summaryrefslogtreecommitdiff
path: root/compiler/nativeGen
Commit message (Collapse)AuthorAgeFilesLines
...
* codegen: fix memset unroll for small bytearrays, add 64-bit setsArtem Pyanykh2019-04-091-25/+55
| | | | | | | | | | | | | | | | | | | | | | Fixes #16052 When the offset in `setByteArray#` is statically known, we can provide better alignment guarantees then just 1 byte. Also, memset can now do 64-bit wide sets. The current memset intrinsic is not optimal however and can be improved for the case when we know that we deal with (baseAddress at known alignment) + offset For instance, on 64-bit `setByteArray# s 1# 23# 0#` given that bytearray is 8 bytes aligned could be unrolled into `movb, movw, movl, movq, movq`; but currently it is `movb x23` since alignment of 1 is all we can embed into MO_Memset op.
* Add support for bitreverse primopAlexandre2019-04-014-1/+17
| | | | | | This commit includes the necessary changes in code and documentation to support a primop that reverses a word's bits. It also includes a test.
* Update Wiki URLs to point to GitLabTakenobu Tani2019-03-252-2/+2
| | | | | | | | | | | | | | | | | | | | | | | This moves all URL references to Trac Wiki to their corresponding GitLab counterparts. This substitution is classified as follows: 1. Automated substitution using sed with Ben's mapping rule [1] Old: ghc.haskell.org/trac/ghc/wiki/XxxYyy... New: gitlab.haskell.org/ghc/ghc/wikis/xxx-yyy... 2. Manual substitution for URLs containing `#` index Old: ghc.haskell.org/trac/ghc/wiki/XxxYyy...#Zzz New: gitlab.haskell.org/ghc/ghc/wikis/xxx-yyy...#zzz 3. Manual substitution for strings starting with `Commentary` Old: Commentary/XxxYyy... New: commentary/xxx-yyy... See also !539 [1]: https://gitlab.haskell.org/bgamari/gitlab-migration/blob/master/wiki-mapping.json
* PPC NCG: Use liveness information in CmmCallPeter Trommler2019-03-154-42/+49
| | | | | | | | | | | | | | | | | We make liveness information for global registers available on `JMP` and `BCTR`, which were the last instructions missing. With complete liveness information we do not need to reserve global registers in `freeReg` anymore. Moreover we assign R9 and R10 to callee saves registers. Cleanup by removing `Reg_Su`, which was unused, from `freeReg` and removing unused register definitions. The calculation of the number of floating point registers is too conservative. Just follow X86 and specify the constants directly. Overall on PowerPC this results in 0.3 % smaller code size in nofib while runtime is slightly better in some tests.
* Update Trac ticket URLs to point to GitLabRyan Scott2019-03-153-5/+5
| | | | | This moves all URL references to Trac tickets to their corresponding GitLab counterparts.
* NCG: correctly escape path strings on Windows (#16389)Sylvain Henry2019-03-092-2/+4
| | | | | GHC native code generator generates .incbin and .file directives. We need to escape those strings correctly on Windows (see #16389).
* Rip out object splittingBen Gamari2019-03-057-65/+27
| | | | | | | | | | | | | | | The splitter is an evil Perl script that processes assembler code. Its job can be done better by the linker's --gc-sections flag. GHC passes this flag to the linker whenever -split-sections is passed on the command line. This is based on @DemiMarie's D2768. Fixes Trac #11315 Fixes Trac #9832 Fixes Trac #8964 Fixes Trac #8685 Fixes Trac #8629
* Don't wrap the entry map for LiveInfo in Maybe.klebinger.andreas@gmx.at2019-02-156-26/+27
| | | | | | | | | | | | | It never really encoded a invariant. * The linear register allocator just did partial pattern matches * The graph allocator just set it to (Just mapEmpty) for Nothing So I changed LiveInfo to directly contain the map. Further natCmmTopToLive which filled in Nothing is no longer exported. Instead we know call cmmTopLiveness which changes the type AND fills in the map.
* NCG: fast compilation of very large strings (#16190)Sylvain Henry2019-02-144-12/+51
| | | | | | | | | | This patch adds an optimization into the NCG: for large strings (threshold configurable via -fbinary-blob-threshold=NNN flag), instead of printing `.asciz "..."` in the generated ASM source, we print `.incbin "tmpXXX.dat"` and we dump the contents of the string into a temporary "tmpXXX.dat" file. See the note for more details.
