1 /*
2 * Copyright © 2011 Red Hat All Rights Reserved.
3 * Copyright © 2014 Advanced Micro Devices, Inc.
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining
7 * a copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
15 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
16 * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
17 * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS
18 * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
20 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
22 *
23 * The above copyright notice and this permission notice (including the
24 * next paragraph) shall be included in all copies or substantial portions
25 * of the Software.
26 */
27
28 /* Contact:
29 * Marek Olšák <maraeo@gmail.com>
30 */
31
32 #include "amdgpu_winsys.h"
33 #include "util/u_format.h"
34
35 #ifndef CIASICIDGFXENGINE_SOUTHERNISLAND
36 #define CIASICIDGFXENGINE_SOUTHERNISLAND 0x0000000A
37 #endif
38
39
amdgpu_surface_sanity(const struct pipe_resource * tex)40 static int amdgpu_surface_sanity(const struct pipe_resource *tex)
41 {
42 /* all dimension must be at least 1 ! */
43 if (!tex->width0 || !tex->height0 || !tex->depth0 ||
44 !tex->array_size)
45 return -EINVAL;
46
47 switch (tex->nr_samples) {
48 case 0:
49 case 1:
50 case 2:
51 case 4:
52 case 8:
53 break;
54 default:
55 return -EINVAL;
56 }
57
58 switch (tex->target) {
59 case PIPE_TEXTURE_1D:
60 if (tex->height0 > 1)
61 return -EINVAL;
62 /* fall through */
63 case PIPE_TEXTURE_2D:
64 case PIPE_TEXTURE_RECT:
65 if (tex->depth0 > 1 || tex->array_size > 1)
66 return -EINVAL;
67 break;
68 case PIPE_TEXTURE_3D:
69 if (tex->array_size > 1)
70 return -EINVAL;
71 break;
72 case PIPE_TEXTURE_1D_ARRAY:
73 if (tex->height0 > 1)
74 return -EINVAL;
75 /* fall through */
76 case PIPE_TEXTURE_CUBE:
77 case PIPE_TEXTURE_2D_ARRAY:
78 case PIPE_TEXTURE_CUBE_ARRAY:
79 if (tex->depth0 > 1)
80 return -EINVAL;
81 break;
82 default:
83 return -EINVAL;
84 }
85 return 0;
86 }
87
allocSysMem(const ADDR_ALLOCSYSMEM_INPUT * pInput)88 static void *ADDR_API allocSysMem(const ADDR_ALLOCSYSMEM_INPUT * pInput)
89 {
90 return malloc(pInput->sizeInBytes);
91 }
92
freeSysMem(const ADDR_FREESYSMEM_INPUT * pInput)93 static ADDR_E_RETURNCODE ADDR_API freeSysMem(const ADDR_FREESYSMEM_INPUT * pInput)
94 {
95 free(pInput->pVirtAddr);
96 return ADDR_OK;
97 }
98
amdgpu_addr_create(struct amdgpu_winsys * ws)99 ADDR_HANDLE amdgpu_addr_create(struct amdgpu_winsys *ws)
100 {
101 ADDR_CREATE_INPUT addrCreateInput = {0};
102 ADDR_CREATE_OUTPUT addrCreateOutput = {0};
103 ADDR_REGISTER_VALUE regValue = {0};
104 ADDR_CREATE_FLAGS createFlags = {{0}};
105 ADDR_E_RETURNCODE addrRet;
106
107 addrCreateInput.size = sizeof(ADDR_CREATE_INPUT);
108 addrCreateOutput.size = sizeof(ADDR_CREATE_OUTPUT);
109
110 regValue.noOfBanks = ws->amdinfo.mc_arb_ramcfg & 0x3;
