[Git][ghc/ghc][wip/ncg-simd] 62 commits: rts: fix checkClosure error message
sheaf (@sheaf)
gitlab at gitlab.haskell.org
Tue Sep 17 12:25:30 UTC 2024
sheaf pushed to branch wip/ncg-simd at Glasgow Haskell Compiler / GHC
Commits:
0d3bc2fa by Cheng Shao at 2024-09-04T07:20:06-04:00
rts: fix checkClosure error message
This patch fixes an error message in checkClosure() when the closure
has already been evacuated. The previous logic was meant to print the
evacuated closure's type in the error message, but it was completely
wrong, given info was not really an info table, but a tagged pointer
that points to the closure's new address.
- - - - -
fb0a4e5c by Sven Tennie at 2024-09-04T07:20:43-04:00
MO_AcquireFence: Less restrictive barrier
GCC and CLang translate the built-in `atomic_thread_fence(memory_order_acquire)`
to `dmb ishld`, which is a bit less restrictive than `dmb ish` (which
also implies stores.)
- - - - -
a45f1488 by Fendor at 2024-09-04T20:22:00-04:00
testsuite: Add support to capture performance metrics via 'perf'
Performance metrics collected via 'perf' can be more accurate for
run-time performance than GHC's rts, due to the usage of hardware
counters.
We allow performance tests to also record PMU events according to 'perf
list'.
- - - - -
ce61fca5 by Fendor at 2024-09-04T20:22:00-04:00
gitlab-ci: Add nightly job for running the testsuite with perf profiling support
- - - - -
6dfb9471 by Fendor at 2024-09-04T20:22:00-04:00
Enable perf profiling for compiler performance tests
- - - - -
da306610 by sheaf at 2024-09-04T20:22:41-04:00
RecordCon lookup: don't allow a TyCon
This commit adds extra logic when looking up a record constructor.
If GHC.Rename.Env.lookupOccRnConstr returns a TyCon (as it may, due to
the logic explained in Note [Pattern to type (P2T) conversion]),
we emit an error saying that the data constructor is not in scope.
This avoids the compiler falling over shortly thereafter, in the call to
'lookupConstructorInfo' inside 'GHC.Rename.Env.lookupRecFieldOcc',
because the record constructor would not have been a ConLike.
Fixes #25056
- - - - -
9c354beb by Matthew Pickering at 2024-09-04T20:23:16-04:00
Use deterministic names for temporary files
When there are multiple threads they can race to create a temporary
file, in some situations the thread will create ghc_1.c and in some it
will create ghc_2.c. This filename ends up in the debug info for object
files after compiling a C file, therefore contributes to object
nondeterminism.
In order to fix this we store a prefix in `TmpFs` which serves to
namespace temporary files. The prefix is populated from the counter in
TmpFs when the TmpFs is forked. Therefore the TmpFs must be forked
outside the thread which consumes it, in a deterministic order, so each
thread always receives a TmpFs with the same prefix.
This assumes that after the initial TmpFs is created, all other TmpFs
are created from forking the original TmpFs. Which should have been try
anyway as otherwise there would be file collisions and non-determinism.
Fixes #25224
- - - - -
59906975 by Hécate Kleidukos at 2024-09-05T10:57:15-04:00
Silence x-partial in Haddock.Backends.Xhtml
This is an unfortunate consequence of two mechanisms:
* GHC provides (possibly-empty) lists of names
* The functions that retrieve those names are not equipped to do error
reporting, and thus accept these lists at face value. They will have
to be attached an effect for error reporting in a later refactoring
- - - - -
8afbab62 by Hécate Kleidukos at 2024-09-05T10:57:15-04:00
hadrian: Support loading haddock in ghci
There is one tricky aspect with wired-in packages where the boot package
is built with `-this-unit-id ghc` but the dependency is reported as
`-package-id ghc-9.6...`. This has never been fixed in GHC as the
situation of loading wired-in packages into the multi-repl seems like
quite a niche feature that is always just easier to workaround.