* Fix Int overflow on 32 bit platformPeter Trommler2019-02-101-1/+1
|
* Stack: fix name mangling.Tamar Christina2019-02-091-1/+1
|
* Allow resizing the stack for the graph allocator.klebinger.andreas@gmx.at2019-02-086-36/+105
| | | | | | | | | | The graph allocator now dynamically resizes the number of stack slots when running into the limit. This fixes #8657. Also loop membership of basic blocks is now available in the register allocator for cost heuristics.
* Remove unused importsSebastian Graf2019-02-021-3/+0
|
* PPC NCG: Promote integers to word size in C callsPeter Trommler2019-01-311-13/+23
| | | | Fixes #16222
* Small optimizations to BlockLayout.klebinger.andreas@gmx.at2019-01-311-39/+31
| | | | | | | | | | * Remove `takeL/R 1` occurences by lastOL/headOL. * Make BlockChain a OrdList newtype by removing the set of blocks. Initially BlockChain contained both, a set for membership test and a ordered list of blocks. The set is not used for any performance sensitive lookups so we get rid of it.
* Replace BlockSequence with OrdList in BlockLayout.hsklebinger.andreas@gmx.at2019-01-311-76/+23
| | | | | OrdList does the same thing and more so there is no reason to have both.
* Optimize pprASCIISylvain Henry2019-01-311-12/+23
| | | | | | | | | * Use `ByteString.foldr` instead of `(List.foldr . BS.unpack)` * Avoid calling `chr` and its test that checks for invalid Unicode codepoints: we stay in the ASCII range so we know we're ok * Avoid calling `isPrint` (unsafe FFI call): we can check the ASCII printable range directly * Use bit operations (`unsafeShiftR`, `.&.`) instead of `div` and `mod`
* Use ByteString to represent Cmm string literals (#16198)Sylvain Henry2019-01-313-4/+8
| | | | Also used ByteString in some other relevant places
* Compile count{Leading,Trailing}Zeros to corresponding x86_64 instructions ↵Dmitry Ivanov2019-01-303-28/+63
| | | | | | | under -mbmi2 This works similarly to existing implementation for popCount. Trac ticket: #16086.
* Revert "Batch merge"Ben Gamari2019-01-303-63/+28
| | | | This reverts commit 76c8fd674435a652c75a96c85abbf26f1f221876.
* Batch mergeBen Gamari2019-01-303-28/+63
|
* Fix regDotColor for amd64.klebinger.andreas@gmx.at2019-01-272-31/+48
| | | | | Add missing color mappings to regDotColor for amd64. Also set fakeRegs to red instead of xmm regs.
* A few typofixesGabor Greif2019-01-231-1/+1
|
* PPC NCG: Rename constructorsPeter Trommler2019-01-171-28/+29
| | | | | Rename constructors in calling convention data type to reflect the fact that they represent an ELF ABI not only a Linux ABI.
* Fix tab and improve whitespacePeter Trommler2019-01-171-7/+8
|
* PPC NCG: Register definitions for all 64-bit systemsPeter Trommler2019-01-171-7/+3
|
* PPC NCG: GOT declaration for all 64-bit ELF systemsPeter Trommler2019-01-171-5/+3
|
* PPC NCG: Make `stackHeaderSize` more generalPeter Trommler2019-01-171-7/+6
|
* PPC NCG: Make calling convention more generalPeter Trommler2019-01-171-6/+5
| | | | | All operating systems except AIX and Darwin follow the ELF specification.
* PPC NCG: Refactor stack allocation codePeter Trommler2019-01-162-26/+15
| | | | | There is only one place where UPDATE_SP was used. Instead of the UPDATE_SP pseudo instruction build the list of instructions directly.
* PPC NCG: Reduce memory consumption emitting string literalsPeter Trommler2019-01-134-66/+55
|
* PPC NCG: Remove Darwin supportPeter Trommler2019-01-015-206/+42
| | | | | | | Support for Mac OS X on PowerPC has been dropped by Apple years ago. We follow suit and remove PowerPC support for Darwin. Fixes #16106.
