• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright 2009 Marek Olšák <maraeo@gmail.com>
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  * on the rights to use, copy, modify, merge, publish, distribute, sub
8  * license, and/or sell copies of the Software, and to permit persons to whom
9  * the Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice (including the next
12  * paragraph) shall be included in all copies or substantial portions of the
13  * Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
22 
23 #include "r300_context.h"
24 #include "r300_emit.h"
25 #include "r300_texture.h"
26 #include "r300_reg.h"
27 
28 #include "util/format/u_format.h"
29 #include "util/half_float.h"
30 #include "util/u_pack_color.h"
31 #include "util/u_surface.h"
32 
33 enum r300_blitter_op /* bitmask */
34 {
35     R300_STOP_QUERY         = 1,
36     R300_SAVE_TEXTURES      = 2,
37     R300_SAVE_FRAMEBUFFER   = 4,
38     R300_IGNORE_RENDER_COND = 8,
39 
40     R300_CLEAR         = R300_STOP_QUERY,
41 
42     R300_CLEAR_SURFACE = R300_STOP_QUERY | R300_SAVE_FRAMEBUFFER,
43 
44     R300_COPY          = R300_STOP_QUERY | R300_SAVE_FRAMEBUFFER |
45                          R300_SAVE_TEXTURES | R300_IGNORE_RENDER_COND,
46 
47     R300_BLIT          = R300_STOP_QUERY | R300_SAVE_FRAMEBUFFER |
48                          R300_SAVE_TEXTURES,
49 
50     R300_DECOMPRESS    = R300_STOP_QUERY | R300_IGNORE_RENDER_COND,
51 };
52 
r300_blitter_begin(struct r300_context * r300,enum r300_blitter_op op)53 static void r300_blitter_begin(struct r300_context* r300, enum r300_blitter_op op)
54 {
55     if ((op & R300_STOP_QUERY) && r300->query_current) {
56         r300->blitter_saved_query = r300->query_current;
57         r300_stop_query(r300);
58     }
59 
60     /* Yeah we have to save all those states to ensure the blitter operation
61      * is really transparent. The states will be restored by the blitter once
62      * copying is done. */
63     util_blitter_save_blend(r300->blitter, r300->blend_state.state);
64     util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state.state);
65     util_blitter_save_stencil_ref(r300->blitter, &(r300->stencil_ref));
66     util_blitter_save_rasterizer(r300->blitter, r300->rs_state.state);
67     util_blitter_save_fragment_shader(r300->blitter, r300->fs.state);
68     util_blitter_save_vertex_shader(r300->blitter, r300->vs_state.state);
69     util_blitter_save_viewport(r300->blitter, &r300->viewport);
70     util_blitter_save_scissor(r300->blitter, r300->scissor_state.state);
71     util_blitter_save_sample_mask(r300->blitter, *(unsigned*)r300->sample_mask.state);
72     util_blitter_save_vertex_buffer_slot(r300->blitter, r300->vertex_buffer);
73     util_blitter_save_vertex_elements(r300->blitter, r300->velems);
74 
75     if (op & R300_SAVE_FRAMEBUFFER) {
76         util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state);
77     }
78 
79     if (op & R300_SAVE_TEXTURES) {
80         struct r300_textures_state* state =
81             (struct r300_textures_state*)r300->textures_state.state;
82 
83         util_blitter_save_fragment_sampler_states(
84             r300->blitter, state->sampler_state_count,
85             (void**)state->sampler_states);
86 
87         util_blitter_save_fragment_sampler_views(
88             r300->blitter, state->sampler_view_count,
89             (struct pipe_sampler_view**)state->sampler_views);
90     }
91 
92     if (op & R300_IGNORE_RENDER_COND) {
93         /* Save the flag. */
94         r300->blitter_saved_skip_rendering = r300->skip_rendering+1;
95         r300->skip_rendering = FALSE;
96     } else {
97         r300->blitter_saved_skip_rendering = 0;
98     }
99 }
100 
r300_blitter_end(struct r300_context * r300)101 static void r300_blitter_end(struct r300_context *r300)
102 {
103     if (r300->blitter_saved_query) {
104         r300_resume_query(r300, r300->blitter_saved_query);
105         r300->blitter_saved_query = NULL;
106     }
107 
108     if (r300->blitter_saved_skip_rendering) {
109         /* Restore the flag. */
110         r300->skip_rendering = r300->blitter_saved_skip_rendering-1;
111     }
112 }
113 
r300_depth_clear_cb_value(enum pipe_format format,const float * rgba)114 static uint32_t r300_depth_clear_cb_value(enum pipe_format format,
