1 /* 2 Copyright (C) Intel Corp. 2006. All Rights Reserved. 3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to 4 develop this 3D driver. 5 6 Permission is hereby granted, free of charge, to any person obtaining 7 a copy of this software and associated documentation files (the 8 "Software"), to deal in the Software without restriction, including 9 without limitation the rights to use, copy, modify, merge, publish, 10 distribute, sublicense, and/or sell copies of the Software, and to 11 permit persons to whom the Software is furnished to do so, subject to 12 the following conditions: 13 14 The above copyright notice and this permission notice (including the 15 next paragraph) shall be included in all copies or substantial 16 portions of the Software. 17 18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. 21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE 22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 25 26 **********************************************************************/ 27 /* 28 * Authors: 29 * Keith Whitwell <keith@tungstengraphics.com> 30 */ 31 32 33 #ifndef BRW_WM_H 34 #define BRW_WM_H 35 36 #include <stdbool.h> 37 38 #include "program/prog_instruction.h" 39 #include "brw_context.h" 40 #include "brw_eu.h" 41 #include "brw_program.h" 42 43 #define SATURATE (1<<5) 44 45 /* A big lookup table is used to figure out which and how many 46 * additional regs will inserted before the main payload in the WM 47 * program execution. These mainly relate to depth and stencil 48 * processing and the early-depth-test optimization. 49 */ 50 #define IZ_PS_KILL_ALPHATEST_BIT 0x1 51 #define IZ_PS_COMPUTES_DEPTH_BIT 0x2 52 #define IZ_DEPTH_WRITE_ENABLE_BIT 0x4 53 #define IZ_DEPTH_TEST_ENABLE_BIT 0x8 54 #define IZ_STENCIL_WRITE_ENABLE_BIT 0x10 55 #define IZ_STENCIL_TEST_ENABLE_BIT 0x20 56 #define IZ_BIT_MAX 0x40 57 58 #define AA_NEVER 0 59 #define AA_SOMETIMES 1 60 #define AA_ALWAYS 2 61 62 struct brw_wm_prog_key { 63 uint8_t iz_lookup; 64 GLuint stats_wm:1; 65 GLuint flat_shade:1; 66 GLuint nr_color_regions:5; 67 GLuint sample_alpha_to_coverage:1; 68 GLuint render_to_fbo:1; 69 GLuint clamp_fragment_color:1; 70 GLuint line_aa:2; 71 72 GLbitfield proj_attrib_mask; /**< one bit per fragment program attribute */ 73 74 GLushort drawable_height; 75 GLbitfield64 vp_outputs_written; 76 GLuint program_string_id:32; 77 78 struct brw_sampler_prog_key_data tex; 79 }; 80 81 82 /* A bit of a glossary: 83 * 84 * brw_wm_value: A computed value or program input. Values are 85 * constant, they are created once and are never modified. When a 86 * fragment program register is written or overwritten, new values are 87 * created fresh, preserving the rule that values are constant. 88 * 89 * brw_wm_ref: A reference to a value. Wherever a value used is by an 90 * instruction or as a program output, that is tracked with an 91 * instance of this struct. All references to a value occur after it 92 * is created. After the last reference, a value is dead and can be 93 * discarded. 94 * 95 * brw_wm_grf: Represents a physical hardware register. May be either 96 * empty or hold a value. Register allocation is the process of 97 * assigning values to grf registers. This occurs in pass2 and the 98 * brw_wm_grf struct is not used before that. 99 * 100 * Fragment program registers: These are time-varying constructs that 101 * are hard to reason about and which we translate away in pass0. A 102 * single fragment program register element (eg. temp[0].x) will be 103 * translated to one or more brw_wm_value structs, one for each time 104 * that temp[0].x is written to during the program. 105 */ 106 107 108 109 /* Used in pass2 to track register allocation. 