nvc0: do upload-time fixups for interpolation parameters
[mesa.git] / src / gallium / drivers / nouveau / codegen / nv50_ir_target.cpp
1 /*
2 * Copyright 2011 Christoph Bumiller
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 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
21 */
22
23 #include "codegen/nv50_ir.h"
24 #include "codegen/nv50_ir_target.h"
25
26 namespace nv50_ir {
27
28 const uint8_t Target::operationSrcNr[] =
29 {
30 0, 0, // NOP, PHI
31 0, 0, 0, 0, // UNION, SPLIT, MERGE, CONSTRAINT
32 1, 1, 2, // MOV, LOAD, STORE
33 2, 2, 2, 2, 2, 3, 3, 3, // ADD, SUB, MUL, DIV, MOD, MAD, FMA, SAD
34 1, 1, 1, // ABS, NEG, NOT
35 2, 2, 2, 2, 2, // AND, OR, XOR, SHL, SHR
36 2, 2, 1, // MAX, MIN, SAT
37 1, 1, 1, 1, // CEIL, FLOOR, TRUNC, CVT
38 3, 3, 3, 2, 3, 3, // SET_AND,OR,XOR, SET, SELP, SLCT
39 1, 1, 1, 1, 1, 1, // RCP, RSQ, LG2, SIN, COS, EX2
40 1, 1, 1, 1, 1, 2, // EXP, LOG, PRESIN, PREEX2, SQRT, POW
41 0, 0, 0, 0, 0, // BRA, CALL, RET, CONT, BREAK,
42 0, 0, 0, // PRERET,CONT,BREAK
43 0, 0, 0, 0, 0, 0, // BRKPT, JOINAT, JOIN, DISCARD, EXIT, MEMBAR
44 1, 1, 1, 2, 1, 2, // VFETCH, PFETCH, AFETCH, EXPORT, LINTERP, PINTERP
45 1, 1, // EMIT, RESTART
46 1, 1, 1, // TEX, TXB, TXL,
47 1, 1, 1, 1, 1, 1, 2, // TXF, TXQ, TXD, TXG, TXLQ, TEXCSAA, TEXPREP
48 1, 1, 2, 2, 2, 2, 2, // SULDB, SULDP, SUSTB, SUSTP, SUREDB, SUREDP, SULEA
49 3, 3, 3, 3, // SUBFM, SUCLAMP, SUEAU, MADSP
50 0, // TEXBAR
51 1, 1, // DFDX, DFDY
52 1, 2, 1, 2, 0, 0, // RDSV, WRSV, PIXLD, QUADOP, QUADON, QUADPOP
53 2, 3, 2, 1, 3, // POPCNT, INSBF, EXTBF, BFIND, PERMT
54 2, 2, // ATOM, BAR
55 2, 2, 2, 2, 3, 2, // VADD, VAVG, VMIN, VMAX, VSAD, VSET,
56 2, 2, 2, 1, // VSHR, VSHL, VSEL, CCTL
57 3, // SHFL
58 0
59 };
60
61 const OpClass Target::operationClass[] =
62 {
63 // NOP; PHI; UNION, SPLIT, MERGE, CONSTRAINT
64 OPCLASS_OTHER,
65 OPCLASS_PSEUDO,
66 OPCLASS_PSEUDO, OPCLASS_PSEUDO, OPCLASS_PSEUDO, OPCLASS_PSEUDO,
67 // MOV; LOAD; STORE
68 OPCLASS_MOVE,
69 OPCLASS_LOAD,
70 OPCLASS_STORE,
71 // ADD, SUB, MUL; DIV, MOD; MAD, FMA, SAD
72 OPCLASS_ARITH, OPCLASS_ARITH, OPCLASS_ARITH,
73 OPCLASS_ARITH, OPCLASS_ARITH,
74 OPCLASS_ARITH, OPCLASS_ARITH, OPCLASS_ARITH,
75 // ABS, NEG; NOT, AND, OR, XOR; SHL, SHR
76 OPCLASS_CONVERT, OPCLASS_CONVERT,
77 OPCLASS_LOGIC, OPCLASS_LOGIC, OPCLASS_LOGIC, OPCLASS_LOGIC,
78 OPCLASS_SHIFT, OPCLASS_SHIFT,
79 // MAX, MIN
80 OPCLASS_COMPARE, OPCLASS_COMPARE,
81 // SAT, CEIL, FLOOR, TRUNC; CVT
82 OPCLASS_CONVERT, OPCLASS_CONVERT, OPCLASS_CONVERT, OPCLASS_CONVERT,
83 OPCLASS_CONVERT,
84 // SET(AND,OR,XOR); SELP, SLCT
85 OPCLASS_COMPARE, OPCLASS_COMPARE, OPCLASS_COMPARE, OPCLASS_COMPARE,
86 OPCLASS_COMPARE, OPCLASS_COMPARE,
87 // RCP, RSQ, LG2, SIN, COS; EX2, EXP, LOG, PRESIN, PREEX2; SQRT, POW
88 OPCLASS_SFU, OPCLASS_SFU, OPCLASS_SFU, OPCLASS_SFU, OPCLASS_SFU,
89 OPCLASS_SFU, OPCLASS_SFU, OPCLASS_SFU, OPCLASS_SFU, OPCLASS_SFU,
90 OPCLASS_SFU, OPCLASS_SFU,