115                                           const float* rgba)
116 {
117     union util_color uc;
118     util_pack_color(rgba, format, &uc);
119 
120     if (util_format_get_blocksizebits(format) == 32)
121         return uc.ui[0];
122     else
123         return uc.us | (uc.us << 16);
124 }
125 
r300_cbzb_clear_allowed(struct r300_context * r300,unsigned clear_buffers)126 static boolean r300_cbzb_clear_allowed(struct r300_context *r300,
127                                        unsigned clear_buffers)
128 {
129     struct pipe_framebuffer_state *fb =
130         (struct pipe_framebuffer_state*)r300->fb_state.state;
131 
132     /* Only color clear allowed, and only one colorbuffer. */
133     if ((clear_buffers & ~PIPE_CLEAR_COLOR) != 0 || fb->nr_cbufs != 1 || !fb->cbufs[0])
134         return FALSE;
135 
136     return r300_surface(fb->cbufs[0])->cbzb_allowed;
137 }
138 
r300_fast_zclear_allowed(struct r300_context * r300,unsigned clear_buffers)139 static boolean r300_fast_zclear_allowed(struct r300_context *r300,
140                                         unsigned clear_buffers)
141 {
142     struct pipe_framebuffer_state *fb =
143         (struct pipe_framebuffer_state*)r300->fb_state.state;
144 
145     return r300_resource(fb->zsbuf->texture)->tex.zmask_dwords[fb->zsbuf->u.tex.level] != 0;
146 }
147 
r300_hiz_clear_allowed(struct r300_context * r300)148 static boolean r300_hiz_clear_allowed(struct r300_context *r300)
149 {
150     struct pipe_framebuffer_state *fb =
151         (struct pipe_framebuffer_state*)r300->fb_state.state;
152 
153     return r300_resource(fb->zsbuf->texture)->tex.hiz_dwords[fb->zsbuf->u.tex.level] != 0;
154 }
155 
r300_depth_clear_value(enum pipe_format format,double depth,unsigned stencil)156 static uint32_t r300_depth_clear_value(enum pipe_format format,
157                                        double depth, unsigned stencil)
158 {
159     switch (format) {
160         case PIPE_FORMAT_Z16_UNORM:
161         case PIPE_FORMAT_X8Z24_UNORM:
162             return util_pack_z(format, depth);
163 
164         case PIPE_FORMAT_S8_UINT_Z24_UNORM:
165             return util_pack_z_stencil(format, depth, stencil);
166 
167         default:
168             assert(0);
169             return 0;
170     }
171 }
172 
r300_hiz_clear_value(double depth)173 static uint32_t r300_hiz_clear_value(double depth)
174 {
175     uint32_t r = (uint32_t)(CLAMP(depth, 0, 1) * 255.5);
176     assert(r <= 255);
177     return r | (r << 8) | (r << 16) | (r << 24);
178 }
179 
r300_set_clear_color(struct r300_context * r300,const union pipe_color_union * color)180 static void r300_set_clear_color(struct r300_context *r300,
181                                  const union pipe_color_union *color)
182 {
183     struct pipe_framebuffer_state *fb =
184         (struct pipe_framebuffer_state*)r300->fb_state.state;
185     union util_color uc;
186 
187     memset(&uc, 0, sizeof(uc));
188     util_pack_color(color->f, fb->cbufs[0]->format, &uc);
189 
190     if (fb->cbufs[0]->format == PIPE_FORMAT_R16G16B16A16_FLOAT ||
191         fb->cbufs[0]->format == PIPE_FORMAT_R16G16B16X16_FLOAT) {
192         /* (0,1,2,3) maps to (B,G,R,A) */
193         r300->color_clear_value_gb = uc.h[0] | ((uint32_t)uc.h[1] << 16);
194         r300->color_clear_value_ar = uc.h[2] | ((uint32_t)uc.h[3] << 16);
195     } else {
196         r300->color_clear_value = uc.ui[0];
197     }
198 }
199 
200 DEBUG_GET_ONCE_BOOL_OPTION(hyperz, "RADEON_HYPERZ", FALSE)
201 
202 /* Clear currently bound buffers. */
r300_clear(struct pipe_context * pipe,unsigned buffers,const struct pipe_scissor_state * scissor_state,const union pipe_color_union * color,double depth,unsigned stencil)203 static void r300_clear(struct pipe_context* pipe,
204                        unsigned buffers,
205                        const struct pipe_scissor_state *scissor_state,
206                        const union pipe_color_union *color,
207                        double depth,
208                        unsigned stencil)
209 {
210     /* My notes about Zbuffer compression:
211      *
212      * 1) The zbuffer must be micro-tiled and whole microtiles must be
213      *    written if compression is enabled. If microtiling is disabled,
214      *    it locks up.
