vc4_qir.h revision 64122b16ce74a3fb65269bab325c651c26ccd2d0
1/* 2 * Copyright © 2014 Broadcom 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 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * 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 NONINFRINGEMENT. IN NO EVENT SHALL 18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 21 * IN THE SOFTWARE. 22 */ 23 24#ifndef VC4_QIR_H 25#define VC4_QIR_H 26 27#include <stdio.h> 28#include <stdlib.h> 29#include <stdbool.h> 30#include <stdint.h> 31#include <string.h> 32 33#include "util/u_simple_list.h" 34#include "tgsi/tgsi_parse.h" 35 36enum qfile { 37 QFILE_NULL, 38 QFILE_TEMP, 39 QFILE_VARY, 40 QFILE_UNIF, 41}; 42 43struct qreg { 44 enum qfile file; 45 uint32_t index; 46}; 47 48enum qop { 49 QOP_UNDEF, 50 QOP_MOV, 51 QOP_FADD, 52 QOP_FSUB, 53 QOP_FMUL, 54 QOP_MUL24, 55 QOP_FMIN, 56 QOP_FMAX, 57 QOP_FMINABS, 58 QOP_FMAXABS, 59 QOP_ADD, 60 QOP_SUB, 61 QOP_SHL, 62 QOP_SHR, 63 QOP_ASR, 64 QOP_MIN, 65 QOP_MAX, 66 QOP_AND, 67 QOP_OR, 68 QOP_XOR, 69 QOP_NOT, 70 71 /* Sets the flag register according to src. */ 72 QOP_SF, 73 74 /* Note: Orderings of these compares must be the same as in 75 * qpu_defines.h. Selects the src[0] if the ns flag bit is set, 76 * otherwise 0. */ 77 QOP_SEL_X_0_ZS, 78 QOP_SEL_X_0_ZC, 79 QOP_SEL_X_0_NS, 80 QOP_SEL_X_0_NC, 81 /* Selects the src[0] if the ns flag bit is set, otherwise src[1]. */ 82 QOP_SEL_X_Y_ZS, 83 QOP_SEL_X_Y_ZC, 84 QOP_SEL_X_Y_NS, 85 QOP_SEL_X_Y_NC, 86 87 QOP_FTOI, 88 QOP_ITOF, 89 QOP_RCP, 90 QOP_RSQ, 91 QOP_EXP2, 92 QOP_LOG2, 93 QOP_VW_SETUP, 94 QOP_VR_SETUP, 95 QOP_PACK_SCALED, 96 QOP_PACK_COLORS, 97 QOP_VPM_WRITE, 98 QOP_VPM_READ, 99 QOP_TLB_DISCARD_SETUP, 100 QOP_TLB_STENCIL_SETUP, 101 QOP_TLB_Z_WRITE, 102 QOP_TLB_COLOR_WRITE, 103 QOP_TLB_COLOR_READ, 104 QOP_VARY_ADD_C, 105 106 QOP_FRAG_X, 107 QOP_FRAG_Y, 108 QOP_FRAG_Z, 109 QOP_FRAG_W, 110 111 QOP_UNPACK_8A, 112 QOP_UNPACK_8B, 113 QOP_UNPACK_8C, 114 QOP_UNPACK_8D, 115 116 /** Texture x coordinate parameter write */ 117 QOP_TEX_S, 118 /** Texture y coordinate parameter write */ 119 QOP_TEX_T, 120 /** Texture border color parameter or cube map z coordinate write */ 121 QOP_TEX_R, 122 /** Texture LOD bias parameter write */ 123 QOP_TEX_B, 124 /** 125 * Signal of texture read being necessary and then reading r4 into 126 * the destination 127 */ 128 QOP_TEX_RESULT, 129 QOP_R4_UNPACK_A, 130 QOP_R4_UNPACK_B, 131 QOP_R4_UNPACK_C, 132 QOP_R4_UNPACK_D 133}; 134 135struct simple_node { 136 struct simple_node *next; 137 struct simple_node *prev; 138}; 139 140struct qinst { 141 struct simple_node link; 142 143 enum qop op; 144 struct qreg dst; 145 struct qreg *src; 146}; 147 148enum qstage { 149 /** 150 * Coordinate shader, runs during binning, before the VS, and just 151 * outputs position. 152 */ 153 QSTAGE_COORD, 154 QSTAGE_VERT, 155 QSTAGE_FRAG, 156}; 157 158enum quniform_contents { 159 /** 160 * Indicates that a constant 32-bit value is copied from the program's 161 * uniform contents. 162 */ 163 QUNIFORM_CONSTANT, 164 /** 165 * Indicates that the program's uniform contents are used as an index 166 * into the GL uniform storage. 167 */ 168 QUNIFORM_UNIFORM, 169 170 /** @{ 171 * Scaling factors from clip coordinates to relative to the viewport 172 * center. 