| 1 |
/* |
| 2 |
* Copyright 2010 Christoph Bumiller |
| 3 |
* |
| 4 |
* Permission is hereby granted, free of charge, to any person obtaining a |
| 5 |
* copy of this software and associated documentation files (the "Software"), |
| 6 |
* to deal in the Software without restriction, including without limitation |
| 7 |
* the rights to use, copy, modify, merge, publish, distribute, sublicense, |
| 8 |
* and/or sell copies of the Software, and to permit persons to whom the |
| 9 |
* Software is furnished to do so, subject to the following conditions: |
| 10 |
* |
| 11 |
* The above copyright notice and this permission notice shall be included in |
| 12 |
* all copies or substantial portions of the Software. |
| 13 |
* |
| 14 |
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 15 |
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 16 |
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
| 17 |
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR |
| 18 |
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, |
| 19 |
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR |
| 20 |
* OTHER DEALINGS IN THE SOFTWARE. |
| 21 |
*/ |
| 22 |
|
| 23 |
#define NVC0_PUSH_EXPLICIT_SPACE_CHECKING |
| 24 |
|
| 25 |
#include "pipe/p_context.h" |
| 26 |
#include "pipe/p_state.h" |
| 27 |
#include "util/u_draw.h" |
| 28 |
#include "util/u_inlines.h" |
| 29 |
#include "util/format/u_format.h" |
| 30 |
#include "translate/translate.h" |
| 31 |
|
| 32 |
#include "nvc0/nvc0_context.h" |
| 33 |
#include "nvc0/nvc0_query_hw.h" |
| 34 |
#include "nvc0/nvc0_resource.h" |
| 35 |
|
| 36 |
#include "nvc0/nvc0_3d.xml.h" |
| 37 |
|
| 38 |
void |
| 39 |
nvc0_vertex_state_delete(struct pipe_context *pipe, |
| 40 |
void *hwcso) |
| 41 |
{ |
| 42 |
struct nvc0_vertex_stateobj *so = hwcso; |
| 43 |
|
| 44 |
if (so->translate) |
| 45 |
so->translate->release(so->translate); |
| 46 |
FREE(hwcso); |
| 47 |
} |
| 48 |
|
| 49 |
void * |
| 50 |
nvc0_vertex_state_create(struct pipe_context *pipe, |
| 51 |
unsigned num_elements, |
| 52 |
const struct pipe_vertex_element *elements) |
| 53 |
{ |
| 54 |
struct nvc0_context *nvc0 = nvc0_context(pipe); |
| 55 |
struct nvc0_vertex_stateobj *so; |
| 56 |
struct translate_key transkey; |
| 57 |
unsigned i; |
| 58 |
unsigned src_offset_max = 0; |
| 59 |
|
| 60 |
so = CALLOC(1, sizeof(*so) + |
| 61 |
num_elements * sizeof(struct nvc0_vertex_element)); |
| 62 |
if (!so) |
| 63 |
return NULL; |
| 64 |
so->num_elements = num_elements; |
| 65 |
so->instance_elts = 0; |
| 66 |
so->instance_bufs = 0; |
| 67 |
so->shared_slots = false; |
| 68 |
so->need_conversion = false; |
| 69 |
|
| 70 |
memset(so->vb_access_size, 0, sizeof(so->vb_access_size)); |
| 71 |
|
| 72 |
for (i = 0; i < PIPE_MAX_ATTRIBS; ++i) |
| 73 |
so->min_instance_div[i] = 0xffffffff; |
| 74 |
|
| 75 |
transkey.nr_elements = 0; |
| 76 |
transkey.output_stride = 0; |
| 77 |
|
| 78 |
for (i = 0; i < num_elements; ++i) { |
| 79 |
const struct pipe_vertex_element *ve = &elements[i]; |
| 80 |
const unsigned vbi = ve->vertex_buffer_index; |
| 81 |
unsigned size; |
| 82 |
enum pipe_format fmt = ve->src_format; |
| 83 |
|
| 84 |
so->element[i].pipe = elements[i]; |
| 85 |
so->element[i].state = nvc0_vertex_format[fmt].vtx; |
| 86 |
|
| 87 |
if (!so->element[i].state) { |
| 88 |
switch (util_format_get_nr_components(fmt)) { |
| 89 |
case 1: fmt = PIPE_FORMAT_R32_FLOAT; break; |
| 90 |
case 2: fmt = PIPE_FORMAT_R32G32_FLOAT; break; |
| 91 |
case 3: fmt = PIPE_FORMAT_R32G32B32_FLOAT; break; |
| 92 |
case 4: fmt = PIPE_FORMAT_R32G32B32A32_FLOAT; break; |
| 93 |
default: |
| 94 |
assert(0); |
| 95 |
FREE(so); |
| 96 |
return NULL; |
| 97 |
} |
| 98 |
so->element[i].state = nvc0_vertex_format[fmt].vtx; |
| 99 |
so->need_conversion = true; |
| 100 |
util_debug_message(&nouveau_context(pipe)->debug, FALLBACK, |
| 101 |
"Converting vertex element %d, no hw format %s", |
| 102 |
i, util_format_name(ve->src_format)); |
| 103 |
} |
| 104 |
size = util_format_get_blocksize(fmt); |
| 105 |
|
| 106 |
src_offset_max = MAX2(src_offset_max, ve->src_offset); |
| 107 |
|
| 108 |
if (so->vb_access_size[vbi] < (ve->src_offset + size)) |
| 109 |
so->vb_access_size[vbi] = ve->src_offset + size; |
| 110 |
|
| 111 |
if (unlikely(ve->instance_divisor)) { |
| 112 |
so->instance_elts |= 1 << i; |
| 113 |
so->instance_bufs |= 1 << vbi; |
| 114 |
if (ve->instance_divisor < so->min_instance_div[vbi]) |
| 115 |
so->min_instance_div[vbi] = ve->instance_divisor; |
| 116 |
} |
| 117 |
|
| 118 |
so->strides[vbi] = ve->src_stride; |
| 119 |
if (!ve->src_stride && nvc0->screen->eng3d->oclass < GM107_3D_CLASS) |
| 120 |
so->constant_vbos |= 1 << vbi; |
| 121 |
|
| 122 |
if (1) { |
| 123 |
unsigned ca; |
| 124 |
unsigned j = transkey.nr_elements++; |
| 125 |
|
| 126 |
ca = util_format_description(fmt)->channel[0].size / 8; |
| 127 |
if (ca != 1 && ca != 2) |
| 128 |
ca = 4; |
| 129 |
|
| 130 |
transkey.element[j].type = TRANSLATE_ELEMENT_NORMAL; |
| 131 |
transkey.element[j].input_format = ve->src_format; |
| 132 |
transkey.element[j].input_buffer = vbi; |
| 133 |
transkey.element[j].input_offset = ve->src_offset; |
| 134 |
transkey.element[j].instance_divisor = ve->instance_divisor; |
| 135 |
|
| 136 |
transkey.output_stride = align(transkey.output_stride, ca); |
| 137 |
transkey.element[j].output_format = fmt; |
| 138 |
transkey.element[j].output_offset = transkey.output_stride; |
| 139 |
transkey.output_stride += size; |
| 140 |
|
| 141 |
so->element[i].state_alt = so->element[i].state; |
| 142 |
so->element[i].state_alt |= transkey.element[j].output_offset << 7; |
| 143 |
} |
| 144 |
|
| 145 |
so->element[i].state |= i << NVC0_3D_VERTEX_ATTRIB_FORMAT_BUFFER__SHIFT; |
| 146 |
} |
| 147 |
transkey.output_stride = align(transkey.output_stride, 4); |
| 148 |
|
| 149 |
so->size = transkey.output_stride; |
