1 |
879 |
markom |
/* memory.c -- OpenRISC Custom Unit Compiler, memory optimization and scheduling
|
2 |
|
|
* Copyright (C) 2002 Marko Mlinar, markom@opencores.org
|
3 |
|
|
*
|
4 |
|
|
* This file is part of OpenRISC 1000 Architectural Simulator.
|
5 |
|
|
*
|
6 |
|
|
* This program is free software; you can redistribute it and/or modify
|
7 |
|
|
* it under the terms of the GNU General Public License as published by
|
8 |
|
|
* the Free Software Foundation; either version 2 of the License, or
|
9 |
|
|
* (at your option) any later version.
|
10 |
|
|
*
|
11 |
|
|
* This program is distributed in the hope that it will be useful,
|
12 |
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
13 |
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
14 |
|
|
* GNU General Public License for more details.
|
15 |
|
|
*
|
16 |
|
|
* You should have received a copy of the GNU General Public License
|
17 |
|
|
* along with this program; if not, write to the Free Software
|
18 |
|
|
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
|
19 |
|
|
|
20 |
|
|
#include <stdio.h>
|
21 |
|
|
#include <stdlib.h>
|
22 |
|
|
#include <stdarg.h>
|
23 |
|
|
#include <assert.h>
|
24 |
897 |
markom |
#include "sim-config.h"
|
25 |
879 |
markom |
#include "cuc.h"
|
26 |
|
|
#include "insn.h"
|
27 |
|
|
|
28 |
|
|
/* Checks for memory conflicts between two instructions; returns 1 if detected
|
29 |
|
|
|
30 |
|
|
static int check_memory_conflict (cuc_func *f, cuc_insn *a, cuc_insn *b, int otype)
|
31 |
|
|
{
|
32 |
|
|
switch (otype) {
|
33 |
897 |
markom |
case MO_EXACT: /* exact */
|
34 |
|
|
case MO_STRONG: /* strong */
|
35 |
879 |
markom |
return 1;
|
36 |
897 |
markom |
case MO_WEAK: /* weak */
|
37 |
879 |
markom |
assert (a->type & IT_MEMORY);
|
38 |
|
|
assert (b->type & IT_MEMORY);
|
39 |
|
|
if ((a->opt[1] & OPT_REF) && f->INSN(a->op[1]).index == II_ADD
|
40 |
|
|
&&(b->opt[1] & OPT_REF) && f->INSN(b->op[1]).index == II_ADD) {
|
41 |
|
|
int aw, bw;
|
42 |
|
|
assert ((aw = II_MEM_WIDTH (a->index)) >= 0);
|
43 |
|
|
assert ((bw = II_MEM_WIDTH (b->index)) >= 0);
|
44 |
|
|
|
45 |
|
|
a = &f->INSN(a->op[1]);
|
46 |
|
|
b = &f->INSN(b->op[1]);
|
47 |
|
|
if (a->opt[1] != b->opt[1] || a->op[1] != b->op[1]
|
48 |
|
|
|| a->opt[2] != OPT_CONST || b->opt[2] != OPT_CONST) return 1;
|
49 |
|
|
|
50 |
|
|
/* Check if they overlap */
|
51 |
|
|
if (a->op[2] >= b->op[2] && a->op[2] < b->op[2] + bw) return 1;
|
52 |
|
|
if (b->op[2] >= a->op[2] && b->op[2] < a->op[2] + aw) return 1;
|
53 |
|
|
return 0;
|
54 |
|
|
} else return 1;
|
55 |
897 |
markom |
case MO_NONE: /* none */
|
56 |
879 |
markom |
return 0;
|
57 |
|
|
default:
|
58 |
|
|
assert (0);
|
59 |
|
|
}
|
60 |
|
|
return 1;
|
61 |
|
|
}
|
62 |
|
|
|
63 |
|
|
/* Adds memory dependencies based on ordering type:
