Skip to content

Latest commit

 

History

History
159 lines (142 loc) · 4.64 KB

available-compiler-flags.md

File metadata and controls

159 lines (142 loc) · 4.64 KB
Raw

Available Compiler Flags

Prev | Table of Contents | Next

Convenient TOML lists:

All Flags

In your configuration file flit-config.toml (see FLiT Configuration File), you specify compiler flags for each of the compilers. Only one compiler flag will be used with one optimization level. If you want to have specific flag combinations, you can place it in the list, such as "-mavx2 -mfma -funsafe-math-optimizations". Below is the original default list for the supported compilers:

Flag GCC Clang Intel
-fassociative-math X X
-fcx-fortran-rules X
-fcx-limited-range X X
-fexcess-precision=fast X X
-fexcess-precision=standard X
-ffinite-math-only X X
-ffloat-store X X X
-ffp-contract=on X X
-fma X
-fmerge-all-constants X X X
-fno-trapping-math X X
-fp-model fast=1 X
-fp-model fast=2 X
-fp-model=double X
-fp-model=extended X
-fp-model=precise X
-fp-model=source X
-fp-model=strict X
-fp-port X
-freciprocal-math X X
-frounding-math X X X
-fsignaling-nans X X
-fsingle-precision-constant X X
-ftz X
-funsafe-math-optimizations X X
-march=core-avx2 X X
-mavx X X X
-mavx2 -mfma X X X
-mfpmath=sse -mtune=native X X X
-mp1 X
-no-fma X
-no-ftz X
-no-prec-div X
-prec-div X

For your convenience, here are toml-style lists that can be copied into your flit-config.toml file directly:

GCC

Note: in some versions of python-toml, there is a parsing bug when a list has an empty string in the middle. So simply put it at the end without a comma. This has been fixed in the latest version of python-toml.

You will likely want the empty string in your search space as it represents the absence of any switches.

switches_list = [
  '-fassociative-math',
  '-fcx-fortran-rules',
  '-fcx-limited-range',
  '-fexcess-precision=fast',
  '-ffinite-math-only',
  '-ffloat-store',
  '-ffp-contract=on',
  '-fmerge-all-constants',
  '-fno-trapping-math',
  '-freciprocal-math',
  '-frounding-math',
  '-fsignaling-nans',
  '-funsafe-math-optimizations',
  '-mavx',
  '-mavx2 -mfma',
  '-mfpmath=sse -mtune=native',
  ''
]

Clang

switches_list = [
  '-fassociative-math',
  '-fexcess-precision=fast',
  '-fexcess-precision=standard',
  '-ffinite-math-only',
  '-ffloat-store',
  '-ffp-contract=on',
  '-fmerge-all-constants',
  '-fno-trapping-math',
  '-freciprocal-math',
  '-frounding-math',
  '-fsignaling-nans',
  '-fsingle-precision-constant',
  '-funsafe-math-optimizations',
  '-march=core-avx2',
  '-mavx',
  '-mavx2 -mfma',
  '-mfpmath=sse -mtune=native',
  ''
]

Intel

switches_list = [
  '--use_fast_math',
  '-fcx-limited-range',
  '-ffloat-store',
  '-fma',
  '-fmerge-all-constants',
  '-fp-model fast=1',
  '-fp-model fast=2',
  '-fp-model=double',
  '-fp-model=except',
  '-fp-model=extended',
  '-fp-model=precise',
  '-fp-model=source',
  '-fp-model=strict',
  '-fp-port',
  '-frounding-math',
  '-fsingle-precision-constant',
  '-ftz',
  '-march=core-avx2',
  '-mavx',
  '-mavx2 -mfma',
  '-mfpmath=sse -mtune=native',
  '-mp1',
  '-no-fma',
  '-no-ftz',
  '-no-prec-div',
  '-prec-div',
  ''
]

Prev | Table of Contents | Next