- Jun 04, 2024
-
-
This commit removes some legacy i386 windows related code paths in the RTS, given this target is no longer supported.
-
We have formally dropped i386 windows support (#18487) a long time ago. The stdcall foreign call convention is only used by i386 windows, and the legacy logic around it is a significant maintenance burden for future work that adds arm64 windows support (#24603). Therefore, this patch removes stdcall support completely from the compiler as well as the RTS (#24883): - stdcall is still recognized as a FFI calling convention in Haskell syntax. GHC will now unconditionally emit a warning (-Wunsupported-calling-conventions) and treat it as ccall. - Apart from minimum logic to support the parsing and warning logic, all other code paths related to stdcall has been completely stripped from the compiler. - ghci only supports FFI_DEFAULT_ABI and ccall convention from now on. - FFI foreign export adjustor code on all platforms no longer handles the stdcall case and only handles ccall from now on. - The Win32 specific parts of RTS no longer has special code paths for stdcall. This commit is the final nail on the coffin for i386 windows support. Further commits will perform more housecleaning to strip the legacy code paths and pave way for future arm64 windows support.
-
This commit introduces a new language extension, `-XOrPatterns`, as described in GHC Proposal 522. An or-pattern `pat1; ...; patk` succeeds iff one of the patterns `pat1`, ..., `patk` succeed, in this order. See also the summary `Note [Implmentation of OrPatterns]`. Co-Authored-By:Sebastian Graf <sgraf1337@gmail.com>
-
-
- Jun 03, 2024
-
-
This commit fixes a subtle mistake in alignmentFromTypes that used to generate Unaligned when element type & index type are the same width. Fixes #24930.
-
For each module in a GHCi session, we keep alive one `ModLocation`. A `ModLocation` is fairly inefficiently packed, as `String`s are expensive in memory usage. While benchmarking the agda codebase, we concluded that we keep alive around 11MB of `FilePath`'s, solely retained by `ModLocation`. We provide a more densely packed encoding of `ModLocation`, by moving from `FilePath` to `OsPath`. Further, we migrate the full `Finder` component to `OsPath` to avoid unnecessary transformations. As the `Finder` component is well-encapsulated, this requires only a minimal amount of changes in other modules. We introduce pattern synonym for 'ModLocation' which maintains backwards compatibility and avoids breaking consumers of 'ModLocation'.
-
-
Implements the algorithm from compiler-rt's udiv128by64to64default. This rewrite results in a roughly 24x improvement in runtime on AArch64 (and likely any other arch that uses it).
-
- Jun 02, 2024
-
-
Fixes #24839. Co-authored-by:Sylvain Henry <hsyl20@gmail.com>
-
This patch makes hadrian build in-tree GMP with the --enable-alloca=malloc-notreentrant configure option. We will only need malloc-reentrant when we have threaded RTS and SMP support on wasm32, which will take some time to happen, before which we should use malloc-notreentrant to avoid undesired runtime overhead.
-
This patch disables in-tree GMP FFT code paths for wasm32 target in order to give up some performance of multiplying very large operands in exchange for reduced code size.
-
This patch disables PIC for in-tree GMP on wasm32 target. Enabling PIC unconditionally adds undesired code size and runtime overhead for wasm32.
-
- Jun 01, 2024
-
-
-
For whatever reason, the `Outputable` instance for `FixitySig` simply did not print out namespace specifiers, leading to the confusing `-ddump-splices` output seen in #24911. This patch corrects this oversight. Fixes #24911.
-
This patch fixes missing -ddump-cmm-raw output when compiling .cmm, which is useful for debugging cmm related codegen issues.
-
-
ucd2haskell tool used streamly library which version in cabal was out of date. It is updated to the latest version at hackage with deprecated parts rewritten. Also following fixes were applied to existing code in suppose that from its last run the code structure was changed and now it was required to be up to date with actual folder structures: 1. Ghc module path environment got a suffix with `src`. 2. Generated code got 2.1 `GHC.Internal` prefix for `Data.*`. 2.2 `GHC.Unicode.Internal` swapped on `GHC.Internal.Unicode` according to actual structure.
-
The debian12 image in this commit has llvm 18 installed.
-
-
- May 31, 2024
-
-
-
'UniqSet Unique' represents a set of uniques as a 'Map Unique Unique', which is wasting space (associated key/value are always the same). Fix #23572 and #23605
-
When configure fails to find LLC/OPT/LLVMAS within supported version range, it used to set "llc"/"opt"/"clang" as fallback values. This behavior is particularly troublesome when the user has llc/opt/clang with other versions in their PATH and run the testsuite, since hadrian will incorrectly assume have_llvm=True and pass that to the testsuite driver, resulting in annoying optllvm test failures (#23186). If configure determines llc/opt/clang wouldn't work, then we shouldn't pretend it'll work at all, and the bindist configure will invoke FIND_LLVM_PROG check again at install time anyway.
-
hadrian may invoke findExecutable "" at run-time due to a certain program is not found by configure script. Which is fine and findExecutable is supposed to return Nothing in this case. However, on Windows there's a directory bug that throws an exception (see https://github.com/haskell/directory/issues/180), so we might as well use a wrapper for findExecutable and handle exceptions gracefully.
