Faster table compute in mpn_sec_powm

[Marco Bodrato]

Ciao,

Il Sab, 24 Marzo 2018 3:37 pm, Torbjörn Granlund ha scritto:
>   B^(2j)   = (B^j)^2      (i.e., even values of i)
>   B^(2j+1) = (B^(2j))*B   (i.e., odd values of i)

> I haven't benchmarked this change, but it is clearly never a slowdown.

maybe a slowdown on the few systems with a non-zero SQR_BASECASE_THRESHOLD?

May I suggest:

**
/tmp/extdiff.03cT8S/gmp-repo.2a3b146aee90/mpn/generic/sec_powm.c	2018-03-25
00:47:27.000000000 +0100
--- /home/bodrato/gmp-repo/mpn/generic/sec_powm.c	2018-03-26
03:21:22.401588338 +0200
***************
*** 116,128 ****
     size.  */
  #define mpn_local_sqr(rp,up,n,tp) mpn_sqr_basecase(rp,up,n)
  #else
  /* Else use mpn_sqr_basecase for its allowed sizes, else
mpn_mul_basecase.  */
  #define mpn_local_sqr(rp,up,n,tp) \
    do {									\
!     if (BELOW_THRESHOLD (n, SQR_BASECASE_LIM))				\
        mpn_sqr_basecase (rp, up, n);					\
      else								\
        mpn_mul_basecase(rp, up, n, up, n);				\
    } while (0)
  #endif

--- 116,129 ----
     size.  */
  #define mpn_local_sqr(rp,up,n,tp) mpn_sqr_basecase(rp,up,n)
  #else
  /* Else use mpn_sqr_basecase for its allowed sizes, else
mpn_mul_basecase.  */
  #define mpn_local_sqr(rp,up,n,tp) \
    do {									\
!     if (ABOVE_THRESHOLD (n, SQR_BASECASE_THRESHOLD) &&			\
! 	BELOW_THRESHOLD (n, SQR_BASECASE_LIM))				\
        mpn_sqr_basecase (rp, up, n);					\
      else								\
        mpn_mul_basecase(rp, up, n, up, n);				\
    } while (0)
  #endif

... but this threshold might be used also for the other definition of
_local_sqr ... moreover, both definitions discard the parameter tp, can we
get rid of it?

Ĝis,
m

-- 
http://bodrato.it/