| 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 |
#include "pipe/p_defines.h" |
| 24 |
#include "util/u_framebuffer.h" |
| 25 |
#include "util/u_upload_mgr.h" |
| 26 |
|
| 27 |
#include "nvc0/nvc0_context.h" |
| 28 |
#include "nvc0/nvc0_screen.h" |
| 29 |
#include "nvc0/nvc0_resource.h" |
| 30 |
|
| 31 |
|
| 32 |
#include "xf86drm.h" |
| 33 |
#include "drm-uapi/nouveau_drm.h" |
| 34 |
|
| 35 |
|
| 36 |
static void |
| 37 |
nvc0_svm_migrate(struct pipe_context *pipe, unsigned num_ptrs, |
| 38 |
const void* const* ptrs, const size_t *sizes, |
| 39 |
bool to_device, bool mem_undefined) |
| 40 |
{ |
| 41 |
struct nvc0_context *nvc0 = nvc0_context(pipe); |
| 42 |
struct nouveau_screen *screen = &nvc0->screen->base; |
| 43 |
int fd = screen->drm->fd; |
| 44 |
unsigned i; |
| 45 |
|
| 46 |
for (i = 0; i < num_ptrs; i++) { |
| 47 |
struct drm_nouveau_svm_bind args; |
| 48 |
uint64_t cmd, prio, target; |
| 49 |
|
| 50 |
args.va_start = (uint64_t)(uintptr_t)ptrs[i]; |
| 51 |
if (sizes && sizes[i]) { |
| 52 |
args.va_end = (uint64_t)(uintptr_t)ptrs[i] + sizes[i]; |
| 53 |
args.npages = DIV_ROUND_UP(args.va_end - args.va_start, 0x1000); |
| 54 |
} else { |
| 55 |
args.va_end = 0; |
| 56 |
args.npages = 0; |
| 57 |
} |
| 58 |
args.stride = 0; |
| 59 |
|
| 60 |
args.reserved0 = 0; |
| 61 |
args.reserved1 = 0; |
| 62 |
|
| 63 |
prio = 0; |
| 64 |
cmd = NOUVEAU_SVM_BIND_COMMAND__MIGRATE; |
| 65 |
target = to_device ? NOUVEAU_SVM_BIND_TARGET__GPU_VRAM : 0; |
| 66 |
|
| 67 |
args.header = cmd << NOUVEAU_SVM_BIND_COMMAND_SHIFT; |
| 68 |
args.header |= prio << NOUVEAU_SVM_BIND_PRIORITY_SHIFT; |
| 69 |
args.header |= target << NOUVEAU_SVM_BIND_TARGET_SHIFT; |
| 70 |
|
| 71 |
/* This is best effort, so no garanty whatsoever */ |
| 72 |
drmCommandWrite(fd, DRM_NOUVEAU_SVM_BIND, |
| 73 |
&args, sizeof(args)); |
| 74 |
} |
| 75 |
} |
| 76 |
|
| 77 |
|
| 78 |
static void |
| 79 |
nvc0_flush(struct pipe_context *pipe, |
| 80 |
struct pipe_fence_handle **fence, |
| 81 |
unsigned flags) |
| 82 |
{ |
| 83 |
struct nvc0_context *nvc0 = nvc0_context(pipe); |
| 84 |
|
| 85 |
if (fence) |
| 86 |
nouveau_fence_ref(nvc0->base.fence, (struct nouveau_fence **)fence); |
| 87 |
|
| 88 |
PUSH_KICK(nvc0->base.pushbuf); /* fencing handled in kick_notify */ |
| 89 |
|
| 90 |
nouveau_context_update_frame_stats(&nvc0->base); |
| 91 |
} |
| 92 |
|
| 93 |
static void |
| 94 |
nvc0_texture_barrier(struct pipe_context *pipe, unsigned flags) |
| 95 |
{ |
| 96 |
struct nouveau_pushbuf *push = nvc0_context(pipe)->base.pushbuf; |
| 97 |
|
| 98 |
IMMED_NVC0(push, NVC0_3D(SERIALIZE), 0); |
| 99 |
IMMED_NVC0(push, NVC0_3D(TEX_CACHE_CTL), 0); |
| 100 |
} |
| 101 |
|
| 102 |
static void |
| 103 |
nvc0_memory_barrier(struct pipe_context *pipe, unsigned flags) |
| 104 |
{ |
| 105 |
