fix --help display
[cparser] / driver / firm_opt.c
index 73923ab..ada3dda 100644 (file)
@@ -1,12 +1,9 @@
 /**
- *
- * @file firm_opt.c -- Firm-generating back end optimizations.
- *
- * (C) 2005-2009  Michael Beck   beck@ipd.info.uni-karlsruhe.de
- *
- * $Id$
+ * (C) 2005-2010
+ * @file
+ * @author Michael Beck, Matthias Braun
+ * @brief Firm-generating back end optimizations.
  */
-
 #include <config.h>
 
 #include <stdbool.h>
 #include <stdbool.h>
 #include <assert.h>
 #include <libfirm/firm.h>
-#include <libfirm/be.h>
-
 
 #include "firm_opt.h"
-#include "firm_codegen.h"
-#include "firm_cmdline.h"
 #include "firm_timing.h"
 #include "ast2firm.h"
+#include "adt/strutil.h"
+#include "adt/util.h"
+
+/* optimization settings */
+struct a_firm_opt {
+       bool     const_folding;   /**< enable constant folding */
+       bool     cse;             /**< enable common-subexpression elimination */
+       bool     confirm;         /**< enable Confirm optimization */
+       bool     muls;            /**< enable architecture dependent mul optimization */
+       bool     divs;            /**< enable architecture dependent div optimization */
+       bool     mods;            /**< enable architecture dependent mod optimization */
+       bool     alias_analysis;  /**< enable Alias Analysis */
+       bool     strict_alias;    /**< enable strict Alias Analysis (using type based AA) */
+       bool     no_alias;        /**< no aliasing possible. */
+       bool     verify;          /**< Firm verifier setting */
+       bool     check_all;       /**< enable checking all Firm phases */
+       int      clone_threshold; /**< The threshold value for procedure cloning. */
+       unsigned inline_maxsize;  /**< Maximum function size for inlining. */
+       unsigned inline_threshold;/**< Inlining benefice threshold. */
+};
 
-#if defined(_DEBUG) || defined(FIRM_DEBUG)
-#define DBG(x)  dbg_printf x
-#else
-#define DBG(x) ((void)0)
-#endif /* _DEBUG || FIRM_DEBUG */
-
-
-/** dump all the graphs depending on cond */
-#define DUMP_ALL(cond, suffix)                             \
-  do {                                                     \
-    if (cond) {                                            \
-      timer_push(TV_VCG_DUMP);                             \
-      if (firm_dump.no_blocks)                             \
-        dump_all_ir_graphs(dump_ir_graph, suffix);         \
-      else if (firm_dump.extbb)                            \
-        dump_all_ir_graphs(dump_ir_extblock_graph, suffix);\
-      else                                                 \
-        dump_all_ir_graphs(dump_ir_block_graph, suffix);   \
-      timer_pop();                                         \
-    }                                                      \
-  } while (0)
-
-/** dump all control flow graphs depending on cond */
-#define DUMP_ALL_CFG(cond, suffix)                      \
-  do {                                                  \
-    if (cond) {                                         \
-      timer_push(TV_VCG_DUMP);                          \
-        dump_all_ir_graphs(dump_cfg, suffix);           \
-      timer_pop();                                      \
-    }                                                   \
-  } while (0)
-
-/** check all graphs depending on cond */
-#define CHECK_ALL(cond)                                 \
-  do {                                                  \
-    if (cond) {                                         \
-      int ii;                                           \
-      timer_push(TV_VERIFY);                            \
-      for (ii = get_irp_n_irgs() - 1; ii >= 0; --ii)    \
-        irg_verify(get_irp_irg(ii), VRFY_ENFORCE_SSA);  \
-      timer_pop();                                      \
-    }                                                   \
-  } while (0)
-
-
-
-/** dump graphs irg depending on cond */
-#define DUMP_ONE(cond, irg, suffix)                     \
-  do {                                                  \
-    if (cond) {                                         \
-      timer_push(TV_VCG_DUMP);                          \
-      if (firm_dump.no_blocks)                          \
-        dump_ir_graph(irg, suffix);                     \
-      else if (firm_dump.extbb)                         \
-        dump_ir_extblock_graph(irg, suffix);            \
-      else                                              \
-        dump_ir_block_graph(irg, suffix);               \
-      timer_pop();                                      \
-    }                                                   \
-  } while (0)
-
-/** dump control flow graph irg depending on cond */
-#define DUMP_ONE_CFG(cond, irg, suffix)                 \
-  do {                                                  \
-    if (cond) {                                         \
-      timer_push(TV_VCG_DUMP);                          \
-      dump_cfg(irg, suffix);                            \
-      timer_pop();                                      \
-    }                                                   \
-  } while (0)
-
-/** check a graph irg depending on cond */
-#define CHECK_ONE(cond, irg)                            \
-  do {                                                  \
-    if (cond) {                                         \
-      timer_push(TV_VERIFY);                            \
-        irg_verify(irg, VRFY_ENFORCE_SSA);              \
-      timer_pop();                                      \
-    }                                                   \
-  } while (0)
-
-
-/* set by the backend parameters */
-static const ir_settings_arch_dep_t *ad_param = NULL;
-static create_intrinsic_fkt *arch_create_intrinsic = NULL;
-static void *create_intrinsic_ctx = NULL;
-static const ir_settings_if_conv_t *if_conv_info = NULL;
-
-/* entities of runtime functions */
-ir_entity_ptr rts_entities[rts_max];
+/** statistic options */
+typedef enum a_firmstat_selection_tag {
+       STAT_NONE        = 0x00000000,
+       STAT_BEFORE_OPT  = 0x00000001,
+       STAT_AFTER_OPT   = 0x00000002,
+       STAT_AFTER_LOWER = 0x00000004,
+       STAT_FINAL_IR    = 0x00000008,
+       STAT_FINAL       = 0x00000010,
+} a_firmstat_selection;
+
+/* dumping options */
+struct a_firm_dump {
+       bool debug_print;   /**< enable debug print */
+       bool all_types;     /**< dump the All_types graph */
+       bool ir_graph;      /**< dump all graphs */
+       bool all_phases;    /**< dump the IR graph after all phases */
+       bool statistic;     /**< Firm statistic setting */
+       bool stat_pattern;  /**< enable Firm statistic pattern */
+       bool stat_dag;      /**< enable Firm DAG statistic */
+};
 
-/**
- * factory for setting architecture dependent parameters
- */
-static const ir_settings_arch_dep_t *arch_factory(void)
-{
-  static const ir_settings_arch_dep_t param = {
-      1,   /* also use subs */
-      4,   /* maximum shifts */
-     31,   /* maximum shift amount */
-     NULL, /* use default evaluator */
-
-      1, /* allow Mulhs */
-      1, /* allow Mulus */
-     32  /* Mulh allowed up to 32 bit */
-  };
-
-  return ad_param ? ad_param : &param;
-}
+struct a_firm_be_opt {
+       bool selection;
+       bool node_stat;
+};
 
-/**
- * Map runtime functions.
- */
-static void rts_map(void) {
-  static const struct {
-    ir_entity_ptr *ent; /**< address of the rts entity */
-    i_mapper_func func; /**< mapper function. */
-  } mapper[] = {
-    /* integer */
-    { &rts_entities[rts_abs],     i_mapper_abs },
-    { &rts_entities[rts_labs],    i_mapper_abs },
-    { &rts_entities[rts_llabs],   i_mapper_abs },
-    { &rts_entities[rts_imaxabs], i_mapper_abs },
-
-    /* double -> double */
-    { &rts_entities[rts_fabs],    i_mapper_abs },
-    { &rts_entities[rts_sqrt],    i_mapper_sqrt },
-    { &rts_entities[rts_cbrt],    i_mapper_cbrt },
-    { &rts_entities[rts_pow],     i_mapper_pow },
-    { &rts_entities[rts_exp],     i_mapper_exp },
-    { &rts_entities[rts_exp2],    i_mapper_exp },
-    { &rts_entities[rts_exp10],   i_mapper_exp },
-    { &rts_entities[rts_log],     i_mapper_log },
-    { &rts_entities[rts_log2],    i_mapper_log2 },
-    { &rts_entities[rts_log10],   i_mapper_log10 },
-    { &rts_entities[rts_sin],     i_mapper_sin },
-    { &rts_entities[rts_cos],     i_mapper_cos },
-    { &rts_entities[rts_tan],     i_mapper_tan },
-    { &rts_entities[rts_asin],    i_mapper_asin },
-    { &rts_entities[rts_acos],    i_mapper_acos },
-    { &rts_entities[rts_atan],    i_mapper_atan },
-    { &rts_entities[rts_sinh],    i_mapper_sinh },
-    { &rts_entities[rts_cosh],    i_mapper_cosh },
-    { &rts_entities[rts_tanh],    i_mapper_tanh },
-
-    /* float -> float */
-    { &rts_entities[rts_fabsf],   i_mapper_abs },
-    { &rts_entities[rts_sqrtf],   i_mapper_sqrt },
-    { &rts_entities[rts_cbrtf],   i_mapper_cbrt },
-    { &rts_entities[rts_powf],    i_mapper_pow },
-    { &rts_entities[rts_expf],    i_mapper_exp },
-    { &rts_entities[rts_exp2f],   i_mapper_exp },
-    { &rts_entities[rts_exp10f],  i_mapper_exp },
-    { &rts_entities[rts_logf],    i_mapper_log },
-    { &rts_entities[rts_log2f],   i_mapper_log2 },
-    { &rts_entities[rts_log10f],  i_mapper_log10 },
-    { &rts_entities[rts_sinf],    i_mapper_sin },
-    { &rts_entities[rts_cosf],    i_mapper_cos },
-    { &rts_entities[rts_tanf],    i_mapper_tan },
-    { &rts_entities[rts_asinf],   i_mapper_asin },
-    { &rts_entities[rts_acosf],   i_mapper_acos },
-    { &rts_entities[rts_atanf],   i_mapper_atan },
-    { &rts_entities[rts_sinhf],   i_mapper_sinh },
-    { &rts_entities[rts_coshf],   i_mapper_cosh },
-    { &rts_entities[rts_tanhf],   i_mapper_tanh },
-
-    /* long double -> long double */
-    { &rts_entities[rts_fabsl],   i_mapper_abs },
-    { &rts_entities[rts_sqrtl],   i_mapper_sqrt },
-    { &rts_entities[rts_cbrtl],   i_mapper_cbrt },
-    { &rts_entities[rts_powl],    i_mapper_pow },
-    { &rts_entities[rts_expl],    i_mapper_exp },
-    { &rts_entities[rts_exp2l],   i_mapper_exp },
-    { &rts_entities[rts_exp10l],  i_mapper_exp },
-    { &rts_entities[rts_logl],    i_mapper_log },
-    { &rts_entities[rts_log2l],   i_mapper_log2 },
-    { &rts_entities[rts_log10l],  i_mapper_log10 },
-    { &rts_entities[rts_sinl],    i_mapper_sin },
-    { &rts_entities[rts_cosl],    i_mapper_cos },
-    { &rts_entities[rts_tanl],    i_mapper_tan },
-    { &rts_entities[rts_asinl],   i_mapper_asin },
-    { &rts_entities[rts_acosl],   i_mapper_acos },
-    { &rts_entities[rts_atanl],   i_mapper_atan },
-    { &rts_entities[rts_sinhl],   i_mapper_sinh },
-    { &rts_entities[rts_coshl],   i_mapper_cosh },
-    { &rts_entities[rts_tanhl],   i_mapper_tanh },
-
-    /* string */
-    { &rts_entities[rts_strcmp],  i_mapper_strcmp },
-    { &rts_entities[rts_strncmp], i_mapper_strncmp },
-    { &rts_entities[rts_strcpy],  i_mapper_strcpy },
-    { &rts_entities[rts_strlen],  i_mapper_strlen },
-    { &rts_entities[rts_memcpy],  i_mapper_memcpy },
-    { &rts_entities[rts_mempcpy], i_mapper_mempcpy },
-    { &rts_entities[rts_memmove], i_mapper_memmove },
-    { &rts_entities[rts_memset],  i_mapper_memset },
-    { &rts_entities[rts_memcmp],  i_mapper_memcmp }
-  };
-  i_record rec[sizeof(mapper)/sizeof(mapper[0])];
-  unsigned i, n_map;
-
-  for (i = n_map = 0; i < sizeof(mapper)/sizeof(mapper[0]); ++i)
-    if (*mapper[i].ent != NULL) {
-      rec[n_map].i_call.kind     = INTRINSIC_CALL;
-      rec[n_map].i_call.i_ent    = *mapper[i].ent;
-      rec[n_map].i_call.i_mapper = mapper[i].func;
-      rec[n_map].i_call.ctx      = NULL;
-      rec[n_map].i_call.link     = NULL;
-      ++n_map;
-  }  /* if */
-  if (n_map > 0)
-    lower_intrinsics(rec, n_map, /* part_block_used=*/0);
-}  /* rts_map */
+/* optimization settings */
+static struct a_firm_opt firm_opt = {
+       .const_folding    =  true,
+       .cse              =  true,
+       .confirm          =  true,
+       .muls             =  true,
+       .divs             =  true,
+       .mods             =  true,
+       .alias_analysis   =  true,
+       .strict_alias     =  false,
+       .no_alias         =  false,
+       .verify           =  FIRM_VERIFICATION_ON,
+       .check_all        =  true,
+       .clone_threshold  =  DEFAULT_CLONE_THRESHOLD,
+       .inline_maxsize   =  750,
+       .inline_threshold =  0,
+};
 
