projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
r5xx: Unbreak MAX and MIN.
[mesa.git] / src / mesa / drivers / dri / r300 / r500_fragprog.c
1 /*
2 * Copyright (C) 2005 Ben Skeggs.
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 /**
29 * \file
30 *
31 * \author Ben Skeggs <darktama@iinet.net.au>
32 *
33 * \author Jerome Glisse <j.glisse@gmail.com>
34 *
35 * \author Corbin Simpson <MostAwesomeDude@gmail.com>
36 *
37 * \todo Depth write, WPOS/FOGC inputs
38 *
39 * \todo FogOption
40 *
41 * \todo Verify results of opcodes for accuracy, I've only checked them in
42 * specific cases.
43 */
44
45 #include "glheader.h"
46 #include "macros.h"
47 #include "enums.h"
48 #include "shader/prog_instruction.h"
49 #include "shader/prog_parameter.h"
50 #include "shader/prog_print.h"
51
52 #include "r300_context.h"
53 #include "r500_fragprog.h"
54 #include "r300_reg.h"
55 #include "r300_state.h"
56
57 /*
58 * Useful macros and values
59 */
60 #define ERROR(fmt, args...) do { \
61 fprintf(stderr, "%s::%s(): " fmt "\n", \
62 __FILE__, __FUNCTION__, ##args); \
63 fp->error = GL_TRUE; \
64 } while(0)
65
66 #define COMPILE_STATE struct r300_pfs_compile_state *cs = fp->cs
67
68 #define R500_US_NUM_TEMP_REGS 128
69 #define R500_US_NUM_CONST_REGS 256
70
71 /* "Register" flags */
72 #define REG_CONSTANT (1 << 8)
73 #define REG_SRC_REL (1 << 9)
74 #define REG_DEST_REL (1 << 7)
75
76 /* Swizzle tools */
77 #define R500_SWIZZLE_ZERO 4
78 #define R500_SWIZZLE_HALF 5
79 #define R500_SWIZZLE_ONE 6
80 #define R500_SWIZ_RGB_ZERO ((4 << 0) | (4 << 3) | (4 << 6))
81 #define R500_SWIZ_RGB_ONE ((6 << 0) | (6 << 3) | (6 << 6))
82 #define R500_SWIZ_RGB_RGB ((0 << 0) | (1 << 3) | (2 << 6))
83 /* Swizzles for inst2 */
84 #define MAKE_SWIZ_TEX_STRQ(x) (x << 8)
85 #define MAKE_SWIZ_TEX_RGBA(x) (x << 24)
86 /* Swizzles for inst3 */
87 #define MAKE_SWIZ_RGB_A(x) (x << 2)
88 #define MAKE_SWIZ_RGB_B(x) (x << 15)
89 /* Swizzles for inst4 */
90 #define MAKE_SWIZ_ALPHA_A(x) (x << 14)
91 #define MAKE_SWIZ_ALPHA_B(x) (x << 21)
92 /* Swizzle for inst5 */
93 #define MAKE_SWIZ_RGBA_C(x) (x << 14)
94 #define MAKE_SWIZ_ALPHA_C(x) (x << 27)
95
96 static inline GLuint make_rgb_swizzle(struct prog_src_register src) {
97 GLuint swiz = 0x0;
98 GLuint temp;
99 /* This could be optimized, but it should be plenty fast already. */
100 int i;
101 for (i = 0; i < 3; i++) {
102 temp = (src.Swizzle >> i*3) & 0x7;
103 /* Fix SWIZZLE_ONE */
104 if (temp == 5) temp++;
105 swiz += temp << i*3;
106 }
107 return swiz;
108 }
109
110 static inline GLuint make_alpha_swizzle(struct prog_src_register src) {
111 GLuint swiz = (src.Swizzle >> 12) & 0x7;
112 if (swiz == 5) swiz++;
113 return swiz;
114 }
115
116 static inline GLuint make_strq_swizzle(struct prog_src_register src) {
117 GLuint swiz = 0x0;
118 GLuint temp = src.Swizzle;
119 int i;
120 for (i = 0; i < 4; i++) {
121 swiz += (temp & 0x3) << i*2;
122 temp >>= 3;
123 }
124 return swiz;
125 }
126
127 static int get_temp(struct r500_fragment_program *fp, int slot) {
128
129 COMPILE_STATE;
130
131 int r = slot;
132
133 while (cs->inputs[r].refcount != 0) {
134 /* Crap, taken. */
135 r++;
136 }
137
138 fp->temp_reg_offset = r - slot;
139
140 if (r >= R500_US_NUM_TEMP_REGS) {
141 ERROR("Out of hardware temps!\n");
142 return 0;
143 }
144
145 if (r > fp->max_temp_idx)
146 fp->max_temp_idx = r;
147
148 return r;
149 }
150
151 /* Borrowed verbatim from r300_fragprog since it hasn't changed. */