struct nvc0_context *nvc0 = nvc0_context(pipe); |
| 106 |
struct nouveau_pushbuf *push = nvc0->base.pushbuf; |
| 107 |
int i, s; |
| 108 |
|
| 109 |
if (!(flags & ~PIPE_BARRIER_UPDATE)) |
| 110 |
return; |
| 111 |
|
| 112 |
if (flags & PIPE_BARRIER_MAPPED_BUFFER) { |
| 113 |
for (i = 0; i < nvc0->num_vtxbufs; ++i) { |
| 114 |
if (!nvc0->vtxbuf[i].buffer.resource && !nvc0->vtxbuf[i].is_user_buffer) |
| 115 |
continue; |
| 116 |
if (nvc0->vtxbuf[i].buffer.resource->flags & PIPE_RESOURCE_FLAG_MAP_PERSISTENT) |
| 117 |
nvc0->base.vbo_dirty = true; |
| 118 |
} |
| 119 |
|
| 120 |
for (s = 0; s < 5 && !nvc0->cb_dirty; ++s) { |
| 121 |
uint32_t valid = nvc0->constbuf_valid[s]; |
| 122 |
|
| 123 |
while (valid && !nvc0->cb_dirty) { |
| 124 |
const unsigned i = ffs(valid) - 1; |
| 125 |
struct pipe_resource *res; |
| 126 |
|
| 127 |
valid &= ~(1 << i); |
| 128 |
if (nvc0->constbuf[s][i].user) |
| 129 |
continue; |
| 130 |
|
| 131 |
res = nvc0->constbuf[s][i].u.buf; |
| 132 |
if (!res) |
| 133 |
continue; |
| 134 |
|
| 135 |
if (res->flags & PIPE_RESOURCE_FLAG_MAP_PERSISTENT) |
| 136 |
nvc0->cb_dirty = true; |
| 137 |
} |
| 138 |
} |
| 139 |
} else { |
| 140 |
/* Pretty much any writing by shaders needs a serialize after |
| 141 |
* it. Especially when moving between 3d and compute pipelines, but even |
| 142 |
* without that. |
| 143 |
*/ |
| 144 |
IMMED_NVC0(push, NVC0_3D(SERIALIZE), 0); |
| 145 |
} |
| 146 |
|
| 147 |
/* If we're going to texture from a buffer/image written by a shader, we |
| 148 |
* must flush the texture cache. |
| 149 |
*/ |
| 150 |
if (flags & PIPE_BARRIER_TEXTURE) |
| 151 |
IMMED_NVC0(push, NVC0_3D(TEX_CACHE_CTL), 0); |
| 152 |
|
| 153 |
if (flags & PIPE_BARRIER_CONSTANT_BUFFER) |
| 154 |
nvc0->cb_dirty = true; |
| 155 |
if (flags & (PIPE_BARRIER_VERTEX_BUFFER | PIPE_BARRIER_INDEX_BUFFER)) |
| 156 |
nvc0->base.vbo_dirty = true; |
| 157 |
} |
| 158 |
|
| 159 |
static void |
| 160 |
nvc0_emit_string_marker(struct pipe_context *pipe, const char *str, int len) |
| 161 |
{ |
| 162 |
struct nouveau_pushbuf *push = nvc0_context(pipe)->base.pushbuf; |
| 163 |
int string_words = len / 4; |
| 164 |
int data_words; |
| 165 |
|
| 166 |
if (len <= 0) |
| 167 |
return; |
| 168 |
string_words = MIN2(string_words, NV04_PFIFO_MAX_PACKET_LEN); |
| 169 |
if (string_words == NV04_PFIFO_MAX_PACKET_LEN) |
| 170 |
data_words = string_words; |
| 171 |
else |
| 172 |
data_words = string_words + !!(len & 3); |
| 173 |
BEGIN_NIC0(push, SUBC_3D(NV04_GRAPH_NOP), data_words); |
| 174 |
if (string_words) |
| 175 |
PUSH_DATAp(push, str, string_words); |
| 176 |
if (string_words != data_words) { |
| 177 |
int data = 0; |
| 178 |
memcpy(&data, &str[string_words * 4], len & 3); |
| 179 |
PUSH_DATA (push, data); |
| 180 |
} |
| 181 |
} |
| 182 |
|
| 183 |
static enum pipe_reset_status |
