1/*
2 * Copyright 2008 Ben Skeggs
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 "nv50/nv50_context.h"
24#include "nv50/nv50_resource.h"
25#include "nv50/g80_texture.xml.h"
26#include "nv50/g80_defs.xml.h"
27
28#include "util/u_format.h"
29
30static inline uint32_t
31nv50_tic_swizzle(const struct nv50_format *fmt, unsigned swz, bool tex_int)
32{
33   switch (swz) {
34   case PIPE_SWIZZLE_X  : return fmt->tic.src_x;
35   case PIPE_SWIZZLE_Y: return fmt->tic.src_y;
36   case PIPE_SWIZZLE_Z : return fmt->tic.src_z;
37   case PIPE_SWIZZLE_W: return fmt->tic.src_w;
38   case PIPE_SWIZZLE_1:
39      return tex_int ? G80_TIC_SOURCE_ONE_INT : G80_TIC_SOURCE_ONE_FLOAT;
40   case PIPE_SWIZZLE_0:
41   default:
42      return G80_TIC_SOURCE_ZERO;
43   }
44}
45
46struct pipe_sampler_view *
47nv50_create_sampler_view(struct pipe_context *pipe,
48                         struct pipe_resource *res,
49                         const struct pipe_sampler_view *templ)
50{
51   uint32_t flags = 0;
52
53   if (templ->target == PIPE_TEXTURE_RECT || templ->target == PIPE_BUFFER)
54      flags |= NV50_TEXVIEW_SCALED_COORDS;
55
56   return nv50_create_texture_view(pipe, res, templ, flags, templ->target);
57}
58
59struct pipe_sampler_view *
60nv50_create_texture_view(struct pipe_context *pipe,
61                         struct pipe_resource *texture,
62                         const struct pipe_sampler_view *templ,
63                         uint32_t flags,
64                         enum pipe_texture_target target)
65{
66   const uint32_t class_3d = nouveau_context(pipe)->screen->class_3d;
67   const struct util_format_description *desc;
68   const struct nv50_format *fmt;
69   uint64_t addr;
70   uint32_t *tic;
71   uint32_t swz[4];
72   uint32_t depth;
73   struct nv50_tic_entry *view;
74   struct nv50_miptree *mt = nv50_miptree(texture);
75   bool tex_int;
76
77   view = MALLOC_STRUCT(nv50_tic_entry);
78   if (!view)
79      return NULL;
80
81   view->pipe = *templ;
82   view->pipe.reference.count = 1;
83   view->pipe.texture = NULL;
84   view->pipe.context = pipe;
85
86   view->id = -1;
87
88   pipe_resource_reference(&view->pipe.texture, texture);
89
90   tic = &view->tic[0];
91
92   desc = util_format_description(view->pipe.format);
93
94   /* TIC[0] */
95
96   fmt = &nv50_format_table[view->pipe.format];
97
98   tex_int = util_format_is_pure_integer(view->pipe.format);
99
100   swz[0] = nv50_tic_swizzle(fmt, view->pipe.swizzle_r, tex_int);
101   swz[1] = nv50_tic_swizzle(fmt, view->pipe.swizzle_g, tex_int);
102   swz[2] = nv50_tic_swizzle(fmt, view->pipe.swizzle_b, tex_int);
103   swz[3] = nv50_tic_swizzle(fmt, view->pipe.swizzle_a, tex_int);
104   tic[0] = (fmt->tic.format << G80_TIC_0_COMPONENTS_SIZES__SHIFT) |
105            (fmt->tic.type_r << G80_TIC_0_R_DATA_TYPE__SHIFT) |
106            (fmt->tic.type_g << G80_TIC_0_G_DATA_TYPE__SHIFT) |
107            (fmt->tic.type_b << G80_TIC_0_B_DATA_TYPE__SHIFT) |
108            (fmt->tic.type_a << G80_TIC_0_A_DATA_TYPE__SHIFT) |
109            (swz[0] << G80_TIC_0_X_SOURCE__SHIFT) |
110            (swz[1] << G80_TIC_0_Y_SOURCE__SHIFT) |
111            (swz[2] << G80_TIC_0_Z_SOURCE__SHIFT) |
112            (swz[3] << G80_TIC_0_W_SOURCE__SHIFT);