-static int *irg_dump_no;
+/* dumping options */
+static struct a_firm_dump firm_dump = {
+       .debug_print  = false,
+       .all_types    = false,
+       .ir_graph     = false,
+       .all_phases   = false,
+       .statistic    = STAT_NONE,
+       .stat_pattern = 0,
+       .stat_dag     = 0,
+};
 
-static void dump_graph_count(ir_graph *const irg, const char *const suffix)
-{
-  char name[64];
-  snprintf(name, sizeof(name), "-%02d_%s", irg_dump_no[get_irg_idx(irg)]++, suffix);
-  DUMP_ONE(1, irg, name);
-}
+#define X(a)  a, sizeof(a)-1
+
+/** Parameter description structure */
+static const struct params {
+  const char *option;      /**< name of the option */
+  size_t     opt_len;      /**< length of the option string */
+  bool       *flag;        /**< address of variable to set/reset */
+  bool       set;          /**< iff true, variable will be set, else reset */
+  const char *description; /**< description of this option */
+} firm_options[] = {
+  /* firm optimization options */
+  { X("no-opt"),                 NULL,                       0, "disable all FIRM optimizations" },
+  { X("cse"),                    &firm_opt.cse,              1, "enable common subexpression elimination" },
+  { X("no-cse"),                 &firm_opt.cse,              0, "disable common subexpression elimination" },
+  { X("const-fold"),             &firm_opt.const_folding,    1, "enable constant folding" },
+  { X("no-const-fold"),          &firm_opt.const_folding,    0, "disable constant folding" },
+  { X("inline-max-size=<size>"), NULL,                       0, "set maximum size for function inlining" },
+  { X("inline-threshold=<size>"),NULL,                       0, "set benefice threshold for function inlining" },
+  { X("confirm"),                &firm_opt.confirm,          1, "enable Confirm optimization" },
+  { X("no-confirm"),             &firm_opt.confirm,          0, "disable Confirm optimization" },
+  { X("opt-mul"),                &firm_opt.muls,             0, "enable multiplication optimization" },
+  { X("no-opt-mul"),             &firm_opt.muls,             0, "disable multiplication optimization" },
+  { X("opt-div"),                &firm_opt.divs,             0, "enable division optimization" },
+  { X("no-opt-div"),             &firm_opt.divs,             0, "disable division optimization" },
+  { X("opt-mod"),                &firm_opt.mods,             0, "enable remainder optimization" },
+  { X("no-opt-mod"),             &firm_opt.mods,             0, "disable remainder optimization" },
+  { X("opt-alias"),              &firm_opt.alias_analysis,   1, "enable alias analysis" },
+  { X("no-opt-alias"),           &firm_opt.alias_analysis,   0, "disable alias analysis" },
+  { X("alias"),                  &firm_opt.no_alias,         0, "aliasing occurs" },
+  { X("no-alias"),               &firm_opt.no_alias,         1, "no aliasing occurs" },
+  { X("strict-aliasing"),        &firm_opt.strict_alias,     1, "strict alias rules" },
+  { X("no-strict-aliasing"),     &firm_opt.strict_alias,     0, "strict alias rules" },
+  { X("clone-threshold=<value>"),NULL,                       0, "set clone threshold to <value>" },
+
+  /* other firm regarding options */
+  { X("verify-off"),             &firm_opt.verify,           FIRM_VERIFICATION_OFF,    "disable node verification" },
+  { X("verify-on"),              &firm_opt.verify,           FIRM_VERIFICATION_ON,     "enable node verification" },
+  { X("verify-report"),          &firm_opt.verify,           FIRM_VERIFICATION_REPORT, "node verification, report only" },
+  { X("check-all"),              &firm_opt.check_all,        1, "enable checking all Firm phases" },
+  { X("no-check-all"),           &firm_opt.check_all,        0, "disable checking all Firm phases" },
+
+  /* dumping */
+  { X("dump-ir"),                &firm_dump.ir_graph,        1, "dump IR graph" },
+  { X("dump-all-types"),         &firm_dump.all_types,       1, "dump graph of all types" },
+  { X("dump-all-phases"),        &firm_dump.all_phases,      1, "dump graphs for all optimization phases" },
+  { X("dump-filter=<string>"),   NULL,                       0, "set dumper filter" },
+
+  /* misc */
+  { X("stat-before-opt"),        &firm_dump.statistic,       STAT_BEFORE_OPT,  "Firm statistic output before optimizations" },
+  { X("stat-after-opt"),         &firm_dump.statistic,       STAT_AFTER_OPT,   "Firm statistic output after optimizations" },
+  { X("stat-after-lower"),       &firm_dump.statistic,       STAT_AFTER_LOWER, "Firm statistic output after lowering" },
+  { X("stat-final-ir"),          &firm_dump.statistic,       STAT_FINAL_IR,    "Firm statistic after final optimization" },
+  { X("stat-final"),             &firm_dump.statistic,       STAT_FINAL,       "Firm statistic after code generation" },
+  { X("stat-pattern"),           &firm_dump.stat_pattern,    1, "Firm statistic calculates most used pattern" },
+  { X("stat-dag"),               &firm_dump.stat_dag,        1, "Firm calculates DAG statistics" },
+};
 
-#if 0
-static void dump_graph_cfg_count(ir_graph *const irg, const char *const suffix)
-{
-  char name[64];
-  snprintf(name, sizeof(name), "-%02d_%s", irg_dump_no[get_irg_idx(irg)]++, suffix);
-  DUMP_ONE_CFG(1, irg, name);
-}
-#endif
+#undef X
 
-static void dump_all_count(const char *const suffix)
-{
-  const int n_irgs = get_irp_n_irgs();
-  int i;
+static ir_timer_t *t_vcg_dump;
+static ir_timer_t *t_verify;
+static ir_timer_t *t_all_opt;
+static ir_timer_t *t_backend;
+static bool do_irg_opt(ir_graph *irg, const char *name);
 
-  for (i = 0; i < n_irgs; ++i)
-    dump_graph_count(get_irp_irg(i), suffix);
-}
+/** dump all the graphs depending on cond */
 
-#define DUMP_ONE_C(cond, irg, suffix)    \
-  do {                                   \
-    if (cond) {                          \
-      dump_graph_count((irg), (suffix)); \
-    }                                    \
-  } while (0)
-
-#define DUMP_ONE_CFG_C(cond, irg, suffix)    \
-  do {                                       \
-    if (cond) {                              \
-      dump_graph_cfg_count((irg), (suffix)); \
-    }                                        \
-  } while (0)
-
-#define DUMP_ALL_C(cond, suffix) \
-  do {                           \
-    if (cond) {                  \
-      dump_all_count((suffix));  \
-    }                            \
-  } while (0)
-
-static void remove_unused_functions(void)
+static void dump_all(const char *suffix)
 {
-       ir_entity **keep_methods;
-       int         arr_len;
-
-       /* Analysis that finds the free methods,
-          i.e. methods that are dereferenced.
-          Optimizes polymorphic calls :-). */
-       cgana(&arr_len, &keep_methods);
+       if (!firm_dump.ir_graph)
+               return;
 
-       /* Remove methods that are never called. */
-       gc_irgs(arr_len, keep_methods);
-       free(keep_methods);
+       timer_push(t_vcg_dump);
+       dump_all_ir_graphs(suffix);
+       timer_pop(t_vcg_dump);
 }
 