111 regValue.gbAddrConfig = ws->amdinfo.gb_addr_cfg;
112 regValue.noOfRanks = (ws->amdinfo.mc_arb_ramcfg & 0x4) >> 2;
113
114 regValue.backendDisables = ws->amdinfo.backend_disable[0];
115 regValue.pTileConfig = ws->amdinfo.gb_tile_mode;
116 regValue.noOfEntries = ARRAY_SIZE(ws->amdinfo.gb_tile_mode);
117 if (ws->info.chip_class == SI) {
118 regValue.pMacroTileConfig = NULL;
119 regValue.noOfMacroEntries = 0;
120 } else {
121 regValue.pMacroTileConfig = ws->amdinfo.gb_macro_tile_mode;
122 regValue.noOfMacroEntries = ARRAY_SIZE(ws->amdinfo.gb_macro_tile_mode);
123 }
124
125 createFlags.value = 0;
126 createFlags.useTileIndex = 1;
127 createFlags.degradeBaseLevel = 1;
128 createFlags.useHtileSliceAlign = 1;
129
130 addrCreateInput.chipEngine = CIASICIDGFXENGINE_SOUTHERNISLAND;
131 addrCreateInput.chipFamily = ws->family;
132 addrCreateInput.chipRevision = ws->rev_id;
133 addrCreateInput.createFlags = createFlags;
134 addrCreateInput.callbacks.allocSysMem = allocSysMem;
135 addrCreateInput.callbacks.freeSysMem = freeSysMem;
136 addrCreateInput.callbacks.debugPrint = 0;
137 addrCreateInput.regValue = regValue;
138
139 addrRet = AddrCreate(&addrCreateInput, &addrCreateOutput);
140 if (addrRet != ADDR_OK)
141 return NULL;
142
143 return addrCreateOutput.hLib;
144 }
145
compute_level(struct amdgpu_winsys * ws,const struct pipe_resource * tex,struct radeon_surf * surf,bool is_stencil,unsigned level,bool compressed,ADDR_COMPUTE_SURFACE_INFO_INPUT * AddrSurfInfoIn,ADDR_COMPUTE_SURFACE_INFO_OUTPUT * AddrSurfInfoOut,ADDR_COMPUTE_DCCINFO_INPUT * AddrDccIn,ADDR_COMPUTE_DCCINFO_OUTPUT * AddrDccOut,ADDR_COMPUTE_HTILE_INFO_INPUT * AddrHtileIn,ADDR_COMPUTE_HTILE_INFO_OUTPUT * AddrHtileOut)146 static int compute_level(struct amdgpu_winsys *ws,
147 const struct pipe_resource *tex,
148 struct radeon_surf *surf, bool is_stencil,
149 unsigned level, bool compressed,
150 ADDR_COMPUTE_SURFACE_INFO_INPUT *AddrSurfInfoIn,
151 ADDR_COMPUTE_SURFACE_INFO_OUTPUT *AddrSurfInfoOut,
152 ADDR_COMPUTE_DCCINFO_INPUT *AddrDccIn,
153 ADDR_COMPUTE_DCCINFO_OUTPUT *AddrDccOut,
154 ADDR_COMPUTE_HTILE_INFO_INPUT *AddrHtileIn,
155 ADDR_COMPUTE_HTILE_INFO_OUTPUT *AddrHtileOut)
156 {
157 struct radeon_surf_level *surf_level;
158 ADDR_E_RETURNCODE ret;
159
160 AddrSurfInfoIn->mipLevel = level;
161 AddrSurfInfoIn->width = u_minify(tex->width0, level);
162 AddrSurfInfoIn->height = u_minify(tex->height0, level);
163
164 if (tex->target == PIPE_TEXTURE_3D)
165 AddrSurfInfoIn->numSlices = u_minify(tex->depth0, level);
166 else if (tex->target == PIPE_TEXTURE_CUBE)
167 AddrSurfInfoIn->numSlices = 6;
168 else
169 AddrSurfInfoIn->numSlices = tex->array_size;
170
171 if (level > 0) {
172 /* Set the base level pitch. This is needed for calculation
173 * of non-zero levels. */
174 if (is_stencil)