| 150 |
so->translate = translate_create(&transkey); |
| 151 |
|
| 152 |
if (so->instance_elts || src_offset_max >= (1 << 14)) |
| 153 |
return so; |
| 154 |
so->shared_slots = true; |
| 155 |
|
| 156 |
for (i = 0; i < num_elements; ++i) { |
| 157 |
const unsigned b = elements[i].vertex_buffer_index; |
| 158 |
const unsigned s = elements[i].src_offset; |
| 159 |
so->element[i].state &= ~NVC0_3D_VERTEX_ATTRIB_FORMAT_BUFFER__MASK; |
| 160 |
so->element[i].state |= b << NVC0_3D_VERTEX_ATTRIB_FORMAT_BUFFER__SHIFT; |
| 161 |
so->element[i].state |= s << NVC0_3D_VERTEX_ATTRIB_FORMAT_OFFSET__SHIFT; |
| 162 |
} |
| 163 |
return so; |
| 164 |
} |
| 165 |
|
| 166 |
#define NVC0_3D_VERTEX_ATTRIB_INACTIVE \ |
| 167 |
NVC0_3D_VERTEX_ATTRIB_FORMAT_TYPE_FLOAT | \ |
| 168 |
NVC0_3D_VERTEX_ATTRIB_FORMAT_SIZE_32 | NVC0_3D_VERTEX_ATTRIB_FORMAT_CONST |
| 169 |
|
| 170 |
#define VTX_ATTR(a, c, t, s) \ |
| 171 |
((NVC0_3D_VTX_ATTR_DEFINE_TYPE_##t) | \ |
| 172 |
(NVC0_3D_VTX_ATTR_DEFINE_SIZE_##s) | \ |
| 173 |
((a) << NVC0_3D_VTX_ATTR_DEFINE_ATTR__SHIFT) | \ |
| 174 |
((c) << NVC0_3D_VTX_ATTR_DEFINE_COMP__SHIFT)) |
| 175 |
|
| 176 |
static void |
| 177 |
nvc0_set_constant_vertex_attrib(struct nvc0_context *nvc0, const unsigned a) |
| 178 |
{ |
| 179 |
struct nouveau_pushbuf *push = nvc0->base.pushbuf; |
| 180 |
struct pipe_vertex_element *ve = &nvc0->vertex->element[a].pipe; |
| 181 |
struct pipe_vertex_buffer *vb = &nvc0->vtxbuf[ve->vertex_buffer_index]; |
| 182 |
uint32_t mode; |
| 183 |
const struct util_format_description *desc; |
| 184 |
void *dst; |
| 185 |
const void *src = (const uint8_t *)vb->buffer.user + ve->src_offset; |
| 186 |
assert(vb->is_user_buffer); |
| 187 |
|
| 188 |
desc = util_format_description(ve->src_format); |
| 189 |
|
| 190 |
PUSH_SPACE(push, 6); |
| 191 |
BEGIN_NVC0(push, NVC0_3D(VTX_ATTR_DEFINE), 5); |
| 192 |
dst = &push->cur[1]; |
| 193 |
util_format_unpack_rgba(ve->src_format, dst, src, 1); |
| 194 |
if (desc->channel[0].pure_integer) { |
| 195 |
if (desc->channel[0].type == UTIL_FORMAT_TYPE_SIGNED) { |
| 196 |
mode = VTX_ATTR(a, 4, SINT, 32); |
| 197 |
} else { |
| 198 |
mode = VTX_ATTR(a, 4, UINT, 32); |
| 199 |
} |
| 200 |
} else { |
| 201 |
mode = VTX_ATTR(a, 4, FLOAT, 32); |
| 202 |
} |
| 203 |
push->cur[0] = mode; |
| 204 |
push->cur += 5; |
| 205 |
} |
| 206 |
|
| 207 |
static inline void |
| 208 |
nvc0_user_vbuf_range(struct nvc0_context *nvc0, int vbi, |
| 209 |
uint32_t *base, uint32_t *size) |
| 210 |
{ |
| 211 |
if (unlikely(nvc0->vertex->instance_bufs & (1 << vbi))) { |
| 212 |
const uint32_t div = nvc0->vertex->min_instance_div[vbi]; |
| 213 |
*base = nvc0->instance_off * nvc0->vertex->strides[vbi]; |
| 214 |
*size = (nvc0->instance_max / div) * nvc0->vertex->strides[vbi] + |
| 215 |
nvc0->vertex->vb_access_size[vbi]; |
| 216 |
} else { |
| 217 |
/* NOTE: if there are user buffers, we *must* have index bounds */ |
| 218 |
assert(nvc0->vb_elt_limit != ~0); |
| 219 |
*base = nvc0->vb_elt_first * nvc0->vertex->strides[vbi]; |
| 220 |
*size = nvc0->vb_elt_limit * nvc0->vertex->strides[vbi] + |
| 221 |
nvc0->vertex->vb_access_size[vbi]; |
| 222 |
} |
| 223 |
} |
| 224 |
|
| 225 |
static inline void |
| 226 |
nvc0_release_user_vbufs(struct nvc0_context *nvc0) |
| 227 |
{ |
| 228 |
if (nvc0->vbo_user) { |
| 229 |
nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_VTX_TMP); |
| 230 |
nouveau_scratch_done(&nvc0->base); |
| 231 |
} |
| 232 |
} |
| 233 |
|
| 234 |
static void |
| 235 |
nvc0_update_user_vbufs(struct nvc0_context *nvc0) |
| 236 |
{ |
| 237 |
uint64_t address[PIPE_MAX_ATTRIBS]; |
| 238 |
struct nouveau_pushbuf *push = nvc0->base.pushbuf; |
| 239 |
int i; |
| 240 |
uint32_t written = 0; |
| 241 |
|
| 242 |
PUSH_SPACE(push, nvc0->vertex->num_elements * 8); |
| 243 |
for (i = 0; i < nvc0->vertex->num_elements; ++i) { |
| 244 |
struct pipe_vertex_element *ve = &nvc0->vertex->element[i].pipe; |
| 245 |
const unsigned b = ve->vertex_buffer_index; |
| 246 |
struct pipe_vertex_buffer *vb = &nvc0->vtxbuf[b]; |
| 247 |
uint32_t base, size; |
| 248 |
|
| 249 |
if (!(nvc0->vbo_user & (1 << b))) |
| 250 |
continue; |
| 251 |
if (nvc0->constant_vbos & (1 << b)) { |
| 252 |
nvc0_set_constant_vertex_attrib(nvc0, i); |
| 253 |
continue; |
| 254 |
} |
| 255 |
nvc0_user_vbuf_range(nvc0, b, &base, &size); |
| 256 |
|
| 257 |
if (!(written & (1 << b))) { |
| 258 |
struct nouveau_bo *bo; |
| 259 |
const uint32_t bo_flags = NOUVEAU_BO_RD | NOUVEAU_BO_GART; |
| 260 |
written |= 1 << b; |
| 261 |
address[b] = nouveau_scratch_data(&nvc0->base, vb->buffer.user, |
| 262 |
base, size, &bo); |
| 263 |
if (bo) |
| 264 |
BCTX_REFN_bo(nvc0->bufctx_3d, 3D_VTX_TMP, bo_flags, bo); |
| 265 |
|
| 266 |
NOUVEAU_DRV_STAT(&nvc0->screen->base, user_buffer_upload_bytes, size); |
| 267 |
} |
| 268 |
|
| 269 |
BEGIN_1IC0(push, NVC0_3D(MACRO_VERTEX_ARRAY_SELECT), 5); |
| 270 |
PUSH_DATA (push, i); |
| 271 |
PUSH_DATAh(push, address[b] + base + size - 1); |
| 272 |
PUSH_DATA (push, address[b] + base + size - 1); |
| 273 |
PUSH_DATAh(push, address[b] + ve->src_offset); |
| 274 |
PUSH_DATA (push, address[b] + ve->src_offset); |
| 275 |
} |
| 276 |
nvc0->base.vbo_dirty = true; |
| 277 |
} |
| 278 |
|
| 279 |
static void |
| 280 |
nvc0_update_user_vbufs_shared(struct nvc0_context *nvc0) |
| 281 |
{ |
| 282 |
struct nouveau_pushbuf *push = nvc0->base.pushbuf; |
| 283 |
uint32_t mask = nvc0->vbo_user & ~nvc0->constant_vbos; |
| 284 |
|
| 285 |
PUSH_SPACE(push, nvc0->num_vtxbufs * 8); |
| 286 |
while (mask) { |
| 287 |
struct nouveau_bo *bo; |
| 288 |
const uint32_t bo_flags = NOUVEAU_BO_RD | NOUVEAU_BO_GART; |
| 289 |
uint64_t address; |
| 290 |
uint32_t base, size; |
| 291 |
const int b = ffs(mask) - 1; |
| 292 |
mask &= ~(1 << b); |
| 293 |
|
| 294 |
nvc0_user_vbuf_range(nvc0, b, &base, &size); |
| 295 |
|
| 296 |
