Merge branch 'mesa_7_6_branch' into mesa_7_7_branch
[mesa.git] / src / mesa / drivers / dri / r300 / compiler / r500_fragprog.c
1 /*
2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3 *
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining
7 * a copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sublicense, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial
16 * portions of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 */
27
28 #include "r500_fragprog.h"
29
30 #include <stdio.h>
31
32 #include "../r300_reg.h"
33
34 static struct rc_src_register shadow_ambient(struct radeon_compiler * c, int tmu)
35 {
36 struct rc_src_register reg = { 0, };
37
38 reg.File = RC_FILE_CONSTANT;
39 reg.Index = rc_constants_add_state(&c->Program.Constants, RC_STATE_SHADOW_AMBIENT, tmu);
40 reg.Swizzle = RC_SWIZZLE_WWWW;
41 return reg;
42 }
43
44 /**
45 * Transform TEX, TXP, TXB, and KIL instructions in the following way:
46 * - implement texture compare (shadow extensions)
47 * - extract non-native source / destination operands
48 */
49 int r500_transform_TEX(
50 struct radeon_compiler * c,
51 struct rc_instruction * inst,
52 void* data)
53 {
54 struct r300_fragment_program_compiler *compiler =
55 (struct r300_fragment_program_compiler*)data;
56
57 if (inst->U.I.Opcode != RC_OPCODE_TEX &&
58 inst->U.I.Opcode != RC_OPCODE_TXB &&
59 inst->U.I.Opcode != RC_OPCODE_TXP &&
60 inst->U.I.Opcode != RC_OPCODE_KIL)
61 return 0;
62
63 /* ARB_shadow & EXT_shadow_funcs */
64 if (inst->U.I.Opcode != RC_OPCODE_KIL &&
65 c->Program.ShadowSamplers & (1 << inst->U.I.TexSrcUnit)) {
66 rc_compare_func comparefunc = compiler->state.unit[inst->U.I.TexSrcUnit].texture_compare_func;
67
68 if (comparefunc == RC_COMPARE_FUNC_NEVER || comparefunc == RC_COMPARE_FUNC_ALWAYS) {
69 inst->U.I.Opcode = RC_OPCODE_MOV;
70
71 if (comparefunc == RC_COMPARE_FUNC_ALWAYS) {
72 inst->U.I.SrcReg[0].File = RC_FILE_NONE;
73 inst->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_1111;
74 } else {
75 inst->U.I.SrcReg[0] = shadow_ambient(c, inst->U.I.TexSrcUnit);
76 }
77
78 return 1;
79 } else {
80 rc_compare_func comparefunc = compiler->state.unit[inst->U.I.TexSrcUnit].texture_compare_func;
81 unsigned int depthmode = compiler->state.unit[inst->U.I.TexSrcUnit].depth_texture_mode;
82 struct rc_instruction * inst_rcp = rc_insert_new_instruction(c, inst);
83 struct rc_instruction * inst_mad = rc_insert_new_instruction(c, inst_rcp);
84 struct rc_instruction * inst_cmp = rc_insert_new_instruction(c, inst_mad);
85 int pass, fail;
86
87 inst_rcp->U.I.Opcode = RC_OPCODE_RCP;
88 inst_rcp->U.I.DstReg.File = RC_FILE_TEMPORARY;
89 inst_rcp->U.I.DstReg.Index = rc_find_free_temporary(c);
90 inst_rcp->U.I.DstReg.WriteMask = RC_MASK_W;
91 inst_rcp->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
92 inst_rcp->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_WWWW;
93
94 inst_cmp->U.I.DstReg = inst->U.I.DstReg;
95 inst->U.I.DstReg.File = RC_FILE_TEMPORARY;
96 inst->U.I.DstReg.Index = rc_find_free_temporary(c);
97 inst->U.I.DstReg.WriteMask = RC_MASK_XYZW;
98
99 inst_mad->U.I.Opcode = RC_OPCODE_MAD;