175 AddrSurfInfoIn->basePitch = surf->stencil_level[0].nblk_x;
176 else
177 AddrSurfInfoIn->basePitch = surf->level[0].nblk_x;
178
179 /* Convert blocks to pixels for compressed formats. */
180 if (compressed)
181 AddrSurfInfoIn->basePitch *= surf->blk_w;
182 }
183
184 ret = AddrComputeSurfaceInfo(ws->addrlib,
185 AddrSurfInfoIn,
186 AddrSurfInfoOut);
187 if (ret != ADDR_OK) {
188 return ret;
189 }
190
191 surf_level = is_stencil ? &surf->stencil_level[level] : &surf->level[level];
192 surf_level->offset = align64(surf->surf_size, AddrSurfInfoOut->baseAlign);
193 surf_level->slice_size = AddrSurfInfoOut->sliceSize;
194 surf_level->nblk_x = AddrSurfInfoOut->pitch;
195 surf_level->nblk_y = AddrSurfInfoOut->height;
196
197 switch (AddrSurfInfoOut->tileMode) {
198 case ADDR_TM_LINEAR_ALIGNED:
199 surf_level->mode = RADEON_SURF_MODE_LINEAR_ALIGNED;
200 break;
201 case ADDR_TM_1D_TILED_THIN1:
202 surf_level->mode = RADEON_SURF_MODE_1D;
203 break;
204 case ADDR_TM_2D_TILED_THIN1:
205 surf_level->mode = RADEON_SURF_MODE_2D;
206 break;
207 default:
208 assert(0);
209 }
210
211 if (is_stencil)
212 surf->stencil_tiling_index[level] = AddrSurfInfoOut->tileIndex;
213 else
214 surf->tiling_index[level] = AddrSurfInfoOut->tileIndex;
215
216 surf->surf_size = surf_level->offset + AddrSurfInfoOut->surfSize;
217
218 /* Clear DCC fields at the beginning. */
219 surf_level->dcc_offset = 0;
220
221 /* The previous level's flag tells us if we can use DCC for this level. */
222 if (AddrSurfInfoIn->flags.dccCompatible &&
223 (level == 0 || AddrDccOut->subLvlCompressible)) {
224 AddrDccIn->colorSurfSize = AddrSurfInfoOut->surfSize;
225 AddrDccIn->tileMode = AddrSurfInfoOut->tileMode;
226 AddrDccIn->tileInfo = *AddrSurfInfoOut->pTileInfo;
227 AddrDccIn->tileIndex = AddrSurfInfoOut->tileIndex;
228 AddrDccIn->macroModeIndex = AddrSurfInfoOut->macroModeIndex;
229
230 ret = AddrComputeDccInfo(ws->addrlib,
231 AddrDccIn,
232 AddrDccOut);
233
234 if (ret == ADDR_OK) {
235 surf_level->dcc_offset = surf->dcc_size;
236 surf_level->dcc_fast_clear_size = AddrDccOut->dccFastClearSize;
237 surf->num_dcc_levels = level + 1;
238 surf->dcc_size = surf_level->dcc_offset + AddrDccOut->dccRamSize;
239 surf->dcc_alignment = MAX2(surf->dcc_alignment, AddrDccOut->dccRamBaseAlign);
240 }
241 }
242
243 /* TC-compatible HTILE. */
244 if (!is_stencil &&
245 AddrSurfInfoIn->flags.depth &&
246 AddrSurfInfoIn->flags.tcCompatible &&
247 surf_level->mode == RADEON_SURF_MODE_2D &&
248 level == 0) {
249 AddrHtileIn->flags.tcCompatible = 1;
250 AddrHtileIn->pitch = AddrSurfInfoOut->pitch;
251 AddrHtileIn->height = AddrSurfInfoOut->height;
252 AddrHtileIn->numSlices = AddrSurfInfoOut->depth;
253 AddrHtileIn->blockWidth = ADDR_HTILE_BLOCKSIZE_8;
254 AddrHtileIn->blockHeight = ADDR_HTILE_BLOCKSIZE_8;