-static int firm_const_exists;
+/* entities of runtime functions */
+ir_entity *rts_entities[rts_max];
 
-static void do_optimize_funccalls(void)
+/**
+ * Map runtime functions.
+ */
+static void rts_map(void)
 {
-       optimize_funccalls(firm_const_exists, NULL);
-}
+       static const struct {
+               ir_entity   **ent; /**< address of the rts entity */
+               i_mapper_func func; /**< mapper function. */
+       } mapper[] = {
+               /* integer */
+               { &rts_entities[rts_abs],     i_mapper_abs },
+               { &rts_entities[rts_labs],    i_mapper_abs },
+               { &rts_entities[rts_llabs],   i_mapper_abs },
+               { &rts_entities[rts_imaxabs], i_mapper_abs },
+
+               /* double -> double */
+               { &rts_entities[rts_fabs],    i_mapper_abs },
+               { &rts_entities[rts_sqrt],    i_mapper_sqrt },
+               { &rts_entities[rts_cbrt],    i_mapper_cbrt },
+               { &rts_entities[rts_pow],     i_mapper_pow },
+               { &rts_entities[rts_exp],     i_mapper_exp },
+               { &rts_entities[rts_exp2],    i_mapper_exp },
+               { &rts_entities[rts_exp10],   i_mapper_exp },
+               { &rts_entities[rts_log],     i_mapper_log },
+               { &rts_entities[rts_log2],    i_mapper_log2 },
+               { &rts_entities[rts_log10],   i_mapper_log10 },
+               { &rts_entities[rts_sin],     i_mapper_sin },
+               { &rts_entities[rts_cos],     i_mapper_cos },
+               { &rts_entities[rts_tan],     i_mapper_tan },
+               { &rts_entities[rts_asin],    i_mapper_asin },
+               { &rts_entities[rts_acos],    i_mapper_acos },
+               { &rts_entities[rts_atan],    i_mapper_atan },
+               { &rts_entities[rts_sinh],    i_mapper_sinh },
+               { &rts_entities[rts_cosh],    i_mapper_cosh },
+               { &rts_entities[rts_tanh],    i_mapper_tanh },
+
+               /* float -> float */
+               { &rts_entities[rts_fabsf],   i_mapper_abs },
+               { &rts_entities[rts_sqrtf],   i_mapper_sqrt },
+               { &rts_entities[rts_cbrtf],   i_mapper_cbrt },
+               { &rts_entities[rts_powf],    i_mapper_pow },
+               { &rts_entities[rts_expf],    i_mapper_exp },
+               { &rts_entities[rts_exp2f],   i_mapper_exp },
+               { &rts_entities[rts_exp10f],  i_mapper_exp },
+               { &rts_entities[rts_logf],    i_mapper_log },
+               { &rts_entities[rts_log2f],   i_mapper_log2 },
+               { &rts_entities[rts_log10f],  i_mapper_log10 },
+               { &rts_entities[rts_sinf],    i_mapper_sin },
+               { &rts_entities[rts_cosf],    i_mapper_cos },
+               { &rts_entities[rts_tanf],    i_mapper_tan },
+               { &rts_entities[rts_asinf],   i_mapper_asin },
+               { &rts_entities[rts_acosf],   i_mapper_acos },
+               { &rts_entities[rts_atanf],   i_mapper_atan },
+               { &rts_entities[rts_sinhf],   i_mapper_sinh },
+               { &rts_entities[rts_coshf],   i_mapper_cosh },
+               { &rts_entities[rts_tanhf],   i_mapper_tanh },
+
+               /* long double -> long double */
+               { &rts_entities[rts_fabsl],   i_mapper_abs },
+               { &rts_entities[rts_sqrtl],   i_mapper_sqrt },
+               { &rts_entities[rts_cbrtl],   i_mapper_cbrt },
+               { &rts_entities[rts_powl],    i_mapper_pow },
+               { &rts_entities[rts_expl],    i_mapper_exp },
+               { &rts_entities[rts_exp2l],   i_mapper_exp },
+               { &rts_entities[rts_exp10l],  i_mapper_exp },
+               { &rts_entities[rts_logl],    i_mapper_log },
+               { &rts_entities[rts_log2l],   i_mapper_log2 },
+               { &rts_entities[rts_log10l],  i_mapper_log10 },
+               { &rts_entities[rts_sinl],    i_mapper_sin },
+               { &rts_entities[rts_cosl],    i_mapper_cos },
+               { &rts_entities[rts_tanl],    i_mapper_tan },
+               { &rts_entities[rts_asinl],   i_mapper_asin },
+               { &rts_entities[rts_acosl],   i_mapper_acos },
+               { &rts_entities[rts_atanl],   i_mapper_atan },
+               { &rts_entities[rts_sinhl],   i_mapper_sinh },
+               { &rts_entities[rts_coshl],   i_mapper_cosh },
+               { &rts_entities[rts_tanhl],   i_mapper_tanh },
+
+               /* string */
+               { &rts_entities[rts_strcmp],  i_mapper_strcmp },
+               { &rts_entities[rts_strncmp], i_mapper_strncmp },
+               { &rts_entities[rts_strcpy],  i_mapper_strcpy },
+               { &rts_entities[rts_strlen],  i_mapper_strlen },
+               { &rts_entities[rts_memcpy],  i_mapper_memcpy },
+               { &rts_entities[rts_mempcpy], i_mapper_mempcpy },
+               { &rts_entities[rts_memmove], i_mapper_memmove },
+               { &rts_entities[rts_memset],  i_mapper_memset },
+               { &rts_entities[rts_memcmp],  i_mapper_memcmp }
+       };
+       i_record rec[lengthof(mapper)];
+       size_t   n_map = 0;
+
+       for (size_t i = 0; i != lengthof(mapper); ++i) {
+               if (*mapper[i].ent != NULL) {
+                       rec[n_map].i_call.kind     = INTRINSIC_CALL;
+                       rec[n_map].i_call.i_ent    = *mapper[i].ent;
+                       rec[n_map].i_call.i_mapper = mapper[i].func;
+                       rec[n_map].i_call.ctx      = NULL;
+                       rec[n_map].i_call.link     = NULL;
+                       ++n_map;
+               }
+       }
 
-static void do_gcse(ir_graph *irg)
-{
-       set_opt_global_cse(1);
-       optimize_graph_df(irg);
-       place_code(irg);
-       set_opt_global_cse(0);
+       if (n_map > 0)
+               lower_intrinsics(rec, n_map, /* part_block_used=*/0);
 }
 
-static void do_lower_highlevel(ir_graph *irg)
-{
-       lower_highlevel_graph(irg, firm_opt.lower_bitfields);
-}
+static int *irg_dump_no;
 
-static void do_if_conv(ir_graph *irg)
+static void do_stred(ir_graph *irg)
 {
-       opt_if_conv(irg, if_conv_info);
+       opt_osr(irg, osr_flag_default | osr_flag_keep_reg_pressure | osr_flag_ignore_x86_shift);
 }
 
-static void do_stred(ir_graph *irg)
+static void after_inline_opt(ir_graph *irg)
 {
-       opt_osr(irg, osr_flag_default | osr_flag_keep_reg_pressure | osr_flag_ignore_x86_shift);
+       do_irg_opt(irg, "scalar-replace");
+       do_irg_opt(irg, "local");
+       do_irg_opt(irg, "control-flow");
+       do_irg_opt(irg, "combo");
 }
 
 static void do_inline(void)
 {
-       inline_functions(firm_opt.inline_maxsize, firm_opt.inline_threshold);
+       inline_functions(firm_opt.inline_maxsize, firm_opt.inline_threshold,
+                        after_inline_opt);
 }
 
 static void do_cloning(void)
@@ -341,128 +305,210 @@ static void do_cloning(void)
        proc_cloning((float) firm_opt.clone_threshold);
 }
 
-static void do_lower_switch(ir_graph *irg)
+static void do_lower_mux(ir_graph *irg)
 {
-       lower_switch(irg, firm_opt.spare_size);
+       lower_mux(irg, NULL);
+}
+
+static void do_gcse(ir_graph *irg)
+{
+       set_opt_global_cse(1);
+       optimize_graph_df(irg);
+       set_opt_global_cse(0);
 }
 
 typedef enum opt_target {
-       OPT_TARGET_IRG,
-       OPT_TARGET_IRP
-} opt_target;
+       OPT_TARGET_IRG, /**< optimization function works on a single graph */
+       OPT_TARGET_IRP  /**< optimization function works on the complete program */
+} opt_target_t;
+
+typedef enum opt_flags {
+       OPT_FLAG_NONE         = 0,
+       OPT_FLAG_ENABLED      = 1 << 0, /**< enable the optimization */
+       OPT_FLAG_NO_DUMP      = 1 << 1, /**< don't dump after transformation */
+       OPT_FLAG_NO_VERIFY    = 1 << 2, /**< don't verify after transformation */
+       OPT_FLAG_HIDE_OPTIONS = 1 << 3, /**< do not automatically process
+                                            -foptions for this transformation */
+       OPT_FLAG_ESSENTIAL    = 1 << 4, /**< output won't work without this pass
+                                            so we need it even with -O0 */
+} opt_flags_t;
 
 typedef void (*transform_irg_func)(ir_graph *irg);
 typedef void (*transform_irp_func)(void);
-typedef void (*func_ptr_t)(void);
 
 typedef struct {
-       opt_target  target;
-       const char *name;
-       func_ptr_t  func;
-       bool        enabled;
-       bool        dump;
-       bool        verify;
-       int         timer;
+       opt_target_t  target;
+       const char   *name;
+       union {
+               transform_irg_func transform_irg;
+               transform_irp_func transform_irp;
+       } u;
+       const char   *description;
+       opt_flags_t   flags;
+       ir_timer_t   *timer;
 } opt_config_t;
 