* PPC NCG: Simple 64-bit condition code on 32-bitPeter Trommler2018-12-301-3/+48
|
* PPC NCG: Generate MO_?_QuotRem for subword sizesPeter Trommler2018-12-111-21/+24
| | | | | | | | | | | | | | | Handle Int*QuotRemOP and Word*QuotRemOp in PPC NCG. Refactor common code with remainder operation. Test Plan: validate (I validated on Linux powerpc64le and x86_64) Reviewers: erikd, hvr, bgamari, simonmar Reviewed By: bgamari Subscribers: rwbarton, carter Differential Revision: https://phabricator.haskell.org/D5323
* PPC NCG: Implement MachOps for smaller sizesPeter Trommler2018-12-111-161/+146
| | | | | | | | | | | | | | | | | Generate code for MachOps with smaller than wordsize data. Refactor conversion MachOps. Fixes #15854 Test Plan: validate (I validated on powerpc64le and x86_64 Linux) Reviewers: bgamari, hvr, erikd, simonmar Subscribers: rwbarton, carter GHC Trac Issues: #15854 Differential Revision: https://phabricator.haskell.org/D5300
* Taming the Kind Inference MonsterSimon Peyton Jones2018-11-291-1/+0
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | My original goal was (Trac #15809) to move towards using level numbers as the basis for deciding which type variables to generalise, rather than searching for the free varaibles of the environment. However it has turned into a truly major refactoring of the kind inference engine. Let's deal with the level-numbers part first: * Augment quantifyTyVars to calculate the type variables to quantify using level numbers, and compare the result with the existing approach. That is; no change in behaviour, just a WARNing if the two approaches give different answers. * To do this I had to get the level number right when calling quantifyTyVars, and this entailed a bit of care, especially in the code for kind-checking type declarations. * However, on the way I was able to eliminate or simplify a number of calls to solveEqualities. This work is incomplete: I'm not /using/ level numbers yet. When I subsequently get rid of any remaining WARNings in quantifyTyVars, that the level-number answers differ from the current answers, then I can rip out the current "free vars of the environment" stuff. Anyway, this led me into deep dive into kind inference for type and class declarations, which is an increasingly soggy part of GHC. Richard already did some good work recently in commit 5e45ad10ffca1ad175b10f6ef3327e1ed8ba25f3 Date: Thu Sep 13 09:56:02 2018 +0200 Finish fix for #14880. The real change that fixes the ticket is described in Note [Naughty quantification candidates] in TcMType. but I kept turning over stones. So this patch has ended up with a pretty significant refactoring of that code too. Kind inference for types and classes ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ * Major refactoring in the way we generalise the inferred kind of a TyCon, in kcTyClGroup. Indeed, I made it into a new top-level function, generaliseTcTyCon. Plus a new Note to explain it Note [Inferring kinds for type declarations]. * We decided (Trac #15592) not to treat class type variables specially when dealing with Inferred/Specified/Required for associated types. That simplifies things quite a bit. I also rewrote Note [Required, Specified, and Inferred for types] * Major refactoring of the crucial function kcLHsQTyVars: I split it into kcLHsQTyVars_Cusk and kcLHsQTyVars_NonCusk because the two are really quite different. The CUSK case is almost entirely rewritten, and is much easier because of our new decision not to treat the class variables specially * I moved all the error checks from tcTyClTyVars (which was a bizarre place for it) into generaliseTcTyCon and/or the CUSK case of kcLHsQTyVars. Now tcTyClTyVars is extremely simple. * I got rid of all the all the subtleties in tcImplicitTKBndrs. Indeed now there is no difference between tcImplicitTKBndrs and kcImplicitTKBndrs; there is now a single bindImplicitTKBndrs. Same for kc/tcExplicitTKBndrs. None of them monkey with level numbers, nor build implication constraints. scopeTyVars is gone entirely, as is kcLHsQTyVarBndrs. It's vastly simpler. I found I could get rid of kcLHsQTyVarBndrs entirely, in favour of the bnew bindExplicitTKBndrs. Quantification ~~~~~~~~~~~~~~ * I now deal with the "naughty quantification candidates" of the previous patch in candidateQTyVars, rather than in quantifyTyVars; see Note [Naughty quantification candidates] in TcMType. I also killed off closeOverKindsCQTvs in favour of the same strategy that we use for tyCoVarsOfType: namely, close over kinds at the occurrences. And candidateQTyVars no longer needs a gbl_tvs argument. * Passing the ContextKind, rather than the expected kind itself, to tc_hs_sig_type_and_gen makes it easy to allocate the expected result kind (when we are in inference mode) at the right level. Type families ~~~~~~~~~~~~~~ * I did a major rewrite of the impenetrable tcFamTyPats. The result is vastly more comprehensible. * I got rid of kcDataDefn entirely, quite a big function. * I re-did the way that checkConsistentFamInst works, so that