255 AddrHtileIn->pTileInfo = AddrSurfInfoOut->pTileInfo;
256 AddrHtileIn->tileIndex = AddrSurfInfoOut->tileIndex;
257 AddrHtileIn->macroModeIndex = AddrSurfInfoOut->macroModeIndex;
258
259 ret = AddrComputeHtileInfo(ws->addrlib,
260 AddrHtileIn,
261 AddrHtileOut);
262
263 if (ret == ADDR_OK) {
264 surf->htile_size = AddrHtileOut->htileBytes;
265 surf->htile_alignment = AddrHtileOut->baseAlign;
266 }
267 }
268
269 return 0;
270 }
271
272 #define G_009910_MICRO_TILE_MODE(x) (((x) >> 0) & 0x03)
273 #define G_009910_MICRO_TILE_MODE_NEW(x) (((x) >> 22) & 0x07)
274
set_micro_tile_mode(struct radeon_surf * surf,struct radeon_info * info)275 static void set_micro_tile_mode(struct radeon_surf *surf,
276 struct radeon_info *info)
277 {
278 uint32_t tile_mode = info->si_tile_mode_array[surf->tiling_index[0]];
279
280 if (info->chip_class >= CIK)
281 surf->micro_tile_mode = G_009910_MICRO_TILE_MODE_NEW(tile_mode);
282 else
283 surf->micro_tile_mode = G_009910_MICRO_TILE_MODE(tile_mode);
284 }
285
cik_get_macro_tile_index(struct radeon_surf * surf)286 static unsigned cik_get_macro_tile_index(struct radeon_surf *surf)
287 {
288 unsigned index, tileb;
289
290 tileb = 8 * 8 * surf->bpe;
291 tileb = MIN2(surf->tile_split, tileb);
292
293 for (index = 0; tileb > 64; index++)
294 tileb >>= 1;
295
296 assert(index < 16);
297 return index;
298 }
299
amdgpu_surface_init(struct radeon_winsys * rws,const struct pipe_resource * tex,unsigned flags,unsigned bpe,enum radeon_surf_mode mode,struct radeon_surf * surf)300 static int amdgpu_surface_init(struct radeon_winsys *rws,
301 const struct pipe_resource *tex,
302 unsigned flags, unsigned bpe,
303 enum radeon_surf_mode mode,
304 struct radeon_surf *surf)
305 {
306 struct amdgpu_winsys *ws = (struct amdgpu_winsys*)rws;
307 unsigned level;
308 bool compressed;
309 ADDR_COMPUTE_SURFACE_INFO_INPUT AddrSurfInfoIn = {0};
310 ADDR_COMPUTE_SURFACE_INFO_OUTPUT AddrSurfInfoOut = {0};
311 ADDR_COMPUTE_DCCINFO_INPUT AddrDccIn = {0};
312 ADDR_COMPUTE_DCCINFO_OUTPUT AddrDccOut = {0};
313 ADDR_COMPUTE_HTILE_INFO_INPUT AddrHtileIn = {0};
314 ADDR_COMPUTE_HTILE_INFO_OUTPUT AddrHtileOut = {0};
315 ADDR_TILEINFO AddrTileInfoIn = {0};
316 ADDR_TILEINFO AddrTileInfoOut = {0};
317 int r;
318
319 r = amdgpu_surface_sanity(tex);
320 if (r)
321 return r;
322
323 AddrSurfInfoIn.size = sizeof(ADDR_COMPUTE_SURFACE_INFO_INPUT);
324 AddrSurfInfoOut.size = sizeof(ADDR_COMPUTE_SURFACE_INFO_OUTPUT);
325 AddrDccIn.size = sizeof(ADDR_COMPUTE_DCCINFO_INPUT);
326 AddrDccOut.size = sizeof(ADDR_COMPUTE_DCCINFO_OUTPUT);
327 AddrHtileIn.size = sizeof(ADDR_COMPUTE_HTILE_INFO_INPUT);
328 AddrHtileOut.size = sizeof(ADDR_COMPUTE_HTILE_INFO_OUTPUT);
329 AddrSurfInfoOut.pTileInfo = &AddrTileInfoOut;
330
331 surf->blk_w = util_format_get_blockwidth(tex->format);
332 surf->blk_h = util_format_get_blockheight(tex->format);
