| 1 |
|
| 2 |
#include "pipe/p_context.h" |
| 3 |
#include "pipe/p_state.h" |
| 4 |
#include "util/u_inlines.h" |
| 5 |
#include "util/format/u_format.h" |
| 6 |
#include "translate/translate.h" |
| 7 |
|
| 8 |
#include "nvc0/nvc0_context.h" |
| 9 |
#include "nvc0/nvc0_resource.h" |
| 10 |
|
| 11 |
#include "nvc0/nvc0_3d.xml.h" |
| 12 |
|
| 13 |
struct push_context { |
| 14 |
struct nouveau_pushbuf *push; |
| 15 |
|
| 16 |
struct translate *translate; |
| 17 |
void *dest; |
| 18 |
const void *idxbuf; |
| 19 |
|
| 20 |
uint32_t vertex_size; |
| 21 |
uint32_t restart_index; |
| 22 |
uint32_t start_instance; |
| 23 |
uint32_t instance_id; |
| 24 |
|
| 25 |
bool prim_restart; |
| 26 |
bool need_vertex_id; |
| 27 |
|
| 28 |
struct { |
| 29 |
bool enabled; |
| 30 |
bool value; |
| 31 |
uint8_t width; |
| 32 |
unsigned stride; |
| 33 |
const uint8_t *data; |
| 34 |
} edgeflag; |
| 35 |
}; |
| 36 |
|
| 37 |
static void nvc0_push_upload_vertex_ids(struct push_context *, |
| 38 |
struct nvc0_context *, |
| 39 |
const struct pipe_draw_info *, |
| 40 |
const struct pipe_draw_start_count_bias *draw); |
| 41 |
|
| 42 |
static void |
| 43 |
nvc0_push_context_init(struct nvc0_context *nvc0, struct push_context *ctx) |
| 44 |
{ |
| 45 |
ctx->push = nvc0->base.pushbuf; |
| 46 |
|
| 47 |
ctx->translate = nvc0->vertex->translate; |
| 48 |
ctx->vertex_size = nvc0->vertex->size; |
| 49 |
ctx->instance_id = 0; |
| 50 |
|
| 51 |
ctx->need_vertex_id = |
| 52 |
nvc0->vertprog->vp.need_vertex_id && (nvc0->vertex->num_elements < 32); |
| 53 |
|
| 54 |
ctx->edgeflag.value = true; |
| 55 |
ctx->edgeflag.enabled = nvc0->vertprog->vp.edgeflag < PIPE_MAX_ATTRIBS; |
| 56 |
|
| 57 |
/* silence warnings */ |
| 58 |
ctx->edgeflag.data = NULL; |
| 59 |
ctx->edgeflag.stride = 0; |
| 60 |
ctx->edgeflag.width = 0; |
| 61 |
} |
| 62 |
|
| 63 |
static inline void |
| 64 |
nvc0_vertex_configure_translate(struct nvc0_context *nvc0, int32_t index_bias) |
| 65 |
{ |
| 66 |
struct translate *translate = nvc0->vertex->translate; |
| 67 |
unsigned i; |
| 68 |
|
| 69 |
for (i = 0; i < nvc0->num_vtxbufs; ++i) { |
| 70 |
const uint8_t *map; |
| 71 |
const struct pipe_vertex_buffer *vb = &nvc0->vtxbuf[i]; |
| 72 |
|
| 73 |
if (likely(vb->is_user_buffer)) |
| 74 |
map = (const uint8_t *)vb->buffer.user; |
| 75 |
else { |
| 76 |
if (!vb->buffer.resource) |
| 77 |
continue; |
| 78 |
|
| 79 |
map = nouveau_resource_map_offset(&nvc0->base, |
| 80 |
nv04_resource(vb->buffer.resource), vb->buffer_offset, NOUVEAU_BO_RD); |
| 81 |
} |
| 82 |
|
| 83 |
if (index_bias && !unlikely(nvc0->vertex->instance_bufs & (1 << i))) |
| 84 |
map += (intptr_t)index_bias * nvc0->vertex->strides[i]; |
| 85 |
|
| 86 |
translate->set_buffer(translate, i, map, nvc0->vertex->strides[i], ~0); |
| 87 |
} |
| 88 |
} |
| 89 |
|
| 90 |
static inline void |
| 91 |
nvc0_push_map_idxbuf(struct push_context *ctx, struct nvc0_context *nvc0, |
| 92 |
const struct pipe_draw_info *info) |
| 93 |
{ |
| 94 |
if (!info->has_user_indices) { |
| 95 |
struct nv04_resource *buf = nv04_resource(info->index.resource); |
| 96 |
ctx->idxbuf = nouveau_resource_map_offset( |
| 97 |
&nvc0->base, buf, 0, NOUVEAU_BO_RD); |
| 98 |
} else { |
| 99 |
ctx->idxbuf = info->index.user; |
| 100 |
} |
| 101 |
} |
| 102 |
|
| 103 |
static inline void |
| 104 |
nvc0_push_map_edgeflag(struct push_context *ctx, struct nvc0_context *nvc0, |
| 105 |
int32_t index_bias) |
| 106 |
{ |
| 107 |
unsigned attr = nvc0->vertprog->vp.edgeflag; |
| 108 |
struct pipe_vertex_element *ve = &nvc0->vertex->element[attr].pipe; |
| 109 |