152 static GLuint emit_const4fv(struct r500_fragment_program *fp,
153 const GLfloat * cp)
154 {
155 GLuint reg = 0x0;
156 int index;
157
158 for (index = 0; index < fp->const_nr; ++index) {
159 if (fp->constant[index] == cp)
160 break;
161 }
162
163 if (index >= fp->const_nr) {
164 if (index >= R500_US_NUM_CONST_REGS) {
165 ERROR("Out of hw constants!\n");
166 return reg;
167 }
168
169 fp->const_nr++;
170 fp->constant[index] = cp;
171 }
172
173 reg = index | REG_CONSTANT;
174 return reg;
175 }
176
177 static GLuint make_src(struct r500_fragment_program *fp, struct prog_src_register src) {
178 COMPILE_STATE;
179 GLuint reg;
180 switch (src.File) {
181 case PROGRAM_TEMPORARY:
182 reg = src.Index + fp->temp_reg_offset;
183 break;
184 case PROGRAM_INPUT:
185 reg = cs->inputs[src.Index].reg;
186 break;
187 case PROGRAM_STATE_VAR:
188 case PROGRAM_NAMED_PARAM:
189 case PROGRAM_CONSTANT:
190 reg = emit_const4fv(fp, fp->mesa_program.Base.Parameters->
191 ParameterValues[src.Index]);
192 break;
193 default:
194 ERROR("Can't handle src.File %x\n", src.File);
195 reg = 0x0;
196 break;
197 }
198 return reg;
199 }
200
201 static GLuint make_dest(struct r500_fragment_program *fp, struct prog_dst_register dest) {
202 GLuint reg;
203 switch (dest.File) {
204 case PROGRAM_TEMPORARY:
205 reg = dest.Index + fp->temp_reg_offset;
206 break;
207 case PROGRAM_OUTPUT:
208 /* Eventually we may need to handle multiple
209 * rendering targets... */
210 reg = dest.Index;
211 break;
212 default:
213 ERROR("Can't handle dest.File %x\n", dest.File);
214 reg = 0x0;
215 break;
216 }
217 return reg;
218 }
219
220 static void emit_tex(struct r500_fragment_program *fp,
221 struct prog_instruction *fpi, int opcode, int dest, int counter)
222 {
223 int hwsrc, hwdest;
224 GLuint mask;
225
226 mask = fpi->DstReg.WriteMask << 11;
227 hwsrc = make_src(fp, fpi->SrcReg[0]);
228
229 fp->inst[counter].inst0 = R500_INST_TYPE_TEX | mask
230 | R500_INST_TEX_SEM_WAIT;
231
232 fp->inst[counter].inst1 = R500_TEX_ID(fpi->TexSrcUnit)
233 | R500_TEX_SEM_ACQUIRE | R500_TEX_IGNORE_UNCOVERED;
234
235 if (fpi->TexSrcTarget == TEXTURE_RECT_INDEX)
236 fp->inst[counter].inst1 |= R500_TEX_UNSCALED;
237
238 switch (opcode) {
239 case OPCODE_TEX:
240 fp->inst[counter].inst1 |= R500_TEX_INST_LD;
241 break;
242 case OPCODE_TXB:
243 fp->inst[counter].inst1 |= R500_TEX_INST_LODBIAS;
244 break;
245 case OPCODE_TXP:
246 fp->inst[counter].inst1 |= R500_TEX_INST_PROJ;
247 break;
248 default:
249 ERROR("emit_tex can't handle opcode %x\n", opcode);
250 }
251
252 fp->inst[counter].inst2 = R500_TEX_SRC_ADDR(hwsrc)
253 /* | MAKE_SWIZ_TEX_STRQ(make_strq_swizzle(fpi->SrcReg[0])) */
254 | R500_TEX_SRC_S_SWIZ_R | R500_TEX_SRC_T_SWIZ_G
255 | R500_TEX_SRC_R_SWIZ_B | R500_TEX_SRC_Q_SWIZ_A
256 | R500_TEX_DST_ADDR(dest)
257 | R500_TEX_DST_R_SWIZ_R | R500_TEX_DST_G_SWIZ_G
258 | R500_TEX_DST_B_SWIZ_B | R500_TEX_DST_A_SWIZ_A;
259
260
261
262 fp->inst[counter].inst3 = 0x0;
263 fp->inst[counter].inst4 = 0x0;
264 fp->inst[counter].inst5 = 0x0;
265 }
266
267 static void dumb_shader(struct r500_fragment_program *fp)
268 {
269 fp->inst[0].inst0 = R500_INST_TYPE_TEX
270 | R500_INST_TEX_SEM_WAIT
271 | R500_INST_RGB_WMASK_R
272 | R500_INST_RGB_WMASK_G
273 | R500_INST_RGB_WMASK_B
274 | R500_INST_ALPHA_WMASK
275 | R500_INST_RGB_CLAMP
276 | R500_INST_ALPHA_CLAMP;
277 fp->inst[0].inst1 = R500_TEX_ID(0)