333 surf->bpe = bpe;
334 surf->flags = flags;
335
336 compressed = surf->blk_w == 4 && surf->blk_h == 4;
337
338 /* MSAA and FMASK require 2D tiling. */
339 if (tex->nr_samples > 1 ||
340 (flags & RADEON_SURF_FMASK))
341 mode = RADEON_SURF_MODE_2D;
342
343 /* DB doesn't support linear layouts. */
344 if (flags & (RADEON_SURF_Z_OR_SBUFFER) &&
345 mode < RADEON_SURF_MODE_1D)
346 mode = RADEON_SURF_MODE_1D;
347
348 /* Set the requested tiling mode. */
349 switch (mode) {
350 case RADEON_SURF_MODE_LINEAR_ALIGNED:
351 AddrSurfInfoIn.tileMode = ADDR_TM_LINEAR_ALIGNED;
352 break;
353 case RADEON_SURF_MODE_1D:
354 AddrSurfInfoIn.tileMode = ADDR_TM_1D_TILED_THIN1;
355 break;
356 case RADEON_SURF_MODE_2D:
357 AddrSurfInfoIn.tileMode = ADDR_TM_2D_TILED_THIN1;
358 break;
359 default:
360 assert(0);
361 }
362
363 /* The format must be set correctly for the allocation of compressed
364 * textures to work. In other cases, setting the bpp is sufficient. */
365 if (compressed) {
366 switch (bpe) {
367 case 8:
368 AddrSurfInfoIn.format = ADDR_FMT_BC1;
369 break;
370 case 16:
371 AddrSurfInfoIn.format = ADDR_FMT_BC3;
372 break;
373 default:
374 assert(0);
375 }
376 }
377 else {
378 AddrDccIn.bpp = AddrSurfInfoIn.bpp = bpe * 8;
379 }
380
381 AddrDccIn.numSamples = AddrSurfInfoIn.numSamples =
382 tex->nr_samples ? tex->nr_samples : 1;
383 AddrSurfInfoIn.tileIndex = -1;
384
385 /* Set the micro tile type. */
386 if (flags & RADEON_SURF_SCANOUT)
387 AddrSurfInfoIn.tileType = ADDR_DISPLAYABLE;
388 else if (flags & (RADEON_SURF_Z_OR_SBUFFER | RADEON_SURF_FMASK))
389 AddrSurfInfoIn.tileType = ADDR_DEPTH_SAMPLE_ORDER;
390 else
391 AddrSurfInfoIn.tileType = ADDR_NON_DISPLAYABLE;
392
393 AddrSurfInfoIn.flags.color = !(flags & RADEON_SURF_Z_OR_SBUFFER);
394 AddrSurfInfoIn.flags.depth = (flags & RADEON_SURF_ZBUFFER) != 0;
395 AddrSurfInfoIn.flags.cube = tex->target == PIPE_TEXTURE_CUBE;
396 AddrSurfInfoIn.flags.fmask = (flags & RADEON_SURF_FMASK) != 0;
397 AddrSurfInfoIn.flags.display = (flags & RADEON_SURF_SCANOUT) != 0;
398 AddrSurfInfoIn.flags.pow2Pad = tex->last_level > 0;
399 AddrSurfInfoIn.flags.tcCompatible = (flags & RADEON_SURF_TC_COMPATIBLE_HTILE) != 0;
400
401 /* Only degrade the tile mode for space if TC-compatible HTILE hasn't been
402 * requested, because TC-compatible HTILE requires 2D tiling.
403 */
404 AddrSurfInfoIn.flags.degrade4Space = !AddrSurfInfoIn.flags.tcCompatible &&
405 !AddrSurfInfoIn.flags.fmask &&
406 tex->nr_samples <= 1 &&
407 (flags & RADEON_SURF_OPTIMIZE_FOR_SPACE);
408 AddrSurfInfoIn.flags.opt4Space = AddrSurfInfoIn.flags.degrade4Space;
409
410 /* DCC notes:
411 * - If we add MSAA support, keep in mind that CB can't decompress 8bpp
412 * with samples >= 4.
413 * - Mipmapped array textures have low performance (discovered by a closed
414 * driver team).