| 184 |
nvc0_get_device_reset_status(struct pipe_context *pipe) |
| 185 |
{ |
| 186 |
return PIPE_NO_RESET; |
| 187 |
} |
| 188 |
|
| 189 |
static void |
| 190 |
nvc0_context_unreference_resources(struct nvc0_context *nvc0) |
| 191 |
{ |
| 192 |
unsigned s, i; |
| 193 |
|
| 194 |
nouveau_bufctx_del(&nvc0->bufctx_3d); |
| 195 |
nouveau_bufctx_del(&nvc0->bufctx); |
| 196 |
nouveau_bufctx_del(&nvc0->bufctx_cp); |
| 197 |
|
| 198 |
nvc0_framebuffer_init(&nvc0->base.pipe, NULL, nvc0->fb_cbufs, &nvc0->fb_zsbuf); |
| 199 |
util_unreference_framebuffer_state(&nvc0->framebuffer); |
| 200 |
|
| 201 |
for (i = 0; i < nvc0->num_vtxbufs; ++i) |
| 202 |
pipe_vertex_buffer_unreference(&nvc0->vtxbuf[i]); |
| 203 |
|
| 204 |
for (s = 0; s < 6; ++s) { |
| 205 |
for (i = 0; i < nvc0->num_textures[s]; ++i) |
| 206 |
pipe_sampler_view_reference(&nvc0->textures[s][i], NULL); |
| 207 |
|
| 208 |
for (i = 0; i < NVC0_MAX_PIPE_CONSTBUFS; ++i) |
| 209 |
if (!nvc0->constbuf[s][i].user) |
| 210 |
pipe_resource_reference(&nvc0->constbuf[s][i].u.buf, NULL); |
| 211 |
|
| 212 |
for (i = 0; i < NVC0_MAX_BUFFERS; ++i) |
| 213 |
pipe_resource_reference(&nvc0->buffers[s][i].buffer, NULL); |
| 214 |
|
| 215 |
for (i = 0; i < NVC0_MAX_IMAGES; ++i) { |
| 216 |
pipe_resource_reference(&nvc0->images[s][i].resource, NULL); |
| 217 |
if (nvc0->screen->base.class_3d >= GM107_3D_CLASS) |
| 218 |
pipe_sampler_view_reference(&nvc0->images_tic[s][i], NULL); |
| 219 |
} |
| 220 |
} |
| 221 |
|
| 222 |
for (i = 0; i < nvc0->num_tfbbufs; ++i) |
| 223 |
pipe_so_target_reference(&nvc0->tfbbuf[i], NULL); |
| 224 |
|
| 225 |
for (i = 0; i < nvc0->global_residents.size / sizeof(struct pipe_resource *); |
| 226 |
++i) { |
| 227 |
struct pipe_resource **res = util_dynarray_element( |
| 228 |
&nvc0->global_residents, struct pipe_resource *, i); |
| 229 |
pipe_resource_reference(res, NULL); |
| 230 |
} |
| 231 |
util_dynarray_fini(&nvc0->global_residents); |
| 232 |
|
| 233 |
if (nvc0->tcp_empty) |
| 234 |
nvc0->base.pipe.delete_tcs_state(&nvc0->base.pipe, nvc0->tcp_empty); |
| 235 |
} |
| 236 |
|
| 237 |
static void |
| 238 |
nvc0_destroy(struct pipe_context *pipe) |
| 239 |
{ |
| 240 |
struct nvc0_context *nvc0 = nvc0_context(pipe); |
| 241 |
|
| 242 |
simple_mtx_lock(&nvc0->screen->state_lock); |
| 243 |
if (nvc0->screen->cur_ctx == nvc0) { |
| 244 |
nvc0->screen->cur_ctx = NULL; |
| 245 |
nvc0->screen->save_state = nvc0->state; |
| 246 |
nvc0->screen->save_state.tfb = NULL; |
| 247 |
} |
| 248 |
simple_mtx_unlock(&nvc0->screen->state_lock); |
| 249 |
|
| 250 |
if (nvc0->base.pipe.stream_uploader) |
| 251 |
u_upload_destroy(nvc0->base.pipe.stream_uploader); |
| 252 |
|
| 253 |
/* Unset bufctx, we don't want to revalidate any resources after the flush. |
| 254 |