struct pipe_vertex_buffer *vb = &nvc0->vtxbuf[ve->vertex_buffer_index]; |
| 110 |
struct nv04_resource *buf = nv04_resource(vb->buffer.resource); |
| 111 |
|
| 112 |
ctx->edgeflag.stride = ve->src_stride; |
| 113 |
ctx->edgeflag.width = util_format_get_blocksize(ve->src_format); |
| 114 |
if (!vb->is_user_buffer) { |
| 115 |
unsigned offset = vb->buffer_offset + ve->src_offset; |
| 116 |
ctx->edgeflag.data = nouveau_resource_map_offset(&nvc0->base, |
| 117 |
buf, offset, NOUVEAU_BO_RD); |
| 118 |
} else { |
| 119 |
ctx->edgeflag.data = (const uint8_t *)vb->buffer.user + ve->src_offset; |
| 120 |
} |
| 121 |
|
| 122 |
if (index_bias) |
| 123 |
ctx->edgeflag.data += (intptr_t)index_bias * ve->src_stride; |
| 124 |
} |
| 125 |
|
| 126 |
static inline unsigned |
| 127 |
prim_restart_search_i08(const uint8_t *elts, unsigned push, uint8_t index) |
| 128 |
{ |
| 129 |
unsigned i; |
| 130 |
for (i = 0; i < push && elts[i] != index; ++i); |
| 131 |
return i; |
| 132 |
} |
| 133 |
|
| 134 |
static inline unsigned |
| 135 |
prim_restart_search_i16(const uint16_t *elts, unsigned push, uint16_t index) |
| 136 |
{ |
| 137 |
unsigned i; |
| 138 |
for (i = 0; i < push && elts[i] != index; ++i); |
| 139 |
return i; |
| 140 |
} |
| 141 |
|
| 142 |
static inline unsigned |
| 143 |
prim_restart_search_i32(const uint32_t *elts, unsigned push, uint32_t index) |
| 144 |
{ |
| 145 |
unsigned i; |
| 146 |
for (i = 0; i < push && elts[i] != index; ++i); |
| 147 |
return i; |
| 148 |
} |
| 149 |
|
| 150 |
static inline bool |
| 151 |
ef_value_8(const struct push_context *ctx, uint32_t index) |
| 152 |
{ |
| 153 |
uint8_t *pf = (uint8_t *)&ctx->edgeflag.data[index * ctx->edgeflag.stride]; |
| 154 |
return !!*pf; |
| 155 |
} |
| 156 |
|
| 157 |
static inline bool |
| 158 |
ef_value_32(const struct push_context *ctx, uint32_t index) |
| 159 |
{ |
| 160 |
uint32_t *pf = (uint32_t *)&ctx->edgeflag.data[index * ctx->edgeflag.stride]; |
| 161 |
return !!*pf; |
| 162 |
} |
| 163 |
|
| 164 |
static inline bool |
| 165 |
ef_toggle(struct push_context *ctx) |
| 166 |
{ |
| 167 |
ctx->edgeflag.value = !ctx->edgeflag.value; |
| 168 |
return ctx->edgeflag.value; |
| 169 |
} |
| 170 |
|
| 171 |
static inline unsigned |
| 172 |
ef_toggle_search_i08(struct push_context *ctx, const uint8_t *elts, unsigned n) |
| 173 |
{ |
| 174 |
unsigned i; |
| 175 |
bool ef = ctx->edgeflag.value; |
| 176 |
if (ctx->edgeflag.width == 1) |
| 177 |
for (i = 0; i < n && ef_value_8(ctx, elts[i]) == ef; ++i); |
| 178 |
else |
| 179 |
for (i = 0; i < n && ef_value_32(ctx, elts[i]) == ef; ++i); |
| 180 |
return i; |
| 181 |
} |
| 182 |
|
| 183 |
static inline unsigned |
| 184 |
ef_toggle_search_i16(struct push_context *ctx, const uint16_t *elts, unsigned n) |
| 185 |
{ |
| 186 |
unsigned i; |
| 187 |
bool ef = ctx->edgeflag.value; |
| 188 |
if (ctx->edgeflag.width == 1) |
| 189 |
for (i = 0; i < n && ef_value_8(ctx, elts[i]) == ef; ++i); |
| 190 |
else |
| 191 |
for (i = 0; i < n && ef_value_32(ctx, elts[i]) == ef; ++i); |
| 192 |
return i; |
| 193 |
} |
| 194 |
|
| 195 |
static inline unsigned |
| 196 |
ef_toggle_search_i32(struct push_context *ctx, const uint32_t *elts, unsigned n) |
| 197 |
{ |
| 198 |
unsigned i; |
| 199 |
bool ef = ctx->edgeflag.value; |
| 200 |
if (ctx->edgeflag.width == 1) |
| 201 |
for (i = 0; i < n && ef_value_8(ctx, elts[i]) == ef; ++i); |
| 202 |
else |
| 203 |
for (i = 0; i < n && ef_value_32(ctx, elts[i]) == ef; ++i); |
| 204 |
return i; |
| 205 |
} |
| 206 |
|
| 207 |
static inline unsigned |
| 208 |