address = nouveau_scratch_data(&nvc0->base, nvc0->vtxbuf[b].buffer.user, |
| 297 |
base, size, &bo); |
| 298 |
if (bo) |
| 299 |
BCTX_REFN_bo(nvc0->bufctx_3d, 3D_VTX_TMP, bo_flags, bo); |
| 300 |
|
| 301 |
BEGIN_1IC0(push, NVC0_3D(MACRO_VERTEX_ARRAY_SELECT), 5); |
| 302 |
PUSH_DATA (push, b); |
| 303 |
PUSH_DATAh(push, address + base + size - 1); |
| 304 |
PUSH_DATA (push, address + base + size - 1); |
| 305 |
PUSH_DATAh(push, address); |
| 306 |
PUSH_DATA (push, address); |
| 307 |
|
| 308 |
NOUVEAU_DRV_STAT(&nvc0->screen->base, user_buffer_upload_bytes, size); |
| 309 |
} |
| 310 |
|
| 311 |
mask = nvc0->state.constant_elts; |
| 312 |
while (mask) { |
| 313 |
int i = ffs(mask) - 1; |
| 314 |
mask &= ~(1 << i); |
| 315 |
nvc0_set_constant_vertex_attrib(nvc0, i); |
| 316 |
} |
| 317 |
} |
| 318 |
|
| 319 |
static void |
| 320 |
nvc0_validate_vertex_buffers(struct nvc0_context *nvc0) |
| 321 |
{ |
| 322 |
struct nouveau_pushbuf *push = nvc0->base.pushbuf; |
| 323 |
const struct nvc0_vertex_stateobj *vertex = nvc0->vertex; |
| 324 |
uint32_t refd = 0; |
| 325 |
unsigned i; |
| 326 |
|
| 327 |
PUSH_SPACE(push, vertex->num_elements * 8); |
| 328 |
for (i = 0; i < vertex->num_elements; ++i) { |
| 329 |
const struct nvc0_vertex_element *ve; |
| 330 |
const struct pipe_vertex_buffer *vb; |
| 331 |
struct nv04_resource *res; |
| 332 |
unsigned b; |
| 333 |
unsigned limit, offset; |
| 334 |
|
| 335 |
if (nvc0->state.constant_elts & (1 << i)) |
| 336 |
continue; |
| 337 |
ve = &vertex->element[i]; |
| 338 |
b = ve->pipe.vertex_buffer_index; |
| 339 |
vb = &nvc0->vtxbuf[b]; |
| 340 |
|
| 341 |
if (nvc0->vbo_user & (1 << b)) { |
| 342 |
if (!(nvc0->constant_vbos & (1 << b))) { |
| 343 |
if (ve->pipe.instance_divisor) { |
| 344 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_DIVISOR(i)), 1); |
| 345 |
PUSH_DATA (push, ve->pipe.instance_divisor); |
| 346 |
} |
| 347 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 1); |
| 348 |
PUSH_DATA (push, (1 << 12) | vertex->strides[b]); |
| 349 |
} |
| 350 |
/* address/value set in nvc0_update_user_vbufs */ |
| 351 |
continue; |
| 352 |
} |
| 353 |
res = nv04_resource(vb->buffer.resource); |
| 354 |
offset = ve->pipe.src_offset + vb->buffer_offset; |
| 355 |
limit = vb->buffer.resource->width0 - 1; |
| 356 |
|
| 357 |
if (unlikely(ve->pipe.instance_divisor)) { |
| 358 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 4); |
| 359 |
PUSH_DATA (push, NVC0_3D_VERTEX_ARRAY_FETCH_ENABLE | vertex->strides[b]); |
| 360 |
PUSH_DATAh(push, res->address + offset); |
| 361 |
PUSH_DATA (push, res->address + offset); |
| 362 |
PUSH_DATA (push, ve->pipe.instance_divisor); |
| 363 |
} else { |
| 364 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 3); |
| 365 |
PUSH_DATA (push, NVC0_3D_VERTEX_ARRAY_FETCH_ENABLE | vertex->strides[b]); |
| 366 |
PUSH_DATAh(push, res->address + offset); |
| 367 |
PUSH_DATA (push, res->address + offset); |
| 368 |
} |
| 369 |
|
| 370 |
if (nvc0->screen->eng3d->oclass < TU102_3D_CLASS) |
| 371 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_LIMIT_HIGH(i)), 2); |
| 372 |
else |
| 373 |
BEGIN_NVC0(push, SUBC_3D(TU102_3D_VERTEX_ARRAY_LIMIT_HIGH(i)), 2); |
| 374 |
PUSH_DATAh(push, res->address + limit); |
| 375 |
PUSH_DATA (push, res->address + limit); |
| 376 |
|
| 377 |
if (!(refd & (1 << b))) { |
| 378 |
refd |= 1 << b; |
| 379 |
BCTX_REFN(nvc0->bufctx_3d, 3D_VTX, res, RD); |
| 380 |
} |
| 381 |
} |
| 382 |
if (nvc0->vbo_user) |
| 383 |
nvc0_update_user_vbufs(nvc0); |
| 384 |
} |
| 385 |
|
| 386 |
static void |
| 387 |
nvc0_validate_vertex_buffers_shared(struct nvc0_context *nvc0) |
| 388 |
{ |
| 389 |
struct nouveau_pushbuf *push = nvc0->base.pushbuf; |
| 390 |
unsigned b; |
| 391 |
const uint32_t mask = nvc0->vbo_user; |
| 392 |
|
| 393 |
PUSH_SPACE(push, nvc0->num_vtxbufs * 8 + nvc0->vertex->num_elements); |
| 394 |
for (b = 0; b < nvc0->num_vtxbufs; ++b) { |
| 395 |
struct pipe_vertex_buffer *vb = &nvc0->vtxbuf[b]; |
| 396 |
struct nv04_resource *buf; |
| 397 |
uint32_t offset, limit; |
| 398 |
|
| 399 |
if (mask & (1 << b)) { |
| 400 |
if (!(nvc0->constant_vbos & (1 << b))) { |
| 401 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(b)), 1); |
| 402 |
PUSH_DATA (push, NVC0_3D_VERTEX_ARRAY_FETCH_ENABLE | nvc0->vertex->strides[b]); |
| 403 |
} |
| 404 |
/* address/value set in nvc0_update_user_vbufs_shared */ |
| 405 |
continue; |
| 406 |
} else if (!vb->buffer.resource) { |
| 407 |
/* there can be holes in the vertex buffer lists */ |
| 408 |
IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(b)), 0); |
| 409 |
continue; |
| 410 |
} |
| 411 |
buf = nv04_resource(vb->buffer.resource); |
| 412 |
offset = vb->buffer_offset; |
| 413 |
limit = buf->base.width0 - 1; |
| 414 |
|
| 415 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(b)), 3); |
| 416 |
PUSH_DATA (push, NVC0_3D_VERTEX_ARRAY_FETCH_ENABLE | nvc0->vertex->strides[b]); |
| 417 |
PUSH_DATAh(push, buf->address + offset); |
| 418 |
PUSH_DATA (push, buf->address + offset); |
| 419 |
|
| 420 |
if (nvc0->screen->eng3d->oclass < TU102_3D_CLASS) |
| 421 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_LIMIT_HIGH(b)), 2); |
| 422 |
else |
| 423 |
BEGIN_NVC0(push, SUBC_3D(TU102_3D_VERTEX_ARRAY_LIMIT_HIGH(b)), 2); |
| 424 |
PUSH_DATAh(push, buf->address + limit); |
| 425 |
PUSH_DATA (push, buf->address + limit); |
| 426 |
|
| 427 |
BCTX_REFN(nvc0->bufctx_3d, 3D_VTX, buf, RD); |
| 428 |
} |
| 429 |
/* If there are more elements than buffers, we might not have unset |
| 430 |
* fetching on the later elements. |
| 431 |
*/ |
| 432 |
for (; b < nvc0->vertex->num_elements; ++b) |
| 433 |
IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(b)), 0); |
| 434 |
|
| 435 |
if (nvc0->vbo_user) |
| 436 |
nvc0_update_user_vbufs_shared(nvc0); |
| 437 |
} |
| 438 |
|
| 439 |
void |
| 440 |
nvc0_vertex_arrays_validate(struct nvc0_context *nvc0) |