* Other contexts will always set their bufctx again on action calls. |
| 255 |
*/ |
| 256 |
nouveau_pushbuf_bufctx(nvc0->base.pushbuf, NULL); |
| 257 |
PUSH_KICK(nvc0->base.pushbuf); |
| 258 |
|
| 259 |
nvc0_context_unreference_resources(nvc0); |
| 260 |
nvc0_blitctx_destroy(nvc0); |
| 261 |
|
| 262 |
list_for_each_entry_safe(struct nvc0_resident, pos, &nvc0->tex_head, list) { |
| 263 |
list_del(&pos->list); |
| 264 |
free(pos); |
| 265 |
} |
| 266 |
|
| 267 |
list_for_each_entry_safe(struct nvc0_resident, pos, &nvc0->img_head, list) { |
| 268 |
list_del(&pos->list); |
| 269 |
free(pos); |
| 270 |
} |
| 271 |
|
| 272 |
nouveau_fence_cleanup(&nvc0->base); |
| 273 |
nouveau_context_destroy(&nvc0->base); |
| 274 |
} |
| 275 |
|
| 276 |
void |
| 277 |
nvc0_default_kick_notify(struct nouveau_context *context) |
| 278 |
{ |
| 279 |
struct nvc0_context *nvc0 = nvc0_context(&context->pipe); |
| 280 |
|
| 281 |
_nouveau_fence_next(context); |
| 282 |
_nouveau_fence_update(context->screen, true); |
| 283 |
|
| 284 |
nvc0->state.flushed = true; |
| 285 |
} |
| 286 |
|
| 287 |
static int |
| 288 |
nvc0_invalidate_resource_storage(struct nouveau_context *ctx, |
| 289 |
struct pipe_resource *res, |
| 290 |
int ref) |
| 291 |
{ |
| 292 |
struct nvc0_context *nvc0 = nvc0_context(&ctx->pipe); |
| 293 |
unsigned s, i; |
| 294 |
|
| 295 |
if (res->bind & PIPE_BIND_RENDER_TARGET) { |
| 296 |
for (i = 0; i < nvc0->framebuffer.nr_cbufs; ++i) { |
| 297 |
if (nvc0->framebuffer.cbufs[i].texture == res) { |
| 298 |
nvc0->dirty_3d |= NVC0_NEW_3D_FRAMEBUFFER; |
| 299 |
nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_FB); |
| 300 |
if (!--ref) |
| 301 |
return ref; |
| 302 |
} |
| 303 |
} |
| 304 |
} |
| 305 |
if (res->bind & PIPE_BIND_DEPTH_STENCIL) { |
| 306 |
if (nvc0->framebuffer.zsbuf.texture == res) { |
| 307 |
nvc0->dirty_3d |= NVC0_NEW_3D_FRAMEBUFFER; |
| 308 |
nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_FB); |
| 309 |
if (!--ref) |
| 310 |
return ref; |
| 311 |
} |
| 312 |
} |
| 313 |
|
| 314 |
if (res->target == PIPE_BUFFER) { |
| 315 |
for (i = 0; i < nvc0->num_vtxbufs; ++i) { |
| 316 |
if (nvc0->vtxbuf[i].buffer.resource == res) { |
| 317 |
nvc0->dirty_3d |= NVC0_NEW_3D_ARRAYS; |
| 318 |
nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_VTX); |
| 319 |
if (!--ref) |
| 320 |
return ref; |
| 321 |
} |
| 322 |
} |
| 323 |
|
| 324 |
for (s = 0; s < 6; ++s) { |
| 325 |
for (i = 0; i < nvc0->num_textures[s]; ++i) { |
| 326 |
if (nvc0->textures[s][i] && |
| 327 |
nvc0->textures[s][i]->texture == res) { |
| 328 |
nvc0->textures_dirty[s] |= 1 << i; |
| 329 |
if (unlikely(s == 5)) { |
| 330 |
nvc0->dirty_cp |= NVC0_NEW_CP_TEXTURES; |
| 331 |
nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_TEX(i)); |
| 332 |
} else { |
| 333 |
nvc0->dirty_3d |= NVC0_NEW_3D_TEXTURES; |
| 334 |
nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_TEX(s, i)); |
| 335 |
} |
| 336 |
if (!--ref) |
| 337 |
return ref; |
| 338 |
} |
| 339 |
} |
| 340 |
} |
| 341 |
|
| 342 |
for (s = 0; s < 6; ++s) { |
| 343 |
for (i = 0; i < NVC0_MAX_PIPE_CONSTBUFS; ++i) { |
| 344 |
if (!(nvc0->constbuf_valid[s] & (1 << i))) |
| 345 |
continue; |
| 346 |
if (!nvc0->constbuf[s][i].user && |
| 347 |
nvc0->constbuf[s][i].u.buf == res) { |
| 348 |
nvc0->constbuf_dirty[s] |= 1 << i; |
| 349 |
if (unlikely(s == 5)) { |
| 350 |
nvc0->dirty_cp |= NVC0_NEW_CP_CONSTBUF; |
| 351 |
nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_CB(i)); |
| 352 |
} else { |
| 353 |
nvc0->dirty_3d |= NVC0_NEW_3D_CONSTBUF; |
| 354 |
nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_CB(s, i)); |
| 355 |
} |
| 356 |
if (!--ref) |
| 357 |
return ref; |
| 358 |
} |
| 359 |
} |
| 360 |
} |
| 361 |
|
| 362 |
for (s = 0; s < 6; ++s) { |
| 363 |
for (i = 0; i < NVC0_MAX_BUFFERS; ++i) { |
| 364 |
if (nvc0->buffers[s][i].buffer == res) { |
| 365 |
nvc0->buffers_dirty[s] |= 1 << i; |
| 366 |
if (unlikely(s == 5)) { |
| 367 |
nvc0->dirty_cp |= NVC0_NEW_CP_BUFFERS; |
| 368 |
nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_BUF); |
| 369 |
} else { |
| 370 |
nvc0->dirty_3d |= NVC0_NEW_3D_BUFFERS; |
| 371 |
nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_BUF); |
| 372 |
} |
| 373 |
if (!--ref) |
| 374 |
return ref; |
| 375 |
} |
| 376 |
} |
| 377 |
} |
| 378 |
|
| 379 |
for (s = 0; s < 6; ++s) { |
| 380 |
for (i = 0; i < NVC0_MAX_IMAGES; ++i) { |
| 381 |
if (nvc0->images[s][i].resource == res) { |
| 382 |
nvc0->images_dirty[s] |= 1 << i; |
| 383 |
if (unlikely(s == 5)) { |
| 384 |
nvc0->dirty_cp |= NVC0_NEW_CP_SURFACES; |
| 385 |
nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_SUF); |
| 386 |
} else { |
| 387 |
nvc0->dirty_3d |= NVC0_NEW_3D_SURFACES; |
| 388 |
nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_SUF); |
| 389 |
} |
| 390 |
} |
| 391 |
if (!--ref) |
| 392 |
return ref; |
| 393 |
} |
| 394 |
} |
| 395 |
} |
| 396 |
|
| 397 |
return ref; |
| 398 |
} |
| 399 |
|
| 400 |
static void |
| 401 |
nvc0_context_get_sample_position(struct pipe_context *, unsigned, unsigned, |
| 402 |
float *); |
| 403 |
|
| 404 |
struct pipe_context * |
| 405 |
nvc0_create(struct pipe_screen *pscreen, void *priv, unsigned ctxflags) |
| 406 |
{ |
| 407 |
struct nvc0_screen *screen = nvc0_screen(pscreen); |
| 408 |
struct nvc0_context *nvc0; |
| 409 |
struct pipe_context *pipe; |
| 410 |
int ret; |
| 411 |
uint32_t flags; |
| 412 |
|
| 413 |
nvc0 = CALLOC_STRUCT(nvc0_context); |
| 414 |
if (!nvc0) |
| 415 |
return NULL; |
| 416 |
pipe = &nvc0->base.pipe; |
| 417 |
|
| 418 |
if (!nvc0_blitctx_create(nvc0)) |
| 419 |
goto out_err; |
| 420 |
|
| 421 |
if (nouveau_context_init(&nvc0->base, &screen->base)) |
