Commit Graph

61 Commits

Author SHA1 Message Date
Alex Crichton
e2a724ce18 Update the object crate to 0.27.0 (#3465)
Mostly just keeping us up to date with changes there since we somewhat
heavily rely on it now.
2021-10-20 10:52:06 -05:00
bjorn3
d78f436daf Remove reloc_constant
It is no longer used by the new backends
2021-10-10 14:43:55 +02:00
bjorn3
2db3b5b9df Remove code offsets from Function (#3412)
* Remove code offsets from Function

* Remove reloc_jt and fix wasmtime-cranelift
2021-10-07 15:54:00 +02:00
Benjamin Bouvier
43a86f14d5 Remove more old backend ISA concepts (#3402)
This also paves the way for unifying TargetIsa and MachBackend, since now they map one to one. In theory the two traits could be merged, which would be nice to limit the number of total concepts. Also they have quite different responsibilities, so it might be fine to keep them separate.

Interestingly, this PR started as removing RegInfo from the TargetIsa trait since the adapter returned a dummy value there. From the fallout, noticed that all Display implementations didn't needed an ISA anymore (since these were only used to render ISA specific registers). Also the whole family of RegInfo / ValueLoc / RegUnit was exclusively used for the old backend, and these could be removed. Notably, some IR instructions needed to be removed, because they were using RegUnit too: this was the oddball of regfill / regmove / regspill / copy_special, which were IR instructions inserted by the old regalloc. Fare thee well!
2021-10-04 10:36:12 +02:00
Benjamin Bouvier
bae4ec6427 Remove ancient register allocation (#3401) 2021-09-30 21:27:23 +02:00
Afonso Bordado
b8ad99e435 aarch64: Implement TLS ELF GD Relocations
Implement the `TlsValue` opcode in the aarch64 backend for ELF_GD.

This is a little bit unusual as the default TLS mechanism for aarch64 is TLS Descriptors in other compilers.
However currently we only recognize elf_gd so lets start with that as a TLS implementation.
2021-06-24 12:21:44 +01:00
bjorn3
7d23a9daf7 Lower Arm64Call relocations in cranelift-object 2021-06-10 12:54:25 +02:00
StackDoubleFlow
9637bc5a09 Fix cranelift Module and ObjectModule docs links (#2852) 2021-04-21 06:29:02 -07:00
bjorn3
cc89111463 Support declaring anonymous functions and data objects 2021-03-21 18:00:26 +01:00
Will Robson
38926fb1fc cranelift-module: Add support for passing a StackMapSink when defining functions
Fixes #2738

This follows the convention set by the existing method of passing a
TrapSink by adding another argument for a StackMapSink.
2021-03-19 00:02:15 +00:00
bjorn3
a710fc4425 Make Module object safe 2021-02-22 16:00:37 +01:00
Chris Fallin
76fb9a08c7 cranelift-object: reject symbol names with NUL byte without panic'ing.
Avoid a `panic!()`, and return a proper error, on a NUL byte. We hit a
null-byte check inside the `object` crate otherwise; this blocks fuzzing
when testing via a write-object-file-and-dlopen flow.
2020-12-17 09:53:56 -08:00
bjorn3
b7a93c2321 Remove reloc_block
It isn't called and all reloc sinks either ignore it or panic when it is
called.
2020-11-11 12:36:17 +01:00
bjorn3
1dc27c93a4 Introduce DataDescription::all_relocs to dedup some code 2020-11-11 12:36:17 +01:00
bjorn3
844a52e96a Don't unnecessarily take &self for some ModuleDeclarations methods 2020-11-11 11:43:09 +01:00
bjorn3
856f799ade Make some things more consistent between define_function and define_function_bytes 2020-11-11 11:38:28 +01:00
Alex Crichton
8af2dbfbac Allow offloading compilation in cranelift-object (#2371)
This commit is a slight refactoring of the `Module` trait and backend in
`cranelift-object`. The goal is to enable parallelization of compilation
when using `cranelift-object`. Currently this is difficult because
`ObjectModule::define_function` requires `&mut self`. This instead
soups up the `define_function_bytes` interface to handle relocations so
compilation can happen externally before defining it in a `Module`. This
also means that `define_function` is now a convenience wrapper around
`define_function_bytes`.
2020-11-06 09:56:44 -06:00
bjorn3
b061694491 Rustfmt and update docs 2020-10-01 09:53:23 +02:00
bjorn3
d84ca235d2 Remove Backend trait and turn Module into a trait 2020-09-30 19:52:57 +02:00
bjorn3
84c6ec3214 Move alignment config from declare_data to define_data 2020-09-30 19:23:23 +02:00
bjorn3
b44c5bb2be Move ModuleDeclarations to backends 2020-09-30 18:33:29 +02:00
bjorn3
7a6e909efe Move a bit more logic out of Module 2020-09-30 17:31:08 +02:00
bjorn3
c2ffcdc6d0 Move logic out of more Module methods 2020-09-30 17:12:33 +02:00
bjorn3
80f4ecf9b5 Move almost all logic out of Module 2020-09-30 16:50:24 +02:00
bjorn3
588a4be0b3 Store Compiled* in the backend instead of Module 2020-09-30 16:18:57 +02:00
bjorn3
405b9e2875 Remove finalize_* from the Backend trait
Instead let the `finish` method perform finalization
2020-09-30 14:20:39 +02:00
bjorn3
59f95083b1 Remove write_data_funcaddr and write_data_dataaddr
They are unimplemented by all backends
2020-09-30 13:59:26 +02:00
bjorn3
7608749647 Merge finalize_definitions into finish 2020-09-30 13:58:13 +02:00
bjorn3
4483c3740a Remove get_finalized_* 2020-09-30 13:53:01 +02:00
bjorn3
6161acfba5 Minor simplification 2020-09-30 12:33:26 +02:00
bjorn3
b1187b5507 Merge ModuleNamespace and ModuleContents 2020-09-30 12:29:22 +02:00
Pat Hickey
1fabb051b0 Merge pull request #2218 from bjorn3/per_function_sections
cranelift-object: Support per function sections
2020-09-22 11:28:58 -07:00
bjorn3
863d4272a7 cranelift-object: Support per function sections
This makes it possible for the linker to omit unused functions when
--gc-sections is passed.
2020-09-22 14:04:15 +02:00
bjorn3
deb20c129a Fix relocated readonly data in custom sections
Lld doesn't allow relocations in readonly sections
2020-09-21 13:38:42 +02:00
Joshua Nelson
d28abad441 Upgrade to target-lexicon 0.11
This allows downstream library users to use `CDataModel` without having
to install two different versions of target-lexicon.
2020-09-15 11:40:09 -07:00
Nick Fitzgerald
05bf9ea3f3 Rename "Stackmap" to "StackMap"
And "stackmap" to "stack_map".