| 441 |
{ |
| 442 |
struct nouveau_pushbuf *push = nvc0->base.pushbuf; |
| 443 |
struct nvc0_vertex_stateobj *vertex = nvc0->vertex; |
| 444 |
struct nvc0_vertex_element *ve; |
| 445 |
uint32_t const_vbos; |
| 446 |
unsigned i; |
| 447 |
uint8_t vbo_mode; |
| 448 |
bool update_vertex; |
| 449 |
|
| 450 |
nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_VTX); |
| 451 |
|
| 452 |
assert(vertex); |
| 453 |
if (unlikely(vertex->need_conversion) || |
| 454 |
unlikely(nvc0->vertprog->vp.edgeflag < PIPE_MAX_ATTRIBS)) { |
| 455 |
vbo_mode = 3; |
| 456 |
} else if (nvc0->vbo_user & ~nvc0->constant_vbos) { |
| 457 |
vbo_mode = nvc0->vbo_push_hint ? 1 : 0; |
| 458 |
} else { |
| 459 |
vbo_mode = 0; |
| 460 |
} |
| 461 |
const_vbos = vbo_mode ? 0 : nvc0->constant_vbos; |
| 462 |
|
| 463 |
update_vertex = (nvc0->dirty_3d & NVC0_NEW_3D_VERTEX) || |
| 464 |
(const_vbos != nvc0->state.constant_vbos) || |
| 465 |
(vbo_mode != nvc0->state.vbo_mode); |
| 466 |
|
| 467 |
if (update_vertex) { |
| 468 |
const unsigned n = MAX2(vertex->num_elements, nvc0->state.num_vtxelts); |
| 469 |
|
| 470 |
simple_mtx_assert_locked(&nvc0->screen->state_lock); |
| 471 |
nvc0->state.constant_vbos = const_vbos; |
| 472 |
nvc0->state.constant_elts = 0; |
| 473 |
nvc0->state.num_vtxelts = vertex->num_elements; |
| 474 |
nvc0->state.vbo_mode = vbo_mode; |
| 475 |
|
| 476 |
if (unlikely(vbo_mode)) { |
| 477 |
if (unlikely(nvc0->state.instance_elts & 3)) { |
| 478 |
/* translate mode uses only 2 vertex buffers */ |
| 479 |
nvc0->state.instance_elts &= ~3; |
| 480 |
PUSH_SPACE(push, 3); |
| 481 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_PER_INSTANCE(0)), 2); |
| 482 |
PUSH_DATA (push, 0); |
| 483 |
PUSH_DATA (push, 0); |
| 484 |
} |
| 485 |
|
| 486 |
PUSH_SPACE(push, n * 2 + 4); |
| 487 |
|
| 488 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ATTRIB_FORMAT(0)), n); |
| 489 |
for (i = 0; i < vertex->num_elements; ++i) |
| 490 |
PUSH_DATA(push, vertex->element[i].state_alt); |
| 491 |
for (; i < n; ++i) |
| 492 |
PUSH_DATA(push, NVC0_3D_VERTEX_ATTRIB_INACTIVE); |
| 493 |
|
| 494 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(0)), 1); |
| 495 |
PUSH_DATA (push, (1 << 12) | vertex->size); |
| 496 |
for (i = 1; i < n; ++i) |
| 497 |
IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 0); |
| 498 |
} else { |
| 499 |
uint32_t *restrict data; |
| 500 |
|
| 501 |
if (unlikely(vertex->instance_elts != nvc0->state.instance_elts)) { |
| 502 |
nvc0->state.instance_elts = vertex->instance_elts; |
| 503 |
assert(n); /* if (n == 0), both masks should be 0 */ |
| 504 |
PUSH_SPACE(push, 3); |
| 505 |
BEGIN_NVC0(push, NVC0_3D(MACRO_VERTEX_ARRAY_PER_INSTANCE), 2); |
| 506 |
PUSH_DATA (push, n); |
| 507 |
PUSH_DATA (push, vertex->instance_elts); |
| 508 |
} |
| 509 |
|
| 510 |
PUSH_SPACE(push, n * 2 + 1); |
| 511 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ATTRIB_FORMAT(0)), n); |
| 512 |
data = push->cur; |
| 513 |
push->cur += n; |
| 514 |
for (i = 0; i < vertex->num_elements; ++i) { |
| 515 |
ve = &vertex->element[i]; |
| 516 |
data[i] = ve->state; |
| 517 |
if (unlikely(const_vbos & (1 << ve->pipe.vertex_buffer_index))) { |
| 518 |
nvc0->state.constant_elts |= 1 << i; |
| 519 |
data[i] |= NVC0_3D_VERTEX_ATTRIB_FORMAT_CONST; |
| 520 |
IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 0); |
| 521 |
} |
| 522 |
} |
| 523 |
for (; i < n; ++i) { |
| 524 |
data[i] = NVC0_3D_VERTEX_ATTRIB_INACTIVE; |
| 525 |
IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 0); |
| 526 |
} |
| 527 |
} |
| 528 |
} |
| 529 |
if (nvc0->state.vbo_mode) /* using translate, don't set up arrays here */ |
| 530 |
return; |
| 531 |
|
| 532 |
if (vertex->shared_slots) |
| 533 |
nvc0_validate_vertex_buffers_shared(nvc0); |
| 534 |
else |
| 535 |
nvc0_validate_vertex_buffers(nvc0); |
| 536 |
} |
| 537 |
|
| 538 |
#define NVC0_PRIM_GL_CASE(n) \ |
| 539 |
case MESA_PRIM_##n: return NVC0_3D_VERTEX_BEGIN_GL_PRIMITIVE_##n |
| 540 |
|
| 541 |
static inline unsigned |
| 542 |
nvc0_prim_gl(unsigned prim) |
| 543 |
{ |
| 544 |
switch (prim) { |
| 545 |
NVC0_PRIM_GL_CASE(POINTS); |
| 546 |
NVC0_PRIM_GL_CASE(LINES); |
| 547 |
NVC0_PRIM_GL_CASE(LINE_LOOP); |
| 548 |
NVC0_PRIM_GL_CASE(LINE_STRIP); |
| 549 |
NVC0_PRIM_GL_CASE(TRIANGLES); |
| 550 |
NVC0_PRIM_GL_CASE(TRIANGLE_STRIP); |
| 551 |
NVC0_PRIM_GL_CASE(TRIANGLE_FAN); |
| 552 |
NVC0_PRIM_GL_CASE(QUADS); |
| 553 |
NVC0_PRIM_GL_CASE(QUAD_STRIP); |
| 554 |
NVC0_PRIM_GL_CASE(POLYGON); |
| 555 |
NVC0_PRIM_GL_CASE(LINES_ADJACENCY); |
| 556 |
NVC0_PRIM_GL_CASE(LINE_STRIP_ADJACENCY); |
| 557 |
NVC0_PRIM_GL_CASE(TRIANGLES_ADJACENCY); |
| 558 |
NVC0_PRIM_GL_CASE(TRIANGLE_STRIP_ADJACENCY); |
| 559 |
NVC0_PRIM_GL_CASE(PATCHES); |
| 560 |
default: |
| 561 |
return NVC0_3D_VERTEX_BEGIN_GL_PRIMITIVE_POINTS; |
| 562 |
} |
| 563 |
} |
| 564 |
|
| 565 |
static void |
| 566 |
nvc0_draw_vbo_kick_notify(struct nouveau_context *context) |
| 567 |
{ |
| 568 |
_nouveau_fence_update(context->screen, true); |
| 569 |
} |
| 570 |
|
| 571 |
static void |
| 572 |
nvc0_draw_arrays(struct nvc0_context *nvc0, |
| 573 |
unsigned mode, unsigned start, unsigned count, |
| 574 |
unsigned instance_count) |
| 575 |
{ |
| 576 |
struct nouveau_pushbuf *push = nvc0->base.pushbuf; |
| 577 |
unsigned prim; |
| 578 |
|
| 579 |
if (nvc0->state.index_bias) { |
| 580 |
/* index_bias is implied 0 if !info->index_size (really ?) */ |
| 581 |
/* TODO: can we deactivate it for the VERTEX_BUFFER_FIRST command ? */ |
| 582 |
PUSH_SPACE(push, 2); |
| 583 |
IMMED_NVC0(push, NVC0_3D(VB_ELEMENT_BASE), 0); |
| 584 |
IMMED_NVC0(push, NVC0_3D(VERTEX_ID_BASE), 0); |
| 585 |
nvc0->state.index_bias = 0; |
| 586 |
} |
| 587 |
|
| 588 |
prim = nvc0_prim_gl(mode); |
| 589 |
|
| 590 |
while (instance_count--) { |
| 591 |
PUSH_SPACE(push, 6); |