278 | R500_TEX_INST_LD
279 | R500_TEX_SEM_ACQUIRE
280 | R500_TEX_IGNORE_UNCOVERED;
281 fp->inst[0].inst2 = R500_TEX_SRC_ADDR(0)
282 | R500_TEX_SRC_S_SWIZ_R
283 | R500_TEX_SRC_T_SWIZ_G
284 | R500_TEX_DST_ADDR(0)
285 | R500_TEX_DST_R_SWIZ_R
286 | R500_TEX_DST_G_SWIZ_G
287 | R500_TEX_DST_B_SWIZ_B
288 | R500_TEX_DST_A_SWIZ_A;
289 fp->inst[0].inst3 = R500_DX_ADDR(0)
290 | R500_DX_S_SWIZ_R
291 | R500_DX_T_SWIZ_R
292 | R500_DX_R_SWIZ_R
293 | R500_DX_Q_SWIZ_R
294 | R500_DY_ADDR(0)
295 | R500_DY_S_SWIZ_R
296 | R500_DY_T_SWIZ_R
297 | R500_DY_R_SWIZ_R
298 | R500_DY_Q_SWIZ_R;
299 fp->inst[0].inst4 = 0x0;
300 fp->inst[0].inst5 = 0x0;
301
302 fp->inst[1].inst0 = R500_INST_TYPE_OUT |
303 R500_INST_TEX_SEM_WAIT |
304 R500_INST_LAST |
305 R500_INST_RGB_OMASK_R |
306 R500_INST_RGB_OMASK_G |
307 R500_INST_RGB_OMASK_B |
308 R500_INST_ALPHA_OMASK;
309 fp->inst[1].inst1 = R500_RGB_ADDR0(0) |
310 R500_RGB_ADDR1(0) |
311 R500_RGB_ADDR1_CONST |
312 R500_RGB_ADDR2(0) |
313 R500_RGB_ADDR2_CONST |
314 R500_RGB_SRCP_OP_1_MINUS_2RGB0;
315 fp->inst[1].inst2 = R500_ALPHA_ADDR0(0) |
316 R500_ALPHA_ADDR1(0) |
317 R500_ALPHA_ADDR1_CONST |
318 R500_ALPHA_ADDR2(0) |
319 R500_ALPHA_ADDR2_CONST |
320 R500_ALPHA_SRCP_OP_1_MINUS_2A0;
321 fp->inst[1].inst3 = R500_ALU_RGB_SEL_A_SRC0 |
322 R500_ALU_RGB_R_SWIZ_A_R |
323 R500_ALU_RGB_G_SWIZ_A_G |
324 R500_ALU_RGB_B_SWIZ_A_B |
325 R500_ALU_RGB_SEL_B_SRC0 |
326 R500_ALU_RGB_R_SWIZ_B_1 |
327 R500_ALU_RGB_B_SWIZ_B_1 |
328 R500_ALU_RGB_G_SWIZ_B_1;
329 fp->inst[1].inst4 = R500_ALPHA_OP_MAD |
330 R500_ALPHA_SWIZ_A_A |
331 R500_ALPHA_SWIZ_B_1;
332 fp->inst[1].inst5 = R500_ALU_RGBA_OP_MAD |
333 R500_ALU_RGBA_R_SWIZ_0 |
334 R500_ALU_RGBA_G_SWIZ_0 |
335 R500_ALU_RGBA_B_SWIZ_0 |
336 R500_ALU_RGBA_A_SWIZ_0;
337
338 fp->cs->nrslots = 2;
339 fp->translated = GL_TRUE;
340 }
341
342 /* static void emit_alu(struct r500_fragment_program *fp) {
343 * } */
344
345 static void emit_mov(struct r500_fragment_program *fp, int counter, struct prog_src_register src, GLuint dest) {
346 /* The r3xx shader uses MAD to implement MOV. We are using CMP, since
347 * it is technically more accurate and recommended by ATI/AMD. */
348 GLuint src_reg = make_src(fp, src);
349 fp->inst[counter].inst0 = R500_INST_TYPE_ALU | R500_INST_TEX_SEM_WAIT;
350 fp->inst[counter].inst1 = R500_RGB_ADDR0(src_reg);
351 fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src_reg);
352 fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
353 | MAKE_SWIZ_RGB_A(make_rgb_swizzle(src))
354 | R500_ALU_RGB_SEL_B_SRC0
355 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(src))
356 | R500_ALU_RGB_OMOD_DISABLE;
357 fp->inst[counter].inst4 = R500_ALPHA_OP_CMP
358 | R500_ALPHA_ADDRD(dest)
359 | R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_alpha_swizzle(src))
360 | R500_ALPHA_SEL_B_SRC0 | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(src))
361 | R500_ALPHA_OMOD_DISABLE;
362 fp->inst[counter].inst5 = R500_ALU_RGBA_OP_CMP
363 | R500_ALU_RGBA_ADDRD(dest)
364 | MAKE_SWIZ_RGBA_C(R500_SWIZ_RGB_ZERO)
365 | MAKE_SWIZ_ALPHA_C(R500_SWIZZLE_ZERO);
366 }
367
368 static GLboolean parse_program(struct r500_fragment_program *fp)
369 {
370 struct gl_fragment_program *mp = &fp->mesa_program;
371 const struct prog_instruction *inst = mp->Base.Instructions;
372 struct prog_instruction *fpi;
373 GLuint src[3], dest, temp[2];
374 int flags, pixel_mask = 0, output_mask = 0, counter = 0;
375