This commit is purely mechanical.
2020-08-07 10:08:44 -07:00
Pat Hickey
4f72a7483b Merge pull request #1836 from carlokok/feature/object_file_section
Cranelift: Module data apis should allow specifying the object file section
2020-06-17 10:49:09 -07:00
Carlo Kok
b2549d1fde Rust fmt fixes 2020-06-17 06:46:58 +02:00
Carlo Kok
0b613caad1 Address concerns in pullrequests. 2020-06-15 20:46:04 +02:00
bjorn3
9788b02dd5 Bump object to 0.19.0 (#1767)
* Bump object to 0.19.0
2020-06-12 15:37:04 -05:00
Carlo Kok
38531b8f42 Rust fmt fixes 2020-06-08 19:56:21 +02:00
Carlo Kok
4ff01aa898 Extra assertions and fix for Faerie. 2020-06-08 07:22:53 +02:00
Carlo Kok
6b47079884 Cranelift: Module data apis should allow specifying the object file section #1640 2020-06-06 16:08:21 +02:00
Benjamin Bouvier
f4c4a84b84 cranelift codegen: pass source locations with external relocations; 2020-04-07 11:52:39 +02:00
Nathan Froyd
83602e05b6 cranelift-object: mark output as using non-executable stacks
The `faerie` crate does this by default.  `object` itself does not, but
it seems reasonable to make this change for parity between
`cranelift-faerie` and `cranelift-object`.
2020-04-01 12:20:53 -04:00
Nathan Froyd
dcabb55776 change Module::define_function to take TrapSink instances
Experience with the `define_function` API has shown that returning
borrowed slices of `TrapSite` is not ideal: the returned slice
represents a borrow on the entire `Module`, which makes calling back
into methods taking `&mut self` a bit tricky.

To eliminate the problem, let's require the callers of `define_function`
to provide `TrapSink` instances.  This style of API enables them to
control when and how traps are collected, and makes the `object` and
`faerie` backends simpler/more efficient by not having to worry about
trap collection.
2020-03-24 13:36:01 -04:00
bjorn3
d54611dac8 Update object to 0.18 (#1381) 2020-03-23 08:56:51 -05:00
bjorn3
b4562c62e3 Add support for hidden visibility 2020-03-17 17:51:12 +01:00
Nathan Froyd
2bb3096342 change interfaces to use slices instead of Vec 2020-03-05 10:19:01 -05:00
Nathan Froyd
0f49a830c9 cranelift-module: expose trap information when defining functions
The current interface of `cranelift-module` requires consumers who want
to be informed about traps to discover that information through
`Module::Product`, which is backend-specific.  Since it's advantageous
to manipulate this information in a backend-agnostic way, this patch
changes `Module::define_function{,_bytes}` to return information about
the traps contained in the function being defined.
2020-03-03 11:37:15 -05:00