| 592 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_BEGIN_GL), 1); |
| 593 |
PUSH_DATA (push, prim); |
| 594 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_BUFFER_FIRST), 2); |
| 595 |
PUSH_DATA (push, start); |
| 596 |
PUSH_DATA (push, count); |
| 597 |
IMMED_NVC0(push, NVC0_3D(VERTEX_END_GL), 0); |
| 598 |
|
| 599 |
prim |= NVC0_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT; |
| 600 |
} |
| 601 |
NOUVEAU_DRV_STAT(&nvc0->screen->base, draw_calls_array, 1); |
| 602 |
} |
| 603 |
|
| 604 |
static void |
| 605 |
nvc0_draw_elements_inline_u08(struct nouveau_pushbuf *push, const uint8_t *map, |
| 606 |
unsigned start, unsigned count) |
| 607 |
{ |
| 608 |
map += start; |
| 609 |
|
| 610 |
if (count & 3) { |
| 611 |
unsigned i; |
| 612 |
PUSH_SPACE(push, 4); |
| 613 |
BEGIN_NIC0(push, NVC0_3D(VB_ELEMENT_U32), count & 3); |
| 614 |
for (i = 0; i < (count & 3); ++i) |
| 615 |
PUSH_DATA(push, *map++); |
| 616 |
count &= ~3; |
| 617 |
} |
| 618 |
while (count) { |
| 619 |
unsigned i, nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN * 4) / 4; |
| 620 |
|
| 621 |
PUSH_SPACE(push, nr + 1); |
| 622 |
BEGIN_NIC0(push, NVC0_3D(VB_ELEMENT_U8), nr); |
| 623 |
for (i = 0; i < nr; ++i) { |
| 624 |
PUSH_DATA(push, |
| 625 |
(map[3] << 24) | (map[2] << 16) | (map[1] << 8) | map[0]); |
| 626 |
map += 4; |
| 627 |
} |
| 628 |
count -= nr * 4; |
| 629 |
} |
| 630 |
} |
| 631 |
|
| 632 |
static void |
| 633 |
nvc0_draw_elements_inline_u16(struct nouveau_pushbuf *push, const uint16_t *map, |
| 634 |
unsigned start, unsigned count) |
| 635 |
{ |
| 636 |
map += start; |
| 637 |
|
| 638 |
if (count & 1) { |
| 639 |
count &= ~1; |
| 640 |
PUSH_SPACE(push, 2); |
| 641 |
BEGIN_NVC0(push, NVC0_3D(VB_ELEMENT_U32), 1); |
| 642 |
PUSH_DATA (push, *map++); |
| 643 |
} |
| 644 |
while (count) { |
| 645 |
unsigned i, nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN * 2) / 2; |
| 646 |
|
| 647 |
PUSH_SPACE(push, nr + 1); |
| 648 |
BEGIN_NIC0(push, NVC0_3D(VB_ELEMENT_U16), nr); |
| 649 |
for (i = 0; i < nr; ++i) { |
| 650 |
PUSH_DATA(push, (map[1] << 16) | map[0]); |
| 651 |
map += 2; |
| 652 |
} |
| 653 |
count -= nr * 2; |
| 654 |
} |
| 655 |
} |
| 656 |
|
| 657 |
static void |
| 658 |
nvc0_draw_elements_inline_u32(struct nouveau_pushbuf *push, const uint32_t *map, |
| 659 |
unsigned start, unsigned count) |
| 660 |
{ |
| 661 |
map += start; |
| 662 |
|
| 663 |
while (count) { |
| 664 |
const unsigned nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN); |
| 665 |
|
| 666 |
PUSH_SPACE(push, nr + 1); |
| 667 |
BEGIN_NIC0(push, NVC0_3D(VB_ELEMENT_U32), nr); |
| 668 |
PUSH_DATAp(push, map, nr); |
| 669 |
|
| 670 |
map += nr; |
| 671 |
count -= nr; |
| 672 |
} |
| 673 |
} |
| 674 |
|
| 675 |
static void |
| 676 |
nvc0_draw_elements_inline_u32_short(struct nouveau_pushbuf *push, |
| 677 |
const uint32_t *map, |
| 678 |
unsigned start, unsigned count) |
| 679 |
{ |
| 680 |
map += start; |
| 681 |
|
| 682 |
if (count & 1) { |
| 683 |
count--; |
| 684 |
PUSH_SPACE(push, 2); |
| 685 |
BEGIN_NVC0(push, NVC0_3D(VB_ELEMENT_U32), 1); |
| 686 |
PUSH_DATA (push, *map++); |
| 687 |
} |
| 688 |
while (count) { |
| 689 |
unsigned i, nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN * 2) / 2; |
| 690 |
|
| 691 |
PUSH_SPACE(push, nr + 1); |
| 692 |
BEGIN_NIC0(push, NVC0_3D(VB_ELEMENT_U16), nr); |
| 693 |
for (i = 0; i < nr; ++i) { |
| 694 |
PUSH_DATA(push, (map[1] << 16) | map[0]); |
| 695 |
map += 2; |
| 696 |
} |
| 697 |
count -= nr * 2; |
| 698 |
} |
| 699 |
} |
| 700 |
|
| 701 |
static void |
| 702 |
nvc0_draw_elements(struct nvc0_context *nvc0, bool shorten, |
| 703 |
const struct pipe_draw_info *info, |
| 704 |
unsigned mode, unsigned start, unsigned count, |
| 705 |
unsigned instance_count, int32_t index_bias, |
| 706 |
unsigned index_size) |
| 707 |
{ |
| 708 |
struct nouveau_pushbuf *push = nvc0->base.pushbuf; |
| 709 |
unsigned prim; |
| 710 |
|
| 711 |
prim = nvc0_prim_gl(mode); |
| 712 |
|
| 713 |
if (index_bias != nvc0->state.index_bias) { |
| 714 |
PUSH_SPACE(push, 4); |
| 715 |
BEGIN_NVC0(push, NVC0_3D(VB_ELEMENT_BASE), 1); |
| 716 |
PUSH_DATA (push, index_bias); |
| 717 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_ID_BASE), 1); |
| 718 |
PUSH_DATA (push, index_bias); |
| 719 |
nvc0->state.index_bias = index_bias; |
| 720 |
} |
| 721 |
|
| 722 |
if (!info->has_user_indices) { |
| 723 |
PUSH_SPACE(push, 1); |
| 724 |
IMMED_NVC0(push, NVC0_3D(VERTEX_BEGIN_GL), prim); |
| 725 |
do { |
| 726 |
PUSH_SPACE(push, 7); |
| 727 |
BEGIN_NVC0(push, NVC0_3D(INDEX_BATCH_FIRST), 2); |
| 728 |
PUSH_DATA (push, start); |
| 729 |
PUSH_DATA (push, count); |
| 730 |
if (--instance_count) { |
| 731 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_END_GL), 2); |
| 732 |
PUSH_DATA (push, 0); |
| 733 |
PUSH_DATA (push, prim | NVC0_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT); |
| 734 |
} |
| 735 |
} while (instance_count); |
| 736 |
IMMED_NVC0(push, NVC0_3D(VERTEX_END_GL), 0); |
| 737 |
} else { |
| 738 |
const void *data = info->index.user; |
| 739 |
|
| 740 |
while (instance_count--) { |
| 741 |
PUSH_SPACE(push, 2); |
| 742 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_BEGIN_GL), 1); |
| 743 |
PUSH_DATA (push, prim); |
| 744 |
switch (index_size) { |
| 745 |
case 1: |
| 746 |
nvc0_draw_elements_inline_u08(push, data, start, count); |
| 747 |
break; |
| 748 |
case 2: |
| 749 |
nvc0_draw_elements_inline_u16(push, data, start, count); |
| 750 |
break; |
| 751 |
case 4: |
| 752 |
if (shorten) |
| 753 |
nvc0_draw_elements_inline_u32_short(push, data, start, count); |
| 754 |
else |
| 755 |
nvc0_draw_elements_inline_u32(push, data, start, count); |
| 756 |
break; |
| 757 |
default: |
| 758 |
assert(0); |
| 759 |
return; |
| 760 |
} |
| 761 |
PUSH_SPACE(push, 1); |
| 762 |
IMMED_NVC0(push, NVC0_3D(VERTEX_END_GL), 0); |
| 763 |
|
| 764 |