ef_toggle_search_seq(struct push_context *ctx, unsigned start, unsigned n) |
| 209 |
{ |
| 210 |
unsigned i; |
| 211 |
bool ef = ctx->edgeflag.value; |
| 212 |
if (ctx->edgeflag.width == 1) |
| 213 |
for (i = 0; i < n && ef_value_8(ctx, start++) == ef; ++i); |
| 214 |
else |
| 215 |
for (i = 0; i < n && ef_value_32(ctx, start++) == ef; ++i); |
| 216 |
return i; |
| 217 |
} |
| 218 |
|
| 219 |
static inline void * |
| 220 |
nvc0_push_setup_vertex_array(struct nvc0_context *nvc0, const unsigned count) |
| 221 |
{ |
| 222 |
struct nouveau_pushbuf *push = nvc0->base.pushbuf; |
| 223 |
struct nouveau_bo *bo; |
| 224 |
uint64_t va; |
| 225 |
const unsigned size = count * nvc0->vertex->size; |
| 226 |
|
| 227 |
void *const dest = nouveau_scratch_get(&nvc0->base, size, &va, &bo); |
| 228 |
|
| 229 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_START_HIGH(0)), 2); |
| 230 |
PUSH_DATAh(push, va); |
| 231 |
PUSH_DATA (push, va); |
| 232 |
|
| 233 |
if (nvc0->screen->eng3d->oclass < TU102_3D_CLASS) |
| 234 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_LIMIT_HIGH(0)), 2); |
| 235 |
else |
| 236 |
BEGIN_NVC0(push, SUBC_3D(TU102_3D_VERTEX_ARRAY_LIMIT_HIGH(0)), 2); |
| 237 |
PUSH_DATAh(push, va + size - 1); |
| 238 |
PUSH_DATA (push, va + size - 1); |
| 239 |
|
| 240 |
BCTX_REFN_bo(nvc0->bufctx_3d, 3D_VTX_TMP, NOUVEAU_BO_GART | NOUVEAU_BO_RD, |
| 241 |
bo); |
| 242 |
PUSH_VAL(push); |
| 243 |
|
| 244 |
return dest; |
| 245 |
} |
| 246 |
|
| 247 |
static void |
| 248 |
disp_vertices_i08(struct push_context *ctx, unsigned start, unsigned count) |
| 249 |
{ |
| 250 |
struct nouveau_pushbuf *push = ctx->push; |
| 251 |
struct translate *translate = ctx->translate; |
| 252 |
const uint8_t *restrict elts = (uint8_t *)ctx->idxbuf + start; |
| 253 |
unsigned pos = 0; |
| 254 |
|
| 255 |
do { |
| 256 |
unsigned nR = count; |
| 257 |
|
| 258 |
if (unlikely(ctx->prim_restart)) |
| 259 |
nR = prim_restart_search_i08(elts, nR, ctx->restart_index); |
| 260 |
|
| 261 |
translate->run_elts8(translate, elts, nR, |
| 262 |
ctx->start_instance, ctx->instance_id, ctx->dest); |
| 263 |
count -= nR; |
| 264 |
ctx->dest += nR * ctx->vertex_size; |
| 265 |
|
| 266 |
while (nR) { |
| 267 |
unsigned nE = nR; |
| 268 |
|
| 269 |
if (unlikely(ctx->edgeflag.enabled)) |
| 270 |
nE = ef_toggle_search_i08(ctx, elts, nR); |
| 271 |
|
| 272 |
PUSH_SPACE(push, 4); |
| 273 |
if (likely(nE >= 2)) { |
| 274 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_BUFFER_FIRST), 2); |
| 275 |
PUSH_DATA (push, pos); |
| 276 |
PUSH_DATA (push, nE); |
| 277 |
} else |
| 278 |
if (nE) { |
| 279 |
if (pos <= 0xff) { |
| 280 |
IMMED_NVC0(push, NVC0_3D(VB_ELEMENT_U32), pos); |
| 281 |
} else { |
| 282 |
BEGIN_NVC0(push, NVC0_3D(VB_ELEMENT_U32), 1); |
| 283 |
PUSH_DATA (push, pos); |
| 284 |
} |
| 285 |
} |
| 286 |
if (unlikely(nE != nR)) |
| 287 |
IMMED_NVC0(push, NVC0_3D(EDGEFLAG), ef_toggle(ctx)); |
| 288 |
|
| 289 |
pos += nE; |
| 290 |
elts += nE; |
| 291 |
nR -= nE; |
| 292 |
} |
| 293 |
if (count) { |
| 294 |
BEGIN_NVC0(push, NVC0_3D(VB_ELEMENT_U32), 1); |
| 295 |
PUSH_DATA (push, 0xffffffff); |
| 296 |
++elts; |
| 297 |
ctx->dest += ctx->vertex_size; |
| 298 |
++pos; |
| 299 |
--count; |
| 300 |
} |
| 301 |
} while (count); |
| 302 |
} |
| 303 |
|
| 304 |
static void |
| 305 |
disp_vertices_i16(struct push_context *ctx, unsigned start, unsigned count) |
| 306 |
{ |
| 307 |
struct nouveau_pushbuf *push = ctx->push; |
| 308 |
struct translate *translate = ctx->translate; |
| 309 |
const uint16_t *restrict elts = (uint16_t *)ctx->idxbuf + start; |
| 310 |
unsigned pos = 0; |
