[Git][ghc/ghc][wip/ncg-simd] 45 commits: compiler: Turn `FinderCache` into a record of operations so that GHC API clients can
sheaf (@sheaf)
gitlab at gitlab.haskell.org
Tue Jul 9 10:44:53 UTC 2024
sheaf pushed to branch wip/ncg-simd at Glasgow Haskell Compiler / GHC
Commits:
0d170eaf by Zubin Duggal at 2024-07-04T11:08:41-04:00
compiler: Turn `FinderCache` into a record of operations so that GHC API clients can
have full control over how its state is managed by overriding `hsc_FC`.
Also removes the `uncacheModule` function as this wasn't being used by anything
since 1893ba12fe1fa2ade35a62c336594afcd569736e
Fixes #23604
- - - - -
4664997d by Teo Camarasu at 2024-07-04T11:09:18-04:00
Add HasCallStack to T23221
This makes the test a bit easier to debug
- - - - -
66919dcc by Teo Camarasu at 2024-07-04T11:09:18-04:00
rts: use live words to estimate heap size
We use live words rather than live blocks to determine the size of the
heap for determining memory retention.
Most of the time these two metrics align, but they can come apart in
normal usage when using the nonmoving collector.
The nonmoving collector leads to a lot of partially occupied blocks. So,
using live words is more accurate.
They can also come apart when the heap is suffering from high levels
fragmentation caused by small pinned objects, but in this case, the
block size is the more accurate metric. Since this case is best avoided
anyway. It is ok to accept the trade-off that we might try (and
probably) fail to return more memory in this case.
See also the Note [Statistics for retaining memory]
Resolves #23397
- - - - -
8dfca66a by Oleg Grenrus at 2024-07-04T11:09:55-04:00
Add reflections of GHC.TypeLits/Nats type families
-------------------------
Metric Increase:
ghc_experimental_dir
ghc_experimental_so
-------------------------
- - - - -
6c469bd2 by Adam Gundry at 2024-07-04T11:10:33-04:00
Correct -Wpartial-fields warning to say "Definition" rather than "Use"
Fixes #24710. The message and documentation for `-Wpartial-fields` were
misleading as (a) the warning occurs at definition sites rather than use
sites, and (b) the warning relates to the definition of a field independently
of the selector function (e.g. because record updates are also partial).
- - - - -
977b6b64 by Max Ulidtko at 2024-07-04T11:11:11-04:00
GHCi: Support local Prelude
Fixes #10920, an issue where GHCi bails out when started alongside a
file named Prelude.hs or Prelude.lhs (even empty file suffices).
The in-source Note [GHCi and local Preludes] documents core reasoning.
Supplementary changes:
* add debug traces for module lookups under -ddump-if-trace;
* drop stale comment in GHC.Iface.Load;
* reduce noise in -v3 traces from GHC.Utils.TmpFs;
* new test, which also exercizes HomeModError.
- - - - -
87cf4111 by Ryan Scott at 2024-07-04T11:11:47-04:00
Add missing gParPat in cvtp's ViewP case
When converting a `ViewP` using `cvtp`, we need to ensure that the view pattern
is parenthesized so that the resulting code will parse correctly when
roundtripped back through GHC's parser.
Fixes #24894.
- - - - -
b05613c5 by Adam Gundry at 2024-07-04T11:12:23-04:00
Use structured error representation for module cycle errors (see #18516)
This removes the re-export of cyclicModuleErr from the top-level GHC module.
- - - - -
70389749 by Adam Gundry at 2024-07-04T11:12:23-04:00
Use structured error representation when reloading a nonexistent module
- - - - -
680ade3d by sheaf at 2024-07-04T11:12:23-04:00
Use structured errors for a Backpack instantiation error
- - - - -
97c6d6de by sheaf at 2024-07-04T11:12:23-04:00
Move mkFileSrcSpan to GHC.Unit.Module.Location
- - - - -
f9e7bd9b by Adriaan Leijnse at 2024-07-04T11:12:59-04:00
ttg: Remove SourceText from OverloadedLabel
Progress towards #21592
- - - - -
00d63245 by Alexander Foremny at 2024-07-04T11:12:59-04:00
AST: GHC.Prelude -> Prelude
Refactor occurrences to GHC.Prelude with Prelude within
Language/Haskell.
