nvc0: do not set tiled mode on gart bo when fence debugging is used
[mesa.git] / src / gallium / drivers / r300 / r300_tgsi_to_rc.c
1 /*
2 * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com>
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 COPYRIGHT HOLDER(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 #include "r300_tgsi_to_rc.h"
24
25 #include "compiler/radeon_compiler.h"
26
27 #include "tgsi/tgsi_info.h"
28 #include "tgsi/tgsi_parse.h"
29 #include "tgsi/tgsi_scan.h"
30 #include "tgsi/tgsi_util.h"
31
32 static unsigned translate_opcode(unsigned opcode)
33 {
34 switch(opcode) {
35 case TGSI_OPCODE_ARL: return RC_OPCODE_ARL;
36 case TGSI_OPCODE_MOV: return RC_OPCODE_MOV;
37 case TGSI_OPCODE_LIT: return RC_OPCODE_LIT;
38 case TGSI_OPCODE_RCP: return RC_OPCODE_RCP;
39 case TGSI_OPCODE_RSQ: return RC_OPCODE_RSQ;
40 case TGSI_OPCODE_EXP: return RC_OPCODE_EXP;
41 case TGSI_OPCODE_LOG: return RC_OPCODE_LOG;
42 case TGSI_OPCODE_MUL: return RC_OPCODE_MUL;
43 case TGSI_OPCODE_ADD: return RC_OPCODE_ADD;
44 case TGSI_OPCODE_DP3: return RC_OPCODE_DP3;
45 case TGSI_OPCODE_DP4: return RC_OPCODE_DP4;
46 case TGSI_OPCODE_DST: return RC_OPCODE_DST;
47 case TGSI_OPCODE_MIN: return RC_OPCODE_MIN;
48 case TGSI_OPCODE_MAX: return RC_OPCODE_MAX;
49 case TGSI_OPCODE_SLT: return RC_OPCODE_SLT;
50 case TGSI_OPCODE_SGE: return RC_OPCODE_SGE;
51 case TGSI_OPCODE_MAD: return RC_OPCODE_MAD;
52 case TGSI_OPCODE_SUB: return RC_OPCODE_SUB;
53 case TGSI_OPCODE_LRP: return RC_OPCODE_LRP;
54 case TGSI_OPCODE_CND: return RC_OPCODE_CND;
55 /* case TGSI_OPCODE_DP2A: return RC_OPCODE_DP2A; */
56 /* gap */
57 case TGSI_OPCODE_FRC: return RC_OPCODE_FRC;
58 case TGSI_OPCODE_CLAMP: return RC_OPCODE_CLAMP;
59 case TGSI_OPCODE_FLR: return RC_OPCODE_FLR;
60 case TGSI_OPCODE_ROUND: return RC_OPCODE_ROUND;
61 case TGSI_OPCODE_EX2: return RC_OPCODE_EX2;
62 case TGSI_OPCODE_LG2: return RC_OPCODE_LG2;
63 case TGSI_OPCODE_POW: return RC_OPCODE_POW;
64 case TGSI_OPCODE_XPD: return RC_OPCODE_XPD;
65 /* gap */
66 case TGSI_OPCODE_ABS: return RC_OPCODE_ABS;
67 /* case TGSI_OPCODE_RCC: return RC_OPCODE_RCC; */
68 case TGSI_OPCODE_DPH: return RC_OPCODE_DPH;
69 case TGSI_OPCODE_COS: return RC_OPCODE_COS;
70 case TGSI_OPCODE_DDX: return RC_OPCODE_DDX;
71 case TGSI_OPCODE_DDY: return RC_OPCODE_DDY;
72 case TGSI_OPCODE_KILP: return RC_OPCODE_KILP;
73 /* case TGSI_OPCODE_PK2H: return RC_OPCODE_PK2H; */
74 /* case TGSI_OPCODE_PK2US: return RC_OPCODE_PK2US; */
75 /* case TGSI_OPCODE_PK4B: return RC_OPCODE_PK4B; */
76 /* case TGSI_OPCODE_PK4UB: return RC_OPCODE_PK4UB; */
77 /* case TGSI_OPCODE_RFL: return RC_OPCODE_RFL; */
78 case TGSI_OPCODE_SEQ: return RC_OPCODE_SEQ;
