r300: Zero-initialize register for NV_vertex_program
[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 "radeon_compiler.h"
26 #include "radeon_program.h"
27
28 #include "tgsi/tgsi_parse.h"
29 #include "tgsi/tgsi_scan.h"
30 #include "tgsi/tgsi_util.h"
31
32
33 static unsigned translate_opcode(unsigned opcode)
34 {
35 switch(opcode) {
36 case TGSI_OPCODE_ARL: return OPCODE_ARL;
37 case TGSI_OPCODE_MOV: return OPCODE_MOV;
38 case TGSI_OPCODE_LIT: return OPCODE_LIT;
39 case TGSI_OPCODE_RCP: return OPCODE_RCP;
40 case TGSI_OPCODE_RSQ: return OPCODE_RSQ;
41 case TGSI_OPCODE_EXP: return OPCODE_EXP;
42 case TGSI_OPCODE_LOG: return OPCODE_LOG;
43 case TGSI_OPCODE_MUL: return OPCODE_MUL;
44 case TGSI_OPCODE_ADD: return OPCODE_ADD;
45 case TGSI_OPCODE_DP3: return OPCODE_DP3;
46 case TGSI_OPCODE_DP4: return OPCODE_DP4;
47 case TGSI_OPCODE_DST: return OPCODE_DST;
48 case TGSI_OPCODE_MIN: return OPCODE_MIN;
49 case TGSI_OPCODE_MAX: return OPCODE_MAX;
50 case TGSI_OPCODE_SLT: return OPCODE_SLT;
51 case TGSI_OPCODE_SGE: return OPCODE_SGE;
52 case TGSI_OPCODE_MAD: return OPCODE_MAD;
53 case TGSI_OPCODE_SUB: return OPCODE_SUB;
54 case TGSI_OPCODE_LRP: return OPCODE_LRP;
55 /* case TGSI_OPCODE_CND: return OPCODE_CND; */
56 case TGSI_OPCODE_DP2A: return OPCODE_DP2A;
57 /* gap */
58 case TGSI_OPCODE_FRC: return OPCODE_FRC;
59 /* case TGSI_OPCODE_CLAMP: return OPCODE_CLAMP; */
60 case TGSI_OPCODE_FLR: return OPCODE_FLR;
61 /* case TGSI_OPCODE_ROUND: return OPCODE_ROUND; */
62 case TGSI_OPCODE_EX2: return OPCODE_EX2;
63 case TGSI_OPCODE_LG2: return OPCODE_LG2;
64 case TGSI_OPCODE_POW: return OPCODE_POW;
65 case TGSI_OPCODE_XPD: return OPCODE_XPD;
66 /* gap */
67 case TGSI_OPCODE_ABS: return OPCODE_ABS;
68 case TGSI_OPCODE_RCC: return OPCODE_RCC;
69 case TGSI_OPCODE_DPH: return OPCODE_DPH;
70 case TGSI_OPCODE_COS: return OPCODE_COS;
71 case TGSI_OPCODE_DDX: return OPCODE_DDX;
72 case TGSI_OPCODE_DDY: return OPCODE_DDY;
73 /* case TGSI_OPCODE_KILP: return OPCODE_KILP; */
74 case TGSI_OPCODE_PK2H: return OPCODE_PK2H;
75 case TGSI_OPCODE_PK2US: return OPCODE_PK2US;
76 case TGSI_OPCODE_PK4B: return OPCODE_PK4B;
77 case TGSI_OPCODE_PK4UB: return OPCODE_PK4UB;
78 case TGSI_OPCODE_RFL: return OPCODE_RFL;
79 case TGSI_OPCODE_SEQ: return OPCODE_SEQ;
80 case TGSI_OPCODE_SFL: return OPCODE_SFL;
81 case TGSI_OPCODE_SGT: return OPCODE_SGT;
82 case TGSI_OPCODE_SIN: return OPCODE_SIN;
83 case TGSI_OPCODE_SLE: return OPCODE_SLE;
84 case TGSI_OPCODE_SNE: return OPCODE_SNE;
85 case TGSI_OPCODE_STR: return OPCODE_STR;
86 case TGSI_OPCODE_TEX: return OPCODE_TEX;
87 case TGSI_OPCODE_TXD: return OPCODE_TXD;
88 case TGSI_OPCODE_TXP: return OPCODE_TXP;
89 case TGSI_OPCODE_UP2H: return OPCODE_UP2H;
90 case TGSI_OPCODE_UP2US: return OPCODE_UP2US;
91 case TGSI_OPCODE_UP4B: return OPCODE_UP4B;
92 case TGSI_OPCODE_UP4UB: return OPCODE_UP4UB;
93 case TGSI_OPCODE_X2D: return OPCODE_X2D;