113
114   addr = mt->base.address;
115
116   depth = MAX2(mt->base.base.array_size, mt->base.base.depth0);
117
118   if (mt->base.base.array_size > 1) {
119      /* there doesn't seem to be a base layer field in TIC */
120      addr += view->pipe.u.tex.first_layer * mt->layer_stride;
121      depth = view->pipe.u.tex.last_layer - view->pipe.u.tex.first_layer + 1;
122   }
123
124   tic[2] = 0x10001000 | G80_TIC_2_BORDER_SOURCE_COLOR;
125
126   if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB)
127      tic[2] |= G80_TIC_2_SRGB_CONVERSION;
128
129   if (!(flags & NV50_TEXVIEW_SCALED_COORDS))
130      tic[2] |= G80_TIC_2_NORMALIZED_COORDS;
131
132   if (unlikely(!nouveau_bo_memtype(nv04_resource(texture)->bo))) {
133      if (target == PIPE_BUFFER) {
134         addr += view->pipe.u.buf.offset;
135         tic[2] |= G80_TIC_2_LAYOUT_PITCH | G80_TIC_2_TEXTURE_TYPE_ONE_D_BUFFER;
136         tic[3] = 0;
137         tic[4] = /* width */
138            view->pipe.u.buf.size / (desc->block.bits / 8);
139         tic[5] = 0;
140      } else {
141         tic[2] |= G80_TIC_2_LAYOUT_PITCH | G80_TIC_2_TEXTURE_TYPE_TWO_D_NO_MIPMAP;
142         tic[3] = mt->level[0].pitch;
143         tic[4] = mt->base.base.width0;
144         tic[5] = (1 << 16) | (mt->base.base.height0);
145      }
146      tic[6] =
147      tic[7] = 0;
148      tic[1] = addr;
149      tic[2] |= addr >> 32;
150      return &view->pipe;
151   }
152
153   tic[1] = addr;
154   tic[2] |= (addr >> 32) & 0xff;
155
156   tic[2] |=
157      ((mt->level[0].tile_mode & 0x0f0) << (22 - 4)) |
158      ((mt->level[0].tile_mode & 0xf00) << (25 - 8));
159
160   switch (target) {
161   case PIPE_TEXTURE_1D:
162      tic[2] |= G80_TIC_2_TEXTURE_TYPE_ONE_D;
163      break;
164   case PIPE_TEXTURE_2D:
165      tic[2] |= G80_TIC_2_TEXTURE_TYPE_TWO_D;
166      break;
167   case PIPE_TEXTURE_RECT:
168      tic[2] |= G80_TIC_2_TEXTURE_TYPE_TWO_D_NO_MIPMAP;
169      break;
170   case PIPE_TEXTURE_3D:
171      tic[2] |= G80_TIC_2_TEXTURE_TYPE_THREE_D;
172      break;
173   case PIPE_TEXTURE_CUBE:
174      depth /= 6;
175      tic[2] |= G80_TIC_2_TEXTURE_TYPE_CUBEMAP;
176      break;
177   case PIPE_TEXTURE_1D_ARRAY:
178      tic[2] |= G80_TIC_2_TEXTURE_TYPE_ONE_D_ARRAY;
179      break;
180   case PIPE_TEXTURE_2D_ARRAY:
181      tic[2] |= G80_TIC_2_TEXTURE_TYPE_TWO_D_ARRAY;
182      break;
183   case PIPE_TEXTURE_CUBE_ARRAY:
184      depth /= 6;
185      tic[2] |= G80_TIC_2_TEXTURE_TYPE_CUBE_ARRAY;
186      break;
187   case PIPE_BUFFER:
188      assert(0); /* should be linear and handled above ! */
189      tic[2] |= G80_TIC_2_TEXTURE_TYPE_ONE_D_BUFFER | G80_TIC_2_LAYOUT_PITCH;
190      break;
191   default:
192      unreachable("unexpected/invalid texture target");
193   }
194
195   tic[3] = (flags & NV50_TEXVIEW_FILTER_MSAA8) ? 0x20000000 : 0x00300000;
196
197   tic[4] = (1 << 31) | (mt->base.base.width0 << mt->ms_x);
198
199   tic[5] = (mt->base.base.height0 << mt->ms_y) & 0xffff;
200   tic[5] |= depth << 16;
201   if (class_3d > NV50_3D_CLASS)
202      tic[5] |= mt->base.base.last_level << G80_TIC_5_MAP_MIP_LEVEL__SHIFT;
203   else
204      tic[5] |= view->pipe.u.tex.last_level << G80_TIC_5_MAP_MIP_LEVEL__SHIFT;
205
