Re: [PATCH] Fix PR70726

[Jakub Jelinek]

On Tue, Apr 19, 2016 at 01:03:51PM +0200, Richard Biener wrote:
> I am testing the following to fix PR70726.
> 
> Bootstrap and regtest running on x86_64-unknown-linux-gnu.
> 
> Richard.
> 
> 2016-04-19  Richard Biener  
> 
>   PR tree-optimization/70726
>   * tree-vect-stmts.c (vectorizable_shift): Do not use scalar
>   shift amounts from a pattern stmt operand.
> 
>   * g++.dg/vect/pr70726.cc: New testcase.

Looks reasonable to me.

Jakub

[PATCH] Fix PR70726

[Richard Biener]

I am testing the following to fix PR70726.

Bootstrap and regtest running on x86_64-unknown-linux-gnu.

Richard.

2016-04-19  Richard Biener  

PR tree-optimization/70726
* tree-vect-stmts.c (vectorizable_shift): Do not use scalar
shift amounts from a pattern stmt operand.

* g++.dg/vect/pr70726.cc: New testcase.

Index: gcc/tree-vect-stmts.c
===
*** gcc/tree-vect-stmts.c   (revision 235188)
--- gcc/tree-vect-stmts.c   (working copy)
*** vectorizable_shift (gimple *stmt, gimple
*** 4532,4537 
--- 4532,4547 
if (!operand_equal_p (gimple_assign_rhs2 (slpstmt), op1, 0))
  scalar_shift_arg = false;
}
+ 
+   /* If the shift amount is computed by a pattern stmt we cannot
+  use the scalar amount directly thus give up and use a vector
+shift.  */
+   if (dt[1] == vect_internal_def)
+   {
+ gimple *def = SSA_NAME_DEF_STMT (op1);
+ if (is_pattern_stmt_p (vinfo_for_stmt (def)))
+   scalar_shift_arg = false;
+   }
  }
else
  {
Index: gcc/testsuite/g++.dg/vect/pr70726.cc
===
*** gcc/testsuite/g++.dg/vect/pr70726.cc(revision 0)
--- gcc/testsuite/g++.dg/vect/pr70726.cc(working copy)
***
*** 0 
--- 1,19 
+ // { dg-do compile }
+ // { dg-additional-options "-Ofast" }
+ // { dg-additional-options "-mavx2" { target x86_64-*-* i?86-*-* } }
+ 
+ extern long a;
+ extern int b[100];
+ extern unsigned c[5][5][2][4][2][2][3];
+ void fn1() {
+   for (int d = 0; d < 2; d = d + 1)
+ for (int e = 0; e < 5; e = e + 1)
+   for (int f = 0; f < 3; f = f + 1)
+ for (int g = 0; g < 3; g = g + 1)
+   for (int h = 0; h < 2; h = h + 1)
+ for (int i = 0; i < 4; i = i + 1)
+   for (int j = 0; j < 2; j = j + 1)
+ for (int k = 0; k < 2; k = k + 1)
+   for (int l = 0; l < 3; l = l + 1)
+ c[d][e][h][i][j][k][l] = a << b[f * 5 + g] + 4;
+ }