| 1 |
#include <strings.h> |
| 2 |
#include "pipe/p_context.h" |
| 3 |
#include "pipe/p_defines.h" |
| 4 |
#include "pipe/p_state.h" |
| 5 |
#include "util/compiler.h" |
| 6 |
#include "util/u_dynarray.h" |
| 7 |
#include "util/u_debug.h" |
| 8 |
#include "util/u_memory.h" |
| 9 |
|
| 10 |
#include "pipe/p_shader_tokens.h" |
| 11 |
#include "tgsi/tgsi_parse.h" |
| 12 |
#include "tgsi/tgsi_dump.h" |
| 13 |
|
| 14 |
#include "draw/draw_context.h" |
| 15 |
|
| 16 |
#include "nv_object.xml.h" |
| 17 |
#include "nouveau_debug.h" |
| 18 |
#include "nv30/nv30-40_3d.xml.h" |
| 19 |
#include "nv30/nv30_state.h" |
| 20 |
|
| 21 |
/* TODO (at least...): |
| 22 |
* 1. Indexed consts + ARL |
| 23 |
* 3. NV_vp11, NV_vp2, NV_vp3 features |
| 24 |
* - extra arith opcodes |
| 25 |
* - branching |
| 26 |
* - texture sampling |
| 27 |
* - indexed attribs |
| 28 |
* - indexed results |
| 29 |
* 4. bugs |
| 30 |
*/ |
| 31 |
|
| 32 |
#include "nv30/nv30_vertprog.h" |
| 33 |
#include "nv30/nv40_vertprog.h" |
| 34 |
|
| 35 |
struct nvfx_loop_entry { |
| 36 |
unsigned brk_target; |
| 37 |
unsigned cont_target; |
| 38 |
}; |
| 39 |
|
| 40 |
struct nvfx_vpc { |
| 41 |
struct pipe_shader_state pipe; |
| 42 |
struct nv30_vertprog *vp; |
| 43 |
struct tgsi_shader_info* info; |
| 44 |
|
| 45 |
struct nv30_vertprog_exec *vpi; |
| 46 |
|
| 47 |
unsigned r_temps; |
| 48 |
unsigned r_temps_discard; |
| 49 |
struct nvfx_reg r_result[PIPE_MAX_SHADER_OUTPUTS]; |
| 50 |
struct nvfx_reg *r_address; |
| 51 |
struct nvfx_reg *r_temp; |
| 52 |
struct nvfx_reg *r_const; |
| 53 |
struct nvfx_reg r_0_1; |
| 54 |
|
| 55 |
struct nvfx_reg *imm; |
| 56 |
unsigned nr_imm; |
| 57 |
|
| 58 |
int hpos_idx; |
| 59 |
int cvtx_idx; |
| 60 |
|
| 61 |
unsigned is_nv4x; |
| 62 |
|
| 63 |
struct util_dynarray label_relocs; |
| 64 |
struct util_dynarray loop_stack; |
| 65 |
}; |
| 66 |
|
| 67 |
static struct nvfx_reg |
| 68 |
temp(struct nvfx_vpc *vpc) |
| 69 |
{ |
| 70 |
int idx = ffs(~vpc->r_temps) - 1; |
| 71 |
|
| 72 |
if (idx < 0 || (!vpc->is_nv4x && idx >= 16)) { |
| 73 |
NOUVEAU_ERR("out of temps!!\n"); |
| 74 |
return nvfx_reg(NVFXSR_TEMP, 0); |
| 75 |
} |
| 76 |
|
| 77 |
vpc->r_temps |= (1 << idx); |
| 78 |
vpc->r_temps_discard |= (1 << idx); |
| 79 |
return nvfx_reg(NVFXSR_TEMP, idx); |
| 80 |
} |
| 81 |
|
| 82 |
static inline void |
| 83 |
release_temps(struct nvfx_vpc *vpc) |
| 84 |
{ |
| 85 |
vpc->r_temps &= ~vpc->r_temps_discard; |
| 86 |
vpc->r_temps_discard = 0; |
| 87 |
} |
| 88 |
|
| 89 |
static struct nvfx_reg |
| 90 |
constant(struct nvfx_vpc *vpc, int pipe, float x, float y, float z, float w) |
| 91 |
{ |
| 92 |
struct nv30_vertprog *vp = vpc->vp; |
| 93 |
struct nv30_vertprog_data *vpd; |
| 94 |
int idx; |
| 95 |
|
| 96 |
if (pipe >= 0) { |
| 97 |
for (idx = 0; idx < vp->nr_consts; idx++) { |
| 98 |
if (vp->consts[idx].index == pipe) |
| 99 |
return nvfx_reg(NVFXSR_CONST, idx); |
| 100 |
} |
| 101 |
} |
| 102 |
|
| 103 |
idx = vp->nr_consts++; |
| 104 |
vp->consts = realloc(vp->consts, sizeof(*vpd) * vp->nr_consts); |
| 105 |
vpd = &vp->consts[idx]; |
| 106 |
|
| 107 |
vpd->index = pipe; |
| 108 |
vpd->value[0] = x; |
| 109 |
vpd->value[1] = y; |
| 110 |
vpd->value[2] = z; |
| 111 |
vpd->value[3] = w; |
| 112 |
return nvfx_reg(NVFXSR_CONST, idx); |
| 113 |
} |
| 114 |
|
| 115 |
#define arith(s,t,o,d,m,s0,s1,s2) \ |
| 116 |
nvfx_insn((s), (NVFX_VP_INST_SLOT_##t << 7) | NVFX_VP_INST_##t##_OP_##o, -1, (d), (m), (s0), (s1), (s2)) |
| 117 |
|
| 118 |
static void |
| 119 |
emit_src(struct nvfx_vpc *vpc, uint32_t *hw, |
| 120 |
int pos, struct nvfx_src src) |
| 121 |
{ |
| 122 |
struct nv30_vertprog *vp = vpc->vp; |
| 123 |
uint32_t sr = 0; |
| 124 |
struct nvfx_relocation reloc; |
| 125 |
|
| 126 |
switch (src.reg.type) { |
| 127 |
case NVFXSR_TEMP: |
| 128 |
sr |= (NVFX_VP(SRC_REG_TYPE_TEMP) << NVFX_VP(SRC_REG_TYPE_SHIFT)); |
| 129 |
sr |= (src.reg.index << NVFX_VP(SRC_TEMP_SRC_SHIFT)); |
| 130 |
break; |
| 131 |
case NVFXSR_INPUT: |
| 132 |
sr |= (NVFX_VP(SRC_REG_TYPE_INPUT) << |
| 133 |
NVFX_VP(SRC_REG_TYPE_SHIFT)); |
| 134 |
vp->ir |= (1 << src.reg.index); |
| 135 |
hw[1] |= (src.reg.index << NVFX_VP(INST_INPUT_SRC_SHIFT)); |
| 136 |
break; |
| 137 |
case NVFXSR_CONST: |
| 138 |
sr |= (NVFX_VP(SRC_REG_TYPE_CONST) << |
| 139 |
NVFX_VP(SRC_REG_TYPE_SHIFT)); |
| 140 |
if (src.reg.index < 256 && src.reg.index >= -256) { |
| 141 |
reloc.location = vp->nr_insns - 1; |
| 142 |
reloc.target = src.reg.index; |
| 143 |
util_dynarray_append(&vp->const_relocs, reloc); |
| 144 |
} else { |
| 145 |
hw[1] |= (src.reg.index << NVFX_VP(INST_CONST_SRC_SHIFT)) & |
| 146 |
NVFX_VP(INST_CONST_SRC_MASK); |
| 147 |
} |
| 148 |
break; |
| 149 |
case NVFXSR_NONE: |
| 150 |
sr |= (NVFX_VP(SRC_REG_TYPE_INPUT) << |
| 151 |
NVFX_VP(SRC_REG_TYPE_SHIFT)); |
| 152 |
break; |
| 153 |
default: |
| 154 |
assert(0); |
| 155 |
} |
| 156 |
|
| 157 |
if (src.negate) |
| 158 |
sr |= NVFX_VP(SRC_NEGATE); |
| 159 |
|
| 160 |
if (src.abs) |
| 161 |