 static opt_config_t opts[] = {
-       { OPT_TARGET_IRP, "rts",         (func_ptr_t) rts_map,                 true, true,  true,  -1 },
-       { OPT_TARGET_IRG, "combo",       (func_ptr_t) combo,                   true, true,  true,  TV_COMBO },
-       { OPT_TARGET_IRG, "controlflow", (func_ptr_t) optimize_cf,             true, true,  true,  TV_CF_OPT },
-       { OPT_TARGET_IRG, "local",       (func_ptr_t) optimize_graph_df,       true, true,  true,  TV_LOCAL_OPT },
-       { OPT_TARGET_IRP, "gc_irgs",     (func_ptr_t) remove_unused_functions, true, false, false, TV_CGANA },
-       { OPT_TARGET_IRP, "tailrec",     (func_ptr_t) opt_tail_recursion,      true, true,  true,  TV_TAIL_REC },
-       { OPT_TARGET_IRP, "funccalls",   (func_ptr_t) do_optimize_funccalls,   true, true,  true,  TV_REAL_FUNC_CALL },
-       { OPT_TARGET_IRP, "lowerconst",  (func_ptr_t) lower_const_code,        true, false, false, -1 },
-       { OPT_TARGET_IRG, "onereturn",   (func_ptr_t) normalize_one_return,    true, false, false, -1 },
-       { OPT_TARGET_IRG, "scalar",      (func_ptr_t) scalar_replacement_opt,  true, true,  true,  TV_SCALAR_REPLACE },
-       { OPT_TARGET_IRG, "reassoc",     (func_ptr_t) optimize_reassociation,  true, true,  true,  TV_REASSOCIATION },
-       { OPT_TARGET_IRG, "gcse",        (func_ptr_t) do_gcse,                 true, true,  true,  TV_CODE_PLACE },
-       { OPT_TARGET_IRG, "place",       (func_ptr_t) place_code,              true, true,  true,  TV_CODE_PLACE },
-       { OPT_TARGET_IRG, "confirm",     (func_ptr_t) construct_confirms,      true, true,  true,  TV_CONFIRM_CREATE },
-       { OPT_TARGET_IRG, "ldst",        (func_ptr_t) optimize_load_store,     true, true,  true,  TV_LOAD_STORE },
-       { OPT_TARGET_IRG, "sync",        (func_ptr_t) opt_sync,                true, true,  true,  -1 },
-       { OPT_TARGET_IRG, "lower",       (func_ptr_t) do_lower_highlevel,      true, true,  true,  -1 },
-       { OPT_TARGET_IRG, "deconv",      (func_ptr_t) conv_opt,                true, true,  true,  TV_DECONV },
-       { OPT_TARGET_IRG, "jumpthreading", (func_ptr_t) opt_jumpthreading,     true, true,  true,  TV_JUMPTHREADING },
-       { OPT_TARGET_IRG, "remove_confirms", (func_ptr_t) remove_confirms,     true, false, false, TV_CONFIRM_CREATE },
-       { OPT_TARGET_IRG, "gvnpre",      (func_ptr_t) do_gvn_pre,              true, true,  true,  -1 },
-       { OPT_TARGET_IRG, "ifconv",      (func_ptr_t) do_if_conv,              true, true,  true,  TV_IF_CONV },
-       { OPT_TARGET_IRG, "bool",        (func_ptr_t) opt_bool,                true, true,  true,  -1 },
-       { OPT_TARGET_IRG, "shape-blocks",(func_ptr_t) shape_blocks,            true, true,  true,  TV_END_MELT },
-       { OPT_TARGET_IRG, "stred",       (func_ptr_t) do_stred,                true, true,  true,  TV_OSR },
-       { OPT_TARGET_IRG, "dead",        (func_ptr_t) dead_node_elimination,   true, false, true,  TV_DEAD_NODE },
-       { OPT_TARGET_IRP, "inline",      (func_ptr_t) do_inline,               true, true,  true,  -1 },
-       { OPT_TARGET_IRP, "clone",       (func_ptr_t) do_cloning,              true, true,  true,  -1 },
-       { OPT_TARGET_IRG, "lower_switch", (func_ptr_t) do_lower_switch,        true, true,  true,  -1 },
+#define IRG(a, b, c, d) { OPT_TARGET_IRG, a, .u.transform_irg = (transform_irg_func)b, c, d }
+#define IRP(a, b, c, d) { OPT_TARGET_IRP, a, .u.transform_irp = b,                     c, d }
+       IRG("bool",              opt_bool,                 "bool simplification",                                   OPT_FLAG_NONE),
+       IRG("combo",             combo,                    "combined CCE, UCE and GVN",                             OPT_FLAG_NONE),
+       IRG("confirm",           construct_confirms,       "confirm optimization",                                  OPT_FLAG_HIDE_OPTIONS),
+       IRG("control-flow",      optimize_cf,              "optimization of control-flow",                          OPT_FLAG_HIDE_OPTIONS),
+       IRG("dead",              dead_node_elimination,    "dead node elimination",                                 OPT_FLAG_HIDE_OPTIONS | OPT_FLAG_NO_DUMP | OPT_FLAG_NO_VERIFY),
+       IRG("deconv",            conv_opt,                 "conv node elimination",                                 OPT_FLAG_NONE),
+       IRG("fp-vrp",            fixpoint_vrp,             "fixpoint value range propagation",                      OPT_FLAG_NONE),
+       IRG("frame",             opt_frame_irg,            "remove unused frame entities",                          OPT_FLAG_NONE),
+       IRG("gvn-pre",           do_gvn_pre,               "global value numbering partial redundancy elimination", OPT_FLAG_NONE),
+       IRG("if-conversion",     opt_if_conv,              "if-conversion",                                         OPT_FLAG_NONE),
+       IRG("invert-loops",      do_loop_inversion,        "loop inversion",                                        OPT_FLAG_NONE),
+       IRG("ivopts",            do_stred,                 "induction variable strength reduction",                 OPT_FLAG_NONE),
+       IRG("local",             local_opts,               "local graph optimizations",                             OPT_FLAG_HIDE_OPTIONS),
+       IRG("lower",             lower_highlevel_graph,    "lowering",                                              OPT_FLAG_HIDE_OPTIONS | OPT_FLAG_ESSENTIAL),
+       IRG("lower-mux",         do_lower_mux,             "mux lowering",                                          OPT_FLAG_NONE),
+       IRG("opt-load-store",    optimize_load_store,      "load store optimization",                               OPT_FLAG_NONE),
+       IRG("opt-tail-rec",      opt_tail_rec_irg,         "tail-recursion eliminiation",                           OPT_FLAG_NONE),
+       IRG("parallelize-mem",   opt_parallelize_mem,      "parallelize memory",                                    OPT_FLAG_NONE),
+       IRG("gcse",              do_gcse,                  "global common subexpression eliminiation",              OPT_FLAG_NONE),
+       IRG("place",             place_code,               "code placement",                                        OPT_FLAG_NONE),
+       IRG("reassociation",     optimize_reassociation,   "reassociation",                                         OPT_FLAG_NONE),
+       IRG("remove-confirms",   remove_confirms,          "confirm removal",                                       OPT_FLAG_HIDE_OPTIONS | OPT_FLAG_NO_DUMP | OPT_FLAG_NO_VERIFY),
+       IRG("remove-phi-cycles", remove_phi_cycles,        "removal of phi cycles",                                 OPT_FLAG_HIDE_OPTIONS),
+       IRG("scalar-replace",    scalar_replacement_opt,   "scalar replacement",                                    OPT_FLAG_NONE),
+       IRG("shape-blocks",      shape_blocks,             "block shaping",                                         OPT_FLAG_NONE),
+       IRG("thread-jumps",      opt_jumpthreading,        "path-sensitive jumpthreading",                          OPT_FLAG_NONE),
+       IRG("unroll-loops",      do_loop_unrolling,        "loop unrolling",                                        OPT_FLAG_NONE),
+       IRG("vrp",               set_vrp_data,             "value range propagation",                               OPT_FLAG_NONE),
+       IRP("inline",            do_inline,                "inlining",                                              OPT_FLAG_NONE),
+       IRP("lower-const",       lower_const_code,         "lowering of constant code",                             OPT_FLAG_HIDE_OPTIONS | OPT_FLAG_NO_DUMP | OPT_FLAG_NO_VERIFY | OPT_FLAG_ESSENTIAL),
+       IRP("target-lowering",   be_lower_for_target,      "lowering necessary for target architecture",            OPT_FLAG_HIDE_OPTIONS | OPT_FLAG_ESSENTIAL),
+       IRP("opt-func-call",     optimize_funccalls,       "function call optimization",                            OPT_FLAG_NONE),
+       IRP("opt-proc-clone",    do_cloning,               "procedure cloning",                                     OPT_FLAG_NONE),
+       IRP("remove-unused",     garbage_collect_entities, "removal of unused functions/variables",                 OPT_FLAG_NO_DUMP | OPT_FLAG_NO_VERIFY),
+       IRP("rts",               rts_map,                  "optimization of known library functions",               OPT_FLAG_NONE),
+       IRP("opt-cc",            mark_private_methods,     "calling conventions optimization",                      OPT_FLAG_NONE),
+#undef IRP
+#undef IRG
 };
-static const int n_opts = sizeof(opts) / sizeof(opts[0]);
+
+#define FOR_EACH_OPT(i) for (opt_config_t *i = opts; i != endof(opts); ++i)
 
 static opt_config_t *get_opt(const char *name)
 {
-       int i;
-       for (i = 0; i < n_opts; ++i) {
-               opt_config_t *config = &opts[i];
-               if (strcmp(config->name, name) == 0)
+       FOR_EACH_OPT(config) {
+               if (streq(config->name, name))
                        return config;
        }
 
-       assert(0);
        return NULL;
 }
 
 static void set_opt_enabled(const char *name, bool enabled)
 {
        opt_config_t *config = get_opt(name);
-       config->enabled = enabled;
+       config->flags = (config->flags & ~OPT_FLAG_ENABLED)
+               | (enabled ? OPT_FLAG_ENABLED : 0);
 }
 
-static void do_irg_opt(ir_graph *irg, const char *name)
+static bool get_opt_enabled(const char *name)
 {
-       transform_irg_func  func;
-       ir_graph           *old_irg;
-       opt_config_t       *config = get_opt(name);
-       assert(config->target == OPT_TARGET_IRG);
-       if (!config->enabled)
-               return;
+       opt_config_t *config = get_opt(name);
+       return (config->flags & OPT_FLAG_ENABLED) != 0;
+}
 
-       if (config->timer != -1)
-               timer_push(config->timer);
+/**
+ * perform an optimization on a single graph
+ *
+ * @return  true if something changed, false otherwise
+ */
+static bool do_irg_opt(ir_graph *irg, const char *name)
+{
+       opt_config_t *const config = get_opt(name);
+       assert(config != NULL);
+       assert(config->target == OPT_TARGET_IRG);
+       if (! (config->flags & OPT_FLAG_ENABLED))
+               return false;
 
