This implements `airSplat` for the native WebAssembly backend when
the features 'simd128' or 'relaxed-simd' are enabled. The commit
supports splat where the value lives in the linear memory segment,
as well as on the stack. This saves a lot of instruction cost.
When it detects the element type is not 8, 16, 32 or 64 bits,
the backend will instead use the same strategy as if the features
where disabled.
This adds the opcodes for both the simd128 and relaxed-simd features.
Those instructions are required by the self-hosted WebAssembly
backend. Additionally, this also adds the new `v128` Valtype which
is required to represent a 128bit simd value. SIMD values that do
not have exactly 128 bits will be represented differently.
This fixes a class of bugs on macOS where a segfault happening in
a loaded dylib with no debug info would cause a panic in the panic
handler instead of simply noting that the dylib has no valid debug
info via `error.MissingDebugInfo`. An example could be code linking
some system dylib and causing some routine to segfault on say invalid
pointer value, which should normally cause Zig to print an incomplete
stack trace anchored at the currently loaded image and backtrace all
the way back to the Zig binary with valid debug info. Currently, in
a situation like this we would trigger a panic within a panic.
On x86 interestingly I can see a reduction in codesize by 1 instruction with this.
While not necessarily faster, it might still reduce codesize a bit and this ordering is also more logical
because it follows ASCII table order. Rust's std uses this ordering too.
See https://zig.godbolt.org/z/PqodY8YqY for the difference.
This makes it so that we no longer use a LUT (Look-Up Table):
* The code is much simpler and easier to understand now.
* Using a LUT means we rely on a warm cache.
Relying on the cache like this results in inconsistent performance and in many cases codegen will be worse.
Also as @topolarity once pointed out, in some cases while it seems like the code may branch, it actually doesn't:
https://github.com/ziglang/zig/pull/11629#issuecomment-1213641429
* Other languages' standard libraries don't do this either.
JFF I wanted to see what other languages codegen compared to us now:
https://rust.godbolt.org/z/Te4ax9Edf, https://zig.godbolt.org/z/nTbYedWKv
So we are pretty much on par or better than other languages now.
* Improve and remove duplicate doNotOptimizeAway() implementations
We currently have two doNotOptimizeAway() implementations, one in
std.math and the other one in std.mem.
Maybe we should deprecate one. In the meantime, the std.math one
now just calls the std.mem one.
In a comptime environment, just ignore the value. Previously,
std.mem.doNotOptimizeAway() did not work at comptime.
If the value fits in a CPU register, just tell the compiler we
need that value to be computed, without clobbering anything else.
Only clobber all possibly escaped memory on pointers or large arrays.
Add tests by the way since we didn't had any (we had, but only
indirect ones).
- In #13720, expectEqualBytes was added as a standalone function
- In #13723, expectEqualSlices was made to use expectEqualBytes when the type was u8
- In this commit, expectEqualSlices has fully absorbed expectEqualBytes, and expectEqualBytes itself has been removed
For non-`u8` types, expectEqualSlices will now work similarly to expectEqualBytes (highlighting diffs in red), but will use a full line for each index and therefore will only print a maximum of 16 indexes.
This reverts commit d4adf4420071397d993bac629a9da27b33c67ca3.
Unfortunately, this is not the right place to check if AES functions
are being used at comptime or not.
There are still a few occurrences of "stage1" in the standard library
and self-hosted compiler source, however, these instances need a bit
more careful inspection to ensure no breakage.
This branch largely reverts 58f961f4cb9875bbce3070969438ecf08f392c9f. I
would like to revisit the proposal to modify the standard library in
this way and think more carefully about it before adding isAbsolute()
checks everywhere.
Instead of checking for absolute paths and current working directories
in various file system operations, there is one simple solution: allow
overriding `std.fs.cwd` on WASI.
os.realpath is back to causing a compile error when used on WASI. This
caused a compile error in the Sema handling of `@src()`. The compiler
should never call realpath, so the commit that made this change is
reverted (95ab942184427e7c9b840d71f4d093931e3e48fb). If this breaks
debug info, a different strategy is needed to solve it other than using
realpath.
I also removed the preopens code and replaced it with something much
simpler. There is no longer any global state in the standard library.
Additionally-
* os.openat no longer does an unnecessary fstat on WASI when O.WRONLY
is not provided.
* os.chdir is back to causing a compile error on WASI.