hw[0] |= (1 << (21 + pos)); |
| 162 |
|
| 163 |
sr |= ((src.swz[0] << NVFX_VP(SRC_SWZ_X_SHIFT)) | |
| 164 |
(src.swz[1] << NVFX_VP(SRC_SWZ_Y_SHIFT)) | |
| 165 |
(src.swz[2] << NVFX_VP(SRC_SWZ_Z_SHIFT)) | |
| 166 |
(src.swz[3] << NVFX_VP(SRC_SWZ_W_SHIFT))); |
| 167 |
|
| 168 |
if(src.indirect) { |
| 169 |
if(src.reg.type == NVFXSR_CONST) |
| 170 |
hw[3] |= NVFX_VP(INST_INDEX_CONST); |
| 171 |
else if(src.reg.type == NVFXSR_INPUT) |
| 172 |
hw[0] |= NVFX_VP(INST_INDEX_INPUT); |
| 173 |
else |
| 174 |
assert(0); |
| 175 |
|
| 176 |
if(src.indirect_reg) |
| 177 |
hw[0] |= NVFX_VP(INST_ADDR_REG_SELECT_1); |
| 178 |
hw[0] |= src.indirect_swz << NVFX_VP(INST_ADDR_SWZ_SHIFT); |
| 179 |
} |
| 180 |
|
| 181 |
switch (pos) { |
| 182 |
case 0: |
| 183 |
hw[1] |= ((sr & NVFX_VP(SRC0_HIGH_MASK)) >> |
| 184 |
NVFX_VP(SRC0_HIGH_SHIFT)) << NVFX_VP(INST_SRC0H_SHIFT); |
| 185 |
hw[2] |= (sr & NVFX_VP(SRC0_LOW_MASK)) << |
| 186 |
NVFX_VP(INST_SRC0L_SHIFT); |
| 187 |
break; |
| 188 |
case 1: |
| 189 |
hw[2] |= sr << NVFX_VP(INST_SRC1_SHIFT); |
| 190 |
break; |
| 191 |
case 2: |
| 192 |
hw[2] |= ((sr & NVFX_VP(SRC2_HIGH_MASK)) >> |
| 193 |
NVFX_VP(SRC2_HIGH_SHIFT)) << NVFX_VP(INST_SRC2H_SHIFT); |
| 194 |
hw[3] |= (sr & NVFX_VP(SRC2_LOW_MASK)) << |
| 195 |
NVFX_VP(INST_SRC2L_SHIFT); |
| 196 |
break; |
| 197 |
default: |
| 198 |
assert(0); |
| 199 |
} |
| 200 |
} |
| 201 |
|
| 202 |
static void |
| 203 |
emit_dst(struct nvfx_vpc *vpc, uint32_t *hw, |
| 204 |
int slot, struct nvfx_reg dst) |
| 205 |
{ |
| 206 |
struct nv30_vertprog *vp = vpc->vp; |
| 207 |
|
| 208 |
switch (dst.type) { |
| 209 |
case NVFXSR_NONE: |
| 210 |
if(!vpc->is_nv4x) |
| 211 |
hw[0] |= NV30_VP_INST_DEST_TEMP_ID_MASK; |
| 212 |
else { |
| 213 |
hw[3] |= NV40_VP_INST_DEST_MASK; |
| 214 |
if (slot == 0) |
| 215 |
hw[0] |= NV40_VP_INST_VEC_DEST_TEMP_MASK; |
| 216 |
else |
| 217 |
hw[3] |= NV40_VP_INST_SCA_DEST_TEMP_MASK; |
| 218 |
} |
| 219 |
break; |
| 220 |
case NVFXSR_TEMP: |
| 221 |
if(!vpc->is_nv4x) |
| 222 |
hw[0] |= (dst.index << NV30_VP_INST_DEST_TEMP_ID_SHIFT); |
| 223 |
else { |
| 224 |
hw[3] |= NV40_VP_INST_DEST_MASK; |
| 225 |
if (slot == 0) |
| 226 |
hw[0] |= (dst.index << NV40_VP_INST_VEC_DEST_TEMP_SHIFT); |
| 227 |
else |
| 228 |
hw[3] |= (dst.index << NV40_VP_INST_SCA_DEST_TEMP_SHIFT); |
| 229 |
} |
| 230 |
break; |
| 231 |
case NVFXSR_OUTPUT: |
| 232 |
/* TODO: this may be wrong because on nv30 COL0 and BFC0 are swapped */ |
| 233 |
if(vpc->is_nv4x) { |
| 234 |
switch (dst.index) { |
| 235 |
case NV30_VP_INST_DEST_CLP(0): |
| 236 |
dst.index = NVFX_VP(INST_DEST_FOGC); |
| 237 |
vp->or |= (1 << 6); |
| 238 |
break; |
| 239 |
case NV30_VP_INST_DEST_CLP(1): |
| 240 |
dst.index = NVFX_VP(INST_DEST_FOGC); |
| 241 |
vp->or |= (1 << 7); |
| 242 |
break; |
| 243 |
case NV30_VP_INST_DEST_CLP(2): |
| 244 |
dst.index = NVFX_VP(INST_DEST_FOGC); |
| 245 |
vp->or |= (1 << 8); |
| 246 |
break; |
| 247 |
case NV30_VP_INST_DEST_CLP(3): |
| 248 |
dst.index = NVFX_VP(INST_DEST_PSZ); |
| 249 |
vp->or |= (1 << 9); |
| 250 |
break; |
| 251 |
case NV30_VP_INST_DEST_CLP(4): |
| 252 |
dst.index = NVFX_VP(INST_DEST_PSZ); |
| 253 |
vp->or |= (1 << 10); |
| 254 |
break; |
| 255 |
case NV30_VP_INST_DEST_CLP(5): |
| 256 |
dst.index = NVFX_VP(INST_DEST_PSZ); |
| 257 |
vp->or |= (1 << 11); |
| 258 |
break; |
| 259 |
case NV40_VP_INST_DEST_COL0: vp->or |= (1 << 0); break; |
| 260 |
case NV40_VP_INST_DEST_COL1: vp->or |= (1 << 1); break; |
| 261 |
case NV40_VP_INST_DEST_BFC0: vp->or |= (1 << 2); break; |
| 262 |
case NV40_VP_INST_DEST_BFC1: vp->or |= (1 << 3); break; |
| 263 |
case NV40_VP_INST_DEST_FOGC: vp->or |= (1 << 4); break; |
| 264 |
case NV40_VP_INST_DEST_PSZ : vp->or |= (1 << 5); break; |
| 265 |
} |
| 266 |
} |
| 267 |
|
| 268 |
if(!vpc->is_nv4x) { |
| 269 |
hw[3] |= (dst.index << NV30_VP_INST_DEST_SHIFT); |
| 270 |
hw[0] |= NV30_VP_INST_VEC_DEST_TEMP_MASK; |
| 271 |
|
| 272 |
/*XXX: no way this is entirely correct, someone needs to |
| 273 |
* figure out what exactly it is. |
| 274 |
*/ |
| 275 |
hw[3] |= 0x800; |
| 276 |
} else { |
| 277 |
hw[3] |= (dst.index << NV40_VP_INST_DEST_SHIFT); |
| 278 |
if (slot == 0) { |
| 279 |
hw[0] |= NV40_VP_INST_VEC_RESULT; |
| 280 |
hw[0] |= NV40_VP_INST_VEC_DEST_TEMP_MASK; |
| 281 |
} else { |
| 282 |
hw[3] |= NV40_VP_INST_SCA_RESULT; |
| 283 |
hw[3] |= NV40_VP_INST_SCA_DEST_TEMP_MASK; |
| 284 |
} |
| 285 |
} |
| 286 |
break; |
| 287 |
default: |
| 288 |
assert(0); |
| 289 |
} |
| 290 |
} |
| 291 |
|
| 292 |
static void |
| 293 |
nvfx_vp_emit(struct nvfx_vpc *vpc, struct nvfx_insn insn) |
| 294 |
{ |
| 295 |
struct nv30_vertprog *vp = vpc->vp; |
| 296 |
unsigned slot = insn.op >> 7; |
| 297 |
unsigned op = insn.op & 0x7f; |
| 298 |
uint32_t *hw; |
| 299 |
|
| 300 |
vp->insns = realloc(vp->insns, ++vp->nr_insns * sizeof(*vpc->vpi)); |
| 301 |
vpc->vpi = &vp->insns[vp->nr_insns - 1]; |
| 302 |
memset(vpc->vpi, 0, sizeof(*vpc->vpi)); |
| 303 |
|
| 304 |
hw = vpc->vpi->data; |
| 305 |
|
| 306 |
