r200_vertprog.c revision 122629f27925a9dc50029bebc5079f87f416a7e1
1/************************************************************************** 2 3Copyright (C) 2005 Aapo Tahkola. 4 5All Rights Reserved. 6 7Permission is hereby granted, free of charge, to any person obtaining a 8copy of this software and associated documentation files (the "Software"), 9to deal in the Software without restriction, including without limitation 10on the rights to use, copy, modify, merge, publish, distribute, sub 11license, and/or sell copies of the Software, and to permit persons to whom 12the Software is furnished to do so, subject to the following conditions: 13 14The above copyright notice and this permission notice (including the next 15paragraph) shall be included in all copies or substantial portions of the 16Software. 17 18THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 19IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 20FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL 21THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, 22DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 23OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 24USE OR OTHER DEALINGS IN THE SOFTWARE. 25 26**************************************************************************/ 27 28/* 29 * Authors: 30 * Aapo Tahkola <aet@rasterburn.org> 31 */ 32#include "glheader.h" 33#include "macros.h" 34#include "enums.h" 35#include "program.h" 36 37#include "r200_context.h" 38#include "r200_vertprog.h" 39#include "r200_ioctl.h" 40#include "r200_tcl.h" 41#include "program_instruction.h" 42#include "tnl/tnl.h" 43 44#if SWIZZLE_X != VSF_IN_COMPONENT_X || \ 45 SWIZZLE_Y != VSF_IN_COMPONENT_Y || \ 46 SWIZZLE_Z != VSF_IN_COMPONENT_Z || \ 47 SWIZZLE_W != VSF_IN_COMPONENT_W || \ 48 SWIZZLE_ZERO != VSF_IN_COMPONENT_ZERO || \ 49 SWIZZLE_ONE != VSF_IN_COMPONENT_ONE || \ 50 WRITEMASK_X != VSF_FLAG_X || \ 51 WRITEMASK_Y != VSF_FLAG_Y || \ 52 WRITEMASK_Z != VSF_FLAG_Z || \ 53 WRITEMASK_W != VSF_FLAG_W 54#error Cannot change these! 55#endif 56 57#define SCALAR_FLAG (1<<31) 58#define FLAG_MASK (1<<31) 59#define OP_MASK (0xf) /* we are unlikely to have more than 15 */ 60#define OPN(operator, ip) {#operator, OPCODE_##operator, ip} 61 62static struct{ 63 char *name; 64 int opcode; 65 unsigned long ip; /* number of input operands and flags */ 66}op_names[]={ 67 OPN(ABS, 1), 68 OPN(ADD, 2), 69 OPN(ARL, 1|SCALAR_FLAG), 70 OPN(DP3, 2), 71 OPN(DP4, 2), 72 OPN(DPH, 2), 73 OPN(DST, 2), 74 OPN(EX2, 1|SCALAR_FLAG), 75 OPN(EXP, 1|SCALAR_FLAG), 76 OPN(FLR, 1), 77 OPN(FRC, 1), 78 OPN(LG2, 1|SCALAR_FLAG), 79 OPN(LIT, 1), 80 OPN(LOG, 1|SCALAR_FLAG), 81 OPN(MAD, 3), 82 OPN(MAX, 2), 83 OPN(MIN, 2), 84 OPN(MOV, 1), 85 OPN(MUL, 2), 86 OPN(POW, 2|SCALAR_FLAG), 87 OPN(RCP, 1|SCALAR_FLAG), 88 OPN(RSQ, 1|SCALAR_FLAG), 89 OPN(SGE, 2), 90 OPN(SLT, 2), 91 OPN(SUB, 2), 92 OPN(SWZ, 1), 93 OPN(XPD, 2), 94 OPN(PRINT, 0), 95 OPN(END, 0), 96}; 97#undef OPN 98 99static GLboolean r200VertexProgUpdateParams(GLcontext *ctx, struct r200_vertex_program *vp) 100{ 101 r200ContextPtr rmesa = R200_CONTEXT( ctx ); 102 GLfloat *fcmd = (GLfloat *)&rmesa->hw.vpp[0].cmd[VPP_CMD_0 + 1]; 103 int pi; 104 struct gl_vertex_program *mesa_vp = (void *)vp; 105 struct gl_program_parameter_list *paramList; 106 drm_radeon_cmd_header_t tmp; 107 108 R200_STATECHANGE( rmesa, vpp[0] ); 109 R200_STATECHANGE( rmesa, vpp[1] ); 110 assert(mesa_vp->Base.Parameters); 111 _mesa_load_state_parameters(ctx, mesa_vp->Base.Parameters); 112 paramList = mesa_vp->Base.Parameters; 113 114 if(paramList->NumParameters > R200_VSF_MAX_PARAM){ 115 fprintf(stderr, "%s:Params exhausted\n", __FUNCTION__); 116 return GL_FALSE; 117 } 118 119 for(pi = 0; pi < paramList->NumParameters; pi++) { 120 switch(paramList->Parameters[pi].Type) { 121 case PROGRAM_STATE_VAR: 122 case PROGRAM_NAMED_PARAM: 123 //fprintf(stderr, "%s", vp->Parameters->Parameters[pi].Name); 124 case PROGRAM_CONSTANT: 125 *fcmd++ = paramList->ParameterValues[pi][0]; 126 *fcmd++ = paramList->ParameterValues[pi][1]; 127 *fcmd++ = paramList->ParameterValues[pi][2]; 128 *fcmd++ = paramList->ParameterValues[pi][3]; 129 break; 130 default: 131 _mesa_problem(NULL, "Bad param type in %s", __FUNCTION__); 132 break; 133 } 134 if (pi == 95) { 135 fcmd = (GLfloat *)rmesa->hw.vpp[1].cmd[VPP_CMD_0 + 1]; 136 } 137 } 138 /* hack up the cmd_size so not the whole state atom is emitted always. */ 139 rmesa->hw.vpp[0].cmd_size = 140 1 + 4 * ((paramList->NumParameters > 96) ? 96 : paramList->NumParameters); 141 tmp.i = rmesa->hw.vpp[0].cmd[VPP_CMD_0]; 142 tmp.veclinear.count = (paramList->NumParameters > 96) ? 