110 */ 111 struct brw_wm_grf { 112 struct brw_wm_value *value; 113 GLuint nextuse; 114 }; 115 116 struct brw_wm_value { 117 struct brw_reg hw_reg; /* emitted to this reg, may not always be there */ 118 struct brw_wm_ref *lastuse; 119 struct brw_wm_grf *resident; 120 GLuint contributes_to_output:1; 121 GLuint spill_slot:16; /* if non-zero, spill immediately after calculation */ 122 }; 123 124 struct brw_wm_ref { 125 struct brw_reg hw_reg; /* nr filled in in pass2, everything else, pass0 */ 126 struct brw_wm_value *value; 127 struct brw_wm_ref *prevuse; 128 GLuint unspill_reg:7; /* unspill to reg */ 129 GLuint emitted:1; 130 GLuint insn:24; 131 }; 132 133 struct brw_wm_constref { 134 const struct brw_wm_ref *ref; 135 GLfloat constval; 136 }; 137 138 139 struct brw_wm_instruction { 140 struct brw_wm_value *dst[4]; 141 struct brw_wm_ref *src[3][4]; 142 GLuint opcode:8; 143 GLuint saturate:1; 144 GLuint writemask:4; 145 GLuint tex_unit:4; /* texture unit for TEX, TXD, TXP instructions */ 146 GLuint tex_idx:4; /* TEXTURE_1D,2D,3D,CUBE,RECT_INDEX source target */ 147 GLuint tex_shadow:1; /* do shadow comparison? */ 148 GLuint eot:1; /* End of thread indicator for FB_WRITE*/ 149 GLuint target:10; /* target binding table index for FB_WRITE*/ 150 }; 151 152 153 #define BRW_WM_MAX_INSN (MAX_PROGRAM_INSTRUCTIONS*3 + FRAG_ATTRIB_MAX + 3) 154 #define BRW_WM_MAX_GRF 128 /* hardware limit */ 155 #define BRW_WM_MAX_VREG (BRW_WM_MAX_INSN * 4) 156 #define BRW_WM_MAX_REF (BRW_WM_MAX_INSN * 12) 157 #define BRW_WM_MAX_PARAM 256 158 #define BRW_WM_MAX_CONST 256 159 #define BRW_WM_MAX_SUBROUTINE 16 160 161 /* used in masks next to WRITEMASK_*. */ 162 #define SATURATE (1<<5) 163 164 165 /* New opcodes to track internal operations required for WM unit. 166 * These are added early so that the registers used can be tracked, 167 * freed and reused like those of other instructions. 168 */ 169 #define WM_PIXELXY (MAX_OPCODE) 170 #define WM_DELTAXY (MAX_OPCODE + 1) 171 #define WM_PIXELW (MAX_OPCODE + 2) 172 #define WM_LINTERP (MAX_OPCODE + 3) 173 #define WM_PINTERP (MAX_OPCODE + 4) 174 #define WM_CINTERP (MAX_OPCODE + 5) 175 #define WM_WPOSXY (MAX_OPCODE + 6) 176 #define WM_FB_WRITE (MAX_OPCODE + 7) 177 #define WM_FRONTFACING (MAX_OPCODE + 8) 178 #define MAX_WM_OPCODE (MAX_OPCODE + 9) 179 180 #define PROGRAM_PAYLOAD (PROGRAM_FILE_MAX) 181 #define NUM_FILES (PROGRAM_PAYLOAD + 1) 182 183 #define PAYLOAD_DEPTH (FRAG_ATTRIB_MAX) 184 #define PAYLOAD_W (FRAG_ATTRIB_MAX + 1) 185 #define PAYLOAD_FP_REG_MAX (FRAG_ATTRIB_MAX + 2) 186 187 struct brw_wm_compile { 188 struct brw_compile func; 189 struct brw_wm_prog_key key; 190 struct brw_wm_prog_data prog_data; 191 192 struct brw_fragment_program *fp; 193 194 GLfloat (*env_param)[4]; 195 196 enum { 197 START, 198 PASS2_DONE 199 } state; 200 201 uint8_t source_depth_reg; 202 uint8_t source_w_reg; 203 uint8_t aa_dest_stencil_reg; 204 uint8_t dest_depth_reg; 205 uint8_t barycentric_coord_reg[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT]; 206 uint8_t nr_payload_regs; 207 GLuint computes_depth:1; /* could be derived from program string */ 208 GLuint source_depth_to_render_target:1; 209 GLuint runtime_check_aads_emit:1; 210 211 /* Initial pass - translate fp instructions to fp instructions, 212 * simplifying and adding instructions for interpolation and 213 * framebuffer writes. 214 */ 215 struct prog_instruction *prog_instructions; 216 GLuint nr_fp_insns; 217 GLuint fp_temp; 218 GLuint fp_interp_emitted; 219 220 struct prog_src_register pixel_xy; 221 struct prog_src_register delta_xy; 222 struct prog_src_register pixel_w; 223 224 225 struct brw_wm_value *vreg; 226 GLuint nr_vreg; 227 228 struct brw_wm_value creg[BRW_WM_MAX_PARAM]; 229 GLuint nr_creg; 230 231 struct { 232 struct brw_wm_value depth[4]; /* includes r0/r1 */ 233 struct brw_wm_value input_interp[FRAG_ATTRIB_MAX]; 234 } payload; 235 236 237 const struct brw_wm_ref *pass0_fp_reg[NUM_FILES][256][4]; 238 239 struct brw_wm_ref undef_ref; 240 