100 inst_mad->U.I.DstReg.File = RC_FILE_TEMPORARY;
101 inst_mad->U.I.DstReg.Index = rc_find_free_temporary(c);
102 inst_mad->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
103 inst_mad->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_ZZZZ;
104 inst_mad->U.I.SrcReg[1].File = RC_FILE_TEMPORARY;
105 inst_mad->U.I.SrcReg[1].Index = inst_rcp->U.I.DstReg.Index;
106 inst_mad->U.I.SrcReg[1].Swizzle = RC_SWIZZLE_WWWW;
107 inst_mad->U.I.SrcReg[2].File = RC_FILE_TEMPORARY;
108 inst_mad->U.I.SrcReg[2].Index = inst->U.I.DstReg.Index;
109 if (depthmode == 0) /* GL_LUMINANCE */
110 inst_mad->U.I.SrcReg[2].Swizzle = RC_MAKE_SWIZZLE(RC_SWIZZLE_X, RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_Z);
111 else if (depthmode == 2) /* GL_ALPHA */
112 inst_mad->U.I.SrcReg[2].Swizzle = RC_SWIZZLE_WWWW;
113
114 /* Recall that SrcReg[0] is tex, SrcReg[2] is r and:
115 * r < tex <=> -tex+r < 0
116 * r >= tex <=> not (-tex+r < 0 */
117 if (comparefunc == RC_COMPARE_FUNC_LESS || comparefunc == RC_COMPARE_FUNC_GEQUAL)
118 inst_mad->U.I.SrcReg[2].Negate = inst_mad->U.I.SrcReg[2].Negate ^ RC_MASK_XYZW;
119 else
120 inst_mad->U.I.SrcReg[0].Negate = inst_mad->U.I.SrcReg[0].Negate ^ RC_MASK_XYZW;
121
122 inst_cmp->U.I.Opcode = RC_OPCODE_CMP;
123 /* DstReg has been filled out above */
124 inst_cmp->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
125 inst_cmp->U.I.SrcReg[0].Index = inst_mad->U.I.DstReg.Index;
126
127 if (comparefunc == RC_COMPARE_FUNC_LESS || comparefunc == RC_COMPARE_FUNC_GREATER) {
128 pass = 1;
129 fail = 2;
130 } else {
131 pass = 2;
132 fail = 1;
133 }
134
135 inst_cmp->U.I.SrcReg[pass].File = RC_FILE_NONE;
136 inst_cmp->U.I.SrcReg[pass].Swizzle = RC_SWIZZLE_1111;
137 inst_cmp->U.I.SrcReg[fail] = shadow_ambient(c, inst->U.I.TexSrcUnit);
138 }
139 }
140
141 /* Cannot write texture to output registers */
142 if (inst->U.I.Opcode != RC_OPCODE_KIL && inst->U.I.DstReg.File != RC_FILE_TEMPORARY) {
143 struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst);
144
145 inst_mov->U.I.Opcode = RC_OPCODE_MOV;
146 inst_mov->U.I.DstReg = inst->U.I.DstReg;
147 inst_mov->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
148 inst_mov->U.I.SrcReg[0].Index = rc_find_free_temporary(c);
149
150 inst->U.I.DstReg.File = RC_FILE_TEMPORARY;
151 inst->U.I.DstReg.Index = inst_mov->U.I.SrcReg[0].Index;
152 inst->U.I.DstReg.WriteMask = RC_MASK_XYZW;
153 }
154
155 /* Cannot read texture coordinate from constants file */
156 if (inst->U.I.SrcReg[0].File != RC_FILE_TEMPORARY && inst->U.I.SrcReg[0].File != RC_FILE_INPUT) {
157 struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst->Prev);
158
159 inst_mov->U.I.Opcode = RC_OPCODE_MOV;
160 inst_mov->U.I.DstReg.File = RC_FILE_TEMPORARY;
161 inst_mov->U.I.DstReg.Index = rc_find_free_temporary(c);
162 inst_mov->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
163
164 reset_srcreg(&inst->U.I.SrcReg[0]);
165 inst->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
166 inst->U.I.SrcReg[0].Index = inst_mov->U.I.DstReg.Index;
167 }
168
169 return 1;
170 }
171
172 /**
173 * Rewrite IF instructions to use the ALU result special register.
