Domagoj Šarić
Domagoj Šarić
Global cache blocking values and thread scheduling don't map optimally to the new big.LITTLE systems with 'global task scheduling' (where all CPU tasks are active at the same time).
Even after integration with the cpuinfo library the init_hwinfo() function still uses fixed hardcoded values for ARM CPUs. I guess the same cpuinfo-based code should be used for all architectures...
Please add ARM BF16 availability info.
cpuinfo seems to suppose that the functionality in subject is available only in AArch64 mode/builds yet this may not be correct: https://gcc.gnu.org/legacy-ml/gcc-patches/2013-01/msg01441.html https://bugzilla.mozilla.org/show_bug.cgi?id=1077321
### Description Under NDK the linked Clang issue https://github.com/llvm/llvm-project/issues/58933 is even worse - linking fails (recognizes the warning as error): C/C++: ld: error: call to _ZN2MB2NN6Namespace4ThingIfEC2ERKS3_ marked "dontcall-warn": Don't call...
Switch to using dependencies (such as kumi) as submodules (or packages or...) instead of as 'flattened detail headers'. (makes collaboration easier)
I often have a need to perform some transformation on a shape (or a pair of shapes) (implicitly when doing transformations on an ndarray which affects its shape) - and...
Somewhat related to https://github.com/jfalcou/kiwaku/issues/83. Please provide an build time option to (not) use/provide iostream functionality (eg. ifdef wrapping of std::ostream& operator
via constructors or make/factory function (to avoid std::vector style () vs {} construction ambiguities)