prim |= NVC0_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT; |
| 765 |
} |
| 766 |
} |
| 767 |
NOUVEAU_DRV_STAT(&nvc0->screen->base, draw_calls_indexed, 1); |
| 768 |
} |
| 769 |
|
| 770 |
static void |
| 771 |
nvc0_draw_stream_output(struct nvc0_context *nvc0, |
| 772 |
const struct pipe_draw_info *info, |
| 773 |
const struct pipe_draw_indirect_info *indirect) |
| 774 |
{ |
| 775 |
struct nouveau_pushbuf *push = nvc0->base.pushbuf; |
| 776 |
struct nvc0_so_target *so = nvc0_so_target(indirect->count_from_stream_output); |
| 777 |
struct nv04_resource *res = nv04_resource(so->pipe.buffer); |
| 778 |
unsigned mode = nvc0_prim_gl(info->mode); |
| 779 |
unsigned num_instances = info->instance_count; |
| 780 |
|
| 781 |
if (res->status & NOUVEAU_BUFFER_STATUS_GPU_WRITING) { |
| 782 |
res->status &= ~NOUVEAU_BUFFER_STATUS_GPU_WRITING; |
| 783 |
PUSH_SPACE(push, 2); |
| 784 |
IMMED_NVC0(push, NVC0_3D(SERIALIZE), 0); |
| 785 |
nvc0_hw_query_fifo_wait(nvc0, nvc0_query(so->pq)); |
| 786 |
if (nvc0->screen->eng3d->oclass < GM107_3D_CLASS) |
| 787 |
IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_FLUSH), 0); |
| 788 |
|
| 789 |
NOUVEAU_DRV_STAT(&nvc0->screen->base, gpu_serialize_count, 1); |
| 790 |
} |
| 791 |
|
| 792 |
while (num_instances--) { |
| 793 |
PUSH_SPACE_EX(push, 16, 0, 1); |
| 794 |
BEGIN_NVC0(push, NVC0_3D(VERTEX_BEGIN_GL), 1); |
| 795 |
PUSH_DATA (push, mode); |
| 796 |
BEGIN_NVC0(push, NVC0_3D(DRAW_TFB_BASE), 1); |
| 797 |
PUSH_DATA (push, 0); |
| 798 |
BEGIN_NVC0(push, NVC0_3D(DRAW_TFB_STRIDE), 1); |
| 799 |
PUSH_DATA (push, so->stride); |
| 800 |
BEGIN_NVC0(push, NVC0_3D(DRAW_TFB_BYTES), 1); |
| 801 |
nvc0_hw_query_pushbuf_submit(push, nvc0_query(so->pq), 0x4); |
| 802 |
IMMED_NVC0(push, NVC0_3D(VERTEX_END_GL), 0); |
| 803 |
|
| 804 |
mode |= NVC0_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT; |
| 805 |
} |
| 806 |
} |
| 807 |
|
| 808 |
static void |
| 809 |
nvc0_draw_indirect(struct nvc0_context *nvc0, const struct pipe_draw_info *info, |
| 810 |
unsigned drawid_offset, |
| 811 |
const struct pipe_draw_indirect_info *indirect) |
| 812 |
{ |
| 813 |
struct nouveau_pushbuf *push = nvc0->base.pushbuf; |
| 814 |
struct nv04_resource *buf = nv04_resource(indirect->buffer); |
| 815 |
struct nv04_resource *buf_count = nv04_resource(indirect->indirect_draw_count); |
| 816 |
unsigned size, macro, count = indirect->draw_count, drawid = drawid_offset; |
| 817 |
uint32_t offset = buf->offset + indirect->offset; |
| 818 |
struct nvc0_screen *screen = nvc0->screen; |
| 819 |
|
| 820 |
PUSH_SPACE(push, 7); |
| 821 |
|
| 822 |
/* must make FIFO wait for engines idle before continuing to process */ |
| 823 |
if ((buf->fence_wr && !nouveau_fence_signalled(buf->fence_wr)) || |
| 824 |
(buf_count && buf_count->fence_wr && |
| 825 |
!nouveau_fence_signalled(buf_count->fence_wr))) { |
| 826 |
IMMED_NVC0(push, SUBC_3D(NV10_SUBCHAN_REF_CNT), 0); |
| 827 |
} |
| 828 |
|
| 829 |
/* Queue things up to let the macros write params to the driver constbuf */ |
| 830 |
BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3); |
| 831 |
PUSH_DATA (push, NVC0_CB_AUX_SIZE); |
| 832 |
PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(0)); |
| 833 |
PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(0)); |
| 834 |
BEGIN_NVC0(push, NVC0_3D(CB_POS), 1); |
| 835 |
PUSH_DATA (push, NVC0_CB_AUX_DRAW_INFO); |
| 836 |
|
| 837 |
if (info->index_size) { |
| 838 |
assert(!info->has_user_indices); |
| 839 |
assert(nouveau_resource_mapped_by_gpu(info->index.resource)); |
| 840 |
size = 5; |
| 841 |
if (buf_count) |
| 842 |
macro = NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT_COUNT; |
| 843 |
else |
| 844 |
macro = NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT; |
| 845 |
} else { |
| 846 |
if (nvc0->state.index_bias) { |
| 847 |
/* index_bias is implied 0 if !info->index_size (really ?) */ |
| 848 |
IMMED_NVC0(push, NVC0_3D(VB_ELEMENT_BASE), 0); |
| 849 |
IMMED_NVC0(push, NVC0_3D(VERTEX_ID_BASE), 0); |
| 850 |
nvc0->state.index_bias = 0; |
| 851 |
} |
| 852 |
size = 4; |
| 853 |
if (buf_count) |
| 854 |
macro = NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT_COUNT; |
| 855 |
else |
| 856 |
macro = NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT; |
| 857 |
} |
| 858 |
|
| 859 |
/* If the stride is not the natural stride, we have to stick a separate |
| 860 |
* push data reference for each draw. Otherwise it can all go in as one. |
| 861 |
* Of course there is a maximum packet size, so we have to break things up |
| 862 |
* along those borders as well. |
| 863 |
*/ |
| 864 |
while (count) { |
| 865 |
unsigned draws = count, pushes, i; |
| 866 |
if (indirect->stride == size * 4) { |
| 867 |
draws = MIN2(draws, (NV04_PFIFO_MAX_PACKET_LEN - 4) / size); |
| 868 |
pushes = 1; |
| 869 |
} else { |
| 870 |
draws = MIN2(draws, 32); |
| 871 |
pushes = draws; |
| 872 |
} |
| 873 |
|
| 874 |
PUSH_SPACE_EX(push, 16, 0, pushes + !!buf_count); |
| 875 |
PUSH_REF1(push, buf->bo, NOUVEAU_BO_RD | buf->domain); |
| 876 |
if (buf_count) |
| 877 |
PUSH_REF1(push, buf_count->bo, NOUVEAU_BO_RD | buf_count->domain); |
| 878 |
PUSH_DATA(push, |
| 879 |
NVC0_FIFO_PKHDR_1I(0, macro, 3 + !!buf_count + draws * size)); |
| 880 |
PUSH_DATA(push, nvc0_prim_gl(info->mode)); |
| 881 |
PUSH_DATA(push, drawid); |
| 882 |
PUSH_DATA(push, draws); |
| 883 |
if (buf_count) { |
| 884 |
nouveau_pushbuf_data(push, |
| 885 |
buf_count->bo, |
| 886 |
buf_count->offset + indirect->indirect_draw_count_offset, |
| 887 |
NVC0_IB_ENTRY_1_NO_PREFETCH | 4); |
| 888 |
} |
| 889 |
if (pushes == 1) { |
| 890 |
nouveau_pushbuf_data(push, |
| 891 |
buf->bo, offset, |
| 892 |
NVC0_IB_ENTRY_1_NO_PREFETCH | (size * 4 * draws)); |
| 893 |
offset += draws * indirect->stride; |
| 894 |
} else { |
| 895 |
for (i = 0; i < pushes; i++) { |
| 896 |