Progress towards #21592
- - - - -
cc846ea5 by Alexander Foremny at 2024-07-04T11:12:59-04:00
AST: remove occurrences of GHC.Unit.Module.ModuleName
`GHC.Unit.Module` re-exports `ModuleName` from
`Language.Haskell.Syntax.Module.Name`.
Progress towards #21592
- - - - -
24c7d287 by Fabian Kirchner at 2024-07-04T11:12:59-04:00
AST: move Data instance definition for ModuleName to GHC.Unit.Types
To remove the dependency on GHC.Utils.Misc inside
Language.Haskell.Syntax.Module.Name, the instance definition is moved
from there into GHC.Unit.Types.
Progress towards #21592
- - - - -
6cbba381 by Fabian Kirchner at 2024-07-04T11:12:59-04:00
AST: move negateOverLitVal into GHC.Hs.Lit
The function negateOverLitVal is not used within Language.Haskell and
therefore can be moved to the respective module inside GHC.Hs.
Progress towards #21592
- - - - -
611aa7c6 by Fabian Kirchner at 2024-07-04T11:12:59-04:00
AST: move conDetailsArity into GHC.Rename.Module
The function conDetailsArity is only used inside GHC.Rename.Module. We
therefore move it there from Language.Haskell.Syntax.Lit.
Progress towards #21592
- - - - -
1b968d16 by Mauricio at 2024-07-04T11:12:59-04:00
AST: Remove GHC.Utils.Assert from GHC
Simple cleanup.
Progress towards #21592
- - - - -
3d192e5d by Fabian Kirchner at 2024-07-04T11:12:59-04:00
ttg: extract Specificity, ForAllTyFlag and helper functions from GHC.Types.Var
Progress towards #21592
Specificity, ForAllTyFlag and its' helper functions are extracted from
GHC.Types.Var and moved into a new module Language.Haskell.Syntax.Specificity.
Note: Eventually (i.e. after Language.Haskell.Syntax.Decls does not depend on
GHC.* anymore) these should be moved into Language.Haskell.Syntax.Decls.
At this point, this would cause cyclic dependencies.
- - - - -
257d1adc by Adowrath at 2024-07-04T11:12:59-04:00
ttg: Split HsSrcBang, remove ref to DataCon from Syntax.Type
Progress towards #21592
This splits HsSrcBang up, creating the new HsBang within
`Language.Haskell.Syntax.Basic`. `HsBang` holds the unpackedness
and strictness information, while `HsSrcBang` only adds the SourceText
for usage within the compiler directly.
Inside the AST, to preserve the SourceText, it is hidden behind the
pre-existing extension point `XBindTy`. All other occurrences of
`HsSrcBang` were adapted to deconstruct the inner `HsBang`, and when
interacting with the `BindTy` constructor, the hidden `SourceText` is
extracted/inserted into the `XBindTy` extension point.
`GHC.Core.DataCon` exports both `HsSrcBang` and `HsBang` for
convenience. A constructor function `mkHsSrcBang` that takes all
individual components has been added.
Two exceptions has been made though:
- The `Outputable HsSrcBang` instance is replaced by
`Outputable HsBang`. While being only GHC-internal, the only place
it's used is in outputting `HsBangTy` constructors -- which already
have `HsBang`. It wouldn't make sense to reconstruct a `HsSrcBang` just
to ignore the `SourceText` anyway.
- The error `TcRnUnexpectedAnnotation` did not use the `SourceText`,
so it too now only holds a `HsBang`.