| 422 |
goto out_err; |
| 423 |
nvc0->base.kick_notify = nvc0_default_kick_notify; |
| 424 |
nvc0->base.pushbuf->rsvd_kick = 5; |
| 425 |
|
| 426 |
ret = nouveau_bufctx_new(nvc0->base.client, 2, &nvc0->bufctx); |
| 427 |
if (!ret) |
| 428 |
ret = nouveau_bufctx_new(nvc0->base.client, NVC0_BIND_3D_COUNT, |
| 429 |
&nvc0->bufctx_3d); |
| 430 |
if (!ret) |
| 431 |
ret = nouveau_bufctx_new(nvc0->base.client, NVC0_BIND_CP_COUNT, |
| 432 |
&nvc0->bufctx_cp); |
| 433 |
if (ret) |
| 434 |
goto out_err; |
| 435 |
|
| 436 |
nvc0->screen = screen; |
| 437 |
pipe->screen = pscreen; |
| 438 |
pipe->priv = priv; |
| 439 |
pipe->stream_uploader = u_upload_create_default(pipe); |
| 440 |
if (!pipe->stream_uploader) |
| 441 |
goto out_err; |
| 442 |
pipe->const_uploader = pipe->stream_uploader; |
| 443 |
|
| 444 |
pipe->destroy = nvc0_destroy; |
| 445 |
|
| 446 |
pipe->draw_vbo = nvc0_draw_vbo; |
| 447 |
pipe->clear = nvc0_clear; |
| 448 |
pipe->launch_grid = (nvc0->screen->base.class_3d >= NVE4_3D_CLASS) ? |
| 449 |
nve4_launch_grid : nvc0_launch_grid; |
| 450 |
|
| 451 |
pipe->svm_migrate = nvc0_svm_migrate; |
| 452 |
|
| 453 |
pipe->flush = nvc0_flush; |
| 454 |
pipe->texture_barrier = nvc0_texture_barrier; |
| 455 |
pipe->memory_barrier = nvc0_memory_barrier; |
| 456 |
pipe->get_sample_position = nvc0_context_get_sample_position; |
| 457 |
pipe->emit_string_marker = nvc0_emit_string_marker; |
| 458 |
pipe->get_device_reset_status = nvc0_get_device_reset_status; |
| 459 |
|
| 460 |
nvc0_init_query_functions(nvc0); |
| 461 |
nvc0_init_surface_functions(nvc0); |
| 462 |
nvc0_init_state_functions(nvc0); |
| 463 |
nvc0_init_transfer_functions(nvc0); |
| 464 |
nvc0_init_resource_functions(pipe); |
| 465 |
if (nvc0->screen->base.class_3d >= NVE4_3D_CLASS) |
| 466 |
nvc0_init_bindless_functions(pipe); |
| 467 |
|
| 468 |
list_inithead(&nvc0->tex_head); |
| 469 |
list_inithead(&nvc0->img_head); |
| 470 |
|
| 471 |
nvc0->base.invalidate_resource_storage = nvc0_invalidate_resource_storage; |
| 472 |
|
| 473 |
pipe->create_video_codec = nvc0_create_decoder; |
| 474 |
pipe->create_video_buffer = nvc0_video_buffer_create; |
| 475 |
|
| 476 |
/* shader builtin library is per-screen, but we need a context for m2mf */ |
| 477 |
nvc0_program_library_upload(nvc0); |
| 478 |
nvc0_program_init_tcp_empty(nvc0); |
| 479 |
if (!nvc0->tcp_empty) |
| 480 |
goto out_err; |
| 481 |
/* set the empty tctl prog on next draw in case one is never set */ |
| 482 |
nvc0->dirty_3d |= NVC0_NEW_3D_TCTLPROG; |
| 483 |
|
| 484 |
/* Do not bind the COMPUTE driver constbuf at screen initialization because |
| 485 |
* CBs are aliased between 3D and COMPUTE, but make sure it will be bound if |
| 486 |
* a grid is launched later. */ |
| 487 |
nvc0->dirty_cp |= NVC0_NEW_CP_DRIVERCONST; |