376 if (!inst || inst[0].Opcode == OPCODE_END) {
377 ERROR("The program is empty!\n");
378 return GL_FALSE;
379 }
380
381 for (fpi = mp->Base.Instructions; fpi->Opcode != OPCODE_END; fpi++) {
382
383 if (fpi->Opcode != OPCODE_KIL) {
384 dest = make_dest(fp, fpi->DstReg);
385
386 pixel_mask = fpi->DstReg.WriteMask << 11;
387 output_mask = fpi->DstReg.WriteMask << 14;
388 }
389
390 switch (fpi->Opcode) {
391 case OPCODE_ABS:
392 emit_mov(fp, counter, fpi->SrcReg[0], dest);
393 fp->inst[counter].inst0 |= pixel_mask;
394 fp->inst[counter].inst3 |= R500_ALU_RGB_MOD_A_ABS
395 | R500_ALU_RGB_MOD_B_ABS;
396 fp->inst[counter].inst4 |= R500_ALPHA_MOD_A_ABS
397 | R500_ALPHA_MOD_B_ABS;
398 break;
399 case OPCODE_ADD:
400 src[0] = make_src(fp, fpi->SrcReg[0]);
401 src[1] = make_src(fp, fpi->SrcReg[1]);
402 /* Variation on MAD: 1*src0+src1 */
403 fp->inst[counter].inst0 = R500_INST_TYPE_ALU
404 | pixel_mask;
405 fp->inst[counter].inst1 = R500_RGB_ADDR0(src[0])
406 | R500_RGB_ADDR1(src[1]) | R500_RGB_ADDR2(0);
407 fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0])
408 | R500_ALPHA_ADDR1(src[1]) | R500_ALPHA_ADDR2(0);
409 fp->inst[counter].inst3 = /* 1 */
410 MAKE_SWIZ_RGB_A(R500_SWIZ_RGB_ONE)
411 | R500_ALU_RGB_SEL_B_SRC0 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[0]));
412 fp->inst[counter].inst4 = R500_ALPHA_OP_MAD
413 | R500_ALPHA_ADDRD(dest)
414 | MAKE_SWIZ_ALPHA_A(R500_SWIZZLE_ONE)
415 | R500_ALPHA_SEL_B_SRC0 | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(fpi->SrcReg[0]));
416 fp->inst[counter].inst5 = R500_ALU_RGBA_OP_MAD
417 | R500_ALU_RGBA_ADDRD(dest)
418 | R500_ALU_RGBA_SEL_C_SRC1
419 | MAKE_SWIZ_RGBA_C(make_rgb_swizzle(fpi->SrcReg[1]))
420 | R500_ALU_RGBA_ALPHA_SEL_C_SRC1
421 | MAKE_SWIZ_ALPHA_C(make_alpha_swizzle(fpi->SrcReg[1]));
422 break;
423 case OPCODE_CMP:
424 src[0] = make_src(fp, fpi->SrcReg[0]);
425 src[1] = make_src(fp, fpi->SrcReg[1]);
426 src[2] = make_src(fp, fpi->SrcReg[2]);
427 fp->inst[counter].inst0 = R500_INST_TYPE_ALU
428 | pixel_mask;
429 fp->inst[counter].inst1 = R500_RGB_ADDR0(src[0])
430 | R500_RGB_ADDR1(src[1]) | R500_RGB_ADDR2(src[2]);
431 fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0])
432 | R500_ALPHA_ADDR1(src[1]) | R500_ALPHA_ADDR2(src[2]);
433 fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
434 | MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]))
435 | R500_ALU_RGB_SEL_B_SRC1 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[1]));
436 fp->inst[counter].inst4 = R500_ALPHA_OP_CMP
437 | R500_ALPHA_ADDRD(dest)
438 | R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_alpha_swizzle(fpi->SrcReg[0]))
439 | R500_ALPHA_SEL_B_SRC1 | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(fpi->SrcReg[1]));
440 fp->inst[counter].inst5 = R500_ALU_RGBA_OP_CMP
441 | R500_ALU_RGBA_ADDRD(dest)
442 | R500_ALU_RGBA_SEL_C_SRC2
443 | MAKE_SWIZ_RGBA_C(make_rgb_swizzle(fpi->SrcReg[2]))
444 | R500_ALU_RGBA_ALPHA_SEL_C_SRC2
445 | MAKE_SWIZ_ALPHA_C(make_alpha_swizzle(fpi->SrcReg[2]));
446 break;
447 case OPCODE_DP3:
448 src[0] = make_src(fp, fpi->SrcReg[0]);
449 src[1] = make_src(fp, fpi->SrcReg[1]);
450 fp->inst[counter].inst0 = R500_INST_TYPE_ALU
451 | R500_INST_TEX_SEM_WAIT | pixel_mask;
452 fp->inst[counter].inst1 = R500_RGB_ADDR0(src[0])
453 | R500_RGB_ADDR1(src[1]);
454 fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0])
455 | R500_ALPHA_ADDR1(src[1]);
456 fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