96 : paramList->NumParameters; 143 rmesa->hw.vpp[0].cmd[VPP_CMD_0] = tmp.i; 144 if (paramList->NumParameters > 96) { 145 rmesa->hw.vpp[1].cmd_size = 1 + 4 * (paramList->NumParameters - 96); 146 tmp.i = rmesa->hw.vpp[1].cmd[VPP_CMD_0]; 147 tmp.veclinear.count = paramList->NumParameters - 96; 148 rmesa->hw.vpp[1].cmd[VPP_CMD_0] = tmp.i; 149 } 150 return GL_TRUE; 151} 152 153static __inline unsigned long t_dst_mask(GLuint mask) 154{ 155 /* WRITEMASK_* is equivalent to VSF_FLAG_* */ 156 return mask & VSF_FLAG_ALL; 157} 158 159static unsigned long t_dst(struct prog_dst_register *dst) 160{ 161 switch(dst->File) { 162 case PROGRAM_TEMPORARY: 163 return ((dst->Index << R200_VPI_OUT_REG_INDEX_SHIFT) 164 | R200_VSF_OUT_CLASS_TMP); 165 case PROGRAM_OUTPUT: 166 switch (dst->Index) { 167 case VERT_RESULT_HPOS: 168 return R200_VSF_OUT_CLASS_RESULT_POS; 169 case VERT_RESULT_COL0: 170 return R200_VSF_OUT_CLASS_RESULT_COLOR; 171 case VERT_RESULT_COL1: 172 return ((1 << R200_VPI_OUT_REG_INDEX_SHIFT) 173 | R200_VSF_OUT_CLASS_RESULT_COLOR); 174 case VERT_RESULT_FOGC: 175 return R200_VSF_OUT_CLASS_RESULT_FOGC; 176 case VERT_RESULT_TEX0: 177 case VERT_RESULT_TEX1: 178 case VERT_RESULT_TEX2: 179 case VERT_RESULT_TEX3: 180 case VERT_RESULT_TEX4: 181 case VERT_RESULT_TEX5: 182 return (((dst->Index - VERT_RESULT_TEX0) << R200_VPI_OUT_REG_INDEX_SHIFT) 183 | R200_VSF_OUT_CLASS_RESULT_TEXC); 184 case VERT_RESULT_PSIZ: 185 return R200_VSF_OUT_CLASS_RESULT_POINTSIZE; 186 default: 187 fprintf(stderr, "problem in %s, unknown dst output reg %d\n", __FUNCTION__, dst->Index); 188 exit(0); 189 return 0; 190 } 191 case PROGRAM_ADDRESS: 192 assert (dst->Index == 0); 193 return R200_VSF_OUT_CLASS_ADDR; 194 default: 195 fprintf(stderr, "problem in %s, unknown register type %d\n", __FUNCTION__, dst->File); 196 exit(0); 197 return 0; 198 } 199} 200 201static unsigned long t_src_class(enum register_file file) 202{ 203 204 switch(file){ 205 case PROGRAM_TEMPORARY: 206 return VSF_IN_CLASS_TMP; 207 208 case PROGRAM_INPUT: 209 return VSF_IN_CLASS_ATTR; 210 211 case PROGRAM_LOCAL_PARAM: 212 case PROGRAM_ENV_PARAM: 213 case PROGRAM_NAMED_PARAM: 214 case PROGRAM_STATE_VAR: 215 return VSF_IN_CLASS_PARAM; 216 /* 217 case PROGRAM_OUTPUT: 218 case PROGRAM_WRITE_ONLY: 219 case PROGRAM_ADDRESS: 220 */ 221 default: 222 fprintf(stderr, "problem in %s", __FUNCTION__); 223 exit(0); 224 } 225} 226 227static __inline unsigned long t_swizzle(GLubyte swizzle) 228{ 229/* this is in fact a NOP as the Mesa SWIZZLE_* are all identical to VSF_IN_COMPONENT_* */ 230 return swizzle; 231} 232 233#if 0 234static void vp_dump_inputs(struct r200_vertex_program *vp, char *caller) 235{ 236 int i; 237 238 if(vp == NULL){ 239 fprintf(stderr, "vp null in call to %s from %s\n", __FUNCTION__, caller); 240 return ; 241 } 242 243 fprintf(stderr, "%s:<", caller); 244 for(i=0; i < VERT_ATTRIB_MAX; i++) 245 fprintf(stderr, "%d ", vp->inputs[i]); 246 fprintf(stderr, ">\n"); 247 248} 249#endif 250 251static unsigned long t_src_index(struct r200_vertex_program *vp, struct prog_src_register *src) 252{ 253/* 254 int i; 255 int max_reg = -1; 256*/ 257 if(src->File == PROGRAM_INPUT){ 258/* if(vp->inputs[src->Index] != -1) 259 return vp->inputs[src->Index]; 260 261 for(i=0; i < VERT_ATTRIB_MAX; i++) 262 if(vp->inputs[i] > max_reg) 263 max_reg = vp->inputs[i]; 264 265 vp->inputs[src->Index] = max_reg+1;*/ 266 267 //vp_dump_inputs(vp, __FUNCTION__); 268 assert(vp->inputs[src->Index] != -1); 269 return vp->inputs[src->Index]; 270 } else { 271 if (src->Index < 0) { 272 fprintf(stderr, "WARNING negative offsets for indirect addressing do not work\n"); 273 return 0; 274 } 275 return src->Index; 276 } 277} 278 279static unsigned long t_src(struct r200_vertex_program *vp, struct prog_src_register *src) 280{ 281 282 return MAKE_VSF_SOURCE(t_src_index(vp, src), 283 t_swizzle(GET_SWZ(src->Swizzle, 0)), 284 t_swizzle(GET_SWZ(src->Swizzle, 1)), 285 t_swizzle(GET_SWZ(src->Swizzle, 2)), 286 t_swizzle(GET_SWZ(src->Swizzle, 3)), 287 t_src_class(src->File), 288 src->NegateBase) | (src->RelAddr << 4); 289} 290 291static unsigned long t_src_scalar(struct r200_vertex_program *vp, struct prog_src_register *src) 292{ 293 294 return MAKE_VSF_SOURCE(t_src_index(vp, src), 295 t_swizzle(GET_SWZ(src->Swizzle, 0)), 296 t_swizzle(GET_SWZ(src->Swizzle, 0)), 297 t_swizzle(GET_SWZ(src->Swizzle, 0)), 298 t_swizzle(GET_SWZ(src->Swizzle, 0)), 299 t_src_class(src->File), 300 src->NegateBase ? VSF_FLAG_ALL : VSF_FLAG_NONE) | (src->RelAddr << 4); 301} 302 303static unsigned long t_opcode(enum prog_opcode opcode) 304{ 305 306 switch(opcode){ 307 case OPCODE_ADD: return R200_VPI_OUT_OP_ADD; 308 /* FIXME: ARL works fine, but negative offsets won't work - fglrx just 309 * seems to ignore neg offsets which isn't quite correct... 