Go to most recent revision | Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
4358 | Serge | 1 | /* |
2 | * Copyright 2013 Vadim Girlin |
||
3 | * |
||
4 | * Permission is hereby granted, free of charge, to any person obtaining a |
||
5 | * copy of this software and associated documentation files (the "Software"), |
||
6 | * to deal in the Software without restriction, including without limitation |
||
7 | * on the rights to use, copy, modify, merge, publish, distribute, sub |
||
8 | * license, and/or sell copies of the Software, and to permit persons to whom |
||
9 | * the Software is furnished to do so, subject to the following conditions: |
||
10 | * |
||
11 | * The above copyright notice and this permission notice (including the next |
||
12 | * paragraph) shall be included in all copies or substantial portions of the |
||
13 | * Software. |
||
14 | * |
||
15 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
||
16 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
||
17 | * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL |
||
18 | * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, |
||
19 | * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR |
||
20 | * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE |
||
21 | * USE OR OTHER DEALINGS IN THE SOFTWARE. |
||
22 | * |
||
23 | * Authors: |
||
24 | * Vadim Girlin |
||
25 | */ |
||
26 | |||
27 | #include "sb_bc.h" |
||
28 | #include "sb_shader.h" |
||
29 | #include "sb_pass.h" |
||
30 | |||
31 | namespace r600_sb { |
||
32 | |||
33 | static const char* chans = "xyzw01?_"; |
||
34 | |||
35 | static const char* vec_bs[] = { |
||
36 | "VEC_012", "VEC_021", "VEC_120", "VEC_102", "VEC_201", "VEC_210" |
||
37 | }; |
||
38 | |||
39 | static const char* scl_bs[] = { |
||
40 | "SCL_210", "SCL_122", "SCL_212", "SCL_221" |
||
41 | }; |
||
42 | |||
43 | |||
44 | bool bc_dump::visit(cf_node& n, bool enter) { |
||
45 | if (enter) { |
||
46 | |||
47 | id = n.bc.id << 1; |
||
48 | |||
49 | if ((n.bc.op_ptr->flags & CF_ALU) && n.bc.is_alu_extended()) { |
||
50 | dump_dw(id, 2); |
||
51 | id += 2; |
||
52 | sblog << "\n"; |
||
53 | } |
||
54 | |||
55 | dump_dw(id, 2); |
||
56 | dump(n); |
||
57 | |||
58 | if (n.bc.op_ptr->flags & CF_CLAUSE) { |
||
59 | id = n.bc.addr << 1; |
||
60 | new_group = 1; |
||
61 | } |
||
62 | } |
||
63 | return true; |
||
64 | } |
||
65 | |||
66 | bool bc_dump::visit(alu_node& n, bool enter) { |
||
67 | if (enter) { |
||
68 | sblog << " "; |
||
69 | dump_dw(id, 2); |
||
70 | |||
71 | if (new_group) { |
||
72 | sblog.print_w(++group_index, 5); |
||
73 | sblog << " "; |
||
74 | } else |
||
75 | sblog << " "; |
||
76 | |||
77 | dump(n); |
||
78 | id += 2; |
||
79 | |||
80 | new_group = n.bc.last; |
||
81 | } else { |
||
82 | if (n.bc.last) { |
||
83 | alu_group_node *g = |
||
84 | static_cast |
||
85 | assert(g); |
||
86 | for (unsigned k = 0; k < g->literals.size(); ++k) { |