if (insn.cc_test != NVFX_COND_TR) |
| 307 |
hw[0] |= NVFX_VP(INST_COND_TEST_ENABLE); |
| 308 |
hw[0] |= (insn.cc_test << NVFX_VP(INST_COND_SHIFT)); |
| 309 |
hw[0] |= ((insn.cc_swz[0] << NVFX_VP(INST_COND_SWZ_X_SHIFT)) | |
| 310 |
(insn.cc_swz[1] << NVFX_VP(INST_COND_SWZ_Y_SHIFT)) | |
| 311 |
(insn.cc_swz[2] << NVFX_VP(INST_COND_SWZ_Z_SHIFT)) | |
| 312 |
(insn.cc_swz[3] << NVFX_VP(INST_COND_SWZ_W_SHIFT))); |
| 313 |
if(insn.cc_update) |
| 314 |
hw[0] |= NVFX_VP(INST_COND_UPDATE_ENABLE); |
| 315 |
|
| 316 |
if(insn.sat) { |
| 317 |
assert(vpc->is_nv4x); |
| 318 |
if(vpc->is_nv4x) |
| 319 |
hw[0] |= NV40_VP_INST_SATURATE; |
| 320 |
} |
| 321 |
|
| 322 |
if(!vpc->is_nv4x) { |
| 323 |
if(slot == 0) |
| 324 |
hw[1] |= (op << NV30_VP_INST_VEC_OPCODE_SHIFT); |
| 325 |
else { |
| 326 |
hw[0] |= ((op >> 4) << NV30_VP_INST_SCA_OPCODEH_SHIFT); |
| 327 |
hw[1] |= ((op & 0xf) << NV30_VP_INST_SCA_OPCODEL_SHIFT); |
| 328 |
} |
| 329 |
// hw[3] |= NVFX_VP(INST_SCA_DEST_TEMP_MASK); |
| 330 |
// hw[3] |= (mask << NVFX_VP(INST_VEC_WRITEMASK_SHIFT)); |
| 331 |
|
| 332 |
if (insn.dst.type == NVFXSR_OUTPUT) { |
| 333 |
if (slot) |
| 334 |
hw[3] |= (insn.mask << NV30_VP_INST_SDEST_WRITEMASK_SHIFT); |
| 335 |
else |
| 336 |
hw[3] |= (insn.mask << NV30_VP_INST_VDEST_WRITEMASK_SHIFT); |
| 337 |
} else { |
| 338 |
if (slot) |
| 339 |
hw[3] |= (insn.mask << NV30_VP_INST_STEMP_WRITEMASK_SHIFT); |
| 340 |
else |
| 341 |
hw[3] |= (insn.mask << NV30_VP_INST_VTEMP_WRITEMASK_SHIFT); |
| 342 |
} |
| 343 |
} else { |
| 344 |
if (slot == 0) { |
| 345 |
hw[1] |= (op << NV40_VP_INST_VEC_OPCODE_SHIFT); |
| 346 |
hw[3] |= NV40_VP_INST_SCA_DEST_TEMP_MASK; |
| 347 |
hw[3] |= (insn.mask << NV40_VP_INST_VEC_WRITEMASK_SHIFT); |
| 348 |
} else { |
| 349 |
hw[1] |= (op << NV40_VP_INST_SCA_OPCODE_SHIFT); |
| 350 |
hw[0] |= NV40_VP_INST_VEC_DEST_TEMP_MASK ; |
| 351 |
hw[3] |= (insn.mask << NV40_VP_INST_SCA_WRITEMASK_SHIFT); |
| 352 |
} |
| 353 |
} |
| 354 |
|
| 355 |
emit_dst(vpc, hw, slot, insn.dst); |
| 356 |
emit_src(vpc, hw, 0, insn.src[0]); |
| 357 |
emit_src(vpc, hw, 1, insn.src[1]); |
| 358 |
emit_src(vpc, hw, 2, insn.src[2]); |
| 359 |
|
| 360 |
// if(insn.src[0].indirect || op == NVFX_VP_INST_VEC_OP_ARL) |
| 361 |
// hw[3] |= NV40_VP_INST_SCA_RESULT; |
| 362 |
} |
| 363 |
|
| 364 |
static inline struct nvfx_src |
| 365 |
tgsi_src(struct nvfx_vpc *vpc, const struct tgsi_full_src_register *fsrc) { |
| 366 |
struct nvfx_src src; |
| 367 |
|
| 368 |
switch (fsrc->Register.File) { |
| 369 |
case TGSI_FILE_INPUT: |
| 370 |
src.reg = nvfx_reg(NVFXSR_INPUT, fsrc->Register.Index); |
| 371 |
break; |
| 372 |
case TGSI_FILE_CONSTANT: |
| 373 |
if(fsrc->Register.Indirect) { |
| 374 |
src.reg = vpc->r_const[0]; |
| 375 |
src.reg.index = fsrc->Register.Index; |
| 376 |
} else { |
| 377 |
src.reg = vpc->r_const[fsrc->Register.Index]; |
| 378 |
} |
| 379 |
break; |
| 380 |
case TGSI_FILE_IMMEDIATE: |
| 381 |
src.reg = vpc->imm[fsrc->Register.Index]; |
| 382 |
break; |
| 383 |
case TGSI_FILE_TEMPORARY: |
| 384 |
src.reg = vpc->r_temp[fsrc->Register.Index]; |
| 385 |
break; |
| 386 |
default: |
| 387 |
NOUVEAU_ERR("bad src file\n"); |
| 388 |
src.reg.index = 0; |
| 389 |
src.reg.type = -1; |
| 390 |
break; |
| 391 |
} |
| 392 |
|
| 393 |
src.abs = fsrc->Register.Absolute; |
| 394 |
src.negate = fsrc->Register.Negate; |
| 395 |
src.swz[0] = fsrc->Register.SwizzleX; |
| 396 |
src.swz[1] = fsrc->Register.SwizzleY; |
| 397 |
src.swz[2] = fsrc->Register.SwizzleZ; |
| 398 |
src.swz[3] = fsrc->Register.SwizzleW; |
| 399 |
src.indirect = 0; |
| 400 |
src.indirect_reg = 0; |
| 401 |
src.indirect_swz = 0; |
| 402 |
|
| 403 |
if(fsrc->Register.Indirect) { |
| 404 |
if(fsrc->Indirect.File == TGSI_FILE_ADDRESS && |
| 405 |
(fsrc->Register.File == TGSI_FILE_CONSTANT || |
| 406 |
fsrc->Register.File == TGSI_FILE_INPUT)) { |
| 407 |
src.indirect = 1; |
| 408 |
src.indirect_reg = fsrc->Indirect.Index; |
| 409 |
src.indirect_swz = fsrc->Indirect.Swizzle; |
| 410 |
} else { |
| 411 |
src.reg.index = 0; |
| 412 |
src.reg.type = -1; |
| 413 |
} |
| 414 |
} |
| 415 |
|
| 416 |
return src; |
| 417 |
} |
| 418 |
|
| 419 |
static inline struct nvfx_reg |
| 420 |
tgsi_dst(struct nvfx_vpc *vpc, const struct tgsi_full_dst_register *fdst) { |
| 421 |
struct nvfx_reg dst; |
| 422 |
|
| 423 |
switch (fdst->Register.File) { |
| 424 |
case TGSI_FILE_NULL: |
| 425 |
dst = nvfx_reg(NVFXSR_NONE, 0); |
| 426 |
break; |
| 427 |
case TGSI_FILE_OUTPUT: |
| 428 |
dst = vpc->r_result[fdst->Register.Index]; |
| 429 |
break; |
| 430 |
case TGSI_FILE_TEMPORARY: |
| 431 |
dst = vpc->r_temp[fdst->Register.Index]; |
| 432 |
break; |
| 433 |
case TGSI_FILE_ADDRESS: |
| 434 |
dst = vpc->r_address[fdst->Register.Index]; |
| 435 |
break; |
| 436 |
default: |
| 437 |
NOUVEAU_ERR("bad dst file %i\n", fdst->Register.File); |
| 438 |
dst.index = 0; |
| 439 |
dst.type = 0; |
| 440 |
break; |
| 441 |
} |
| 442 |
|
| 443 |
return dst; |
| 444 |
} |
| 445 |
|
| 446 |
static inline int |