- - - - -
6cac9eb8 by Matthew Pickering at 2024-09-05T10:57:15-04:00
hadrian/multi: Load all targets when ./hadrian/ghci-multi is called
This seems to make a bit more sense than just loading `ghc` component
(and dependencies).
- - - - -
7d84df86 by Matthew Pickering at 2024-09-05T10:57:51-04:00
ci: Beef up determinism interface test
There have recently been some determinism issues with the simplifier and
documentation. We enable more things to test in the ABI test to check
that we produce interface files deterministically.
- - - - -
5456e02e by Sylvain Henry at 2024-09-06T11:57:01+02:00
Transform some StgRhsClosure into StgRhsCon after unarisation (#25166)
Before unarisation we may have code like:
Test.foo :: Test.D
[GblId, Unf=OtherCon []] =
\u []
case (# |_| #) [GHC.Types.(##)] of sat_sAw [Occ=Once1] {
__DEFAULT -> Test.D [GHC.Types.True sat_sAw];
};
After unarisation we get:
Test.foo :: Test.D
[GblId, Unf=OtherCon []] =
{} \u [] Test.D [GHC.Types.True 2#];
Notice that it's still an Updatable closure for no reason anymore. This
patch transforms appropriate StgRhsClosures into StgRhsCons after
unarisation, allowing these closures to be statically allocated. Now we
get the expected:
Test.foo :: Test.D
[GblId, Unf=OtherCon []] =
Test.D! [GHC.Types.True 2#];
Fix #25166
To avoid duplicating code, this patch refactors the mk(Top)StgRhs
functions and put them in a GHC.Stg.Make module alongside the new
mk(Top)StgRhsCon_maybe functions.
- - - - -
958b4518 by Hécate Kleidukos at 2024-09-06T16:40:56-04:00
haddock: Add missing requirements.txt for the online manual
- - - - -
573f9833 by Sven Tennie at 2024-09-08T09:58:21+00:00
AArch64: Implement takeRegRegMoveInstr
This has likely been forgotten.
- - - - -
20b0de7d by Hécate Kleidukos at 2024-09-08T14:19:28-04:00
haddock: Configuration fix for ReadTheDocs
- - - - -
03055c71 by Sylvain Henry at 2024-09-09T14:58:15-04:00
JS: fake support for native adjustors (#25159)
The JS backend doesn't support adjustors (I believe) and in any case if
it ever supports them it will be a native support, not one via libffi.
- - - - -
5bf0e6bc by Sylvain Henry at 2024-09-09T14:58:56-04:00
JS: remove redundant h$lstat
It was introduced a second time by mistake in
27dceb42376c34b99a38e36a33b2abc346ed390f (cf #25190)
- - - - -
ffbc2ab0 by Simon Peyton Jones at 2024-09-10T00:40:37-04:00
Refactor only newSysLocalDs
* Change newSysLocalDs to take a scaled type
* Add newSysLocalMDs that takes a type and makes a ManyTy local
Lots of files touched, nothing deep.
- - - - -
7124e4ad by Simon Peyton Jones at 2024-09-10T00:40:37-04:00
Don't introduce 'nospec' on the LHS of a RULE
This patch address #25160. The main payload is:
* When desugaring the LHS of a RULE, do not introduce the `nospec` call
for non-canonical evidence. See GHC.Core.InstEnv
Note [Coherence and specialisation: overview]
The `nospec` call usually introdued in `dsHsWrapper`, but we don't want it
on the LHS of a RULE (that's what caused #25160). So now `dsHsWrapper` takes
a flag to say if it's on the LHS of a RULE. See wrinkle (NC1) in
`Note [Desugaring non-canonical evidence]` in GHC.HsToCore.Binds.
But I think this flag will go away again when I have finished with my
(entirely separate) speciaise-on-values patch (#24359).
All this meant I had to re-understand the `nospec` stuff and coherence, and
that in turn made me do some refactoring, and add a lot of new documentation
The big change is that in GHC.Core.InstEnv, I changed
the /type synonym/ `Canonical` into
a /data type/ `CanonicalEvidence`
and documented it a lot better.
