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