| 447 |
tgsi_mask(uint tgsi) |
| 448 |
{ |
| 449 |
int mask = 0; |
| 450 |
|
| 451 |
if (tgsi & TGSI_WRITEMASK_X) mask |= NVFX_VP_MASK_X; |
| 452 |
if (tgsi & TGSI_WRITEMASK_Y) mask |= NVFX_VP_MASK_Y; |
| 453 |
if (tgsi & TGSI_WRITEMASK_Z) mask |= NVFX_VP_MASK_Z; |
| 454 |
if (tgsi & TGSI_WRITEMASK_W) mask |= NVFX_VP_MASK_W; |
| 455 |
return mask; |
| 456 |
} |
| 457 |
|
| 458 |
static bool |
| 459 |
nvfx_vertprog_parse_instruction(struct nvfx_vpc *vpc, |
| 460 |
unsigned idx, const struct tgsi_full_instruction *finst) |
| 461 |
{ |
| 462 |
struct nvfx_src src[3], tmp; |
| 463 |
struct nvfx_reg dst; |
| 464 |
struct nvfx_reg final_dst; |
| 465 |
struct nvfx_src none = nvfx_src(nvfx_reg(NVFXSR_NONE, 0)); |
| 466 |
struct nvfx_insn insn; |
| 467 |
struct nvfx_relocation reloc; |
| 468 |
struct nvfx_loop_entry loop; |
| 469 |
bool sat = false; |
| 470 |
int mask; |
| 471 |
int ai = -1, ci = -1, ii = -1; |
| 472 |
int i; |
| 473 |
unsigned sub_depth = 0; |
| 474 |
|
| 475 |
for (i = 0; i < finst->Instruction.NumSrcRegs; i++) { |
| 476 |
const struct tgsi_full_src_register *fsrc; |
| 477 |
|
| 478 |
fsrc = &finst->Src[i]; |
| 479 |
if (fsrc->Register.File == TGSI_FILE_TEMPORARY) { |
| 480 |
src[i] = tgsi_src(vpc, fsrc); |
| 481 |
} |
| 482 |
} |
| 483 |
|
| 484 |
for (i = 0; i < finst->Instruction.NumSrcRegs; i++) { |
| 485 |
const struct tgsi_full_src_register *fsrc; |
| 486 |
|
| 487 |
fsrc = &finst->Src[i]; |
| 488 |
|
| 489 |
switch (fsrc->Register.File) { |
| 490 |
case TGSI_FILE_INPUT: |
| 491 |
if (ai == -1 || ai == fsrc->Register.Index) { |
| 492 |
ai = fsrc->Register.Index; |
| 493 |
src[i] = tgsi_src(vpc, fsrc); |
| 494 |
} else { |
| 495 |
src[i] = nvfx_src(temp(vpc)); |
| 496 |
nvfx_vp_emit(vpc, arith(0, VEC, MOV, src[i].reg, NVFX_VP_MASK_ALL, |
| 497 |
tgsi_src(vpc, fsrc), none, none)); |
| 498 |
} |
| 499 |
break; |
| 500 |
case TGSI_FILE_CONSTANT: |
| 501 |
if ((ci == -1 && ii == -1) || |
| 502 |
ci == fsrc->Register.Index) { |
| 503 |
ci = fsrc->Register.Index; |
| 504 |
src[i] = tgsi_src(vpc, fsrc); |
| 505 |
} else { |
| 506 |
src[i] = nvfx_src(temp(vpc)); |
| 507 |
nvfx_vp_emit(vpc, arith(0, VEC, MOV, src[i].reg, NVFX_VP_MASK_ALL, |
| 508 |
tgsi_src(vpc, fsrc), none, none)); |
| 509 |
} |
| 510 |
break; |
| 511 |
case TGSI_FILE_IMMEDIATE: |
| 512 |
if ((ci == -1 && ii == -1) || |
| 513 |
ii == fsrc->Register.Index) { |
| 514 |
ii = fsrc->Register.Index; |
| 515 |
src[i] = tgsi_src(vpc, fsrc); |
| 516 |
} else { |
| 517 |
src[i] = nvfx_src(temp(vpc)); |
| 518 |
nvfx_vp_emit(vpc, arith(0, VEC, MOV, src[i].reg, NVFX_VP_MASK_ALL, |
| 519 |
tgsi_src(vpc, fsrc), none, none)); |
| 520 |
} |
| 521 |
break; |
| 522 |
case TGSI_FILE_TEMPORARY: |
| 523 |
/* handled above */ |
| 524 |
break; |
| 525 |
default: |
| 526 |
NOUVEAU_ERR("bad src file\n"); |
| 527 |
return false; |
| 528 |
} |
| 529 |
} |
| 530 |
|
| 531 |
for (i = 0; i < finst->Instruction.NumSrcRegs; i++) { |
| 532 |
if(src[i].reg.type < 0) |
| 533 |
return false; |
| 534 |
} |
| 535 |
|
| 536 |
if(finst->Dst[0].Register.File == TGSI_FILE_ADDRESS && |
| 537 |
finst->Instruction.Opcode != TGSI_OPCODE_ARL) |
| 538 |
return false; |
| 539 |
|
| 540 |
final_dst = dst = tgsi_dst(vpc, &finst->Dst[0]); |
| 541 |
mask = tgsi_mask(finst->Dst[0].Register.WriteMask); |
| 542 |
if(finst->Instruction.Saturate) { |
| 543 |
assert(finst->Instruction.Opcode != TGSI_OPCODE_ARL); |
| 544 |
if (vpc->is_nv4x) |
| 545 |
sat = true; |
| 546 |
else |
| 547 |
if(dst.type != NVFXSR_TEMP) |
| 548 |
dst = temp(vpc); |
| 549 |
} |
| 550 |
|
| 551 |
switch (finst->Instruction.Opcode) { |
| 552 |
case TGSI_OPCODE_ADD: |
| 553 |
nvfx_vp_emit(vpc, arith(sat, VEC, ADD, dst, mask, src[0], none, src[1])); |
| 554 |
break; |
| 555 |
case TGSI_OPCODE_ARL: |
| 556 |
nvfx_vp_emit(vpc, arith(0, VEC, ARL, dst, mask, src[0], none, none)); |
| 557 |
break; |
| 558 |
case TGSI_OPCODE_CEIL: |
| 559 |
tmp = nvfx_src(temp(vpc)); |
| 560 |
nvfx_vp_emit(vpc, arith(0, VEC, FLR, tmp.reg, mask, neg(src[0]), none, none)); |
| 561 |
nvfx_vp_emit(vpc, arith(sat, VEC, MOV, dst, mask, neg(tmp), none, none)); |
| 562 |
break; |
| 563 |
case TGSI_OPCODE_CMP: |
| 564 |
insn = arith(0, VEC, MOV, none.reg, mask, src[0], none, none); |
| 565 |
insn.cc_update = 1; |
| 566 |
nvfx_vp_emit(vpc, insn); |
| 567 |
|
| 568 |
insn = arith(sat, VEC, MOV, dst, mask, src[2], none, none); |
| 569 |
insn.cc_test = NVFX_COND_GE; |
| 570 |
nvfx_vp_emit(vpc, insn); |
| 571 |
|
| 572 |
insn = arith(sat, VEC, MOV, dst, mask, src[1], none, none); |
| 573 |
insn.cc_test = NVFX_COND_LT; |
| 574 |
nvfx_vp_emit(vpc, insn); |
| 575 |
break; |
| 576 |
case TGSI_OPCODE_COS: |
| 577 |
nvfx_vp_emit(vpc, arith(sat, SCA, COS, dst, mask, none, none, src[0])); |
| 578 |
break; |
| 579 |
case TGSI_OPCODE_DP2: |
| 580 |
tmp = nvfx_src(temp(vpc)); |
| 581 |
nvfx_vp_emit(vpc, arith(0, VEC, MUL, tmp.reg, NVFX_VP_MASK_X | NVFX_VP_MASK_Y, src[0], src[1], none)); |
| 582 |
