r300g: implement ARR opcode
[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_DP2A: return RC_OPCODE_DP2A; */
55 case TGSI_OPCODE_FRC: return RC_OPCODE_FRC;
56 case TGSI_OPCODE_CLAMP: return RC_OPCODE_CLAMP;
57 case TGSI_OPCODE_FLR: return RC_OPCODE_FLR;
58 case TGSI_OPCODE_ROUND: return RC_OPCODE_ROUND;
59 case TGSI_OPCODE_EX2: return RC_OPCODE_EX2;
60 case TGSI_OPCODE_LG2: return RC_OPCODE_LG2;
61 case TGSI_OPCODE_POW: return RC_OPCODE_POW;
62 case TGSI_OPCODE_XPD: return RC_OPCODE_XPD;
63 case TGSI_OPCODE_ABS: return RC_OPCODE_ABS;
64 case TGSI_OPCODE_DPH: return RC_OPCODE_DPH;
65 case TGSI_OPCODE_COS: return RC_OPCODE_COS;
66 case TGSI_OPCODE_DDX: return RC_OPCODE_DDX;
67 case TGSI_OPCODE_DDY: return RC_OPCODE_DDY;
68 case TGSI_OPCODE_KILL: return RC_OPCODE_KILP;
69 /* case TGSI_OPCODE_PK2H: return RC_OPCODE_PK2H; */
70 /* case TGSI_OPCODE_PK2US: return RC_OPCODE_PK2US; */
71 /* case TGSI_OPCODE_PK4B: return RC_OPCODE_PK4B; */
72 /* case TGSI_OPCODE_PK4UB: return RC_OPCODE_PK4UB; */
73 case TGSI_OPCODE_SEQ: return RC_OPCODE_SEQ;
74 case TGSI_OPCODE_SGT: return RC_OPCODE_SGT;
75 case TGSI_OPCODE_SIN: return RC_OPCODE_SIN;
76 case TGSI_OPCODE_SLE: return RC_OPCODE_SLE;
77 case TGSI_OPCODE_SNE: return RC_OPCODE_SNE;
78 case TGSI_OPCODE_TEX: return RC_OPCODE_TEX;
79 case TGSI_OPCODE_TXD: return RC_OPCODE_TXD;
80 case TGSI_OPCODE_TXP: return RC_OPCODE_TXP;
81 /* case TGSI_OPCODE_UP2H: return RC_OPCODE_UP2H; */
82 /* case TGSI_OPCODE_UP2US: return RC_OPCODE_UP2US; */
83 /* case TGSI_OPCODE_UP4B: return RC_OPCODE_UP4B; */
84 /* case TGSI_OPCODE_UP4UB: return RC_OPCODE_UP4UB; */
85 case TGSI_OPCODE_ARR: return RC_OPCODE_ARR;
86 /* case TGSI_OPCODE_CAL: return RC_OPCODE_CAL; */
87 /* case TGSI_OPCODE_RET: return RC_OPCODE_RET; */
88 case TGSI_OPCODE_SSG: return RC_OPCODE_SSG;
89 case TGSI_OPCODE_CMP: return RC_OPCODE_CMP;
90 case TGSI_OPCODE_SCS: return RC_OPCODE_SCS;
91 case TGSI_OPCODE_TXB: return RC_OPCODE_TXB;
92 /* case TGSI_OPCODE_DIV: return RC_OPCODE_DIV; */
93 case TGSI_OPCODE_DP2: return RC_OPCODE_DP2;
94 case TGSI_OPCODE_TXL: return RC_OPCODE_TXL;
95 case TGSI_OPCODE_BRK: return RC_OPCODE_BRK;
96 case TGSI_OPCODE_IF: return RC_OPCODE_IF;
97 case TGSI_OPCODE_BGNLOOP: return RC_OPCODE_BGNLOOP;
98 case TGSI_OPCODE_ELSE: return RC_OPCODE_ELSE;
99 case TGSI_OPCODE_ENDIF: return RC_OPCODE_ENDIF;
100 case TGSI_OPCODE_ENDLOOP: return RC_OPCODE_ENDLOOP;
101 /* case TGSI_OPCODE_PUSHA: return RC_OPCODE_PUSHA; */
102 /* case TGSI_OPCODE_POPA: return RC_OPCODE_POPA; */
103 case TGSI_OPCODE_CEIL: return RC_OPCODE_CEIL;
104 /* case TGSI_OPCODE_I2F: return RC_OPCODE_I2F; */
105 /* case TGSI_OPCODE_NOT: return RC_OPCODE_NOT; */
106 case TGSI_OPCODE_TRUNC: return RC_OPCODE_TRUNC;
107 /* case TGSI_OPCODE_SHL: return RC_OPCODE_SHL; */
108 /* case TGSI_OPCODE_ISHR: return RC_OPCODE_SHR; */
