/*
- * Copyright (C) 1995-2007 University of Karlsruhe. All right reserved.
+ * Copyright (C) 1995-2008 University of Karlsruhe. All right reserved.
*
* This file is part of libFirm.
*
* @file
* @brief conv node optimisation
* @author Matthias Braun, Christoph Mallon
- * @version $Id: condeval.c 13543 2007-04-29 19:29:02Z beck $
+ * @version $Id$
*
* Try to minimize the number of conv nodes by changing modes of operations.
* The typical example is the following structure:
* TODO: * try to optimize cmp modes
* * decide when it is useful to move the convs through phis
*/
-#ifdef HAVE_CONFIG_H
#include "config.h"
-#endif
+
+#include "iroptimize.h"
#include <assert.h>
#include "debug.h"
#include "ircons.h"
#include "irgmod.h"
+#include "irgopt.h"
#include "irnode_t.h"
#include "iredges_t.h"
#include "irgwalk.h"
DEBUG_ONLY(static firm_dbg_module_t *dbg);
+static inline int imin(int a, int b) { return a < b ? a : b; }
+
static
int is_optimizable_node(const ir_node *node)
{
- if(is_Const(node)) {
- ir_mode *mode = get_irn_mode(node);
- /* tarval module is incomplete and can't convert floats to ints */
- if(!mode_is_int(mode))
- return 0;
- return 1;
+ switch (get_irn_opcode(node)) {
+ case iro_Add:
+ case iro_And:
+ case iro_Eor:
+ case iro_Minus:
+ case iro_Mul:
+ case iro_Not:
+ case iro_Or:
+ case iro_Phi:
+ case iro_Shl:
+ case iro_Sub:
+ return 1;
+
+ default: return 0;
}
- return is_Add(node) || is_Sub(node) || is_Mul(node);
+}
+
+static tarval* conv_const_tv(const ir_node* cnst, ir_mode* dest_mode)
+{
+ return tarval_convert_to(get_Const_tarval(cnst), dest_mode);
+}
+
+static
+int is_downconv(ir_mode *src_mode, ir_mode *dest_mode)
+{
+ return
+ mode_is_int(src_mode) &&
+ mode_is_int(dest_mode) &&
+ get_mode_size_bits(dest_mode) <= get_mode_size_bits(src_mode);
}
static
int get_conv_costs(const ir_node *node, ir_mode *dest_mode)
{
ir_mode *mode = get_irn_mode(node);
+ size_t arity;
+ size_t i;
+ int costs;
- if(mode == dest_mode)
+ if (mode == dest_mode)
return 0;
- /* TODO... */
- if(!is_Const(node) && get_irn_n_edges(node) > 1) {
- DB((dbg, LEVEL_3, "multi outs at %+F\n", node));
- return 1;
+ if (is_Const(node)) {
+ /* TODO tarval module is incomplete and can't convert floats to ints */
+ return conv_const_tv(node, dest_mode) == tarval_bad ? 1 : 0;
}
- if(is_Conv(node)) {
- return get_conv_costs(get_Conv_op(node), dest_mode) - 1;
+ if (is_Conv(node) &&
+ is_downconv(mode, dest_mode) &&
+ get_irn_mode(get_Conv_op(node)) == dest_mode) {
+ return -1;
}
- if(is_optimizable_node(node)) {
- int i;
- int arity = get_irn_arity(node);
- int costs = 0;
+ if (get_irn_n_edges(node) > 1) {
+ DB((dbg, LEVEL_3, "multi outs at %+F\n", node));
+ return 1;
+ }
- for(i = 0; i < arity; ++i) {
- ir_node *pred = get_irn_n(node, i);
- costs += get_conv_costs(pred, dest_mode);
+#if 0 // TODO
+ /* Take the minimum of the conversion costs for Phi predecessors as only one
+ * branch is actually executed at a time */
+ if (is_Phi(node)) {
+ size_t i;
+ size_t arity = get_Phi_n_preds(node);
+ int costs;
+
+ costs = get_conv_costs(get_Phi_pred(node, 0), dest_mode);
+ for (i = 1; i < arity; ++i) {
+ ir_node *pred = get_Phi_pred(node, i);
+ int c = get_conv_costs(pred, dest_mode);
+ if (c < costs) costs = c;
}
return costs;
}
+#endif
+
+ if (!is_downconv(mode, dest_mode)) {
+ return 1;
+ }
+
+ if (is_Conv(node)) {
+ return get_conv_costs(get_Conv_op(node), dest_mode) - 1;
+ }
+
+ if (!is_optimizable_node(node)) {
+ return 1;
+ }
- return 1;
+ costs = 0;
+ // The shift count does not participate in the conv optimisation
+ arity = is_Shl(node) ? 1 : get_irn_arity(node);
+ for (i = 0; i < arity; ++i) {
+ ir_node *pred = get_irn_n(node, i);
+ costs += imin(get_conv_costs(pred, dest_mode), 1);
+ }
+
+ return costs;
+}
+
+static ir_node *place_conv(ir_node *node, ir_mode *dest_mode)
+{