- - - - -
24757fec by Mauricio at 2024-07-04T11:12:59-04:00
AST: Moved definitions that use GHC.Utils.Panic to GHC namespace
Progress towards #21592
- - - - -
9be49379 by Mike Pilgrem at 2024-07-04T11:13:41-04:00
Fix #25032 Refer to Cabal's `includes` field, not `include-files`
- - - - -
9e2ecf14 by Andrew Lelechenko at 2024-07-04T11:14:17-04:00
base: fix more missing changelog entries
- - - - -
a82121b3 by Peter Trommler at 2024-07-04T11:14:53-04:00
X86 NCG: Fix argument promotion in foreign C calls
Promote 8 bit and 16 bit signed arguments by sign extension.
Fixes #25018
- - - - -
fab13100 by Bryan Richter at 2024-07-04T11:15:29-04:00
Add .gitlab/README.md with creds instructions
- - - - -
564981bd by Matthew Pickering at 2024-07-05T07:35:29-04:00
configure: Set LD_STAGE0 appropiately when 9.10.1 is used as a boot compiler
In 9.10.1 the "ld command" has been removed, so we fall back to using
the more precise "merge objects command" when it's available as
LD_STAGE0 is only used to set the object merging command in hadrian.
Fixes #24949
- - - - -
a949c792 by Matthew Pickering at 2024-07-05T07:35:29-04:00
hadrian: Don't build ghci object files for ./hadrian/ghci target
There is some convoluted logic which determines whether we build ghci
object files are not. In any case, if you set `ghcDynPrograms = pure
False` then it forces them to be built.
Given we aren't ever building executables with this flavour it's fine
to leave `ghcDynPrograms` as the default and it should be a bit faster
to build less.
Also fixes #24949
- - - - -
48bd8f8e by Matthew Pickering at 2024-07-05T07:36:06-04:00
hadrian: Remove STG dump from ticky_ghc flavour transformer
This adds 10-15 minutes to build time, it is a better strategy to
precisely enable dumps for the modules which show up prominently in a
ticky profile.
Given I am one of the only people regularly building ticky compilers I
think it's worthwhile to remove these.
Fixes #23635
- - - - -
5b1aefb7 by Matthew Pickering at 2024-07-05T07:36:06-04:00
hadrian: Add dump_stg flavour transformer
This allows you to write `--flavour=default+ticky_ghc+dump_stg` if you
really want STG for all modules.
- - - - -
ab2b60b6 by Sven Tennie at 2024-07-08T15:03:41-04:00
AArch64: Simplify stmtToInstrs type
There's no need to hand `Nothing`s around... (there was no case with a
`BlockId`.)
- - - - -
71a7fa8c by Sven Tennie at 2024-07-08T15:03:41-04:00
AArch64: Simplify stmtsToInstrs type
The `BlockId` parameter (`bid`) is never used, only handed around.
Deleting it simplifies the surrounding code.
- - - - -
8bf6fd68 by Simon Peyton Jones at 2024-07-08T15:04:17-04:00
Fix eta-expansion in Prep
As #25033 showed, we were eta-expanding in a way that broke a join point,
which messed up Note [CorePrep invariants].
The fix is rather easy. See Wrinkle (EA1) of
Note [Eta expansion of arguments in CorePrep]
- - - - -
96acf823 by Sjoerd Visscher at 2024-07-09T06:16:14-04:00
One-shot Haddock
- - - - -
74ec4c06 by Sjoerd Visscher at 2024-07-09T06:16:14-04:00
Remove haddock-stdout test option
Superseded by output handling of Hadrian
- - - - -
ed8a8f0b by Rodrigo Mesquita at 2024-07-09T06:16:51-04:00
ghc-boot: Relax Cabal bound
Fixes #25013
- - - - -
3f9548fe by Matthew Pickering at 2024-07-09T06:17:36-04:00
ci: Unset ALEX/HAPPY variables when testing bootstrap jobs
Ticket #24826 reports a regression in 9.10.1 when building from a source
distribution. This patch is an attempt to reproduce the issue on CI by
more aggressively removing `alex` and `happy` from the environment.
