[Bug tree-optimization/66036] strided group loads are not vectorized

[Bug tree-optimization/66036] strided group loads are not vectorized

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66036

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2015-05-06
             Blocks|                            |53947
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org
     Ever confirmed|0                           |1
           Severity|normal                      |enhancement

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Mine.


Referenced Bugs:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=53947
[Bug 53947] [meta-bug] vectorizer missed-optimizations

[Bug tree-optimization/66036] New: strided group loads are not vectorized

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66036

            Bug ID: 66036
           Summary: strided group loads are not vectorized
           Product: gcc
           Version: 5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: rguenth at gcc dot gnu.org
  Target Milestone: ---

For example

struct Xd {
    double x;
    double y;
};
double testd (struct Xd *x, int stride, int n)
{
  int i;
  double sum = 0.;
  for (i = 0; i < n; ++i)
    { 
      sum += x[i*stride].x;
      sum += x[i*stride].y;
    }
  return sum;
}

or similar cases without reduction (simple case)

int testi (int *p, short *q, int stride, int n)
{
  int i;
  for (i = 0; i < n; ++i)
    {
      q[i*4+0] = p[i*stride+0];
      q[i*4+1] = p[i*stride+1];
      q[i*4+2] = p[i*stride+2];
      q[i*4+3] = p[i*stride+3];
    }
}

or the more complex case

int testi2 (int *q, short *p, int stride, int n)
{
  int i;
  for (i = 0; i < n; ++i)
    {
      q[i*4+0] = p[i*stride+0];
      q[i*4+1] = p[i*stride+1];
      q[i*4+2] = p[i*stride+2];
      q[i*4+3] = p[i*stride+3];
    }
}

because here the SLP group has smaller-than-vector size and thus requires
two "scalar" loads and a vector build from them (x86_64 movhlps/movulps).
The more complex form happens in SPEC CPUv6.

[Bug tree-optimization/66036] strided group loads are not vectorized

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66036

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Reduction case with unrolling

struct Xf {
    float x;
    float y;
};
float testf (struct Xf *x, int stride, int n)
{
  int i;
  float sum = 0.;
  for (i = 0; i < n; ++i)
    { 
      sum += x[i*stride].x;
      sum += x[i*stride].y;
    }
  return sum;
}

[Bug tree-optimization/66036] strided group loads are not vectorized

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66036

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed.

[Bug tree-optimization/66036] strided group loads are not vectorized

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66036

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
Author: rguenth
Date: Fri May  8 15:13:55 2015
New Revision: 222914

URL: https://gcc.gnu.org/viewcvs?rev=222914&root=gcc&view=rev
Log:
2015-05-08  Richard Biener  <rguenther@suse.de>

        PR tree-optimization/66036
        * tree-vect-data-refs.c (vect_compute_data_ref_alignment):
        Handle strided group loads.
        (vect_verify_datarefs_alignment): Likewise.
        (vect_enhance_data_refs_alignment): Likewise.
        (vect_analyze_group_access): Likewise.
        (vect_analyze_data_ref_access): Likewise.
        (vect_analyze_data_ref_accesses): Likewise.
        * tree-vect-stmts.c (vect_model_load_cost): Likewise.
        (vectorizable_load): Likewise.

        * gcc.dg/vect/slp-41.c: New testcase.

Added:
    trunk/gcc/testsuite/gcc.dg/vect/slp-41.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-vect-data-refs.c
    trunk/gcc/tree-vect-stmts.c