|
64 |
|
|
|
65 |
|
|
void add_memory_dep (cuc_func *f, int otype)
|
66 |
|
|
{
|
67 |
|
|
int b, i;
|
68 |
|
|
dep_list *all_mem = NULL;
|
69 |
|
|
|
70 |
|
|
for (b = 0; b < f->num_bb; b++) {
|
71 |
|
|
cuc_insn *insn = f->bb[b].insn;
|
72 |
|
|
for (i = 0; i < f->bb[b].ninsn; i++)
|
73 |
|
|
if (insn[i].type & IT_MEMORY) {
|
74 |
|
|
dep_list *tmp = all_mem;
|
75 |
|
|
while (tmp) {
|
76 |
|
|
//printf ("%x %x\n", REF (b,i), tmp->ref);
|
77 |
|
|
if (check_memory_conflict (f, &insn[i], &f->INSN(tmp->ref), otype))
|
78 |
|
|
add_dep (&insn[i].dep, tmp->ref);
|
79 |
|
|
tmp = tmp->next;
|
80 |
|
|
}
|
81 |
|
|
add_dep (&all_mem, REF (b, i));
|
82 |
|
|
}
|
83 |
|
|
}
|
84 |
|
|
dispose_list (&all_mem);
|
85 |
|
|
}
|
86 |
|
|
|
87 |
|
|
/* returns nonzero if a < b */
|
88 |
|
|
int mem_ordering_cmp (cuc_func *f, cuc_insn *a, cuc_insn *b)
|
89 |
|
|
{
|
90 |
|
|
assert (a->type & IT_MEMORY);
|
91 |
|
|
assert (b->type & IT_MEMORY);
|
92 |
|
|
if ((a->opt[1] & OPT_REF) && f->INSN(a->op[1]).index == II_ADD
|
93 |
|
|
&&(b->opt[1] & OPT_REF) && f->INSN(b->op[1]).index == II_ADD) {
|
94 |
|
|
a = &f->INSN(a->op[1]);
|
95 |
|
|
b = &f->INSN(b->op[1]);
|
96 |
|
|
if (a->opt[1] != b->opt[1] || a->op[1] != b->op[1]
|
97 |
|
|
|| a->opt[2] != OPT_CONST || b->opt[2] != OPT_CONST) return 0;
|
98 |
|
|
|
99 |
|
|
/* Order linearly, we can then join them to bursts */
|
100 |
|
|
return a->op[2] < b->op[2];
|
101 |
|
|
} else return 0;
|
102 |
|
|
}
|
103 |
|
|
|
104 |
|
|
/* Schedule memory accesses
|
105 |
|
|
|
106 |
|
|
void schedule_memory (cuc_func *f, int otype)
|
107 |
|
|
{
|
108 |
|
|
int b, i, j;
|
109 |
|
|
f->nmsched = 0;
|
110 |
|
|
|
111 |
|
|
for (b = 0; b < f->num_bb; b++) {
|
112 |
|
|
cuc_insn *insn = f->bb[b].insn;
|
113 |
|
|
for (i = 0; i < f->bb[b].ninsn; i++)
|
114 |
|
|
if (insn[i].type & IT_MEMORY) {
|
115 |
|
|
f->msched[f->nmsched++] = REF (b, i);
|
116 |
897 |
markom |
if (otype == MO_NONE || otype == MO_WEAK) insn[i].type |= IT_FLAG1; /* mark unscheduled */
|
117 |
879 |
markom |
}
|
118 |
|
|
}
|
119 |
|
|
#if 0
|
120 |
|
|
for (i = 0; i < f->nmsched; i++)
|
121 |
|
|
printf ("[%i]%i%c ", f->msched[i], f->mtype[i] & MT_WIDTH, (f->mtype[i] & MT_BURST) ? (f->mtype[i] & MT_BURSTE) ? 'E' : 'B' : ' ');
|
122 |
|
|
printf ("\n");
|
123 |
|
|
#endif
|
124 |
|
|
|
125 |
|
|
/* We can reorder just more loose types
|
126 |
|
|
We assume, that memory accesses are currently in valid (but not neccesserly)
|
127 |
|
|
optimal order */
|
128 |
897 |
markom |
if (otype == MO_WEAK || otype == MO_NONE) {
|
129 |
879 |
markom |
for (i = 0; i < f->nmsched; i++) {
|
130 |
|
|
int best = i;
|
131 |
|
|
int tmp;
|
132 |
|
|
for (j = i + 1; j < f->nmsched; j++) if (REF_BB(f->msched[j]) == REF_BB(f->msched[best])) {
|
133 |
|
|