310 */ 311 case OPCODE_ARL: return R200_VPI_OUT_OP_ARL; 312 case OPCODE_DP4: return R200_VPI_OUT_OP_DOT; 313 case OPCODE_DST: return R200_VPI_OUT_OP_DST; 314 case OPCODE_EX2: return R200_VPI_OUT_OP_EX2; 315 case OPCODE_EXP: return R200_VPI_OUT_OP_EXP; 316 case OPCODE_FRC: return R200_VPI_OUT_OP_FRC; 317 case OPCODE_LG2: return R200_VPI_OUT_OP_LG2; 318 case OPCODE_LIT: return R200_VPI_OUT_OP_LIT; 319 case OPCODE_LOG: return R200_VPI_OUT_OP_LOG; 320 case OPCODE_MAX: return R200_VPI_OUT_OP_MAX; 321 case OPCODE_MIN: return R200_VPI_OUT_OP_MIN; 322 case OPCODE_MUL: return R200_VPI_OUT_OP_MUL; 323 case OPCODE_RCP: return R200_VPI_OUT_OP_RCP; 324 case OPCODE_RSQ: return R200_VPI_OUT_OP_RSQ; 325 case OPCODE_SGE: return R200_VPI_OUT_OP_SGE; 326 case OPCODE_SLT: return R200_VPI_OUT_OP_SLT; 327 328 default: 329 fprintf(stderr, "%s: Should not be called with opcode %d!", __FUNCTION__, opcode); 330 } 331 exit(-1); 332 return 0; 333} 334 335static unsigned long op_operands(enum prog_opcode opcode) 336{ 337 int i; 338 339 /* Can we trust mesas opcodes to be in order ? */ 340 for(i=0; i < sizeof(op_names) / sizeof(*op_names); i++) 341 if(op_names[i].opcode == opcode) 342 return op_names[i].ip; 343 344 fprintf(stderr, "op %d not found in op_names\n", opcode); 345 exit(-1); 346 return 0; 347} 348 349/* TODO: Get rid of t_src_class call */ 350#define CMP_SRCS(a, b) (((a.RelAddr != b.RelAddr) || (a.Index != b.Index)) && \ 351 ((t_src_class(a.File) == VSF_IN_CLASS_PARAM && \ 352 t_src_class(b.File) == VSF_IN_CLASS_PARAM) || \ 353 (t_src_class(a.File) == VSF_IN_CLASS_ATTR && \ 354 t_src_class(b.File) == VSF_IN_CLASS_ATTR))) \ 355 356/* fglrx on rv250 codes up unused sources as follows: 357 unused but necessary sources are same as previous source, zero-ed out. 358 unnecessary sources are same as previous source but with VSF_IN_CLASS_NONE set. 359 i.e. an add (2 args) has its 2nd arg (if you use it as mov) zero-ed out, and 3rd arg 360 set to VSF_IN_CLASS_NONE. Not sure if strictly necessary. */ 361 362/* use these simpler definitions. Must obviously not be used with not yet set up regs. 363 Those are NOT semantically equivalent to the r300 ones, requires code changes */ 364#define ZERO_SRC_0 (((o_inst->src0 & ~(0xfff << R200_VPI_IN_X_SHIFT)) \ 365 | ((R200_VPI_IN_SELECT_ZERO << R200_VPI_IN_X_SHIFT) \ 366 | (R200_VPI_IN_SELECT_ZERO << R200_VPI_IN_Y_SHIFT) \ 367 | (R200_VPI_IN_SELECT_ZERO << R200_VPI_IN_Z_SHIFT) \ 368 | (R200_VPI_IN_SELECT_ZERO << R200_VPI_IN_W_SHIFT)))) 369 370#define ZERO_SRC_1 (((o_inst->src1 & ~(0xfff << R200_VPI_IN_X_SHIFT)) \ 371 | ((R200_VPI_IN_SELECT_ZERO << R200_VPI_IN_X_SHIFT) \ 372 | (R200_VPI_IN_SELECT_ZERO << R200_VPI_IN_Y_SHIFT) \ 373 | (R200_VPI_IN_SELECT_ZERO << R200_VPI_IN_Z_SHIFT) \ 374 | (R200_VPI_IN_SELECT_ZERO << R200_VPI_IN_W_SHIFT)))) 375 376#define ZERO_SRC_2 (((o_inst->src2 & ~(0xfff << R200_VPI_IN_X_SHIFT)) \ 377 | ((R200_VPI_IN_SELECT_ZERO << R200_VPI_IN_X_SHIFT) \ 378 | (R200_VPI_IN_SELECT_ZERO << R200_VPI_IN_Y_SHIFT) \ 379 | (R200_VPI_IN_SELECT_ZERO << R200_VPI_IN_Z_SHIFT) \ 380 | (R200_VPI_IN_SELECT_ZERO << R200_VPI_IN_W_SHIFT)))) 381 382#define UNUSED_SRC_0 ((o_inst->src0 & ~15) | 9) 383 384#define UNUSED_SRC_1 ((o_inst->src1 & ~15) | 9) 385 386#define UNUSED_SRC_2 ((o_inst->src2 & ~15) | 9) 387 388 389/* DP4 version seems to trigger some hw peculiarity - fglrx does this on r200 however */ 390#define PREFER_DP4 391 392static GLboolean r200_translate_vertex_program(struct r200_vertex_program *vp) 393{ 394 struct gl_vertex_program *mesa_vp = (void *)vp; 395 struct prog_instruction *vpi; 396 int i; 397 VERTEX_SHADER_INSTRUCTION *o_inst; 398 unsigned long operands; 399 int are_srcs_scalar; 400 unsigned long hw_op; 401 402 vp->native = GL_FALSE; 403 404 if ((mesa_vp->Base.InputsRead & 405 ~(VERT_BIT_POS | VERT_BIT_NORMAL | VERT_BIT_COLOR0 | VERT_BIT_COLOR1 | 406 VERT_BIT_FOG | VERT_BIT_TEX0 | VERT_BIT_TEX1 | VERT_BIT_TEX2 | 407 VERT_BIT_TEX3 | VERT_BIT_TEX4 | VERT_BIT_TEX5)) != 0) { 408 if (R200_DEBUG & DEBUG_FALLBACKS) { 409 fprintf(stderr, "can't handle vert prog inputs 0x%x\n", 410 mesa_vp->Base.InputsRead); 411 } 412 return GL_FALSE; 413 } 414 415 if (mesa_vp->IsNVProgram) { 416 /* subtle differences in spec like guaranteed initialized regs could cause 417 headaches. Might want to remove the driconf option to enable it completely */ 418 return GL_FALSE; 419 } 420 /* Initial value should be last tmp reg that hw supports. 421 Strangely enough r300 doesnt mind even though these would be out of range. 422 Smart enough to realize that it doesnt need it? */ 423 int u_temp_i = R200_VSF_MAX_TEMPS - 1; 424 struct prog_src_register src[3]; 425 426/* if (getenv("R300_VP_SAFETY")) { 427 WARN_ONCE("R300_VP_SAFETY enabled.