That in turn made me realise that CalLStacks were being treated with a
bit of a hack, which I documented in `Note [CallStack and ExecptionContext hack]`.
- - - - -
663daf8d by Simon Peyton Jones at 2024-09-10T00:40:37-04:00
Add defaulting of equalities
This MR adds one new defaulting strategy to the top-level
defaulting story: see Note [Defaulting equalities] in GHC.Tc.Solver.
This resolves #25029 and #25125, which showed that users were
accidentally relying on a GHC bug, which was fixed by
commit 04f5bb85c8109843b9ac2af2a3e26544d05e02f4
Author: Simon Peyton Jones <simon.peytonjones at gmail.com>
Date: Wed Jun 12 17:44:59 2024 +0100
Fix untouchability test
This MR fixes #24938. The underlying problem was tha the test for
"does this implication bring in scope any equalities" was plain wrong.
This fix gave rise to a number of user complaints; but the improved
defaulting story of this MR largely resolves them.
On the way I did a bit of refactoring, of course
* Completely restructure the extremely messy top-level defaulting
code. The new code is in GHC.Tc.Solver.tryDefaulting, and is much,
much, much esaier to grok.
- - - - -
e28cd021 by Andrzej Rybczak at 2024-09-10T00:41:18-04:00
Don't name a binding pattern
It's a keyword when PatternSynonyms are set.
- - - - -
b09571e2 by Simon Peyton Jones at 2024-09-10T00:41:54-04:00
Do not use an error thunk for an absent dictionary
In worker/wrapper we were using an error thunk for an absent dictionary,
but that works very badly for -XDictsStrict, or even (as #24934 showed)
in some complicated cases involving strictness analysis and unfoldings.
This MR just uses RubbishLit for dictionaries. Simple.
No test case, sadly because our only repro case is rather complicated.
- - - - -
8bc9f5f6 by Hécate Kleidukos at 2024-09-10T00:42:34-04:00
haddock: Remove support for applehelp format in the Manual
- - - - -
9ca15506 by doyougnu at 2024-09-10T10:46:38-04:00
RTS linker: add support for hidden symbols (#25191)
Add linker support for hidden symbols. We basically treat them as weak
symbols.
Patch upstreamed from haskell.nix
Co-authored-by: Sylvain Henry <sylvain at haskus.fr>
Co-authored-by: Moritz Angermann <moritz.angermann at gmail.com>
- - - - -
3b2dc826 by Sven Tennie at 2024-09-10T10:47:14-04:00
Fix C warnings (#25237)
GCC 14 treats the fixed warnings as errors by default. I.e. we're
gaining GCC 14 compatibility with these fixes.
- - - - -
05715994 by Sylvain Henry at 2024-09-10T10:47:55-04:00
JS: fix codegen of static string data
Before this patch, when string literals are made trivial, we would
generate `h$("foo")` instead of `h$str("foo")`. This was
introduced by mistake in 6bd850e887b82c5a28bdacf5870d3dc2fc0f5091.
- - - - -
949ebced by Hécate Kleidukos at 2024-09-10T19:19:40-04:00
haddock: Re-organise cross-OS compatibility layer
- - - - -
84ac9a99 by Hécate Kleidukos at 2024-09-10T19:19:40-04:00
haddock: Remove CPP for obsolete GHC and Cabal versions
- - - - -
370d1599 by Hécate Kleidukos at 2024-09-10T19:19:40-04:00
haddock: Move the changelog file to the 'extra-doc-files' section in the cabal file
- - - - -
cfbff65a by Simon Peyton Jones at 2024-09-10T19:20:16-04:00
Add ZonkAny and document it
This MR fixed #24817 by adding ZonkAny, which takes a Nat
argument.
See Note [Any types] in GHC.Builtin.Types, especially
wrinkle (Any4).
- - - - -
0167e472 by Matthew Pickering at 2024-09-11T02:41:42-04:00
hadrian: Make sure ffi headers are built before using a compiler
When we are using ffi adjustors then we rely on `ffi.h` and
`ffitarget.h` files during code generation when compiling stubs.