it allows alpha-renaming of invisible arguments. * The interaction of kind signatures and family instances is tricky. Type families: see Note [Apparently-nullary families] Data families: see Note [Result kind signature for a data family instance] and Note [Eta-reduction for data families] * The consistent instantation of an associated type family is tricky. See Note [Checking consistent instantiation] and Note [Matching in the consistent-instantation check] in TcTyClsDecls. It's now checked in TcTyClsDecls because that is when we have the relevant info to hand. * I got tired of the compromises in etaExpandFamInst, so I did the job properly by adding a field cab_eta_tvs to CoAxBranch. See Coercion.etaExpandCoAxBranch. tcInferApps and friends ~~~~~~~~~~~~~~~~~~~~~~~ * I got rid of the mysterious and horrible ClsInstInfo argument to tcInferApps, checkExpectedKindX, and various checkValid functions. It was horrible! * I got rid of [Type] result of tcInferApps. This list was used only in tcFamTyPats, when checking the LHS of a type instance; and if there is a cast in the middle, the list is meaningless. So I made tcInferApps simpler, and moved the complexity (not much) to tcInferApps. Result: tcInferApps is now pretty comprehensible again. * I refactored the many function in TcMType that instantiate skolems. Smaller things * I rejigged the error message in checkValidTelescope; I think it's quite a bit better now. * checkValidType was not rejecting constraints in a kind signature forall (a :: Eq b => blah). blah2 That led to further errors when we then do an ambiguity check. So I make checkValidType reject it more aggressively. * I killed off quantifyConDecl, instead calling kindGeneralize directly. * I fixed an outright bug in tyCoVarsOfImplic, where we were not colleting the tyvar of the kind of the skolems * Renamed ClsInstInfo to AssocInstInfo, and made it into its own data type * Some fiddling around with pretty-printing of family instances which was trickier than I thought. I wanted wildcards to print as plain "_" in user messages, although they each need a unique identity in the CoAxBranch. Some other oddments * Refactoring around the trace messages from reportUnsolved. * A bit of extra tc-tracing in TcHsSyn.commitFlexi This patch fixes a raft of bugs, and includes tests for them. * #14887 * #15740 * #15764 * #15789 * #15804 * #15817 * #15870 * #15874 * #15881
* Fix unused-import warningsDavid Eichmann2018-11-225-6/+0
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This patch fixes a fairly long-standing bug (dating back to 2015) in RdrName.bestImport, namely commit 9376249b6b78610db055a10d05f6592d6bbbea2f Author: Simon Peyton Jones <simonpj@microsoft.com> Date: Wed Oct 28 17:16:55 2015 +0000 Fix unused-import stuff in a better way In that patch got the sense of the comparison back to front, and thereby failed to implement the unused-import rules described in Note [Choosing the best import declaration] in RdrName This led to Trac #13064 and #15393 Fixing this bug revealed a bunch of unused imports in libraries; the ones in the GHC repo are part of this commit. The two important changes are * Fix the bug in bestImport * Modified the rules by adding (a) in Note [Choosing the best import declaration] in RdrName Reason: the previosu rules made Trac #5211 go bad again. And the new rule (a) makes sense to me. In unravalling this I also ended up doing a few other things * Refactor RnNames.ImportDeclUsage to use a [GlobalRdrElt] for the things that are used, rather than [AvailInfo]. This is simpler and more direct. * Rename greParentName to greParent_maybe, to follow GHC naming conventions * Delete dead code RdrName.greUsedRdrName Bumps a few submodules. Reviewers: hvr, goldfire, bgamari, simonmar, jrtc27 Subscribers: rwbarton, carter Differential Revision: https://phabricator.haskell.org/D5312
* Fixup the new code layout patch for SplitObjs.klebinger.andreas@gmx.at2018-11-221-3/+7
| | | | | | | | | | | | | | | When splitting objects we sometimes generate dummy CmmProcs containing bottom in some fields. Code introduced in the new code layout patch looked at these which blew up the compiler. Now we instead check first if the function actually contains code. Reviewers: bgamari Subscribers: simonpj, rwbarton, carter Differential Revision: https://phabricator.haskell.org/D5357
* Rename literal constructorsSylvain Henry2018-11-225-32/+32
| | | | | | | | | | | | | | | | | | | | | | | | In a previous patch we replaced some built-in literal constructors (MachInt, MachWord, etc.) with a single LitNumber constructor. In this patch we replace the `Mach` prefix of the remaining constructors with `Lit` for consistency (e.g., LitChar, LitLabel, etc.). Sadly the name `LitString` was already taken for a kind of FastString and it would become misleading to have both `LitStr` (literal constructor renamed after `MachStr`) and `LitString` (FastString variant). Hence this patch renames the FastString variant `PtrString` (which is more accurate) and the literal string constructor now uses the least surprising `LitString` name. Both `Literal` and `LitString/PtrString` have recently seen breaking changes so doing this kind of renaming now shouldn't harm much. Reviewers: hvr, goldfire, bgamari, simonmar, jrtc27, tdammers Subscribers: tdammers, rwbarton, thomie, carter Differential Revision: https://phabricator.haskell.org/D4881