\n"); 428 429 vpi = malloc((mesa_vp->Base.NumInstructions + VSF_MAX_FRAGMENT_TEMPS) * sizeof(struct prog_instruction)); 430 memset(vpi, 0, VSF_MAX_FRAGMENT_TEMPS * sizeof(struct prog_instruction)); 431 432 for (i=0; i < VSF_MAX_FRAGMENT_TEMPS; i++) { 433 vpi[i].Opcode = OPCODE_MOV; 434 vpi[i].StringPos = 0; 435 vpi[i].Data = 0; 436 437 vpi[i].DstReg.File = PROGRAM_TEMPORARY; 438 vpi[i].DstReg.Index = i; 439 vpi[i].DstReg.WriteMask = WRITEMASK_XYZW; 440 vpi[i].DstReg.CondMask = COND_TR; 441 442 vpi[i].SrcReg[0].File = PROGRAM_STATE_VAR; 443 vpi[i].SrcReg[0].Index = 0; 444 vpi[i].SrcReg[0].Swizzle = MAKE_SWIZZLE4(SWIZZLE_ONE, SWIZZLE_ONE, SWIZZLE_ONE, SWIZZLE_ONE); 445 } 446 447 memcpy(&vpi[i], mesa_vp->Base.Instructions, mesa_vp->Base.NumInstructions * sizeof(struct prog_instruction)); 448 449 free(mesa_vp->Base.Instructions); 450 451 mesa_vp->Base.Instructions = vpi; 452 453 mesa_vp->Base.NumInstructions += VSF_MAX_FRAGMENT_TEMPS; 454 vpi = &mesa_vp->Base.Instructions[mesa_vp->Base.NumInstructions-1]; 455 456 assert(vpi->Opcode == OPCODE_END); 457 }*/ 458/* FIXME: is changing the prog safe to do here? */ 459 if (mesa_vp->IsPositionInvariant) { 460 struct gl_program_parameter_list *paramList; 461 GLint tokens[6] = { STATE_MATRIX, STATE_MVP, 0, 0, 0, STATE_MATRIX }; 462 463#ifdef PREFER_DP4 464 tokens[5] = STATE_MATRIX; 465#else 466 tokens[5] = STATE_MATRIX_TRANSPOSE; 467#endif 468 paramList = mesa_vp->Base.Parameters; 469 470 vpi = malloc((mesa_vp->Base.NumInstructions + 4) * sizeof(struct prog_instruction)); 471 memset(vpi, 0, 4 * sizeof(struct prog_instruction)); 472 473 for (i=0; i < 4; i++) { 474 GLint idx; 475 tokens[3] = tokens[4] = i; 476 idx = _mesa_add_state_reference(paramList, tokens); 477#ifdef PREFER_DP4 478 vpi[i].Opcode = OPCODE_DP4; 479 vpi[i].StringPos = 0; 480 vpi[i].Data = 0; 481 482 vpi[i].DstReg.File = PROGRAM_OUTPUT; 483 vpi[i].DstReg.Index = VERT_RESULT_HPOS; 484 vpi[i].DstReg.WriteMask = 1 << i; 485 vpi[i].DstReg.CondMask = COND_TR; 486 487 vpi[i].SrcReg[0].File = PROGRAM_STATE_VAR; 488 vpi[i].SrcReg[0].Index = idx; 489 vpi[i].SrcReg[0].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W); 490 491 vpi[i].SrcReg[1].File = PROGRAM_INPUT; 492 vpi[i].SrcReg[1].Index = VERT_ATTRIB_POS; 493 vpi[i].SrcReg[1].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W); 494#else 495 if (i == 0) 496 vpi[i].Opcode = OPCODE_MUL; 497 else 498 vpi[i].Opcode = OPCODE_MAD; 499 500 vpi[i].StringPos = 0; 501 vpi[i].Data = 0; 502 503 if (i == 3) 504 vpi[i].DstReg.File = PROGRAM_OUTPUT; 505 else 506 vpi[i].DstReg.File = PROGRAM_TEMPORARY; 507 vpi[i].DstReg.Index = 0; 508 vpi[i].DstReg.WriteMask = 0xf; 509 vpi[i].DstReg.CondMask = COND_TR; 510 511 vpi[i].SrcReg[0].File = PROGRAM_STATE_VAR; 512 vpi[i].SrcReg[0].Index = idx; 513 vpi[i].SrcReg[0].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W); 514 515 vpi[i].SrcReg[1].File = PROGRAM_INPUT; 516 vpi[i].SrcReg[1].Index = VERT_ATTRIB_POS; 517 vpi[i].SrcReg[1].Swizzle = MAKE_SWIZZLE4(i, i, i, i); 518 519 if (i > 0) { 520 vpi[i].SrcReg[2].File = PROGRAM_TEMPORARY; 521 vpi[i].SrcReg[2].Index = 0; 522 vpi[i].SrcReg[2].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W); 523 } 524#endif 525 } 526 527 memcpy(&vpi[i], mesa_vp->Base.Instructions, mesa_vp->Base.NumInstructions * sizeof(struct prog_instruction)); 528 529 free(mesa_vp->Base.Instructions); 530 531 mesa_vp->Base.Instructions = vpi; 532 533 mesa_vp->Base.NumInstructions += 4; 534 vpi = &mesa_vp->Base.Instructions[mesa_vp->Base.NumInstructions-1]; 535 536 assert(vpi->Opcode == OPCODE_END); 537 538 mesa_vp->Base.InputsRead |= (1 << VERT_ATTRIB_POS); 539 mesa_vp->Base.OutputsWritten |= (1 << VERT_RESULT_HPOS); 540 541 //fprintf(stderr, "IsPositionInvariant is set!\n"); 542 //_mesa_print_program(&mesa_vp->Base); 543 } 544 545 vp->pos_end = 0; 546 mesa_vp->Base.NumNativeInstructions = 0; 547 mesa_vp->Base.NumNativeParameters = mesa_vp->Base.Parameters->NumParameters; 548 549 for(i=0; i < VERT_ATTRIB_MAX; i++) 550 vp->inputs[i] = -1; 551/* fglrx uses fixed inputs as follows for conventional attribs. 552 generic attribs use non-fixed assignment, fglrx will always use the lowest attrib values available. 553 There are 12 generic attribs possible, corresponding to attrib 0, 2-11 and 13 in a hw vertex prog. 554 attr 1 and 12 are not available for generic attribs as those cannot be made vec4 (correspond to 555 vertex normal/weight) 556 attr 0 is pos, R200_VTX_XY1|R200_VTX_Z1|R200_VTX_W1 in R200_SE_VTX_FMT_0 557 attr 2-5 use colors 0-3 (R200_VTX_FP_RGBA << R200_VTX_COLOR_0/1/2/3_SHIFT in R200_SE_VTX_FMT_0) 558 attr 6-11 use tex 0-5 (4 << R200_VTX_TEX0/1/2/3/4/5_COMP_CNT_SHIFT in R200_SE_VTX_FMT_1) 559 attr 13 uses vtx1 pos (R200_VTX_XY1|R200_VTX_Z1|R200_VTX_W1 in R200_SE_VTX_FMT_0) 560 generic attribs would require some more work (dma regions, renaming). */ 561 562/* may look different when using idx buf / input_route instead of se_vtx_fmt? */ 563 vp->inputs[VERT_ATTRIB_POS] = 0; 564 vp->inputs[VERT_ATTRIB_WEIGHT] = 12; 565 vp->inputs[VERT_ATTRIB_NORMAL] = 1; 566 vp->inputs[VERT_ATTRIB_COLOR0] = 2; 567 vp->inputs[VERT_ATTRIB_COLOR1] = 3; 568 vp->inputs[VERT_ATTRIB_FOG] = 15; 569 vp->inputs[VERT_ATTRIB_TEX0] = 6; 570 vp->inputs[VERT_ATTRIB_TEX1] = 7; 571 vp->inputs[VERT_ATTRIB_TEX2] = 8; 572 vp->inputs[VERT_ATTRIB_TEX3] = 9; 573 vp->inputs[VERT_ATTRIB_TEX4] = 10; 574 vp->inputs[VERT_ATTRIB_TEX5] = 11; 575/* attr 4,5 and 13 are only used with generic attribs. 