174 */
175 int r500_transform_IF(
176 struct radeon_compiler * c,
177 struct rc_instruction * inst,
178 void* data)
179 {
180 if (inst->U.I.Opcode != RC_OPCODE_IF)
181 return 0;
182
183 struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst->Prev);
184 inst_mov->U.I.Opcode = RC_OPCODE_MOV;
185 inst_mov->U.I.DstReg.WriteMask = 0;
186 inst_mov->U.I.WriteALUResult = RC_ALURESULT_W;
187 inst_mov->U.I.ALUResultCompare = RC_COMPARE_FUNC_NOTEQUAL;
188 inst_mov->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
189 inst_mov->U.I.SrcReg[0].Swizzle = combine_swizzles4(inst_mov->U.I.SrcReg[0].Swizzle,
190 RC_SWIZZLE_UNUSED, RC_SWIZZLE_UNUSED, RC_SWIZZLE_UNUSED, RC_SWIZZLE_X);
191
192 inst->U.I.SrcReg[0].File = RC_FILE_SPECIAL;
193 inst->U.I.SrcReg[0].Index = RC_SPECIAL_ALU_RESULT;
194 inst->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_XYZW;
195 inst->U.I.SrcReg[0].Negate = 0;
196
197 return 1;
198 }
199
200 static int r500_swizzle_is_native(rc_opcode opcode, struct rc_src_register reg)
201 {
202 unsigned int relevant;
203 int i;
204
205 if (opcode == RC_OPCODE_TEX ||
206 opcode == RC_OPCODE_TXB ||
207 opcode == RC_OPCODE_TXP ||
208 opcode == RC_OPCODE_KIL) {
209 if (reg.Abs)
210 return 0;
211
212 if (opcode == RC_OPCODE_KIL && (reg.Swizzle != RC_SWIZZLE_XYZW || reg.Negate != RC_MASK_NONE))
213 return 0;
214
215 if (reg.Negate)
216 reg.Negate ^= RC_MASK_XYZW;
217
218 for(i = 0; i < 4; ++i) {
219 unsigned int swz = GET_SWZ(reg.Swizzle, i);
220 if (swz == RC_SWIZZLE_UNUSED) {
221 reg.Negate &= ~(1 << i);
222 continue;
223 }
224 if (swz >= 4)
225 return 0;
226 }
227
228 if (reg.Negate)
229 return 0;
230
231 return 1;
232 } else if (opcode == RC_OPCODE_DDX || opcode == RC_OPCODE_DDY) {
233 /* DDX/MDH and DDY/MDV explicitly ignore incoming swizzles;
234 * if it doesn't fit perfectly into a .xyzw case... */
235 if (reg.Swizzle == RC_SWIZZLE_XYZW && !reg.Abs && !reg.Negate)
236 return 1;
237
238 return 0;
239 } else {
240 /* ALU instructions support almost everything */
241 if (reg.Abs)
242 return 1;
243
244 relevant = 0;
245 for(i = 0; i < 3; ++i) {
246 unsigned int swz = GET_SWZ(reg.Swizzle, i);
247 if (swz != RC_SWIZZLE_UNUSED && swz != RC_SWIZZLE_ZERO)
248 relevant |= 1 << i;
249 }
250 if ((reg.Negate & relevant) && ((reg.Negate & relevant) != relevant))
251 return 0;
252
253 return 1;
254 }
255 }
256
257 /**
258 * Split source register access.
259 *
260 * The only thing we *cannot* do in an ALU instruction is per-component
261 * negation.
262 */
263 static void r500_swizzle_split(struct rc_src_register src, unsigned int usemask,
264 struct rc_swizzle_split * split)
265 {
266 unsigned int negatebase[2] = { 0, 0 };
267 int i;
268
269 for(i = 0; i < 4; ++i) {
270 unsigned int swz = GET_SWZ(src.Swizzle, i);
271 if (swz == RC_SWIZZLE_UNUSED || !GET_BIT(usemask, i))
272 continue;
273 negatebase[GET_BIT(src.Negate, i)] |= 1 << i;
274 }
275
276 split->NumPhases = 0;
277
278 for(i = 0; i <= 1; ++i) {
279 if (!negatebase[i])
280 continue;
281
282 split->Phase[split->NumPhases++] = negatebase[i];
283 }
284 }
285
286 struct rc_swizzle_caps r500_swizzle_caps = {
287 .IsNative = r500_swizzle_is_native,
288 .Split = r500_swizzle_split