| 311 |
|
| 312 |
do { |
| 313 |
unsigned nR = count; |
| 314 |
|
| 315 |
if (unlikely(ctx->prim_restart)) |
| 316 |
nR = prim_restart_search_i16(elts, nR, ctx->restart_index); |
| 317 |
|
| 318 |
translate->run_elts16(translate, elts, nR, |
| 319 |
ctx->start_instance, ctx->instance_id, ctx->dest); |
| 320 |
count -= nR; |
| 321 |
ctx->dest += nR * ctx->vertex_size; |
| 322 |
|
| 323 |
while (nR) { |
| 324 |
unsigned nE = nR; |
| 325 |
|
| 326 |
if (unlikely(ctx->edgeflag.enabled)) |
| 327 |
nE = ef_toggle_search_i16(ctx, elts, nR); |
| 328 |
|
| 329 |
PUSH_SPACE(push, 4); |
| 330 |
if (likely(nE >= 2)) { |
| 331 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_BUFFER_FIRST), 2); |
| 332 |
PUSH_DATA (push, pos); |
| 333 |
PUSH_DATA (push, nE); |
| 334 |
} else |
| 335 |
if (nE) { |
| 336 |
if (pos <= 0xff) { |
| 337 |
IMMED_NVC0(push, NVC0_3D(VB_ELEMENT_U32), pos); |
| 338 |
} else { |
| 339 |
BEGIN_NVC0(push, NVC0_3D(VB_ELEMENT_U32), 1); |
| 340 |
PUSH_DATA (push, pos); |
| 341 |
} |
| 342 |
} |
| 343 |
if (unlikely(nE != nR)) |
| 344 |
IMMED_NVC0(push, NVC0_3D(EDGEFLAG), ef_toggle(ctx)); |
| 345 |
|
| 346 |
pos += nE; |
| 347 |
elts += nE; |
| 348 |
nR -= nE; |
| 349 |
} |
| 350 |
if (count) { |
| 351 |
BEGIN_NVC0(push, NVC0_3D(VB_ELEMENT_U32), 1); |
| 352 |
PUSH_DATA (push, 0xffffffff); |
| 353 |
++elts; |
| 354 |
ctx->dest += ctx->vertex_size; |
| 355 |
++pos; |
| 356 |
--count; |
| 357 |
} |
| 358 |
} while (count); |
| 359 |
} |
| 360 |
|
| 361 |
static void |
| 362 |
disp_vertices_i32(struct push_context *ctx, unsigned start, unsigned count) |
| 363 |
{ |
| 364 |
struct nouveau_pushbuf *push = ctx->push; |
| 365 |
struct translate *translate = ctx->translate; |
| 366 |
const uint32_t *restrict elts = (uint32_t *)ctx->idxbuf + start; |
| 367 |
unsigned pos = 0; |
| 368 |
|
| 369 |
do { |
| 370 |
unsigned nR = count; |
| 371 |
|
| 372 |
if (unlikely(ctx->prim_restart)) |
| 373 |
nR = prim_restart_search_i32(elts, nR, ctx->restart_index); |
| 374 |
|
| 375 |
translate->run_elts(translate, elts, nR, |
| 376 |
ctx->start_instance, ctx->instance_id, ctx->dest); |
| 377 |
count -= nR; |
| 378 |
ctx->dest += nR * ctx->vertex_size; |
| 379 |
|
| 380 |
while (nR) { |
| 381 |
unsigned nE = nR; |
| 382 |
|
| 383 |
if (unlikely(ctx->edgeflag.enabled)) |
| 384 |
nE = ef_toggle_search_i32(ctx, elts, nR); |
| 385 |
|
| 386 |
PUSH_SPACE(push, 4); |
| 387 |
if (likely(nE >= 2)) { |
| 388 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_BUFFER_FIRST), 2); |
| 389 |
PUSH_DATA (push, pos); |
| 390 |
PUSH_DATA (push, nE); |
| 391 |
} else |
| 392 |
if (nE) { |
| 393 |
if (pos <= 0xff) { |
| 394 |
IMMED_NVC0(push, NVC0_3D(VB_ELEMENT_U32), pos); |
| 395 |
} else { |
| 396 |
BEGIN_NVC0(push, NVC0_3D(VB_ELEMENT_U32), 1); |
| 397 |
PUSH_DATA (push, pos); |
| 398 |
} |
| 399 |
} |
| 400 |
if (unlikely(nE != nR)) |
| 401 |
IMMED_NVC0(push, NVC0_3D(EDGEFLAG), ef_toggle(ctx)); |
| 402 |
|
| 403 |
pos += nE; |
| 404 |
elts += nE; |
| 405 |
nR -= nE; |
| 406 |
} |
| 407 |
if (count) { |
| 408 |
BEGIN_NVC0(push, NVC0_3D(VB_ELEMENT_U32), 1); |
| 409 |
PUSH_DATA (push, 0xffffffff); |
| 410 |
++elts; |
| 411 |
ctx->dest += ctx->vertex_size; |
| 412 |
++pos; |
| 413 |
--count; |
| 414 |
} |
| 415 |
} while (count); |
| 416 |
} |
| 417 |
|
| 418 |
static void |
| 419 |
disp_vertices_seq(struct push_context *ctx, unsigned start, unsigned count) |
| 420 |
{ |
| 421 |
struct nouveau_pushbuf *push = ctx->push; |
| 422 |
struct translate *translate = ctx->translate; |
| 423 |
unsigned pos = 0; |
| 424 |
|
| 425 |