91 // BRA, CALL, RET; CONT, BREAK, PRE(RET,CONT,BREAK); BRKPT, JOINAT, JOIN
92 OPCLASS_FLOW, OPCLASS_FLOW, OPCLASS_FLOW,
93 OPCLASS_FLOW, OPCLASS_FLOW, OPCLASS_FLOW, OPCLASS_FLOW, OPCLASS_FLOW,
94 OPCLASS_FLOW, OPCLASS_FLOW, OPCLASS_FLOW,
95 // DISCARD, EXIT
96 OPCLASS_FLOW, OPCLASS_FLOW,
97 // MEMBAR
98 OPCLASS_CONTROL,
99 // VFETCH, PFETCH, AFETCH, EXPORT
100 OPCLASS_LOAD, OPCLASS_OTHER, OPCLASS_OTHER, OPCLASS_STORE,
101 // LINTERP, PINTERP
102 OPCLASS_SFU, OPCLASS_SFU,
103 // EMIT, RESTART
104 OPCLASS_CONTROL, OPCLASS_CONTROL,
105 // TEX, TXB, TXL, TXF; TXQ, TXD, TXG, TXLQ; TEXCSAA, TEXPREP
106 OPCLASS_TEXTURE, OPCLASS_TEXTURE, OPCLASS_TEXTURE, OPCLASS_TEXTURE,
107 OPCLASS_TEXTURE, OPCLASS_TEXTURE, OPCLASS_TEXTURE, OPCLASS_TEXTURE,
108 OPCLASS_TEXTURE, OPCLASS_TEXTURE,
109 // SULDB, SULDP, SUSTB, SUSTP; SUREDB, SUREDP, SULEA
110 OPCLASS_SURFACE, OPCLASS_SURFACE, OPCLASS_ATOMIC, OPCLASS_SURFACE,
111 OPCLASS_SURFACE, OPCLASS_SURFACE, OPCLASS_SURFACE,
112 // SUBFM, SUCLAMP, SUEAU, MADSP
113 OPCLASS_OTHER, OPCLASS_OTHER, OPCLASS_OTHER, OPCLASS_ARITH,
114 // TEXBAR
115 OPCLASS_OTHER,
116 // DFDX, DFDY, RDSV, WRSV; PIXLD, QUADOP, QUADON, QUADPOP
117 OPCLASS_OTHER, OPCLASS_OTHER, OPCLASS_OTHER, OPCLASS_OTHER,
118 OPCLASS_OTHER, OPCLASS_OTHER, OPCLASS_CONTROL, OPCLASS_CONTROL,
119 // POPCNT, INSBF, EXTBF, BFIND; PERMT
120 OPCLASS_BITFIELD, OPCLASS_BITFIELD, OPCLASS_BITFIELD, OPCLASS_BITFIELD,
121 OPCLASS_BITFIELD,
122 // ATOM, BAR
123 OPCLASS_ATOMIC, OPCLASS_CONTROL,
124 // VADD, VAVG, VMIN, VMAX
125 OPCLASS_VECTOR, OPCLASS_VECTOR, OPCLASS_VECTOR, OPCLASS_VECTOR,
126 // VSAD, VSET, VSHR, VSHL
127 OPCLASS_VECTOR, OPCLASS_VECTOR, OPCLASS_VECTOR, OPCLASS_VECTOR,
128 // VSEL, CCTL
129 OPCLASS_VECTOR, OPCLASS_CONTROL,
130 // SHFL
131 OPCLASS_OTHER,
132 OPCLASS_PSEUDO // LAST
133 };
134
135
136 extern Target *getTargetGM107(unsigned int chipset);
137 extern Target *getTargetNVC0(unsigned int chipset);
138 extern Target *getTargetNV50(unsigned int chipset);
139
140 Target *Target::create(unsigned int chipset)
141 {
142 STATIC_ASSERT(Elements(operationSrcNr) == OP_LAST + 1);
143 STATIC_ASSERT(Elements(operationClass) == OP_LAST + 1);
144 switch (chipset & ~0xf) {
145 case 0x110:
146 return getTargetGM107(chipset);
147 case 0xc0:
148 case 0xd0:
149 case 0xe0:
150 case 0xf0:
151 case 0x100:
152 return getTargetNVC0(chipset);
153 case 0x50:
154 case 0x80:
155 case 0x90:
156 case 0xa0:
157 return getTargetNV50(chipset);
158 default:
159 ERROR("unsupported target: NV%x\n", chipset);
160 return 0;
161 }
162 }
163
164 void Target::destroy(Target *targ)
165 {
166 delete targ;
167 }
168
169 CodeEmitter::CodeEmitter(const Target *target) : targ(target), interpInfo(NULL)
170 {
171 }
172
173 void
174 CodeEmitter::setCodeLocation(void *ptr, uint32_t size)
175 {
176 code = reinterpret_cast<uint32_t *>(ptr);
177 codeSize = 0;
178 codeSizeLimit = size;
179 }
180
181 void
182 CodeEmitter::printBinary() const
