Unfortunately, the self-hosted SPIR-V backend is quite tightly coupled
with the self-hosted SPIR-V linker through its `Object` concept (which
is much like `llvm.Object`). Reworking this would be too much work for
this branch. So, for now, I have introduced a special case (similar to
the LLVM backend's special case) to the codegen logic when using this
backend. We will want to delete this special case at some point, but it
need not block this work.
My original goal here was just to get the self-hosted Wasm backend
compiling again after the pipeline change, but it turned out that from
there it was pretty simple to entirely eliminate the shared state
between `codegen.wasm` and `link.Wasm`. As such, this commit not only
fixes the backend, but makes it the second backend (after CBE) to
support the new 1:N:1 threading model.
The idea here is that instead of the linker calling into codegen,
instead codegen should run before we touch the linker, and after MIR is
produced, it is sent to the linker. Aside from simplifying the call
graph (by preventing N linkers from each calling into M codegen
backends!), this has the huge benefit that it is possible to
parallellize codegen separately from linking. The threading model can
look like this:
* 1 semantic analysis thread, which generates AIR
* N codegen threads, which process AIR into MIR
* 1 linker thread, which emits MIR to the binary
The codegen threads are also responsible for `Air.Legalize` and
`Air.Liveness`; it's more efficient to do this work here instead of
blocking the main thread for this trivially parallel task.
I have repurposed the `Zcu.Feature.separate_thread` backend feature to
indicate support for this 1:N:1 threading pattern. This commit makes the
C backend support this feature, since it was relatively easy to divorce
from `link.C`: it just required eliminating some shared buffers. Other
backends don't currently support this feature. In fact, they don't even
compile -- the next few commits will fix them back up.
The main goal of this commit is to make it easier to decouple codegen
from the linkers by being able to do LLVM codegen without going through
the `link.File`; however, this ended up being a nice refactor anyway.
Previously, every linker stored an optional `llvm.Object`, which was
populated when using LLVM for the ZCU *and* linking an output binary;
and `Zcu` also stored an optional `llvm.Object`, which was used only
when we needed LLVM for the ZCU (e.g. for `-femit-llvm-bc`) but were not
emitting a binary.
This situation was incredibly silly. It meant there were N+1 places the
LLVM object might be instead of just 1, and it meant that every linker
had to start a bunch of methods by checking for an LLVM object, and just
dispatching to the corresponding method on *it* instead if it was not
`null`.
Instead, we now always store the LLVM object on the `Zcu` -- which makes
sense, because it corresponds to the object emitted by, well, the Zig
Compilation Unit! The linkers now mostly don't make reference to LLVM.
`Compilation` makes sure to emit the LLVM object if necessary before
calling `flush`, so it is ready for the linker. Also, all of the
`link.File` methods which act on the ZCU -- like `updateNav` -- now
check for the LLVM object in `link.zig` instead of in every single
individual linker implementation. Notably, the change to LLVM emit
improves this rather ludicrous call chain in the `-fllvm -flld` case:
* Compilation.flush
* link.File.flush
* link.Elf.flush
* link.Elf.linkWithLLD
* link.Elf.flushModule
* link.emitLlvmObject
* Compilation.emitLlvmObject
* llvm.Object.emit
Replacing it with this one:
* Compilation.flush
* llvm.Object.emit
...although we do currently still end up in `link.Elf.linkWithLLD` to do
the actual linking. The logic for invoking LLD should probably also be
unified at least somewhat; I haven't done that in this commit.
To my knowledge, the only platforms that actually *require* PIE are Fuchsia and
Android, and the latter *only* when building a dynamically-linked executable.
OpenBSD and macOS both strongly encourage using PIE by default, but it isn't
technically required. So for the latter platforms, we enable it by default but
don't enforce it.
Also, importantly, if we're building an object file or a static library, and the
user hasn't explicitly told us whether to build PIE or non-PIE code (and the
target doesn't require PIE), we should *not* default to PIE. Doing so produces
code that cannot be linked into non-PIE output. In other words, building an
object file or a static library as PIE is an optimization only to be done when
the user knows that it'll end up in a PIE executable in the end.