- - - - -
aba2c9d4 by Andrea Bedini at 2024-07-09T06:17:36-04:00
hadrian: Ignore build-tool-depends fields in cabal files
hadrian does not utilise the build-tool-depends fields in cabal files
and their presence can cause issues when building source distribution
(see #24826)
Ideally Cabal would support building "full" source distributions which
would remove the need for workarounds in hadrian but for now we can
patch the build-tool-depends out of the cabal files.
Fixes #24826
- - - - -
12bb9e7b by Matthew Pickering at 2024-07-09T06:18:12-04:00
testsuite: Don't attempt to link when checking whether a way is supported
It is sufficient to check that the simple test file compiles as it will
fail if there are not the relevant library files for the requested way.
If you break a way so badly that even a simple executable fails to link
(as I did for profiled dynamic way), it will just mean the tests for
that way are skipped on CI rather than displayed.
- - - - -
2ff70ba0 by sheaf at 2024-07-09T12:43:32+02:00
Testsuite: use py-cpuinfo to compute CPU features
This replaces the rather hacky logic we had in place for checking
CPU features. In particular, this means that feature availability now
works properly on Windows.
- - - - -
61fc81ac by sheaf at 2024-07-09T12:44:05+02:00
The X86 SIMD patch.
This commit adds support for 128 bit wide SIMD vectors and vector
operations to GHC's X86 native code generator.
Main changes:
- Introduction of vector formats (`GHC.CmmToAsm.Format`)
- Introduction of 128-bit virtual register (`GHC.Platform.Reg`),
and removal of unused Float virtual register.
- Refactor of `GHC.Platform.Reg.Class.RegClass`: it now only contains
two classes, `RcInteger` (for general purpose registers) and `RcFloatOrVector`
(for registers that can be used for scalar floating point values as well
as vectors).
- Modify `GHC.CmmToAsm.X86.Instr.regUsageOfInstr` to keep track
of which format each register is used at, so that the register
allocator can know if it needs to spill the entire vector register
or just the lower 64 bits.
- Modify spill/load/reg-2-reg code to account for vector registers
(`GHC.CmmToAsm.X86.Instr.{mkSpillInstr, mkLoadInstr, mkRegRegMoveInstr, takeRegRegMoveInstr}`).
- Modify the register allocator code (`GHC.CmmToAsm.Reg.*`) to propagate
the format we are storing in any given register, for instance changing
`Reg` to `RegFormat` or `GlobalReg` to `GlobalRegUse`.
- Add logic to lower vector `MachOp`s to X86 assembly
(see `GHC.CmmToAsm.X86.CodeGen`)
- Minor cleanups to genprimopcode, to remove the llvm_only attribute
which is no longer applicable.
Tests for this feature are provided in the "testsuite/tests/simd" directory.
Fixes #7741
Keeping track of register formats adds a small memory overhead to the
register allocator (in particular, regUsageOfInstr now allocates more
to keep track of the `Format` each register is used at). This explains
the following metric increases.
-------------------------
Metric Increase:
T12707
T13035
T13379
T3294
T4801
T5321FD
T5321Fun
T783
-------------------------
- - - - -
de5a3e48 by sheaf at 2024-07-09T12:44:08+02:00
Add vector fused multiply-add operations
This commit adds fused multiply add operations such as `fmaddDoubleX2#`.
These are handled both in the X86 NCG and the LLVM backends.
- - - - -
c218d2ff by sheaf at 2024-07-09T12:44:09+02:00
Add vector shuffle primops
This adds vector shuffle primops, such as
```
shuffleFloatX4# :: FloatX4# -> FloatX4# -> (# Int#, Int#, Int#, Int# #) -> FloatX4#
```
which shuffle the components of the input two vectors into the output vector.