if (mem_ordering_cmp (f, &f->INSN (f->msched[j]), &f->INSN(f->msched[best]))) {
|
134 |
|
|
/* Check dependencies */
|
135 |
|
|
dep_list *t = f->INSN(f->msched[j]).dep;
|
136 |
|
|
while (t) {
|
137 |
|
|
if (f->INSN(t->ref).type & IT_FLAG1) break;
|
138 |
|
|
t = t->next;
|
139 |
|
|
}
|
140 |
|
|
if (!t) best = j; /* no conflicts -> ok */
|
141 |
|
|
}
|
142 |
|
|
}
|
143 |
|
|
|
144 |
|
|
/* we have to shift instructions up, to maintain valid dependencies
|
145 |
|
|
and make space for best candidate */
|
146 |
|
|
|
147 |
|
|
/* make local copy */
|
148 |
|
|
tmp = f->msched[best];
|
149 |
|
|
for (j = best; j > i; j--) f->msched[j] = f->msched[j - 1];
|
150 |
|
|
f->msched[i] = tmp;
|
151 |
|
|
f->INSN(f->msched[i]).type &= ~IT_FLAG1; /* mark scheduled */
|
152 |
|
|
}
|
153 |
|
|
}
|
154 |
|
|
|
155 |
|
|
#if 0
|
156 |
|
|
for (i = 0; i < f->nmsched; i++)
|
157 |
|
|
printf ("[%i]%i%c ", f->msched[i], f->mtype[i] & MT_WIDTH, (f->mtype[i] & MT_BURST) ? (f->mtype[i] & MT_BURSTE) ? 'E' : 'B' : ' ');
|
158 |
|
|
printf ("\n");
|
159 |
|
|
#endif
|
160 |
|
|
|
161 |
904 |
markom |
/* Assign memory types */
|
162 |
879 |
markom |
for (i = 0; i < f->nmsched; i++) {
|
163 |
|
|
cuc_insn *a = &f->INSN(f->msched[i]);
|
164 |
907 |
markom |
f->mtype[i] = !II_IS_LOAD(a->index) ? MT_STORE : MT_LOAD;
|
165 |
879 |
markom |
f->mtype[i] |= II_MEM_WIDTH (a->index);
|
166 |
|
|
if (a->type & IT_SIGNED) f->mtype[i] |= MT_SIGNED;
|
167 |
|
|
}
|
168 |
|
|
|
169 |
904 |
markom |
/* Check if they address the same location, so we can join them */
|
170 |
|
|
if (otype == MO_WEAK || otype == MO_NONE) {
|
171 |
|
|
for (i = 1, j = 1; i < f->nmsched; i++)
|
172 |
|
|
/* Exclude memory stores and different memory types */
|
173 |
907 |
markom |
if (f->mtype[i - 1] == f->mtype[i] && f->mtype[i] & MT_LOAD) {
|
174 |
904 |
markom |
cuc_insn *a = &f->INSN(f->msched[i - 1]);
|
175 |
|
|
cuc_insn *b = &f->INSN(f->msched[i]);
|
176 |
|
|
if ((a->opt[1] & OPT_REF) && f->INSN(a->op[1]).index == II_ADD
|
177 |
|
|
&&(b->opt[1] & OPT_REF) && f->INSN(b->op[1]).index == II_ADD) {
|
178 |
|
|
a = &f->INSN(a->op[1]);
|
179 |
|
|
b = &f->INSN(b->op[1]);
|
180 |
|
|
/* Not in usual form? */
|
181 |
|
|
if (a->opt[1] != b->opt[1] || a->op[1] != b->op[1]
|
182 |
|
|
|| a->opt[2] != OPT_CONST || b->opt[2] != OPT_CONST) goto keep;
|
183 |
|
|
|
184 |
|
|
//printf ("%i %i, ", a->op[2], b->op[2]);
|
185 |
|
|
|
186 |
|
|
/* Check if they are the same => do not copy */
|
187 |
|
|
if (a->op[2] == b->op[2]
|
188 |
|
|
&& REF_BB(f->msched[i - 1]) == REF_BB(f->msched[i])) {
|
189 |
|
|
/* yes => remove actual instruction */
|
190 |
|
|
int t1 = MIN (f->msched[i - 1], f->msched[i]);
|
191 |
|
|
int t2 = MAX (f->msched[i - 1], f->msched[i]);
|
192 |
|
|
int b, i, j;
|
193 |
|
|
cucdebug (2, "Removing %x_%x and using %x_%x instead.\n",
