The following patch fixes us miscounting the number of scalar
instructions for BB vectorization leading to vectorizations that
are not profitable.
A simple fix is to count each scalar stmt at most once.
Bootstrapped and tested on x86_64-unknown-linux-gnu, applied.
Richard.
2016-01-18 Richard Biener <rguent...@suse.de>
PR tree-optimization/69297
* tree-vect-slp.c (vect_bb_slp_scalar_cost): Count each scalar
stmt at most once.
(vect_bb_vectorization_profitable_p): Clear visited flag again.
Index: gcc/tree-vect-slp.c
===================================================================
*** gcc/tree-vect-slp.c (revision 232496)
--- gcc/tree-vect-slp.c (working copy)
*************** vect_bb_slp_scalar_cost (basic_block bb,
*** 2409,2414 ****
--- 2409,2419 ----
if ((*life)[i])
continue;
+ /* Count scalar stmts only once. */
+ if (gimple_visited_p (stmt))
+ continue;
+ gimple_set_visited (stmt, true);
+
stmt_info = vinfo_for_stmt (stmt);
if (STMT_VINFO_DATA_REF (stmt_info))
{
*************** vect_bb_vectorization_profitable_p (bb_v
*** 2451,2456 ****
--- 2456,2466 ----
&life);
}
+ /* Unset visited flag. */
+ for (gimple_stmt_iterator gsi = bb_vinfo->region_begin;
+ gsi_stmt (gsi) != gsi_stmt (bb_vinfo->region_end); gsi_next (&gsi))
+ gimple_set_visited (gsi_stmt (gsi), false);
+
/* Complete the target-specific cost calculation. */
finish_cost (BB_VINFO_TARGET_COST_DATA (bb_vinfo), &vec_prologue_cost,
&vec_inside_cost, &vec_epilogue_cost);
Index: gcc/testsuite/gcc.dg/vect/costmodel/x86_64/costmodel-pr69297.c
===================================================================
*** gcc/testsuite/gcc.dg/vect/costmodel/x86_64/costmodel-pr69297.c
(revision 0)
--- gcc/testsuite/gcc.dg/vect/costmodel/x86_64/costmodel-pr69297.c
(working copy)
***************
*** 0 ****
--- 1,83 ----
+ /* { dg-do compile } */
+ /* { dg-additional-options "-march=core-avx2 -fdump-tree-slp-details" } */
+
+ #define abs(x) (x) < 0 ? -(x) : (x)
+ int
+ foo (int* diff)
+ {
+ int k, satd = 0, m[16], d[16];
+
+ m[ 0] = diff[ 0] + diff[12];
+ m[ 4] = diff[ 4] + diff[ 8];
+ m[ 8] = diff[ 4] - diff[ 8];
+ m[12] = diff[ 0] - diff[12];
+ m[ 1] = diff[ 1] + diff[13];
+ m[ 5] = diff[ 5] + diff[ 9];
+ m[ 9] = diff[ 5] - diff[ 9];
+ m[13] = diff[ 1] - diff[13];
+ m[ 2] = diff[ 2] + diff[14];
+ m[ 6] = diff[ 6] + diff[10];
+ m[10] = diff[ 6] - diff[10];
+ m[14] = diff[ 2] - diff[14];
+ m[ 3] = diff[ 3] + diff[15];
+ m[ 7] = diff[ 7] + diff[11];
+ m[11] = diff[ 7] - diff[11];
+ m[15] = diff[ 3] - diff[15];
+
+ d[ 0] = m[ 0] + m[ 4];
+ d[ 8] = m[ 0] - m[ 4];
+ d[ 4] = m[ 8] + m[12];
+ d[12] = m[12] - m[ 8];
+ d[ 1] = m[ 1] + m[ 5];
+ d[ 9] = m[ 1] - m[ 5];
+ d[ 5] = m[ 9] + m[13];
+ d[13] = m[13] - m[ 9];
+ d[ 2] = m[ 2] + m[ 6];
+ d[10] = m[ 2] - m[ 6];
+ d[ 6] = m[10] + m[14];
+ d[14] = m[14] - m[10];
+ d[ 3] = m[ 3] + m[ 7];
+ d[11] = m[ 3] - m[ 7];
+ d[ 7] = m[11] + m[15];
+ d[15] = m[15] - m[11];
+
+ m[ 0] = d[ 0] + d[ 3];
+ m[ 1] = d[ 1] + d[ 2];
+ m[ 2] = d[ 1] - d[ 2];
+ m[ 3] = d[ 0] - d[ 3];
+ m[ 4] = d[ 4] + d[ 7];
+ m[ 5] = d[ 5] + d[ 6];
+ m[ 6] = d[ 5] - d[ 6];
+ m[ 7] = d[ 4] - d[ 7];
+ m[ 8] = d[ 8] + d[11];
+ m[ 9] = d[ 9] + d[10];
+ m[10] = d[ 9] - d[10];
+ m[11] = d[ 8] - d[11];
+ m[12] = d[12] + d[15];
+ m[13] = d[13] + d[14];
+ m[14] = d[13] - d[14];
+ m[15] = d[12] - d[15];
+
+ d[ 0] = m[ 0] + m[ 1];
+ d[ 1] = m[ 0] - m[ 1];
+ d[ 2] = m[ 2] + m[ 3];
+ d[ 3] = m[ 3] - m[ 2];
+ d[ 4] = m[ 4] + m[ 5];
+ d[ 5] = m[ 4] - m[ 5];
+ d[ 6] = m[ 6] + m[ 7];
+ d[ 7] = m[ 7] - m[ 6];
+ d[ 8] = m[ 8] + m[ 9];
+ d[ 9] = m[ 8] - m[ 9];
+ d[10] = m[10] + m[11];
+ d[11] = m[11] - m[10];
+ d[12] = m[12] + m[13];
+ d[13] = m[12] - m[13];
+ d[14] = m[14] + m[15];
+ d[15] = m[15] - m[14];
+ for (k=0; k<16; k++)
+ satd += abs(d[k]);
+ return satd;
+ }
+
+ /* { dg-final { scan-tree-dump "vectorization is not profitable" "slp1" } } */
+ /* { dg-final { scan-tree-dump-not "basic block vectorized" "slp1" } } */
