README

TODO: Clarify and add more documentation all around.

All of the *.cpp files in this folder will be compiled under all compiler
flags specified by CPU_CAPABILITY_FLAGS in aten/src/ATen/CMakeLists.txt.

The purpose of this is to allow the compilation with various compiler
flags to enable features such as AVX instructions, while using runtime
dispatch, which makes sure only valid instructions will be used on any
given platform.

Vec256.h provides a generic implementation of a vec256 type that allows
the programmer to write code packing various primitives (such as floats)
within 256bit registers. vec256 defines various operators such as + and *
and provides functions to allow operations such as max, min, etc.

As an example ReduceOpsKernel.cpp implements a generic kernel_ that reduces
an entire array using a given associative binary operation such as +.

More explicity, calling kernel_ with template argument std::plus will cause
it to sum up the entire array into a single value.

ReduceOpsKernel.cpp uses the CPU_CAPABILITY_* macros to "know" under which
compiler flags it is currently compiled. This allows the programmer to write
generic code, which will be compiled under multipled compilation settings.

../ReduceOps.cpp now includes the header ReduceOpsKernel.h, which contains
a generic definition of sumImplAll. This function allows the user to reduce
over a dimension or all dimensions. The appropiate capability is chosen at
runtime using cpuinfo. If the current platform has avx, sumImpl will be set
to umImplAll<CPUCapability::AVX>.