* [PATCH] Fix PR56787, missed vectorization with CLOBBERs
@ 2013-05-28 12:07 Richard Biener
2013-05-28 12:34 ` Jakub Jelinek
0 siblings, 1 reply; 3+ messages in thread
From: Richard Biener @ 2013-05-28 12:07 UTC (permalink / raw)
To: gcc-patches; +Cc: Michael Matz, Jakub Jelinek
The following fixes vectorization of loops with CLOBBER statements
in the way of ignoring them during analysis and removing them during
vectorization (because we ignored them for store/load placement
and do not handle them in case we need to unroll the loop body).
Bootstrap / regtest pending on x86_64-unknown-linux-gnu.
Do you see a better way of handling the CLOBBERs that is worthwhile?
Thanks,
Richard.
2013-05-28 Richard Biener <rguenther@suse.de>
PR tree-optimization/56787
* tree-vect-data-refs.c (vect_analyze_data_refs): Drop clobbers
from the list of data references.
* tree-vect-loop.c (vect_determine_vectorization_factor): Skip
clobbers.
(vect_analyze_loop_operations): Likewise.
(vect_transform_loop): Remove clobbers.
* gcc.dg/vect/pr56787.c: New testcase.
Index: gcc/tree-vect-data-refs.c
===================================================================
*** gcc/tree-vect-data-refs.c (revision 199374)
--- gcc/tree-vect-data-refs.c (working copy)
*************** vect_analyze_data_refs (loop_vec_info lo
*** 2861,2866 ****
--- 2861,2867 ----
bool gather = false;
int vf;
+ again:
if (!dr || !DR_REF (dr))
{
if (dump_enabled_p ())
*************** vect_analyze_data_refs (loop_vec_info lo
*** 2872,2877 ****
--- 2873,2891 ----
stmt = DR_STMT (dr);
stmt_info = vinfo_for_stmt (stmt);
+ /* Discard clobbers from the dataref vector. We will remove
+ clobber stmts during vectorization. */
+ if (gimple_clobber_p (stmt))
+ {
+ if (i == datarefs.length () - 1)
+ {
+ datarefs.pop ();
+ break;
+ }
+ datarefs[i] = datarefs.pop ();
+ goto again;
+ }
+
/* Check that analysis of the data-ref succeeded. */
if (!DR_BASE_ADDRESS (dr) || !DR_OFFSET (dr) || !DR_INIT (dr)
|| !DR_STEP (dr))
Index: gcc/tree-vect-loop.c
===================================================================
*** gcc/tree-vect-loop.c (revision 199374)
--- gcc/tree-vect-loop.c (working copy)
*************** vect_determine_vectorization_factor (loo
*** 270,277 ****
gcc_assert (stmt_info);
/* Skip stmts which do not need to be vectorized. */
! if (!STMT_VINFO_RELEVANT_P (stmt_info)
! && !STMT_VINFO_LIVE_P (stmt_info))
{
if (STMT_VINFO_IN_PATTERN_P (stmt_info)
&& (pattern_stmt = STMT_VINFO_RELATED_STMT (stmt_info))
--- 270,278 ----
gcc_assert (stmt_info);
/* Skip stmts which do not need to be vectorized. */
! if ((!STMT_VINFO_RELEVANT_P (stmt_info)
! && !STMT_VINFO_LIVE_P (stmt_info))
! || gimple_clobber_p (stmt))
{
if (STMT_VINFO_IN_PATTERN_P (stmt_info)
&& (pattern_stmt = STMT_VINFO_RELATED_STMT (stmt_info))
*************** vect_analyze_loop_operations (loop_vec_i
*** 1431,1437 ****
for (si = gsi_start_bb (bb); !gsi_end_p (si); gsi_next (&si))
{
gimple stmt = gsi_stmt (si);
! if (!vect_analyze_stmt (stmt, &need_to_vectorize, NULL))
return false;
}
} /* bbs */
--- 1432,1439 ----
for (si = gsi_start_bb (bb); !gsi_end_p (si); gsi_next (&si))
{
gimple stmt = gsi_stmt (si);
! if (!gimple_clobber_p (stmt)
! && !vect_analyze_stmt (stmt, &need_to_vectorize, NULL))
return false;
}
} /* bbs */
*************** vect_transform_loop (loop_vec_info loop_
*** 5595,5601 ****
if (transform_pattern_stmt)
stmt = pattern_stmt;
else
! stmt = gsi_stmt (si);
if (dump_enabled_p ())
{
--- 5597,5613 ----
if (transform_pattern_stmt)
stmt = pattern_stmt;
else
! {
! stmt = gsi_stmt (si);
! /* During vectorization remove existing clobber stmts. */
! if (gimple_clobber_p (stmt))
! {
! unlink_stmt_vdef (stmt);
! gsi_remove (&si, true);
! release_defs (stmt);
! continue;
! }
! }
if (dump_enabled_p ())
{
Index: gcc/testsuite/gcc.dg/vect/pr56787.c
===================================================================