| 488 |
|
| 489 |
/* now that there are no more opportunities for errors, set the current |
| 490 |
* context if there isn't already one. |
| 491 |
*/ |
| 492 |
simple_mtx_lock(&screen->state_lock); |
| 493 |
if (!screen->cur_ctx) { |
| 494 |
nvc0->state = screen->save_state; |
| 495 |
screen->cur_ctx = nvc0; |
| 496 |
} |
| 497 |
simple_mtx_unlock(&screen->state_lock); |
| 498 |
|
| 499 |
nouveau_pushbuf_bufctx(nvc0->base.pushbuf, nvc0->bufctx); |
| 500 |
PUSH_SPACE(nvc0->base.pushbuf, 8); |
| 501 |
|
| 502 |
/* add permanently resident buffers to bufctxts */ |
| 503 |
|
| 504 |
flags = NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RD; |
| 505 |
|
| 506 |
BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->uniform_bo); |
| 507 |
BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->txc); |
| 508 |
if (screen->compute) { |
| 509 |
BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->uniform_bo); |
| 510 |
BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->txc); |
| 511 |
} |
| 512 |
|
| 513 |
flags = NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RDWR; |
| 514 |
|
| 515 |
if (screen->poly_cache) |
| 516 |
BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->poly_cache); |
| 517 |
if (screen->compute) |
| 518 |
BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->tls); |
| 519 |
|
| 520 |
flags = NOUVEAU_BO_GART | NOUVEAU_BO_WR; |
| 521 |
|
| 522 |
BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->fence.bo); |
| 523 |
BCTX_REFN_bo(nvc0->bufctx, FENCE, flags, screen->fence.bo); |
| 524 |
if (screen->compute) |
| 525 |
BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->fence.bo); |
| 526 |
|
| 527 |
nvc0->base.scratch.bo_size = 2 << 20; |
| 528 |
|
| 529 |
memset(nvc0->tex_handles, ~0, sizeof(nvc0->tex_handles)); |
| 530 |
|
| 531 |
nvc0->global_residents = UTIL_DYNARRAY_INIT; |
| 532 |
|
| 533 |
// Make sure that the first TSC entry has SRGB conversion bit set, since we |
| 534 |
// use it as a fallback on Fermi for TXF, and on Kepler+ generations for |
| 535 |
// FBFETCH handling (which also uses TXF). |
| 536 |
// |
| 537 |
// NOTE: Preliminary testing suggests that this isn't necessary at all at |
| 538 |
// least on GM20x (untested on Kepler). However this is ~free, so no reason |
| 539 |
// not to do it. |
| 540 |
if (!screen->tsc.entries[0]) |
| 541 |
nvc0_upload_tsc0(nvc0); |
| 542 |
|
| 543 |
// On Fermi, mark samplers dirty so that the proper binding can happen |
| 544 |
if (screen->base.class_3d < NVE4_3D_CLASS) { |
| 545 |
for (int s = 0; s < 6; s++) |
| 546 |
nvc0->samplers_dirty[s] = 1; |
| 547 |
nvc0->dirty_3d |= NVC0_NEW_3D_SAMPLERS; |
| 548 |
nvc0->dirty_cp |= NVC0_NEW_CP_SAMPLERS; |
| 549 |
} |
| 550 |
|
| 551 |
nouveau_fence_new(&nvc0->base, &nvc0->base.fence); |
| 552 |
|
| 553 |