||
87 | sblog << " "; |
||
88 | dump_dw(id, 1); |
||
89 | id += 1; |
||
90 | sblog << "\n"; |
||
91 | } |
||
92 | |||
93 | id = (id + 1) & ~1u; |
||
94 | } |
||
95 | } |
||
96 | |||
97 | return false; |
||
98 | } |
||
99 | |||
100 | bool bc_dump::visit(fetch_node& n, bool enter) { |
||
101 | if (enter) { |
||
102 | sblog << " "; |
||
103 | dump_dw(id, 3); |
||
104 | dump(n); |
||
105 | id += 4; |
||
106 | } |
||
107 | return false; |
||
108 | } |
||
109 | |||
110 | static void fill_to(sb_ostringstream &s, int pos) { |
||
111 | int l = s.str().length(); |
||
112 | if (l < pos) |
||
113 | s << std::string(pos-l, ' '); |
||
114 | } |
||
115 | |||
116 | void bc_dump::dump(cf_node& n) { |
||
117 | sb_ostringstream s; |
||
118 | s << n.bc.op_ptr->name; |
||
119 | |||
120 | if (n.bc.op_ptr->flags & CF_EXP) { |
||
121 | static const char *exp_type[] = {"PIXEL", "POS ", "PARAM"}; |
||
122 | |||
123 | fill_to(s, 18); |
||
124 | s << " " << exp_type[n.bc.type] << " "; |
||
125 | |||
126 | if (n.bc.burst_count) { |
||
127 | sb_ostringstream s2; |
||
128 | s2 << n.bc.array_base << "-" << n.bc.array_base + n.bc.burst_count; |
||
129 | s.print_wl(s2.str(), 5); |
||
130 | s << " R" << n.bc.rw_gpr << "-" << |
||
131 | n.bc.rw_gpr + n.bc.burst_count << "."; |
||
132 | } else { |
||
133 | s.print_wl(n.bc.array_base, 5); |
||
134 | s << " R" << n.bc.rw_gpr << "."; |
||
135 | } |
||
136 | |||
137 | for (int k = 0; k < 4; ++k) |
||
138 | s << chans[n.bc.sel[k]]; |
||
139 | |||
140 | } else if (n.bc.op_ptr->flags & (CF_STRM | CF_RAT)) { |
||
141 | static const char *exp_type[] = {"WRITE", "WRITE_IND", "WRITE_ACK", |
||
142 | "WRITE_IND_ACK"}; |
||
143 | fill_to(s, 18); |
||
144 | s << " " << exp_type[n.bc.type] << " "; |
||
145 | s.print_wl(n.bc.array_base, 5); |
||
146 | s << " R" << n.bc.rw_gpr << "."; |
||
147 | for (int k = 0; k < 4; ++k) |
||
148 | s << ((n.bc.comp_mask & (1 << k)) ? chans[k] : '_'); |
||
149 | |||
150 | if ((n.bc.op_ptr->flags & CF_RAT) && (n.bc.type & 1)) { |
||
151 | s << ", @R" << n.bc.index_gpr << ".xyz"; |
||
152 | } |
||
153 | |||
154 | s << " ES:" << n.bc.elem_size; |
||
155 | |||
156 | } else { |
||
157 | |||
158 | if (n.bc.op_ptr->flags & CF_CLAUSE) { |
||
159 | s << " " << n.bc.count+1; |
||
160 | } |
||
161 | |||
162 | s << " @" << (n.bc.addr << 1); |
||
163 | |||
164 | if (n.bc.op_ptr->flags & CF_ALU) { |
||
165 | |||
166 | for (int k = 0; k < 4; ++k) { |
||
167 | bc_kcache &kc = n.bc.kc[k]; |
||
168 | if (kc.mode) { |
||
169 | s << " KC" << k << "[CB" << kc.bank << ":" << |
||
170 | (kc.addr << 4) << "-" << |
||
171 | (((kc.addr + kc.mode) << 4) - 1) << "]"; |
||
172 | } |
||
173 | } |
||
174 | } |
||
175 | |||
176 | if (n.bc.cond) |
||
177 | s << " CND:" << n.bc.pop_count; |
||
178 | |||
179 | if (n.bc.pop_count) |
||
180 | s << " POP:" << n.bc.pop_count; |
||
181 | } |
||
182 | |||
183 | if (!n.bc.barrier) |
||
