This commit helps with platform code separation (fixes#3). It should also help with the build as we can simply include all `impl_*.c` files regardless of OS / arch.
Note: this patch contains breaking changes in `include/cpu_features_macros.h`
- `CPU_FEATURES_OS_LINUX_OR_ANDROID` does not exist anymore
- `CPU_FEATURES_OS_FREEBSD`, `CPU_FEATURES_OS_ANDROID` and `CPU_FEATURES_OS_LINUX` are now mutually exclusive (i.e. `CPU_FEATURES_OS_ANDROID` does not imply `CPU_FEATURES_OS_LINUX`)
- `CPU_FEATURES_OS_DARWIN` has been renamed into `CPU_FEATURES_OS_MACOS` to be able to target non-Mac Apple products (IOS, TV, WATCH). They are now targetable with `CPU_FEATURES_OS_IPHONE`. This matches Apple naming convention described in [this stackoverflow](https://stackoverflow.com/a/49560690).
This is a non functional change, it allows:
- Getting rid of `unix_features_aggregator`
- Have a single blob describing the features
- Fix wrong mocking of `hwcaps`
Downside: abuse of macros makes the code slightly magical and harder to understand.
It think it's still an improvement over the current situation as there's less repetition and less chances to get something wrong.
Fixes#4. This is based on #115 with a few modifications:
- Removed use of __builtin_cpu_supports since it relies on cpuid and doesn't improve on the current situation,
- Added detection for all of sse, sse2, sse3, ssse3, sse4_1 and sse4_2,
- Added tests for Atom, Nehalem, and P3 processors,
Thx to @gadoofou87 for providing the original PR.
It also removes the need for #92
* Fix SSE detection on non-AVX CPUs
* Fixes typo
* Mock OSX sysctlbyname in tests
* Also update other tests
* FakeCpu is reset between each tests
* Fix conflicting name on Windows
* Disable pre AVX cpu sse detection tests on Windows
* Guard OS specific code with macros
* Fix missing import for tests
* Fix wrong function prototype
* Fix wrong mocking of P3 on Windows
* Completely guard OS specific parts in x86 tests
* Store DWORD instead unsigned long for x86 tests
* [NFC] clang-format codebase
* revert to 80 char columns at the price of uglier table init
* Specifically disabling clang-format for table initialization
* add Ice Lake Server and Sapphire Rapids models
The information contained in this commit was obtained from
"Intel® Architecture Instruction Set Extensions and Future Features Programming Reference" document 319433-040 from
https://software.intel.com/content/www/us/en/develop/download/intel-architecture-instruction-set-extensions-programming-reference.html
Signed-off-by: Jeff Hammond <jeff.r.hammond@intel.com>
* Tiger Lake; Ice Lake NNP-I; SPR string
Signed-off-by: Hammond, Jeff R <jeff.r.hammond@intel.com>
* add AVX512_BF16 and AVX512_VP2INTERSECT detection
Signed-off-by: Hammond, Jeff R <jeff.r.hammond@intel.com>
* correction for KNM features: s/4VBMI2/4FMAPS/g
Signed-off-by: Hammond, Jeff R <jeff.r.hammond@intel.com>
* add AMX/TMUL bits from 319433-040
Signed-off-by: Hammond, Jeff R <jeff.r.hammond@intel.com>
* add Intel copyright
Fixes#128
This adds the following features: dcpodp, sve2, sveaes, svepmull,
svebitperm, svesha3, svesm4, flagm2, frint, svei8mm, svef32mm,
svef64mm, svebf16, i8mm, bf16, dgh and rng.
With these, all features used by Linux 5.8 on AArch64 is supported.
Fixes#126
Visual C++ reported a number of warnings:
..\src\filesystem.c(27): warning C4996: '_open': This function or variable may be unsafe. Consider using _sopen_s instead. To disable deprecation, use _CRT_SECURE_NO_WARNINGS. See online help for details.
..\src\filesystem.c(34): warning C4267: 'function': conversion from 'size_t' to 'unsigned int', possible loss of data
..\src\string_view.c(25): warning C4244: 'return': conversion from '__int64' to 'int', possible loss of data
..\src\string_view.c(41): warning C4244: 'return': conversion from '__int64' to 'int', possible loss of data
..\src\utils\list_cpu_features.c(151): warning C4090: 'function': different 'const' qualifiers
..\src\cpuinfo_x86.c(67): warning C4244: 'return': conversion from 'unsigned __int64' to 'uint32_t', possible loss of data
..\test\filesystem_for_testing.cc(48): warning C4267: 'return': conversion from 'size_t' to 'int', possible loss of data
..\test\filesystem_for_testing.cc(57): warning C4267: 'argument': conversion from 'size_t' to 'int', possible loss of data
..\test\bit_utils_test.cc(26): warning C4267: 'argument': conversion from 'size_t' to 'uint32_t', possible loss of data
..\test\bit_utils_test.cc(32): warning C4267: 'argument': conversion from 'size_t' to 'uint32_t', possible loss of data
..\test\bit_utils_test.cc(37): warning C4267: 'argument': conversion from 'size_t' to 'uint32_t', possible loss of data
Most of them, with the exception of _open() -> _sopen_s() are about
questionable implicit casts. Fixing the types and adding explicit casts
to explicitly mark the intention (and silence the warnings in the process)
* Add support for leaf4 on Intel's x86 arch
* Update cpuinfo_x86.h
* Fix typo
* Force compiler to use C99
* Add Intel x86 leaf2 support
* Fixes after review
* Fix review comments
* First implementation of ndk_compat
* Restrict building of linux_based_hardware_detection to UNIX
* Fix variable declaration and printf formatting
* Restrict ndk compat to UNIX style systems
* Restrict cpu_mask index to 32
* Fix values display in ndk-compat-test
* Addressing comments
Some PowerPC machines can operate in a mode that appears different
to a process than the actual hardware. AT_PLATFORM indicates the
supported instruction set and AT_BASE_PLATFORM indicates the
actual microarchitecture of the hardware.
Signed-off-by: Rashmica Gupta <rashmica.gupta@au1.ibm.com>