nouveau_pushbuf_data(push, |
| 897 |
buf->bo, offset, |
| 898 |
NVC0_IB_ENTRY_1_NO_PREFETCH | (size * 4)); |
| 899 |
offset += indirect->stride; |
| 900 |
} |
| 901 |
} |
| 902 |
count -= draws; |
| 903 |
drawid += draws; |
| 904 |
} |
| 905 |
} |
| 906 |
|
| 907 |
static inline void |
| 908 |
nvc0_update_prim_restart(struct nvc0_context *nvc0, bool en, uint32_t index) |
| 909 |
{ |
| 910 |
struct nouveau_pushbuf *push = nvc0->base.pushbuf; |
| 911 |
|
| 912 |
if (en != nvc0->state.prim_restart) { |
| 913 |
if (en) { |
| 914 |
BEGIN_NVC0(push, NVC0_3D(PRIM_RESTART_ENABLE), 2); |
| 915 |
PUSH_DATA (push, 1); |
| 916 |
PUSH_DATA (push, index); |
| 917 |
} else { |
| 918 |
IMMED_NVC0(push, NVC0_3D(PRIM_RESTART_ENABLE), 0); |
| 919 |
} |
| 920 |
nvc0->state.prim_restart = en; |
| 921 |
} else |
| 922 |
if (en) { |
| 923 |
BEGIN_NVC0(push, NVC0_3D(PRIM_RESTART_INDEX), 1); |
| 924 |
PUSH_DATA (push, index); |
| 925 |
} |
| 926 |
} |
| 927 |
|
| 928 |
static void |
| 929 |
nvc0_draw_single_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info, |
| 930 |
unsigned drawid_offset, |
| 931 |
const struct pipe_draw_indirect_info *indirect, |
| 932 |
const struct pipe_draw_start_count_bias *draws) |
| 933 |
{ |
| 934 |
struct nvc0_context *nvc0 = nvc0_context(pipe); |
| 935 |
struct nouveau_pushbuf *push = nvc0->base.pushbuf; |
| 936 |
struct nvc0_screen *screen = nvc0->screen; |
| 937 |
int s; |
| 938 |
|
| 939 |
simple_mtx_assert_locked(&nvc0->screen->state_lock); |
| 940 |
|
| 941 |
if (nvc0->vertprog->vp.need_draw_parameters && (!indirect || indirect->count_from_stream_output)) { |
| 942 |
PUSH_SPACE(push, 9); |
| 943 |
BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3); |
| 944 |
PUSH_DATA (push, NVC0_CB_AUX_SIZE); |
| 945 |
PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(0)); |
| 946 |
PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(0)); |
| 947 |
BEGIN_1IC0(push, NVC0_3D(CB_POS), 1 + 3); |
| 948 |
PUSH_DATA (push, NVC0_CB_AUX_DRAW_INFO); |
| 949 |
PUSH_DATA (push, info->index_size ? draws->index_bias : 0); |
| 950 |
PUSH_DATA (push, info->start_instance); |
| 951 |
PUSH_DATA (push, drawid_offset); |
| 952 |
} |
| 953 |
|
| 954 |
if (nvc0->screen->base.class_3d < NVE4_3D_CLASS && |
| 955 |
nvc0->seamless_cube_map != nvc0->state.seamless_cube_map) { |
| 956 |
nvc0->state.seamless_cube_map = nvc0->seamless_cube_map; |
| 957 |
PUSH_SPACE(push, 1); |
| 958 |
IMMED_NVC0(push, NVC0_3D(TEX_MISC), |
| 959 |
nvc0->seamless_cube_map ? NVC0_3D_TEX_MISC_SEAMLESS_CUBE_MAP : 0); |
| 960 |
} |
| 961 |
|
| 962 |
nvc0->base.kick_notify = nvc0_draw_vbo_kick_notify; |
| 963 |
|
| 964 |
for (s = 0; s < 5 && !nvc0->cb_dirty; ++s) { |
| 965 |
if (nvc0->constbuf_coherent[s]) |
| 966 |
nvc0->cb_dirty = true; |
| 967 |
} |
| 968 |
|
| 969 |
if (nvc0->cb_dirty) { |
| 970 |
PUSH_SPACE(push, 1); |
| 971 |
IMMED_NVC0(push, NVC0_3D(MEM_BARRIER), 0x1011); |
| 972 |
nvc0->cb_dirty = false; |
| 973 |
} |
| 974 |
|
| 975 |
for (s = 0; s < 5; ++s) { |
| 976 |
if (!nvc0->textures_coherent[s]) |
| 977 |
continue; |
| 978 |
|
| 979 |
PUSH_SPACE(push, nvc0->num_textures[s] * 2); |
| 980 |
|
| 981 |
for (int i = 0; i < nvc0->num_textures[s]; ++i) { |
| 982 |
struct nv50_tic_entry *tic = nv50_tic_entry(nvc0->textures[s][i]); |
| 983 |
if (!(nvc0->textures_coherent[s] & (1 << i))) |
| 984 |
continue; |
| 985 |
|
| 986 |
BEGIN_NVC0(push, NVC0_3D(TEX_CACHE_CTL), 1); |
| 987 |
PUSH_DATA (push, (tic->id << 4) | 1); |
| 988 |
NOUVEAU_DRV_STAT(&nvc0->screen->base, tex_cache_flush_count, 1); |
| 989 |
} |
| 990 |
} |
| 991 |
|
| 992 |
if (nvc0->state.vbo_mode) { |
| 993 |
if (indirect && indirect->buffer) |
| 994 |
nvc0_push_vbo_indirect(nvc0, info, drawid_offset, indirect, &draws[0]); |
| 995 |
else |
| 996 |
nvc0_push_vbo(nvc0, info, indirect, &draws[0]); |
| 997 |
return; |
| 998 |
} |
| 999 |
|
| 1000 |
/* space for base instance, flush, and prim restart */ |
| 1001 |
PUSH_SPACE(push, 8); |
| 1002 |
|
| 1003 |
if (nvc0->state.instance_base != info->start_instance) { |
| 1004 |
nvc0->state.instance_base = info->start_instance; |
| 1005 |
/* NOTE: this does not affect the shader input, should it ? */ |
| 1006 |
BEGIN_NVC0(push, NVC0_3D(VB_INSTANCE_BASE), 1); |
| 1007 |
PUSH_DATA (push, info->start_instance); |
| 1008 |
} |
| 1009 |
|
| 1010 |
nvc0->base.vbo_dirty |= !!nvc0->vtxbufs_coherent; |
| 1011 |
|
| 1012 |
if (!nvc0->base.vbo_dirty && info->index_size && !info->has_user_indices && |
| 1013 |
info->index.resource->flags & PIPE_RESOURCE_FLAG_MAP_COHERENT) |
| 1014 |
nvc0->base.vbo_dirty = true; |
| 1015 |
|
| 1016 |
nvc0_update_prim_restart(nvc0, info->primitive_restart, info->restart_index); |
| 1017 |
|
| 1018 |
if (nvc0->base.vbo_dirty) { |
| 1019 |
if (nvc0->screen->eng3d->oclass < GM107_3D_CLASS) |
| 1020 |
IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_FLUSH), 0); |
| 1021 |
nvc0->base.vbo_dirty = false; |
| 1022 |
} |
| 1023 |
|
| 1024 |
if (unlikely(indirect && indirect->buffer)) { |
| 1025 |
nvc0_draw_indirect(nvc0, info, drawid_offset, indirect); |
| 1026 |
} else |
| 1027 |
if (unlikely(indirect && indirect->count_from_stream_output)) { |
| 1028 |
nvc0_draw_stream_output(nvc0, info, indirect); |
| 1029 |
} else |
| 1030 |
if (info->index_size) { |
| 1031 |
bool shorten = info->index_bounds_valid && info->max_index <= 65535; |
| 1032 |
|
| 1033 |
if (info->primitive_restart && info->restart_index > 65535) |
| 1034 |
shorten = false; |
| 1035 |
|
| 1036 |
nvc0_draw_elements(nvc0, shorten, info, |
| 1037 |
info->mode, draws[0].start, draws[0].count, |
| 1038 |
info->instance_count, draws->index_bias, info->index_size); |
| 1039 |
} else { |
| 1040 |
nvc0_draw_arrays(nvc0, |
| 1041 |
info->mode, draws[0].start, draws[0].count, |
| 1042 |
info->instance_count); |
| 1043 |
} |
| 1044 |
} |
| 1045 |
|
| 1046 |