94 case TGSI_OPCODE_ARA: return OPCODE_ARA;
95 case TGSI_OPCODE_ARR: return OPCODE_ARR;
96 case TGSI_OPCODE_BRA: return OPCODE_BRA;
97 case TGSI_OPCODE_CAL: return OPCODE_CAL;
98 case TGSI_OPCODE_RET: return OPCODE_RET;
99 case TGSI_OPCODE_SSG: return OPCODE_SSG;
100 case TGSI_OPCODE_CMP: return OPCODE_CMP;
101 case TGSI_OPCODE_SCS: return OPCODE_SCS;
102 case TGSI_OPCODE_TXB: return OPCODE_TXB;
103 /* case TGSI_OPCODE_NRM: return OPCODE_NRM; */
104 /* case TGSI_OPCODE_DIV: return OPCODE_DIV; */
105 case TGSI_OPCODE_DP2: return OPCODE_DP2;
106 case TGSI_OPCODE_TXL: return OPCODE_TXL;
107 case TGSI_OPCODE_BRK: return OPCODE_BRK;
108 case TGSI_OPCODE_IF: return OPCODE_IF;
109 /* case TGSI_OPCODE_LOOP: return OPCODE_LOOP; */
110 /* case TGSI_OPCODE_REP: return OPCODE_REP; */
111 case TGSI_OPCODE_ELSE: return OPCODE_ELSE;
112 case TGSI_OPCODE_ENDIF: return OPCODE_ENDIF;
113 case TGSI_OPCODE_ENDLOOP: return OPCODE_ENDLOOP;
114 /* case TGSI_OPCODE_ENDREP: return OPCODE_ENDREP; */
115 case TGSI_OPCODE_PUSHA: return OPCODE_PUSHA;
116 case TGSI_OPCODE_POPA: return OPCODE_POPA;
117 /* case TGSI_OPCODE_CEIL: return OPCODE_CEIL; */
118 /* case TGSI_OPCODE_I2F: return OPCODE_I2F; */
119 case TGSI_OPCODE_NOT: return OPCODE_NOT;
120 case TGSI_OPCODE_TRUNC: return OPCODE_TRUNC;
121 /* case TGSI_OPCODE_SHL: return OPCODE_SHL; */
122 /* case TGSI_OPCODE_SHR: return OPCODE_SHR; */
123 case TGSI_OPCODE_AND: return OPCODE_AND;
124 case TGSI_OPCODE_OR: return OPCODE_OR;
125 /* case TGSI_OPCODE_MOD: return OPCODE_MOD; */
126 case TGSI_OPCODE_XOR: return OPCODE_XOR;
127 /* case TGSI_OPCODE_SAD: return OPCODE_SAD; */
128 /* case TGSI_OPCODE_TXF: return OPCODE_TXF; */
129 /* case TGSI_OPCODE_TXQ: return OPCODE_TXQ; */
130 case TGSI_OPCODE_CONT: return OPCODE_CONT;
131 /* case TGSI_OPCODE_EMIT: return OPCODE_EMIT; */
132 /* case TGSI_OPCODE_ENDPRIM: return OPCODE_ENDPRIM; */
133 /* case TGSI_OPCODE_BGNLOOP2: return OPCODE_BGNLOOP2; */
134 case TGSI_OPCODE_BGNSUB: return OPCODE_BGNSUB;
135 /* case TGSI_OPCODE_ENDLOOP2: return OPCODE_ENDLOOP2; */
136 case TGSI_OPCODE_ENDSUB: return OPCODE_ENDSUB;
137 case TGSI_OPCODE_NOISE1: return OPCODE_NOISE1;
138 case TGSI_OPCODE_NOISE2: return OPCODE_NOISE2;
139 case TGSI_OPCODE_NOISE3: return OPCODE_NOISE3;
140 case TGSI_OPCODE_NOISE4: return OPCODE_NOISE4;
141 case TGSI_OPCODE_NOP: return OPCODE_NOP;
142 /* gap */
143 case TGSI_OPCODE_NRM4: return OPCODE_NRM4;
144 /* case TGSI_OPCODE_CALLNZ: return OPCODE_CALLNZ; */
145 /* case TGSI_OPCODE_IFC: return OPCODE_IFC; */
146 /* case TGSI_OPCODE_BREAKC: return OPCODE_BREAKC; */
147 case TGSI_OPCODE_KIL: return OPCODE_KIL;
148 case TGSI_OPCODE_END: return OPCODE_END;
149 case TGSI_OPCODE_SWZ: return OPCODE_SWZ;
150 }
151
152 fprintf(stderr, "Unknown opcode: %i\n", opcode);
153 abort();
154 }
155
156 static unsigned translate_saturate(unsigned saturate)
157 {