215      *
216      * 2) There is ZMASK RAM which contains a compressed zbuffer.
217      *    Each dword of the Z Mask contains compression information
218      *    for 16 4x4 pixel tiles, that is 2 bits for each tile.
219      *    On chips with 2 Z pipes, every other dword maps to a different
220      *    pipe. On newer chipsets, there is a new compression mode
221      *    with 8x8 pixel tiles per 2 bits.
222      *
223      * 3) The FASTFILL bit has nothing to do with filling. It only tells hw
224      *    it should look in the ZMASK RAM first before fetching from a real
225      *    zbuffer.
226      *
227      * 4) If a pixel is in a cleared state, ZB_DEPTHCLEARVALUE is returned
228      *    during zbuffer reads instead of the value that is actually stored
229      *    in the zbuffer memory. A pixel is in a cleared state when its ZMASK
230      *    is equal to 0. Therefore, if you clear ZMASK with zeros, you may
231      *    leave the zbuffer memory uninitialized, but then you must enable
232      *    compression, so that the ZMASK RAM is actually used.
233      *
234      * 5) Each 4x4 (or 8x8) tile is automatically decompressed and recompressed
235      *    during zbuffer updates. A special decompressing operation should be
236      *    used to fully decompress a zbuffer, which basically just stores all
237      *    compressed tiles in ZMASK to the zbuffer memory.
238      *
239      * 6) For a 16-bit zbuffer, compression causes a hung with one or
240      *    two samples and should not be used.
241      *
242      * 7) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears
243      *    to avoid needless decompression.
244      *
245      * 8) Fastfill must not be used if reading of compressed Z data is disabled
246      *    and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE),
247      *    i.e. it cannot be used to compress the zbuffer.
248      *
249      * 9) ZB_CB_CLEAR does not interact with zbuffer compression in any way.
250      *
251      * - Marek
252      */
253 
254     struct r300_context* r300 = r300_context(pipe);
255     struct pipe_framebuffer_state *fb =
256         (struct pipe_framebuffer_state*)r300->fb_state.state;
257     struct r300_hyperz_state *hyperz =
258         (struct r300_hyperz_state*)r300->hyperz_state.state;
259     uint32_t width = fb->width;
260     uint32_t height = fb->height;
261     uint32_t hyperz_dcv = hyperz->zb_depthclearvalue;
262 
263     /* Use fast Z clear.
264      * The zbuffer must be in micro-tiled mode, otherwise it locks up. */
265     if (buffers & PIPE_CLEAR_DEPTHSTENCIL) {
266         boolean zmask_clear, hiz_clear;
267 
268         /* If both depth and stencil are present, they must be cleared together. */
269         if (fb->zsbuf->texture->format == PIPE_FORMAT_S8_UINT_Z24_UNORM &&
270             (buffers & PIPE_CLEAR_DEPTHSTENCIL) != PIPE_CLEAR_DEPTHSTENCIL) {
271             zmask_clear = FALSE;
272             hiz_clear = FALSE;
273         } else {
274             zmask_clear = r300_fast_zclear_allowed(r300, buffers);
275             hiz_clear = r300_hiz_clear_allowed(r300);
276         }
277 
278         /* If we need Hyper-Z. */
279         if (zmask_clear || hiz_clear) {
280             /* Try to obtain the access to Hyper-Z buffers if we don't have one. */
281             if (!r300->hyperz_enabled &&
282                 (r300->screen->caps.is_r500 || debug_get_option_hyperz())) {
283                 r300->hyperz_enabled =
284                     r300->rws->cs_request_feature(&r300->cs,
285                                                 RADEON_FID_R300_HYPERZ_ACCESS,
286                                                 TRUE);
287                 if (r300->hyperz_enabled) {
288                    /* Need to emit HyperZ buffer regs for the first time. */
289                    r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
290                 }
291             }
292 
293             /* Setup Hyper-Z clears. */
294             if (r300->hyperz_enabled) {
295                 if (zmask_clear) {
296                     hyperz_dcv = hyperz->zb_depthclearvalue =
297                         r300_depth_clear_value(fb->zsbuf->format, depth, stencil);
298 
299                     r300_mark_atom_dirty(r300, &r300->zmask_clear);
300                     r300_mark_atom_dirty(r300, &r300->gpu_flush);
301                     buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
302                 }
303 
304                 if (hiz_clear) {
305                     r300->hiz_clear_value = r300_hiz_clear_value(depth);
306                     r300_mark_atom_dirty(r300, &r300->hiz_clear);
307                     r300_mark_atom_dirty(r300, &r300->gpu_flush);
308                 }
309                 r300->num_z_clears++;
310             }
311         }
312     }
313 
314     /* Use fast color clear for an AA colorbuffer.