NB: the indices must be compile time literals, to match the X86 SHUFPD
instruction immediate and the LLVM shufflevector instruction.
These are handled in the X86 NCG and the LLVM backend.
Tested in simd009.
- - - - -
cbfd1bb8 by sheaf at 2024-07-09T12:44:09+02:00
Add Broadcast MachOps
This adds proper MachOps for broadcast instructions, allowing us to
produce better code for broadcasting a value than simply packing that
value (doing many vector insertions in a row).
These are lowered in the X86 NCG and LLVM backends. In the LLVM backend,
it uses the previously introduced shuffle instructions.
- - - - -
d2080986 by sheaf at 2024-07-09T12:44:09+02:00
Fix treatment of signed zero in vector negation
This commit fixes the handling of signed zero in floating-point vector
negation.
A slight hack was introduced to work around the fact that Cmm doesn't
currently have a notion of signed floating point literals
(see get_float_broadcast_value_reg). This can be removed once CmmFloat
can express the value -0.0.
The simd006 test has been updated to use a stricter notion of equality
of floating-point values, which ensure the validity of this change.
- - - - -
87939b37 by sheaf at 2024-07-09T12:44:09+02:00
Use xmm registers in genapply
This commit updates genapply to use xmm, ymm and zmm registers, for
stg_ap_v16/stg_ap_v32/stg_ap_v64, respectively.
It also updates the Cmm lexer and parser to produce Cmm vectors rather
than 128/256/512 bit wide scalars for V16/V32/V64, removing bits128,
bits256 and bits512 in favour of vectors.
The Cmm Lint check is weakened for vectors, as (in practice, e.g. on X86)
it is okay to use a single vector register to hold multiple different
types of data, and we don't know just from seeing e.g. "XMM1" how to
interpret the 128 bits of data within.
Fixes #25062
- - - - -
30 changed files:
- .gitlab-ci.yml
- + .gitlab/README.md
- .gitlab/ci.sh
- compiler/GHC.hs
- compiler/GHC/Builtin/Types.hs
- compiler/GHC/Builtin/primops.txt.pp
- compiler/GHC/ByteCode/Asm.hs
- compiler/GHC/Cmm.hs
- compiler/GHC/Cmm/CallConv.hs
- compiler/GHC/Cmm/Graph.hs
- compiler/GHC/Cmm/Lexer.x
- compiler/GHC/Cmm/Lint.hs
- compiler/GHC/Cmm/Liveness.hs
- compiler/GHC/Cmm/MachOp.hs
- compiler/GHC/Cmm/Node.hs
- compiler/GHC/Cmm/Opt.hs
- compiler/GHC/Cmm/Parser.y
- compiler/GHC/Cmm/ProcPoint.hs
- compiler/GHC/Cmm/Reg.hs
- compiler/GHC/Cmm/Sink.hs
- compiler/GHC/Cmm/Type.hs
- compiler/GHC/CmmToAsm.hs
- compiler/GHC/CmmToAsm/AArch64.hs
- compiler/GHC/CmmToAsm/AArch64/CodeGen.hs
- compiler/GHC/CmmToAsm/AArch64/Instr.hs
- compiler/GHC/CmmToAsm/AArch64/Ppr.hs
- compiler/GHC/CmmToAsm/AArch64/Regs.hs
- compiler/GHC/CmmToAsm/Config.hs
- compiler/GHC/CmmToAsm/Format.hs
- compiler/GHC/CmmToAsm/Instr.hs
The diff was not included because it is too large.
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/f47ef2d5f999e43af1dcc333142368d91ee5d3af...87939b374863c9c4cb0222521b59b63f622a62de
--
This project does not include diff previews in email notifications.
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/f47ef2d5f999e43af1dcc333142368d91ee5d3af...87939b374863c9c4cb0222521b59b63f622a62de
You're receiving this email because of your account on gitlab.haskell.org.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-commits/attachments/20240709/87997c16/attachment-0001.html>
More information about the ghc-commits
mailing list