289 };
290
291 static char *toswiz(int swiz_val) {
292 switch(swiz_val) {
293 case 0: return "R";
294 case 1: return "G";
295 case 2: return "B";
296 case 3: return "A";
297 case 4: return "0";
298 case 5: return "1/2";
299 case 6: return "1";
300 case 7: return "U";
301 }
302 return NULL;
303 }
304
305 static char *toop(int op_val)
306 {
307 char *str = NULL;
308 switch (op_val) {
309 case 0: str = "MAD"; break;
310 case 1: str = "DP3"; break;
311 case 2: str = "DP4"; break;
312 case 3: str = "D2A"; break;
313 case 4: str = "MIN"; break;
314 case 5: str = "MAX"; break;
315 case 6: str = "Reserved"; break;
316 case 7: str = "CND"; break;
317 case 8: str = "CMP"; break;
318 case 9: str = "FRC"; break;
319 case 10: str = "SOP"; break;
320 case 11: str = "MDH"; break;
321 case 12: str = "MDV"; break;
322 }
323 return str;
324 }
325
326 static char *to_alpha_op(int op_val)
327 {
328 char *str = NULL;
329 switch (op_val) {
330 case 0: str = "MAD"; break;
331 case 1: str = "DP"; break;
332 case 2: str = "MIN"; break;
333 case 3: str = "MAX"; break;
334 case 4: str = "Reserved"; break;
335 case 5: str = "CND"; break;
336 case 6: str = "CMP"; break;
337 case 7: str = "FRC"; break;
338 case 8: str = "EX2"; break;
339 case 9: str = "LN2"; break;
340 case 10: str = "RCP"; break;
341 case 11: str = "RSQ"; break;
342 case 12: str = "SIN"; break;
343 case 13: str = "COS"; break;
344 case 14: str = "MDH"; break;
345 case 15: str = "MDV"; break;
346 }
347 return str;
348 }
349
350 static char *to_mask(int val)
351 {
352 char *str = NULL;
353 switch(val) {
354 case 0: str = "NONE"; break;
355 case 1: str = "R"; break;
356 case 2: str = "G"; break;
357 case 3: str = "RG"; break;
358 case 4: str = "B"; break;
359 case 5: str = "RB"; break;
360 case 6: str = "GB"; break;
361 case 7: str = "RGB"; break;
362 case 8: str = "A"; break;
363 case 9: str = "AR"; break;
364 case 10: str = "AG"; break;
365 case 11: str = "ARG"; break;
366 case 12: str = "AB"; break;
367 case 13: str = "ARB"; break;
368 case 14: str = "AGB"; break;
369 case 15: str = "ARGB"; break;
370 }
371 return str;
372 }
373
374 static char *to_texop(int val)
375 {
376 switch(val) {
377 case 0: return "NOP";
378 case 1: return "LD";
379 case 2: return "TEXKILL";
380 case 3: return "PROJ";
381 case 4: return "LODBIAS";
382 case 5: return "LOD";
383 case 6: return "DXDY";
384 }
385 return NULL;
386 }
387
388 void r500FragmentProgramDump(struct rX00_fragment_program_code *c)
389 {
390 struct r500_fragment_program_code *code = &c->code.r500;
391 fprintf(stderr, "R500 Fragment Program:\n--------\n");
392
393 int n;
394 uint32_t inst;
395 uint32_t inst0;
396 char *str = NULL;
397
398 for (n = 0; n < code->inst_end+1; n++) {
399 inst0 = inst = code->inst[n].inst0;
400 fprintf(stderr,"%d\t0:CMN_INST 0x%08x:", n, inst);
401 switch(inst & 0x3) {
402 case R500_INST_TYPE_ALU: str = "ALU"; break;
403 case R500_INST_TYPE_OUT: str = "OUT"; break;
404 case R500_INST_TYPE_FC: str = "FC"; break;
405 case R500_INST_TYPE_TEX: str = "TEX"; break;
406 };
407 fprintf(stderr,"%s %s %s %s %s ", str,
408 inst & R500_INST_TEX_SEM_WAIT ? "TEX_WAIT" : "",
409 inst & R500_INST_LAST ? "LAST" : "",
