Michael found a nice trick to implement sets for float compares

[libfirm] / ir / be / ia32 / ia32_architecture.c

diff --git a/ir/be/ia32/ia32_architecture.c b/ir/be/ia32/ia32_architecture.c
index 351aebb..6e3e141 100644 (file)
--- a/ir/be/ia32/ia32_architecture.c
+++ b/ir/be/ia32/ia32_architecture.c
@@ -130,6 +130,7 @@ enum cpu_support {
 };
 
 static int         opt_size             = 0;
+static int         emit_machcode        = 0;
 static cpu_support arch                 = cpu_generic;
 static cpu_support opt_arch             = cpu_generic;
 static int         use_sse2             = 0;
@@ -220,6 +221,8 @@ static const lc_opt_table_entry_t ia32_architecture_options[] = {
                          &opt_cc, 1),
        LC_OPT_ENT_BIT("unsafe_floatconv", "do unsafe floating point controlword "
                       "optimisations", &opt_unsafe_floatconv, 1),
+       LC_OPT_ENT_BOOL("machcode", "output machine code instead of assembler",
+                       &emit_machcode),
        LC_OPT_LAST
 };
 
@@ -509,8 +512,10 @@ void ia32_setup_cg_config(void)
        c->use_sse_prefetch     = FLAGS(arch, (arch_feature_3DNowE | arch_feature_sse1));
        c->use_3dnow_prefetch   = FLAGS(arch, arch_feature_3DNow);
        c->use_popcnt           = FLAGS(arch, (arch_feature_sse4_2 | arch_feature_sse4a));
+       c->use_i486             = (arch & arch_mask) >= arch_i486;
        c->optimize_cc          = opt_cc;
        c->use_unsafe_floatconv = opt_unsafe_floatconv;
+       c->emit_machcode        = emit_machcode;
 
        c->function_alignment       = arch_costs->function_alignment;
        c->label_alignment          = arch_costs->label_alignment;