/* XXX: This will read the data corresponding to the primitive restart index, |
| 426 |
* maybe we should avoid that ? |
| 427 |
*/ |
| 428 |
translate->run(translate, start, count, |
| 429 |
ctx->start_instance, ctx->instance_id, ctx->dest); |
| 430 |
do { |
| 431 |
unsigned nr = count; |
| 432 |
|
| 433 |
if (unlikely(ctx->edgeflag.enabled)) |
| 434 |
nr = ef_toggle_search_seq(ctx, start + pos, nr); |
| 435 |
|
| 436 |
PUSH_SPACE(push, 4); |
| 437 |
if (likely(nr)) { |
| 438 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_BUFFER_FIRST), 2); |
| 439 |
PUSH_DATA (push, pos); |
| 440 |
PUSH_DATA (push, nr); |
| 441 |
} |
| 442 |
if (unlikely(nr != count)) |
| 443 |
IMMED_NVC0(push, NVC0_3D(EDGEFLAG), ef_toggle(ctx)); |
| 444 |
|
| 445 |
pos += nr; |
| 446 |
count -= nr; |
| 447 |
} while (count); |
| 448 |
} |
| 449 |
|
| 450 |
|
| 451 |
#define NVC0_PRIM_GL_CASE(n) \ |
| 452 |
case MESA_PRIM_##n: return NVC0_3D_VERTEX_BEGIN_GL_PRIMITIVE_##n |
| 453 |
|
| 454 |
static inline unsigned |
| 455 |
nvc0_prim_gl(unsigned prim) |
| 456 |
{ |
| 457 |
switch (prim) { |
| 458 |
NVC0_PRIM_GL_CASE(POINTS); |
| 459 |
NVC0_PRIM_GL_CASE(LINES); |
| 460 |
NVC0_PRIM_GL_CASE(LINE_LOOP); |
| 461 |
NVC0_PRIM_GL_CASE(LINE_STRIP); |
| 462 |
NVC0_PRIM_GL_CASE(TRIANGLES); |
| 463 |
NVC0_PRIM_GL_CASE(TRIANGLE_STRIP); |
| 464 |
NVC0_PRIM_GL_CASE(TRIANGLE_FAN); |
| 465 |
NVC0_PRIM_GL_CASE(QUADS); |
| 466 |
NVC0_PRIM_GL_CASE(QUAD_STRIP); |
| 467 |
NVC0_PRIM_GL_CASE(POLYGON); |
| 468 |
NVC0_PRIM_GL_CASE(LINES_ADJACENCY); |
| 469 |
NVC0_PRIM_GL_CASE(LINE_STRIP_ADJACENCY); |
| 470 |
NVC0_PRIM_GL_CASE(TRIANGLES_ADJACENCY); |
| 471 |
NVC0_PRIM_GL_CASE(TRIANGLE_STRIP_ADJACENCY); |
| 472 |
NVC0_PRIM_GL_CASE(PATCHES); |
| 473 |
default: |
| 474 |
return NVC0_3D_VERTEX_BEGIN_GL_PRIMITIVE_POINTS; |
| 475 |
} |
| 476 |
} |
| 477 |
|
| 478 |
typedef struct { |
| 479 |
uint32_t count; |
| 480 |
uint32_t primCount; |
| 481 |
uint32_t first; |
| 482 |
uint32_t baseInstance; |
| 483 |
} DrawArraysIndirectCommand; |
| 484 |
|
| 485 |
typedef struct { |
| 486 |
uint32_t count; |
| 487 |
uint32_t primCount; |
| 488 |
uint32_t firstIndex; |
| 489 |
int32_t baseVertex; |
| 490 |
uint32_t baseInstance; |
| 491 |
} DrawElementsIndirectCommand; |
| 492 |
|
| 493 |
void |
| 494 |
nvc0_push_vbo_indirect(struct nvc0_context *nvc0, const struct pipe_draw_info *info, |
| 495 |
unsigned drawid_offset, |
| 496 |
const struct pipe_draw_indirect_info *indirect, |
| 497 |
const struct pipe_draw_start_count_bias *draw) |
| 498 |
{ |
| 499 |
/* The strategy here is to just read the commands from the indirect buffer |
| 500 |
* and do the draws. This is suboptimal, but will only happen in the case |
| 501 |
* that conversion is required for FIXED or DOUBLE inputs. |
| 502 |
*/ |
| 503 |
struct nvc0_screen *screen = nvc0->screen; |
| 504 |
struct nouveau_pushbuf *push = nvc0->base.pushbuf; |
| 505 |
struct nv04_resource *buf = nv04_resource(indirect->buffer); |
| 506 |
struct nv04_resource *buf_count = nv04_resource(indirect->indirect_draw_count); |
| 507 |
unsigned i; |
| 508 |
|
| 509 |
unsigned draw_count = indirect->draw_count; |
| 510 |
if (buf_count) { |
| 511 |
uint32_t *count = nouveau_resource_map_offset( |
| 512 |
&nvc0->base, buf_count, indirect->indirect_draw_count_offset, |
| 513 |
NOUVEAU_BO_RD); |
| 514 |
draw_count = *count; |
| 515 |
} |
| 516 |
|
| 517 |
uint8_t *buf_data = nouveau_resource_map_offset( |
| 518 |
&nvc0->base, buf, indirect->offset, NOUVEAU_BO_RD); |
| 519 |
struct pipe_draw_info single = *info; |
| 520 |