206   tic[6] = (mt->ms_x > 1) ? 0x88000000 : 0x03000000; /* sampling points */
207
208   if (class_3d > NV50_3D_CLASS)
209      tic[7] = (view->pipe.u.tex.last_level << 4) | view->pipe.u.tex.first_level;
210   else
211      tic[7] = 0;
212
213   if (unlikely(!(tic[2] & G80_TIC_2_NORMALIZED_COORDS)))
214      if (mt->base.base.last_level)
215         tic[5] &= ~G80_TIC_5_MAP_MIP_LEVEL__MASK;
216
217   return &view->pipe;
218}
219
220static void
221nv50_update_tic(struct nv50_context *nv50, struct nv50_tic_entry *tic,
222                struct nv04_resource *res)
223{
224   uint64_t address = res->address;
225   if (res->base.target != PIPE_BUFFER)
226      return;
227   address += tic->pipe.u.buf.offset;
228   if (tic->tic[1] == (uint32_t)address &&
229       (tic->tic[2] & 0xff) == address >> 32)
230      return;
231
232   nv50_screen_tic_unlock(nv50->screen, tic);
233   tic->id = -1;
234   tic->tic[1] = address;
235   tic->tic[2] &= 0xffffff00;
236   tic->tic[2] |= address >> 32;
237}
238
239static bool
240nv50_validate_tic(struct nv50_context *nv50, int s)
241{
242   struct nouveau_pushbuf *push = nv50->base.pushbuf;
243   struct nouveau_bo *txc = nv50->screen->txc;
244   unsigned i;
245   bool need_flush = false;
246
247   assert(nv50->num_textures[s] <= PIPE_MAX_SAMPLERS);
248   for (i = 0; i < nv50->num_textures[s]; ++i) {
249      struct nv50_tic_entry *tic = nv50_tic_entry(nv50->textures[s][i]);
250      struct nv04_resource *res;
251
252      if (!tic) {
253         BEGIN_NV04(push, NV50_3D(BIND_TIC(s)), 1);
254         PUSH_DATA (push, (i << 1) | 0);
255         continue;
256      }
257      res = &nv50_miptree(tic->pipe.texture)->base;
258      nv50_update_tic(nv50, tic, res);
259
260      if (tic->id < 0) {
261         tic->id = nv50_screen_tic_alloc(nv50->screen, tic);
262
263         BEGIN_NV04(push, NV50_2D(DST_FORMAT), 2);
264         PUSH_DATA (push, G80_SURFACE_FORMAT_R8_UNORM);
265         PUSH_DATA (push, 1);
266         BEGIN_NV04(push, NV50_2D(DST_PITCH), 5);
267         PUSH_DATA (push, 262144);
268         PUSH_DATA (push, 65536);
269         PUSH_DATA (push, 1);
270         PUSH_DATAh(push, txc->offset);
271         PUSH_DATA (push, txc->offset);
272         BEGIN_NV04(push, NV50_2D(SIFC_BITMAP_ENABLE), 2);
273         PUSH_DATA (push, 0);
274         PUSH_DATA (push, G80_SURFACE_FORMAT_R8_UNORM);
275         BEGIN_NV04(push, NV50_2D(SIFC_WIDTH), 10);
276         PUSH_DATA (push, 32);
277         PUSH_DATA (push, 1);
278         PUSH_DATA (push, 0);
279         PUSH_DATA (push, 1);
280         PUSH_DATA (push, 0);
281         PUSH_DATA (push, 1);
282         PUSH_DATA (push, 0);
283         PUSH_DATA (push, tic->id * 32);
284         PUSH_DATA (push, 0);
285         PUSH_DATA (push, 0);
286         BEGIN_NI04(push, NV50_2D(SIFC_DATA), 8);
287         PUSH_DATAp(push, &tic->tic[0], 8);
288
289         need_flush = true;
290      } else
291      if (res->status & NOUVEAU_BUFFER_STATUS_GPU_WRITING) {
292         BEGIN_NV04(push, NV50_3D(TEX_CACHE_CTL), 1);
293         PUSH_DATA (push, 0x20);
294      }
295
296      nv50->screen->tic.lock[tic->id / 32] |= 1 << (tic->id % 32);
297
298      res->status &= ~NOUVEAU_BUFFER_STATUS_GPU_WRITING;
299      res->status |= NOUVEAU_BUFFER_STATUS_GPU_READING;
