroll] Use the upper bound of the loop trip count to fullly unroll a loop"

2016-10-12 21:29:38 +00:00

AMDGPU

AMDGPU: Remove some old intrinsic uses from tests

2016-02-11 06:02:01 +00:00

PowerPC

[LoopUnroll] Fix a PowerPC test broken by r277524.

2016-08-02 21:43:25 +00:00

X86

Adds the ability to use an epilog remainder loop during loop unrolling and makes

2016-04-05 12:19:35 +00:00

2004-05-13-DontUnrollTooMuch.ll

…

2005-03-06-BadLoopInfoUpdate.ll

…

2006-08-24-MultiBlockLoop.ll

…

2007-04-16-PhiUpdate.ll

…

2007-05-05-UnrollMiscomp.ll

…

2007-05-09-UnknownTripCount.ll

…

2007-11-05-Crash.ll

…

2011-08-08-PhiUpdate.ll

…

2011-08-09-IVSimplify.ll

…

2011-08-09-PhiUpdate.ll

…

2011-10-01-NoopTrunc.ll

…

2012-04-09-unroll-indirectbr.ll

…

basic.ll

…

convergent.ll

[LoopUnroll] Respect the convergent attribute.

2016-03-14 23:15:34 +00:00

ephemeral.ll

…

full-unroll-bad-cost.ll

…

full-unroll-crashers.ll

[LoopUnrollAnalyzer] Fix a bug in UnrolledInstAnalyzer::visitLoad.

2016-06-23 14:31:31 +00:00

full-unroll-heuristics-2.ll

[LoopUnrollAnalyzer] Take into account cost of instructions controlling branches, along with their operands.

2016-05-18 21:20:12 +00:00

full-unroll-heuristics-cmp.ll

[LoopUnrollAnalyzer] Don't treat gep-instructions with simplified offset as simplified.

2016-05-13 01:42:34 +00:00

full-unroll-heuristics-dce.ll

Revert "Revert "[Unroll] Implement a conservative and monotonically increasing cost tracking system during the full unroll heuristic analysis that avoids counting any instruction cost until that instruction becomes "live" through a side-effect or use outside the...""

2016-05-13 21:23:25 +00:00

full-unroll-heuristics-geps.ll

Revert "Revert "[Unroll] Implement a conservative and monotonically increasing cost tracking system during the full unroll heuristic analysis that avoids counting any instruction cost until that instruction becomes "live" through a side-effect or use outside the...""

2016-05-13 21:23:25 +00:00

full-unroll-heuristics-phi-prop.ll

[Unroll] Improve the brute force loop unroll estimate by propagating

2015-08-03 20:32:27 +00:00

full-unroll-heuristics.ll

…

full-unroll-keep-first-exit.ll

[LoopUnroll] Keep the loop test only on the first iteration of max-or-zero loops

2016-10-21 11:08:48 +00:00

high-cost-trip-count-computation.ll

Adds the ability to use an epilog remainder loop during loop unrolling and makes

2016-04-05 12:19:35 +00:00

ignore-annotation-intrinsic-cost.ll

…

loop-remarks-with-hotness.ll

[LoopUnroll] Include hotness of region in opt remark

2016-07-29 19:29:47 +00:00

loop-remarks.ll

…

partial-unroll-const-bounds.ll

The patch refactors unroll pass.

2016-05-27 23:15:06 +00:00

pr10813.ll

…

pr11361.ll

…

pr14167.ll

…

pr18861.ll

[Tests] Add one more case to LoopUnroll/pr18861.ll for better coverage.

2015-10-02 19:21:52 +00:00

pr27157.ll

[LoopUnroll] Fix the way we update DT after complete unrolling.

2016-04-06 21:47:12 +00:00

pr28132.ll

[LoopUnroll] Don't crash trying to unroll loop with EH pad exit

2016-06-15 00:19:56 +00:00

rebuild_lcssa.ll

[LoopSimplify] When simplifying phis in loop-simplify, do it only if it preserves LCSSA form.

2016-09-27 21:03:45 +00:00

runtime-loop1.ll

[LoopUnroll] Switch the default value of -unroll-runtime-epilog back to its original value.

2016-08-02 21:24:14 +00:00

runtime-loop2.ll

[LoopUnroll] Switch the default value of -unroll-runtime-epilog back to its original value.

2016-08-02 21:24:14 +00:00

runtime-loop3.ll

…

runtime-loop4.ll

[LoopUnroll] Switch the default value of -unroll-runtime-epilog back to its original value.

2016-08-02 21:24:14 +00:00

runtime-loop5.ll

[LoopUnroll] Switch the default value of -unroll-runtime-epilog back to its original value.

2016-08-02 21:24:14 +00:00

runtime-loop.ll

[LoopUnroll] Switch the default value of -unroll-runtime-epilog back to its original value.

2016-08-02 21:24:14 +00:00

scevunroll.ll

Revert "[IndVarSimplify] Rewrite loop exit values with their initial values from loop preheader"

2015-11-03 07:14:39 +00:00

shifted-tripcount.ll

…

tripcount-overflow.ll

[LoopUnroll] Switch the default value of -unroll-runtime-epilog back to its original value.

2016-08-02 21:24:14 +00:00

unloop.ll

[LoopUnroll] Include hotness of region in opt remark

2016-07-29 19:29:47 +00:00

unroll-cleanup.ll

[LoopUnroll] Switch the default value of -unroll-runtime-epilog back to its original value.

2016-08-02 21:24:14 +00:00

unroll-cleanuppad.ll

[LoopUnroll] Unroll loops which have exit blocks to EH pads

2016-05-03 03:57:40 +00:00

unroll-count.ll

The patch set unroll disable pragma when unroll

2016-06-08 20:21:24 +00:00

unroll-heuristics-pgo.ll

Use profile info to adjust loop unroll threshold.

2016-11-17 01:17:02 +00:00

unroll-opt-attribute.ll

Loop unroller: set thresholds for optsize and minsize functions to zero

2016-05-10 21:45:55 +00:00

unroll-pragmas-disabled.ll

…

unroll-pragmas.ll

Revert test change in r282894 as it's broken in some platforms.

2016-09-30 19:25:23 +00:00

update-loop-info-in-subloops.ll

…