* NCG: New code layout algorithm.Andreas Klebinger2018-11-1718-319/+2021
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Summary: This patch implements a new code layout algorithm. It has been tested for x86 and is disabled on other platforms. Performance varies slightly be CPU/Machine but in general seems to be better by around 2%. Nofib shows only small differences of about +/- ~0.5% overall depending on flags/machine performance in other benchmarks improved significantly. Other benchmarks includes at least the benchmarks of: aeson, vector, megaparsec, attoparsec, containers, text and xeno. While the magnitude of gains differed three different CPUs where tested with all getting faster although to differing degrees. I tested: Sandy Bridge(Xeon), Haswell, Skylake * Library benchmark results summarized: * containers: ~1.5% faster * aeson: ~2% faster * megaparsec: ~2-5% faster * xml library benchmarks: 0.2%-1.1% faster * vector-benchmarks: 1-4% faster * text: 5.5% faster On average GHC compile times go down, as GHC compiled with the new layout is faster than the overhead introduced by using the new layout algorithm, Things this patch does: * Move code responsilbe for block layout in it's own module. * Move the NcgImpl Class into the NCGMonad module. * Extract a control flow graph from the input cmm. * Update this cfg to keep it in sync with changes during asm codegen. This has been tested on x64 but should work on x86. Other platforms still use the old codelayout. * Assign weights to the edges in the CFG based on type and limited static analysis which are then used for block layout. * Once we have the final code layout eliminate some redundant jumps. In particular turn a sequences of: jne .foo jmp .bar foo: into je bar foo: .. Test Plan: ci Reviewers: bgamari, jmct, jrtc27, simonmar, simonpj, RyanGlScott Reviewed By: RyanGlScott Subscribers: RyanGlScott, trommler, jmct, carter, thomie, rwbarton GHC Trac Issues: #15124 Differential Revision: https://phabricator.haskell.org/D4726
* Add Int8# and Word8#Michal Terepeta2018-11-023-15/+108
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This is the first step of implementing: https://github.com/ghc-proposals/ghc-proposals/pull/74 The main highlights/changes: primops.txt.pp gets two new sections for two new primitive types for signed and unsigned 8-bit integers (Int8# and Word8 respectively) along with basic arithmetic and comparison operations. PrimRep/RuntimeRep get two new constructors for them. All of the primops translate into the existing MachOPs. For CmmCalls the codegen will now zero-extend the values at call site (so that they can be moved to the right register) and then truncate them back their original width. x86 native codegen needed some updates, since it wasn't able to deal with the new widths, but all the changes are quite localized. LLVM backend seems to just work. This is the second attempt at merging this, after the first attempt in D4475 had to be backed out due to regressions on i386. Bumps binary submodule. Signed-off-by: Michal Terepeta <michal.terepeta@gmail.com> Test Plan: ./validate (on both x86-{32,64}) Reviewers: bgamari, hvr, goldfire, simonmar Subscribers: rwbarton, carter Differential Revision: https://phabricator.haskell.org/D5258
* Fix rare undefined asm temp end label error in x86Zejun Wu2018-10-281-7/+8
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Summary: Encountered assembly error due to undefined label `.LcaDcU_info_end` for following code generated by `pprFrameProc`: ``` .Lsat_sa8fp{v}_info_fde_end: .long .Lblock{v caDcU}_info_fde_end-.Lblock{v caDcU}_info_fde .Lblock{v caDcU}_info_fde: .long _nbHlD-.Lsection_frame .quad block{v caDcU}_info-1 .quad .Lblock{v caDcU}_info_end-block{v caDcU}_info+1 .byte 1 ``` This diff fixed the error. Test Plan: ./validate Also the case where we used to have assembly error is now fixed. Unfortunately, I have limited insight here and cannot get a small enough repro or test case for this. Ben says: > I think I see: Previously we only produced end symbols for the info > tables of top-level procedures. However, blocks within a procedure may > also have info tables, we will dutifully generate debug information for > and consequently we get undefined symbols. Reviewers: simonmar, scpmw, last_g, bgamari Reviewed By: bgamari Subscribers: rwbarton, carter Differential Revision: https://phabricator.haskell.org/D5246
* Revert "Add Int8# and Word8#"Ben Gamari2018-10-093-97/+11
| | | | | | | | | This unfortunately broke i386 support since it introduced references to byte-sized registers that don't exist on that architecture. Reverts binary submodule This reverts commit 5d5307f943d7581d7013ffe20af22233273fba06.