183 {
184 uint32_t *bin = code - codeSize / 4;
185 INFO("program binary (%u bytes)", codeSize);
186 for (unsigned int pos = 0; pos < codeSize / 4; ++pos) {
187 if ((pos % 8) == 0)
188 INFO("\n");
189 INFO("%08x ", bin[pos]);
190 }
191 INFO("\n");
192 }
193
194 static inline uint32_t sizeToBundlesNVE4(uint32_t size)
195 {
196 return (size + 55) / 56;
197 }
198
199 void
200 CodeEmitter::prepareEmission(Program *prog)
201 {
202 for (ArrayList::Iterator fi = prog->allFuncs.iterator();
203 !fi.end(); fi.next()) {
204 Function *func = reinterpret_cast<Function *>(fi.get());
205 func->binPos = prog->binSize;
206 prepareEmission(func);
207
208 // adjust sizes & positions for schedulding info:
209 if (prog->getTarget()->hasSWSched) {
210 uint32_t adjPos = func->binPos;
211 BasicBlock *bb = NULL;
212 for (int i = 0; i < func->bbCount; ++i) {
213 bb = func->bbArray[i];
214 int32_t adjSize = bb->binSize;
215 if (adjPos % 64) {
216 adjSize -= 64 - adjPos % 64;
217 if (adjSize < 0)
218 adjSize = 0;
219 }
220 adjSize = bb->binSize + sizeToBundlesNVE4(adjSize) * 8;
221 bb->binPos = adjPos;
222 bb->binSize = adjSize;
223 adjPos += adjSize;
224 }
225 if (bb)
226 func->binSize = adjPos - func->binPos;
227 }
228
229 prog->binSize += func->binSize;
230 }
231 }
232
233 void
234 CodeEmitter::prepareEmission(Function *func)
235 {
236 func->bbCount = 0;
237 func->bbArray = new BasicBlock * [func->cfg.getSize()];
238
239 BasicBlock::get(func->cfg.getRoot())->binPos = func->binPos;
240
241 for (IteratorRef it = func->cfg.iteratorCFG(); !it->end(); it->next())
242 prepareEmission(BasicBlock::get(*it));
243 }
244
245 void
246 CodeEmitter::prepareEmission(BasicBlock *bb)
247 {
248 Instruction *i, *next;
249 Function *func = bb->getFunction();
250 int j;
251 unsigned int nShort;
252
253 for (j = func->bbCount - 1; j >= 0 && !func->bbArray[j]->binSize; --j);
254
255 for (; j >= 0; --j) {
256 BasicBlock *in = func->bbArray[j];
257 Instruction *exit = in->getExit();
258
259 if (exit && exit->op == OP_BRA && exit->asFlow()->target.bb == bb) {
260 in->binSize -= 8;
261 func->binSize -= 8;
262
263 for (++j; j < func->bbCount; ++j)
264 func->bbArray[j]->binPos -= 8;
265
266 in->remove(exit);
267 }
268 bb->binPos = in->binPos + in->binSize;
269 if (in->binSize) // no more no-op branches to bb
270 break;
271 }
272 func->bbArray[func->bbCount++] = bb;
273
274 if (!bb->getExit())
275 return;
276
277 // determine encoding size, try to group short instructions
278 nShort = 0;
279 for (i = bb->getEntry(); i; i = next) {
280 next = i->next;
281
282 if (i->op == OP_MEMBAR && !targ->isOpSupported(OP_MEMBAR, TYPE_NONE)) {
283 bb->remove(i);
284 continue;
285 }
286
287 i->encSize = getMinEncodingSize(i);
288 if (next && i->encSize < 8)
289 ++nShort;
290 else
291 if ((nShort & 1) && next && getMinEncodingSize(next) == 4) {
292 if (i->isCommutationLegal(i->next)) {
293 bb->permuteAdjacent(i, next);
