Summary:
This avoids a ptrtoint by just using the clang builtin. This is clang
specific but only clang can compile GPU code anyway so I do not bother
with a fallback.
Summary:
Now that we have `malloc` we can implement `realloc` efficiently. This
uses the known chunk sizes to avoid unnecessary allocations. We just
return nullptr for NVPTX. I'd remove the list for the entrypoint but
then the libc++ code would stop working. When someone writes the NVPTX
support this will be trivial.
Summary:
In the GPU allocator we reinterpret cast from a void pointer. We know
that an actual object was constructed there according to the C++ object
model, but to make it fully standards compliant we need to 'launder' it
to forward that information to the compiler. Add this function and call
it as appropriate.
This implementation has been compiled with the [pigweed toolchain](https://pigweed.dev/toolchain.html) and tested on:
- Raspberry Pi Pico 2 with the following options\
`--target=armv8m.main-none-eabi`
`-march=armv8m.main+fp+dsp`
`-mcpu=cortex-m33`
- Raspberry Pi Pico with the following options\
`--target=armv6m-none-eabi`
`-march=armv6m`
`-mcpu=cortex-m0+`
They both compile down to a little bit more than 200 bytes and are between 2 and 10 times faster than byte per byte copies.
For best performance the following options can be set in the `libc/config/baremetal/arm/config.json`
```
{
"codegen": {
"LIBC_CONF_KEEP_FRAME_POINTER": {
"value": false
}
},
"general": {
"LIBC_ADD_NULL_CHECKS": {
"value": false
}
}
}
```
Invalid mbstate_t should set errno to EINVAL.
Changed Error return for the internal functions and added tests for the
public functions.
Co-authored-by: Sriya Pratipati <sriyap@google.com>
This PR fixes the feature detection for RISC-V floating-point support in
LLVM's libc implementation.
The `__riscv_flen` macro represents the floating-point register width in
bits (32, 64, or 128). Since Extension D is specifically documented as
implying F, we can use simple >= comparisons to detect them.
For half-precision support, the implementation checks for the Zfhmin
extension as RVA22 and RVA23 profiles only require Zfhmin rather than
the full Zfh extension. Zfh also implies Zfhmin, so checking for Zfhmin
should cover all cases.
isComplete previously meant different things for different conversion
directions.
Refactored bytes_processed to bytes_stored which now consistently
increments on every push and decrements on pop making both directions
more consistent with each other
Updating x87 floating point register significantly affect the
performance of the functions.
All the floating point exception reads will merge the results from both
mxcsr and x87 registers anyway.
If you enable `LIBC_CONF_PRINTF_FLOAT_TO_STR_USE_FLOAT320` and use a
`%f` style printf format directive to print a nonzero number too small
to show up in the output digits, e.g. `printf("%.2f", 0.001)`, then the
output would be intermittently incorrect, because
`DyadicFloat::as_mantissa_type_rounded` would try to shift the 320-bit
mantissa right by more than 320 bits, invoking the 'undefined behavior'
clause commented in the `shift()` function in `big_int.h`.
There were already tests in the libc test suite exercising this case,
e.g. the subnormal tests in `LlvmLibcSPrintfTest.FloatDecimalConv` use
`%f` at the default precision of 6 decimal places on tiny numbers such
as 2^-1027. But because the behavior is undefined, they don't visibly
fail all the time, and in all previous test runs we'd tried with
USE_FLOAT320, they had got lucky.
The fix is simply to detect an out-of-range right shift before doing it,
and instead just set the output value to zero.
In #94078, `write_to_stdout` had not been fully implemented. However,
now that it has been implemented, to conform with the C standard
(7.23.6.3. The printf function, specifically point 2), we use `stdout`.
This issue is tracked in #94685.
- Also prefer `static constexpr`
- Made it explicit that we are writing to `stdout`
Removed strcmp, strlen, and memset calls from table.h and replaced them
with internal functions.
---------
Co-authored-by: Sriya Pratipati <sriyap@google.com>
The previous internal fcntl implementation modified errno directly, this
patch fixes that. This patch also moves open and close into OSUtil since
they are used in multiple places. There are more places that need
similar cleanup but only got comments in this patch to keep it
relatively reviewable.
Related to: https://github.com/llvm/llvm-project/issues/143937
Summary:
We need to set the bitfield memory to zero because the system does not
guarantee zeroed out memory. Even if fresh pages are zero, the system
allows re-use so we would need a `kfd` level API to skip this step.
Because we can't this patch updates the logic to perform the zero
initialization wave-parallel. This reduces the amount of time it takes
to allocate a fresh by up to a tenth.
This has the unfortunate side effect that the control flow is more
convoluted and we waste some extra registers, but it's worth it to
reduce the slab allocation latency.
After changing mbstate_t to mbstate we forgot to change the
character_converter files to reflect it.
Co-authored-by: Sriya Pratipati <sriyap@google.com>