nvfx_vp_emit(vpc, arith(sat, VEC, ADD, dst, mask, swz(tmp, X, X, X, X), none, swz(tmp, Y, Y, Y, Y))); |
| 583 |
break; |
| 584 |
case TGSI_OPCODE_DP3: |
| 585 |
nvfx_vp_emit(vpc, arith(sat, VEC, DP3, dst, mask, src[0], src[1], none)); |
| 586 |
break; |
| 587 |
case TGSI_OPCODE_DP4: |
| 588 |
nvfx_vp_emit(vpc, arith(sat, VEC, DP4, dst, mask, src[0], src[1], none)); |
| 589 |
break; |
| 590 |
case TGSI_OPCODE_DST: |
| 591 |
nvfx_vp_emit(vpc, arith(sat, VEC, DST, dst, mask, src[0], src[1], none)); |
| 592 |
break; |
| 593 |
case TGSI_OPCODE_EX2: |
| 594 |
nvfx_vp_emit(vpc, arith(sat, SCA, EX2, dst, mask, none, none, src[0])); |
| 595 |
break; |
| 596 |
case TGSI_OPCODE_EXP: |
| 597 |
nvfx_vp_emit(vpc, arith(sat, SCA, EXP, dst, mask, none, none, src[0])); |
| 598 |
break; |
| 599 |
case TGSI_OPCODE_FLR: |
| 600 |
nvfx_vp_emit(vpc, arith(sat, VEC, FLR, dst, mask, src[0], none, none)); |
| 601 |
break; |
| 602 |
case TGSI_OPCODE_FRC: |
| 603 |
nvfx_vp_emit(vpc, arith(sat, VEC, FRC, dst, mask, src[0], none, none)); |
| 604 |
break; |
| 605 |
case TGSI_OPCODE_LG2: |
| 606 |
nvfx_vp_emit(vpc, arith(sat, SCA, LG2, dst, mask, none, none, src[0])); |
| 607 |
break; |
| 608 |
case TGSI_OPCODE_LIT: |
| 609 |
nvfx_vp_emit(vpc, arith(sat, SCA, LIT, dst, mask, none, none, src[0])); |
| 610 |
break; |
| 611 |
case TGSI_OPCODE_LOG: |
| 612 |
nvfx_vp_emit(vpc, arith(sat, SCA, LOG, dst, mask, none, none, src[0])); |
| 613 |
break; |
| 614 |
case TGSI_OPCODE_LRP: |
| 615 |
tmp = nvfx_src(temp(vpc)); |
| 616 |
nvfx_vp_emit(vpc, arith(0, VEC, MAD, tmp.reg, mask, neg(src[0]), src[2], src[2])); |
| 617 |
nvfx_vp_emit(vpc, arith(sat, VEC, MAD, dst, mask, src[0], src[1], tmp)); |
| 618 |
break; |
| 619 |
case TGSI_OPCODE_MAD: |
| 620 |
nvfx_vp_emit(vpc, arith(sat, VEC, MAD, dst, mask, src[0], src[1], src[2])); |
| 621 |
break; |
| 622 |
case TGSI_OPCODE_MAX: |
| 623 |
nvfx_vp_emit(vpc, arith(sat, VEC, MAX, dst, mask, src[0], src[1], none)); |
| 624 |
break; |
| 625 |
case TGSI_OPCODE_MIN: |
| 626 |
nvfx_vp_emit(vpc, arith(sat, VEC, MIN, dst, mask, src[0], src[1], none)); |
| 627 |
break; |
| 628 |
case TGSI_OPCODE_MOV: |
| 629 |
nvfx_vp_emit(vpc, arith(sat, VEC, MOV, dst, mask, src[0], none, none)); |
| 630 |
break; |
| 631 |
case TGSI_OPCODE_MUL: |
| 632 |
nvfx_vp_emit(vpc, arith(sat, VEC, MUL, dst, mask, src[0], src[1], none)); |
| 633 |
break; |
| 634 |
case TGSI_OPCODE_NOP: |
| 635 |
break; |
| 636 |
case TGSI_OPCODE_POW: |
| 637 |
tmp = nvfx_src(temp(vpc)); |
| 638 |
nvfx_vp_emit(vpc, arith(0, SCA, LG2, tmp.reg, NVFX_VP_MASK_X, none, none, swz(src[0], X, X, X, X))); |
| 639 |
nvfx_vp_emit(vpc, arith(0, VEC, MUL, tmp.reg, NVFX_VP_MASK_X, swz(tmp, X, X, X, X), swz(src[1], X, X, X, X), none)); |
| 640 |
nvfx_vp_emit(vpc, arith(sat, SCA, EX2, dst, mask, none, none, swz(tmp, X, X, X, X))); |
| 641 |
break; |
| 642 |
case TGSI_OPCODE_RCP: |
| 643 |
nvfx_vp_emit(vpc, arith(sat, SCA, RCP, dst, mask, none, none, src[0])); |
| 644 |
break; |
| 645 |
case TGSI_OPCODE_RSQ: |
| 646 |
nvfx_vp_emit(vpc, arith(sat, SCA, RSQ, dst, mask, none, none, abs(src[0]))); |
| 647 |
break; |
| 648 |
case TGSI_OPCODE_SEQ: |
| 649 |
nvfx_vp_emit(vpc, arith(sat, VEC, SEQ, dst, mask, src[0], src[1], none)); |
| 650 |
break; |
| 651 |
case TGSI_OPCODE_SGE: |
| 652 |
nvfx_vp_emit(vpc, arith(sat, VEC, SGE, dst, mask, src[0], src[1], none)); |
| 653 |
break; |
| 654 |
case TGSI_OPCODE_SGT: |
| 655 |
nvfx_vp_emit(vpc, arith(sat, VEC, SGT, dst, mask, src[0], src[1], none)); |
| 656 |
break; |
| 657 |
case TGSI_OPCODE_SIN: |
| 658 |
nvfx_vp_emit(vpc, arith(sat, SCA, SIN, dst, mask, none, none, src[0])); |
| 659 |
break; |
| 660 |
case TGSI_OPCODE_SLE: |
| 661 |
nvfx_vp_emit(vpc, arith(sat, VEC, SLE, dst, mask, src[0], src[1], none)); |
| 662 |
break; |
| 663 |
case TGSI_OPCODE_SLT: |
| 664 |
nvfx_vp_emit(vpc, arith(sat, VEC, SLT, dst, mask, src[0], src[1], none)); |
| 665 |
break; |
| 666 |
case TGSI_OPCODE_SNE: |
| 667 |
nvfx_vp_emit(vpc, arith(sat, VEC, SNE, dst, mask, src[0], src[1], none)); |
| 668 |
break; |
| 669 |
case TGSI_OPCODE_SSG: |
| 670 |
nvfx_vp_emit(vpc, arith(sat, VEC, SSG, dst, mask, src[0], none, none)); |
| 671 |
break; |
| 672 |
case TGSI_OPCODE_TRUNC: |
| 673 |
tmp = nvfx_src(temp(vpc)); |
| 674 |
insn = arith(0, VEC, MOV, none.reg, mask, src[0], none, none); |
| 675 |
insn.cc_update = 1; |
| 676 |
nvfx_vp_emit(vpc, insn); |
| 677 |
|
| 678 |
nvfx_vp_emit(vpc, arith(0, VEC, FLR, tmp.reg, mask, abs(src[0]), none, none)); |
| 679 |
nvfx_vp_emit(vpc, arith(sat, VEC, MOV, dst, mask, tmp, none, none)); |
| 680 |
|
| 681 |
insn = arith(sat, VEC, MOV, dst, mask, neg(tmp), none, none); |
| 682 |
insn.cc_test = NVFX_COND_LT; |
| 683 |
nvfx_vp_emit(vpc, insn); |
| 684 |
break; |
| 685 |
case TGSI_OPCODE_IF: |
| 686 |
insn = arith(0, VEC, MOV, none.reg, NVFX_VP_MASK_X, src[0], none, none); |
| 687 |
insn.cc_update = 1; |
| 688 |
nvfx_vp_emit(vpc, insn); |
| 689 |
|
| 690 |
reloc.location = vpc->vp->nr_insns; |
| 691 |
reloc.target = finst->Label.Label + 1; |
