rustc: Handle modules in "fat" LTO more robustly
When performing a "fat" LTO the compiler has a whole mess of codegen
units that it links together. To do this it needs to select one module
as a "base" module and then link everything else into this module.
Previously LTO passes assume that there's at least one module in-memory
to link into, but nowadays that's not always true! With incremental
compilation modules may actually largely be cached and it may be
possible that there's no in-memory modules to work with.
This commit updates the logic of the LTO backend to handle modules a bit
more uniformly during a fat LTO. This commit immediately splits them
into two lists, one serialized and one in-memory. The in-memory list is
then searched for the largest module and failing that we simply
deserialize the first serialized module and link into that. This
refactoring avoids juggling three lists, two of which are serialized
modules and one of which is half serialized and half in-memory.
Closes#63349
When performing a "fat" LTO the compiler has a whole mess of codegen
units that it links together. To do this it needs to select one module
as a "base" module and then link everything else into this module.
Previously LTO passes assume that there's at least one module in-memory
to link into, but nowadays that's not always true! With incremental
compilation modules may actually largely be cached and it may be
possible that there's no in-memory modules to work with.
This commit updates the logic of the LTO backend to handle modules a bit
more uniformly during a fat LTO. This commit immediately splits them
into two lists, one serialized and one in-memory. The in-memory list is
then searched for the largest module and failing that we simply
deserialize the first serialized module and link into that. This
refactoring avoids juggling three lists, two of which are serialized
modules and one of which is half serialized and half in-memory.
Closes#63349
Closure types have been moved to the namespace where they
are defined, and both closure and generator type names now
include the disambiguator.
This fixes an exception when lldb prints nested closures.
Fixes#57822
Do not generate allocations for zero sized allocations
Alternative to https://github.com/rust-lang/rust/issues/62487
r? @eddyb
There are other places where we could do this, too, but that would cause `static FOO: () = ();` to not have a unique address
Deduplicate rustc_demangle in librustc_codegen_llvm
This commit removes the crates.io dependency of `rustc-demangle` from
`rustc_codegen_llvm`. This crate is actually already pulled in to part
of the `librustc_driver` build and with the upcoming pipelining
implementation in Cargo it causes build issues if `rustc-demangle` is
left to its own devices.
This is not currently required, but once pipelining is enabled for
rustc's own build it will be required to build correctly.
This commit removes the crates.io dependency of `rustc-demangle` from
`rustc_codegen_llvm`. This crate is actually already pulled in to part
of the `librustc_driver` build and with the upcoming pipelining
implementation in Cargo it causes build issues if `rustc-demangle` is
left to its own devices.
This is not currently required, but once pipelining is enabled for
rustc's own build it will be required to build correctly.
rustc: Update wasm32 support for LLVM 9
This commit brings in a number of minor updates for rustc's support for
the wasm target which has changed in the LLVM 9 update. Notable updates
include:
* The compiler now no longer manually inserts the `producers` section,
instead relying on LLVM to do so. LLVM uses the `llvm.ident` metadata
for the `processed-by` directive (which is now emitted on the wasm
target in this PR) and it uses debuginfo to figure out what `language`
to put in the `producers` section.
* Threaded WebAssembly code now requires different flags to be passed
with LLD. In LLD we now pass:
* `--shared-memory` - required since objects are compiled with
atomics. This also means that the generated memory will be marked as
`shared`.
* `--max-memory=1GB` - required with the `--shared-memory` argument
since shared memories in WebAssembly must have a maximum size. The
1GB number is intended to be a conservative estimate for rustc, but
it should be overridable with `-C link-arg` if necessary.
* `--passive-segments` - this has become the default for multithreaded
memory, but when compiling a threaded module all data segments need
to be marked as passive to ensure they don't re-initialize memory
for each thread. This will also cause LLD to emit a synthetic
function to initialize memory which users will have to arrange to
call.
* The `__heap_base` and `__data_end` globals are explicitly exported
since they're now hidden by default due to the `--export` flags we
pass to LLD.
This commit brings in a number of minor updates for rustc's support for
the wasm target which has changed in the LLVM 9 update. Notable updates
include:
* The compiler now no longer manually inserts the `producers` section,
instead relying on LLVM to do so. LLVM uses the `llvm.ident` metadata
for the `processed-by` directive (which is now emitted on the wasm
target in this PR) and it uses debuginfo to figure out what `language`
to put in the `producers` section.
* Threaded WebAssembly code now requires different flags to be passed
with LLD. In LLD we now pass:
* `--shared-memory` - required since objects are compiled with
atomics. This also means that the generated memory will be marked as
`shared`.
* `--max-memory=1GB` - required with the `--shared-memory` argument
since shared memories in WebAssembly must have a maximum size. The
1GB number is intended to be a conservative estimate for rustc, but
it should be overridable with `-C link-arg` if necessary.
* `--passive-segments` - this has become the default for multithreaded
memory, but when compiling a threaded module all data segments need
to be marked as passive to ensure they don't re-initialize memory
for each thread. This will also cause LLD to emit a synthetic
function to initialize memory which users will have to arrange to
call.
* The `__heap_base` and `__data_end` globals are explicitly exported
since they're now hidden by default due to the `--export` flags we
pass to LLD.
Exception for specific cases like linting, additional passes should
be going into the module pass manager (even if they are function
passes). The separate function pass manager is only used for very
early optimization passes.
Rather than apparending passes to the MPM, use the OptimizerLast
and EnabledOnOptLevel0 pass manager builder extension hooks, which
allow adding passes directly before finalization (alias
canonicalization and name-anon-globals).
The main effect and purpose of this change is to add sanitizer
passes at the end of the pipeline, which is where they belong.
In LLVM 9 the address sanitizer can't be used as a pass in the
early function pass manager, because it has a dependence on a
module-level analysis pass.
Prepare for LLVM 9 update
Main changes:
* In preparation for opaque pointer types, the `byval` attribute now takes a type. As such, the argument type needs to be threaded through to the function/callsite attribute application logic.
* On ARM the `+fp-only-sp` and `+d16` features have become `-fp64` and `-d32`. I've switched the target definitions to use the new names, but also added bidirectional emulation so either can be used on any LLVM version for backwards compatibility.
* The datalayout can now specify function pointer alignment. In particular on ARM `Fi8` is specified, which means that function pointer alignment is independent of function alignment. I've added this to our datalayouts to match LLVM (which is something we check) and strip the fnptr alignment for older LLVM versions.
* The fmul/fadd reductions now always respect the accumulator (including for unordered reductions), so we should pass the identity instead of undef.
Open issues:
* https://reviews.llvm.org/D62106 causes linker errors with ld.bdf due to https://sourceware.org/bugzilla/show_bug.cgi?id=24784. To avoid this I've enabled `RelaxELFRelocations`, which results in a GOTPCRELX relocation for `__tls_get_addr` and avoids the issue. However, this is likely not acceptable because relax relocations are not supported by older linker versions. We may need an LLVM option to keep using PLT for `__tls_get_addr` despite `RtLibUseGOT`.
The corresponding llvm-project PR is https://github.com/rust-lang/llvm-project/pull/19.
r? @ghost
Check if the archive has already been added to avoid duplicates
This avoids adding archives multiple times, which results in duplicate
objects in the resulting rlib, leading to symbol collision and link
failures. This could happen when crate contains multiple link attributes
that all reference the same archive.
Now that procedural macros no longer link transitively to libsyntax,
this shouldn't be needed any more! This commit is an experiment in
removing all dynamic libraries from rustc except for librustc_driver
itself. Let's see how far we can get with that!