struct brw_wm_value undef_value; 241 242 struct brw_wm_ref *refs; 243 GLuint nr_refs; 244 245 struct brw_wm_instruction *instruction; 246 GLuint nr_insns; 247 248 struct brw_wm_constref constref[BRW_WM_MAX_CONST]; 249 GLuint nr_constrefs; 250 251 struct brw_wm_grf pass2_grf[BRW_WM_MAX_GRF/2]; 252 253 GLuint grf_limit; 254 GLuint max_wm_grf; 255 GLuint last_scratch; 256 257 GLuint cur_inst; /**< index of current instruction */ 258 259 bool out_of_regs; /**< ran out of GRF registers? */ 260 261 /** Mapping from Mesa registers to hardware registers */ 262 struct { 263 bool inited; 264 struct brw_reg reg; 265 } wm_regs[NUM_FILES][256][4]; 266 267 bool used_grf[BRW_WM_MAX_GRF]; 268 GLuint first_free_grf; 269 struct brw_reg stack; 270 struct brw_reg emit_mask_reg; 271 GLuint tmp_regs[BRW_WM_MAX_GRF]; 272 GLuint tmp_index; 273 GLuint tmp_max; 274 GLuint subroutines[BRW_WM_MAX_SUBROUTINE]; 275 GLuint dispatch_width; 276 277 /** we may need up to 3 constants per instruction (if use_const_buffer) */ 278 struct { 279 GLint index; 280 struct brw_reg reg; 281 } current_const[3]; 282 }; 283 284 285 /** Bits for prog_instruction::Aux field */ 286 #define INST_AUX_EOT 0x1 287 #define INST_AUX_TARGET(T) (T << 1) 288 #define INST_AUX_GET_TARGET(AUX) ((AUX) >> 1) 289 290 291 GLuint brw_wm_nr_args( GLuint opcode ); 292 GLuint brw_wm_is_scalar_result( GLuint opcode ); 293 294 void brw_wm_pass_fp( struct brw_wm_compile *c ); 295 void brw_wm_pass0( struct brw_wm_compile *c ); 296 void brw_wm_pass1( struct brw_wm_compile *c ); 297 void brw_wm_pass2( struct brw_wm_compile *c ); 298 void brw_wm_emit( struct brw_wm_compile *c ); 299 bool brw_wm_arg_can_be_immediate(enum prog_opcode, int arg); 300 void brw_wm_print_value( struct brw_wm_compile *c, 301 struct brw_wm_value *value ); 302 303 void brw_wm_print_ref( struct brw_wm_compile *c, 304 struct brw_wm_ref *ref ); 305 306 void brw_wm_print_insn( struct brw_wm_compile *c, 307 struct brw_wm_instruction *inst ); 308 309 void brw_wm_print_program( struct brw_wm_compile *c, 310 const char *stage ); 311 312 void brw_wm_lookup_iz(struct intel_context *intel, 313 struct brw_wm_compile *c); 314 315 bool brw_wm_fs_emit(struct brw_context *brw, struct brw_wm_compile *c, 316 struct gl_shader_program *prog); 317 318 /* brw_wm_emit.c */ 319 void emit_alu1(struct brw_compile *p, 320 struct brw_instruction *(*func)(struct brw_compile *, 321 struct brw_reg, 322 struct brw_reg), 323 const struct brw_reg *dst, 324 GLuint mask, 325 const struct brw_reg *arg0); 326 void emit_alu2(struct brw_compile *p, 327 struct brw_instruction *(*func)(struct brw_compile *, 328 struct brw_reg, 329 struct brw_reg, 330 struct brw_reg), 331 const struct brw_reg *dst, 332 GLuint mask, 333 const struct brw_reg *arg0, 334 const struct brw_reg *arg1); 335 void emit_cinterp(struct brw_compile *p, 336 const struct brw_reg *dst, 337 GLuint mask, 338 const struct brw_reg *arg0); 339 void emit_cmp(struct brw_compile *p, 340 const struct brw_reg *dst, 341 GLuint mask, 342 const struct brw_reg *arg0, 343 const struct brw_reg *arg1, 344 const struct brw_reg *arg2); 345 void emit_ddxy(struct brw_compile *p, 346 const struct brw_reg *dst, 347 GLuint mask, 348 bool is_ddx, 349 const struct brw_reg *arg0, 350 bool negate_value); 351 void emit_delta_xy(struct brw_compile *p, 352 const struct brw_reg *dst, 353 GLuint mask, 354 const struct brw_reg *arg0); 355 void emit_dp2(struct brw_compile *p, 356 const struct brw_reg *dst, 357 GLuint mask, 358 const struct brw_reg *arg0, 359 const struct brw_reg *arg1); 360 void emit_dp3(struct brw_compile *p, 361 const struct brw_reg *dst, 362 GLuint mask, 363 const struct brw_reg *arg0, 364 const struct brw_reg *arg1); 365 void emit_dp4(struct brw_compile *p, 366 const struct brw_reg *dst, 367 GLuint mask, 368 const struct brw_reg *arg0, 369 const struct brw_reg *arg1); 370 void emit_dph(struct brw_compile *p, 371 const struct brw_reg *dst, 372 GLuint mask, 373 const struct brw_reg *arg0, 374 const struct brw_reg *arg1); 375 void emit_fb_write(struct brw_wm_compile *c, 376 struct brw_reg *arg0, 377 struct brw_reg *arg1, 378 struct brw_reg *arg2, 379 GLuint target, 380 GLuint eot); 381 void emit_frontfacing(struct brw_compile *p, 382 const struct brw_reg *dst, 383 GLuint mask); 384 void emit_linterp(struct brw_compile *p, 385 const struct brw_reg *dst, 386 GLuint mask, 387 const struct brw_reg *arg0, 388 const struct brw_reg *deltas); 389 void emit_lrp(struct brw_compile *p, 390 const struct brw_reg *dst, 391 GLuint mask, 392 const struct brw_reg *arg0, 393 const struct brw_reg *arg1, 394 const struct brw_reg *arg2); 395 void emit_mad(struct brw_compile *p, 396 const struct brw_reg *dst, 397 GLuint mask, 398 const struct brw_reg *arg0, 399 const struct brw_reg *arg1, 400 const struct brw_reg *arg2); 401 void emit_math1(struct brw_wm_compile *c, 402 GLuint function, 403 const struct brw_reg *dst, 404 GLuint mask, 405 const struct brw_reg *arg0); 406 void emit_math2(struct brw_wm_compile *c, 407 GLuint function, 408 const struct brw_reg *dst, 409 GLuint mask, 410 const struct brw_reg *arg0, 411 const struct brw_reg *arg1); 412 void emit_min(struct brw_compile *p, 413 const struct brw_reg *dst, 414 GLuint mask, 415 const struct brw_reg *arg0, 416 const struct brw_reg *arg1); 417 void emit_max(struct brw_compile *p, 418 const struct brw_reg *dst, 419 GLuint mask, 420 const struct brw_reg *arg0, 421 const struct brw_reg *arg1); 422 void emit_pinterp(struct brw_compile *p, 423 const struct brw_reg *dst, 424 GLuint mask, 425 const struct brw_reg *arg0, 426 const struct brw_reg *deltas, 427 const struct brw_reg *w); 428 void emit_pixel_xy(struct brw_wm_compile *c, 429 const struct brw_reg *dst, 430 GLuint mask); 431 void emit_pixel_w(struct brw_wm_compile *c, 432 const struct brw_reg *dst, 433 GLuint mask, 434 const struct brw_reg *arg0, 435 const struct brw_reg *deltas); 436 void emit_sop(struct brw_compile *p, 437 const struct brw_reg *dst, 438 GLuint mask, 439 GLuint cond, 440 const struct brw_reg *arg0, 441 const struct brw_reg *arg1); 442 void emit_sign(struct brw_compile *p, 443 const struct brw_reg *dst, 444 GLuint mask, 445 const struct brw_reg *arg0); 446 void emit_tex(struct brw_wm_compile *c, 447 struct brw_reg *dst, 448 GLuint dst_flags, 449 struct brw_reg *arg, 450 struct brw_reg depth_payload, 451 GLuint tex_idx, 452 GLuint sampler, 453 bool shadow); 454 void emit_txb(struct brw_wm_compile *c, 455 struct brw_reg *dst, 456 GLuint dst_flags, 457 struct brw_reg *arg, 458 struct brw_reg depth_payload, 459 GLuint tex_idx, 460 GLuint sampler); 461 void emit_wpos_xy(struct brw_wm_compile *c, 462 const struct brw_reg *dst, 463 GLuint mask, 464 const struct brw_reg *arg0); 465 void emit_xpd(struct brw_compile *p, 466 const struct brw_reg *dst, 467 GLuint mask, 468 const struct brw_reg *arg0, 469 const struct brw_reg *arg1); 470 471 GLboolean brw_link_shader(struct gl_context *ctx, struct gl_shader_program *prog); 472 struct gl_shader *brw_new_shader(struct gl_context *ctx, GLuint name, GLuint type); 473 struct gl_shader_program *brw_new_shader_program(struct gl_context *ctx, GLuint name); 474 475 bool brw_color_buffer_write_enabled(struct brw_context *brw); 476 bool brw_render_target_supported(struct intel_context *intel, 477 struct gl_renderbuffer *rb); 478 void brw_wm_payload_setup(struct brw_context *brw, 479 struct brw_wm_compile *c); 480 bool do_wm_prog(struct brw_context *brw, 481 struct gl_shader_program *prog, 482 struct brw_fragment_program *fp, 483 struct brw_wm_prog_key *key); 484 void brw_wm_debug_recompile(struct brw_context *brw, 485 struct gl_shader_program *prog, 486 const struct brw_wm_prog_key *key); 487 488 #endif 489