| 692 |
util_dynarray_append(&vpc->label_relocs, reloc); |
| 693 |
|
| 694 |
insn = arith(0, SCA, BRA, none.reg, 0, none, none, none); |
| 695 |
insn.cc_test = NVFX_COND_EQ; |
| 696 |
insn.cc_swz[0] = insn.cc_swz[1] = insn.cc_swz[2] = insn.cc_swz[3] = 0; |
| 697 |
nvfx_vp_emit(vpc, insn); |
| 698 |
break; |
| 699 |
case TGSI_OPCODE_ELSE: |
| 700 |
case TGSI_OPCODE_CAL: |
| 701 |
reloc.location = vpc->vp->nr_insns; |
| 702 |
reloc.target = finst->Label.Label; |
| 703 |
util_dynarray_append(&vpc->label_relocs, reloc); |
| 704 |
|
| 705 |
if(finst->Instruction.Opcode == TGSI_OPCODE_CAL) |
| 706 |
insn = arith(0, SCA, CAL, none.reg, 0, none, none, none); |
| 707 |
else |
| 708 |
insn = arith(0, SCA, BRA, none.reg, 0, none, none, none); |
| 709 |
nvfx_vp_emit(vpc, insn); |
| 710 |
break; |
| 711 |
case TGSI_OPCODE_RET: |
| 712 |
if(sub_depth || !vpc->vp->enabled_ucps) { |
| 713 |
tmp = none; |
| 714 |
tmp.swz[0] = tmp.swz[1] = tmp.swz[2] = tmp.swz[3] = 0; |
| 715 |
nvfx_vp_emit(vpc, arith(0, SCA, RET, none.reg, 0, none, none, tmp)); |
| 716 |
} else { |
| 717 |
reloc.location = vpc->vp->nr_insns; |
| 718 |
reloc.target = vpc->info->num_instructions; |
| 719 |
util_dynarray_append(&vpc->label_relocs, reloc); |
| 720 |
nvfx_vp_emit(vpc, arith(0, SCA, BRA, none.reg, 0, none, none, none)); |
| 721 |
} |
| 722 |
break; |
| 723 |
case TGSI_OPCODE_BGNSUB: |
| 724 |
++sub_depth; |
| 725 |
break; |
| 726 |
case TGSI_OPCODE_ENDSUB: |
| 727 |
--sub_depth; |
| 728 |
break; |
| 729 |
case TGSI_OPCODE_ENDIF: |
| 730 |
/* nothing to do here */ |
| 731 |
break; |
| 732 |
case TGSI_OPCODE_BGNLOOP: |
| 733 |
loop.cont_target = idx; |
| 734 |
loop.brk_target = finst->Label.Label + 1; |
| 735 |
util_dynarray_append(&vpc->loop_stack, loop); |
| 736 |
break; |
| 737 |
case TGSI_OPCODE_ENDLOOP: |
| 738 |
loop = util_dynarray_pop(&vpc->loop_stack, struct nvfx_loop_entry); |
| 739 |
|
| 740 |
reloc.location = vpc->vp->nr_insns; |
| 741 |
reloc.target = loop.cont_target; |
| 742 |
util_dynarray_append(&vpc->label_relocs, reloc); |
| 743 |
|
| 744 |
nvfx_vp_emit(vpc, arith(0, SCA, BRA, none.reg, 0, none, none, none)); |
| 745 |
break; |
| 746 |
case TGSI_OPCODE_CONT: |
| 747 |
loop = util_dynarray_top(&vpc->loop_stack, struct nvfx_loop_entry); |
| 748 |
|
| 749 |
reloc.location = vpc->vp->nr_insns; |
| 750 |
reloc.target = loop.cont_target; |
| 751 |
util_dynarray_append(&vpc->label_relocs, reloc); |
| 752 |
|
| 753 |
nvfx_vp_emit(vpc, arith(0, SCA, BRA, none.reg, 0, none, none, none)); |
| 754 |
break; |
| 755 |
case TGSI_OPCODE_BRK: |
| 756 |
loop = util_dynarray_top(&vpc->loop_stack, struct nvfx_loop_entry); |
| 757 |
|
| 758 |
reloc.location = vpc->vp->nr_insns; |
| 759 |
reloc.target = loop.brk_target; |
| 760 |
util_dynarray_append(&vpc->label_relocs, reloc); |
| 761 |
|
| 762 |
nvfx_vp_emit(vpc, arith(0, SCA, BRA, none.reg, 0, none, none, none)); |
| 763 |
break; |
| 764 |
case TGSI_OPCODE_END: |
| 765 |
assert(!sub_depth); |
| 766 |
if(vpc->vp->enabled_ucps) { |
| 767 |
if(idx != (vpc->info->num_instructions - 1)) { |
| 768 |
reloc.location = vpc->vp->nr_insns; |
| 769 |
reloc.target = vpc->info->num_instructions; |
| 770 |
util_dynarray_append(&vpc->label_relocs, reloc); |
| 771 |
nvfx_vp_emit(vpc, arith(0, SCA, BRA, none.reg, 0, none, none, none)); |
| 772 |
} |
| 773 |
} else { |
| 774 |
if(vpc->vp->nr_insns) |
| 775 |
vpc->vp->insns[vpc->vp->nr_insns - 1].data[3] |= NVFX_VP_INST_LAST; |
| 776 |
nvfx_vp_emit(vpc, arith(0, VEC, NOP, none.reg, 0, none, none, none)); |
| 777 |
vpc->vp->insns[vpc->vp->nr_insns - 1].data[3] |= NVFX_VP_INST_LAST; |
| 778 |
} |
| 779 |
break; |
| 780 |
default: |
| 781 |
NOUVEAU_ERR("invalid opcode %d\n", finst->Instruction.Opcode); |
| 782 |
return false; |
| 783 |
} |
| 784 |
|
| 785 |
if(finst->Instruction.Saturate && !vpc->is_nv4x) { |
| 786 |
if (!vpc->r_0_1.type) |
| 787 |
vpc->r_0_1 = constant(vpc, -1, 0, 1, 0, 0); |
| 788 |
nvfx_vp_emit(vpc, arith(0, VEC, MAX, dst, mask, nvfx_src(dst), swz(nvfx_src(vpc->r_0_1), X, X, X, X), none)); |
| 789 |
nvfx_vp_emit(vpc, arith(0, VEC, MIN, final_dst, mask, nvfx_src(dst), swz(nvfx_src(vpc->r_0_1), Y, Y, Y, Y), none)); |
| 790 |
} |
| 791 |
|
| 792 |
release_temps(vpc); |
| 793 |
return true; |
| 794 |
} |
| 795 |
|
| 796 |
static bool |
| 797 |
nvfx_vertprog_parse_decl_output(struct nvfx_vpc *vpc, |
| 798 |
const struct tgsi_full_declaration *fdec) |
| 799 |
{ |
| 800 |
unsigned num_texcoords = vpc->is_nv4x ? 10 : 8; |
| 801 |
unsigned idx = fdec->Range.First; |
| 802 |
unsigned semantic_index = fdec->Semantic.Index; |
| 803 |
int hw = 0, i; |
| 804 |
|
| 805 |
switch (fdec->Semantic.Name) { |
| 806 |
case TGSI_SEMANTIC_POSITION: |
| 807 |
hw = NVFX_VP(INST_DEST_POS); |
| 808 |
vpc->hpos_idx = idx; |
| 809 |
break; |
| 810 |
case TGSI_SEMANTIC_CLIPVERTEX: |
| 811 |
vpc->r_result[idx] = temp(vpc); |
| 812 |
vpc->r_temps_discard = 0; |
| 813 |
vpc->cvtx_idx = idx; |
| 814 |
return true; |
| 815 |