315      * The CMASK is shared between all colorbuffers, so we use it
316      * if there is only one colorbuffer bound. */
317     if ((buffers & PIPE_CLEAR_COLOR) && fb->nr_cbufs == 1 && fb->cbufs[0] &&
318         r300_resource(fb->cbufs[0]->texture)->tex.cmask_dwords) {
319         /* Try to obtain the access to the CMASK if we don't have one. */
320         if (!r300->cmask_access) {
321             r300->cmask_access =
322                 r300->rws->cs_request_feature(&r300->cs,
323                                               RADEON_FID_R300_CMASK_ACCESS,
324                                               TRUE);
325         }
326 
327         /* Setup the clear. */
328         if (r300->cmask_access) {
329             /* Pair the resource with the CMASK to avoid other resources
330              * accessing it. */
331             if (!r300->screen->cmask_resource) {
332                 mtx_lock(&r300->screen->cmask_mutex);
333                 /* Double checking (first unlocked, then locked). */
334                 if (!r300->screen->cmask_resource) {
335                     /* Don't reference this, so that the texture can be
336                      * destroyed while set in cmask_resource.
337                      * Then in texture_destroy, we set cmask_resource to NULL. */
338                     r300->screen->cmask_resource = fb->cbufs[0]->texture;
339                 }
340                 mtx_unlock(&r300->screen->cmask_mutex);
341             }
342 
343             if (r300->screen->cmask_resource == fb->cbufs[0]->texture) {
344                 r300_set_clear_color(r300, color);
345                 r300_mark_atom_dirty(r300, &r300->cmask_clear);
346                 r300_mark_atom_dirty(r300, &r300->gpu_flush);
347                 buffers &= ~PIPE_CLEAR_COLOR;
348             }
349         }
350     }
351     /* Enable CBZB clear. */
352     else if (r300_cbzb_clear_allowed(r300, buffers)) {
353         struct r300_surface *surf = r300_surface(fb->cbufs[0]);
354 
355         hyperz->zb_depthclearvalue =
356                 r300_depth_clear_cb_value(surf->base.format, color->f);
357 
358         width = surf->cbzb_width;
359         height = surf->cbzb_height;
360 
361         r300->cbzb_clear = TRUE;
362         r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
363     }
364 
365     /* Clear. */
366     if (buffers) {
367         /* Clear using the blitter. */
368         r300_blitter_begin(r300, R300_CLEAR);
369         util_blitter_clear(r300->blitter, width, height, 1,
370                            buffers, color, depth, stencil,
371                            util_framebuffer_get_num_samples(fb) > 1);
372         r300_blitter_end(r300);
373     } else if (r300->zmask_clear.dirty ||
374                r300->hiz_clear.dirty ||
375                r300->cmask_clear.dirty) {
376         /* Just clear zmask and hiz now, this does not use the standard draw
377          * procedure. */
378         /* Calculate zmask_clear and hiz_clear atom sizes. */
379         unsigned dwords =
380             r300->gpu_flush.size +
381             (r300->zmask_clear.dirty ? r300->zmask_clear.size : 0) +
382             (r300->hiz_clear.dirty ? r300->hiz_clear.size : 0) +
383             (r300->cmask_clear.dirty ? r300->cmask_clear.size : 0) +
384             r300_get_num_cs_end_dwords(r300);
385 
386         /* Reserve CS space. */
387         if (!r300->rws->cs_check_space(&r300->cs, dwords, false)) {
388             r300_flush(&r300->context, PIPE_FLUSH_ASYNC, NULL);
389         }
390 
391         /* Emit clear packets. */
392         r300_emit_gpu_flush(r300, r300->gpu_flush.size, r300->gpu_flush.state);
393         r300->gpu_flush.dirty = FALSE;
394 
395         if (r300->zmask_clear.dirty) {
396             r300_emit_zmask_clear(r300, r300->zmask_clear.size,
397                                   r300->zmask_clear.state);
398             r300->zmask_clear.dirty = FALSE;
399         }
400         if (r300->hiz_clear.dirty) {
401             r300_emit_hiz_clear(r300, r300->hiz_clear.size,
402                                 r300->hiz_clear.state);
403             r300->hiz_clear.dirty = FALSE;
404         }
405         if (r300->cmask_clear.dirty) {
406             r300_emit_cmask_clear(r300, r300->cmask_clear.size,
407                                   r300->cmask_clear.state);
408             r300->cmask_clear.dirty = FALSE;
409         }
410     } else {
411         assert(0);
412     }
413 
414     /* Disable CBZB clear. */
415     if (r300->cbzb_clear) {
416         r300->cbzb_clear = FALSE;
417         hyperz->zb_depthclearvalue = hyperz_dcv;
418         r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
419     }
420 
421     /* Enable fastfill and/or hiz.