-       old_irg          = current_ir_graph;
+       ir_graph *const old_irg = current_ir_graph;
        current_ir_graph = irg;
 
-       func = (transform_irg_func) config->func;
-       func(irg);
+       timer_push(config->timer);
+       config->u.transform_irg(irg);
+       timer_pop(config->timer);
 
-       if (config->timer != -1)
-               timer_pop();
+       if (firm_dump.all_phases && firm_dump.ir_graph) {
+               dump_ir_graph(irg, name);
+       }
 
-       DUMP_ONE_C(firm_dump.ir_graph && firm_dump.all_phases, irg, config->name);
-       CHECK_ONE(firm_opt.check_all, irg);
+       if (firm_opt.check_all) {
+               timer_push(t_verify);
+               irg_verify(irg, VERIFY_ENFORCE_SSA);
+               timer_pop(t_verify);
+       }
 
        current_ir_graph = old_irg;
+       return true;
 }
 
 static void do_irp_opt(const char *name)
 {
-       transform_irp_func  func;
-       opt_config_t       *config = get_opt(name);
+       opt_config_t *const config = get_opt(name);
        assert(config->target == OPT_TARGET_IRP);
-       if (!config->enabled)
+       if (! (config->flags & OPT_FLAG_ENABLED))
                return;
 
-       if (config->timer != -1)
-               timer_push(config->timer);
+       timer_push(config->timer);
+       config->u.transform_irp();
+       timer_pop(config->timer);
 
-       func = (transform_irp_func) config->func;
-       func();
+       if (firm_dump.ir_graph && firm_dump.all_phases) {
+               int i;
+               for (i = get_irp_n_irgs() - 1; i >= 0; --i) {
+                       ir_graph *irg = get_irp_irg(i);
+                       dump_ir_graph(irg, name);
+               }
+       }
 
-       DUMP_ALL_C(firm_dump.ir_graph && firm_dump.all_phases, config->name);
-       CHECK_ALL(firm_opt.check_all);
+       if (firm_opt.check_all) {
+               int i;
+               timer_push(t_verify);
+               for (i = get_irp_n_irgs() - 1; i >= 0; --i) {
+                       irg_verify(get_irp_irg(i), VERIFY_ENFORCE_SSA);
+               }
+               timer_pop(t_verify);
+       }
+}
 
-       if (config->timer != -1)
-               timer_pop();
+/**
+ * Enable transformations which should be always safe (and cheap) to perform
+ */
+static void enable_safe_defaults(void)
+{
+       set_opt_enabled("remove-unused", true);
+       set_opt_enabled("opt-tail-rec", true);
+       set_opt_enabled("opt-func-call", true);
+       set_opt_enabled("reassociation", true);
+       set_opt_enabled("control-flow", true);
+       set_opt_enabled("local", true);
+       set_opt_enabled("lower-const", true);
+       set_opt_enabled("scalar-replace", true);
+       set_opt_enabled("place", true);
+       set_opt_enabled("gcse", true);
+       set_opt_enabled("confirm", true);
+       set_opt_enabled("opt-load-store", true);
+       set_opt_enabled("lower", true);
+       set_opt_enabled("deconv", true);
+       set_opt_enabled("remove-confirms", true);
+       set_opt_enabled("ivopts", true);
+       set_opt_enabled("dead", true);
+       set_opt_enabled("remove-phi-cycles", true);
+       set_opt_enabled("frame", true);
+       set_opt_enabled("combo", true);
+       set_opt_enabled("invert-loops", true);
+       set_opt_enabled("target-lowering", true);
+       set_opt_enabled("rts", true);
+       set_opt_enabled("parallelize-mem", true);
+       set_opt_enabled("opt-cc", true);
 }
 
 /**
@@ -472,325 +518,128 @@ static void do_irp_opt(const char *name)
  */
 static void do_firm_optimizations(const char *input_filename)
 {
-  int      i;
-  unsigned aa_opt;
-
-  /* FIXME: cloning might ADD new graphs. */
-  irg_dump_no = calloc(get_irp_last_idx(), sizeof(*irg_dump_no));
-
-  set_opt_strength_red(firm_opt.strength_red);
-  set_opt_scalar_replacement(firm_opt.scalar_replace);
-  set_opt_auto_create_sync(firm_opt.auto_sync);
-  set_opt_alias_analysis(firm_opt.alias_analysis);
-  set_opt_combo(firm_opt.combo);
-
-  aa_opt = aa_opt_no_opt;
-  if (firm_opt.strict_alias)
-    aa_opt |= aa_opt_type_based | aa_opt_byte_type_may_alias;
-  if (firm_opt.no_alias)
-    aa_opt = aa_opt_no_alias;
-
-  set_irp_memory_disambiguator_options(aa_opt);
-
-  /* parameter passing code should set them directly sometime... */
-  set_opt_enabled("rts", !firm_opt.freestanding);
-  set_opt_enabled("gc_irgs", firm_opt.remove_unused);
-  set_opt_enabled("tailrec", firm_opt.tail_rec);
-  set_opt_enabled("funccalls", firm_opt.func_calls);
-  set_opt_enabled("gcse", firm_opt.gcse);
-  set_opt_enabled("place", !firm_opt.gcse);
-  set_opt_enabled("confirm", firm_opt.confirm);
-  set_opt_enabled("remove_confirms", firm_opt.confirm);
-  set_opt_enabled("sync", firm_opt.sync);
-  set_opt_enabled("ldst", firm_opt.load_store);
-  set_opt_enabled("deconv", firm_opt.deconv);
-  set_opt_enabled("jumpthreading", firm_opt.jumpthreading);
-  set_opt_enabled("gvnpre", firm_opt.gvn_pre);
-  set_opt_enabled("ifconv", firm_opt.if_conversion);
-  set_opt_enabled("bool", firm_opt.bool_opt);
-  set_opt_enabled("shape-blocks", firm_opt.shape_blocks);
-  set_opt_enabled("inline", firm_opt.do_inline);
-  set_opt_enabled("clone", firm_opt.cloning);
-  set_opt_enabled("combo", firm_opt.combo);
-
-  timer_start(TV_ALL_OPT);
-
-  do_irp_opt("rts");
-
-  /* first step: kill dead code */
-  if (firm_opt.combo) {
-    for (i = 0; i < get_irp_n_irgs(); i++) {
-      ir_graph *irg = get_irp_irg(i);
-      do_irg_opt(irg, "combo");
-    }
-  } else {
-    for (i = 0; i < get_irp_n_irgs(); i++) {
-      ir_graph *irg = get_irp_irg(i);
-      do_irg_opt(irg, "local");
-      do_irg_opt(irg, "controlflow");
-    }
-  }
-
-  do_irp_opt("gc_irgs");
-  do_irp_opt("tailrec");
-  do_irp_opt("funccalls");
-  do_irp_opt("lowerconst");
-
-  for (i = 0; i < get_irp_n_irgs(); i++) {
-    ir_graph *irg = get_irp_irg(i);
-
-    do_irg_opt(irg, "scalar");
-    do_irg_opt(irg, "local");
-    do_irg_opt(irg, "reassoc");
-    do_irg_opt(irg, "local");
-    do_irg_opt(irg, "gcse");
-
-    if (firm_opt.confirm) {
-      /* Confirm construction currently can only handle blocks with only one control
-         flow predecessor. Calling optimize_cf here removes Bad predecessors and help
-         the optimization of switch constructs. */
-      do_irg_opt(irg, "controlflow");
-      do_irg_opt(irg, "confirm");
-      do_irg_opt(irg, "local");
-    }
-
-    do_irg_opt(irg, "controlflow");
-    do_irg_opt(irg, "ldst");
-    do_irg_opt(irg, "lower");
-    do_irg_opt(irg, "deconv");
-    do_irg_opt(irg, "jumpthreading");
-    do_irg_opt(irg, "remove_confirms");
-    do_irg_opt(irg, "gvnpre");
-    do_irg_opt(irg, "place");
-    do_irg_opt(irg, "controlflow");
-
-    if (firm_opt.if_conversion) {
-      do_irg_opt(irg, "ifconv");
-      do_irg_opt(irg, "local");
-      do_irg_opt(irg, "controlflow");
-    }
-
-    do_irg_opt(irg, "bool");
-    do_irg_opt(irg, "shape-blocks");
-    do_irg_opt(irg, "lower_switch");
-    do_irg_opt(irg, "stred");
-    do_irg_opt(irg, "local");
-    do_irg_opt(irg, "dead");
-  }
-
-  do_irp_opt("inline");
-  do_irp_opt("clone");
-
-  for (i = 0; i < get_irp_n_irgs(); i++) {
-    ir_graph *irg = get_irp_irg(i);
-    do_irg_opt(irg, "local");
-    do_irg_opt(irg, "controlflow");
-    do_irg_opt(irg, "jumpthreading");
-    do_irg_opt(irg, "local");
-    do_irg_opt(irg, "controlflow");
-  }
-
-  if (firm_dump.ir_graph) {
-    /* recompute backedges for nicer dumps */
-    for (i = 0; i < get_irp_n_irgs(); i++)
-      construct_cf_backedges(get_irp_irg(i));
-  }
-
-  do_irp_opt("gc_irgs");
-
-  DUMP_ALL(firm_dump.ir_graph, "-opt");
-  /* verify optimized graphs */
-  for (i = get_irp_n_irgs() - 1; i >= 0; --i) {
-    ir_graph *irg = get_irp_irg(i);
-    irg_verify(irg, VRFY_ENFORCE_SSA);
-  }
-
-  if (firm_dump.statistic & STAT_AFTER_OPT)
-    stat_dump_snapshot(input_filename, "opt");
-
-  timer_stop(TV_ALL_OPT);
-}  /* do_firm_optimizations */
+       size_t   i;
+       unsigned aa_opt;
 