*** gcc/testsuite/gcc.dg/vect/pr56787.c (revision 0)
--- gcc/testsuite/gcc.dg/vect/pr56787.c (working copy)
***************
*** 0 ****
--- 1,35 ----
+ /* { dg-do compile } */
+ /* { dg-require-effective-target vect_float } */
+
+ inline void
+ bar (const float s[5], float z[3][5])
+ {
+ float a = s[0], b = s[1], c = s[2], d = s[3], e = s[4];
+ float f = 1.0f / a;
+ float u = f * b, v = f * c, w = f * d;
+ float p = 0.4f * (e - 0.5f * (b * u + c * v + d * w));
+ z[0][3] = b * w;
+ z[1][3] = c * w;
+ z[2][3] = d * w + p;
+ }
+
+ void
+ foo (unsigned long n, const float *__restrict u0,
+ const float *__restrict u1, const float *__restrict u2,
+ const float *__restrict u3, const float *__restrict u4,
+ const float *__restrict s0, const float *__restrict s1,
+ const float *__restrict s2, float *__restrict t3,
+ float *__restrict t4)
+ {
+ unsigned long i;
+ for (i = 0; i < n; i++)
+ {
+ float u[5], f[3][5];
+ u[0] = u0[i]; u[1] = u1[i]; u[2] = u2[i]; u[3] = u3[i]; u[4] = u4[i];
+ bar (u, f);
+ t3[i] = s0[i] * f[0][3] + s1[i] * f[1][3] + s2[i] * f[2][3];
+ }
+ }
+
+ /* { dg-final { scan-tree-dump "vectorized 1 loops" "vect" } } */
+ /* { dg-final { cleanup-tree-dump "vect" } } */
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] Fix PR56787, missed vectorization with CLOBBERs
2013-05-28 12:07 [PATCH] Fix PR56787, missed vectorization with CLOBBERs Richard Biener
@ 2013-05-28 12:34 ` Jakub Jelinek
2013-05-28 12:38 ` Richard Biener
0 siblings, 1 reply; 3+ messages in thread
From: Jakub Jelinek @ 2013-05-28 12:34 UTC (permalink / raw)
To: Richard Biener; +Cc: gcc-patches, Michael Matz
On Tue, May 28, 2013 at 02:07:25PM +0200, Richard Biener wrote:
> The following fixes vectorization of loops with CLOBBER statements
> in the way of ignoring them during analysis and removing them during
> vectorization (because we ignored them for store/load placement
> and do not handle them in case we need to unroll the loop body).
>
> Bootstrap / regtest pending on x86_64-unknown-linux-gnu.
>
> Do you see a better way of handling the CLOBBERs that is worthwhile?
I think this is fine. Do you plan to backport it to 4.8.2 eventually?
> 2013-05-28 Richard Biener <rguenther@suse.de>
>
> PR tree-optimization/56787
> * tree-vect-data-refs.c (vect_analyze_data_refs): Drop clobbers
> from the list of data references.
> * tree-vect-loop.c (vect_determine_vectorization_factor): Skip
> clobbers.
> (vect_analyze_loop_operations): Likewise.
> (vect_transform_loop): Remove clobbers.
>
> * gcc.dg/vect/pr56787.c: New testcase.
Jakub
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] Fix PR56787, missed vectorization with CLOBBERs
2013-05-28 12:34 ` Jakub Jelinek
@ 2013-05-28 12:38 ` Richard Biener
0 siblings, 0 replies; 3+ messages in thread
From: Richard Biener @ 2013-05-28 12:38 UTC (permalink / raw)
To: Jakub Jelinek; +Cc: gcc-patches, Michael Matz
On Tue, 28 May 2013, Jakub Jelinek wrote:
> On Tue, May 28, 2013 at 02:07:25PM +0200, Richard Biener wrote:
> > The following fixes vectorization of loops with CLOBBER statements
> > in the way of ignoring them during analysis and removing them during
> > vectorization (because we ignored them for store/load placement
> > and do not handle them in case we need to unroll the loop body).
> >
> > Bootstrap / regtest pending on x86_64-unknown-linux-gnu.
> >
> > Do you see a better way of handling the CLOBBERs that is worthwhile?
>
> I think this is fine. Do you plan to backport it to 4.8.2 eventually?
Yes.
Richard.
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2013-05-28 12:38 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2013-05-28 12:07 [PATCH] Fix PR56787, missed vectorization with CLOBBERs Richard Biener
2013-05-28 12:34 ` Jakub Jelinek
2013-05-28 12:38 ` Richard Biener
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).