422      *
423      * If we cleared zmask/hiz, it's in use now. The Hyper-Z state update
424      * looks if zmask/hiz is in use and programs hardware accordingly. */
425     if (r300->zmask_in_use || r300->hiz_in_use) {
426         r300_mark_atom_dirty(r300, &r300->hyperz_state);
427     }
428 }
429 
430 /* Clear a region of a color surface to a constant value. */
r300_clear_render_target(struct pipe_context * pipe,struct pipe_surface * dst,const union pipe_color_union * color,unsigned dstx,unsigned dsty,unsigned width,unsigned height,bool render_condition_enabled)431 static void r300_clear_render_target(struct pipe_context *pipe,
432                                      struct pipe_surface *dst,
433                                      const union pipe_color_union *color,
434                                      unsigned dstx, unsigned dsty,
435                                      unsigned width, unsigned height,
436                                      bool render_condition_enabled)
437 {
438     struct r300_context *r300 = r300_context(pipe);
439 
440     r300_blitter_begin(r300, R300_CLEAR_SURFACE |
441                        (render_condition_enabled ? 0 : R300_IGNORE_RENDER_COND));
442     util_blitter_clear_render_target(r300->blitter, dst, color,
443                                      dstx, dsty, width, height);
444     r300_blitter_end(r300);
445 }
446 
447 /* Clear a region of a depth stencil surface. */
r300_clear_depth_stencil(struct pipe_context * pipe,struct pipe_surface * dst,unsigned clear_flags,double depth,unsigned stencil,unsigned dstx,unsigned dsty,unsigned width,unsigned height,bool render_condition_enabled)448 static void r300_clear_depth_stencil(struct pipe_context *pipe,
449                                      struct pipe_surface *dst,
450                                      unsigned clear_flags,
451                                      double depth,
452                                      unsigned stencil,
453                                      unsigned dstx, unsigned dsty,
454                                      unsigned width, unsigned height,
455                                      bool render_condition_enabled)
456 {
457     struct r300_context *r300 = r300_context(pipe);
458     struct pipe_framebuffer_state *fb =
459         (struct pipe_framebuffer_state*)r300->fb_state.state;
460 
461     if (r300->zmask_in_use && !r300->locked_zbuffer) {
462         if (fb->zsbuf->texture == dst->texture) {
463             r300_decompress_zmask(r300);
464         }
465     }
466 
467     /* XXX Do not decompress ZMask of the currently-set zbuffer. */
468     r300_blitter_begin(r300, R300_CLEAR_SURFACE |
469                        (render_condition_enabled ? 0 : R300_IGNORE_RENDER_COND));
470     util_blitter_clear_depth_stencil(r300->blitter, dst, clear_flags, depth, stencil,
471                                      dstx, dsty, width, height);
472     r300_blitter_end(r300);
473 }
474 
r300_decompress_zmask(struct r300_context * r300)475 void r300_decompress_zmask(struct r300_context *r300)
476 {
477     struct pipe_framebuffer_state *fb =
478         (struct pipe_framebuffer_state*)r300->fb_state.state;
479 
480     if (!r300->zmask_in_use || r300->locked_zbuffer)
481         return;
482 
483     r300->zmask_decompress = TRUE;
484     r300_mark_atom_dirty(r300, &r300->hyperz_state);
485 
486     r300_blitter_begin(r300, R300_DECOMPRESS);
487     util_blitter_custom_clear_depth(r300->blitter, fb->width, fb->height, 0,
488                                     r300->dsa_decompress_zmask);
489     r300_blitter_end(r300);
490 
491     r300->zmask_decompress = FALSE;
492     r300->zmask_in_use = FALSE;
493     r300_mark_atom_dirty(r300, &r300->hyperz_state);
494 }
495 
