This fixes a GRAPHITE code-generation issue by eliding ISL AST plus
for large power-of-two values that don't affect the result.
I intentionally didn't extend this to other values with the same
property as I'd like to see testcases.
Bootstrap and regtest running on x86_64-unknown-linux-gnu.
Richard.
2018-01-23 Richard Biener <rguent...@suse.de>
PR tree-optimization/82819
* graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
code generating pluses that are no-ops in the target precision.
* gcc.dg/graphite/pr82819.c: New testcase.
Index: gcc/graphite-isl-ast-to-gimple.c
===================================================================
--- gcc/graphite-isl-ast-to-gimple.c (revision 256977)
+++ gcc/graphite-isl-ast-to-gimple.c (working copy)
@@ -326,7 +326,8 @@ binary_op_to_tree (tree type, __isl_take
/* From our constraint generation we may get modulo operations that
we cannot represent explicitely but that are no-ops for TYPE.
Elide those. */
- if (expr_type == isl_ast_op_pdiv_r
+ if ((expr_type == isl_ast_op_pdiv_r
+ || expr_type == isl_ast_op_add)
&& isl_ast_expr_get_type (arg_expr) == isl_ast_expr_int
&& (wi::exact_log2 (widest_int_from_isl_expr_int (arg_expr))
>= TYPE_PRECISION (type)))
Index: gcc/testsuite/gcc.dg/graphite/pr82819.c
===================================================================
--- gcc/testsuite/gcc.dg/graphite/pr82819.c (nonexistent)
+++ gcc/testsuite/gcc.dg/graphite/pr82819.c (working copy)
@@ -0,0 +1,22 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -floop-nest-optimize" } */
+
+short int *ts;
+
+void
+c2 (unsigned long long int s4, int ns)
+{
+ short int *b2 = (short int *)&ns;
+
+ while (ns != 0)
+ {
+ int xn;
+
+ for (xn = 0; xn < 3; ++xn)
+ for (*b2 = 0; *b2 < 2; ++*b2)
+ s4 += xn;
+ if (s4 != 0)
+ b2 = ts;
+ ++ns;
+ }
+}
