https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95018
Bug ID: 95018
Summary: Excessive unrolling for Fortran library array handling
Product: gcc
Version: 11.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: tkoenig at gcc dot gnu.org
Target Milestone: ---
Created attachment 48488
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=48488&action=edit
Generated assembly
The code generated for in_pack_i4.c from libgfortran on POWER9 is huge
and, presumably, slow; I assume that other code in the library is
similarly affected.
The problem manifests itself in the unrolling of the loops which
do all the work:
while (src)
{
/* Copy the data. */
*(dest++) = *src;
/* Advance to the next element. */
src += stride0;
count[0]++;
/* Advance to the next source element. */
index_type n = 0;
while (count[n] == extent[n])
{
/* When we get to the end of a dimension, reset it and increment
the next dimension. */
count[n] = 0;
/* We could precalculate these products, but this is a less
frequently used path so probably not worth it. */
src -= stride[n] * extent[n];
n++;
if (n == dim)
{
src = NULL;
break;
}
else
{
count[n]++;
src += stride[n];
}
}
}
return destptr;
}
One problem here is the while (count[n] == extent[n]) loop.
This is an odometer algorithm to look for the next element to
go to. Most of the times, the while is false, so it is definitely
not good.
x86_64 does not appear to be affected.