300
301      BCTX_REFN(nv50->bufctx_3d, 3D_TEXTURES, res, RD);
302
303      BEGIN_NV04(push, NV50_3D(BIND_TIC(s)), 1);
304      PUSH_DATA (push, (tic->id << 9) | (i << 1) | 1);
305   }
306   for (; i < nv50->state.num_textures[s]; ++i) {
307      BEGIN_NV04(push, NV50_3D(BIND_TIC(s)), 1);
308      PUSH_DATA (push, (i << 1) | 0);
309   }
310   if (nv50->num_textures[s]) {
311      BEGIN_NV04(push, NV50_3D(CB_ADDR), 1);
312      PUSH_DATA (push, ((NV50_CB_AUX_TEX_MS_OFFSET + 16 * s * 2 * 4) << (8 - 2)) | NV50_CB_AUX);
313      BEGIN_NI04(push, NV50_3D(CB_DATA(0)), nv50->num_textures[s] * 2);
314      for (i = 0; i < nv50->num_textures[s]; i++) {
315         struct nv50_tic_entry *tic = nv50_tic_entry(nv50->textures[s][i]);
316         struct nv50_miptree *res;
317
318         if (!tic) {
319            PUSH_DATA (push, 0);
320            PUSH_DATA (push, 0);
321            continue;
322         }
323         res = nv50_miptree(tic->pipe.texture);
324         PUSH_DATA (push, res->ms_x);
325         PUSH_DATA (push, res->ms_y);
326      }
327   }
328   nv50->state.num_textures[s] = nv50->num_textures[s];
329
330   return need_flush;
331}
332
333void nv50_validate_textures(struct nv50_context *nv50)
334{
335   bool need_flush;
336
337   need_flush  = nv50_validate_tic(nv50, 0);
338   need_flush |= nv50_validate_tic(nv50, 1);
339   need_flush |= nv50_validate_tic(nv50, 2);
340
341   if (need_flush) {
342      BEGIN_NV04(nv50->base.pushbuf, NV50_3D(TIC_FLUSH), 1);
343      PUSH_DATA (nv50->base.pushbuf, 0);
344   }
345}
346
347static bool
348nv50_validate_tsc(struct nv50_context *nv50, int s)
349{
350   struct nouveau_pushbuf *push = nv50->base.pushbuf;
351   unsigned i;
352   bool need_flush = false;
353
354   assert(nv50->num_samplers[s] <= PIPE_MAX_SAMPLERS);
355   for (i = 0; i < nv50->num_samplers[s]; ++i) {
356      struct nv50_tsc_entry *tsc = nv50_tsc_entry(nv50->samplers[s][i]);
357
358      if (!tsc) {
359         BEGIN_NV04(push, NV50_3D(BIND_TSC(s)), 1);
360         PUSH_DATA (push, (i << 4) | 0);
361         continue;
362      }
363      nv50->seamless_cube_map = tsc->seamless_cube_map;
364      if (tsc->id < 0) {
365         tsc->id = nv50_screen_tsc_alloc(nv50->screen, tsc);
366
367         nv50_sifc_linear_u8(&nv50->base, nv50->screen->txc,
368                             65536 + tsc->id * 32,
369                             NOUVEAU_BO_VRAM, 32, tsc->tsc);
370         need_flush = true;
371      }
372      nv50->screen->tsc.lock[tsc->id / 32] |= 1 << (tsc->id % 32);
373
374      BEGIN_NV04(push, NV50_3D(BIND_TSC(s)), 1);
375      PUSH_DATA (push, (tsc->id << 12) | (i << 4) | 1);
376   }
377   for (; i < nv50->state.num_samplers[s]; ++i) {
378      BEGIN_NV04(push, NV50_3D(BIND_TSC(s)), 1);
379      PUSH_DATA (push, (i << 4) | 0);
380   }
381   nv50->state.num_samplers[s] = nv50->num_samplers[s];
382
383   // TXF, in unlinked tsc mode, will always use sampler 0. So we have to
384   // ensure that it remains bound. Its contents don't matter, all samplers we
385   // ever create have the SRGB_CONVERSION bit set, so as long as the first
386   // entry is initialized, we're good to go. This is the only bit that has
387   // any effect on what TXF does.