457 | MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]))
458 | R500_ALU_RGB_SEL_B_SRC1 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[1]));
459 fp->inst[counter].inst4 = R500_ALPHA_OP_DP
460 | R500_ALPHA_ADDRD(dest)
461 | R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_alpha_swizzle(fpi->SrcReg[0]))
462 | R500_ALPHA_SEL_B_SRC1 | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(fpi->SrcReg[1]));
463 fp->inst[counter].inst5 = R500_ALU_RGBA_OP_DP3
464 | R500_ALU_RGBA_ADDRD(dest);
465 break;
466 case OPCODE_DP4:
467 src[0] = make_src(fp, fpi->SrcReg[0]);
468 src[1] = make_src(fp, fpi->SrcReg[1]);
469 /* Based on DP3 */
470 fp->inst[counter].inst0 = R500_INST_TYPE_ALU
471 | R500_INST_TEX_SEM_WAIT | pixel_mask;
472 fp->inst[counter].inst1 = R500_RGB_ADDR0(src[0])
473 | R500_RGB_ADDR1(src[1]);
474 fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0])
475 | R500_ALPHA_ADDR1(src[1]);
476 fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
477 | MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]))
478 | R500_ALU_RGB_SEL_B_SRC1 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[1]));
479 fp->inst[counter].inst4 = R500_ALPHA_OP_DP
480 | R500_ALPHA_ADDRD(dest)
481 | R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_alpha_swizzle(fpi->SrcReg[0]))
482 | R500_ALPHA_SEL_B_SRC1 | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(fpi->SrcReg[1]));
483 fp->inst[counter].inst5 = R500_ALU_RGBA_OP_DP4
484 | R500_ALU_RGBA_ADDRD(dest);
485 break;
486 case OPCODE_MAD:
487 src[0] = make_src(fp, fpi->SrcReg[0]);
488 src[1] = make_src(fp, fpi->SrcReg[1]);
489 src[2] = make_src(fp, fpi->SrcReg[2]);
490 fp->inst[counter].inst0 = R500_INST_TYPE_ALU
491 | pixel_mask;
492 fp->inst[counter].inst1 = R500_RGB_ADDR0(src[0])
493 | R500_RGB_ADDR1(src[1]) | R500_RGB_ADDR2(src[2]);
494 fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0])
495 | R500_ALPHA_ADDR1(src[1]) | R500_ALPHA_ADDR2(src[2]);
496 fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
497 | MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]))
498 | R500_ALU_RGB_SEL_B_SRC1 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[1]));
499 fp->inst[counter].inst4 = R500_ALPHA_OP_MAD
500 | R500_ALPHA_ADDRD(dest)
501 | R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_alpha_swizzle(fpi->SrcReg[0]))
502 | R500_ALPHA_SEL_B_SRC1 | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(fpi->SrcReg[1]));
503 fp->inst[counter].inst5 = R500_ALU_RGBA_OP_MAD
504 | R500_ALU_RGBA_ADDRD(dest)
505 | R500_ALU_RGBA_SEL_C_SRC2
506 | MAKE_SWIZ_RGBA_C(make_rgb_swizzle(fpi->SrcReg[2]))
507 | R500_ALU_RGBA_ALPHA_SEL_C_SRC2
508 | MAKE_SWIZ_ALPHA_C(make_alpha_swizzle(fpi->SrcReg[2]));
509 break;
510 case OPCODE_MAX:
511 src[0] = make_src(fp, fpi->SrcReg[0]);
512 src[1] = make_src(fp, fpi->SrcReg[1]);
513 fp->inst[counter].inst0 = R500_INST_TYPE_ALU | pixel_mask;
514 fp->inst[counter].inst1 = R500_RGB_ADDR0(src[0]) | R500_RGB_ADDR1(src[1]);
515 fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0]) | R500_ALPHA_ADDR1(src[1]);
516 fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
517 | MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]))
518 | R500_ALU_RGB_SEL_B_SRC1
519 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[1]));
520 fp->inst[counter].inst4 = R500_ALPHA_OP_MAX
521 | R500_ALPHA_ADDRD(dest)
522 | R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_alpha_swizzle(fpi->SrcReg[0]))