+ ir_node *block = get_nodes_block(node);
+ ir_node *conv = new_r_Conv(current_ir_graph, block, node, dest_mode);
+ return conv;
}
static
ir_node *conv_transform(ir_node *node, ir_mode *dest_mode)
{
- size_t arity;
- size_t i;
+ ir_mode *mode = get_irn_mode(node);
+ size_t arity;
+ size_t i;
- if (get_irn_mode(node) == dest_mode)
+ if (mode == dest_mode)
return node;
if (is_Const(node)) {
- tarval *tv = tarval_convert_to(get_Const_tarval(node), dest_mode);
- assert(get_tarval_mode(tv) == dest_mode);
- return new_Const(dest_mode, tv);
+ /* TODO tarval module is incomplete and can't convert floats to ints */
+ tarval *tv = conv_const_tv(node, dest_mode);
+ if (tv == tarval_bad) {
+ return place_conv(node, dest_mode);
+ } else {
+ return new_Const(tv);
+ }
+ }
+
+ if (is_Conv(node) &&
+ is_downconv(mode, dest_mode) &&
+ get_irn_mode(get_Conv_op(node)) == dest_mode) {
+ return get_Conv_op(node);
}
- if (!is_optimizable_node(node) || get_irn_n_edges(node) > 1) {
- ir_node *block = get_nodes_block(node);
- ir_node *conv = new_r_Conv(current_ir_graph, block, node, dest_mode);
- return conv;
+ if (get_irn_n_edges(node) > 1) {
+ return place_conv(node, dest_mode);
}
- if (is_Conv(node))
+ if (!is_downconv(mode, dest_mode)) {
+ return place_conv(node, dest_mode);
+ }
+
+ if (is_Conv(node)) {
return conv_transform(get_Conv_op(node), dest_mode);
+ }
- arity = get_irn_arity(node);
+ if (!is_optimizable_node(node)) {
+ return place_conv(node, dest_mode);
+ }
+
+ // The shift count does not participate in the conv optimisation
+ arity = is_Shl(node) ? 1 : get_irn_arity(node);
for (i = 0; i < arity; i++) {
ir_node *pred = get_irn_n(node, i);
- ir_node *transformed = conv_transform(pred, dest_mode);
+ ir_node *transformed;
+ if (get_conv_costs(pred, dest_mode) > 0) {
+ transformed = place_conv(pred, dest_mode);
+ } else {
+ transformed = conv_transform(pred, dest_mode);
+ }
set_irn_n(node, i, transformed);
}
set_irn_mode(node, dest_mode);
return node;
}
-static
-int is_downconv(ir_mode *src_mode, ir_mode *dest_mode)
-{
- if(!mode_is_int(src_mode) || !mode_is_int(dest_mode))
- return 0;
- if(get_mode_size_bits(dest_mode) >= get_mode_size_bits(src_mode))
- return 0;
-
- return 1;
-}
-
-/* TODO, backends can't handle and it's probably not more efficient on most
- archs */
+/* TODO, backends (at least ia32) can't handle it at the moment,
+ and it's probably not more efficient on most archs */
#if 0
static
void try_optimize_cmp(ir_node *node)
}
#endif
+static char changed;
+
static
void conv_opt_walker(ir_node *node, void *data)
{
ir_mode *pred_mode;
ir_mode *mode;
int costs;
+ (void) data;
#if 0
if(is_Cmp(node)) {
}
#endif
- if(!is_Conv(node))
+ if (!is_Conv(node))
return;
pred = get_Conv_op(node);
mode = get_irn_mode(node);
pred_mode = get_irn_mode(pred);
- if(!is_downconv(pred_mode, mode))
+ if (mode_is_reference(mode) || mode_is_reference(pred_mode))
return;
- costs = get_conv_costs(pred, mode);
+ if (!is_Phi(pred) && !is_downconv(pred_mode, mode))
+ return;
+
+ /* - 1 for the initial conv */
+ costs = get_conv_costs(pred, mode) - 1;
DB((dbg, LEVEL_2, "Costs for %+F -> %+F: %d\n", node, pred, costs));
- if (costs >= 0) return;
+ if (costs > 0) return;
transformed = conv_transform(pred, mode);
- exchange(node, transformed);
+ if (node != transformed) {
+ exchange(node, transformed);
+ changed = 1;
+ }
}
-void conv_opt(ir_graph *irg)
+int conv_opt(ir_graph *irg)
{
+ char invalidate = 0;
FIRM_DBG_REGISTER(dbg, "firm.opt.conv");
DB((dbg, LEVEL_1, "===> Performing conversion optimization on %+F\n", irg));
edges_assure(irg);
- irg_walk_graph(irg, conv_opt_walker, NULL, NULL);
+ do {
+ changed = 0;
+ irg_walk_graph(irg, NULL, conv_opt_walker, NULL);
+ local_optimize_graph(irg);
+ invalidate |= changed;
+ } while (changed);
+
+ if (invalidate) {
+ set_irg_outs_inconsistent(irg);
+ }
+ return invalidate;
}