109 /* case TGSI_OPCODE_AND: return RC_OPCODE_AND; */
110 /* case TGSI_OPCODE_OR: return RC_OPCODE_OR; */
111 /* case TGSI_OPCODE_MOD: return RC_OPCODE_MOD; */
112 /* case TGSI_OPCODE_XOR: return RC_OPCODE_XOR; */
113 /* case TGSI_OPCODE_SAD: return RC_OPCODE_SAD; */
114 /* case TGSI_OPCODE_TXF: return RC_OPCODE_TXF; */
115 /* case TGSI_OPCODE_TXQ: return RC_OPCODE_TXQ; */
116 case TGSI_OPCODE_CONT: return RC_OPCODE_CONT;
117 /* case TGSI_OPCODE_EMIT: return RC_OPCODE_EMIT; */
118 /* case TGSI_OPCODE_ENDPRIM: return RC_OPCODE_ENDPRIM; */
119 /* case TGSI_OPCODE_BGNLOOP2: return RC_OPCODE_BGNLOOP2; */
120 /* case TGSI_OPCODE_BGNSUB: return RC_OPCODE_BGNSUB; */
121 /* case TGSI_OPCODE_ENDLOOP2: return RC_OPCODE_ENDLOOP2; */
122 /* case TGSI_OPCODE_ENDSUB: return RC_OPCODE_ENDSUB; */
123 case TGSI_OPCODE_NOP: return RC_OPCODE_NOP;
124 /* case TGSI_OPCODE_CALLNZ: return RC_OPCODE_CALLNZ; */
125 /* case TGSI_OPCODE_BREAKC: return RC_OPCODE_BREAKC; */
126 case TGSI_OPCODE_KILL_IF: return RC_OPCODE_KIL;
127 }
128
129 fprintf(stderr, "r300: Unknown TGSI/RC opcode: %s\n", tgsi_get_opcode_name(opcode));
130 return RC_OPCODE_ILLEGAL_OPCODE;
131 }
132
133 static unsigned translate_saturate(unsigned saturate)
134 {
135 switch(saturate) {
136 default:
137 fprintf(stderr, "Unknown saturate mode: %i\n", saturate);
138 /* fall-through */
139 case TGSI_SAT_NONE: return RC_SATURATE_NONE;
140 case TGSI_SAT_ZERO_ONE: return RC_SATURATE_ZERO_ONE;
141 }
142 }
143
144 static unsigned translate_register_file(unsigned file)
145 {
146 switch(file) {
147 case TGSI_FILE_CONSTANT: return RC_FILE_CONSTANT;
148 case TGSI_FILE_IMMEDIATE: return RC_FILE_CONSTANT;
149 case TGSI_FILE_INPUT: return RC_FILE_INPUT;
150 case TGSI_FILE_OUTPUT: return RC_FILE_OUTPUT;
151 default:
152 fprintf(stderr, "Unhandled register file: %i\n", file);
153 /* fall-through */
154 case TGSI_FILE_TEMPORARY: return RC_FILE_TEMPORARY;
155 case TGSI_FILE_ADDRESS: return RC_FILE_ADDRESS;
156 }
157 }
158
159 static int translate_register_index(
160 struct tgsi_to_rc * ttr,
161 unsigned file,
162 int index)
163 {
164 if (file == TGSI_FILE_IMMEDIATE)
165 return ttr->immediate_offset + index;
166
167 return index;
168 }
169
170 static void transform_dstreg(
171 struct tgsi_to_rc * ttr,
172 struct rc_dst_register * dst,
173 struct tgsi_full_dst_register * src)
174 {
175 dst->File = translate_register_file(src->Register.File);
176 dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);
177 dst->WriteMask = src->Register.WriteMask;
178
179 if (src->Register.Indirect) {
180 ttr->error = TRUE;
181 fprintf(stderr, "r300: Relative addressing of destination operands "
182 "is unsupported.\n");
183 }
184 }
185
186 static void transform_srcreg(
187 struct tgsi_to_rc * ttr,
188 struct rc_src_register * dst,
189 struct tgsi_full_src_register * src)
190 {
191 unsigned i, j;