576 Haven't seen attr 14 used, maybe that's for the hw pointsize vec1 (which is 577 not possibe to use with vertex progs as it is lacking in vert prog specification) */ 578 579 assert(mesa_vp->Base.OutputsWritten & (1 << VERT_RESULT_HPOS)); 580 581 vp->translated = GL_TRUE; 582 583 o_inst = vp->instr; 584 for(vpi = mesa_vp->Base.Instructions; vpi->Opcode != OPCODE_END; vpi++, o_inst++){ 585 if (u_temp_i < mesa_vp->Base.NumTemporaries) { 586 if (R200_DEBUG & DEBUG_FALLBACKS) { 587 fprintf(stderr, "Ran out of temps, num temps %d, us %d\n", mesa_vp->Base.NumTemporaries, u_temp_i); 588 } 589 return GL_FALSE; 590 } 591 u_temp_i = R200_VSF_MAX_TEMPS - 1; 592 if(o_inst - vp->instr >= R200_VSF_MAX_INST) { 593 mesa_vp->Base.NumNativeInstructions = 129; 594 if (R200_DEBUG & DEBUG_FALLBACKS) { 595 fprintf(stderr, "more than 128 native instructions\n"); 596 } 597 return GL_FALSE; 598 } 599 600 operands = op_operands(vpi->Opcode); 601 are_srcs_scalar = operands & SCALAR_FLAG; 602 operands &= OP_MASK; 603 604 for(i = 0; i < operands; i++) 605 src[i] = vpi->SrcReg[i]; 606 607 if(operands == 3){ 608 if( CMP_SRCS(src[1], src[2]) || CMP_SRCS(src[0], src[2]) ){ 609 o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_ADD, 610 (u_temp_i << R200_VPI_OUT_REG_INDEX_SHIFT) | R200_VSF_OUT_CLASS_TMP, 611 VSF_FLAG_ALL); 612 613 o_inst->src0 = MAKE_VSF_SOURCE(t_src_index(vp, &src[2]), 614 SWIZZLE_X, SWIZZLE_Y, 615 SWIZZLE_Z, SWIZZLE_W, 616 t_src_class(src[2].File), VSF_FLAG_NONE) | (src[2].RelAddr << 4); 617 618 o_inst->src1 = ZERO_SRC_0; 619 o_inst->src2 = UNUSED_SRC_1; 620 o_inst++; 621 622 src[2].File = PROGRAM_TEMPORARY; 623 src[2].Index = u_temp_i; 624 src[2].RelAddr = 0; 625 u_temp_i--; 626 } 627 } 628 629 if(operands >= 2){ 630 if( CMP_SRCS(src[1], src[0]) ){ 631 o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_ADD, 632 (u_temp_i << R200_VPI_OUT_REG_INDEX_SHIFT) | R200_VSF_OUT_CLASS_TMP, 633 VSF_FLAG_ALL); 634 635 o_inst->src0 = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), 636 SWIZZLE_X, SWIZZLE_Y, 637 SWIZZLE_Z, SWIZZLE_W, 638 t_src_class(src[0].File), VSF_FLAG_NONE) | (src[0].RelAddr << 4); 639 640 o_inst->src1 = ZERO_SRC_0; 641 o_inst->src2 = UNUSED_SRC_1; 642 o_inst++; 643 644 src[0].File = PROGRAM_TEMPORARY; 645 src[0].Index = u_temp_i; 646 src[0].RelAddr = 0; 647 u_temp_i--; 648 } 649 } 650 651 /* These ops need special handling. */ 652 switch(vpi->Opcode){ 653 case OPCODE_POW: 654/* pow takes only one argument, first scalar is in slot x, 2nd in slot z (other slots don't matter). 655 So may need to insert additional instruction */ 656 if ((src[0].File == src[1].File) && 657 (src[0].Index == src[1].Index)) { 658 o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_POW, t_dst(&vpi->DstReg), 659 t_dst_mask(vpi->DstReg.WriteMask)); 660 o_inst->src0 = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), 661 t_swizzle(GET_SWZ(src[0].Swizzle, 0)), 662 SWIZZLE_ZERO, 663 t_swizzle(GET_SWZ(src[1].Swizzle, 0)), 664 SWIZZLE_ZERO, 665 t_src_class(src[0].File), 666 src[0].NegateBase) | (src[0].RelAddr << 4); 667 o_inst->src1 = UNUSED_SRC_0; 668 o_inst->src2 = UNUSED_SRC_0; 669 } 670 else { 671 o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_ADD, 672 (u_temp_i << R200_VPI_OUT_REG_INDEX_SHIFT) | R200_VSF_OUT_CLASS_TMP, 673 VSF_FLAG_ALL); 674 o_inst->src0 = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), 675 t_swizzle(GET_SWZ(src[0].Swizzle, 0)), 676 SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ZERO, 677 t_src_class(src[0].File), 678 src[0].NegateBase ? VSF_FLAG_ALL : VSF_FLAG_NONE) | (src[0].RelAddr << 4); 679 o_inst->src1 = MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), 680 SWIZZLE_ZERO, SWIZZLE_ZERO, 681 t_swizzle(GET_SWZ(src[1].Swizzle, 0)), SWIZZLE_ZERO, 682 t_src_class(src[1].File), 683 src[1].NegateBase ? VSF_FLAG_ALL : VSF_FLAG_NONE) | (src[1].RelAddr << 4); 684 o_inst->src2 = UNUSED_SRC_1; 685 o_inst++; 686 687 o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_POW, t_dst(&vpi->DstReg), 688 t_dst_mask(vpi->DstReg.WriteMask)); 689 o_inst->src0 = MAKE_VSF_SOURCE(u_temp_i, 690 VSF_IN_COMPONENT_X, 691 VSF_IN_COMPONENT_Y, 692 VSF_IN_COMPONENT_Z, 693 VSF_IN_COMPONENT_W, 694 VSF_IN_CLASS_TMP, 695 VSF_FLAG_NONE); 696 o_inst->src1 = UNUSED_SRC_0; 697 o_inst->src2 = UNUSED_SRC_0; 698 u_temp_i--; 699 } 700 goto next; 701 702 case OPCODE_MOV://ADD RESULT 1.X Y Z W PARAM 0{} {X Y Z W} PARAM 0{} {ZERO ZERO ZERO ZERO} 703 case OPCODE_SWZ: 704 o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_ADD, t_dst(&vpi->DstReg), 705 t_dst_mask(vpi->DstReg.WriteMask)); 706 o_inst->src0 = t_src(vp, &src[0]); 707 o_inst->src1 = ZERO_SRC_0; 708 o_inst->src2 = UNUSED_SRC_1; 709 goto next; 710 711 case OPCODE_MAD: 712 hw_op=(src[0].File == PROGRAM_TEMPORARY && 713 src[1].File == PROGRAM_TEMPORARY && 714 src[2].File == PROGRAM_TEMPORARY) ? R200_VPI_OUT_OP_MAD_2 : R200_VPI_OUT_OP_MAD; 715 716 o_inst->op = MAKE_VSF_OP(hw_op, t_dst(&vpi->DstReg), 