* Add Int8# and Word8#Michal Terepeta2018-10-073-11/+97
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This is the first step of implementing: https://github.com/ghc-proposals/ghc-proposals/pull/74 The main highlights/changes: - `primops.txt.pp` gets two new sections for two new primitive types for signed and unsigned 8-bit integers (`Int8#` and `Word8` respectively) along with basic arithmetic and comparison operations. `PrimRep`/`RuntimeRep` get two new constructors for them. All of the primops translate into the existing `MachOP`s. - For `CmmCall`s the codegen will now zero-extend the values at call site (so that they can be moved to the right register) and then truncate them back their original width. - x86 native codegen needed some updates, since it wasn't able to deal with the new widths, but all the changes are quite localized. LLVM backend seems to just work. Bumps binary submodule. Signed-off-by: Michal Terepeta <michal.terepeta@gmail.com> Test Plan: ./validate with new tests Reviewers: hvr, goldfire, bgamari, simonmar Subscribers: Abhiroop, dfeuer, rwbarton, thomie, carter Differential Revision: https://phabricator.haskell.org/D4475
* Invert FP conditions to eliminate the explicit NaN check.klebinger.andreas@gmx.at2018-09-181-22/+78
| | | | | | | | | | | | | | | | | | | | | | | | | | Summary: Optimisation: we don't have to test the parity flag if we know the test has already excluded the unordered case: eg > and >= test for a zero carry flag, which can only occur for ordered operands. By reversing comparisons we can avoid testing the parity for < and <= as well. This works since: * If any of the arguments is an NaN CF gets set. Resulting in a false result. * Since this allows us to rule out NaN we can exchange the arguments and invert the direction of the arrows. Test Plan: ci/nofib Reviewers: carter, bgamari, alpmestan Reviewed By: alpmestan Subscribers: alpmestan, simonpj, jmct, rwbarton, thomie GHC Trac Issues: #15196 Differential Revision: https://phabricator.haskell.org/D4990
* Mark code related symbols as @function not @objectSergei Azovskov2018-09-141-5/+64
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Summary: This diff is a part of the bigger project which goal is to improve common profiling tools support (perf) for GHC binaries. A similar job was already done and reverted in the past: * https://phabricator.haskell.org/rGHCb1f453e16f0ce11a2ab18cc4c350bdcbd36299a6 * https://phabricator.haskell.org/rGHCf1f3c4f50650110ad0f700d6566a44c515b0548f Reasoning: `Perf` and similar tools build in memory symbol table from the .symtab section of the ELF file to display human-readable function names instead of the addresses in the output. `Perf` uses only two types of symbols: `@function` and `@notype` but GHC is not capable to produce any `@function` symbols so the `perf` output is pretty useless (All the haskell symbols that you can see in `perf` now are `@notype` internal symbols extracted by mistake/hack). The changes: * mark code related symbols as @function * small hack to mark InfoTable symbols as code if TABLES_NEXT_TO_CODE is true Limitations: * The perf symbolization support is not complete after this patch but I'm working on the second patch. * Constructor symbols are not supported. To fix that we can issue extra local symbols which mark code sections as code and will be only used for debug. Test Plan: tests any additional ideas? Perf output on stock ghc 8.4.1: ``` 9.78% FibbSlow FibbSlow [.] ckY_info 9.59% FibbSlow FibbSlow [.] cjqd_info 7.17% FibbSlow FibbSlow [.] c3sg_info 6.62% FibbSlow FibbSlow [.] c1X_info 5.32% FibbSlow FibbSlow [.] cjsX_info 4.18% FibbSlow FibbSlow [.] s3rN_info 3.82% FibbSlow FibbSlow [.] c2m_info 3.68% FibbSlow FibbSlow [.] cjlJ_info 3.26% FibbSlow FibbSlow [.] c3sb_info 3.19% FibbSlow FibbSlow [.] cjPQ_info 3.05% FibbSlow FibbSlow [.] cjQd_info 2.97% FibbSlow FibbSlow [.] cjAB_info 2.78% FibbSlow FibbSlow [.] cjzP_info 2.40% FibbSlow FibbSlow [.] cjOS_info 2.38% FibbSlow FibbSlow [.] s3rK_info 2.27% FibbSlow FibbSlow [.] cjq0_info 2.18% FibbSlow FibbSlow [.] cKQ_info 2.13% FibbSlow FibbSlow [.] cjSl_info 1.99% FibbSlow FibbSlow [.] s3rL_info 1.98% FibbSlow FibbSlow [.] c2cC_info 1.80% FibbSlow FibbSlow [.] s3rO_info 1.37% FibbSlow FibbSlow [.] c2f2_info ... ``` Perf output on patched ghc: ``` 7.97% FibbSlow FibbSlow [.] c3rM_info 6.75% FibbSlow FibbSlow [.] 