294 next->encSize = 4;
295 next = i;
296 i = i->prev;
297 ++nShort;
298 } else
299 if (i->isCommutationLegal(i->prev) && next->next) {
300 bb->permuteAdjacent(i->prev, i);
301 next->encSize = 4;
302 next = next->next;
303 bb->binSize += 4;
304 ++nShort;
305 } else {
306 i->encSize = 8;
307 i->prev->encSize = 8;
308 bb->binSize += 4;
309 nShort = 0;
310 }
311 } else {
312 i->encSize = 8;
313 if (nShort & 1) {
314 i->prev->encSize = 8;
315 bb->binSize += 4;
316 }
317 nShort = 0;
318 }
319 bb->binSize += i->encSize;
320 }
321
322 if (bb->getExit()->encSize == 4) {
323 assert(nShort);
324 bb->getExit()->encSize = 8;
325 bb->binSize += 4;
326
327 if ((bb->getExit()->prev->encSize == 4) && !(nShort & 1)) {
328 bb->binSize += 8;
329 bb->getExit()->prev->encSize = 8;
330 }
331 }
332 assert(!bb->getEntry() || (bb->getExit() && bb->getExit()->encSize == 8));
333
334 func->binSize += bb->binSize;
335 }
336
337 void
338 Program::emitSymbolTable(struct nv50_ir_prog_info *info)
339 {
340 unsigned int n = 0, nMax = allFuncs.getSize();
341
342 info->bin.syms =
343 (struct nv50_ir_prog_symbol *)MALLOC(nMax * sizeof(*info->bin.syms));
344
345 for (ArrayList::Iterator fi = allFuncs.iterator();
346 !fi.end();
347 fi.next(), ++n) {
348 Function *f = (Function *)fi.get();
349 assert(n < nMax);
350
351 info->bin.syms[n].label = f->getLabel();
352 info->bin.syms[n].offset = f->binPos;
353 }
354
355 info->bin.numSyms = n;
356 }
357
358 bool
359 Program::emitBinary(struct nv50_ir_prog_info *info)
360 {
361 CodeEmitter *emit = target->getCodeEmitter(progType);
362
363 emit->prepareEmission(this);
364
365 if (dbgFlags & NV50_IR_DEBUG_BASIC)
366 this->print();
367
368 if (!binSize) {
369 code = NULL;
370 return false;
371 }
372 code = reinterpret_cast<uint32_t *>(MALLOC(binSize));
373 if (!code)
374 return false;
375 emit->setCodeLocation(code, binSize);
376
377 for (ArrayList::Iterator fi = allFuncs.iterator(); !fi.end(); fi.next()) {
378 Function *fn = reinterpret_cast<Function *>(fi.get());
379
380 assert(emit->getCodeSize() == fn->binPos);
381
382 for (int b = 0; b < fn->bbCount; ++b) {
383 for (Instruction *i = fn->bbArray[b]->getEntry(); i; i = i->next) {
384 emit->emitInstruction(i);
385 if (i->sType == TYPE_F64 || i->dType == TYPE_F64)
386 info->io.fp64 = true;
387 }
388 }
389 }
390 info->bin.relocData = emit->getRelocInfo();
391 info->bin.interpData = emit->getInterpInfo();
392
393 emitSymbolTable(info);
394
395 // the nvc0 driver will print the binary iself together with the header
396 if ((dbgFlags & NV50_IR_DEBUG_BASIC) && getTarget()->getChipset() < 0xc0)
397 emit->printBinary();
398
399 delete emit;
400 return true;
401 }
402
403 #define RELOC_ALLOC_INCREMENT 8
404
405 bool
406 CodeEmitter::addReloc(RelocEntry::Type ty, int w, uint32_t data, uint32_t m,
407 int s)
408 {
409 unsigned int n = relocInfo ? relocInfo->count : 0;
410
411 if (!(n % RELOC_ALLOC_INCREMENT)) {