717 t_dst_mask(vpi->DstReg.WriteMask)); 718 o_inst->src0 = t_src(vp, &src[0]); 719#if 0 720if ((o_inst - vp->instr) == 31) { 721/* fix up the broken vertex program of quake4 demo... */ 722o_inst->src1 = MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), 723 SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, 724 t_src_class(src[1].File), 725 src[1].NegateBase) | (src[1].RelAddr << 4); 726o_inst->src2 = MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), 727 SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, 728 t_src_class(src[1].File), 729 src[1].NegateBase) | (src[1].RelAddr << 4); 730} 731else { 732 o_inst->src1 = t_src(vp, &src[1]); 733 o_inst->src2 = t_src(vp, &src[2]); 734} 735#else 736 o_inst->src1 = t_src(vp, &src[1]); 737 o_inst->src2 = t_src(vp, &src[2]); 738#endif 739 goto next; 740 741 case OPCODE_DP3://DOT RESULT 1.X Y Z W PARAM 0{} {X Y Z ZERO} PARAM 0{} {X Y Z ZERO} 742 o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_DOT, t_dst(&vpi->DstReg), 743 t_dst_mask(vpi->DstReg.WriteMask)); 744 745 o_inst->src0 = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), 746 t_swizzle(GET_SWZ(src[0].Swizzle, 0)), 747 t_swizzle(GET_SWZ(src[0].Swizzle, 1)), 748 t_swizzle(GET_SWZ(src[0].Swizzle, 2)), 749 SWIZZLE_ZERO, 750 t_src_class(src[0].File), 751 src[0].NegateBase) | (src[0].RelAddr << 4); 752 753 o_inst->src1 = MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), 754 t_swizzle(GET_SWZ(src[1].Swizzle, 0)), 755 t_swizzle(GET_SWZ(src[1].Swizzle, 1)), 756 t_swizzle(GET_SWZ(src[1].Swizzle, 2)), 757 SWIZZLE_ZERO, 758 t_src_class(src[1].File), 759 src[1].NegateBase) | (src[1].RelAddr << 4); 760 761 o_inst->src2 = UNUSED_SRC_1; 762 goto next; 763 764 case OPCODE_DPH://DOT RESULT 1.X Y Z W PARAM 0{} {X Y Z ONE} PARAM 0{} {X Y Z W} 765 o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_DOT, t_dst(&vpi->DstReg), 766 t_dst_mask(vpi->DstReg.WriteMask)); 767 768 o_inst->src0 = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), 769 t_swizzle(GET_SWZ(src[0].Swizzle, 0)), 770 t_swizzle(GET_SWZ(src[0].Swizzle, 1)), 771 t_swizzle(GET_SWZ(src[0].Swizzle, 2)), 772 VSF_IN_COMPONENT_ONE, 773 t_src_class(src[0].File), 774 src[0].NegateBase) | (src[0].RelAddr << 4); 775 o_inst->src1 = t_src(vp, &src[1]); 776 o_inst->src2 = UNUSED_SRC_1; 777 goto next; 778 779 case OPCODE_SUB://ADD RESULT 1.X Y Z W TMP 0{} {X Y Z W} PARAM 1{X Y Z W } {X Y Z W} neg Xneg Yneg Zneg W 780 o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_ADD, t_dst(&vpi->DstReg), 781 t_dst_mask(vpi->DstReg.WriteMask)); 782 783 o_inst->src0 = t_src(vp, &src[0]); 784 o_inst->src1 = MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), 785 t_swizzle(GET_SWZ(src[1].Swizzle, 0)), 786 t_swizzle(GET_SWZ(src[1].Swizzle, 1)), 787 t_swizzle(GET_SWZ(src[1].Swizzle, 2)), 788 t_swizzle(GET_SWZ(src[1].Swizzle, 3)), 789 t_src_class(src[1].File), 790 (!src[1].NegateBase) ? VSF_FLAG_ALL : VSF_FLAG_NONE) | (src[1].RelAddr << 4); 791 o_inst->src2 = UNUSED_SRC_1; 792 goto next; 793 794 case OPCODE_ABS://MAX RESULT 1.X Y Z W PARAM 0{} {X Y Z W} PARAM 0{X Y Z W } {X Y Z W} neg Xneg Yneg Zneg W 795 o_inst->op=MAKE_VSF_OP(R200_VPI_OUT_OP_MAX, t_dst(&vpi->DstReg), 796 t_dst_mask(vpi->DstReg.WriteMask)); 797 798 o_inst->src0=t_src(vp, &src[0]); 799 o_inst->src1=MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), 800 t_swizzle(GET_SWZ(src[0].Swizzle, 0)), 801 t_swizzle(GET_SWZ(src[0].Swizzle, 1)), 802 t_swizzle(GET_SWZ(src[0].Swizzle, 2)), 803 t_swizzle(GET_SWZ(src[0].Swizzle, 3)), 804 t_src_class(src[0].File), 805 (!src[0].NegateBase) ? VSF_FLAG_ALL : VSF_FLAG_NONE) | (src[0].RelAddr << 4); 806 o_inst->src2 = UNUSED_SRC_1; 807 goto next; 808 809 case OPCODE_FLR: 810 /* FRC TMP 0.X Y Z W PARAM 0{} {X Y Z W} 811 ADD RESULT 1.X Y Z W PARAM 0{} {X Y Z W} TMP 0{X Y Z W } {X Y Z W} neg Xneg Yneg Zneg W */ 812 813 o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_FRC, 814 (u_temp_i << R200_VPI_OUT_REG_INDEX_SHIFT) | R200_VSF_OUT_CLASS_TMP, 815 t_dst_mask(vpi->DstReg.WriteMask)); 816 817 o_inst->src0 = t_src(vp, &src[0]); 818 o_inst->src1 = UNUSED_SRC_0; 819 o_inst->src2 = UNUSED_SRC_1; 820 o_inst++; 821 822 o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_ADD, t_dst(&vpi->DstReg), 823 t_dst_mask(vpi->DstReg.WriteMask)); 824 825 o_inst->src0 = t_src(vp, &src[0]); 826 o_inst->src1 = MAKE_VSF_SOURCE(u_temp_i, 827 VSF_IN_COMPONENT_X, 828 VSF_IN_COMPONENT_Y, 829 VSF_IN_COMPONENT_Z, 830 VSF_IN_COMPONENT_W, 831 VSF_IN_CLASS_TMP, 832 /* Not 100% sure about this */ 833 (!src[0].NegateBase) ? VSF_FLAG_ALL : VSF_FLAG_NONE/*VSF_FLAG_ALL*/); 834 835 o_inst->src2 = UNUSED_SRC_0; 836 u_temp_i--; 837 goto next; 838 839 case OPCODE_XPD: 840 /* mul r0, r1.yzxw, r2.zxyw 841 mad r0, -r2.yzxw, r1.zxyw, r0 842 NOTE: might need MAD_2 843 */ 844 845 o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_MUL, 846 (u_temp_i << R200_VPI_OUT_REG_INDEX_SHIFT) | R200_VSF_OUT_CLASS_TMP, 847 t_dst_mask(vpi->DstReg.WriteMask)); 848 849 o_inst->src0 = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), 850 t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // y 851 t_swizzle(GET_SWZ(src[0].Swizzle, 2)), // z 852 t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // x 853 t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // w 854 t_src_class(src[0].File), 855 src[0].NegateBase) | (src[0].RelAddr << 4); 856 857 o_inst->src1 = MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), 858 t_swizzle(GET_SWZ(src[1].Swizzle, 2)), // z 859 t_swizzle(GET_SWZ(src[1].Swizzle, 0)), // x 860 t_swizzle(GET_SWZ(src[1].Swizzle, 1)), // y 861 t_swizzle(GET_SWZ(src[1].Swizzle, 3)), // w 862 t_src_class(src[1].File), 863 src[1].NegateBase) | (src[1].RelAddr << 4); 864 865 o_inst->src2 = UNUSED_SRC_1; 866 o_inst++; 867 u_temp_i--; 868 869 o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_MAD, t_dst(&vpi->DstReg), 870 t_dst_mask(vpi->DstReg.WriteMask)); 871 872 o_inst->src0 = MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), 873 t_swizzle(GET_SWZ(src[1].Swizzle, 1)), // y 874 t_swizzle(GET_SWZ(src[1].Swizzle, 2)), // z 875 t_swizzle(GET_SWZ(src[1].Swizzle, 0)), // x 876 t_swizzle(GET_SWZ(src[1].Swizzle, 3)), // w 877 t_src_class(src[1].File), 878 (!src[1].NegateBase) ? VSF_FLAG_ALL : VSF_FLAG_NONE) | (src[1].RelAddr << 4); 879 880 o_inst->src1 = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), 881 t_swizzle(GET_SWZ(src[0].Swizzle, 2)), // z 882 t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // x 883 t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // y 884 t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // w 885 t_src_class(src[0].File), 886 src[0].NegateBase) | (src[0].RelAddr << 4); 887 888 o_inst->src2 = MAKE_VSF_SOURCE(u_temp_i+1, 889 VSF_IN_COMPONENT_X, 890 VSF_IN_COMPONENT_Y, 891 VSF_IN_COMPONENT_Z, 892 VSF_IN_COMPONENT_W, 893 VSF_IN_CLASS_TMP, 894 VSF_FLAG_NONE); 895 goto next; 896 897 case OPCODE_END: 898 break; 899 default: 900 break; 901 } 902 903 o_inst->op = MAKE_VSF_OP(t_opcode(vpi->Opcode), t_dst(&vpi->DstReg), 904 t_dst_mask(vpi->DstReg.WriteMask)); 905 906 if(are_srcs_scalar){ 907 switch(operands){ 908 case 1: 909 o_inst->src0 = t_src_scalar(vp, &src[0]); 910 o_inst->src1 = UNUSED_SRC_0; 911 o_inst->src2 = UNUSED_SRC_1; 912 break; 913 914 case 2: 915 o_inst->src0 = t_src_scalar(vp, &src[0]); 916 o_inst->src1 = t_src_scalar(vp, &src[1]); 917 o_inst->src2 = UNUSED_SRC_1; 918 break; 919 920 case 3: 921 o_inst->src0 = t_src_scalar(vp, &src[0]); 922 o_inst->src1 = t_src_scalar(vp, &src[1]); 923 o_inst->src2 = t_src_scalar(vp, &src[2]); 924 break; 925 926 default: 927 fprintf(stderr, "illegal number of operands %lu\n", operands); 928 exit(-1); 929 break; 930 } 931 } else { 932 switch(operands){ 933 case 1: 934 o_inst->src0 = t_src(vp, &src[0]); 935 o_inst->src1 = UNUSED_SRC_0; 936 o_inst->src2 = UNUSED_SRC_1; 937 break; 938 939 case 2: 940 o_inst->src0 = t_src(vp, &src[0]); 941 o_inst->src1 = t_src(vp, &src[1]); 942 o_inst->src2 = UNUSED_SRC_1; 943 break; 944 945 case 3: 946 o_inst->src0 = t_src(vp, &src[0]); 947 o_inst->src1 = t_src(vp, &src[1]); 948 o_inst->src2 = t_src(vp, &src[2]); 949 break; 950 951 default: 952 fprintf(stderr, "illegal number of operands %lu\n", operands); 953 exit(-1); 954 break; 955 } 956 } 957 next: 958 if ((o_inst->op & R200_VSF_OUT_CLASS_MASK) == R200_VSF_OUT_CLASS_RESULT_POS) { 959 vp->pos_end = (o_inst - vp->instr); 960 } 961 } 962 963 /* need to test again since some instructions require more than one (up to 3) native inst */ 964 if(o_inst - vp->instr > R200_VSF_MAX_INST) { 965 mesa_vp->Base.NumNativeInstructions = 129; 966 if (R200_DEBUG & DEBUG_FALLBACKS) { 967 fprintf(stderr, "more than 128 native instructions\n"); 968 } 969 return GL_FALSE; 970 } 971 vp->native = GL_TRUE; 972 mesa_vp->Base.NumNativeInstructions = (o_inst - vp->instr); 973#if 0 974 fprintf(stderr, "hw program:\n"); 975 for(i=0; i < vp->program.length; i++) 976 fprintf(stderr, "%08x\n", vp->instr[i]); 977#endif 978 return GL_TRUE; 979} 980 981void r200SetupVertexProg( GLcontext *ctx ) { 982 r200ContextPtr rmesa = R200_CONTEXT(ctx); 983 struct r200_vertex_program *vp = (struct r200_vertex_program *)ctx->VertexProgram.Current; 984 GLboolean fallback; 985 GLint i; 986 987 if (!vp->translated) { 988 rmesa->curr_vp_hw = NULL; 989 r200_translate_vertex_program(vp); 990 } 991 /* could optimize setting up vertex progs away for non-tcl hw */ 992 fallback = !