158 switch(saturate) {
159 case TGSI_SAT_NONE: return SATURATE_OFF;
160 case TGSI_SAT_ZERO_ONE: return SATURATE_ZERO_ONE;
161 case TGSI_SAT_MINUS_PLUS_ONE: return SATURATE_PLUS_MINUS_ONE;
162 }
163
164 fprintf(stderr, "Unknown saturate mode: %i\n", saturate);
165 abort();
166 }
167
168 static unsigned translate_register_file(unsigned file)
169 {
170 switch(file) {
171 case TGSI_FILE_CONSTANT: return PROGRAM_CONSTANT;
172 case TGSI_FILE_IMMEDIATE: return PROGRAM_CONSTANT;
173 case TGSI_FILE_INPUT: return PROGRAM_INPUT;
174 case TGSI_FILE_OUTPUT: return PROGRAM_OUTPUT;
175 case TGSI_FILE_TEMPORARY: return PROGRAM_TEMPORARY;
176 case TGSI_FILE_ADDRESS: return PROGRAM_ADDRESS;
177 }
178
179 fprintf(stderr, "Unhandled register file: %i\n", file);
180 abort();
181 }
182
183 static int translate_register_index(
184 struct tgsi_to_rc * ttr,
185 unsigned file,
186 int index)
187 {
188 if (file == TGSI_FILE_IMMEDIATE)
189 return ttr->immediate_offset + index;
190
191 return index;
192 }
193
194 static void transform_dstreg(
195 struct tgsi_to_rc * ttr,
196 struct prog_dst_register * dst,
197 struct tgsi_full_dst_register * src)
198 {
199 dst->File = translate_register_file(src->DstRegister.File);
200 dst->Index = translate_register_index(ttr, src->DstRegister.File, src->DstRegister.Index);
201 dst->WriteMask = src->DstRegister.WriteMask;
202 dst->RelAddr = src->DstRegister.Indirect;
203 }
204
205 static void transform_srcreg(
206 struct tgsi_to_rc * ttr,
207 struct prog_src_register * dst,
208 struct tgsi_full_src_register * src)
209 {
210 dst->File = translate_register_file(src->SrcRegister.File);
211 dst->Index = translate_register_index(ttr, src->SrcRegister.File, src->SrcRegister.Index);
212 dst->RelAddr = src->SrcRegister.Indirect;
213 dst->Swizzle = tgsi_util_get_full_src_register_extswizzle(src, 0);
214 dst->Swizzle |= tgsi_util_get_full_src_register_extswizzle(src, 1) << 3;
215 dst->Swizzle |= tgsi_util_get_full_src_register_extswizzle(src, 2) << 6;
216 dst->Swizzle |= tgsi_util_get_full_src_register_extswizzle(src, 3) << 9;
217 dst->Abs = src->SrcRegisterExtMod.Absolute;
218 dst->Negate =
219 src->SrcRegisterExtSwz.NegateX |
220 (src->SrcRegisterExtSwz.NegateY << 1) |
221 (src->SrcRegisterExtSwz.NegateZ << 2) |
222 (src->SrcRegisterExtSwz.NegateW << 3);
223 dst->Negate ^= src->SrcRegister.Negate ? NEGATE_XYZW : 0;
224 }
225
226 static void transform_texture(struct rc_instruction * dst, struct tgsi_instruction_ext_texture src)
227 {
228 switch(src.Texture) {
229 case TGSI_TEXTURE_1D:
230 dst->I.TexSrcTarget = TEXTURE_1D_INDEX;
231 break;
232 case TGSI_TEXTURE_2D:
233 dst->I.TexSrcTarget = TEXTURE_2D_INDEX;
234 break;
235 case TGSI_TEXTURE_3D:
236 dst->I.TexSrcTarget = TEXTURE_3D_INDEX;
237 break;
238 case TGSI_TEXTURE_CUBE:
239 dst->I.TexSrcTarget = TEXTURE_CUBE_INDEX;
240 break;
241 case TGSI_TEXTURE_RECT:
242 dst->I.TexSrcTarget = TEXTURE_RECT_INDEX;
243 break;
244 case TGSI_TEXTURE_SHADOW1D:
245 dst->I.TexSrcTarget = TEXTURE_1D_INDEX;
246 dst->I.TexShadow = 1;
247 break;
248 case TGSI_TEXTURE_SHADOW2D:
249 dst->I.TexSrcTarget = TEXTURE_2D_INDEX;
250 dst->I.TexShadow = 1;
251 break;
252 case TGSI_TEXTURE_SHADOWRECT:
253 dst->I.TexSrcTarget = TEXTURE_RECT_INDEX;
254 dst->I.TexShadow = 1;
255 break;
256 }
257 }
258
259 static void transform_instruction(struct tgsi_to_rc * ttr, struct tgsi_full_instruction * src)
260 {
261 if (src->Instruction.Opcode == TGSI_OPCODE_END)
262 return;
263
264 struct rc_instruction * dst = rc_insert_new_instruction(ttr->compiler, ttr->compiler->Program.Instructions.Prev);
265 int i;
266
267 dst->I.Opcode = translate_opcode(src->Instruction.Opcode);
268 dst->I.SaturateMode = translate_saturate(src->Instruction.Saturate);
269
270 if (src->Instruction.NumDstRegs)
271 transform_dstreg(ttr, &dst->I.DstReg, &src->FullDstRegisters[0]);
272
273 for(i = 0; i < src->Instruction.NumSrcRegs; ++i) {
274 if (src->FullSrcRegisters[i].SrcRegister.File == TGSI_FILE_SAMPLER)
275 dst->I.TexSrcUnit = src->FullSrcRegisters[i].SrcRegister.Index;
276 else
277 transform_srcreg(ttr, &dst->I.SrcReg[i], &src->FullSrcRegisters[i]);
278 }
279
280 /* Texturing. */
281 transform_texture(dst, src->InstructionExtTexture);
282 }
283
284 static void handle_immediate(struct tgsi_to_rc * ttr, struct tgsi_full_immediate * imm)
285 {
286 struct rc_constant constant;
287 int i;
288
289 constant.Type = RC_CONSTANT_IMMEDIATE;
290 constant.Size = 4;
291 for(i = 0; i < 4; ++i)
292 constant.u.Immediate[i] = imm->u[i].Float;
293 rc_constants_add(&ttr->compiler->Program.Constants, &constant);
294 }
295
296 void r300_tgsi_to_rc(struct tgsi_to_rc * ttr, const struct tgsi_token * tokens)
297 {
298 struct tgsi_parse_context parser;
299 int i;
300
301 /* Allocate constants placeholders.
302 *
303 * Note: What if declared constants are not contiguous? */
304 for(i = 0; i <= ttr->info->file_max[TGSI_FILE_CONSTANT]; ++i) {
305 struct rc_constant constant;
306 memset(&constant, 0, sizeof(constant));
307 constant.Type = RC_CONSTANT_EXTERNAL;
308 constant.Size = 4;
309 constant.u.External = i;
310 rc_constants_add(&ttr->compiler->Program.Constants, &constant);
311 }
312
313 ttr->immediate_offset = ttr->compiler->Program.Constants.Count;
314
315 tgsi_parse_init(&parser, tokens);
316
317 while (!tgsi_parse_end_of_tokens(&parser)) {
318 tgsi_parse_token(&parser);
319
320 switch (parser.FullToken.Token.Type) {
321 case TGSI_TOKEN_TYPE_DECLARATION:
322 break;
323 case TGSI_TOKEN_TYPE_IMMEDIATE:
324 handle_immediate(ttr, &parser.FullToken.FullImmediate);
325 break;
326 case TGSI_TOKEN_TYPE_INSTRUCTION:
327 transform_instruction(ttr, &parser.FullToken.FullInstruction);
328 break;
329 }
330 }
331
332 tgsi_parse_free(&parser);
333
334 rc_calculate_inputs_outputs(ttr->compiler);
335 }
336