192
193 dst->File = translate_register_file(src->Register.File);
194 dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);
195 dst->RelAddr = src->Register.Indirect;
196 dst->Swizzle = tgsi_util_get_full_src_register_swizzle(src, 0);
197 dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 1) << 3;
198 dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 2) << 6;
199 dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 3) << 9;
200 dst->Abs = src->Register.Absolute;
201 dst->Negate = src->Register.Negate ? RC_MASK_XYZW : 0;
202
203 if (src->Register.File == TGSI_FILE_IMMEDIATE) {
204 for (i = 0; i < ttr->imms_to_swizzle_count; i++) {
205 if (ttr->imms_to_swizzle[i].index == src->Register.Index) {
206 dst->File = RC_FILE_TEMPORARY;
207 dst->Index = 0;
208 dst->Swizzle = 0;
209 for (j = 0; j < 4; j++) {
210 dst->Swizzle |= GET_SWZ(ttr->imms_to_swizzle[i].swizzle,
211 tgsi_util_get_full_src_register_swizzle(src, j)) << (j * 3);
212 }
213 break;
214 }
215 }
216 }
217 }
218
219 static void transform_texture(struct rc_instruction * dst, struct tgsi_instruction_texture src,
220 uint32_t *shadowSamplers)
221 {
222 switch(src.Texture) {
223 case TGSI_TEXTURE_1D:
224 dst->U.I.TexSrcTarget = RC_TEXTURE_1D;
225 break;
226 case TGSI_TEXTURE_2D:
227 dst->U.I.TexSrcTarget = RC_TEXTURE_2D;
228 break;
229 case TGSI_TEXTURE_3D:
230 dst->U.I.TexSrcTarget = RC_TEXTURE_3D;
231 break;
232 case TGSI_TEXTURE_CUBE:
233 dst->U.I.TexSrcTarget = RC_TEXTURE_CUBE;
234 break;
235 case TGSI_TEXTURE_RECT:
236 dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;
237 break;
238 case TGSI_TEXTURE_SHADOW1D:
239 dst->U.I.TexSrcTarget = RC_TEXTURE_1D;
240 dst->U.I.TexShadow = 1;
241 *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
242 break;
243 case TGSI_TEXTURE_SHADOW2D:
244 dst->U.I.TexSrcTarget = RC_TEXTURE_2D;
245 dst->U.I.TexShadow = 1;
246 *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
247 break;
248 case TGSI_TEXTURE_SHADOWRECT:
249 dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;
250 dst->U.I.TexShadow = 1;
251 *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
252 break;
253 }
254 dst->U.I.TexSwizzle = RC_SWIZZLE_XYZW;
255 }
256
257 static void transform_instruction(struct tgsi_to_rc * ttr, struct tgsi_full_instruction * src)
258 {
259 struct rc_instruction * dst;
260 int i;
261
262 dst = rc_insert_new_instruction(ttr->compiler, ttr->compiler->Program.Instructions.Prev);
263 dst->U.I.Opcode = translate_opcode(src->Instruction.Opcode);
264 dst->U.I.SaturateMode = translate_saturate(src->Instruction.Saturate);
265
266 if (src->Instruction.NumDstRegs)
267 transform_dstreg(ttr, &dst->U.I.DstReg, &src->Dst[0]);
268
269 for(i = 0; i < src->Instruction.NumSrcRegs; ++i) {
270 if (src->Src[i].Register.File == TGSI_FILE_SAMPLER)
271 dst->U.I.TexSrcUnit = src->Src[i].Register.Index;
