Nicholas Nethercote
1c989472e4
Box CastTarget within PassMode.
...
Because `PassMode::Cast` is by far the largest variant, but is
relatively rare.
This requires making `PassMode` not impl `Copy`, and `Clone` is no
longer necessary. This causes lots of sigil adjusting, but nothing very
notable.
2022-08-26 09:35:28 +10:00
Nathan Stocks
8c93170965
adjust to new error value
2022-08-25 11:06:45 -06:00
bjorn3
cfef0a4f8d
Merge commit 'e9d1a0a7b0b28dd422f1a790ccde532acafbf193' into sync_cg_clif-2022-08-24
2022-08-24 18:40:58 +02:00
bjorn3
ad1d993249
Sync from rust 4a24f08ba43166cfee86d868b3fe8612aec6faca
2022-08-24 18:29:45 +02:00
bjorn3
072fd2b0b1
Add fixme
2022-08-24 14:28:40 +00:00
bjorn3
293223d0cf
Tune drop_excess_capacity
2022-08-24 12:22:01 +02:00
bjorn3
f71c545746
Make sure to count reused cgus towards the count of jobs done
2022-08-23 16:51:06 +00:00
bjorn3
5b4195669e
Add some self profiler calls
2022-08-23 16:32:38 +00:00
bjorn3
d081c20273
Compile functions from clif ir to object code in parallel
2022-08-23 16:32:38 +00:00
bjorn3
1a6323313b
Use correct CguReuse variant
2022-08-23 16:32:20 +00:00
bjorn3
f9d60cf551
Do asm compilation and object file emission in parallel
2022-08-23 16:05:29 +00:00
bjorn3
1a0dfb399c
Add a jobserver based concurrency limiter
2022-08-23 15:44:39 +00:00
bjorn3
5896e5cdfa
Store symbol name as owned string
2022-08-23 15:23:56 +00:00
bjorn3
b1e9d2e1a2
Revert "Avoid masking shift amounts ( #1268 )"
...
This reverts commit 156bda8bc708cca60e9de18743d833c8d97dd7ff.
This breaks the mir_overflow_off rustc test:
https://github.com/bjorn3/rustc_codegen_cranelift/runs/7971362755?check_suite_focus=true#step:7:2904
2022-08-23 11:10:44 +00:00
Afonso Bordado
48c45c481c
Use native scalar fma instruction ( #1267 )
...
Cranelift 0.87 now supports lowering `fma` as a libcall on x86 [0].
With 0.88 enabling the native x86 instruction under the `has_fma` flag.
aarch64 and s390x already support this as a native instruction, so it's
nice that we emit it for those.
We can't lower the SIMD version using the `fma` instruction since the
lowering can fail if the x86 `has_fma` flag is not enabled. Cranelift
doesn't yet know how to fallback for these cases
[0]: 709716bb8e
2022-08-23 12:42:35 +02:00
Afonso Bordado
156bda8bc7
Avoid masking shift amounts ( #1268 )
...
Cranelift 0.87 now follows its own documentation regarding
shift amounts, and implicitly masks them if the arch requires it. [0]
[0]: 0508932174
2022-08-23 12:42:24 +02:00
bjorn3
b14c7337db
Update to Cranelift 0.87.0
2022-08-22 18:53:10 +00:00
Waffle Maybe
5f357c2c51
fix cg cranelift
...
Co-authored-by: bjorn3 <17426603+bjorn3@users.noreply.github.com>
2022-08-21 05:27:14 +04:00
Waffle Maybe
63a137605f
use shorter ptr_mask impl in cg cranelift
...
Co-authored-by: bjorn3 <bjorn3@users.noreply.github.com>
2022-08-21 05:27:14 +04:00
Maybe Waffle
5b32aa0747
Add pointer masking convenience functions
...
This commit adds the following functions all of which have a signature
`pointer, usize -> pointer`:
- `<*mut T>::mask`
- `<*const T>::mask`
- `intrinsics::ptr_mask`
These functions are equivalent to `.map_addr(|a| a & mask)` but they
utilize `llvm.ptrmask` llvm intrinsic.
*masks your pointers*
2022-08-21 05:27:14 +04:00
bjorn3
7dc8f38956
Merge pull request #1266 from bjorn3/parallel_comp_refactor2
...
Refactorings for enabling parallel compilation (part 2)
2022-08-19 12:03:43 +02:00
bjorn3
535c6ddc8b
Small cleanup
2022-08-18 19:03:28 +00:00
bjorn3
6421427b74
Move Span lowering from debuginfo finalization to codegen
...
