433629 - add Box helpers that return Box::retType, for tail-call situations

Status: VERIFIED WONTFIX

(Tamarin Graveyard :: Tracing Virtual Machine, defect, P4)

Description

[Edwin Smith]

Attachment: first draft, not well tested.

the cliche':

Box b;
b.setAtom(...)
return b

the new way

return Box::fromAtom(...)

some unconvincing numbers, except for heapsort, which gets much better. (but which also is unstable.. sometimes its worse).  winxp, core 2 duo.

                                ../tt   .
boids                           5469    5469
boidshack                       875     860
gameoflife                      6000    6000
access-binary-trees             78      78
access-fannkuch                 156     156
access-nbody                    203     172
access-nsieve                   63      78
bitops-3bit-bits-in-byte        15      15
bitops-bits-in-byte             47      47
bitops-bitwise-and              282     265
bitops-nsieve-bits              62      62
crypto-md5                      546     516
crypto-sha1                     47      47
math-cordic                     63      62
math-partial-sums               235     218
math-spectral-norm              31      31
s3d-cube                        172     172
s3d-morph                       94      78
s3d-raytrace                    250     250
string-fasta                    172     156
controlflow-recursive           15      32
chess                           19      18
Crypt                           1078    1047
SOR                             1203    1156
HeapSort                        11422   9625

Comment 1

[Steven Johnson]

in principle, looks nice...

(1) in GCC, marking both inline and FASTCALL prevents inlining, I think
(2) maybe it's tme to choose a new name instead of RetType, since it's not just for ret-ing anymore

Comment 2

[Edwin Smith]

these changes were pushed with other bugfixes mostly by stejohns, making this bug obsolete.