Repository navigation

replace several assembly implementations with an intrinsics based implementation and add AVX512 - #1712

Closed

fenrus75 wants to merge 19 commits into

OpenMathLib:developfrom

Closed

replace several assembly implementations with an intrinsics based implementation and add AVX512#1712
fenrus75 wants to merge 19 commits into
OpenMathLib:developfrom
fenrus75:dgemv

Commits on Aug 4, 2018
Commits on Aug 5, 2018
- various code cleanups and comments
  fenrus75
  committed
  9c29524
- use named arguments in the inline asm
  fenrus75
  committed
  847bbd6
- Use intrinsics instead of inline asm
  fenrus75
  committed
  ae38fa5
- Add AVX512 support to DDOT
  fenrus75
  committed
  34d63df
- fix typo in dsymv avx512 code path
  fenrus75
  committed
  21c6220
- sdot_haswell: similar to ddot: turn into intrinsics based C code that supports AVX512
  fenrus75
  committed
  ef30a72
- saxpy_haswell: Use named arguments in inline asm
  fenrus75
  committed
  d866046
- write saxpy_haswell kernel using C intrinsics and don't disallow inlining
  fenrus75
  committed
  06ea72f
- saxpy_haswell: Add AVX512 support
  fenrus75
  committed
  850b73d
- saxpy_haswell: Go to a more compact intrinsics notation
  fenrus75
  committed
  7af8a54
- daxpy_haswell: Change to C+instrinsics + AVX512 to mimic the change to saxpy_haswell
  fenrus75
  committed
  93aa18b
- Convert dscal_haswell to intrinsics and add AVX512 support
  fenrus75
  committed
  b1cc69e