388   if (!nv50->samplers[s][0]) {
389      BEGIN_NV04(push, NV50_3D(BIND_TSC(s)), 1);
390      PUSH_DATA (push, 1);
391   }
392
393   return need_flush;
394}
395
396void nv50_validate_samplers(struct nv50_context *nv50)
397{
398   bool need_flush;
399
400   need_flush  = nv50_validate_tsc(nv50, 0);
401   need_flush |= nv50_validate_tsc(nv50, 1);
402   need_flush |= nv50_validate_tsc(nv50, 2);
403
404   if (need_flush) {
405      BEGIN_NV04(nv50->base.pushbuf, NV50_3D(TSC_FLUSH), 1);
406      PUSH_DATA (nv50->base.pushbuf, 0);
407   }
408}
409
410/* There can be up to 4 different MS levels (1, 2, 4, 8). To simplify the
411 * shader logic, allow each one to take up 8 offsets.
412 */
413#define COMBINE(x, y) x, y
414#define DUMMY 0, 0
415static const uint32_t msaa_sample_xy_offsets[] = {
416   /* MS1 */
417   COMBINE(0, 0),
418   DUMMY,
419   DUMMY,
420   DUMMY,
421   DUMMY,
422   DUMMY,
423   DUMMY,
424   DUMMY,
425
426   /* MS2 */
427   COMBINE(0, 0),
428   COMBINE(1, 0),
429   DUMMY,
430   DUMMY,
431   DUMMY,
432   DUMMY,
433   DUMMY,
434   DUMMY,
435
436   /* MS4 */
437   COMBINE(0, 0),
438   COMBINE(1, 0),
439   COMBINE(0, 1),
440   COMBINE(1, 1),
441   DUMMY,
442   DUMMY,
443   DUMMY,
444   DUMMY,
445
446   /* MS8 */
447   COMBINE(0, 0),
448   COMBINE(1, 0),
449   COMBINE(0, 1),
450   COMBINE(1, 1),
451   COMBINE(2, 0),
452   COMBINE(3, 0),
453   COMBINE(2, 1),
454   COMBINE(3, 1),
455};
456
457void nv50_upload_ms_info(struct nouveau_pushbuf *push)
458{
459   BEGIN_NV04(push, NV50_3D(CB_ADDR), 1);
460   PUSH_DATA (push, (NV50_CB_AUX_MS_OFFSET << (8 - 2)) | NV50_CB_AUX);
461   BEGIN_NI04(push, NV50_3D(CB_DATA(0)), ARRAY_SIZE(msaa_sample_xy_offsets));
462   PUSH_DATAp(push, msaa_sample_xy_offsets, ARRAY_SIZE(msaa_sample_xy_offsets));
463}
464
465void nv50_upload_tsc0(struct nv50_context *nv50)
466{
467   struct nouveau_pushbuf *push = nv50->base.pushbuf;
468   u32 data[8] = { G80_TSC_0_SRGB_CONVERSION };
469   nv50_sifc_linear_u8(&nv50->base, nv50->screen->txc,
470                       65536 /* + tsc->id * 32 */,
471                       NOUVEAU_BO_VRAM, 32, data);
472   BEGIN_NV04(push, NV50_3D(TSC_FLUSH), 1);
473   PUSH_DATA (push, 0);
474}
475