On Mon, 28 Jan 2013, Jakub Jelinek wrote:
On Mon, Jan 28, 2013 at 04:41:31PM +0100, Marc Glisse wrote:
On Mon, 28 Jan 2013, Jakub Jelinek wrote:
2013-01-28 Jakub Jelinek <ja...@redhat.com>
PR tree-optimization/56125
* tree-ssa-math-opts.c (gimple_expand_builtin_pow): Don't optimize
pow(x,c) into sqrt(x) * powi(x, n/2) or
1.0 / (sqrt(x) * powi(x, abs(n/2))) if c is an integer or when
optimizing for size.
Don't optimize pow(x,c) into powi(x, n/3) * powi(cbrt(x), n%3) or
1.0 / (powi(x, abs(n)/3) * powi(cbrt(x), abs(n)%3)) if 2c is an
integer.
* gcc.dg/pr56125.c: New test.
The testcase now generates a library call to pow, like gcc-4.6. This
is correct, but I am surprised this is considered better than
leaving the original x/(n*n) unchanged... Should that be a different
PR?
The function in question is marked as cold, therefore it should be optimized
for size. The call to pow is certainly shorter than the sqrt,
multiplication, division etc.
There is no sqrt, x/(n*n) is just one mul and one div, whereas with the
call I see one mul, 3 movs to prepare for the call, and the call.
--
Marc Glisse