-/**
- * compute the size of a type (do implicit lowering)
- *
- * @param ty   a Firm type
- */
-static int compute_type_size(ir_type *ty)
-{
-  optimization_state_t state;
-  unsigned             align_all = 1;
-  int                  n, size = 0, set = 0;
-  int                  i, dims, s;
-
-  if (get_type_state(ty) == layout_fixed) {
-    /* do not layout already layouted types again */
-    return 1;
-  }
-
-  if (is_Method_type(ty) || ty == get_glob_type()) {
-    /* no need for size calculation for method types or the global type */
-    return 1;
-  }
-
-  DBG(("compute type size visiting: %s\n", get_type_name(ty)));
-
-  switch (get_type_tpop_code(ty)) {
-  case tpo_class:
-  case tpo_struct:
-    for (i = 0, n = get_compound_n_members(ty); i < n; ++i) {
-      ir_entity *ent  = get_compound_member(ty, i);
-      ir_type *ent_ty = get_entity_type(ent);
-      unsigned align, misalign;
-
-      /* inner functions do not expand the frame */
-      if (is_Method_type(ent_ty) && is_frame_type(ty))
-        continue;
-
-      /* compute member types */
-      if (! compute_type_size(ent_ty))
-        return 0;
-
-      align     = get_type_alignment_bytes(ent_ty);
-      align_all = align > align_all ? align : align_all;
-      misalign  = (align ? size % align : 0);
-      size     += (misalign ? align - misalign : 0);
-
-      set_entity_offset(ent, size);
-      size += get_type_size_bytes(ent_ty);
-
-      DBG(("  member %s %s -> (size: %u, align: %u)\n",
-            get_type_name(ent_ty), get_entity_name(ent),
-            get_type_size_bytes(ent_ty), get_type_alignment_bytes(ent_ty)));
-    }
-    if (align_all > 0 && size % align_all) {
-       DBG(("align of the struct member: %u, type size: %d\n", align_all, size));
-       size += align_all - (size % align_all);
-       DBG(("correcting type-size to %d\n", size));
-    }
-    set_type_alignment_bytes(ty, align_all);
-    set = 1;
-    break;
-
-  case tpo_union:
-    for (i = 0, n = get_union_n_members(ty); i < n; ++i) {
-      ir_entity *ent = get_union_member(ty, i);
-
-      if (! compute_type_size(get_entity_type(ent)))
-        return 0;
-      s = get_type_size_bytes(get_entity_type(ent));
-
-      set_entity_offset(ent, 0);
-      size = (s > size ? s : size);
-    }
-    set = 1;
-    break;
-
-  case tpo_array:
-    dims = get_array_n_dimensions(ty);
-
-    if (! compute_type_size(get_array_element_type(ty)))
-      return 0;
-
-    size = 1;
-
-    save_optimization_state(&state);
-    set_optimize(1);
-    set_opt_constant_folding(1);
-       set_opt_algebraic_simplification(1);
-
-    for (i = 0; i < dims; ++i) {
-      ir_node *lower   = get_array_lower_bound(ty, i);
-      ir_node *upper   = get_array_upper_bound(ty, i);
-      ir_graph *rem    = current_ir_graph;
-      tarval  *tv_lower, *tv_upper;
-      long     val_lower, val_upper;
-
-      current_ir_graph = get_const_code_irg();
-      local_optimize_node(lower);
-      local_optimize_node(upper);
-      current_ir_graph = rem;
-
-      tv_lower = computed_value(lower);
-      tv_upper = computed_value(upper);
-
-      if (tv_lower == tarval_bad || tv_upper == tarval_bad) {
-        /*
-         * we cannot calculate the size of this array yet, it
-         * even might be unknown until the end, like argv[]
-         */
-        restore_optimization_state(&state);
-        return 0;
-      }
-
-      val_upper = get_tarval_long(tv_upper);
-      val_lower = get_tarval_long(tv_lower);
-      size     *= val_upper - val_lower;
-    }
-    restore_optimization_state(&state);
-
-    DBG(("array %s -> (elements: %d, element type size: %d)\n",
-          get_type_name(ty),
-          size, get_type_size_bytes(get_array_element_type(ty))));
-    size *= get_type_size_bytes(get_array_element_type(ty));
-    set = 1;
-    break;
-
-  default:
-    break;
-  }
-
-  if (set) {
-    set_type_size_bytes(ty, size);
-    set_type_state(ty, layout_fixed);
-  }
-
-  DBG(("size: %d\n", get_type_size_bytes(ty)));
-
-  return set;
-}  /* compute_type_size */
+       set_opt_alias_analysis(firm_opt.alias_analysis);
 
-/**
- * layout all non-frame types of the Firm graph
- */
-static void compute_type_sizes(void)
-{
-  int i;
-  ir_type *tp;
+       aa_opt = aa_opt_no_opt;
+       if (firm_opt.strict_alias)
+               aa_opt |= aa_opt_type_based | aa_opt_byte_type_may_alias;
+       if (firm_opt.no_alias)
+               aa_opt = aa_opt_no_alias;
 
-  /* all non-frame other types */
-  for (i = get_irp_n_types() - 1; i >= 0; --i) {
-    tp = get_irp_type(i);
-    compute_type_size(tp);
+       set_irp_memory_disambiguator_options(aa_opt);
 
-    if (is_Method_type(tp)) {
-      tp = get_method_value_res_type(tp);
+       /* parameter passing code should set them directly sometime... */
+       set_opt_enabled("confirm", firm_opt.confirm);
+       set_opt_enabled("remove-confirms", firm_opt.confirm);
 
-      if (tp) {
-        /* we have a value result type for this method, lower */
-        compute_type_size(tp);
-      }
-    }
-  }
-}  /* compute_type_sizes */
+       /* osr supersedes remove_phi_cycles */
+       if (get_opt_enabled("ivopts"))
+               set_opt_enabled("remove-phi-cycles", false);
 
-/**
- * layout all frame-types of the Firm graph
- */
-static void compute_frame_type_sizes(void)
-{
-  int i;
-  ir_graph *irg;
-
-  /* all frame types */
-  for (i = get_irp_n_irgs() - 1; i >= 0; --i) {
-    irg = get_irp_irg(i);
-    /* do not optimize away variables in debug mode */
-    if (firm_opt.debug_mode == DBG_MODE_NONE)
-      opt_frame_irg(irg);
-    compute_type_size(get_irg_frame_type(irg));
-  }
-}  /* compute_frame_type_sizes */
+       timer_start(t_all_opt);
+
+       do_irp_opt("rts");
+
+       /* first step: kill dead code */
+       for (i = 0; i < get_irp_n_irgs(); i++) {
+               ir_graph *irg = get_irp_irg(i);
+               do_irg_opt(irg, "combo");
+               do_irg_opt(irg, "local");
+               do_irg_opt(irg, "control-flow");
+       }
+
+       do_irp_opt("remove-unused");
+       for (i = 0; i < get_irp_n_irgs(); ++i) {
+               ir_graph *irg = get_irp_irg(i);
+               do_irg_opt(irg, "opt-tail-rec");
+       }
+       do_irp_opt("opt-func-call");
+       do_irp_opt("lower-const");
+
+       for (i = 0; i < get_irp_n_irgs(); i++) {
+               ir_graph *irg = get_irp_irg(i);
+
+               do_irg_opt(irg, "scalar-replace");
+               do_irg_opt(irg, "invert-loops");
+               do_irg_opt(irg, "unroll-loops");
+               do_irg_opt(irg, "local");
+               do_irg_opt(irg, "reassociation");
+               do_irg_opt(irg, "local");
+               do_irg_opt(irg, "gcse");
+               do_irg_opt(irg, "place");
+
+               if (firm_opt.confirm) {
+                       /* Confirm construction currently can only handle blocks with only
+                          one control flow predecessor. Calling optimize_cf here removes
+                          Bad predecessors and help the optimization of switch constructs.
+                        */
+                       do_irg_opt(irg, "control-flow");
+                       do_irg_opt(irg, "confirm");
+                       do_irg_opt(irg, "vrp");
+                       do_irg_opt(irg, "local");
+               }
+
+               do_irg_opt(irg, "control-flow");
+               do_irg_opt(irg, "opt-load-store");
+               do_irg_opt(irg, "fp-vrp");
+               do_irg_opt(irg, "lower");
+               do_irg_opt(irg, "deconv");
+               do_irg_opt(irg, "thread-jumps");
+               do_irg_opt(irg, "remove-confirms");
+               do_irg_opt(irg, "gvn-pre");
+               do_irg_opt(irg, "gcse");
+               do_irg_opt(irg, "place");
+               do_irg_opt(irg, "control-flow");
+
+               if (do_irg_opt(irg, "if-conversion")) {
+                       do_irg_opt(irg, "local");
+                       do_irg_opt(irg, "control-flow");
+               }
+               /* this doesn't make too much sense but tests the mux destruction... */
+               do_irg_opt(irg, "lower-mux");
+
+               do_irg_opt(irg, "bool");
+               do_irg_opt(irg, "shape-blocks");
+               do_irg_opt(irg, "ivopts");
+               do_irg_opt(irg, "local");
+               do_irg_opt(irg, "dead");
+       }
+
+       do_irp_opt("inline");
+       do_irp_opt("opt-proc-clone");
+
+       for (i = 0; i < get_irp_n_irgs(); i++) {
+               ir_graph *irg = get_irp_irg(i);
+               do_irg_opt(irg, "local");
+               do_irg_opt(irg, "control-flow");
+               do_irg_opt(irg, "thread-jumps");
+               do_irg_opt(irg, "local");
+               do_irg_opt(irg, "control-flow");
+
+               if( do_irg_opt(irg, "vrp") ) { // if vrp is enabled
+                       do_irg_opt(irg, "local");
+                       do_irg_opt(irg, "vrp");
+                       do_irg_opt(irg, "local");
+                       do_irg_opt(irg, "vrp");
+               }
+       }
+
+       if (firm_dump.ir_graph) {
+               /* recompute backedges for nicer dumps */
+               for (i = 0; i < get_irp_n_irgs(); i++)
+                       construct_cf_backedges(get_irp_irg(i));
+       }
+
+       dump_all("opt");
+
+       if (firm_dump.statistic & STAT_AFTER_OPT)
+               stat_dump_snapshot(input_filename, "opt");
+
+       timer_stop(t_all_opt);
+}
 