case TGSI_SEMANTIC_COLOR: |
| 816 |
if (fdec->Semantic.Index == 0) { |
| 817 |
hw = NVFX_VP(INST_DEST_COL0); |
| 818 |
} else |
| 819 |
if (fdec->Semantic.Index == 1) { |
| 820 |
hw = NVFX_VP(INST_DEST_COL1); |
| 821 |
} else { |
| 822 |
NOUVEAU_ERR("bad colour semantic index\n"); |
| 823 |
return false; |
| 824 |
} |
| 825 |
break; |
| 826 |
case TGSI_SEMANTIC_BCOLOR: |
| 827 |
if (fdec->Semantic.Index == 0) { |
| 828 |
hw = NVFX_VP(INST_DEST_BFC0); |
| 829 |
} else |
| 830 |
if (fdec->Semantic.Index == 1) { |
| 831 |
hw = NVFX_VP(INST_DEST_BFC1); |
| 832 |
} else { |
| 833 |
NOUVEAU_ERR("bad bcolour semantic index\n"); |
| 834 |
return false; |
| 835 |
} |
| 836 |
break; |
| 837 |
case TGSI_SEMANTIC_FOG: |
| 838 |
hw = NVFX_VP(INST_DEST_FOGC); |
| 839 |
break; |
| 840 |
case TGSI_SEMANTIC_PSIZE: |
| 841 |
hw = NVFX_VP(INST_DEST_PSZ); |
| 842 |
break; |
| 843 |
case TGSI_SEMANTIC_GENERIC: |
| 844 |
/* this is really an identifier for VP/FP linkage */ |
| 845 |
semantic_index += 8; |
| 846 |
FALLTHROUGH; |
| 847 |
case TGSI_SEMANTIC_TEXCOORD: |
| 848 |
for (i = 0; i < num_texcoords; i++) { |
| 849 |
if (vpc->vp->texcoord[i] == semantic_index) { |
| 850 |
hw = NVFX_VP(INST_DEST_TC(i)); |
| 851 |
break; |
| 852 |
} |
| 853 |
} |
| 854 |
|
| 855 |
if (i == num_texcoords) { |
| 856 |
vpc->r_result[idx] = nvfx_reg(NVFXSR_NONE, 0); |
| 857 |
return true; |
| 858 |
} |
| 859 |
break; |
| 860 |
case TGSI_SEMANTIC_EDGEFLAG: |
| 861 |
vpc->r_result[idx] = nvfx_reg(NVFXSR_NONE, 0); |
| 862 |
return true; |
| 863 |
default: |
| 864 |
NOUVEAU_ERR("bad output semantic\n"); |
| 865 |
return false; |
| 866 |
} |
| 867 |
|
| 868 |
vpc->r_result[idx] = nvfx_reg(NVFXSR_OUTPUT, hw); |
| 869 |
return true; |
| 870 |
} |
| 871 |
|
| 872 |
static bool |
| 873 |
nvfx_vertprog_prepare(struct nvfx_vpc *vpc) |
| 874 |
{ |
| 875 |
struct tgsi_parse_context p; |
| 876 |
int high_const = -1, high_temp = -1, high_addr = -1, nr_imm = 0, i; |
| 877 |
|
| 878 |
tgsi_parse_init(&p, vpc->pipe.tokens); |
| 879 |
while (!tgsi_parse_end_of_tokens(&p)) { |
| 880 |
const union tgsi_full_token *tok = &p.FullToken; |
| 881 |
|
| 882 |
tgsi_parse_token(&p); |
| 883 |
switch(tok->Token.Type) { |
| 884 |
case TGSI_TOKEN_TYPE_IMMEDIATE: |
| 885 |
nr_imm++; |
| 886 |
break; |
| 887 |
case TGSI_TOKEN_TYPE_DECLARATION: |
| 888 |
{ |
| 889 |
const struct tgsi_full_declaration *fdec; |
| 890 |
|
| 891 |
fdec = &p.FullToken.FullDeclaration; |
| 892 |
switch (fdec->Declaration.File) { |
| 893 |
case TGSI_FILE_TEMPORARY: |
| 894 |
if (fdec->Range.Last > high_temp) { |
| 895 |
high_temp = |
| 896 |
fdec->Range.Last; |
| 897 |
} |
| 898 |
break; |
| 899 |
case TGSI_FILE_ADDRESS: |
| 900 |
if (fdec->Range.Last > high_addr) { |
| 901 |
high_addr = |
| 902 |
fdec->Range.Last; |
| 903 |
} |
| 904 |
break; |
| 905 |
case TGSI_FILE_CONSTANT: |
| 906 |
if (fdec->Range.Last > high_const) { |
| 907 |
high_const = |
| 908 |
fdec->Range.Last; |
| 909 |
} |
| 910 |
break; |
| 911 |
case TGSI_FILE_OUTPUT: |
| 912 |
if (!nvfx_vertprog_parse_decl_output(vpc, fdec)) |
| 913 |
return false; |
| 914 |
break; |
| 915 |
default: |
| 916 |
break; |
| 917 |
} |
| 918 |
} |
| 919 |
break; |
| 920 |
default: |
| 921 |
break; |
| 922 |
} |
| 923 |
} |
| 924 |
tgsi_parse_free(&p); |
| 925 |
|
| 926 |
if (nr_imm) { |
| 927 |
vpc->imm = CALLOC(nr_imm, sizeof(struct nvfx_reg)); |
| 928 |
assert(vpc->imm); |
| 929 |
} |
| 930 |
|
| 931 |
if (++high_temp) { |
| 932 |
vpc->r_temp = CALLOC(high_temp, sizeof(struct nvfx_reg)); |
| 933 |
for (i = 0; i < high_temp; i++) |
| 934 |
vpc->r_temp[i] = temp(vpc); |
| 935 |
} |
| 936 |
|
| 937 |
if (++high_addr) { |
| 938 |
vpc->r_address = CALLOC(high_addr, sizeof(struct nvfx_reg)); |
| 939 |
for (i = 0; i < high_addr; i++) |
| 940 |
vpc->r_address[i] = nvfx_reg(NVFXSR_TEMP, i); |
| 941 |
} |
| 942 |
|
| 943 |
if(++high_const) { |
| 944 |
vpc->r_const = CALLOC(high_const, sizeof(struct nvfx_reg)); |
| 945 |
for (i = 0; i < high_const; i++) |
| 946 |
vpc->r_const[i] = constant(vpc, i, 0, 0, 0, 0); |
| 947 |
} |
| 948 |
|
| 949 |
vpc->r_temps_discard = 0; |
| 950 |
return true; |
| 951 |
} |
| 952 |
|
| 953 |
DEBUG_GET_ONCE_BOOL_OPTION(nvfx_dump_vp, "NVFX_DUMP_VP", false) |
| 954 |
|
| 955 |
bool |
| 956 |
_nvfx_vertprog_translate(uint16_t oclass, struct nv30_vertprog *vp) |
| 957 |
{ |
| 958 |
struct tgsi_parse_context parse; |
| 959 |
struct nvfx_vpc *vpc = NULL; |
| 960 |
struct nvfx_src none = nvfx_src(nvfx_reg(NVFXSR_NONE, 0)); |
| 961 |
struct util_dynarray insns; |
| 962 |
int i, ucps; |
| 963 |
|
| 964 |
vp->translated = false; |
| 965 |
vp->nr_insns = 0; |
| 966 |
vp->nr_consts = 0; |
| 967 |
|
| 968 |
vpc = CALLOC_STRUCT(nvfx_vpc); |
| 969 |
if (!vpc) |
| 970 |
return false; |
| 971 |
vpc->is_nv4x = (oclass >= NV40_3D_CLASS) ? ~0 : 0; |
| 972 |
vpc->vp = vp; |
| 973 |
vpc->pipe = vp->pipe; |
| 974 |
vpc->info = &vp->info; |
| 975 |
vpc->cvtx_idx = -1; |
| 976 |
|
| 977 |
if (!nvfx_vertprog_prepare(vpc)) { |
| 978 |