184 | s << " NO_BARRIER"; |
||
185 | |||
186 | if (n.bc.valid_pixel_mode) |
||
187 | s << " VPM"; |
||
188 | |||
189 | if (n.bc.whole_quad_mode) |
||
190 | s << " WQM"; |
||
191 | |||
192 | if (n.bc.end_of_program) |
||
193 | s << " EOP"; |
||
194 | |||
195 | sblog << s.str() << "\n"; |
||
196 | } |
||
197 | |||
198 | |||
199 | static void print_sel(sb_ostream &s, int sel, int rel, int index_mode, |
||
200 | int need_brackets) { |
||
201 | if (rel && index_mode >= 5 && sel < 128) |
||
202 | s << "G"; |
||
203 | if (rel || need_brackets) { |
||
204 | s << "["; |
||
205 | } |
||
206 | s << sel; |
||
207 | if (rel) { |
||
208 | if (index_mode == 0 || index_mode == 6) |
||
209 | s << "+AR"; |
||
210 | else if (index_mode == 4) |
||
211 | s << "+AL"; |
||
212 | } |
||
213 | if (rel || need_brackets) { |
||
214 | s << "]"; |
||
215 | } |
||
216 | } |
||
217 | |||
218 | static void print_dst(sb_ostream &s, bc_alu &alu) |
||
219 | { |
||
220 | unsigned sel = alu.dst_gpr; |
||
221 | char reg_char = 'R'; |
||
222 | if (sel >= 128 - 4) { // clause temporary gpr |
||
223 | sel -= 128 - 4; |
||
224 | reg_char = 'T'; |
||
225 | } |
||
226 | |||
227 | if (alu.write_mask || alu.op_ptr->src_count == 3) { |
||
228 | s << reg_char; |
||
229 | print_sel(s, sel, alu.dst_rel, alu.index_mode, 0); |
||
230 | } else { |
||
231 | s << "__"; |
||
232 | } |
||
233 | s << "."; |
||
234 | s << chans[alu.dst_chan]; |
||
235 | } |
||
236 | |||
237 | static void print_src(sb_ostream &s, bc_alu &alu, unsigned idx) |
||
238 | { |
||
239 | bc_alu_src *src = &alu.src[idx]; |
||
240 | unsigned sel = src->sel, need_sel = 1, need_chan = 1, need_brackets = 0; |
||
241 | |||
242 | if (src->neg) |
||
243 | s <<"-"; |
||
244 | if (src->abs) |
||
245 | s <<"|"; |
||
246 | |||
247 | if (sel < 128 - 4) { |
||
248 | s << "R"; |
||
249 | } else if (sel < 128) { |
||
250 | s << "T"; |
||
251 | sel -= 128 - 4; |
||
252 | } else if (sel < 160) { |
||
253 | s << "KC0"; |
||
254 | need_brackets = 1; |
||
255 | sel -= 128; |
||
256 | } else if (sel < 192) { |
||
257 | s << "KC1"; |
||
258 | need_brackets = 1; |
||
259 | sel -= 160; |
||
260 | } else if (sel >= 448) { |
||
261 | s << "Param"; |
||
262 | sel -= 448; |
||
263 | } else if (sel >= 288) { |
||
264 | s << "KC3"; |
||
265 | need_brackets = 1; |
||
266 | sel -= 288; |
||
267 | } else if (sel >= 256) { |
||
268 | s << "KC2"; |
||
269 | need_brackets = 1; |
||
270 | sel -= 256; |
||
271 | } else { |
||
272 | need_sel = 0; |
||
273 | need_chan = 0; |
||
274 | switch (sel) { |
||
275 | case ALU_SRC_PS: |
||
276 | s << "PS"; |
||
277 | break; |
||
278 | case ALU_SRC_PV: |
||
279 | s << "PV"; |
||
280 | need_chan = 1; |
||
281 | break; |
||
282 | case ALU_SRC_LITERAL: |
||
283 | s << "[0x"; |
||
284 | s.print_zw_hex(src->value.u, 8); |
||
285 | s << " " << src->value.f << "]"; |
||
286 | need_chan = 1; |
||
287 | break; |
||
288 | case ALU_SRC_0_5: |
||
289 | s << "0.5"; |