79 case TGSI_OPCODE_SFL: return RC_OPCODE_SFL;
80 case TGSI_OPCODE_SGT: return RC_OPCODE_SGT;
81 case TGSI_OPCODE_SIN: return RC_OPCODE_SIN;
82 case TGSI_OPCODE_SLE: return RC_OPCODE_SLE;
83 case TGSI_OPCODE_SNE: return RC_OPCODE_SNE;
84 /* case TGSI_OPCODE_STR: return RC_OPCODE_STR; */
85 case TGSI_OPCODE_TEX: return RC_OPCODE_TEX;
86 case TGSI_OPCODE_TXD: return RC_OPCODE_TXD;
87 case TGSI_OPCODE_TXP: return RC_OPCODE_TXP;
88 /* case TGSI_OPCODE_UP2H: return RC_OPCODE_UP2H; */
89 /* case TGSI_OPCODE_UP2US: return RC_OPCODE_UP2US; */
90 /* case TGSI_OPCODE_UP4B: return RC_OPCODE_UP4B; */
91 /* case TGSI_OPCODE_UP4UB: return RC_OPCODE_UP4UB; */
92 /* case TGSI_OPCODE_X2D: return RC_OPCODE_X2D; */
93 /* case TGSI_OPCODE_ARA: return RC_OPCODE_ARA; */
94 /* case TGSI_OPCODE_ARR: return RC_OPCODE_ARR; */
95 /* case TGSI_OPCODE_BRA: return RC_OPCODE_BRA; */
96 /* case TGSI_OPCODE_CAL: return RC_OPCODE_CAL; */
97 /* case TGSI_OPCODE_RET: return RC_OPCODE_RET; */
98 case TGSI_OPCODE_SSG: return RC_OPCODE_SSG;
99 case TGSI_OPCODE_CMP: return RC_OPCODE_CMP;
100 case TGSI_OPCODE_SCS: return RC_OPCODE_SCS;
101 case TGSI_OPCODE_TXB: return RC_OPCODE_TXB;
102 /* case TGSI_OPCODE_NRM: return RC_OPCODE_NRM; */
103 /* case TGSI_OPCODE_DIV: return RC_OPCODE_DIV; */
104 case TGSI_OPCODE_DP2: return RC_OPCODE_DP2;
105 case TGSI_OPCODE_TXL: return RC_OPCODE_TXL;
106 case TGSI_OPCODE_BRK: return RC_OPCODE_BRK;
107 case TGSI_OPCODE_IF: return RC_OPCODE_IF;
108 case TGSI_OPCODE_BGNLOOP: return RC_OPCODE_BGNLOOP;
109 case TGSI_OPCODE_ELSE: return RC_OPCODE_ELSE;
110 case TGSI_OPCODE_ENDIF: return RC_OPCODE_ENDIF;
111 case TGSI_OPCODE_ENDLOOP: return RC_OPCODE_ENDLOOP;
112 /* case TGSI_OPCODE_PUSHA: return RC_OPCODE_PUSHA; */
113 /* case TGSI_OPCODE_POPA: return RC_OPCODE_POPA; */
114 case TGSI_OPCODE_CEIL: return RC_OPCODE_CEIL;
115 /* case TGSI_OPCODE_I2F: return RC_OPCODE_I2F; */
116 /* case TGSI_OPCODE_NOT: return RC_OPCODE_NOT; */
117 case TGSI_OPCODE_TRUNC: return RC_OPCODE_TRUNC;
118 /* case TGSI_OPCODE_SHL: return RC_OPCODE_SHL; */
119 /* case TGSI_OPCODE_ISHR: return RC_OPCODE_SHR; */
120 /* case TGSI_OPCODE_AND: return RC_OPCODE_AND; */
121 /* case TGSI_OPCODE_OR: return RC_OPCODE_OR; */
122 /* case TGSI_OPCODE_MOD: return RC_OPCODE_MOD; */
123 /* case TGSI_OPCODE_XOR: return RC_OPCODE_XOR; */
124 /* case TGSI_OPCODE_SAD: return RC_OPCODE_SAD; */
125 /* case TGSI_OPCODE_TXF: return RC_OPCODE_TXF; */
126 /* case TGSI_OPCODE_TXQ: return RC_OPCODE_TXQ; */
127 case TGSI_OPCODE_CONT: return RC_OPCODE_CONT;
128 /* case TGSI_OPCODE_EMIT: return RC_OPCODE_EMIT; */
129 /* case TGSI_OPCODE_ENDPRIM: return RC_OPCODE_ENDPRIM; */