FREE(vpc); |
| 979 |
return false; |
| 980 |
} |
| 981 |
|
| 982 |
/* Redirect post-transform vertex position to a temp if user clip |
| 983 |
* planes are enabled. We need to append code to the vtxprog |
| 984 |
* to handle clip planes later. |
| 985 |
*/ |
| 986 |
if (vp->enabled_ucps && vpc->cvtx_idx < 0) { |
| 987 |
vpc->r_result[vpc->hpos_idx] = temp(vpc); |
| 988 |
vpc->r_temps_discard = 0; |
| 989 |
vpc->cvtx_idx = vpc->hpos_idx; |
| 990 |
} |
| 991 |
|
| 992 |
insns = UTIL_DYNARRAY_INIT; |
| 993 |
|
| 994 |
tgsi_parse_init(&parse, vp->pipe.tokens); |
| 995 |
while (!tgsi_parse_end_of_tokens(&parse)) { |
| 996 |
tgsi_parse_token(&parse); |
| 997 |
|
| 998 |
switch (parse.FullToken.Token.Type) { |
| 999 |
case TGSI_TOKEN_TYPE_IMMEDIATE: |
| 1000 |
{ |
| 1001 |
const struct tgsi_full_immediate *imm; |
| 1002 |
|
| 1003 |
imm = &parse.FullToken.FullImmediate; |
| 1004 |
assert(imm->Immediate.DataType == TGSI_IMM_FLOAT32); |
| 1005 |
assert(imm->Immediate.NrTokens == 4 + 1); |
| 1006 |
vpc->imm[vpc->nr_imm++] = |
| 1007 |
constant(vpc, -1, |
| 1008 |
imm->u[0].Float, |
| 1009 |
imm->u[1].Float, |
| 1010 |
imm->u[2].Float, |
| 1011 |
imm->u[3].Float); |
| 1012 |
} |
| 1013 |
break; |
| 1014 |
case TGSI_TOKEN_TYPE_INSTRUCTION: |
| 1015 |
{ |
| 1016 |
const struct tgsi_full_instruction *finst; |
| 1017 |
unsigned idx = insns.size >> 2; |
| 1018 |
util_dynarray_append(&insns, vp->nr_insns); |
| 1019 |
finst = &parse.FullToken.FullInstruction; |
| 1020 |
if (!nvfx_vertprog_parse_instruction(vpc, idx, finst)) |
| 1021 |
goto out; |
| 1022 |
} |
| 1023 |
break; |
| 1024 |
default: |
| 1025 |
break; |
| 1026 |
} |
| 1027 |
} |
| 1028 |
|
| 1029 |
util_dynarray_append(&insns, vp->nr_insns); |
| 1030 |
|
| 1031 |
for(unsigned i = 0; i < vpc->label_relocs.size; i += sizeof(struct nvfx_relocation)) |
| 1032 |
{ |
| 1033 |
struct nvfx_relocation* label_reloc = (struct nvfx_relocation*)((char*)vpc->label_relocs.data + i); |
| 1034 |
struct nvfx_relocation hw_reloc; |
| 1035 |
|
| 1036 |
hw_reloc.location = label_reloc->location; |
| 1037 |
hw_reloc.target = ((unsigned*)insns.data)[label_reloc->target]; |
| 1038 |
|
| 1039 |
//debug_printf("hw %u -> tgsi %u = hw %u\n", hw_reloc.location, label_reloc->target, hw_reloc.target); |
| 1040 |
|
| 1041 |
util_dynarray_append(&vp->branch_relocs, hw_reloc); |
| 1042 |
} |
| 1043 |
util_dynarray_fini(&insns); |
| 1044 |
util_dynarray_trim(&vp->branch_relocs); |
| 1045 |
|
| 1046 |
/* XXX: what if we add a RET before?! make sure we jump here...*/ |
| 1047 |
|
| 1048 |
/* Write out HPOS if it was redirected to a temp earlier */ |
| 1049 |
if (vpc->r_result[vpc->hpos_idx].type != NVFXSR_OUTPUT) { |
| 1050 |
struct nvfx_reg hpos = nvfx_reg(NVFXSR_OUTPUT, |
| 1051 |
NVFX_VP(INST_DEST_POS)); |
| 1052 |
struct nvfx_src htmp = nvfx_src(vpc->r_result[vpc->hpos_idx]); |
| 1053 |
|
| 1054 |
nvfx_vp_emit(vpc, arith(0, VEC, MOV, hpos, NVFX_VP_MASK_ALL, htmp, none, none)); |
| 1055 |
} |
| 1056 |
|
| 1057 |
/* Insert code to handle user clip planes */ |
| 1058 |
ucps = vp->enabled_ucps; |
| 1059 |
while (ucps) { |
| 1060 |
int i = ffs(ucps) - 1; ucps &= ~(1 << i); |
| 1061 |
struct nvfx_reg cdst = nvfx_reg(NVFXSR_OUTPUT, NV30_VP_INST_DEST_CLP(i)); |
| 1062 |
struct nvfx_src ceqn = nvfx_src(nvfx_reg(NVFXSR_CONST, 512 + i)); |
| 1063 |
struct nvfx_src htmp = nvfx_src(vpc->r_result[vpc->cvtx_idx]); |
| 1064 |
unsigned mask; |
| 1065 |
|
| 1066 |
if(vpc->is_nv4x) |
| 1067 |
{ |
| 1068 |
switch (i) { |
| 1069 |
case 0: case 3: mask = NVFX_VP_MASK_Y; break; |
| 1070 |
case 1: case 4: mask = NVFX_VP_MASK_Z; break; |
| 1071 |
case 2: case 5: mask = NVFX_VP_MASK_W; break; |
| 1072 |
default: |
| 1073 |
NOUVEAU_ERR("invalid clip dist #%d\n", i); |
| 1074 |
goto out; |
| 1075 |
} |
| 1076 |
} |
| 1077 |
else |
| 1078 |
mask = NVFX_VP_MASK_X; |
| 1079 |
|
| 1080 |
nvfx_vp_emit(vpc, arith(0, VEC, DP4, cdst, mask, htmp, ceqn, none)); |
| 1081 |
} |
| 1082 |
|
| 1083 |
if (vpc->vp->nr_insns) |
| 1084 |
vpc->vp->insns[vpc->vp->nr_insns - 1].data[3] |= NVFX_VP_INST_LAST; |
| 1085 |
|
| 1086 |
if(debug_get_option_nvfx_dump_vp()) |
| 1087 |
{ |
| 1088 |
debug_printf("\n"); |
| 1089 |
tgsi_dump(vpc->pipe.tokens, 0); |
| 1090 |
|
| 1091 |
debug_printf("\n%s vertex program:\n", vpc->is_nv4x ? "nv4x" : "nv3x"); |
| 1092 |
for (i = 0; i < vp->nr_insns; i++) |
| 1093 |
debug_printf("%3u: %08x %08x %08x %08x\n", i, vp->insns[i].data[0], vp->insns[i].data[1], vp->insns[i].data[2], vp->insns[i].data[3]); |
| 1094 |
debug_printf("\n"); |
| 1095 |
} |
| 1096 |
|
| 1097 |
vp->translated = true; |
| 1098 |
|
| 1099 |
out: |
| 1100 |
tgsi_parse_free(&parse); |
| 1101 |
if (vpc) { |
| 1102 |
util_dynarray_fini(&vpc->label_relocs); |
| 1103 |
util_dynarray_fini(&vpc->loop_stack); |
| 1104 |
FREE(vpc->r_temp); |
| 1105 |
FREE(vpc->r_address); |
| 1106 |
FREE(vpc->r_const); |
| 1107 |
FREE(vpc->imm); |
| 1108 |
FREE(vpc); |
| 1109 |
} |
| 1110 |
|
| 1111 |
return vp->translated; |
| 1112 |
} |