Toom sqr recursion

[Torbjorn Granlund]

Torbjorn Granlund <tg at gmplib.org> writes:

  And in toom44_mul.c and toom4_sqr.c we set MAYBE_*_toom4* using
  MUL_FFT_THRESHOLD when we should really consider TOOM6 or (if TOOM6 = 0)
  TOOM8.  This was forgotten when the 6H and 8H code was merged.

Here's is a possible patch set:

diff -Nrc2 gmp-powm-powm_ui.8ec61b8b6882/mpn/generic/toom44_mul.c gmp-powm-powm_ui/mpn/generic/toom44_mul.c
*** gmp-powm-powm_ui.8ec61b8b6882/mpn/generic/toom44_mul.c	Sun Jan 27 22:05:24 2013
--- gmp-powm-powm_ui/mpn/generic/toom44_mul.c	Sun Jan 27 22:05:24 2013
***************
*** 51,54 ****
--- 51,61 ----
  #define MAYBE_mul_toom44   1
  #else
+ #if MUL_TOOM6H_THRESHOLD != 0
+ #define MUL_NEXTALG_THRESHOLD MUL_TOOM6H_THRESHOLD
+ #elif MUL_TOOM8H_THRESHOLD != 0
+ #define MUL_NEXTALG_THRESHOLD MUL_TOOM8H_THRESHOLD
+ #else
+ #define MUL_NEXTALG_THRESHOLD MUL_FFT_THRESHOLD
+ #endif
  #define MAYBE_mul_basecase						\
    (MUL_TOOM44_THRESHOLD < 4 * MUL_TOOM22_THRESHOLD)
***************
*** 56,60 ****
    (MUL_TOOM44_THRESHOLD < 4 * MUL_TOOM33_THRESHOLD)
  #define MAYBE_mul_toom44						\
!   (MUL_FFT_THRESHOLD >= 4 * MUL_TOOM44_THRESHOLD)
  #endif
  
--- 63,67 ----
    (MUL_TOOM44_THRESHOLD < 4 * MUL_TOOM33_THRESHOLD)
  #define MAYBE_mul_toom44						\
!   (MUL_NEXTALG_THRESHOLD >= 4 * MUL_TOOM44_THRESHOLD)
  #endif
  
diff -Nrc2 gmp-powm-powm_ui.8ec61b8b6882/mpn/generic/toom4_sqr.c gmp-powm-powm_ui/mpn/generic/toom4_sqr.c
*** gmp-powm-powm_ui.8ec61b8b6882/mpn/generic/toom4_sqr.c	Sun Jan 27 22:05:24 2013
--- gmp-powm-powm_ui/mpn/generic/toom4_sqr.c	Sun Jan 27 22:05:24 2013
***************
*** 48,51 ****
--- 48,58 ----
  #define MAYBE_sqr_toom4   1
  #else
+ #if SQR_TOOM6H_THRESHOLD != 0
+ #define SQR_NEXTALG_THRESHOLD SQR_TOOM6H_THRESHOLD
+ #elif SQR_TOOM8H_THRESHOLD != 0
+ #define SQR_NEXTALG_THRESHOLD SQR_TOOM8H_THRESHOLD
+ #else
+ #define SQR_NEXTALG_THRESHOLD SQR_FFT_THRESHOLD
+ #endif
  #define MAYBE_sqr_basecase						\
    (SQR_TOOM4_THRESHOLD < 4 * SQR_TOOM2_THRESHOLD)
***************
*** 53,57 ****
    (SQR_TOOM4_THRESHOLD < 4 * SQR_TOOM3_THRESHOLD)
  #define MAYBE_sqr_toom4							\
!   (SQR_FFT_THRESHOLD >= 4 * SQR_TOOM4_THRESHOLD)
  #endif
  
--- 60,64 ----
    (SQR_TOOM4_THRESHOLD < 4 * SQR_TOOM3_THRESHOLD)
  #define MAYBE_sqr_toom4							\
!   (SQR_NEXTALG_THRESHOLD >= 4 * SQR_TOOM4_THRESHOLD)
  #endif
  


-- 
Torbjörn