This removes the dependency on TyCtxt from the debuginfo finalization
code.
2022-08-18 18:19:40 +00:00
bjorn3
0534a555cc
Make line_program_add_file a DebugContext method
2022-08-18 17:48:22 +00:00
bjorn3
1e57774011
Move set_function_span earlier
2022-08-18 17:23:36 +00:00
bjorn3
01be0ddacf
Move FunctionDebugContext creation to codegen_fn
2022-08-18 17:17:33 +00:00
bjorn3
dbf5457308
Introduce FunctionDebugContext
...
This will make it easier to move TyCtxt requiring operations before clif
ir compilation.
2022-08-18 17:11:41 +00:00
bjorn3
312563f3c4
Use walk_chain for function span too
...
This is a correctness fix
2022-08-18 15:58:22 +00:00
bjorn3
259b21fd46
Remove TyCtxt from DebugContext
...
And explicitly thread it through everwhere it is needed.
2022-08-18 15:25:52 +00:00
bjorn3
e5493a5ea2
Remove Instance param of DebugContext::define_function
2022-08-18 15:25:47 +00:00
bjorn3
a49416da6d
Remove stub local debuginfo implementation
...
It isn't actually wired up and temporarily removing it will make changes
to the debuginfo generation code much simpler.
2022-08-18 15:14:04 +00:00
bjorn3
5fc1366dfa
Register debuginfo for lazy jit shim
2022-08-18 12:55:44 +00:00
bjorn3
df1b25171c
Remove TyCtxt parameter from compile_fn
2022-08-17 13:47:52 +00:00
Afonso Bordado
2079b4bb08
Use stack_store instead of stack_addr+store when building structs
2022-08-17 14:46:05 +01:00
bjorn3
c820b7cd60
Remove TyCtxt field from CodegenCx
2022-08-17 13:43:32 +00:00
bjorn3
b181f2b376
Replace instance param of write_clif_file with symbol_name
2022-08-17 13:07:18 +00:00
bjorn3
701b2747c3
Remove optimize_function
...
It currently doesn't have any optimizations at all.
2022-08-17 13:03:32 +00:00
bjorn3
1662702a88
Work around new asm! usage in measureme
...
This is necessary to fix rustc bootstraps
2022-08-14 16:37:30 +00:00
bjorn3
523f0db7db
Merge pull request #1264 from bjorn3/parallel_comp_refactor
...
Refactorings for enabling parallel compilation (part 1)
2022-08-13 16:53:28 +02:00
bjorn3
9461fd2cb0
Remove TyCtxt parameter from emit_cgu
...
TyCtxt isn't available on background threads.
2022-08-13 12:18:41 +00:00
bjorn3
4c0766ce6c
Move error reporting out of emit_cgu
...
Error reporting requires a Session, which isn't available on background
threads.
2022-08-13 09:03:28 +00:00
bjorn3
c2f0b3a1bf
Move copy to incr comp cache to codegen join phase
...
The copy depends on Session, which is only available on the main thread.
As such the copy can't be done on future codegen threads.
2022-08-12 19:10:16 +00:00
bjorn3
6206c4e927
Stream object file to disk
...
This reduces memory usage and may improve performance slightly.
2022-08-12 18:57:01 +00:00
bjorn3
ab7c706306
Move build_isa call into make_module
2022-08-12 18:40:48 +00:00
bjorn3
d3512b1d8e
Don't attempt to do incr comp for the allocator shim
...
The allocator shim doesn't get reused and the allocator shim is just
under 2kb, so reusing it is likely more expensive than regenerating it.
2022-08-12 18:29:46 +00:00
bjorn3
db7d8a811d
Give fields of ModuleCodegenResult names
2022-08-12 18:29:43 +00:00
bjorn3
f76ca22479
Enable inline asm on macOS
2022-08-12 12:38:48 +00:00
bjorn3
e45f6000a0
Remove the partial linking hack for global asm support
2022-08-12 12:27:47 +00:00
bjorn3
48b312f04a
Don't take TyCtxt as argument for compile_global_asm
...
This allows it to be executed on a background thread.
2022-08-12 11:44:52 +00:00
Ulrich Weigand
aca05cf603
Fix alignment flag for emit_small_memory_copy
...
Do not unconditionally pass the "aligned" MemFlag when calling
emit_small_memory_copy. Instead, allow the back end to rely on
the alignment info passed separately to this routine.
2022-08-12 12:54:41 +02:00