/* Thin wrapper to avoid kicking every 3 ns during multidraw */ |
| 1047 |
|
| 1048 |
void |
| 1049 |
nvc0_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info, |
| 1050 |
unsigned drawid_offset, |
| 1051 |
const struct pipe_draw_indirect_info *indirect, |
| 1052 |
const struct pipe_draw_start_count_bias *draws, |
| 1053 |
unsigned num_draws) |
| 1054 |
{ |
| 1055 |
struct nvc0_context *nvc0 = nvc0_context(pipe); |
| 1056 |
struct nouveau_pushbuf *push = nvc0->base.pushbuf; |
| 1057 |
|
| 1058 |
struct nvc0_screen *screen = nvc0->screen; |
| 1059 |
unsigned vram_domain = NV_VRAM_DOMAIN(&screen->base); |
| 1060 |
unsigned count_total = 0; |
| 1061 |
|
| 1062 |
/* The rest is copied straight from util_multi_draw |
| 1063 |
* |
| 1064 |
* XXX: Properly rewrite vbo handling in nvc0/nv50 |
| 1065 |
* |
| 1066 |
*/ |
| 1067 |
|
| 1068 |
unsigned drawid = drawid_offset; |
| 1069 |
|
| 1070 |
/* dont wanna start trippin */ |
| 1071 |
simple_mtx_lock(&nvc0->screen->state_lock); |
| 1072 |
|
| 1073 |
/* NOTE: caller must ensure that (min_index + index_bias) is >= 0 */ |
| 1074 |
if (info->index_bounds_valid) { |
| 1075 |
nvc0->vb_elt_first = info->min_index + (info->index_size ? draws->index_bias : 0); |
| 1076 |
nvc0->vb_elt_limit = info->max_index - info->min_index; |
| 1077 |
} else { |
| 1078 |
nvc0->vb_elt_first = 0; |
| 1079 |
nvc0->vb_elt_limit = ~0; |
| 1080 |
} |
| 1081 |
nvc0->instance_off = info->start_instance; |
| 1082 |
nvc0->instance_max = info->instance_count - 1; |
| 1083 |
|
| 1084 |
/* Get total amount of draw counts to determine whether to push or upload vertices */ |
| 1085 |
for (unsigned i = 0; i < num_draws; i++) { |
| 1086 |
count_total += draws[i].count; |
| 1087 |
} |
| 1088 |
|
| 1089 |
/* For picking only a few vertices from a large user buffer, push is better, |
| 1090 |
* if index count is larger and we expect repeated vertices, suggest upload. |
| 1091 |
*/ |
| 1092 |
nvc0->vbo_push_hint = |
| 1093 |
(!indirect || indirect->count_from_stream_output) && info->index_size && |
| 1094 |
(nvc0->vb_elt_limit >= (count_total * 2)); |
| 1095 |
|
| 1096 |
if (nvc0->dirty_3d & (NVC0_NEW_3D_ARRAYS | NVC0_NEW_3D_VERTEX)) |
| 1097 |
nvc0->constant_vbos = nvc0->vertex->constant_vbos & nvc0->vbo_user; |
| 1098 |
/* Check whether we want to switch vertex-submission mode. */ |
| 1099 |
if (nvc0->vbo_user && !(nvc0->dirty_3d & (NVC0_NEW_3D_ARRAYS | NVC0_NEW_3D_VERTEX))) { |
| 1100 |
if (nvc0->vbo_push_hint != !!nvc0->state.vbo_mode) |
| 1101 |
if (nvc0->state.vbo_mode != 3) |
| 1102 |
nvc0->dirty_3d |= NVC0_NEW_3D_ARRAYS; |
| 1103 |
|
| 1104 |
if (!(nvc0->dirty_3d & NVC0_NEW_3D_ARRAYS) && nvc0->state.vbo_mode == 0) { |
| 1105 |
if (nvc0->vertex->shared_slots) |
| 1106 |
nvc0_update_user_vbufs_shared(nvc0); |
| 1107 |
else |
| 1108 |
nvc0_update_user_vbufs(nvc0); |
| 1109 |
} |
| 1110 |
} |
| 1111 |
|
| 1112 |
if (info->mode == MESA_PRIM_PATCHES && |
| 1113 |
nvc0->state.patch_vertices != nvc0->patch_vertices) { |
| 1114 |
nvc0->state.patch_vertices = nvc0->patch_vertices; |
| 1115 |
PUSH_SPACE(push, 1); |
| 1116 |
IMMED_NVC0(push, NVC0_3D(PATCH_VERTICES), nvc0->state.patch_vertices); |
| 1117 |
} |
| 1118 |
|
| 1119 |
if (info->index_size && !info->has_user_indices) { |
| 1120 |
struct nv04_resource *buf = nv04_resource(info->index.resource); |
| 1121 |
|
| 1122 |
assert(buf); |
| 1123 |
assert(nouveau_resource_mapped_by_gpu(&buf->base)); |
| 1124 |
|
| 1125 |
PUSH_SPACE(push, 6); |
| 1126 |
if (nvc0->screen->eng3d->oclass < TU102_3D_CLASS) { |
| 1127 |
BEGIN_NVC0(push, NVC0_3D(INDEX_ARRAY_START_HIGH), 5); |
| 1128 |
PUSH_DATAh(push, buf->address); |
| 1129 |
PUSH_DATA (push, buf->address); |
| 1130 |
PUSH_DATAh(push, buf->address + buf->base.width0 - 1); |
| 1131 |
PUSH_DATA (push, buf->address + buf->base.width0 - 1); |
| 1132 |
PUSH_DATA (push, info->index_size >> 1); |
| 1133 |
} else { |
| 1134 |
BEGIN_NVC0(push, NVC0_3D(INDEX_ARRAY_START_HIGH), 2); |
| 1135 |
PUSH_DATAh(push, buf->address); |
| 1136 |
PUSH_DATA (push, buf->address); |
| 1137 |
BEGIN_NVC0(push, SUBC_3D(TU102_3D_INDEX_ARRAY_LIMIT_HIGH), 2); |
| 1138 |
PUSH_DATAh(push, buf->address + buf->base.width0 - 1); |
| 1139 |
PUSH_DATA (push, buf->address + buf->base.width0 - 1); |
| 1140 |
BEGIN_NVC0(push, NVC0_3D(INDEX_FORMAT), 1); |
| 1141 |
PUSH_DATA (push, info->index_size >> 1); |
| 1142 |
} |
| 1143 |
|
| 1144 |
BCTX_REFN(nvc0->bufctx_3d, 3D_IDX, buf, RD); |
| 1145 |
} |
| 1146 |
|
| 1147 |
list_for_each_entry(struct nvc0_resident, resident, &nvc0->tex_head, list) { |
| 1148 |
nvc0_add_resident(nvc0->bufctx_3d, NVC0_BIND_3D_BINDLESS, resident->buf, |
| 1149 |
resident->flags); |
| 1150 |
} |
| 1151 |
|
| 1152 |
list_for_each_entry(struct nvc0_resident, resident, &nvc0->img_head, list) { |
| 1153 |
nvc0_add_resident(nvc0->bufctx_3d, NVC0_BIND_3D_BINDLESS, resident->buf, |
| 1154 |
resident->flags); |
| 1155 |
} |
| 1156 |
|
| 1157 |
BCTX_REFN_bo(nvc0->bufctx_3d, 3D_TEXT, vram_domain | NOUVEAU_BO_RD, |
| 1158 |
screen->text); |
| 1159 |
|
| 1160 |
nvc0_state_validate_3d(nvc0, ~0); |
| 1161 |
|
| 1162 |
for (unsigned i = 0; i < num_draws; i++) { |
| 1163 |
if (indirect || (draws[i].count && info->instance_count)) |
| 1164 |
nvc0_draw_single_vbo(pipe, info, drawid, indirect, &draws[i]); |
| 1165 |
if (info->increment_draw_id) |
| 1166 |
drawid++; |
| 1167 |
} |
| 1168 |
|
| 1169 |
PUSH_KICK(push); |
| 1170 |
simple_mtx_unlock(&nvc0->screen->state_lock); |
| 1171 |
|
| 1172 |
nvc0->base.kick_notify = nvc0_default_kick_notify; |
| 1173 |
|
| 1174 |
nvc0_release_user_vbufs(nvc0); |
| 1175 |
|
| 1176 |
nouveau_pushbuf_bufctx(push, NULL); |
| 1177 |
|
| 1178 |
nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_TEXT); |
| 1179 |
nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_IDX); |
| 1180 |
nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_BINDLESS); |
| 1181 |
} |