The implementations use the x86_64 FPU instructions. These instructions
are extremely slow compared to a polynomial based software
implementation. Also, their accuracy falls drastically once the input
goes beyond 2PI. To improve both the speed and accuracy, we will be
taking the following approach going forward:
1. As a follow up to this CL, we will implement a range reduction algorithm
which will expand the accuracy to the entire double precision range.
2. After that, we will replace the HW instructions with a polynomial
implementation to improve the run time.
After step 2, the implementations will be accurate, performant and target
architecture independent.
Reviewed By: lntue
Differential Revision: https://reviews.llvm.org/D102384
This is a roll forward of D101895 with two additional fixes:
Original Patch description:
> This is a follow up on D101524 which:
>
> - simplifies cpu features detection and usage,
> - flattens target dependent optimizations so it's obvious which implementations are generated,
> - provides an implementation targeting the host (march/mtune=native) for the mem* functions,
> - makes sure all implementations are unittested (provided the host can run them).
Additional fixes:
- Fix uninitialized ALL_CPU_FEATURES
- Use non pseudo microarch as it is only supported from Clang 12 on
Differential Revision: https://reviews.llvm.org/D102233
This reverts commit 541f107871 as the bots
are failing with unknown architecture "x86-64-v*". Will let the original
author decide on the right course of action to correct the problem and
reland.
This is a follow up on D101524 which:
- simplifies cpu features detection and usage,
- flattens target dependent optimizations so it's obvious which implementations are generated,
- provides an implementation targeting the host (march/mtune=native) for the mem* functions,
- makes sure all implementations are unittested (provided the host can run them),
- makes sure all implementations are benchmarkable (provided the host can run them).
Differential Revision: https://reviews.llvm.org/D101895
Current implementation defines LIBC_TARGET_MACHINE with the use of CMAKE_SYSTEM_PROCESSOR.
Unfortunately CMAKE_SYSTEM_PROCESSOR is OS dependent and can produce different results.
An evidence of this is the various matchers used to detect whether the architecture is x86.
This patch normalizes LIBC_TARGET_MACHINE and renames it LIBC_TARGET_ARCHITECTURE.
I've added many architectures but we may want to limit ourselves to x86 and ARM.
Differential Revision: https://reviews.llvm.org/D101524
[libc] Introduce asctime, asctime_r to LLVM libc
asctime and asctime_r share the same common code. They call asctime_internal
a static inline function.
asctime uses snprintf to return the string representation in a buffer.
It uses the following format (26 characters is the buffer size) as per
7.27.3.1 section in http://www.open-std.org/jtc1/sc22/wg14/www/docs/n2478.pdf.
The buf parameter for asctime_r shall point to a buffer of at least 26 bytes.
snprintf(buf, 26, "%.3s %.3s%3d %.2d:%.2d:%.2d %d\n",...)
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D99686
Aligned copy used to be 'destination aligned' for x86 but this decision was reverted in D93457 where we noticed that it was better for ARM to be 'source aligned'.
More benchmarking confirmed that it can be up to 30% faster to align copy to destination for x86. This Patch offers both implementations and switches x86 back to destination aligned.
It also fixes alignment to 32 byte on x86.
Differential Revision: https://reviews.llvm.org/D101296
The current generic implementation of the fmaf function has been moved
to the FPUtil directory. This allows one use the fma operation from
implementations of other math functions like the trignometric functions
without depending on/requiring the fma/fmaf/fmal function targets. If
this pattern ends being convenient, we will switch all generic math
implementations to this pattern.
Reviewed By: lntue
Differential Revision: https://reviews.llvm.org/D100811
This helps us avoid the uncomfortable reinterpret-casts. Avoiding the
reinterpret casts prevents us from tripping the sanitizers as well.
Reviewed By: lntue
Differential Revision: https://reviews.llvm.org/D100360
This is needed to prevent asan/msan instrumentation to redirect CopyBlock to `__asan_memcpy` (resp. `__msan_memcpy`).
These functions would then differ operation to `memcpy` which leads to reentrancy issues.
With this patch, `memcpy` is fully instrumented and covered by asan/msan.
If this turns out to be too expensive, instrumentation can be selectively or fully disabled through the use of the `__attribute__((no_sanitize(address, memory)))` annotation.
Differential Revision: https://reviews.llvm.org/D99598
gmtime and gmtime_r share the same common code. They call gmtime_internal
a static inline function. Thus added only validation tests for gmtime_r.
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D99046
This change doesn't handle TIMEZONE, tm_isdst and leap seconds.
Moved shared code between mktime and gmtime into time_utils.cpp.
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D98467
This option will build LLVM libc as a full libc by itself. In this mode,
it is not expected that it will be mixed with other libcs. The
non-full-build mode will be the default LLVM libc build mode. In a future
where LLVM libc is complete enough, the full libc build will be made the
default mode.
These functions used inline asm to read FPU state. This change adds
explicit unpoisoning in these functions as the sanitizers don't see the
read operations.