Therefore we need to add this dependency to the build system (which this
patch does).
Reproducer, configure with `--enable-libffi-adjustors` and then build
"_build/stage1/libraries/ghc-prim/build/GHC/Types.p_o".
Observe that this fails before this patch and works afterwards.
Fixes #24864
Co-authored-by: Sylvain Henry <sylvain at haskus.fr>
- - - - -
0f696958 by Rodrigo Mesquita at 2024-09-11T02:42:18-04:00
base: Deprecate BCO primops exports from GHC.Exts
See https://github.com/haskell/core-libraries-committee/issues/212.
These reexports will be removed in GHC 9.14.
- - - - -
cf0e7729 by Alan Zimmerman at 2024-09-11T02:42:54-04:00
EPA: Remove Anchor = EpaLocation synonym
This just causes confusion.
- - - - -
8e462f4d by Andrew Lelechenko at 2024-09-11T22:20:37-04:00
Bump submodule deepseq to 1.5.1.0
- - - - -
aa4500ae by Sebastian Graf at 2024-09-11T22:21:13-04:00
User's guide: Fix the "no-backtracking" example of -XOrPatterns (#25250)
Fixes #25250.
- - - - -
1c479c01 by Sven Tennie at 2024-09-12T10:39:38+00:00
RISCV64: Add Native Code Generator (NCG)
This architecture wasn't supported before.
Co-authored-by: Moritz Angermann <moritz.angermann at gmail.com>
- - - - -
51b678e1 by Sven Tennie at 2024-09-12T10:39:38+00:00
Adjust test timings for slower computers
Increase the delays a bit to be able to run these tests on slower
computers.
The reference was a Lichee Pi 4a RISCV64 machine.
- - - - -
a0e41741 by Sven Tennie at 2024-09-12T10:39:38+00:00
RISCV64: Add RTS linker
This architecture wasn't supported before.
Co-authored-by: Moritz Angermann <moritz.angermann at gmail.com>
- - - - -
d365b1d4 by Sven Tennie at 2024-09-12T10:39:38+00:00
RISCV64: Ignore divbyzero test
The architecture's behaviour differs from the test's expectations. See
comment in code why this is okay.
- - - - -
abf3d699 by Sven Tennie at 2024-09-12T10:39:38+00:00
RISCV64: Enable MulMayOflo_full test
It works and thus can be tested.
- - - - -
38c7ea8c by Sven Tennie at 2024-09-12T10:39:38+00:00
RISCV64: LibffiAdjustor: Ensure code caches are flushed
RISCV64 needs a specific code flushing sequence (involving fence.i) when
new code is created/loaded.
- - - - -
7edc6965 by Sven Tennie at 2024-09-12T10:39:38+00:00
RISCV64: Add additional linker symbols for builtins
We're relying on some GCC/Clang builtins. These need to be visible to
the linker (and not be stripped away.)
- - - - -
92ad3d42 by Sven Tennie at 2024-09-12T10:39:38+00:00
RISCV64: Add GHCi support
As we got a RTS linker for this architecture now, we can enable GHCi for
it.
- - - - -
a145f701 by Sven Tennie at 2024-09-12T10:39:38+00:00
RISCV64: Set codeowners of the NCG
- - - - -
8e6d58cf by Sven Tennie at 2024-09-12T10:39:38+00:00
Add test for C calling convention
Ensure that parameters and return values are correctly processed. A
dedicated test (like this) helps to get the subtleties of calling
conventions easily right.
The test is failing for WASM32 and marked as fragile to not forget to
investigate this (#25249).
- - - - -
fff55592 by Torsten Schmits at 2024-09-12T21:50:34-04:00
finder: Add `IsBootInterface` to finder cache keys
- - - - -
cdf530df by Alan Zimmerman at 2024-09-12T21:51:10-04:00
EPA: Sync ghc-exactprint to GHC
- - - - -
1374349b by Sebastian Graf at 2024-09-13T07:52:11-04:00
DmdAnal: Fast path for `multDmdType` (#25196)
This is in order to counter a regression exposed by SpecConstr.