||
290 | break; |
||
291 | case ALU_SRC_M_1_INT: |
||
292 | s << "-1"; |
||
293 | break; |
||
294 | case ALU_SRC_1_INT: |
||
295 | s << "1"; |
||
296 | break; |
||
297 | case ALU_SRC_1: |
||
298 | s << "1.0"; |
||
299 | break; |
||
300 | case ALU_SRC_0: |
||
301 | s << "0"; |
||
302 | break; |
||
303 | default: |
||
304 | s << "??IMM_" << sel; |
||
305 | break; |
||
306 | } |
||
307 | } |
||
308 | |||
309 | if (need_sel) |
||
310 | print_sel(s, sel, src->rel, alu.index_mode, need_brackets); |
||
311 | |||
312 | if (need_chan) { |
||
313 | s << "." << chans[src->chan]; |
||
314 | } |
||
315 | |||
316 | if (src->abs) |
||
317 | s << "|"; |
||
318 | } |
||
319 | void bc_dump::dump(alu_node& n) { |
||
320 | sb_ostringstream s; |
||
321 | static const char *omod_str[] = {"","*2","*4","/2"}; |
||
322 | static const char *slots = "xyzwt"; |
||
323 | |||
324 | s << (n.bc.update_exec_mask ? "M" : " "); |
||
325 | s << (n.bc.update_pred ? "P" : " "); |
||
326 | s << " "; |
||
327 | s << (n.bc.pred_sel>=2 ? (n.bc.pred_sel == 2 ? "0" : "1") : " "); |
||
328 | s << " "; |
||
329 | |||
330 | s << slots[n.bc.slot] << ": "; |
||
331 | |||
332 | s << n.bc.op_ptr->name << omod_str[n.bc.omod] << (n.bc.clamp ? "_sat" : ""); |
||
333 | fill_to(s, 26); |
||
334 | s << " "; |
||
335 | |||
336 | print_dst(s, n.bc); |
||
337 | for (int k = 0; k < n.bc.op_ptr->src_count; ++k) { |
||
338 | s << (k ? ", " : ", "); |
||
339 | print_src(s, n.bc, k); |
||
340 | } |
||
341 | |||
342 | if (n.bc.bank_swizzle) { |
||
343 | fill_to(s, 55); |
||
344 | if (n.bc.slot == SLOT_TRANS) |
||
345 | s << " " << scl_bs[n.bc.bank_swizzle]; |
||
346 | else |
||
347 | s << " " << vec_bs[n.bc.bank_swizzle]; |
||
348 | } |
||
349 | |||
350 | sblog << s.str() << "\n"; |
||
351 | } |
||
352 | |||
353 | int bc_dump::init() { |
||
354 | sb_ostringstream s; |
||
355 | s << "===== SHADER #" << sh.id; |
||
356 | |||
357 | if (sh.optimized) |
||
358 | s << " OPT"; |
||
359 | |||
360 | s << " "; |
||
361 | |||
362 | std::string target = std::string(" ") + |
||
363 | sh.get_full_target_name() + " ====="; |
||
364 | |||
365 | while (s.str().length() + target.length() < 80) |
||
366 | s << "="; |
||
367 | |||
368 | s << target; |
||
369 | |||
370 | sblog << "\n" << s.str() << "\n"; |
||
371 | |||
372 | s.clear(); |
||
373 | |||
374 | if (bc_data) { |
||
375 | s << "===== " << ndw << " dw ===== " << sh.ngpr |
||
376 | << " gprs ===== " << sh.nstack << " stack "; |
||
377 | } |
||
378 | |||
379 | while (s.str().length() < 80) |
||
380 | s << "="; |
||
381 | |||
382 | sblog << s.str() << "\n"; |
||
383 | |||
384 | return 0; |
||
385 | } |
||
386 | |||
387 | int bc_dump::done() { |
||
388 | sb_ostringstream s; |
||
389 | s << "===== SHADER_END "; |
||
390 | |||
391 | while (s.str().length() < 80) |
||
392 | s << "="; |
||
393 | |||
394 | sblog << s.str() << "\n\n"; |
||
395 | |||
396 | return 0; |
||
397 | } |
