Re: [PATCH] Fix PR30028, fold sin, cos to sincos on the tree-level

[Paolo Bonzini <paolo dot bonzini at lu dot unisi dot ch>]

> Sure, I need to do _something_ anyway because of the cost issue.  The
> current code will happily transform
>
>  if (a)
>    s = sin (x);
>  else
>    c = cos (x);
>
> to use sincos.

This may actually not be too expensive at all with a serious 
implementation of sincos (i.e. fsincos, or one that vectorizes the 
calculation of sin and cos).  But I think it's better to first get the 
low-hanging fruit right (i.e. sin and cos in the same basic block).

> Probably not (all real-world testcases I looked at have sin and cos in
> the same BB only).  But the tree-ssa-math-opts pass is run after 
> tree-loop, so too late for PRE to clean up.  Was there any reason for
> this?  If not I'd move it at least before loop.

The reason is that LIM performs reciprocal hoisting on its own.

tree-ssa-math-opts.c is named like this because it is a relatively small 
pass.  It may include more passes (cse reciprocals running after loop, 
cse sin/cos running before loop, etc.).  That the two CSEs have 
different gate functions means that separating the passes might actually 
*improve* compilation time with some option combinations.

Paolo