|
194 |
|
|
REF_BB(t2), REF_I(t2), REF_BB(t1), REF_I(t1));
|
195 |
|
|
change_insn_type (&f->INSN(t2), II_NOP);
|
196 |
|
|
/* Update references */
|
197 |
|
|
for (b = 0; b < f->num_bb; b++)
|
198 |
|
|
for (i = 0; i < f->bb[b].ninsn; i++)
|
199 |
|
|
for (j = 0; j < MAX_OPERANDS; j++)
|
200 |
|
|
if (f->bb[b].insn[i].opt[j] & OPT_REF && f->bb[b].insn[i].op[j] == t2)
|
201 |
|
|
f->bb[b].insn[i].op[j] = t1;
|
202 |
|
|
|
203 |
|
|
} else goto keep;
|
204 |
|
|
}
|
205 |
|
|
} else {
|
206 |
|
|
keep:
|
207 |
|
|
f->msched[j] = f->msched[i];
|
208 |
|
|
f->mtype[j++] = f->mtype[i];
|
209 |
|
|
}
|
210 |
|
|
f->nmsched = j;
|
211 |
|
|
}
|
212 |
|
|
|
213 |
897 |
markom |
if (config.cuc.enable_bursts) {
|
214 |
879 |
markom |
//printf ("\n");
|
215 |
|
|
for (i = 1; i < f->nmsched; i++) {
|
216 |
|
|
cuc_insn *a = &f->INSN(f->msched[i - 1]);
|
217 |
|
|
cuc_insn *b = &f->INSN(f->msched[i]);
|
218 |
|
|
int aw = f->mtype[i - 1] & MT_WIDTH;
|
219 |
|
|
|
220 |
|
|
if ((a->opt[1] & OPT_REF) && f->INSN(a->op[1]).index == II_ADD
|
221 |
|
|
&&(b->opt[1] & OPT_REF) && f->INSN(b->op[1]).index == II_ADD) {
|
222 |
|
|
a = &f->INSN(a->op[1]);
|
223 |
|
|
b = &f->INSN(b->op[1]);
|
224 |
|
|
/* Not in usual form? */
|
225 |
|
|
if (a->opt[1] != b->opt[1] || a->op[1] != b->op[1]
|
226 |
|
|
|| a->opt[2] != OPT_CONST || b->opt[2] != OPT_CONST) continue;
|
227 |
|
|
|
228 |
|
|
//printf ("%i %i, ", a->op[2], b->op[2]);
|
229 |
|
|
|
230 |
|
|
/* Check if they touch together */
|
231 |
|
|
if (a->op[2] + aw == b->op[2]) {
|
232 |
|
|
/* yes => do burst */
|
233 |
|
|
f->mtype[i - 1] &= ~MT_BURSTE;
|
234 |
|
|
f->mtype[i - 1] |= MT_BURST;
|
235 |
|
|
f->mtype[i] |= MT_BURST | MT_BURSTE;
|
236 |
|
|
}
|
237 |
|
|
}
|
238 |
|
|
}
|
239 |
|
|
}
|
240 |
|
|
|
241 |
|
|
#if 0
|
242 |
|
|
printf ("\n");
|
243 |
|
|
for (i = 0; i < f->nmsched; i++)
|
244 |
|
|
printf ("[%i]%i%c ", f->msched[i], f->mtype[i] & MT_WIDTH, (f->mtype[i] & MT_BURST) ? (f->mtype[i] & MT_BURSTE) ? 'E' : 'B' : ' ');
|
245 |
|
|
printf ("\n");
|
246 |
|
|
#endif
|
247 |
|
|
|
248 |
|
|
/* We don't need dependencies in non-memory instructions */
|
249 |
|
|
for (b = 0; b < f->num_bb; b++) {
|
250 |
|
|
cuc_insn *insn = f->bb[b].insn;
|
251 |
|
|
for (i = 0; i < f->bb[b].ninsn; i++) if (!(insn[i].type & IT_MEMORY))
|
252 |
|
|
dispose_list (&insn[i].dep);
|
253 |
|
|
}
|
254 |
|
|
|
255 |
|
|
/* Reduce number of dependecies, keeping just direct dependencies, based on memory schedule */
|
256 |
|
|
{
|
257 |
907 |
markom |
int lastl[3] = {-1, -1, -1};
|
258 |
|
|
int lasts[3] = {-1, -1, -1};
|
259 |
|
|
int lastc[3] = {-1, -1, -1};
|
260 |
|
|
int last_load = -1, last_store = -1, last_call = -1;
|
261 |
879 |
markom |
for (i = 0; i < f->nmsched; i++) {
|
262 |
907 |
markom |