523 | R500_ALPHA_SEL_B_SRC1 | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(fpi->SrcReg[1]));
524 fp->inst[counter].inst5 = R500_ALU_RGBA_OP_MAX
525 | R500_ALU_RGBA_ADDRD(dest);
526 break;
527 case OPCODE_MIN:
528 src[0] = make_src(fp, fpi->SrcReg[0]);
529 src[1] = make_src(fp, fpi->SrcReg[1]);
530 fp->inst[counter].inst0 = R500_INST_TYPE_ALU | pixel_mask;
531 fp->inst[counter].inst1 = R500_RGB_ADDR0(src[0]) | R500_RGB_ADDR1(src[1]);
532 fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0]) | R500_ALPHA_ADDR1(src[1]);
533 fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
534 | MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]))
535 | R500_ALU_RGB_SEL_B_SRC1
536 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[1]));
537 fp->inst[counter].inst4 = R500_ALPHA_OP_MIN
538 | R500_ALPHA_ADDRD(dest)
539 | R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_alpha_swizzle(fpi->SrcReg[0]))
540 | R500_ALPHA_SEL_B_SRC1 | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(fpi->SrcReg[1]));
541 fp->inst[counter].inst5 = R500_ALU_RGBA_OP_MIN
542 | R500_ALU_RGBA_ADDRD(dest);
543 break;
544 case OPCODE_MOV:
545 emit_mov(fp, counter, fpi->SrcReg[0], dest);
546 fp->inst[counter].inst0 |= pixel_mask;
547 break;
548 case OPCODE_MUL:
549 src[0] = make_src(fp, fpi->SrcReg[0]);
550 src[1] = make_src(fp, fpi->SrcReg[1]);
551 /* Variation on MAD: src0*src1+0 */
552 fp->inst[counter].inst0 = R500_INST_TYPE_ALU
553 | R500_INST_TEX_SEM_WAIT | pixel_mask;
554 fp->inst[counter].inst1 = R500_RGB_ADDR0(src[0])
555 | R500_RGB_ADDR1(src[1]);
556 fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0])
557 | R500_ALPHA_ADDR1(src[1]);
558 fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
559 | MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]))
560 | R500_ALU_RGB_SEL_B_SRC1 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[1]));
561 fp->inst[counter].inst4 = R500_ALPHA_OP_MAD
562 | R500_ALPHA_ADDRD(dest)
563 | R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_alpha_swizzle(fpi->SrcReg[0]))
564 | R500_ALPHA_SEL_B_SRC1 | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(fpi->SrcReg[1]));
565 fp->inst[counter].inst5 = R500_ALU_RGBA_OP_MAD
566 | R500_ALU_RGBA_ADDRD(dest)
567 // | R500_ALU_RGBA_SEL_C_SRC2
568 | MAKE_SWIZ_RGBA_C(R500_SWIZ_RGB_ZERO)
569 // | R500_ALU_RGBA_ALPHA_SEL_C_SRC2
570 | MAKE_SWIZ_ALPHA_C(R500_SWIZZLE_ZERO);
571 break;
572 case OPCODE_SUB:
573 src[0] = make_src(fp, fpi->SrcReg[0]);
574 src[1] = make_src(fp, fpi->SrcReg[1]);
575 /* Variation on MAD: 1*src0-src1 */
576 fp->inst[counter].inst0 = R500_INST_TYPE_ALU
577 | pixel_mask;
578 fp->inst[counter].inst1 = R500_RGB_ADDR1(src[0])
579 | R500_RGB_ADDR2(src[1]);
580 fp->inst[counter].inst2 = R500_ALPHA_ADDR1(src[0])
581 | R500_ALPHA_ADDR2(src[1]);
582 fp->inst[counter].inst3 = /* 1 */
583 MAKE_SWIZ_RGB_A(R500_SWIZ_RGB_ONE)
584 | R500_ALU_RGB_SEL_B_SRC1 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[0]));
585 fp->inst[counter].inst4 = R500_ALPHA_OP_MAD
586 | R500_ALPHA_ADDRD(dest)
587 | R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(R500_SWIZZLE_ONE)
588 | R500_ALPHA_SEL_B_SRC1 | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(fpi->SrcReg[0]));
589 fp->inst[counter].inst5 = R500_ALU_RGBA_OP_MAD
590 | R500_ALU_RGBA_ADDRD(dest)
591 | R500_ALU_RGBA_SEL_C_SRC2
592 | MAKE_SWIZ_RGBA_C(make_rgb_swizzle(fpi->SrcReg[1]))