struct pipe_draw_start_count_bias sdraw = *draw; |
| 521 |
for (i = 0; i < draw_count; i++, buf_data += indirect->stride) { |
| 522 |
if (info->index_size) { |
| 523 |
DrawElementsIndirectCommand *cmd = (void *)buf_data; |
| 524 |
sdraw.start = draw->start + cmd->firstIndex; |
| 525 |
sdraw.count = cmd->count; |
| 526 |
single.start_instance = cmd->baseInstance; |
| 527 |
single.instance_count = cmd->primCount; |
| 528 |
sdraw.index_bias = cmd->baseVertex; |
| 529 |
} else { |
| 530 |
DrawArraysIndirectCommand *cmd = (void *)buf_data; |
| 531 |
sdraw.start = cmd->first; |
| 532 |
sdraw.count = cmd->count; |
| 533 |
single.start_instance = cmd->baseInstance; |
| 534 |
single.instance_count = cmd->primCount; |
| 535 |
} |
| 536 |
|
| 537 |
if (nvc0->vertprog->vp.need_draw_parameters) { |
| 538 |
PUSH_SPACE(push, 9); |
| 539 |
BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3); |
| 540 |
PUSH_DATA (push, NVC0_CB_AUX_SIZE); |
| 541 |
PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(0)); |
| 542 |
PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(0)); |
| 543 |
BEGIN_1IC0(push, NVC0_3D(CB_POS), 1 + 3); |
| 544 |
PUSH_DATA (push, NVC0_CB_AUX_DRAW_INFO); |
| 545 |
PUSH_DATA (push, sdraw.index_bias); |
| 546 |
PUSH_DATA (push, single.start_instance); |
| 547 |
PUSH_DATA (push, drawid_offset + i); |
| 548 |
} |
| 549 |
|
| 550 |
nvc0_push_vbo(nvc0, &single, NULL, &sdraw); |
| 551 |
} |
| 552 |
|
| 553 |
nouveau_resource_unmap(buf); |
| 554 |
if (buf_count) |
| 555 |
nouveau_resource_unmap(buf_count); |
| 556 |
} |
| 557 |
|
| 558 |
void |
| 559 |
nvc0_push_vbo(struct nvc0_context *nvc0, const struct pipe_draw_info *info, |
| 560 |
const struct pipe_draw_indirect_info *indirect, |
| 561 |
const struct pipe_draw_start_count_bias *draw) |
| 562 |
{ |
| 563 |
struct push_context ctx; |
| 564 |
unsigned i, index_size; |
| 565 |
unsigned index_bias = info->index_size ? draw->index_bias : 0; |
| 566 |
unsigned inst_count = info->instance_count; |
| 567 |
unsigned vert_count = draw->count; |
| 568 |
unsigned prim; |
| 569 |
|
| 570 |
nvc0_push_context_init(nvc0, &ctx); |
| 571 |
|
| 572 |
nvc0_vertex_configure_translate(nvc0, index_bias); |
| 573 |
|
| 574 |
if (nvc0->state.index_bias) { |
| 575 |
/* this is already taken care of by translate */ |
| 576 |
IMMED_NVC0(ctx.push, NVC0_3D(VB_ELEMENT_BASE), 0); |
| 577 |
nvc0->state.index_bias = 0; |
| 578 |
} |
| 579 |
|
| 580 |
if (unlikely(ctx.edgeflag.enabled)) |
| 581 |
nvc0_push_map_edgeflag(&ctx, nvc0, index_bias); |
| 582 |
|
| 583 |
ctx.prim_restart = info->primitive_restart; |
| 584 |
ctx.restart_index = info->restart_index; |
| 585 |
|
| 586 |
if (info->primitive_restart) { |
| 587 |
/* NOTE: I hope we won't ever need that last index (~0). |
| 588 |
* If we do, we have to disable primitive restart here always and |
| 589 |
* use END,BEGIN to restart. (XXX: would that affect PrimitiveID ?) |
| 590 |
* We could also deactivate PRIM_RESTART_WITH_DRAW_ARRAYS temporarily, |
| 591 |
* and add manual restart to disp_vertices_seq. |
| 592 |
*/ |
| 593 |
BEGIN_NVC0(ctx.push, NVC0_3D(PRIM_RESTART_ENABLE), 2); |
| 594 |
PUSH_DATA (ctx.push, 1); |
| 595 |
PUSH_DATA (ctx.push, info->index_size ? 0xffffffff : info->restart_index); |
| 596 |
} else |
| 597 |
if (nvc0->state.prim_restart) { |
| 598 |
IMMED_NVC0(ctx.push, NVC0_3D(PRIM_RESTART_ENABLE), 0); |
| 599 |
} |
| 600 |
nvc0->state.prim_restart = info->primitive_restart; |
| 601 |
|
| 602 |
if (info->index_size) { |
| 603 |
nvc0_push_map_idxbuf(&ctx, nvc0, info); |
| 604 |
index_size = info->index_size; |
| 605 |
} else { |