 /**
  * do Firm lowering
@@ -799,296 +648,320 @@ static void compute_frame_type_sizes(void)
  */
 static void do_firm_lowering(const char *input_filename)
 {
-  int i;
-
-  if (firm_opt.lower_ll) {
-    lwrdw_param_t init = {
-      1,
-      1,
-      get_atomic_mode(ATOMIC_TYPE_LONGLONG),
-      get_atomic_mode(ATOMIC_TYPE_ULONGLONG),
-      get_atomic_mode(ATOMIC_TYPE_INT),
-      get_atomic_mode(ATOMIC_TYPE_UINT),
-      def_create_intrinsic_fkt,
-      NULL
-    };
-
-
-    if (arch_create_intrinsic) {
-      init.create_intrinsic = arch_create_intrinsic;
-      init.ctx              = create_intrinsic_ctx;
-    }
-    timer_push(TV_DW_LOWER);
-      lower_dw_ops(&init);
-    timer_pop();
-  }
-
-  if (firm_dump.statistic & STAT_AFTER_LOWER)
-    stat_dump_snapshot(input_filename, "low");
-
-  /* verify lowered graphs */
-  timer_push(TV_VERIFY);
-  for (i = get_irp_n_irgs() - 1; i >= 0; --i)
-    irg_verify(get_irp_irg(i), VRFY_ENFORCE_SSA);
-  timer_pop();
-
-  DUMP_ALL(firm_dump.ir_graph, "-low");
-
-  if (firm_opt.enabled) {
-    timer_start(TV_ALL_OPT);
-
-    /* run reassociation first on all graphs BEFORE the architecture dependent optimizations
-       are enabled */
-    for (i = get_irp_n_irgs() - 1; i >= 0; --i) {
-      ir_graph *irg = get_irp_irg(i);
-         do_irg_opt(irg, "reassoc");
-       }
+       int i;
 
-    /* enable architecture dependent optimizations */
-    arch_dep_set_opts((arch_dep_opts_t)
-                      ((firm_opt.muls ? arch_dep_mul_to_shift : arch_dep_none) |
-                      (firm_opt.divs ? arch_dep_div_by_const : arch_dep_none) |
-                      (firm_opt.mods ? arch_dep_mod_by_const : arch_dep_none) ));
+       /* enable architecture dependent optimizations */
+       arch_dep_set_opts((arch_dep_opts_t)
+                       ((firm_opt.muls ? arch_dep_mul_to_shift : arch_dep_none) |
+                        (firm_opt.divs ? arch_dep_div_by_const : arch_dep_none) |
+                        (firm_opt.mods ? arch_dep_mod_by_const : arch_dep_none) ));
+       for (i = get_irp_n_irgs() - 1; i >= 0; --i) {
+               ir_graph *irg = get_irp_irg(i);
+               do_irg_opt(irg, "reassociation");
+               do_irg_opt(irg, "local");
+       }
 
-    for (i = get_irp_n_irgs() - 1; i >= 0; --i) {
-      ir_graph *irg = get_irp_irg(i);
+       do_irp_opt("target-lowering");
 
-      current_ir_graph = irg;
+       if (firm_dump.statistic & STAT_AFTER_LOWER)
+               stat_dump_snapshot(input_filename, "low");
 
-         do_irg_opt(irg, "local");
-         do_irg_opt(irg, "gcse");
-         do_irg_opt(irg, "ldst");
-         do_irg_opt(irg, "local");
-         do_irg_opt(irg, "controlflow");
+       timer_start(t_all_opt);
 
-      if (firm_opt.if_conversion) {
-                 do_irg_opt(irg, "ifconv");
-                 do_irg_opt(irg, "local");
-                 do_irg_opt(irg, "controlflow");
-         }
+       for (i = get_irp_n_irgs() - 1; i >= 0; --i) {
+               ir_graph *irg = get_irp_irg(i);
 
-         do_irg_opt(current_ir_graph, "sync");
-    }
-    timer_stop(TV_ALL_OPT);
+               do_irg_opt(irg, "local");
+               do_irg_opt(irg, "deconv");
+               do_irg_opt(irg, "control-flow");
+               do_irg_opt(irg, "opt-load-store");
+               do_irg_opt(irg, "gcse");
+               do_irg_opt(irg, "place");
+               do_irg_opt(irg, "control-flow");
 
-    DUMP_ALL(firm_dump.ir_graph, "-low-opt");
-  }
+               if (do_irg_opt(irg, "vrp")) {
+                       do_irg_opt(irg, "local");
+                       do_irg_opt(irg, "control-flow");
+                       do_irg_opt(irg, "vrp");
+                       do_irg_opt(irg, "local");
+                       do_irg_opt(irg, "control-flow");
+               }
 
-  if (firm_opt.cc_opt)
-    mark_private_methods();
+               if (do_irg_opt(irg, "if-conversion")) {
+                       do_irg_opt(irg, "local");
+                       do_irg_opt(irg, "control-flow");
+               }
 
-  /* set the phase to low */
-  for (i = get_irp_n_irgs() - 1; i >= 0; --i)
-    set_irg_phase_low(get_irp_irg(i));
+               set_irg_state(irg, IR_GRAPH_STATE_NORMALISATION2);
+               do_irg_opt(irg, "local");
 
-  /* all graphs are lowered, set the irp phase to low */
-  set_irp_phase_state(phase_low);
+               do_irg_opt(irg, "parallelize-mem");
+               do_irg_opt(irg, "frame");
+       }
+       do_irp_opt("remove-unused");
+       do_irp_opt("opt-cc");
+       timer_stop(t_all_opt);
+       dump_all("low-opt");
 
-  if (firm_dump.statistic & STAT_FINAL) {
-    stat_dump_snapshot(input_filename, "final");
-  }
-}  /* do_firm_lowering */
+       if (firm_dump.statistic & STAT_FINAL) {
+               stat_dump_snapshot(input_filename, "final");
+       }
+}
 
 /**
  * Initialize for the Firm-generating back end.
  */
 void gen_firm_init(void)
 {
-  firm_parameter_t params;
-  unsigned         pattern = 0;
+       ir_init();
+       enable_safe_defaults();
 
-  /* the automatic state is only set if inlining is enabled */
-  firm_opt.auto_inline = firm_opt.do_inline;
+       FOR_EACH_OPT(i) {
+               i->timer = ir_timer_new();
+               timer_register(i->timer, i->description);
+       }
+       t_verify = ir_timer_new();
+       timer_register(t_verify, "Firm: verify pass");
+       t_vcg_dump = ir_timer_new();
+       timer_register(t_vcg_dump, "Firm: vcg dumping");
+       t_all_opt = ir_timer_new();
+       timer_register(t_all_opt, "Firm: all optimizations");
+       t_backend = ir_timer_new();
+       timer_register(t_backend, "Firm: backend");
+}
 
-  if (firm_dump.stat_pattern)
-    pattern |= FIRMSTAT_PATTERN_ENABLED;
+static void init_statistics(void)
+{
+       unsigned pattern = 0;
 
-  if (firm_dump.stat_dag)
-    pattern |= FIRMSTAT_COUNT_DAG;
+       if (firm_dump.stat_pattern)
+               pattern |= FIRMSTAT_PATTERN_ENABLED;
 
-  memset(&params, 0, sizeof(params));
-  params.size                  = sizeof(params);
-  params.enable_statistics     = firm_dump.statistic == STAT_NONE ? 0 :
-    FIRMSTAT_ENABLED | FIRMSTAT_COUNT_STRONG_OP | FIRMSTAT_COUNT_CONSTS | pattern;
-  params.initialize_local_func = uninitialized_local_var;
-  params.cc_mask               = 0; /* no regparam, cdecl */
-  params.builtin_dbg           = NULL;
+       if (firm_dump.stat_dag)
+               pattern |= FIRMSTAT_COUNT_DAG;
 
-  ir_init(&params);
+       firm_init_stat(firm_dump.statistic == STAT_NONE ?
+                       0 : FIRMSTAT_ENABLED | FIRMSTAT_COUNT_STRONG_OP
+                       | FIRMSTAT_COUNT_CONSTS | pattern);
+}
 
-  if (firm_be_opt.selection == BE_FIRM_BE) {
-    const backend_params *be_params = be_get_backend_param();
+/**
+ * Called, after the Firm generation is completed,
+ * do all optimizations and backend call here.
+ *
+ * @param out                a file handle for the output, may be NULL
+ * @param input_filename     the name of the (main) source file
+ */
+void generate_code(FILE *out, const char *input_filename)
+{
+       int i;
 
-    firm_opt.lower_ll       = (a_byte) be_params->do_dw_lowering;
+       set_optimize(1);
+       set_opt_constant_folding(firm_opt.const_folding);
+       set_opt_algebraic_simplification(firm_opt.const_folding);
+       set_opt_cse(firm_opt.cse);
+       set_opt_global_cse(0);
 
-    arch_create_intrinsic   = be_params->arch_create_intrinsic_fkt;
-    create_intrinsic_ctx    = be_params->create_intrinsic_ctx;
+       init_statistics();
 
-    ad_param                = be_params->dep_param;
-    if_conv_info            = be_params->if_conv_info;
-  }
+       do_node_verification((firm_verification_t) firm_opt.verify);
 
-  dbg_init(NULL, NULL, dbg_snprint);
-  edges_init_dbg(firm_opt.vrfy_edges);
+       /* the general for dumping option must be set, or the others will not work*/
+       firm_dump.ir_graph = (bool) (firm_dump.ir_graph | firm_dump.all_phases);
 
-  set_opt_precise_exc_context(firm_opt.precise_exc);
-  set_opt_fragile_ops(firm_opt.fragile_ops);
+       ir_add_dump_flags(ir_dump_flag_keepalive_edges
+                       | ir_dump_flag_consts_local | ir_dump_flag_dominance);
+       ir_remove_dump_flags(ir_dump_flag_loops | ir_dump_flag_ld_names);
 
-  /* Sel node cannot produce NULL pointers */
-  set_opt_sel_based_null_check_elim(1);
+       /* FIXME: cloning might ADD new graphs. */
+       irg_dump_no = calloc(get_irp_last_idx(), sizeof(*irg_dump_no));
 
-  /* dynamic dispatch works currently only if whole world scenarios */
-  set_opt_dyn_meth_dispatch(0);
+       if (firm_dump.all_types) {
+               dump_ir_prog_ext(dump_typegraph, "types.vcg");
+       }
 
-  arch_dep_init(arch_factory);
+       dump_all("");
 
-  /* do not run architecture dependent optimizations in building phase */
-  arch_dep_set_opts(arch_dep_none);
+       timer_push(t_verify);
+       tr_verify();
+       timer_pop(t_verify);
 
-  do_node_verification((firm_verification_t) firm_opt.vrfy);
-  if (firm_dump.filter)
-    only_dump_method_with_name(new_id_from_str(firm_dump.filter));
+       /* all graphs are finalized, set the irp phase to high */
+       set_irp_phase_state(phase_high);
 
