forked from OSchip/llvm-project
a17f03bd93
We have several bug reports that could be characterized as "reducing scalarization", and this topic was also raised on llvm-dev recently: http://lists.llvm.org/pipermail/llvm-dev/2020-January/138157.html ...so I'm proposing that we deal with these patterns in a new, lightweight IR vector pass that runs before/after other vectorization passes. There are 4 alternate options that I can think of to deal with this kind of problem (and we've seen various attempts at all of these), but they all have flaws: InstCombine - can't happen without TTI, but we don't want target-specific folds there. SDAG - too late to assist other vectorization passes; TLI is not equipped for these kind of cost queries; limited to a single basic block. CGP - too late to assist other vectorization passes; would need to re-implement basic cleanups like CSE/instcombine. SLP - doesn't fit with existing transforms; limited to a single basic block. This initial patch/transform is based on existing code in AggressiveInstCombine: we walk backwards through the function looking for a pattern match. But we diverge from that cost-independent IR canonicalization pass by using TTI to decide if the vector alternative is profitable. We probably have at least 10 similar bug reports/patterns (binops, constants, inserts, cheap shuffles, etc) that would fit in this pass as follow-up enhancements. It's possible that we could iterate on a worklist to fix-point like InstCombine does, but it's safer to start with a most basic case and evolve from there, so I didn't try to do anything fancy with this initial implementation. Differential Revision: https://reviews.llvm.org/D73480 |
||
|---|---|---|
| .. | ||
| Inputs | ||
| X86 | ||
| 2002-01-31-CallGraph.ll | ||
| 2002-02-24-InlineBrokePHINodes.ll | ||
| 2002-03-11-ConstPropCrash.ll | ||
| 2003-02-19-LoopInfoNestingBug.ll | ||
| 2004-08-16-PackedConstantInlineStore.ll | ||
| 2004-08-16-PackedGlobalConstant.ll | ||
| 2004-08-16-PackedSelect.ll | ||
| 2004-08-16-PackedSimple.ll | ||
| 2004-08-20-PackedControlFlow.ll | ||
| 2006-02-05-PassManager.ll | ||
| 2007-06-05-PassID.ll | ||
| 2007-06-28-PassManager.ll | ||
| 2007-09-10-PassManager.ll | ||
| 2008-02-14-PassManager.ll | ||
| 2008-06-04-FieldSizeInPacked.ll | ||
| 2008-10-06-RemoveDeadPass.ll | ||
| 2008-10-15-MissingSpace.ll | ||
| 2009-03-31-CallGraph.ll | ||
| 2009-06-05-no-implicit-float.ll | ||
| 2009-09-14-function-elements.ll | ||
| 2010-05-06-Printer.ll | ||
| FileCheck-space.txt | ||
| ResponseFile.ll | ||
| attribute-comment.ll | ||
| available-externally-lto.ll | ||
| bcanalyzer-block-info.txt | ||
| bcanalyzer-dump-option.txt | ||
| can-execute.txt | ||
| cfg-printer-branch-weights.ll | ||
| cgscc-devirt-iteration.ll | ||
| cgscc-disconnected-invalidation.ll | ||
| cgscc-iterate-function-mutation.ll | ||
| cgscc-libcall-update.ll | ||
| cgscc-observe-devirt.ll | ||
| cleanup-lcssa.ll | ||
| close-stderr.ll | ||
| codegenprepare-and-debug.ll | ||
| constant-fold-gep-address-spaces.ll | ||
| constant-fold-gep.ll | ||
| cspgo-O2-pipeline.ll | ||
| debugcounter-dce.ll | ||
| debugcounter-earlycse.ll | ||
| debugcounter-newgvn.ll | ||
| debugcounter-predicateinfo.ll | ||
| extract-alias.ll | ||
| extract-linkonce.ll | ||
| extract-weak-odr.ll | ||
| extract.ll | ||
| invalid-commandline-option.ll | ||
| invariant.group.ll | ||
| lint.ll | ||
| lit-globbing.ll | ||
| lit-quoting.txt | ||
| lit-unicode.txt | ||
| llvm-nm-without-aliases.ll | ||
| loop-deletion-printer.ll | ||
| loop-pass-ordering.ll | ||
| loop-pass-printer.ll | ||
| loop-pm-invalidation.ll | ||
| machine-size-remarks.ll | ||
| module-pass-printer.ll | ||
| new-pass-manager.ll | ||
| new-pm-cspgo.ll | ||
| new-pm-defaults.ll | ||
| new-pm-lto-defaults.ll | ||
| new-pm-pgo-O0.ll | ||
| new-pm-pgo.ll | ||
| new-pm-pr42726-cgscc.ll | ||
| new-pm-thinlto-defaults.ll | ||
| new-pm-thinlto-postlink-pgo-defaults.ll | ||
| new-pm-thinlto-postlink-samplepgo-defaults.ll | ||
| new-pm-thinlto-prelink-pgo-defaults.ll | ||
| new-pm-thinlto-prelink-samplepgo-defaults.ll | ||
| opt-O0-pipeline.ll | ||
| opt-O2-pipeline.ll | ||
| opt-O3-pipeline.ll | ||
| opt-Os-pipeline.ll | ||
| opt-bisect-helper.py | ||
| opt-bisect-legacy-pass-manager.ll | ||
| opt-hot-cold-split.ll | ||
| opt-override-frame-pointer.ll | ||
| opt-override-mcpu-mattr.ll | ||
| opt-twice.ll | ||
| optimization-remarks-inline.ll | ||
| optimization-remarks-invalidation.ll | ||
| optimization-remarks-lazy-bfi.ll | ||
| optimize-inrange-gep.ll | ||
| optimize-options.ll | ||
| pass-pipeline-parsing.ll | ||
| pass-pipelines.ll | ||
| pipefail.txt | ||
| pr32085.ll | ||
| print-debug-counter.ll | ||
| print-module-scope.ll | ||
| print-slotindexes.ll | ||
| printer.ll | ||
| scalable-vector-array.ll | ||
| scalable-vector-struct.ll | ||
| scalable-vectors-core-ir.ll | ||
| scc-deleted-printer.ll | ||
| scc-pass-printer.ll | ||
| size-remarks.ll | ||
| spir_cc.ll | ||
| statistic.ll | ||
| time-passes.ll | ||
| unroll-sroa.ll | ||
| writing-to-stdout.ll | ||