415 */
416 AddrSurfInfoIn.flags.dccCompatible = ws->info.chip_class >= VI &&
417 !(flags & RADEON_SURF_Z_OR_SBUFFER) &&
418 !(flags & RADEON_SURF_DISABLE_DCC) &&
419 !compressed && AddrDccIn.numSamples <= 1 &&
420 ((tex->array_size == 1 && tex->depth0 == 1) ||
421 tex->last_level == 0);
422
423 AddrSurfInfoIn.flags.noStencil = (flags & RADEON_SURF_SBUFFER) == 0;
424 AddrSurfInfoIn.flags.compressZ = AddrSurfInfoIn.flags.depth;
425
426 /* noStencil = 0 can result in a depth part that is incompatible with
427 * mipmapped texturing. So set noStencil = 1 when mipmaps are requested (in
428 * this case, we may end up setting stencil_adjusted).
429 *
430 * TODO: update addrlib to a newer version, remove this, and
431 * use flags.matchStencilTileCfg = 1 as an alternative fix.
432 */
433 if (tex->last_level > 0)
434 AddrSurfInfoIn.flags.noStencil = 1;
435
436 /* Set preferred macrotile parameters. This is usually required
437 * for shared resources. This is for 2D tiling only. */
438 if (AddrSurfInfoIn.tileMode >= ADDR_TM_2D_TILED_THIN1 &&
439 surf->bankw && surf->bankh && surf->mtilea && surf->tile_split) {
440 assert(!(flags & RADEON_SURF_FMASK));
441
442 /* If any of these parameters are incorrect, the calculation
443 * will fail. */
444 AddrTileInfoIn.banks = surf->num_banks;
445 AddrTileInfoIn.bankWidth = surf->bankw;
446 AddrTileInfoIn.bankHeight = surf->bankh;
447 AddrTileInfoIn.macroAspectRatio = surf->mtilea;
448 AddrTileInfoIn.tileSplitBytes = surf->tile_split;
449 AddrTileInfoIn.pipeConfig = surf->pipe_config + 1; /* +1 compared to GB_TILE_MODE */
450 AddrSurfInfoIn.flags.degrade4Space = 0;
451 AddrSurfInfoIn.pTileInfo = &AddrTileInfoIn;
452
453 /* If AddrSurfInfoIn.pTileInfo is set, Addrlib doesn't set
454 * the tile index, because we are expected to know it if
455 * we know the other parameters.
456 *
457 * This is something that can easily be fixed in Addrlib.
458 * For now, just figure it out here.
459 * Note that only 2D_TILE_THIN1 is handled here.
460 */
461 assert(!(flags & RADEON_SURF_Z_OR_SBUFFER));
462 assert(AddrSurfInfoIn.tileMode == ADDR_TM_2D_TILED_THIN1);
463
464 if (ws->info.chip_class == SI) {
465 if (AddrSurfInfoIn.tileType == ADDR_DISPLAYABLE) {
466 if (bpe == 2)
467 AddrSurfInfoIn.tileIndex = 11; /* 16bpp */
468 else
469 AddrSurfInfoIn.tileIndex = 12; /* 32bpp */
470 } else {
471 if (bpe == 1)
472 AddrSurfInfoIn.tileIndex = 14; /* 8bpp */
473 else if (bpe == 2)
474 AddrSurfInfoIn.tileIndex = 15; /* 16bpp */
475 else if (bpe == 4)
476 AddrSurfInfoIn.tileIndex = 16; /* 32bpp */
477 else
478 AddrSurfInfoIn.tileIndex = 17; /* 64bpp (and 128bpp) */
479 }
480 } else {
481 /* CIK - VI */
482 if (AddrSurfInfoIn.tileType == ADDR_DISPLAYABLE)
483 AddrSurfInfoIn.tileIndex = 10; /* 2D displayable */
484 else
485 AddrSurfInfoIn.tileIndex = 14; /* 2D non-displayable */
486
487 /* Addrlib doesn't set this if tileIndex is forced like above. */
488 AddrSurfInfoOut.macroModeIndex = cik_get_macro_tile_index(surf);
489 }
490 }
491
492 surf->num_dcc_levels = 0;