412 size_t size = sizeof(RelocInfo) + n * sizeof(RelocEntry);
413 relocInfo = reinterpret_cast<RelocInfo *>(
414 REALLOC(relocInfo, n ? size : 0,
415 size + RELOC_ALLOC_INCREMENT * sizeof(RelocEntry)));
416 if (!relocInfo)
417 return false;
418 if (n == 0)
419 memset(relocInfo, 0, sizeof(RelocInfo));
420 }
421 ++relocInfo->count;
422
423 relocInfo->entry[n].data = data;
424 relocInfo->entry[n].mask = m;
425 relocInfo->entry[n].offset = codeSize + w * 4;
426 relocInfo->entry[n].bitPos = s;
427 relocInfo->entry[n].type = ty;
428
429 return true;
430 }
431
432 bool
433 CodeEmitter::addInterp(int ipa, int reg, InterpApply apply)
434 {
435 unsigned int n = interpInfo ? interpInfo->count : 0;
436
437 if (!(n % RELOC_ALLOC_INCREMENT)) {
438 size_t size = sizeof(InterpInfo) + n * sizeof(InterpEntry);
439 interpInfo = reinterpret_cast<InterpInfo *>(
440 REALLOC(interpInfo, n ? size : 0,
441 size + RELOC_ALLOC_INCREMENT * sizeof(InterpEntry)));
442 if (!interpInfo)
443 return false;
444 if (n == 0)
445 memset(interpInfo, 0, sizeof(InterpInfo));
446 }
447 ++interpInfo->count;
448
449 interpInfo->entry[n] = InterpEntry(ipa, reg, codeSize >> 2);
450 interpInfo->apply = apply;
451
452 return true;
453 }
454
455 void
456 RelocEntry::apply(uint32_t *binary, const RelocInfo *info) const
457 {
458 uint32_t value = 0;
459
460 switch (type) {
461 case TYPE_CODE: value = info->codePos; break;
462 case TYPE_BUILTIN: value = info->libPos; break;
463 case TYPE_DATA: value = info->dataPos; break;
464 default:
465 assert(0);
466 break;
467 }
468 value += data;
469 value = (bitPos < 0) ? (value >> -bitPos) : (value << bitPos);
470
471 binary[offset / 4] &= ~mask;
472 binary[offset / 4] |= value & mask;
473 }
474
475 } // namespace nv50_ir
476
477
478 #include "codegen/nv50_ir_driver.h"
479
480 extern "C" {
481
482 void
483 nv50_ir_relocate_code(void *relocData, uint32_t *code,
484 uint32_t codePos,
485 uint32_t libPos,
486 uint32_t dataPos)
487 {
488 nv50_ir::RelocInfo *info = reinterpret_cast<nv50_ir::RelocInfo *>(relocData);
489
490 info->codePos = codePos;
491 info->libPos = libPos;
492 info->dataPos = dataPos;
493
494 for (unsigned int i = 0; i < info->count; ++i)
495 info->entry[i].apply(code, info);
496 }
497
498 void
499 nv50_ir_change_interp(void *interpData, uint32_t *code,
500 bool force_persample_interp, bool flatshade)
501 {
502 nv50_ir::InterpInfo *info = reinterpret_cast<nv50_ir::InterpInfo *>(
503 interpData);
504
505 // force_persample_interp: all non-flat -> per-sample
506 // flatshade: all color -> flat
507 for (unsigned i = 0; i < info->count; ++i)
508 info->apply(&info->entry[i], code, force_persample_interp, flatshade);
509 }
510
511 void
512 nv50_ir_get_target_library(uint32_t chipset,
513 const uint32_t **code, uint32_t *size)
514 {
515 nv50_ir::Target *targ = nv50_ir::Target::create(chipset);
516 targ->getBuiltinCode(code, size);
517 nv50_ir::Target::destroy(targ);
518 }
519
520 }