We want to be able to build and test the string functions in contexts
like that of Fuchsia where LLVM pieces like tablegen are not available.
Since header generation uses tablegen, we are removing the dependency on
headergen here.
Reviewed By: gchatelet
Differential Revision: https://reviews.llvm.org/D97363
Added tests for invalid dates like the following
Date 1970-01-01 00:00:-1 is treated as 1969-12-31 23:59:59 and seconds
are returned for the modified date.
Tested the code by doing ninja check-libc (and cmake).
Reviewed By: sivachandra, rtenneti
Differential Revision: https://reviews.llvm.org/D96684
Namely, implementations of fegetexceptfflag, fesetexceptflag,
fegetenv, fesetenv, feholdexcept and feupdateenv have been added.
Reviewed By: lntue
Differential Revision: https://reviews.llvm.org/D96935
This change also introduces a new source layout for adding machine
specific and generic implementations. To keep the scope of this change
small, this new pattern is only applied for ceil, ceilf and ceill.
Follow up changes will switch all math functions in to the new pattern.
Reviewed By: lntue
Differential Revision: https://reviews.llvm.org/D95850
`ssize_t` is from POSIX and is not standard unfortunately.
Rewritting the code so it doesn't depend on it.
Differential Revision: https://reviews.llvm.org/D94760
- Adds LLVM_LIBC_IS_DEFINED macro to libc/src/__support/common.h
- Adds a few knobs to memcpy to help with experimentations:
- LLVM_LIBC_MEMCPY_X86_USE_ONLY_REPMOVSB replaces the implementation with a single call to rep;movsb
- LLVM_LIBC_MEMCPY_X86_USE_REPMOVSB_FROM_SIZE customizes where the usage of rep;movsb
Differential Revision: https://reviews.llvm.org/D94692
Use `memcpy` rather than copying bytes one by one, for there might be large
size structs to move.
Reviewed By: gchatelet, sivachandra
Differential Revision: https://reviews.llvm.org/D93195
Standard C allows all standard headers to declare macros for all
their functions. So after possibly including any standard header
like <ctype.h>, it's perfectly normal for any and all of the
functions it declares to be defined as macros. Standard C requires
explicit `#undef` before using that identifier in a way that is not
compatible with function-like macro definitions.
The C standard's rules for this are extended to POSIX as well for
the interfaces it defines, and it's the expected norm for
nonstandard extensions declared by standard C library headers too.
So far the only place this has come up for llvm-libc's code is with
the isascii function in Fuchsia's libc. But other cases can arise
for any standard (or common extension) function names that source
code in llvm-libc is using in nonstandard ways, i.e. as C++
identifiers.
The only correct and robust way to handle the possible inclusion of
standard C library headers when building llvm-libc source code is to
use `#undef` explicitly for each identifier before using it. The
easy and obvious place to do that is in the per-function header.
This requires that all code, such as test code, that might include
any standard C library headers, e.g. via utils/UnitTest/Test.h, make
sure to include those *first* before the per-function header.
This change does that for isascii and its test. But it should be
done uniformly for all the code and documented as a consistent
convention so new implementation files are sure to get this right.
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D94642
isblank and iscntrl were casting an int to a char implicitly and this
was throwing errors under Fuchsia. I've added a static cast to resolve
this issue.
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D94634
It was previously a generated header. It can easily converted to a
generated header if required in future.
Reviewed By: michaelrj
Differential Revision: https://reviews.llvm.org/D94445
this will make sure that all of the functions are using the correct
prototypes. Explained much better in the comments of this diff:
https://reviews.llvm.org/D94195
Summary:
The new macro also inserts the C alias for the C++ implementations
without needing an objcopy based post processing step. The CMake
rules have been updated to reflect this. More CMake cleanup can be
taken up in future rounds and appropriate TODOs have been added for them.
Reviewers: mcgrathr, sivachandra
Subscribers:
We used to align destination buffer instead of source buffer for the loop of block copy.
This is a mistake.
Differential Revision: https://reviews.llvm.org/D93457
A differential fuzzer for these functions has also been added.
Along the way, a small correction has been done to the normal/subnormal
limits of x86 long double values.
Reviewed By: lntue
Differential Revision: https://reviews.llvm.org/D94109
The implementation is exactly the same as rint* as even rint does not
raise any floating point exceptions currently. [Note that the standards
do not specify that floating point exceptions must be raised - they
leave it up to the implementation to choose to raise FE_INEXACT when
rounding non-integral values.]
Reviewed By: lntue
Differential Revision: https://reviews.llvm.org/D94112
A new function to MPFRWrapper has been added, which is used to set up
the unit tests.
Reviewed By: lntue
Differential Revision: https://reviews.llvm.org/D93007
Refactor src/math/hypotf.cpp and test/src/math/hypotf_test.cpp and reuse them for hypot and hypot_test
Differential Revision: https://reviews.llvm.org/D91831
Siva Chandra Reddy