[Git][ghc/ghc][wip/ncg-simd] 35 commits: Extend -reexported-module flag to support module renaming
sheaf (@sheaf)
gitlab at gitlab.haskell.org
Tue Aug 27 16:01:49 UTC 2024
sheaf pushed to branch wip/ncg-simd at Glasgow Haskell Compiler / GHC
Commits:
ee0a9c18 by Matthew Pickering at 2024-08-14T14:27:39-04:00
Extend -reexported-module flag to support module renaming
The -reexported-module flag now supports renaming -rexported-modules.
```
-rexported-module "A as B"
```
This feature is only relevant to multi-component sessions.
Fixes #25139
- - - - -
e9496000 by Arnaud Spiwack at 2024-08-14T14:28:20-04:00
Don't restrict eta-reduction of linear functions
This commit simply removes code. All the supporting implementation has
been done as part of !12883.
Closes #25129
- - - - -
2bb4156e by sheaf at 2024-08-14T14:28:56-04:00
Allow @ character in C labels
Generated symbol names can include the '@' character, for example when using
`__attribute__((vectorcall))`.
- - - - -
7602ca23 by Sylvain Henry at 2024-08-14T14:29:36-04:00
Linker: replace blind tuple with a datatype + docs
- - - - -
bdd77b9e by sheaf at 2024-08-16T12:47:11-04:00
isIrrefutableHsPat: look up ConLikes in the HscEnv
At GhcRn stage, in isIrrefutableHsPat we only looked up data constructors
in the RdrEnv, which meant that we lacked fallibility information for
out-of-scope constructors (which can arise from Template Haskell splices).
Instead, we use 'lookupGREInfo', which looks up the information in
the type environment. This was the correct function to call all along,
but was not used in 572fbc44 due to import cycle reasons. The appropriate
functions, 'irrefutableConLike{Rn,Tc}' have been moved to 'GHC.Rename.Env',
which avoids import cycles.
Fixes #25164
- - - - -
4bee377c by Sylvain Henry at 2024-08-16T12:47:53-04:00
Linker: some refactoring to prepare for #24886
- Rename LoadedBCOs into LazyBCOs
- Bundle SptEntries with CompiledByteCode and removed [SptEntry] field
from the BCOs constructor
- Rename Linkable's LM constructor into Linkable: in the past we had LM
and LP for Module and Package, now we only have the former.
- Rename Unlinked into LinkablePart (and linkableUnlinked into
linkableParts)
- Use NonEmpty to encode invariant in Linkable's linkableParts type
- Add helpers: linkableLibs, linkableBCOs, etc.
- Add documentation
- Remove partial nameOfObject
- Rename nameOfObject_maybe into linkablePartPath
- Rename byteCodeOfObject into linkablePartAllBCOs.
- Refactor linkablePartAllBCOs to avoid a panic if a LazyBCO has a C
stub. Document the fact that LazyBCOs are returned in this case
(contrary to linkableBCOs which only returns non-lazy ones)
Refactoring done while trying to understand how to adapt the linker code
to support the JS backend too (cf #24886).
- - - - -
fa0dbaca by Mario Blažević at 2024-08-17T03:31:32+00:00
Implements the Exportable Named Default proposal (#24305)
This squashed commit adds support for exportable named defaults, the accepted
GHC proposal at https://github.com/ghc-proposals/ghc-proposals/pull/409
The proposal extends the Haskell '98 declarations
default (Int, Double)
which were implicitly always applying to Num class alone, to allow specifying
an arbitrary single-parameter class:
default IsString (Text, String)
The effect of this declaration would be to eliminate the ambiguous type errors
around string literals when OverloadedStrings extension is active. The
declaration by itself has effect only in its module, so the proposal also adds
the ability to export class defaults:
module MyModule (default IsIstring)
Once the language extension is published and established, we can consider using
it in base and other libraries.
See Note [Named default declarations] in GHC.Tc.Gen.Default
for implementation details.
- - - - -
1deba6b2 by Simon Peyton Jones at 2024-08-17T13:58:13-04:00
Make kick-out more selective
This MR revised the crucial kick-out criteria in the constraint solver.
Ticket #24984 showed an example in which
* We were kicking out unnecessarily
* That gave rise to extra work, of course
* But it /also/ led to exponentially-sized coercions due to lack
of sharing in coercions (something we want to fix separately #20264)
This MR sharpens up the kick-out criteria; specifially in (KK2) we look
only under type family applications if (fs>=fw).