173 * 174 * This is used by the coordinate and vertex shaders to produce the 175 * 32-bit entry consisting of 2 16-bit fields with 12.4 signed fixed 176 * point offsets from the viewport ccenter. 177 */ 178 QUNIFORM_VIEWPORT_X_SCALE, 179 QUNIFORM_VIEWPORT_Y_SCALE, 180 /** @} */ 181 182 QUNIFORM_VIEWPORT_Z_OFFSET, 183 QUNIFORM_VIEWPORT_Z_SCALE, 184 185 /** 186 * A reference to a texture config parameter 0 uniform. 187 * 188 * This is a uniform implicitly loaded with a QPU_W_TMU* write, which 189 * defines texture type, miplevels, and such. It will be found as a 190 * parameter to the first QOP_TEX_[STRB] instruction in a sequence. 191 */ 192 QUNIFORM_TEXTURE_CONFIG_P0, 193 194 /** 195 * A reference to a texture config parameter 1 uniform. 196 * 197 * This is a uniform implicitly loaded with a QPU_W_TMU* write, which 198 * defines texture width, height, filters, and wrap modes. It will be 199 * found as a parameter to the second QOP_TEX_[STRB] instruction in a 200 * sequence. 201 */ 202 QUNIFORM_TEXTURE_CONFIG_P1, 203 204 /** A reference to a texture config parameter 2 cubemap stride uniform */ 205 QUNIFORM_TEXTURE_CONFIG_P2, 206 207 QUNIFORM_TEXRECT_SCALE_X, 208 QUNIFORM_TEXRECT_SCALE_Y, 209 210 QUNIFORM_BLEND_CONST_COLOR, 211 QUNIFORM_STENCIL, 212 213 QUNIFORM_ALPHA_REF, 214}; 215 216struct vc4_compile { 217 struct tgsi_parse_context parser; 218 struct qreg *temps; 219 struct qreg *inputs; 220 struct qreg *outputs; 221 struct qreg *consts; 222 uint32_t temps_array_size; 223 uint32_t inputs_array_size; 224 uint32_t outputs_array_size; 225 uint32_t uniforms_array_size; 226 uint32_t consts_array_size; 227 uint32_t num_consts; 228 struct qreg line_x, point_x, point_y; 229 struct qreg discard; 230 231 struct pipe_shader_state *shader_state; 232 struct vc4_key *key; 233 struct vc4_fs_key *fs_key; 234 struct vc4_vs_key *vs_key; 235 236 uint32_t *uniform_data; 237 enum quniform_contents *uniform_contents; 238 uint32_t num_uniforms; 239 uint32_t num_outputs; 240 uint32_t num_texture_samples; 241 uint32_t output_position_index; 242 uint32_t output_color_index; 243 uint32_t output_point_size_index; 244 245 struct qreg undef; 246 enum qstage stage; 247 uint32_t num_temps; 248 struct simple_node instructions; 249 uint32_t immediates[1024]; 250 251 struct simple_node qpu_inst_list; 252 uint64_t *qpu_insts; 253 uint32_t qpu_inst_count; 254 uint32_t qpu_inst_size; 255 uint32_t num_inputs; 256 uint32_t color_inputs; 257}; 258 259struct vc4_compile *qir_compile_init(void); 260void qir_compile_destroy(struct vc4_compile *c); 261struct qinst *qir_inst(enum qop op, struct qreg dst, 262 struct qreg src0, struct qreg src1); 263struct qinst *qir_inst4(enum qop op, struct qreg dst, 264 struct qreg a, 265 struct qreg b, 266 struct qreg c, 267 struct qreg d); 268void qir_remove_instruction(struct qinst *qinst); 269void qir_reorder_uniforms(struct vc4_compile *c); 270void qir_emit(struct vc4_compile *c, struct qinst *inst); 271struct qreg qir_get_temp(struct vc4_compile *c); 272int qir_get_op_nsrc(enum qop qop); 273bool qir_reg_equals(struct qreg a, struct qreg b); 274bool qir_has_side_effects(struct qinst *inst); 275bool qir_depends_on_flags(struct qinst *inst); 276bool qir_writes_r4(struct qinst *inst); 277bool qir_reads_r4(struct qinst *inst); 278 