593 | R500_ALU_RGBA_MOD_C_NEG
594 | R500_ALU_RGBA_ALPHA_SEL_C_SRC2
595 | MAKE_SWIZ_ALPHA_C(make_alpha_swizzle(fpi->SrcReg[1]))
596 | R500_ALU_RGBA_ALPHA_MOD_C_NEG;
597 break;
598 case OPCODE_TEX:
599 emit_tex(fp, fpi, OPCODE_TEX, dest, counter);
600 break;
601 case OPCODE_TXB:
602 emit_tex(fp, fpi, OPCODE_TXB, dest, counter);
603 break;
604 case OPCODE_TXP:
605 emit_tex(fp, fpi, OPCODE_TXP, dest, counter);
606 break;
607 default:
608 ERROR("unknown fpi->Opcode %d\n", fpi->Opcode);
609 break;
610 }
611
612 /* Finishing touches */
613 if (fpi->SaturateMode == SATURATE_ZERO_ONE) {
614 fp->inst[counter].inst0 |= R500_INST_RGB_CLAMP | R500_INST_ALPHA_CLAMP;
615 }
616 if (fpi->DstReg.File == PROGRAM_OUTPUT) {
617 fp->inst[counter].inst0 |= R500_INST_TYPE_OUT | output_mask;
618 }
619
620 counter++;
621
622 if (fp->error)
623 return GL_FALSE;
624
625 }
626
627 /* Finish him! (If it's an ALU/OUT instruction...) */
628 if ((fp->inst[counter-1].inst0 & 0x3) <= 1) {
629 fp->inst[counter-1].inst0 |= R500_INST_TYPE_OUT
630 | R500_INST_TEX_SEM_WAIT | R500_INST_LAST;
631 } else {
632 /* We still need to put an output inst, right? */
633 fp->inst[counter].inst0 = R500_INST_TYPE_OUT
634 | R500_INST_TEX_SEM_WAIT | R500_INST_LAST |
635 output_mask;
636 fp->inst[counter].inst1 = R500_RGB_ADDR0(dest);
637 fp->inst[counter].inst2 = R500_ALPHA_ADDR0(dest);
638 fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
639 | MAKE_SWIZ_RGB_A(R500_SWIZ_RGB_RGB)
640 | R500_ALU_RGB_SEL_B_SRC0
641 | MAKE_SWIZ_RGB_B(R500_SWIZ_RGB_ONE);
642 fp->inst[counter].inst4 = R500_ALPHA_OP_MAD
643 | R500_ALPHA_ADDRD(0)
644 | R500_ALPHA_SEL_A_SRC0 | R500_ALPHA_SEL_B_SRC0
645 | R500_ALPHA_SWIZ_A_A | R500_ALPHA_SWIZ_B_1;
646 fp->inst[counter].inst5 = R500_ALU_RGBA_OP_MAD
647 | R500_ALU_RGBA_ADDRD(0)
648 | MAKE_SWIZ_RGBA_C(R500_SWIZ_RGB_ZERO)
649 | MAKE_SWIZ_ALPHA_C(R500_SWIZZLE_ZERO);
650 counter++;
651 }
652
653 fp->cs->nrslots = counter;
654
655 fp->max_temp_idx++;
656
657 return GL_TRUE;
658 }
659
660 static void init_program(r300ContextPtr r300, struct r500_fragment_program *fp)
661 {
662 struct r300_pfs_compile_state *cs = NULL;
663 struct gl_fragment_program *mp = &fp->mesa_program;
664 struct prog_instruction *fpi;
665 GLuint InputsRead = mp->Base.InputsRead;
666 GLuint temps_used = 0; /* for fp->temps[] */
667 int i, j;
668
669 /* New compile, reset tracking data */
670 fp->optimization =
671 driQueryOptioni(&r300->radeon.optionCache, "fp_optimization");
672 fp->translated = GL_FALSE;
673 fp->error = GL_FALSE;
674 fp->cs = cs = &(R300_CONTEXT(fp->ctx)->state.pfs_compile);
675 fp->cur_node = 0;
676 fp->first_node_has_tex = 0;
677 fp->const_nr = 0;
678 /* Size of pixel stack, plus 1. */
679 fp->max_temp_idx = 1;
680 /* Temp register offset. */
681 fp->temp_reg_offset = 0;
682 fp->node[0].alu_end = -1;
683 fp->node[0].tex_end = -1;
684
685 _mesa_memset(cs, 0, sizeof(*fp->cs));
686 for (i = 0; i < PFS_MAX_ALU_INST; i++) {
687 for (j = 0; j < 3; j++) {
688 cs->slot[i].vsrc[j] = SRC_CONST;
689 cs->slot[i].ssrc[j] = SRC_CONST;
690 }
691 }
692
693 /* Work out what temps the Mesa inputs correspond to, this must match
694 * what setup_rs_unit does, which shouldn't be a problem as rs_unit
695 * configures itself based on the fragprog's InputsRead
696 *
697 * NOTE: this depends on get_hw_temp() allocating registers in order,
698 * starting from register 0, so we're just going to do that instead.