return pipe; |
| 554 |
|
| 555 |
out_err: |
| 556 |
if (nvc0) { |
| 557 |
if (pipe->stream_uploader) |
| 558 |
u_upload_destroy(pipe->stream_uploader); |
| 559 |
if (nvc0->bufctx_3d) |
| 560 |
nouveau_bufctx_del(&nvc0->bufctx_3d); |
| 561 |
if (nvc0->bufctx_cp) |
| 562 |
nouveau_bufctx_del(&nvc0->bufctx_cp); |
| 563 |
if (nvc0->bufctx) |
| 564 |
nouveau_bufctx_del(&nvc0->bufctx); |
| 565 |
FREE(nvc0->blit); |
| 566 |
FREE(nvc0); |
| 567 |
} |
| 568 |
return NULL; |
| 569 |
} |
| 570 |
|
| 571 |
void |
| 572 |
nvc0_bufctx_fence(struct nvc0_context *nvc0, struct nouveau_bufctx *bufctx, |
| 573 |
bool on_flush) |
| 574 |
{ |
| 575 |
struct list_head *list = on_flush ? &bufctx->current : &bufctx->pending; |
| 576 |
struct list_head *it; |
| 577 |
NOUVEAU_DRV_STAT_IFD(unsigned count = 0); |
| 578 |
|
| 579 |
for (it = list->next; it != list; it = it->next) { |
| 580 |
struct nouveau_bufref *ref = (struct nouveau_bufref *)it; |
| 581 |
struct nv04_resource *res = ref->priv; |
| 582 |
if (res) |
| 583 |
nvc0_resource_validate(nvc0, res, (unsigned)ref->priv_data); |
| 584 |
NOUVEAU_DRV_STAT_IFD(count++); |
| 585 |
} |
| 586 |
NOUVEAU_DRV_STAT(&nvc0->screen->base, resource_validate_count, count); |
| 587 |
} |
| 588 |
|
| 589 |
const void * |
| 590 |
nvc0_get_sample_locations(unsigned sample_count) |
| 591 |
{ |
| 592 |
static const uint8_t ms1[1][2] = { { 0x8, 0x8 } }; |
| 593 |
static const uint8_t ms2[2][2] = { |
| 594 |
{ 0x4, 0x4 }, { 0xc, 0xc } }; /* surface coords (0,0), (1,0) */ |
| 595 |
static const uint8_t ms4[4][2] = { |
| 596 |
{ 0x6, 0x2 }, { 0xe, 0x6 }, /* (0,0), (1,0) */ |
| 597 |
{ 0x2, 0xa }, { 0xa, 0xe } }; /* (0,1), (1,1) */ |
| 598 |
static const uint8_t ms8[8][2] = { |
| 599 |
{ 0x1, 0x7 }, { 0x5, 0x3 }, /* (0,0), (1,0) */ |
| 600 |
{ 0x3, 0xd }, { 0x7, 0xb }, /* (0,1), (1,1) */ |
| 601 |
{ 0x9, 0x5 }, { 0xf, 0x1 }, /* (2,0), (3,0) */ |
| 602 |
{ 0xb, 0xf }, { 0xd, 0x9 } }; /* (2,1), (3,1) */ |
| 603 |
|
| 604 |
const uint8_t (*ptr)[2]; |
| 605 |
|
| 606 |
switch (sample_count) { |
| 607 |
case 0: |
| 608 |
case 1: ptr = ms1; break; |
| 609 |
case 2: ptr = ms2; break; |
| 610 |
case 4: ptr = ms4; break; |
| 611 |
case 8: ptr = ms8; break; |
| 612 |
default: |
| 613 |
assert(0); |
| 614 |
return NULL; /* bad sample count -> undefined locations */ |
| 615 |
} |
| 616 |
return ptr; |
| 617 |
} |
| 618 |
|
| 619 |
static void |
| 620 |
nvc0_context_get_sample_position(struct pipe_context *pipe, |
| 621 |
unsigned sample_count, unsigned sample_index, |
| 622 |
float *xy) |
| 623 |
{ |
| 624 |
const uint8_t (*ptr)[2]; |
| 625 |
|
| 626 |
ptr = nvc0_get_sample_locations(sample_count); |
| 627 |
if (!ptr) |
| 628 |
return; |
| 629 |
|
| 630 |
xy[0] = ptr[sample_index][0] * 0.0625f; |
| 631 |
xy[1] = ptr[sample_index][1] * 0.0625f; |
| 632 |
} |