Fixes #25196.
- - - - -
80769bc9 by Andrew Lelechenko at 2024-09-13T07:52:47-04:00
Bump submodule array to 0.5.8.0
- - - - -
49ac3fb8 by Sylvain Henry at 2024-09-16T10:33:01-04:00
Linker: add support for extra built-in symbols (#25155)
See added Note [Extra RTS symbols] and new user guide entry.
Co-authored-by: Hamish Mackenzie <Hamish.K.Mackenzie at gmail.com>
Co-authored-by: Moritz Angermann <moritz.angermann at gmail.com>
- - - - -
185320a3 by sheaf at 2024-09-17T14:24:31+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
-------------------------
- - - - -
b28f75a1 by sheaf at 2024-09-17T14:24:34+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
- - - - -
df6d94f7 by sheaf at 2024-09-17T14:24:34+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.
- - - - -
91458a88 by sheaf at 2024-09-17T14:24:35+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.
- - - - -
9b74fe13 by sheaf at 2024-09-17T14:24:35+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.
- - - - -
bcc6b168 by sheaf at 2024-09-17T14:24:35+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.
- - - - -
7e48a328 by sheaf at 2024-09-17T14:24:36+02:00
Add min/max primops
This commit adds min/max primops, such as
minDouble# :: Double# -> Double# -> Double#
minFloatX4# :: FloatX4# -> FloatX4# -> FloatX4#
minWord16X8# :: Word16X8# -> Word16X8# -> Word16X8#
These are supported in:
- the X86, AArch64 and PowerPC NCGs,
- the LLVM backend,
- the WebAssembly and JavaScript backends.
Fixes #25120
- - - - -
8573fa7d by sheaf at 2024-09-17T14:24:36+02:00
Add test for C calls & SIMD vectors
- - - - -
548a26b1 by sheaf at 2024-09-17T14:24:36+02:00
Add test for #25169
- - - - -
480c07dc by sheaf at 2024-09-17T14:24:36+02:00
Fix #25169 using Plan A from the ticket
We now compile certain low-level Cmm functions in the RTS multiple
times, with different levels of vector support. We then dispatch
at runtime in the RTS, based on what instructions are supported.
See Note [realArgRegsCover] in GHC.Cmm.CallConv.
Fixes #25169
-------------------------
Metric Increase:
T10421
T12425
T1969
T9198
-------------------------
- - - - -
ac8481e8 by sheaf at 2024-09-17T14:25:15+02:00
Fix C calls with SIMD vectors
This commit fixes the code generation for C calls, to take into account
the calling convention.
This is particularly tricky on Windows, where all vectors are expected
to be passed by reference. See Note [The Windows X64 C calling convention]
in GHC.CmmToAsm.X86.CodeGen.
- - - - -
acf56839 by sheaf at 2024-09-17T14:25:15+02:00
LLVM: propagate GlobalRegUse information
This commit ensures we keep track of how any particular global register
is being used in the LLVM backend. This informs the LLVM type
annotations, and avoids type mismatches of the following form:
argument is not of expected type '<2 x double>'
call ccc <2 x double> (<2 x double>)
(<4 x i32> arg)
- - - - -
30 changed files:
- .gitlab/ci.sh
- .gitlab/generate-ci/gen_ci.hs
- .gitlab/jobs.yaml
- CODEOWNERS
- compiler/CodeGen.Platform.h
- compiler/GHC/Builtin/Names.hs
- compiler/GHC/Builtin/Types.hs
- compiler/GHC/Builtin/primops.txt.pp
- compiler/GHC/ByteCode/Asm.hs
- compiler/GHC/Cmm.hs
- compiler/GHC/Cmm/CLabel.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
The diff was not included because it is too large.
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/086ca50e7d242fcc8102df437637496b1dfb4136...acf568390a4296b44032a5e6264be016ac6aef5a
--
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/086ca50e7d242fcc8102df437637496b1dfb4136...acf568390a4296b44032a5e6264be016ac6aef5a
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/20240917/0e27713d/attachment-0001.html>
More information about the ghc-commits
mailing list