In this PR the problem was that when dealing with the gimple assign in the tailcall optimization, we, when the rhs operand is of a vector type, need to create -1 also of a vector type, but build_int_cst doesn't create vectors (ICEs). Instead, we should use build_minus_one_cst because that can create even the VECTOR_TYPE constant (and, it can create even REAL_TYPE/COMPLEX_TYPE), as suggested by Marc.
Regtested/bootstrapped on x86_64-linux, ok for trunk and 4.8? 2013-08-09 Marek Polacek <pola...@redhat.com> Marc Glisse <marc.gli...@inria.fr> PR tree-optimization/57980 * tree-tailcall.c (process_assignment): Call build_minus_one_cst when creating -1 constant. * gcc.dg/pr57980.c: New test. --- gcc/tree-tailcall.c.mp 2013-08-09 16:07:55.778616996 +0200 +++ gcc/tree-tailcall.c 2013-08-09 16:08:00.068635823 +0200 @@ -326,11 +326,7 @@ process_assignment (gimple stmt, gimple_ return true; case NEGATE_EXPR: - if (FLOAT_TYPE_P (TREE_TYPE (op0))) - *m = build_real (TREE_TYPE (op0), dconstm1); - else - *m = build_int_cst (TREE_TYPE (op0), -1); - + *m = build_minus_one_cst (TREE_TYPE (op0)); *ass_var = dest; return true; @@ -339,11 +335,7 @@ process_assignment (gimple stmt, gimple_ *a = fold_build1 (NEGATE_EXPR, TREE_TYPE (non_ass_var), non_ass_var); else { - if (FLOAT_TYPE_P (TREE_TYPE (non_ass_var))) - *m = build_real (TREE_TYPE (non_ass_var), dconstm1); - else - *m = build_int_cst (TREE_TYPE (non_ass_var), -1); - + *m = build_minus_one_cst (TREE_TYPE (non_ass_var)); *a = fold_build1 (NEGATE_EXPR, TREE_TYPE (non_ass_var), non_ass_var); } --- gcc/testsuite/gcc.dg/pr57980.c.mp 2013-08-09 16:07:25.967485356 +0200 +++ gcc/testsuite/gcc.dg/pr57980.c 2013-08-09 16:07:17.967450526 +0200 @@ -0,0 +1,19 @@ +/* PR tree-optimization/57980 */ +/* { dg-do compile } */ +/* { dg-options "-O -foptimize-sibling-calls" } */ + +typedef int V __attribute__ ((vector_size (sizeof (int)))); +extern V f (void); + +V +bar (void) +{ + return -f (); +} + +V +foo (void) +{ + V v = { }; + return v - f (); +} Marek