0x000000000032cfa8 6.63% FibbSlow FibbSlow [.] cifA_info 4.98% FibbSlow FibbSlow [.] integerzmgmp_GHCziIntegerziType_eqIntegerzh_info 4.55% FibbSlow FibbSlow [.] chXn_info 4.52% FibbSlow FibbSlow [.] c3rH_info 4.45% FibbSlow FibbSlow [.] chZB_info 4.04% FibbSlow FibbSlow [.] Main_fibbzuslow_info 4.03% FibbSlow FibbSlow [.] stg_ap_0_fast 3.76% FibbSlow FibbSlow [.] chXA_info 3.67% FibbSlow FibbSlow [.] cifu_info 3.25% FibbSlow FibbSlow [.] ci4r_info 2.64% FibbSlow FibbSlow [.] s3rf_info 2.42% FibbSlow FibbSlow [.] s3rg_info 2.39% FibbSlow FibbSlow [.] integerzmgmp_GHCziIntegerziType_eqInteger_info 2.25% FibbSlow FibbSlow [.] integerzmgmp_GHCziIntegerziType_minusInteger_info 2.17% FibbSlow FibbSlow [.] ghczmprim_GHCziClasses_zeze_info 2.09% FibbSlow FibbSlow [.] cicc_info 2.03% FibbSlow FibbSlow [.] 0x0000000000331e15 2.02% FibbSlow FibbSlow [.] s3ri_info 1.91% FibbSlow FibbSlow [.] 0x0000000000331bb8 1.89% FibbSlow FibbSlow [.] ci4N_info ... ``` Reviewers: simonmar, niteria, bgamari, goldfire Reviewed By: simonmar, bgamari Subscribers: lelf, rwbarton, thomie, carter GHC Trac Issues: #15501 Differential Revision: https://phabricator.haskell.org/D4713
* Fix precision of asinh/acosh/atanh by making them primopsArtem Pelenitsyn2018-08-213-0/+24
| | | | | | | | | | Reviewers: hvr, bgamari, simonmar, jrtc27 Reviewed By: bgamari Subscribers: alpmestan, rwbarton, thomie, carter Differential Revision: https://phabricator.haskell.org/D5034
* Replace most occurences of foldl with foldl'.klebinger.andreas@gmx.at2018-08-215-9/+2
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This patch adds foldl' to GhcPrelude and changes must occurences of foldl to foldl'. This leads to better performance especially for quick builds where GHC does not perform strictness analysis. It does change strictness behaviour when we use foldl' to turn a argument list into function applications. But this is only a drawback if code looks ONLY at the last argument but not at the first. And as the benchmarks show leads to fewer allocations in practice at O2. Compiler performance for Nofib: O2 Allocations: -1 s.d. ----- -0.0% +1 s.d. ----- -0.0% Average ----- -0.0% O2 Compile Time: -1 s.d. ----- -2.8% +1 s.d. ----- +1.3% Average ----- -0.8% O0 Allocations: -1 s.d. ----- -0.2% +1 s.d. ----- -0.1% Average ----- -0.2% Test Plan: ci Reviewers: goldfire, bgamari, simonmar, tdammers, monoidal Reviewed By: bgamari, monoidal Subscribers: tdammers, rwbarton, thomie, carter Differential Revision: https://phabricator.haskell.org/D4929
* Turn on MonadFail desugaring by defaultHerbert Valerio Riedel2018-08-071-2/+8
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Summary: This contains two commits: ---- Make GHC's code-base compatible w/ `MonadFail` There were a couple of use-sites which implicitly used pattern-matches in `do`-notation even though the underlying `Monad` didn't explicitly support `fail` This refactoring turns those use-sites into explicit case discrimations and adds an `MonadFail` instance for `UniqSM` (`UniqSM` was the worst offender so this has been postponed for a follow-up refactoring) --- Turn on MonadFail desugaring by default This finally implements the phase scheduled for GHC 8.6 according to https://prime.haskell.org/wiki/Libraries/Proposals/MonadFail#Transitionalstrategy This also preserves some tests that assumed MonadFail desugaring to be active; all ghc boot libs were already made compatible with this `MonadFail` long ago, so no changes were needed there. Test Plan: Locally performed ./validate --fast Reviewers: bgamari, simonmar, jrtc27, RyanGlScott Reviewed By: bgamari Subscribers: bgamari, RyanGlScott, rwbarton, thomie, carter Differential Revision: https://phabricator.haskell.org/D5028
* stack: fix stack allocations on WindowsTamar Christina2018-07-184-27/+103