-  if (firm_opt.enabled) {
-    set_optimize(1);
-    set_opt_constant_folding(firm_opt.const_folding);
-    set_opt_algebraic_simplification(firm_opt.const_folding);
-    set_opt_cse(firm_opt.cse);
-    set_opt_global_cse(0);
-    set_opt_unreachable_code(1);
-    set_opt_control_flow(firm_opt.control_flow);
-    set_opt_control_flow_weak_simplification(1);
-    set_opt_control_flow_strong_simplification(1);
-  } else {
-    set_optimize(0);
-  }
+       /* BEWARE: kill unreachable code before doing compound lowering */
+       for (i = get_irp_n_irgs() - 1; i >= 0; --i) {
+               ir_graph *irg = get_irp_irg(i);
+               do_irg_opt(irg, "control-flow");
+       }
 
-  /* do not dump entity ld names */
-  dump_ld_names(0);
-}  /* gen_firm_init */
+       if (firm_dump.statistic & STAT_BEFORE_OPT) {
+               stat_dump_snapshot(input_filename, "noopt");
+       }
 
-/**
- * Called, after the Firm generation is completed,
- * do all optimizations and backend call here.
- *
- * @param out                a file handle for the output, may be NULL
- * @param input_filename     the name of the (main) source file
- * @param c_mode             non-zero if "C" was compiled
- * @param new_firm_const_exists  non-zero, if the const attribute was used on functions
- */
-void gen_firm_finish(FILE *out, const char *input_filename, int c_mode, int new_firm_const_exists)
+       do_firm_optimizations(input_filename);
+       do_firm_lowering(input_filename);
+
+       /* set the phase to low */
+       for (i = get_irp_n_irgs() - 1; i >= 0; --i)
+               set_irg_phase_state(get_irp_irg(i), phase_low);
+
+       if (firm_dump.statistic & STAT_FINAL_IR)
+               stat_dump_snapshot(input_filename, "final-ir");
+
+       /* run the code generator */
+       timer_start(t_backend);
+       be_main(out, input_filename);
+       timer_stop(t_backend);
+
+       if (firm_dump.statistic & STAT_FINAL)
+               stat_dump_snapshot(input_filename, "final");
+}
+
+void gen_firm_finish(void)
 {
-  int i;
-
-#if 0
-  if (firm_opt.enable_statev) {
-         char buf[1024];
-         snprintf(buf, sizeof(buf), "%s.ev", input_filename);
-         ir_stat_ev_begin(input_filename, firm_opt.statev_filter);
-         ir_stat_ev_compilation_unit(input_filename);
-  }
-#endif
-
-  firm_const_exists = new_firm_const_exists;
-
-  /* the general for dumping option must be set, or the others will not work */
-  firm_dump.ir_graph
-      = (a_byte) (firm_dump.ir_graph | firm_dump.all_phases | firm_dump.extbb);
-
-  dump_keepalive_edges(1);
-  dump_consts_local(1);
-  dump_dominator_information(1);
-  dump_loop_information(0);
-
-  if (!firm_dump.edge_labels)
-    turn_off_edge_labels();
-
-  if (firm_dump.all_types) {
-    dump_all_types("");
-    if (! c_mode) {
-      dump_class_hierarchy(0, "");
-      dump_class_hierarchy(1, "-with-entities");
-    }
-  }
-
-  /* finalize all graphs */
-  for (i = get_irp_n_irgs() - 1; i >= 0; --i) {
-    ir_graph *irg = get_irp_irg(i);
-
-    irg_finalize_cons(irg);
-    DUMP_ONE(firm_dump.ir_graph, irg, "");
-
-    /* verify the graph */
-    timer_push(TV_VERIFY);
-      irg_verify(irg, VRFY_ENFORCE_SSA);
-    timer_pop();
-  }
-
-  timer_push(TV_VERIFY);
-    tr_vrfy();
-  timer_pop();
-
-  /* all graphs are finalized, set the irp phase to high */
-  set_irp_phase_state(phase_high);
-
-  /* BEWARE: kill unreachable code before doing compound lowering */
-  for (i = get_irp_n_irgs() - 1; i >= 0; --i) {
-    ir_graph *irg = get_irp_irg(i);
-    optimize_cf(irg);
-  }
-
-  /* lower all compound call return values */
-  lower_compound_params();
-
-  /* computes the sizes of all types that are still not computed */
-  compute_type_sizes();
-
-  /* lower copyb nodes */
-  for (i = get_irp_n_irgs() - 1; i >= 0; --i) {
-    ir_graph *irg = get_irp_irg(i);
-    lower_CopyB(irg, 128, 4);
-  }
-
-  if (firm_dump.statistic & STAT_BEFORE_OPT) {
-    stat_dump_snapshot(input_filename, "noopt");
-  }
-
-  if (firm_opt.enabled)
-    do_firm_optimizations(input_filename);
-
-  if (firm_dump.gen_firm_asm) {
-    timer_push(TV_FIRM_ASM);
-      gen_Firm_assembler(input_filename);
-    timer_pop();
-    return;
-  }
-
-  if (firm_opt.lower)
-    do_firm_lowering(input_filename);
-
-  /* computes the sizes of all frame types */
-  compute_frame_type_sizes();
-
-  /* set the phase to low */
-  for (i = get_irp_n_irgs() - 1; i >= 0; --i)
-    set_irg_phase_low(get_irp_irg(i));
-
-  if (firm_dump.statistic & STAT_FINAL_IR)
-    stat_dump_snapshot(input_filename, "final-ir");
-
-  /* run the code generator */
-  if (firm_be_opt.selection != BE_NONE)
-    do_codegen(out, input_filename);
-
-  if (firm_dump.statistic & STAT_FINAL)
-    stat_dump_snapshot(input_filename, "final");
+       ir_finish();
 }
 
-/**
- * Do very early initializations
- */
-void firm_early_init(void)
+static void disable_all_opts(void)
+{
+       firm_opt.cse             = false;
+       firm_opt.confirm         = false;
+       firm_opt.muls            = false;
+       firm_opt.divs            = false;
+       firm_opt.mods            = false;
+       firm_opt.alias_analysis  = false;
+       firm_opt.strict_alias    = false;
+       firm_opt.no_alias        = false;
+
+       FOR_EACH_OPT(config) {
+               if (config->flags & OPT_FLAG_ESSENTIAL) {
+                       config->flags |= OPT_FLAG_ENABLED;
+               } else {
+                       config->flags &= ~OPT_FLAG_ENABLED;
+               }
+       }
+}
+
+static bool firm_opt_option(const char *opt)
+{
+       char const* const rest   = strstart(opt, "no-");
+       bool        const enable = rest ? opt = rest, false : true;
+
+       opt_config_t *config = get_opt(opt);
+       if (config == NULL || (config->flags & OPT_FLAG_HIDE_OPTIONS))
+               return false;
+
+       config->flags &= ~OPT_FLAG_ENABLED;
+       config->flags |= enable ? OPT_FLAG_ENABLED : 0;
+       return true;
+}
+
+void firm_option_help(print_option_help_func print_option_help)
 {
-  /* arg: need this here for command line options */
-  be_opt_register();
+       FOR_EACH_OPT(config) {
+               char buf[1024];
+               char buf2[1024];
+
+               if (config->flags & OPT_FLAG_HIDE_OPTIONS)
+                       continue;
+
+               snprintf(buf, sizeof(buf), "-f%s", config->name);
+               snprintf(buf2, sizeof(buf2), "enable %s", config->description);
+               print_option_help(buf, buf2);
+               snprintf(buf, sizeof(buf), "-fno-%s", config->name);
+               snprintf(buf2, sizeof(buf2), "disable %s", config->description);
+               print_option_help(buf, buf2);
+       }
+
+       for (size_t k = 0; k != lengthof(firm_options); ++k) {
+               char buf[1024];
+               char buf2[1024];
+               snprintf(buf, sizeof(buf), "-f%s", firm_options[k].option);
+               snprintf(buf2, sizeof(buf2), "%s", firm_options[k].description);
+               print_option_help(buf, buf2);
+       }
+}
+
+int firm_option(const char *const opt)
+{
+       char const* val;
+       if ((val = strstart(opt, "dump-filter="))) {
+               ir_set_dump_filter(val);
+               return 1;
+       } else if ((val = strstart(opt, "clone-threshold="))) {
+               sscanf(val, "%d", &firm_opt.clone_threshold);
+               return 1;
+       } else if ((val = strstart(opt, "inline-max-size="))) {
+               sscanf(val, "%u", &firm_opt.inline_maxsize);
+               return 1;
+       } else if ((val = strstart(opt, "inline-threshold="))) {
+               sscanf(val, "%u", &firm_opt.inline_threshold);
+               return 1;
+       } else if (streq(opt, "no-opt")) {
+               disable_all_opts();
+               return 1;
+       }
+
+       size_t const len = strlen(opt);
+       for (size_t i = lengthof(firm_options); i != 0;) {
+               struct params const* const o = &firm_options[--i];
+               if (len == o->opt_len && memcmp(opt, o->option, len) == 0) {
+                       /* statistic options do accumulate */
+                       if (o->flag == &firm_dump.statistic)
+                               *o->flag = (bool) (*o->flag | o->set);
+                       else
+                               *o->flag = o->set;
+
+                       return 1;
+               }
+       }
+
+       /* maybe this enables/disables optimizations */
+       if (firm_opt_option(opt))
+               return 1;
+
+       return 0;
+}
+
+static void set_be_option(const char *arg)
+{
+       int res = be_parse_arg(arg);
+       (void) res;
+       assert(res);
+}
+
+static void set_option(const char *arg)
+{
+       int res = firm_option(arg);
+       (void) res;
+       assert(res);
+}
+
+void choose_optimization_pack(int level)
+{
+       /* apply optimization level */
+       switch(level) {
+       case 0:
+               set_option("no-opt");
+               break;
+       case 1:
+               set_option("no-inline");
+               break;
+       default:
+       case 4:
+               /* use_builtins = true; */
+               /* fallthrough */
+       case 3:
+               set_option("thread-jumps");
+               set_option("if-conversion");
+               /* fallthrough */
+       case 2:
+               set_option("strict-aliasing");
+               set_option("inline");
+               set_option("fp-vrp");
+               set_option("deconv");
+               set_be_option("omitfp");
+               break;
+       }
 }