GitLab

Update cmake -- Ensure that the best vec type is taken from the origianl order Move SCALAR at the end of the reverse list

Remove for all vector types the operators on simd vector type

That is, for example, the previous version was offering operators `+-*/[=]` for native types (`__m128/256/512`) but we do not need them.
In fact, the user should use the inastemp classes instead (which provide these operators).
Therefore, for simplicity, we remove all of them but also the cmake detection.

See merge request !4

Remove for all vector types the operators on simd vector type (and explicitely use operand in the inastemp classes)

Add skylake support

create dependencies between types - add AVX 512 COMMON for knl and skl - add skl files

See merge request !2

Add skylake support -- create dependencies between types - add AVX 512 COMMON for knl and skl - add skl files

Update CI -- test with lots of archs using sde

Tests lots of archs for the CI.

This includes AVX (without AVX2) hardware for example.

See merge request !1

Update CMake -- ensure to check new capacities only if possible, and disable type if capacities are not detected

Core update -- use zero upper only in horizontal operations, but not in exp-avx because it leads to bug in GCC (where some registers are not saved and so are erased)

Update cmake and doc - update the detection of Inastemp included by another project and put the cmake options in the readme

Add GEMM example - might be improved in the future but already provide 40/20 GFlop-s in float/double

Update exponential - use const expr instead of array for coef and write factor computation in a single line

Add a test in the cmakelist to remove warnings when inastemp is not included by another cmake project (message was: current scope has no parent).