410 inst & R500_INST_NOP ? "NOP" : "",
411 inst & R500_INST_ALU_WAIT ? "ALU WAIT" : "");
412 fprintf(stderr,"wmask: %s omask: %s\n", to_mask((inst >> 11) & 0xf),
413 to_mask((inst >> 15) & 0xf));
414
415 switch(inst0 & 0x3) {
416 case 0:
417 case 1:
418 fprintf(stderr,"\t1:RGB_ADDR 0x%08x:", code->inst[n].inst1);
419 inst = code->inst[n].inst1;
420
421 fprintf(stderr,"Addr0: %d%c, Addr1: %d%c, Addr2: %d%c, srcp:%d\n",
422 inst & 0xff, (inst & (1<<8)) ? 'c' : 't',
423 (inst >> 10) & 0xff, (inst & (1<<18)) ? 'c' : 't',
424 (inst >> 20) & 0xff, (inst & (1<<28)) ? 'c' : 't',
425 (inst >> 30));
426
427 fprintf(stderr,"\t2:ALPHA_ADDR 0x%08x:", code->inst[n].inst2);
428 inst = code->inst[n].inst2;
429 fprintf(stderr,"Addr0: %d%c, Addr1: %d%c, Addr2: %d%c, srcp:%d\n",
430 inst & 0xff, (inst & (1<<8)) ? 'c' : 't',
431 (inst >> 10) & 0xff, (inst & (1<<18)) ? 'c' : 't',
432 (inst >> 20) & 0xff, (inst & (1<<28)) ? 'c' : 't',
433 (inst >> 30));
434 fprintf(stderr,"\t3 RGB_INST: 0x%08x:", code->inst[n].inst3);
435 inst = code->inst[n].inst3;
436 fprintf(stderr,"rgb_A_src:%d %s/%s/%s %d rgb_B_src:%d %s/%s/%s %d\n",
437 (inst) & 0x3, toswiz((inst >> 2) & 0x7), toswiz((inst >> 5) & 0x7), toswiz((inst >> 8) & 0x7),
438 (inst >> 11) & 0x3,
439 (inst >> 13) & 0x3, toswiz((inst >> 15) & 0x7), toswiz((inst >> 18) & 0x7), toswiz((inst >> 21) & 0x7),
440 (inst >> 24) & 0x3);
441
442
443 fprintf(stderr,"\t4 ALPHA_INST:0x%08x:", code->inst[n].inst4);
444 inst = code->inst[n].inst4;
445 fprintf(stderr,"%s dest:%d%s alp_A_src:%d %s %d alp_B_src:%d %s %d w:%d\n", to_alpha_op(inst & 0xf),
446 (inst >> 4) & 0x7f, inst & (1<<11) ? "(rel)":"",
447 (inst >> 12) & 0x3, toswiz((inst >> 14) & 0x7), (inst >> 17) & 0x3,
448 (inst >> 19) & 0x3, toswiz((inst >> 21) & 0x7), (inst >> 24) & 0x3,
449 (inst >> 31) & 0x1);
450
451 fprintf(stderr,"\t5 RGBA_INST: 0x%08x:", code->inst[n].inst5);
452 inst = code->inst[n].inst5;
453 fprintf(stderr,"%s dest:%d%s rgb_C_src:%d %s/%s/%s %d alp_C_src:%d %s %d\n", toop(inst & 0xf),
454 (inst >> 4) & 0x7f, inst & (1<<11) ? "(rel)":"",
455 (inst >> 12) & 0x3, toswiz((inst >> 14) & 0x7), toswiz((inst >> 17) & 0x7), toswiz((inst >> 20) & 0x7),
456 (inst >> 23) & 0x3,
457 (inst >> 25) & 0x3, toswiz((inst >> 27) & 0x7), (inst >> 30) & 0x3);
458 break;
459 case 2:
460 break;
461 case 3:
462 inst = code->inst[n].inst1;
463 fprintf(stderr,"\t1:TEX_INST: 0x%08x: id: %d op:%s, %s, %s %s\n", inst, (inst >> 16) & 0xf,
464 to_texop((inst >> 22) & 0x7), (inst & (1<<25)) ? "ACQ" : "",
465 (inst & (1<<26)) ? "IGNUNC" : "", (inst & (1<<27)) ? "UNSCALED" : "SCALED");
466 inst = code->inst[n].inst2;
467 fprintf(stderr,"\t2:TEX_ADDR: 0x%08x: src: %d%s %s/%s/%s/%s dst: %d%s %s/%s/%s/%s\n", inst,
468 inst & 127, inst & (1<<7) ? "(rel)" : "",
469 toswiz((inst >> 8) & 0x3), toswiz((inst >> 10) & 0x3),
470 toswiz((inst >> 12) & 0x3), toswiz((inst >> 14) & 0x3),
471 (inst >> 16) & 127, inst & (1<<23) ? "(rel)" : "",
472 toswiz((inst >> 24) & 0x3), toswiz((inst >> 26) & 0x3),
473 toswiz((inst >> 28) & 0x3), toswiz((inst >> 30) & 0x3));
474
475 fprintf(stderr,"\t3:TEX_DXDY: 0x%08x\n", code->inst[n].inst3);
476 break;
477 }
478 fprintf(stderr,"\n");
479 }
480
481 }