| 606 |
if (unlikely(indirect && indirect->count_from_stream_output)) { |
| 607 |
struct pipe_context *pipe = &nvc0->base.pipe; |
| 608 |
struct nvc0_so_target *targ; |
| 609 |
targ = nvc0_so_target(indirect->count_from_stream_output); |
| 610 |
pipe->get_query_result(pipe, targ->pq, true, (void *)&vert_count); |
| 611 |
vert_count /= targ->stride; |
| 612 |
} |
| 613 |
ctx.idxbuf = NULL; /* shut up warnings */ |
| 614 |
index_size = 0; |
| 615 |
} |
| 616 |
|
| 617 |
ctx.start_instance = info->start_instance; |
| 618 |
|
| 619 |
prim = nvc0_prim_gl(info->mode); |
| 620 |
do { |
| 621 |
PUSH_SPACE(ctx.push, 9); |
| 622 |
|
| 623 |
ctx.dest = nvc0_push_setup_vertex_array(nvc0, vert_count); |
| 624 |
if (unlikely(!ctx.dest)) |
| 625 |
break; |
| 626 |
|
| 627 |
if (unlikely(ctx.need_vertex_id)) |
| 628 |
nvc0_push_upload_vertex_ids(&ctx, nvc0, info, draw); |
| 629 |
|
| 630 |
if (nvc0->screen->eng3d->oclass < GM107_3D_CLASS) |
| 631 |
IMMED_NVC0(ctx.push, NVC0_3D(VERTEX_ARRAY_FLUSH), 0); |
| 632 |
BEGIN_NVC0(ctx.push, NVC0_3D(VERTEX_BEGIN_GL), 1); |
| 633 |
PUSH_DATA (ctx.push, prim); |
| 634 |
switch (index_size) { |
| 635 |
case 1: |
| 636 |
disp_vertices_i08(&ctx, draw->start, vert_count); |
| 637 |
break; |
| 638 |
case 2: |
| 639 |
disp_vertices_i16(&ctx, draw->start, vert_count); |
| 640 |
break; |
| 641 |
case 4: |
| 642 |
disp_vertices_i32(&ctx, draw->start, vert_count); |
| 643 |
break; |
| 644 |
default: |
| 645 |
assert(index_size == 0); |
| 646 |
disp_vertices_seq(&ctx, draw->start, vert_count); |
| 647 |
break; |
| 648 |
} |
| 649 |
PUSH_SPACE(ctx.push, 1); |
| 650 |
IMMED_NVC0(ctx.push, NVC0_3D(VERTEX_END_GL), 0); |
| 651 |
|
| 652 |
if (--inst_count) { |
| 653 |
prim |= NVC0_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT; |
| 654 |
++ctx.instance_id; |
| 655 |
} |
| 656 |
nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_VTX_TMP); |
| 657 |
nouveau_scratch_done(&nvc0->base); |
| 658 |
} while (inst_count); |
| 659 |
|
| 660 |
|
| 661 |
/* reset state and unmap buffers (no-op) */ |
| 662 |
|
| 663 |
if (unlikely(!ctx.edgeflag.value)) { |
| 664 |
PUSH_SPACE(ctx.push, 1); |
| 665 |
IMMED_NVC0(ctx.push, NVC0_3D(EDGEFLAG), 1); |
| 666 |
} |
| 667 |
|
| 668 |
if (unlikely(ctx.need_vertex_id)) { |
| 669 |
PUSH_SPACE(ctx.push, 4); |
| 670 |
IMMED_NVC0(ctx.push, NVC0_3D(VERTEX_ID_REPLACE), 0); |
| 671 |
BEGIN_NVC0(ctx.push, NVC0_3D(VERTEX_ATTRIB_FORMAT(1)), 1); |
| 672 |
PUSH_DATA (ctx.push, |
| 673 |
NVC0_3D_VERTEX_ATTRIB_FORMAT_CONST | |
| 674 |
NVC0_3D_VERTEX_ATTRIB_FORMAT_TYPE_FLOAT | |
| 675 |
NVC0_3D_VERTEX_ATTRIB_FORMAT_SIZE_32); |
| 676 |
IMMED_NVC0(ctx.push, NVC0_3D(VERTEX_ARRAY_FETCH(1)), 0); |
| 677 |
} |
| 678 |
|
| 679 |
if (info->index_size && !info->has_user_indices) |
| 680 |
nouveau_resource_unmap(nv04_resource(info->index.resource)); |
| 681 |
for (i = 0; i < nvc0->num_vtxbufs; ++i) |
| 682 |
nouveau_resource_unmap(nv04_resource(nvc0->vtxbuf[i].buffer.resource)); |
| 683 |
|
| 684 |
NOUVEAU_DRV_STAT(&nvc0->screen->base, draw_calls_fallback_count, 1); |
| 685 |
} |
| 686 |
|
| 687 |
static inline void |
| 688 |
copy_indices_u8(uint32_t *dst, const uint8_t *elts, uint32_t bias, unsigned n) |
| 689 |
{ |
| 690 |
unsigned i; |
| 691 |
for (i = 0; i < n; ++i) |
| 692 |
dst[i] = elts[i] + bias; |
| 693 |
} |
| 694 |
|
| 695 |
static inline void |
| 696 |
copy_indices_u16(uint32_t *dst, const uint16_t *elts, uint32_t bias, unsigned n) |
| 697 |
{ |
| 698 |
unsigned i; |
| 699 |
for (i = 0; i < n; ++i) |
| 700 |
dst[i] = elts[i] + bias; |
| 701 |
} |
| 702 |
|