(vp->native && r200VertexProgUpdateParams(ctx, vp) && 993 rmesa->r200Screen->drmSupportsVertexProgram); 994 TCL_FALLBACK(ctx, R200_TCL_FALLBACK_VERTEX_PROGRAM, fallback); 995 if (fallback) return; 996 997 R200_STATECHANGE( rmesa, pvs ); 998 999 rmesa->hw.pvs.cmd[PVS_CNTL_1] = (0 << R200_PVS_CNTL_1_PROGRAM_START_SHIFT) | 1000 ((vp->mesa_program.Base.NumNativeInstructions - 1) << R200_PVS_CNTL_1_PROGRAM_END_SHIFT) | 1001 (vp->pos_end << R200_PVS_CNTL_1_POS_END_SHIFT); 1002 rmesa->hw.pvs.cmd[PVS_CNTL_2] = (0 << R200_PVS_CNTL_2_PARAM_OFFSET_SHIFT) | 1003 (vp->mesa_program.Base.NumNativeParameters << R200_PVS_CNTL_2_PARAM_COUNT_SHIFT); 1004 1005 /* maybe user clip planes just work with vertex progs... untested */ 1006 if (ctx->Transform.ClipPlanesEnabled) { 1007 R200_STATECHANGE( rmesa, tcl ); 1008 if (vp->mesa_program.IsPositionInvariant) { 1009 rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] |= (ctx->Transform.ClipPlanesEnabled << 2); 1010 } 1011 else { 1012 rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] &= ~(0xfc); 1013 } 1014 } 1015 1016 if (vp != rmesa->curr_vp_hw) { 1017 GLuint count = vp->mesa_program.Base.NumNativeInstructions; 1018 drm_radeon_cmd_header_t tmp; 1019 1020 R200_STATECHANGE( rmesa, vpi[0] ); 1021 R200_STATECHANGE( rmesa, vpi[1] ); 1022 1023 /* FIXME: what about using a memcopy... */ 1024 for (i = 0; (i < 64) && i < count; i++) { 1025 rmesa->hw.vpi[0].cmd[VPI_OPDST_0 + 4 * i] = vp->instr[i].op; 1026 rmesa->hw.vpi[0].cmd[VPI_SRC0_0 + 4 * i] = vp->instr[i].src0; 1027 rmesa->hw.vpi[0].cmd[VPI_SRC1_0 + 4 * i] = vp->instr[i].src1; 1028 rmesa->hw.vpi[0].cmd[VPI_SRC2_0 + 4 * i] = vp->instr[i].src2; 1029 } 1030 /* hack up the cmd_size so not the whole state atom is emitted always. 1031 This may require some more thought, we may emit half progs on lost state, but 1032 hopefully it won't matter? 1033 WARNING: must not use R200_DB_STATECHANGE, this will produce bogus (and rejected) 1034 packet emits (due to the mismatched cmd_size and count in cmd/last_cmd) */ 1035 rmesa->hw.vpi[0].cmd_size = 1 + 4 * ((count > 64) ? 64 : count); 1036 tmp.i = rmesa->hw.vpi[0].cmd[VPI_CMD_0]; 1037 tmp.veclinear.count = (count > 64) ? 64 : count; 1038 rmesa->hw.vpi[0].cmd[VPI_CMD_0] = tmp.i; 1039 if (count > 64) { 1040 for (i = 0; i < (count - 64); i++) { 1041 rmesa->hw.vpi[1].cmd[VPI_OPDST_0 + 4 * i] = vp->instr[i + 64].op; 1042 rmesa->hw.vpi[1].cmd[VPI_SRC0_0 + 4 * i] = vp->instr[i + 64].src0; 1043 rmesa->hw.vpi[1].cmd[VPI_SRC1_0 + 4 * i] = vp->instr[i + 64].src1; 1044 rmesa->hw.vpi[1].cmd[VPI_SRC2_0 + 4 * i] = vp->instr[i + 64].src2; 1045 } 1046 rmesa->hw.vpi[1].cmd_size = 1 + 4 * (count - 64); 1047 tmp.i = rmesa->hw.vpi[1].cmd[VPI_CMD_0]; 1048 tmp.veclinear.count = count - 64; 1049 rmesa->hw.vpi[1].cmd[VPI_CMD_0] = tmp.i; 1050 } 1051 rmesa->curr_vp_hw = vp; 1052 } 1053} 1054 1055 1056static void 1057r200BindProgram(GLcontext *ctx, GLenum target, struct gl_program *prog) 1058{ 1059 r200ContextPtr rmesa = R200_CONTEXT(ctx); 1060 1061 switch(target){ 1062 case GL_VERTEX_PROGRAM_ARB: 1063 rmesa->curr_vp_hw = NULL; 1064 break; 1065 default: 1066 _mesa_problem(ctx, "Target not supported yet!"); 1067 break; 1068 } 1069} 1070 1071static struct gl_program * 1072r200NewProgram(GLcontext *ctx, GLenum target, GLuint id) 1073{ 1074 struct r200_vertex_program *vp; 1075 1076 switch(target){ 1077 case GL_VERTEX_PROGRAM_ARB: 1078 vp = CALLOC_STRUCT(r200_vertex_program); 1079 return _mesa_init_vertex_program(ctx, &vp->mesa_program, target, id); 1080 case GL_FRAGMENT_PROGRAM_ARB: 1081 case GL_FRAGMENT_PROGRAM_NV: 1082 return _mesa_init_fragment_program( ctx, CALLOC_STRUCT(gl_fragment_program), target, id ); 1083 default: 1084 _mesa_problem(ctx, "Bad target in r200NewProgram"); 1085 } 1086 return NULL; 1087} 1088 1089 1090static void 1091r200DeleteProgram(GLcontext *ctx, struct gl_program *prog) 1092{ 1093 _mesa_delete_program(ctx, prog); 1094} 1095 1096static void 1097r200ProgramStringNotify(GLcontext *ctx, GLenum target, struct gl_program *prog) 1098{ 1099 struct r200_vertex_program *vp = (void *)prog; 1100 1101 switch(target) { 1102 case GL_VERTEX_PROGRAM_ARB: 1103 vp->translated = GL_FALSE; 1104 memset(&vp->translated, 0, sizeof(struct r200_vertex_program) - sizeof(struct gl_vertex_program)); 1105 /*r200_translate_vertex_shader(vp);*/ 1106 break; 1107 } 1108 /* need this for tcl fallbacks */ 1109 _tnl_program_string(ctx, target, prog); 1110} 1111 1112static GLboolean 1113r200IsProgramNative(GLcontext *ctx, GLenum target, struct gl_program *prog) 1114{ 1115 struct r200_vertex_program *vp = (void *)prog; 1116 1117 switch(target){ 1118 case GL_VERTEX_STATE_PROGRAM_NV: 1119 case GL_VERTEX_PROGRAM_ARB: 1120 if (!vp->translated) { 1121 r200_translate_vertex_program(vp); 1122 } 1123 /* does not take parameters etc. into account */ 1124 return vp->native; 1125 default: 1126 _mesa_problem(ctx, "Bad target in r200NewProgram"); 1127 } 1128 return 0; 1129} 1130 1131void r200InitShaderFuncs(struct dd_function_table *functions) 1132{ 1133 functions->NewProgram = r200NewProgram; 1134 functions->BindProgram = r200BindProgram; 1135 functions->DeleteProgram = r200DeleteProgram; 1136 functions->ProgramStringNotify = r200ProgramStringNotify; 1137 functions->IsProgramNative = r200IsProgramNative; 1138} 1139