int t = f->mtype[i] & MT_LOAD ? 0 : f->mtype[i] & MT_STORE ? 1 : 2;
|
263 |
879 |
markom |
int maxl = lastl[t];
|
264 |
|
|
int maxs = lasts[t];
|
265 |
907 |
markom |
int maxc = lastc[t];
|
266 |
879 |
markom |
dep_list *tmp = f->INSN(f->msched[i]).dep;
|
267 |
|
|
while (tmp) {
|
268 |
|
|
if (f->INSN(tmp->ref).type & IT_MEMORY && REF_BB(tmp->ref) == REF_BB(f->msched[i])) {
|
269 |
|
|
/* Search for the reference */
|
270 |
|
|
for (j = 0; j < f->nmsched; j++) if (f->msched[j] == tmp->ref) break;
|
271 |
|
|
assert (j < f->nmsched);
|
272 |
907 |
markom |
if (f->mtype[j] & MT_STORE) {
|
273 |
879 |
markom |
if (maxs < j) maxs = j;
|
274 |
907 |
markom |
} else if (f->mtype[j] & MT_LOAD) {
|
275 |
879 |
markom |
if (maxl < j) maxl = j;
|
276 |
907 |
markom |
} else if (f->mtype[j] & MT_CALL) {
|
277 |
|
|
if (maxc < j) maxc = j;
|
278 |
879 |
markom |
}
|
279 |
|
|
}
|
280 |
|
|
tmp = tmp->next;
|
281 |
|
|
}
|
282 |
|
|
dispose_list (&f->INSN(f->msched[i]).dep);
|
283 |
907 |
markom |
if (f->mtype[i] & MT_STORE) {
|
284 |
879 |
markom |
maxs = last_store;
|
285 |
|
|
last_store = i;
|
286 |
907 |
markom |
} else if (f->mtype[i] & MT_LOAD) {
|
287 |
879 |
markom |
maxl = last_load;
|
288 |
|
|
last_load = i;
|
289 |
907 |
markom |
} else if (f->mtype[i] & MT_CALL) {
|
290 |
|
|
maxc = last_call;
|
291 |
|
|
last_call = i;
|
292 |
879 |
markom |
}
|
293 |
|
|
|
294 |
|
|
if (maxl > lastl[t]) {
|
295 |
|
|
add_dep (&f->INSN(f->msched[i]).dep, f->msched[maxl]);
|
296 |
|
|
lastl[t] = maxl;
|
297 |
|
|
}
|
298 |
|
|
if (maxs > lasts[t]) {
|
299 |
|
|
add_dep (&f->INSN(f->msched[i]).dep, f->msched[maxs]);
|
300 |
|
|
lasts[t] = maxs;
|
301 |
|
|
}
|
302 |
907 |
markom |
if (maxc > lastc[t]) {
|
303 |
|
|
add_dep (&f->INSN(f->msched[i]).dep, f->msched[maxc]);
|
304 |
|
|
lastc[t] = maxc;
|
305 |
|
|
}
|
306 |
879 |
markom |
//printf ("%i(%i)> ml %i(%i) ms %i(%i) lastl %i %i lasts %i %i last_load %i last_store %i\n", i, f->msched[i], maxl, f->msched[maxl], maxs, f->msched[maxs], lastl[0], lastl[1], lasts[0], lasts[1], last_load, last_store);
|
307 |
|
|
|
308 |
|
|
/* What we have to wait to finish this BB? */
|
309 |
|
|
if (i + 1 >= f->nmsched || REF_BB(f->msched[i + 1]) != REF_BB(f->msched[i])) {
|
310 |
|
|
if (last_load > lastl[t]) {
|
311 |
|
|
add_dep (&f->bb[REF_BB(f->msched[i])].mdep, f->msched[last_load]);
|
312 |
|
|
lastl[t] = last_load;
|
313 |
|
|
}
|
314 |
|
|
if (last_store > lasts[t]) {
|
315 |
|
|
add_dep (&f->bb[REF_BB(f->msched[i])].mdep, f->msched[last_store]);
|
316 |
|
|
lasts[t] = last_store;
|
317 |
|
|
}
|
318 |
907 |
markom |
if (last_call > lastc[t]) {
|
319 |
|
|
add_dep (&f->bb[REF_BB(f->msched[i])].mdep, f->msched[last_call]);
|
320 |
|
|
lastc[t] = last_call;
|
321 |
|
|
}
|
322 |
879 |
markom |
}
|
323 |
|
|
}
|
324 |
|
|
}
|
325 |
|
|
}
|