This forced me to understand the existing kick-out story, and I ended
up rewriting many of the careful Notes in GHC.Tc.Solver.InertSet.
Especially look at the new `Note [The KickOut Criteria]`
The proof of termination is not air-tight, but it is better than before,
and both Richard and I think it's correct :-).
- - - - -
88488847 by Cheng Shao at 2024-08-18T04:44:01+02:00
testsuite: remove undesired -fasm flag from test ways
This patch removes the -fasm flag from test ways, except ways like
optasm that explicitly state they are meant to be compiled with NCG
backend. Most test ways should use the default codegen backend, and
the precense of -fasm can cause stderr mismatches like this when GHC
is configured with the unregisterised backend:
```
--- /dev/null
+++ /tmp/ghctest-3hydwldj/test spaces/testsuite/tests/profiling/should_compile/prof-late-cc.run/prof-late-cc.comp.stderr.normalised
@@ -0,0 +1,2 @@
+when making flags consistent: warning: [GHC-74335] [-Winconsistent-flags (in -Wdefault)]
+ Target platform uses unregisterised ABI, so compiling via C
*** unexpected failure for prof-late-cc(prof_no_auto)
```
This has been breaking the wasm unreg nightly job since !12595 landed.
- - - - -
3a145315 by Cheng Shao at 2024-08-18T13:05:45-04:00
ghci: fix isMinTTY.h casing for Windows targets
This commit fixes isMinTTY.h casing in isMinTTY.c that's compiled for
Windows targets. While this looks harmless given Windows filesystems
are case-insensitive by default, it does cause a compilation warning
with recent versions of clang, so we might as well fix the casing:
```
driver\ghci\isMinTTY.c:10:10: error:
warning: non-portable path to file '"isMinTTY.h"'; specified path differs in case from file name on disk [-Wnonportable-include-path]
|
10 | #include "isMINTTY.h"
| ^
#include "isMINTTY.h"
^~~~~~~~~~~~
"isMinTTY.h"
1 warning generated.
```
- - - - -
5f972bfb by Zubin Duggal at 2024-08-21T03:18:15-04:00
compiler: Fix pretty printing of ticked prefix constructors (#24237)
- - - - -
ef0a08e7 by Mike Pilgrem at 2024-08-21T03:18:57-04:00
Fix #15773 Clarify further -rtsopts 'defaults' in docs
- - - - -
05a4be58 by Sebastian Graf at 2024-08-21T03:19:33-04:00
Improve efficiency of `assertError` (#24625)
... by moving `lazy` to the exception-throwing branch.
It's all documented in `Note [Strictness of assertError]`.
- - - - -
c29b2b5a by sheaf at 2024-08-21T13:11:30-04:00
GHCi debugger: drop record name spaces for Ids
When binding new local variables at a breakpoint, we should create
Ids with variable namespace, and not record field namespace. Otherwise
the rest of the compiler falls over because the IdDetails are wrong.
Fixes #25109
- - - - -
bd82ac9f by Hécate Kleidukos at 2024-08-21T13:12:12-04:00
base: Final deprecation of GHC.Pack
The timeline mandated by #21461 has come to its term and after two years
and four minor releases, we are finally removing GHC.Pack from base.
Closes #21536
- - - - -
5092dbff by Sylvain Henry at 2024-08-21T13:12:54-04:00
JS: support rubbish static literals (#25177)
Support for rubbish dynamic literals was added in #24664. This patch
does the same for static literals.
Fix #25177
- - - - -
b5a2c061 by Phil de Joux at 2024-08-21T13:13:33-04:00
haddock docs: prefix comes before, postfix comes after
- - - - -
6fde3685 by Marcin Szamotulski at 2024-08-21T23:15:39-04:00
haddock: include package info with --show-interface
- - - - -
7e02111b by Andreas Klebinger at 2024-08-21T23:16:15-04:00
Document the (x86) SIMD macros.
Fixes #25021.
- - - - -
05116c83 by Rodrigo Mesquita at 2024-08-22T10:37:44-04:00
ghc-internal: Derive version from ghc's version
Fixes #25005
- - - - -
73f5897d by Ben Gamari at 2024-08-22T10:37:44-04:00
base: Deprecate GHC.Desugar
See https://github.com/haskell/core-libraries-committee/issues/216.