272 else
273 transform_srcreg(ttr, &dst->U.I.SrcReg[i], &src->Src[i]);
274 }
275
276 /* Texturing. */
277 if (src->Instruction.Texture)
278 transform_texture(dst, src->Texture,
279 &ttr->compiler->Program.ShadowSamplers);
280 }
281
282 static void handle_immediate(struct tgsi_to_rc * ttr,
283 struct tgsi_full_immediate * imm,
284 unsigned index)
285 {
286 struct rc_constant constant;
287 unsigned swizzle = 0;
288 boolean can_swizzle = TRUE;
289 unsigned i;
290
291 for (i = 0; i < 4; i++) {
292 if (imm->u[i].Float == 0.0f) {
293 swizzle |= RC_SWIZZLE_ZERO << (i * 3);
294 } else if (imm->u[i].Float == 0.5f && ttr->use_half_swizzles) {
295 swizzle |= RC_SWIZZLE_HALF << (i * 3);
296 } else if (imm->u[i].Float == 1.0f) {
297 swizzle |= RC_SWIZZLE_ONE << (i * 3);
298 } else {
299 can_swizzle = FALSE;
300 break;
301 }
302 }
303
304 if (can_swizzle) {
305 ttr->imms_to_swizzle[ttr->imms_to_swizzle_count].index = index;
306 ttr->imms_to_swizzle[ttr->imms_to_swizzle_count].swizzle = swizzle;
307 ttr->imms_to_swizzle_count++;
308 } else {
309 constant.Type = RC_CONSTANT_IMMEDIATE;
310 constant.Size = 4;
311 for(i = 0; i < 4; ++i)
312 constant.u.Immediate[i] = imm->u[i].Float;
313 rc_constants_add(&ttr->compiler->Program.Constants, &constant);
314 }
315 }
316
317 void r300_tgsi_to_rc(struct tgsi_to_rc * ttr,
318 const struct tgsi_token * tokens)
319 {
320 struct tgsi_full_instruction *inst;
321 struct tgsi_parse_context parser;
322 unsigned imm_index = 0;
323 int i;
324
325 ttr->error = FALSE;
326
327 /* Allocate constants placeholders.
328 *
329 * Note: What if declared constants are not contiguous? */
330 for(i = 0; i <= ttr->info->file_max[TGSI_FILE_CONSTANT]; ++i) {
331 struct rc_constant constant;
332 memset(&constant, 0, sizeof(constant));
333 constant.Type = RC_CONSTANT_EXTERNAL;
334 constant.Size = 4;
335 constant.u.External = i;
336 rc_constants_add(&ttr->compiler->Program.Constants, &constant);
337 }
338
339 ttr->immediate_offset = ttr->compiler->Program.Constants.Count;
340
341 ttr->imms_to_swizzle = malloc(ttr->info->immediate_count * sizeof(struct swizzled_imms));
342 ttr->imms_to_swizzle_count = 0;
343
344 tgsi_parse_init(&parser, tokens);
345
346 while (!tgsi_parse_end_of_tokens(&parser)) {
347 tgsi_parse_token(&parser);
348
349 switch (parser.FullToken.Token.Type) {
350 case TGSI_TOKEN_TYPE_DECLARATION:
351 break;
352 case TGSI_TOKEN_TYPE_IMMEDIATE:
353 handle_immediate(ttr, &parser.FullToken.FullImmediate, imm_index);
354 imm_index++;
355 break;
356 case TGSI_TOKEN_TYPE_INSTRUCTION:
357 inst = &parser.FullToken.FullInstruction;
358 if (inst->Instruction.Opcode == TGSI_OPCODE_END) {
359 break;
360 }
361
362 transform_instruction(ttr, inst);
363 break;
364 }
365 }
366
367 tgsi_parse_free(&parser);
368
369 free(ttr->imms_to_swizzle);
370
371 rc_calculate_inputs_outputs(ttr->compiler);
372 }