This will allow wasm implementations that wish to insert code into
every loop, for example to insert an interrupt check or a safepoint.
do so without relying on asynchronous signals.
There are two cases:
1. It seems that grep on macOS exits as soon as it finds the first match. This makes cargo unhappy and it prints message like "failed printing to stdout: Broken pipe (os error 32)". The solution is to fully consume the output from cargo. I choose to use tee for this task.
2. When in a strict mode, bash complains that $1 is not defined (when it's actually not defined in case of omitting --install). The solution is to apply bash substitution magic: when $1 is undefined or set to null substitute it with empty string.
* Rename `I32` -> `X86_32` and `I64` -> `X86_64`
* Format file to pass flake8 tests
* Fix comment so lines are under 80 char limit
* Remove trailing whitespace from comment
* Renamed `enc_i64` to `enc_x86_64` as per suggestion from PR
In this case, it's a little nicer to just use more assertions,
which will print their line number indicating how far the test
got, when they fail.
And this allows the tests to be run in no_std configurations.
This allows us to run the tests via a library call rather than just
as a command execution. And, it's a step toward a broader goal, which
is to keep the code in the top-level src directory minimal, with
important functionality exposed as crates.
Refactor the filetests harness so that it can be run as part of
`cargo test`. And begin reorganizing the test harness code in preparation
for moving it out of the src directory.
- Test subcommand files are now named `test_*.rs`.
- cton-util subcommand files now just export their `run` and nothing else.
- src/filetest/mod.rs now also just exports `run` and nothing else.
- Tests are now run in release mode (with debug assertions enabled).
This makes it a little simpler to generate 'match' statements, and
it performs deduplication of identical arms. And it means I don't
have to think about as many strings like '{} {{ {}.. }} => {}'
when I'm trying to think about how instructions work :-).
* Rename `ILBuilder` to `FunctionBuilderContext` and update corresponding
code
* Refactor usages of ILBuilder to become FunctionBuilderContext,
update variable names to reflect this change
* Reformat to ensure that lines stay under 100 char limit
* Apply corrections from `rustfmt` to pass tests
* Rename variables to be more consistent with refactor of ILBuilder
In this case, it's a little nicer to just use more assertions,
which will print their line number indicating how far the test
got, when they fail.
And this allows the tests to be run in no_std configurations.
Compute the bound values for expand_fcvt_to_sint using bitwise integer
arithmetic rather than floating-point arithmetic, to avoid relying on
host floating point arithmetic.
This allows the assertions to be disabled in release builds, so that
the code is faster and smaller, at the expense of not performing the
checks. Assertions can be re-enabled in release builds with the
debug-assertions flag in Cargo.toml, as the top-level Cargo.toml
file does.
Previously, cretonne-wasm used its own Local struct for identifying
local variables. However, now that cretonne-frontend provides a
Variable struct, just use that instead.
Frontends can still use their own types with `ILBuilder` and
`FunctionBuilder`. This just provides a basic `Variable` struct
for frontends that want it.
* Elaborate on Cretonne's rustc backend goals.
* Remove these extra newlines. They don't show up in the rendered form anyway.
* Fix typo.
* The document is meant to speak with the voice of the project.
When relaxing a branch, restrict the set of candidate encodings to those which
have the same input constraints as the original encoding choice. This prevents
situations where relaxation prefers a non-REX-prefixed encoding over a REX
prefixed one because the end of the instruction can be one byte closer to the
destination, in a situation where the encoding needs to be REX-prefixed
because of one of the operand registers.
This also makes the Context class perform encoding verification after
relaxation, to catch similar problems in the future.
Fixes#256.