| 703 |
static inline void |
| 704 |
copy_indices_u32(uint32_t *dst, const uint32_t *elts, uint32_t bias, unsigned n) |
| 705 |
{ |
| 706 |
unsigned i; |
| 707 |
for (i = 0; i < n; ++i) |
| 708 |
dst[i] = elts[i] + bias; |
| 709 |
} |
| 710 |
|
| 711 |
static void |
| 712 |
nvc0_push_upload_vertex_ids(struct push_context *ctx, |
| 713 |
struct nvc0_context *nvc0, |
| 714 |
const struct pipe_draw_info *info, |
| 715 |
const struct pipe_draw_start_count_bias *draw) |
| 716 |
|
| 717 |
{ |
| 718 |
struct nouveau_pushbuf *push = ctx->push; |
| 719 |
struct nouveau_bo *bo; |
| 720 |
uint64_t va; |
| 721 |
uint32_t *data; |
| 722 |
uint32_t format; |
| 723 |
unsigned index_size = info->index_size; |
| 724 |
unsigned i; |
| 725 |
unsigned a = nvc0->vertex->num_elements; |
| 726 |
|
| 727 |
if (!index_size || draw->index_bias) |
| 728 |
index_size = 4; |
| 729 |
data = (uint32_t *)nouveau_scratch_get(&nvc0->base, |
| 730 |
draw->count * index_size, &va, &bo); |
| 731 |
|
| 732 |
BCTX_REFN_bo(nvc0->bufctx_3d, 3D_VTX_TMP, NOUVEAU_BO_GART | NOUVEAU_BO_RD, |
| 733 |
bo); |
| 734 |
PUSH_VAL(push); |
| 735 |
|
| 736 |
if (info->index_size) { |
| 737 |
if (!draw->index_bias) { |
| 738 |
memcpy(data, ctx->idxbuf, draw->count * index_size); |
| 739 |
} else { |
| 740 |
switch (info->index_size) { |
| 741 |
case 1: |
| 742 |
copy_indices_u8(data, ctx->idxbuf, draw->index_bias, draw->count); |
| 743 |
break; |
| 744 |
case 2: |
| 745 |
copy_indices_u16(data, ctx->idxbuf, draw->index_bias, draw->count); |
| 746 |
break; |
| 747 |
default: |
| 748 |
copy_indices_u32(data, ctx->idxbuf, draw->index_bias, draw->count); |
| 749 |
break; |
| 750 |
} |
| 751 |
} |
| 752 |
} else { |
| 753 |
for (i = 0; i < draw->count; ++i) |
| 754 |
data[i] = i + (draw->start + draw->index_bias); |
| 755 |
} |
| 756 |
|
| 757 |
format = (1 << NVC0_3D_VERTEX_ATTRIB_FORMAT_BUFFER__SHIFT) | |
| 758 |
NVC0_3D_VERTEX_ATTRIB_FORMAT_TYPE_UINT; |
| 759 |
|
| 760 |
switch (index_size) { |
| 761 |
case 1: |
| 762 |
format |= NVC0_3D_VERTEX_ATTRIB_FORMAT_SIZE_8; |
| 763 |
break; |
| 764 |
case 2: |
| 765 |
format |= NVC0_3D_VERTEX_ATTRIB_FORMAT_SIZE_16; |
| 766 |
break; |
| 767 |
default: |
| 768 |
format |= NVC0_3D_VERTEX_ATTRIB_FORMAT_SIZE_32; |
| 769 |
break; |
| 770 |
} |
| 771 |
|
| 772 |
PUSH_SPACE(push, 12); |
| 773 |
|
| 774 |
if (unlikely(nvc0->state.instance_elts & 2)) { |
| 775 |
nvc0->state.instance_elts &= ~2; |
| 776 |
IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_PER_INSTANCE(1)), 0); |
| 777 |
} |
| 778 |
|
| 779 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ATTRIB_FORMAT(a)), 1); |
| 780 |
PUSH_DATA (push, format); |
| 781 |
|
| 782 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(1)), 3); |
| 783 |
PUSH_DATA (push, NVC0_3D_VERTEX_ARRAY_FETCH_ENABLE | index_size); |
| 784 |
PUSH_DATAh(push, va); |
| 785 |
PUSH_DATA (push, va); |
| 786 |
|
| 787 |
if (nvc0->screen->eng3d->oclass < TU102_3D_CLASS) |
| 788 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_LIMIT_HIGH(1)), 2); |
| 789 |
else |
| 790 |
BEGIN_NVC0(push, SUBC_3D(TU102_3D_VERTEX_ARRAY_LIMIT_HIGH(1)), 2); |
| 791 |
PUSH_DATAh(push, va + draw->count * index_size - 1); |
| 792 |
PUSH_DATA (push, va + draw->count * index_size - 1); |
| 793 |
|
| 794 |
#define NVC0_3D_VERTEX_ID_REPLACE_SOURCE_ATTR_X(a) \ |
| 795 |
(((0x80 + (a) * 0x10) / 4) << NVC0_3D_VERTEX_ID_REPLACE_SOURCE__SHIFT) |
| 796 |
|
| 797 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ID_REPLACE), 1); |
| 798 |
PUSH_DATA (push, NVC0_3D_VERTEX_ID_REPLACE_SOURCE_ATTR_X(a) | 1); |
| 799 |
} |