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Summary: On Windows one is not allowed to drop the stack by more than a page size. The reason for this is that the OS only allocates enough stack till what the TEB specifies. After that a guard page is placed and the rest of the virtual address space is unmapped. The intention is that doing stack allocations will cause you to hit the guard which will then map the next page in and move the guard. This is done to prevent what in the Linux world is known as stack clash vulnerabilities https://access.redhat.com/security/cve/cve-2017-1000364. There are modules in GHC for which the liveliness analysis thinks the reserved 8KB of spill slots isn't enough. One being DynFlags and the other being Cabal. Though I think the Cabal one is likely a bug: ``` 4d6544: 81 ec 00 46 00 00 sub $0x4600,%esp 4d654a: 8d 85 94 fe ff ff lea -0x16c(%ebp),%eax 4d6550: 3b 83 1c 03 00 00 cmp 0x31c(%ebx),%eax 4d6556: 0f 82 de 8d 02 00 jb 4ff33a <_cLpg_info+0x7a> 4d655c: c7 45 fc 14 3d 50 00 movl $0x503d14,-0x4(%ebp) 4d6563: 8b 75 0c mov 0xc(%ebp),%esi 4d6566: 83 c5 fc add $0xfffffffc,%ebp 4d6569: 66 f7 c6 03 00 test $0x3,%si 4d656e: 0f 85 a6 d7 02 00 jne 503d1a <_cLpb_info+0x6> 4d6574: 81 c4 00 46 00 00 add $0x4600,%esp ``` It allocates nearly 18KB of spill slots for a simple 4 line function and doesn't even use it. Note that this doesn't happen on x64 or when making a validate build. Only when making a build without a validate and build.mk. This and the allocation in DynFlags means the stack allocation will jump over the guard page into unmapped memory areas and GHC or an end program segfaults. The pagesize on x86 Windows is 4KB which means we hit it very easily for these two modules, which explains the total DOA of GHC 32bit for the past 3 releases and the "random" segfaults on Windows. ``` 0:000> bp 00503d29 0:000> gn Breakpoint 0 hit WARNING: Stack overflow detected. The unwound frames are extracted from outside normal stack bounds. eax=03b6b9c9 ebx=00dc90f0 ecx=03cac48c edx=03cac43d esi=03b6b9c9 edi=03abef40 eip=00503d29 esp=013e96fc ebp=03cf8f70 iopl=0 nv up ei pl nz na po nc cs=0023 ss=002b ds=002b es=002b fs=0053 gs=002b efl=00000202 setup+0x103d29: 00503d29 89442440 mov dword ptr [esp+40h],eax ss:002b:013e973c=???????? WARNING: Stack overflow detected. The unwound frames are extracted from outside normal stack bounds. WARNING: Stack overflow detected. The unwound frames are extracted from outside normal stack bounds. 0:000> !teb TEB at 00384000 ExceptionList: 013effcc StackBase: 013f0000 StackLimit: 013eb000 ``` This doesn't fix the liveliness analysis but does fix the allocations, by emitting a function call to `__chkstk_ms` when doing allocations of larger than a page, this will make sure the stack is probed every page so the kernel maps in the next page. `__chkstk_ms` is provided by `libGCC`, which is under the `GNU runtime exclusion license`, so it's safe to link against it, even for proprietary code. (Technically we already do since we link compiled C code in.) For allocations smaller than a page we drop the stack and probe the new address. This avoids the function call and still makes sure we hit the guard if needed. PS: In case anyone is Wondering why we didn't notice this before, it's because we only test x86_64 and on Windows 10. On x86_64 the page size is 8KB and also the kernel is a bit more lenient on Windows 10 in that it seems to catch the segfault and resize the stack if it was unmapped: ``` 0:000> t eax=03b6b9c9 ebx=00dc90f0 ecx=03cac48c edx=03cac43d esi=03b6b9c9 edi=03abef40 eip=00503d2d esp=013e96fc ebp=03cf8f70 iopl=0 nv up ei pl nz na po nc cs=0023 ss=002b ds=002b es=002b fs=0053 gs=002b efl=00000202 setup+0x103d2d: 00503d2d 8b461b mov eax,dword ptr [esi+1Bh] ds:002b:03b6b9e4=03cac431 0:000> !teb TEB at 00384000 ExceptionList: 013effcc StackBase: 013f0000 StackLimit: 013e9000 ``` Likely Windows 10 has a guard page larger than previous versions. This fixes the stack allocations, and as soon as I get the time I will look at the liveliness analysis. I find it highly unlikely that simple Cabal function requires ~2200 spill slots. Test Plan: ./validate Reviewers: simonmar, bgamari Reviewed By: bgamari Subscribers: AndreasK, rwbarton, thomie, carter GHC Trac Issues: #15154 Differential Revision: https://phabricator.haskell.org/D4917
View Lorry Controller Status