130 /* case TGSI_OPCODE_BGNLOOP2: return RC_OPCODE_BGNLOOP2; */
131 /* case TGSI_OPCODE_BGNSUB: return RC_OPCODE_BGNSUB; */
132 /* case TGSI_OPCODE_ENDLOOP2: return RC_OPCODE_ENDLOOP2; */
133 /* case TGSI_OPCODE_ENDSUB: return RC_OPCODE_ENDSUB; */
134 case TGSI_OPCODE_NOP: return RC_OPCODE_NOP;
135 /* gap */
136 /* case TGSI_OPCODE_NRM4: return RC_OPCODE_NRM4; */
137 /* case TGSI_OPCODE_CALLNZ: return RC_OPCODE_CALLNZ; */
138 /* case TGSI_OPCODE_BREAKC: return RC_OPCODE_BREAKC; */
139 case TGSI_OPCODE_KIL: return RC_OPCODE_KIL;
140 }
141
142 fprintf(stderr, "r300: Unknown TGSI/RC opcode: %s\n", tgsi_get_opcode_name(opcode));
143 return RC_OPCODE_ILLEGAL_OPCODE;
144 }
145
146 static unsigned translate_saturate(unsigned saturate)
147 {
148 switch(saturate) {
149 default:
150 fprintf(stderr, "Unknown saturate mode: %i\n", saturate);
151 /* fall-through */
152 case TGSI_SAT_NONE: return RC_SATURATE_NONE;
153 case TGSI_SAT_ZERO_ONE: return RC_SATURATE_ZERO_ONE;
154 }
155 }
156
157 static unsigned translate_register_file(unsigned file)
158 {
159 switch(file) {
160 case TGSI_FILE_CONSTANT: return RC_FILE_CONSTANT;
161 case TGSI_FILE_IMMEDIATE: return RC_FILE_CONSTANT;
162 case TGSI_FILE_INPUT: return RC_FILE_INPUT;
163 case TGSI_FILE_OUTPUT: return RC_FILE_OUTPUT;
164 default:
165 fprintf(stderr, "Unhandled register file: %i\n", file);
166 /* fall-through */
167 case TGSI_FILE_TEMPORARY: return RC_FILE_TEMPORARY;
168 case TGSI_FILE_ADDRESS: return RC_FILE_ADDRESS;
169 }
170 }
171
172 static int translate_register_index(
173 struct tgsi_to_rc * ttr,
174 unsigned file,
175 int index)
176 {
177 if (file == TGSI_FILE_IMMEDIATE)
178 return ttr->immediate_offset + index;
179
180 return index;
181 }
182
183 static void transform_dstreg(
184 struct tgsi_to_rc * ttr,
185 struct rc_dst_register * dst,
186 struct tgsi_full_dst_register * src)
187 {
188 dst->File = translate_register_file(src->Register.File);
189 dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);
190 dst->WriteMask = src->Register.WriteMask;
191
192 if (src->Register.Indirect) {
193 ttr->error = TRUE;
194 fprintf(stderr, "r300: Relative addressing of destination operands "
195 "is unsupported.\n");
196 }
197 }
198
199 static void transform_srcreg(
200 struct tgsi_to_rc * ttr,
201 struct rc_src_register * dst,
202 struct tgsi_full_src_register * src)
203 {
204 unsigned i, j;
205
206 dst->File = translate_register_file(src->Register.File);
207 dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);
208 dst->RelAddr = src->Register.Indirect;
209 dst->Swizzle = tgsi_util_get_full_src_register_swizzle(src, 0);
210 dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 1) << 3;