-
Add hie.yaml to ghc-config project directory so it can be edited using HLS.
-
This patch removes special consideration for ArchWasm32 in cmmDoCmmSwitchPlans, which means the compiler will now disable cmmImplementSwitchPlans for wasm unreg backend, just like unreg backend of other targets. We enabled it in the past to workaround some compile-time panic in older versions of LLVM, but those panics are no longer present, hence no need to keep this workaround.
-
The previous description was inherited from `base` and was inappropriate for `ghc-internal`. Also fix the maintainer and bug reporting fields. Closes #24906.
-
Closes #24863
-
-
Deriving `Typeable` does nothing, and it hasn't done for a long while. There has also been a warning for a long while which warns you about uselessly deriving it but it wasn't enabled in -Wall. Fixes #24784
-
-
When -fsplit-sections is passed to GHC, pass -ffunction-sections -fdata-sections to gcc/clang when building C/C++. Previously, -fsplit-sections was only respected by the NCG/LLVM backends, but not the unregisterised backend; the GHC driver did not pass -fdata-sections and -ffunction-sections to the C compiler, which resulted in excessive executable sizes. Fixes #23381. ------------------------- Metric Decrease: size_hello_artifact size_hello_unicode ------------------------- -
When split sections is enabled, ensure -fsplit-sections is passed to GHC as well when invoking GHC to compile C/C++; and pass -ffunction-sections -fdata-sections to gcc/clang when compiling C/C++ with the hadrian Cc builder. Fixes #23381.
-
...so that they do not cause coerce to bypass the nominal role on the corresponding singleton types when they are imported. See Note [Preventing unsafe coercions for singleton types] and the discussion at #23478. This also introduces unsafeWithSNatCo (and analogues for Char and Symbol) so that users can still access the dangerous coercions that importing the real constructors would allow, but only in a very localized way.
-
- May 30, 2024
-
-
This adds diagrams to the documentation of Arrows, similar to the ones found on https://www.haskell.org/arrows/. It does not add diagrams for ArrowChoice for the time being, mainly because it's not clear to me how to visually distinguish them from the ones for Arrow. Ideally, you might want to do something like highlight the arrows belonging to the same tuple or same Either in common colors, but that's not really possible with unicode.
-
There were some syntax errors in the generation script which were preventing it from running. I have tested this with: ``` nix shell --extra-experimental-features nix-command -f .gitlab/rel_eng -c ghcup-metadata --metadata ghcup-0.0.7.yaml --date="2024-05-27" --pipeline-id=95534 --version=9.11.20240525 ``` which completed successfully.
-
-
This patch fixes the STG_FIELD_OFFSET macro definition by using __builtin_offsetof, which is what gcc/clang uses to implement offsetof in standard C. The previous definition that uses NULL pointer involves subtle undefined behavior in C and thus reported by UndefinedBehaviorSanitizer as well: ``` rts/Capability.h:243:58: runtime error: member access within null pointer of type 'Capability' (aka 'struct Capability_') SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior rts/Capability.h:243:58 ```
-
This patch fixes an unaligned load in nonmoving gc by ensuring the closure address is properly untagged first before attempting to prefetch its header. The unaligned load is reported by UndefinedBehaviorSanitizer: ``` rts/sm/NonMovingMark.c:921:9: runtime error: member access within misaligned address 0x0042005f3a71 for type 'StgClosure' (aka 'struct StgClosure_'), which requires 8 byte alignment 0x0042005f3a71: note: pointer points here 00 00 00 98 43 13 8e 12 7f 00 00 50 3c 5f 00 42 00 00 00 58 17 b7 92 12 7f 00 00 89 cb 5e 00 42 ^ SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior rts/sm/NonMovingMark.c:921:9 ``` This issue had previously gone unnoticed since it didn't really harm runtime correctness, the invalid header address directly loaded from a tagged pointer is only used as prefetch address and will not cause segfaults. However, it still should be corrected because the prefetch would be rendered useless by this issue, and untagging only involves a single bitwise operation without memory access so it's cheap enough to add. -
gc_thread/gen_workspace are required to be aligned by 64 bytes. However, this property has not been properly enforced before, and numerous alignment violations at runtime has been caught by UndefinedBehaviorSanitizer that look like: ``` rts/sm/GC.c:1167:8: runtime error: member access within misaligned address 0x0000027a3390 for type 'gc_thread' (aka 'struct gc_thread_'), which requires 64 byte alignment 0x0000027a3390: note: pointer points here 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ^ SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior rts/sm/GC.c:1167:8 rts/sm/GC.c:1184:13: runtime error: member access within misaligned address 0x0000027a3450 for type 'gen_workspace' (aka 'struct gen_workspace_'), which requires 64 byte alignment 0x0000027a3450: note: pointer points here 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ^ SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior rts/sm/GC.c:1184:13 ``` This patch fixes the gc_thread/gen_workspace misalignment issue by explicitly allocating them with alignment constraint. -
-