493 surf->surf_size = 0;
494 surf->dcc_size = 0;
495 surf->dcc_alignment = 1;
496 surf->htile_size = 0;
497 surf->htile_alignment = 1;
498
499 /* Calculate texture layout information. */
500 for (level = 0; level <= tex->last_level; level++) {
501 r = compute_level(ws, tex, surf, false, level, compressed,
502 &AddrSurfInfoIn, &AddrSurfInfoOut,
503 &AddrDccIn, &AddrDccOut, &AddrHtileIn, &AddrHtileOut);
504 if (r)
505 return r;
506
507 if (level == 0) {
508 surf->surf_alignment = AddrSurfInfoOut.baseAlign;
509 surf->pipe_config = AddrSurfInfoOut.pTileInfo->pipeConfig - 1;
510 set_micro_tile_mode(surf, &ws->info);
511
512 /* For 2D modes only. */
513 if (AddrSurfInfoOut.tileMode >= ADDR_TM_2D_TILED_THIN1) {
514 surf->bankw = AddrSurfInfoOut.pTileInfo->bankWidth;
515 surf->bankh = AddrSurfInfoOut.pTileInfo->bankHeight;
516 surf->mtilea = AddrSurfInfoOut.pTileInfo->macroAspectRatio;
517 surf->tile_split = AddrSurfInfoOut.pTileInfo->tileSplitBytes;
518 surf->num_banks = AddrSurfInfoOut.pTileInfo->banks;
519 surf->macro_tile_index = AddrSurfInfoOut.macroModeIndex;
520 } else {
521 surf->macro_tile_index = 0;
522 }
523 }
524 }
525
526 /* Calculate texture layout information for stencil. */
527 if (flags & RADEON_SURF_SBUFFER) {
528 AddrSurfInfoIn.bpp = 8;
529 AddrSurfInfoIn.flags.depth = 0;
530 AddrSurfInfoIn.flags.stencil = 1;
531 AddrSurfInfoIn.flags.tcCompatible = 0;
532 /* This will be ignored if AddrSurfInfoIn.pTileInfo is NULL. */
533 AddrTileInfoIn.tileSplitBytes = surf->stencil_tile_split;
534
535 for (level = 0; level <= tex->last_level; level++) {
536 r = compute_level(ws, tex, surf, true, level, compressed,
537 &AddrSurfInfoIn, &AddrSurfInfoOut, &AddrDccIn, &AddrDccOut,
538 NULL, NULL);
539 if (r)
540 return r;
541
542 /* DB uses the depth pitch for both stencil and depth. */
543 if (surf->stencil_level[level].nblk_x != surf->level[level].nblk_x)
544 surf->stencil_adjusted = true;
545
546 if (level == 0) {
547 /* For 2D modes only. */
548 if (AddrSurfInfoOut.tileMode >= ADDR_TM_2D_TILED_THIN1) {
549 surf->stencil_tile_split =
550 AddrSurfInfoOut.pTileInfo->tileSplitBytes;
551 }
552 }
553 }
554 }
555
556 /* Recalculate the whole DCC miptree size including disabled levels.
557 * This is what addrlib does, but calling addrlib would be a lot more
558 * complicated.
559 */
560 if (surf->dcc_size && tex->last_level > 0) {
561 surf->dcc_size = align64(surf->surf_size >> 8,
562 ws->info.pipe_interleave_bytes *
563 ws->info.num_tile_pipes);
564 }
565
566 /* Make sure HTILE covers the whole miptree, because the shader reads
567 * TC-compatible HTILE even for levels where it's disabled by DB.
568 */
569 if (surf->htile_size && tex->last_level)
570 surf->htile_size *= 2;
571
572 surf->is_linear = surf->level[0].mode == RADEON_SURF_MODE_LINEAR_ALIGNED;
573 return 0;
574 }
575
amdgpu_surface_init_functions(struct amdgpu_winsys * ws)576 void amdgpu_surface_init_functions(struct amdgpu_winsys *ws)
577 {
578 ws->base.surface_init = amdgpu_surface_init;
579 }
580