211 dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 2) << 6;
212 dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 3) << 9;
213 dst->Abs = src->Register.Absolute;
214 dst->Negate = src->Register.Negate ? RC_MASK_XYZW : 0;
215
216 if (src->Register.File == TGSI_FILE_IMMEDIATE) {
217 for (i = 0; i < ttr->imms_to_swizzle_count; i++) {
218 if (ttr->imms_to_swizzle[i].index == src->Register.Index) {
219 dst->File = RC_FILE_TEMPORARY;
220 dst->Index = 0;
221 dst->Swizzle = 0;
222 for (j = 0; j < 4; j++) {
223 dst->Swizzle |= GET_SWZ(ttr->imms_to_swizzle[i].swizzle,
224 tgsi_util_get_full_src_register_swizzle(src, j)) << (j * 3);
225 }
226 break;
227 }
228 }
229 }
230 }
231
232 static void transform_texture(struct rc_instruction * dst, struct tgsi_instruction_texture src,
233 uint32_t *shadowSamplers)
234 {
235 switch(src.Texture) {
236 case TGSI_TEXTURE_1D:
237 dst->U.I.TexSrcTarget = RC_TEXTURE_1D;
238 break;
239 case TGSI_TEXTURE_2D:
240 dst->U.I.TexSrcTarget = RC_TEXTURE_2D;
241 break;
242 case TGSI_TEXTURE_3D:
243 dst->U.I.TexSrcTarget = RC_TEXTURE_3D;
244 break;
245 case TGSI_TEXTURE_CUBE:
246 dst->U.I.TexSrcTarget = RC_TEXTURE_CUBE;
247 break;
248 case TGSI_TEXTURE_RECT:
249 dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;
250 break;
251 case TGSI_TEXTURE_SHADOW1D:
252 dst->U.I.TexSrcTarget = RC_TEXTURE_1D;
253 dst->U.I.TexShadow = 1;
254 *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
255 break;
256 case TGSI_TEXTURE_SHADOW2D:
257 dst->U.I.TexSrcTarget = RC_TEXTURE_2D;
258 dst->U.I.TexShadow = 1;
259 *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
260 break;
261 case TGSI_TEXTURE_SHADOWRECT:
262 dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;
263 dst->U.I.TexShadow = 1;
264 *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
265 break;
266 }
267 dst->U.I.TexSwizzle = RC_SWIZZLE_XYZW;
268 }
269
270 static void transform_instruction(struct tgsi_to_rc * ttr, struct tgsi_full_instruction * src)
271 {
272 struct rc_instruction * dst;
273 int i;
274
275 dst = rc_insert_new_instruction(ttr->compiler, ttr->compiler->Program.Instructions.Prev);
276 dst->U.I.Opcode = translate_opcode(src->Instruction.Opcode);
277 dst->U.I.SaturateMode = translate_saturate(src->Instruction.Saturate);
278
279 if (src->Instruction.NumDstRegs)
280 transform_dstreg(ttr, &dst->U.I.DstReg, &src->Dst[0]);
281
282 for(i = 0; i < src->Instruction.NumSrcRegs; ++i) {
283 if (src->Src[i].Register.File == TGSI_FILE_SAMPLER)
284 dst->U.I.TexSrcUnit = src->Src[i].Register.Index;
285 else
286 transform_srcreg(ttr, &dst->U.I.SrcReg[i], &src->Src[i]);
287 }
288
289 /* Texturing. */
290 if (src->Instruction.Texture)
291 transform_texture(dst, src->Texture,
