mbox series

[0/3] Optimize CORE-MATH internal tables

Message ID 20250203212546.911043-1-adhemerval.zanella@linaro.org
Headers show
Series Optimize CORE-MATH internal tables | expand

Message

Adhemerval Zanella Netto Feb. 3, 2025, 9:24 p.m. UTC
Some implementation share internal table, which can be optimized
to avoid duplicate data.  The are small code or no changes for x86_64
and powerpc64le, which do not affect performance; but on aarch64 with
gcc-14 I see a slight better code generation due the usage of ldq for
floating point constant loading.

Adhemerval Zanella (3):
  math: Consolidate acosf and asinf internal tables
  math: Consolidate acoshf and asinhf internal tables
  math: Consolidate coshf and sinhf internal tables

 math/Makefile                             |   3 +
 sysdeps/ieee754/flt-32/e_acosf.c          |  36 ++-----
 sysdeps/ieee754/flt-32/e_acoshf.c         | 111 ++------------------
 sysdeps/ieee754/flt-32/e_asincosf_data.c  |  53 ++++++++++
 sysdeps/ieee754/flt-32/e_asincosf_data.h  |  37 +++++++
 sysdeps/ieee754/flt-32/e_asinf.c          |  38 ++-----
 sysdeps/ieee754/flt-32/e_coshf.c          |  38 ++-----
 sysdeps/ieee754/flt-32/e_sincoshf_data.c  |  49 +++++++++
 sysdeps/ieee754/flt-32/e_sincoshf_data.h  |  36 +++++++
 sysdeps/ieee754/flt-32/e_sinhf.c          |  37 ++-----
 sysdeps/ieee754/flt-32/s_asincoshf_data.c | 121 ++++++++++++++++++++++
 sysdeps/ieee754/flt-32/s_asincoshf_data.h |  36 +++++++
 sysdeps/ieee754/flt-32/s_asinhf.c         | 109 ++-----------------
 13 files changed, 382 insertions(+), 322 deletions(-)
 create mode 100644 sysdeps/ieee754/flt-32/e_asincosf_data.c
 create mode 100644 sysdeps/ieee754/flt-32/e_asincosf_data.h
 create mode 100644 sysdeps/ieee754/flt-32/e_sincoshf_data.c
 create mode 100644 sysdeps/ieee754/flt-32/e_sincoshf_data.h
 create mode 100644 sysdeps/ieee754/flt-32/s_asincoshf_data.c
 create mode 100644 sysdeps/ieee754/flt-32/s_asincoshf_data.h

Comments

Paul Zimmermann Feb. 4, 2025, 6:32 a.m. UTC | #1
Dear Adhemerval,

> Some implementation share internal table, which can be optimized
> to avoid duplicate data.  The are small code or no changes for x86_64

The are -> There are ?

Paul