This will be removed in GHC 9.14.
- - - - -
821d0a9a by Cheng Shao at 2024-08-22T10:38:22-04:00
compiler: Store ForeignStubs and foreign C files in interfaces
This data is used alongside Core bindings to reconstruct intermediate
build products when linking Template Haskell splices with bytecode.
Since foreign stubs and files are generated in the pipeline, they were
lost with only Core bindings stored in interfaces.
The interface codec type `IfaceForeign` contains a simplified
representation of `ForeignStubs` and the set of foreign sources that
were manually added by the user.
When the backend phase writes an interface, `mkFullIface` calls
`encodeIfaceForeign` to read foreign source file contents and assemble
`IfaceForeign`.
After the recompilation status check of an upstream module,
`initWholeCoreBindings` calls `decodeIfaceForeign` to restore
`ForeignStubs` and write the contents of foreign sources to the file
system as temporary files.
The restored foreign inputs are then processed by `hscInteractive` in
the same manner as in a regular pipeline.
When linking the stub objects for splices, they are excluded from suffix
adjustment for the interpreter way through a new flag in `Unlinked`.
For details about these processes, please consult Note [Foreign stubs
and TH bytecode linking].
Metric Decrease:
T13701
- - - - -
f0408eeb by Cheng Shao at 2024-08-23T10:37:10-04:00
git: remove a.out and include it in .gitignore
a.out is a configure script byproduct. It was mistakenly checked into
the tree in !13118. This patch removes it, and include it in
.gitignore to prevent a similar error in the future.
- - - - -
1f95c5e4 by Matthew Pickering at 2024-08-23T10:37:46-04:00
docs: Fix code-block syntax on old sphinx version
This code-block directive breaks the deb9 sphinx build.
Fixes #25201
- - - - -
27dceb42 by Sylvain Henry at 2024-08-26T11:05:11-04:00
JS: add basic support for POSIX *at functions (#25190)
openat/fstatat/unlinkat/dup are now used in the recent release of the
`directory` and `file-io` packages.
As such, these functions are (indirectly) used in the following tests
one we'll bump the `directory` submodule (see !13122):
- openFile008
- jsOptimizer
- T20509
- bkpcabal02
- bkpcabal03
- bkpcabal04
- - - - -
1eebd0f9 by sheaf at 2024-08-27T17:58:10+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
-------------------------
- - - - -
b46be3b4 by sheaf at 2024-08-27T17:58:10+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
- - - - -
5526307f by sheaf at 2024-08-27T17:58:10+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.
- - - - -
eeaa8401 by sheaf at 2024-08-27T17:58:10+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.
- - - - -
539a32da by sheaf at 2024-08-27T17:58:11+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.
- - - - -
3de21a17 by sheaf at 2024-08-27T17:58:11+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.
- - - - -
7f64a6f6 by sheaf at 2024-08-27T17:58:11+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
- - - - -
41f31b08 by sheaf at 2024-08-27T17:58:11+02:00
Add test for C calls & SIMD vectors
- - - - -
106031d5 by sheaf at 2024-08-27T17:58:11+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.
- - - - -
f538ca1c by sheaf at 2024-08-27T17:58:11+02:00
Add test for #25169
- - - - -
30 changed files:
- .gitignore
- compiler/GHC/Builtin/primops.txt.pp
- compiler/GHC/ByteCode/Asm.hs
- compiler/GHC/ByteCode/Types.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
- compiler/GHC/CmmToAsm/Config.hs
- compiler/GHC/CmmToAsm/Format.hs
- compiler/GHC/CmmToAsm/Instr.hs
- compiler/GHC/CmmToAsm/PPC.hs
- compiler/GHC/CmmToAsm/PPC/CodeGen.hs
The diff was not included because it is too large.
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/dcbbcdf99c76a134e0d3d8f4d82f9d9b6eac3413...f538ca1cbc7757305a5e635c3226418728a88d80
--
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/dcbbcdf99c76a134e0d3d8f4d82f9d9b6eac3413...f538ca1cbc7757305a5e635c3226418728a88d80
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/20240827/50e01626/attachment-0001.html>
More information about the ghc-commits
mailing list