292 &ttr->compiler->Program.ShadowSamplers);
293 }
294
295 static void handle_immediate(struct tgsi_to_rc * ttr,
296 struct tgsi_full_immediate * imm,
297 unsigned index)
298 {
299 struct rc_constant constant;
300 unsigned swizzle = 0;
301 boolean can_swizzle = TRUE;
302 unsigned i;
303
304 for (i = 0; i < 4; i++) {
305 if (imm->u[i].Float == 0.0f) {
306 swizzle |= RC_SWIZZLE_ZERO << (i * 3);
307 } else if (imm->u[i].Float == 0.5f && ttr->use_half_swizzles) {
308 swizzle |= RC_SWIZZLE_HALF << (i * 3);
309 } else if (imm->u[i].Float == 1.0f) {
310 swizzle |= RC_SWIZZLE_ONE << (i * 3);
311 } else {
312 can_swizzle = FALSE;
313 break;
314 }
315 }
316
317 if (can_swizzle) {
318 ttr->imms_to_swizzle[ttr->imms_to_swizzle_count].index = index;
319 ttr->imms_to_swizzle[ttr->imms_to_swizzle_count].swizzle = swizzle;
320 ttr->imms_to_swizzle_count++;
321 } else {
322 constant.Type = RC_CONSTANT_IMMEDIATE;
323 constant.Size = 4;
324 for(i = 0; i < 4; ++i)
325 constant.u.Immediate[i] = imm->u[i].Float;
326 rc_constants_add(&ttr->compiler->Program.Constants, &constant);
327 }
328 }
329
330 void r300_tgsi_to_rc(struct tgsi_to_rc * ttr,
331 const struct tgsi_token * tokens)
332 {
333 struct tgsi_full_instruction *inst;
334 struct tgsi_parse_context parser;
335 unsigned imm_index = 0;
336 int i;
337
338 ttr->error = FALSE;
339
340 /* Allocate constants placeholders.
341 *
342 * Note: What if declared constants are not contiguous? */
343 for(i = 0; i <= ttr->info->file_max[TGSI_FILE_CONSTANT]; ++i) {
344 struct rc_constant constant;
345 memset(&constant, 0, sizeof(constant));
346 constant.Type = RC_CONSTANT_EXTERNAL;
347 constant.Size = 4;
348 constant.u.External = i;
349 rc_constants_add(&ttr->compiler->Program.Constants, &constant);
350 }
351
352 ttr->immediate_offset = ttr->compiler->Program.Constants.Count;
353
354 ttr->imms_to_swizzle = malloc(ttr->info->immediate_count * sizeof(struct swizzled_imms));
355 ttr->imms_to_swizzle_count = 0;
356
357 tgsi_parse_init(&parser, tokens);
358
359 while (!tgsi_parse_end_of_tokens(&parser)) {
360 tgsi_parse_token(&parser);
361
362 switch (parser.FullToken.Token.Type) {
363 case TGSI_TOKEN_TYPE_DECLARATION:
364 break;
365 case TGSI_TOKEN_TYPE_IMMEDIATE:
366 handle_immediate(ttr, &parser.FullToken.FullImmediate, imm_index);
367 imm_index++;
368 break;
369 case TGSI_TOKEN_TYPE_INSTRUCTION:
370 inst = &parser.FullToken.FullInstruction;
371 if (inst->Instruction.Opcode == TGSI_OPCODE_END) {
372 break;
373 }
374
375 transform_instruction(ttr, inst);
376 break;
377 }
378 }
379
380 tgsi_parse_free(&parser);
381
382 free(ttr->imms_to_swizzle);
383
384 rc_calculate_inputs_outputs(ttr->compiler);
385 }