||
398 | |||
399 | bc_dump::bc_dump(shader& s, bytecode* bc) : |
||
400 | vpass(s), bc_data(), ndw(), id(), |
||
401 | new_group(), group_index() { |
||
402 | |||
403 | if (bc) { |
||
404 | bc_data = bc->data(); |
||
405 | ndw = bc->ndw(); |
||
406 | } |
||
407 | } |
||
408 | |||
409 | void bc_dump::dump(fetch_node& n) { |
||
410 | sb_ostringstream s; |
||
411 | static const char * fetch_type[] = {"VERTEX", "INSTANCE", ""}; |
||
412 | |||
413 | s << n.bc.op_ptr->name; |
||
414 | fill_to(s, 20); |
||
415 | |||
416 | s << "R"; |
||
417 | print_sel(s, n.bc.dst_gpr, n.bc.dst_rel, INDEX_LOOP, 0); |
||
418 | s << "."; |
||
419 | for (int k = 0; k < 4; ++k) |
||
420 | s << chans[n.bc.dst_sel[k]]; |
||
421 | s << ", "; |
||
422 | |||
423 | s << "R"; |
||
424 | print_sel(s, n.bc.src_gpr, n.bc.src_rel, INDEX_LOOP, 0); |
||
425 | s << "."; |
||
426 | |||
427 | unsigned vtx = n.bc.op_ptr->flags & FF_VTX; |
||
428 | unsigned num_src_comp = vtx ? ctx.is_cayman() ? 2 : 1 : 4; |
||
429 | |||
430 | for (unsigned k = 0; k < num_src_comp; ++k) |
||
431 | s << chans[n.bc.src_sel[k]]; |
||
432 | |||
433 | if (vtx && n.bc.offset[0]) { |
||
434 | s << " + " << n.bc.offset[0] << "b "; |
||
435 | } |
||
436 | |||
437 | s << ", RID:" << n.bc.resource_id; |
||
438 | |||
439 | if (vtx) { |
||
440 | s << " " << fetch_type[n.bc.fetch_type]; |
||
441 | if (!ctx.is_cayman() && n.bc.mega_fetch_count) |
||
442 | s << " MFC:" << n.bc.mega_fetch_count; |
||
443 | if (n.bc.fetch_whole_quad) |
||
444 | s << " FWQ"; |
||
445 | s << " UCF:" << n.bc.use_const_fields |
||
446 | << " FMT(DTA:" << n.bc.data_format |
||
447 | << " NUM:" << n.bc.num_format_all |
||
448 | << " COMP:" << n.bc.format_comp_all |
||
449 | << " MODE:" << n.bc.srf_mode_all << ")"; |
||
450 | } else { |
||
451 | s << ", SID:" << n.bc.sampler_id; |
||
452 | if (n.bc.lod_bias) |
||
453 | s << " LB:" << n.bc.lod_bias; |
||
454 | s << " CT:"; |
||
455 | for (unsigned k = 0; k < 4; ++k) |
||
456 | s << (n.bc.coord_type[k] ? "N" : "U"); |
||
457 | for (unsigned k = 0; k < 3; ++k) |
||
458 | if (n.bc.offset[k]) |
||
459 | s << " O" << chans[k] << ":" << n.bc.offset[k]; |
||
460 | } |
||
461 | |||
462 | sblog << s.str() << "\n"; |
||
463 | } |
||
464 | |||
465 | void bc_dump::dump_dw(unsigned dw_id, unsigned count) { |
||
466 | if (!bc_data) |
||
467 | return; |
||
468 | |||
469 | assert(dw_id + count <= ndw); |
||
470 | |||
471 | sblog.print_zw(dw_id, 4); |
||
472 | sblog << " "; |
||
473 | while (count--) { |
||
474 | sblog.print_zw_hex(bc_data[dw_id++], 8); |
||
475 | sblog << " "; |
||
476 | } |
||
477 | } |
||
478 | |||
479 | } // namespace r600_sb><>><>=>><>><>><>><>><>><>>><>>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>>><>><>><>><>>><>><>><>><>><>><>>><>><>><>><>>><>><>><>><>><>><>><>><>><>><>><>><>>><>><>><>><>><>><>><>><>><>><>><>>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>><>>><>>><>>><>>"|"; |