699 */
700
701 /* Texcoords come first */
702 for (i = 0; i < fp->ctx->Const.MaxTextureUnits; i++) {
703 if (InputsRead & (FRAG_BIT_TEX0 << i)) {
704 cs->inputs[FRAG_ATTRIB_TEX0 + i].refcount = 0;
705 cs->inputs[FRAG_ATTRIB_TEX0 + i].reg =
706 fp->temp_reg_offset;
707 fp->temp_reg_offset++;
708 }
709 }
710 InputsRead &= ~FRAG_BITS_TEX_ANY;
711
712 /* fragment position treated as a texcoord */
713 if (InputsRead & FRAG_BIT_WPOS) {
714 cs->inputs[FRAG_ATTRIB_WPOS].refcount = 0;
715 cs->inputs[FRAG_ATTRIB_WPOS].reg =
716 fp->temp_reg_offset;
717 fp->temp_reg_offset++;
718 }
719 InputsRead &= ~FRAG_BIT_WPOS;
720
721 /* Then primary colour */
722 if (InputsRead & FRAG_BIT_COL0) {
723 cs->inputs[FRAG_ATTRIB_COL0].refcount = 0;
724 cs->inputs[FRAG_ATTRIB_COL0].reg =
725 fp->temp_reg_offset;
726 fp->temp_reg_offset++;
727 }
728 InputsRead &= ~FRAG_BIT_COL0;
729
730 /* Secondary color */
731 if (InputsRead & FRAG_BIT_COL1) {
732 cs->inputs[FRAG_ATTRIB_COL1].refcount = 0;
733 cs->inputs[FRAG_ATTRIB_COL1].reg =
734 fp->temp_reg_offset;
735 fp->temp_reg_offset++;
736 }
737 InputsRead &= ~FRAG_BIT_COL1;
738
739 /* Anything else */
740 if (InputsRead) {
741 WARN_ONCE("Don't know how to handle inputs 0x%x\n", InputsRead);
742 /* force read from hwreg 0 for now */
743 for (i = 0; i < 32; i++)
744 if (InputsRead & (1 << i))
745 cs->inputs[i].reg = 0;
746 }
747
748 /* Pre-parse the mesa program, grabbing refcounts on input/temp regs.
749 * That way, we can free up the reg when it's no longer needed
750 */
751 if (!mp->Base.Instructions) {
752 ERROR("No instructions found in program, going to go die now.\n");
753 return;
754 }
755
756 #if 0
757 for (fpi = mp->Base.Instructions; fpi->Opcode != OPCODE_END; fpi++) {
758 int idx;
759 for (i = 0; i < 3; i++) {
760 idx = fpi->SrcReg[i].Index;
761 if (fpi->SrcReg[i].File == PROGRAM_INPUT) {
762 cs->inputs[idx].refcount++;
763 if (fp->max_temp_idx < idx)
764 fp->max_temp_idx = idx;
765 }
766 }
767 }
768 #endif
769
770 fp->max_temp_idx = fp->temp_reg_offset + 1;
771
772 cs->temp_in_use = temps_used;
773 }
774
775 static void update_params(struct r500_fragment_program *fp)
776 {
777 struct gl_fragment_program *mp = &fp->mesa_program;
778
779 /* Ask Mesa nicely to fill in ParameterValues for us */
780 if (mp->Base.Parameters)
781 _mesa_load_state_parameters(fp->ctx, mp->Base.Parameters);
782 }
783
784 void r500TranslateFragmentShader(r300ContextPtr r300,
785 struct r500_fragment_program *fp)
786 {
787
788 struct r300_pfs_compile_state *cs = NULL;
789
790 if (!fp->translated) {
791
792 /* I need to see what I'm working with! */
793 fprintf(stderr, "Mesa program:\n");
794 fprintf(stderr, "-------------\n");
795 _mesa_print_program(&fp->mesa_program.Base);
796 fflush(stdout);
797
798 init_program(r300, fp);
799 cs = fp->cs;
800
801 if (parse_program(fp) == GL_FALSE) {
802 ERROR("Huh. Couldn't parse program. There should be additional errors explaining why.\nUsing dumb shader...\n");
803 dumb_shader(fp);
804 fp->inst_offset = 0;
805 fp->inst_end = cs->nrslots - 1;
806 return;
807 }
808 fp->inst_offset = 0;
809 fp->inst_end = cs->nrslots - 1;
810
811 fp->translated = GL_TRUE;
812 r300UpdateStateParameters(fp->ctx, _NEW_PROGRAM);
813 }
814
815 update_params(fp);
816 }