Closes#21837.
This adds 4 `Legalize.Feature`s:
* `expand_intcast_safe`
* `expand_add_safe`
* `expand_sub_safe`
* `expand_mul_safe`
These do pretty much what they say on the tin. This logic was previously
in Sema, used when `Zcu.Feature.safety_checked_instructions` was not
supported by the backend. That `Zcu.Feature` has been removed in favour
of this legalization.
Nothing interesting here; literally just the bare minimum so I can work on this
on and off in a branch without worrying about merge conflicts in the non-backend
code.
Textual PTX is just assembly language like any other. And if we do ever add
support for emitting PTX object files after reverse engineering the bytecode
format, we'd be emitting ELF files like the CUDA toolchain. So there's really no
need for a special ObjectFormat tag here, nor linker code that treats it as a
distinct format.
These backends are completely unusable at the moment; they can produce neither
assembly files nor object files. So give a nicer error when users try to use
them.
Emscripten currently implements `emscripten_return_address()` by calling
out into JavaScript and parsing a stack trace, which introduces
significant overhead that we would prefer to avoid in release builds.
This is especially problematic for allocators because the generic parts
of `std.mem.Allocator` make frequent use of `@returnAddress`, even
though very few allocator implementations even observe the return
address, which makes allocators nigh unusable for performance-critical
applications like games if the compiler is unable to devirtualize the
allocator calls.
Functions like isMinGW() and isGnuLibC() have a good reason to exist: They look
at multiple components of the target. But functions like isWasm(), isDarwin(),
isGnu(), etc only exist to save 4-8 characters. I don't think this is a good
enough reason to keep them, especially given that:
* It's not immediately obvious to a reader whether target.isDarwin() means the
same thing as target.os.tag.isDarwin() precisely because isMinGW() and similar
functions *do* look at multiple components.
* It's not clear where we would draw the line. The logical conclusion before
this commit would be to also wrap Arch.isX86(), Os.Tag.isSolarish(),
Abi.isOpenHarmony(), etc... this obviously quickly gets out of hand.
* It's nice to just have a single correct way of doing something.
This is all of the expected 0.14.0 progress on #21530, which can now be
postponed once this commit is merged.
This required rewriting the (un)wrap operations since the original
implementations were extremely buggy.
Also adds an easy way to retrigger Sema OPV bugs so that I don't have to
keep updating #22419 all the time.
Sema is arbitrarily scalarizing some operations, which means that when I
try to implement vectorized versions of those operations in a backend,
they are impossible to test due to Sema not producing them. Now, I can
implement them and then temporarily enable the new feature for that
backend in order to test them. Once the backend supports all of them,
the feature can be permanently enabled.
This also deletes the Air instructions `int_from_bool` and
`int_from_ptr`, which are just bitcasts with a fixed result type, since
changing `un_op` to `ty_op` takes up the same amount of memory.
This moves the default value logic to Package.Module.create() instead and makes
it so that Compilation.Config.any_unwind_tables is computed similarly to
any_sanitize_thread, any_fuzz, etc. It turns out that for any_unwind_tables, we
only actually care if unwind tables are enabled at all, not at what level.
The blocker for enabling this feature was my need to debug the emitted
assembly without debug info and having to manually inspect memory to
determine struct contents. However, we now have debug info!
(lldb) v -L foo bar
0x00007fffffffda20: (repro.repro.Foo) foo = {
0x00007fffffffda24: .x = 12
0x00007fffffffda20: .y = 34
}
0x00007fffffffda28: (repro.repro.Bar) bar = {
0x00007fffffffda28: .x = 56
0x00007fffffffda2c: .y = 78
}
Updates #21530
The goal here is to support both levels of unwind tables (sync and async) in
zig cc and zig build. Previously, the LLVM backend always used async tables
while zig cc was partially influenced by whatever was Clang's default.
Ali Cheraghi