279void qir_dump(struct vc4_compile *c); 280void qir_dump_inst(struct vc4_compile *c, struct qinst *inst); 281const char *qir_get_stage_name(enum qstage stage); 282 283void qir_optimize(struct vc4_compile *c); 284bool qir_opt_algebraic(struct vc4_compile *c); 285bool qir_opt_copy_propagation(struct vc4_compile *c); 286bool qir_opt_cse(struct vc4_compile *c); 287bool qir_opt_dead_code(struct vc4_compile *c); 288 289#define QIR_ALU0(name) \ 290static inline struct qreg \ 291qir_##name(struct vc4_compile *c) \ 292{ \ 293 struct qreg t = qir_get_temp(c); \ 294 qir_emit(c, qir_inst(QOP_##name, t, c->undef, c->undef)); \ 295 return t; \ 296} 297 298#define QIR_ALU1(name) \ 299static inline struct qreg \ 300qir_##name(struct vc4_compile *c, struct qreg a) \ 301{ \ 302 struct qreg t = qir_get_temp(c); \ 303 qir_emit(c, qir_inst(QOP_##name, t, a, c->undef)); \ 304 return t; \ 305} 306 307#define QIR_ALU2(name) \ 308static inline struct qreg \ 309qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b) \ 310{ \ 311 struct qreg t = qir_get_temp(c); \ 312 qir_emit(c, qir_inst(QOP_##name, t, a, b)); \ 313 return t; \ 314} 315 316#define QIR_NODST_1(name) \ 317static inline void \ 318qir_##name(struct vc4_compile *c, struct qreg a) \ 319{ \ 320 qir_emit(c, qir_inst(QOP_##name, c->undef, a, c->undef)); \ 321} 322 323#define QIR_NODST_2(name) \ 324static inline void \ 325qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b) \ 326{ \ 327 qir_emit(c, qir_inst(QOP_##name, c->undef, a, b)); \ 328} 329 330QIR_ALU1(MOV) 331QIR_ALU2(FADD) 332QIR_ALU2(FSUB) 333QIR_ALU2(FMUL) 334QIR_ALU2(MUL24) 335QIR_NODST_1(SF) 336QIR_ALU1(SEL_X_0_ZS) 337QIR_ALU1(SEL_X_0_ZC) 338QIR_ALU1(SEL_X_0_NS) 339QIR_ALU1(SEL_X_0_NC) 340QIR_ALU2(SEL_X_Y_ZS) 341QIR_ALU2(SEL_X_Y_ZC) 342QIR_ALU2(SEL_X_Y_NS) 343QIR_ALU2(SEL_X_Y_NC) 344QIR_ALU2(FMIN) 345QIR_ALU2(FMAX) 346QIR_ALU2(FMINABS) 347QIR_ALU2(FMAXABS) 348QIR_ALU1(FTOI) 349QIR_ALU1(ITOF) 350 351QIR_ALU2(ADD) 352QIR_ALU2(SUB) 353QIR_ALU2(SHL) 354QIR_ALU2(SHR) 355QIR_ALU2(ASR) 356QIR_ALU2(MIN) 357QIR_ALU2(MAX) 358QIR_ALU2(AND) 359QIR_ALU2(OR) 360QIR_ALU2(XOR) 361QIR_ALU1(NOT) 362 363QIR_ALU1(RCP) 364QIR_ALU1(RSQ) 365QIR_ALU1(EXP2) 366QIR_ALU1(LOG2) 367QIR_ALU2(PACK_SCALED) 368QIR_ALU1(VARY_ADD_C) 369QIR_NODST_1(VPM_WRITE) 370QIR_NODST_2(TEX_S) 371QIR_NODST_2(TEX_T) 372QIR_NODST_2(TEX_R) 373QIR_NODST_2(TEX_B) 374QIR_ALU0(FRAG_X) 375QIR_ALU0(FRAG_Y) 376QIR_ALU0(FRAG_Z) 377QIR_ALU0(FRAG_W) 378QIR_ALU0(TEX_RESULT) 379QIR_ALU0(TLB_COLOR_READ) 380QIR_NODST_1(TLB_Z_WRITE) 381QIR_NODST_1(TLB_DISCARD_SETUP) 382QIR_NODST_1(TLB_STENCIL_SETUP) 383 384static inline struct qreg 385qir_R4_UNPACK(struct vc4_compile *c, struct qreg r4, int i) 386{ 387 struct qreg t = qir_get_temp(c); 388 qir_emit(c, qir_inst(QOP_R4_UNPACK_A + i, t, r4, c->undef)); 389 return t; 390} 391 392static inline struct qreg 393qir_SEL_X_0_COND(struct vc4_compile *c, int i) 394{ 395 struct qreg t = qir_get_temp(c); 396 qir_emit(c, qir_inst(QOP_R4_UNPACK_A + i, t, c->undef, c->undef)); 397 return t; 398} 399 400static inline struct qreg 401qir_UNPACK_8(struct vc4_compile *c, struct qreg src, int i) 402{ 403 struct qreg t = qir_get_temp(c); 404 qir_emit(c, qir_inst(QOP_UNPACK_8A + i, t, src, c->undef)); 405 return t; 406} 407 408#endif /* VC4_QIR_H */ 409