r300_decompress_zmask_locked_unsafe(struct r300_context * r300)496 void r300_decompress_zmask_locked_unsafe(struct r300_context *r300)
497 {
498     struct pipe_framebuffer_state fb;
499 
500     memset(&fb, 0, sizeof(fb));
501     fb.width = r300->locked_zbuffer->width;
502     fb.height = r300->locked_zbuffer->height;
503     fb.zsbuf = r300->locked_zbuffer;
504 
505     r300->context.set_framebuffer_state(&r300->context, &fb);
506     r300_decompress_zmask(r300);
507 }
508 
r300_decompress_zmask_locked(struct r300_context * r300)509 void r300_decompress_zmask_locked(struct r300_context *r300)
510 {
511     struct pipe_framebuffer_state saved_fb;
512 
513     memset(&saved_fb, 0, sizeof(saved_fb));
514     util_copy_framebuffer_state(&saved_fb, r300->fb_state.state);
515     r300_decompress_zmask_locked_unsafe(r300);
516     r300->context.set_framebuffer_state(&r300->context, &saved_fb);
517     util_unreference_framebuffer_state(&saved_fb);
518 
519     pipe_surface_reference(&r300->locked_zbuffer, NULL);
520 }
521 
r300_is_blit_supported(enum pipe_format format)522 bool r300_is_blit_supported(enum pipe_format format)
523 {
524     const struct util_format_description *desc =
525         util_format_description(format);
526 
527     return desc->layout == UTIL_FORMAT_LAYOUT_PLAIN ||
528            desc->layout == UTIL_FORMAT_LAYOUT_S3TC ||
529            desc->layout == UTIL_FORMAT_LAYOUT_RGTC;
530 }
531 
532 /* Copy a block of pixels from one surface to another. */
r300_resource_copy_region(struct pipe_context * pipe,struct pipe_resource * dst,unsigned dst_level,unsigned dstx,unsigned dsty,unsigned dstz,struct pipe_resource * src,unsigned src_level,const struct pipe_box * src_box)533 static void r300_resource_copy_region(struct pipe_context *pipe,
534                                       struct pipe_resource *dst,
535                                       unsigned dst_level,
536                                       unsigned dstx, unsigned dsty, unsigned dstz,
537                                       struct pipe_resource *src,
538                                       unsigned src_level,
539                                       const struct pipe_box *src_box)
540 {
541     struct pipe_screen *screen = pipe->screen;
542     struct r300_context *r300 = r300_context(pipe);
543     struct pipe_framebuffer_state *fb =
544         (struct pipe_framebuffer_state*)r300->fb_state.state;
545     unsigned src_width0 = r300_resource(src)->tex.width0;
546     unsigned src_height0 = r300_resource(src)->tex.height0;
547     unsigned dst_width0 = r300_resource(dst)->tex.width0;
548     unsigned dst_height0 = r300_resource(dst)->tex.height0;
549     unsigned layout;
550     struct pipe_box box, dstbox;
551     struct pipe_sampler_view src_templ, *src_view;
552     struct pipe_surface dst_templ, *dst_view;
553 
554     /* Fallback for buffers. */
555     if ((dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) ||
556         !r300_is_blit_supported(dst->format)) {
557         util_resource_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
558                                   src, src_level, src_box);
559         return;
560     }
561 
562     /* Can't read MSAA textures. */
563     if (src->nr_samples > 1 || dst->nr_samples > 1) {
564         return;
565     }
566 
567     /* The code below changes the texture format so that the copy can be done
568      * on hardware. E.g. depth-stencil surfaces are copied as RGBA
569      * colorbuffers. */
570 
571     util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
572     util_blitter_default_src_texture(r300->blitter, &src_templ, src, src_level);
573 
574     layout = util_format_description(dst_templ.format)->layout;
575 
576     /* Handle non-renderable plain formats. */
577     if (layout == UTIL_FORMAT_LAYOUT_PLAIN &&
578         (!screen->is_format_supported(screen, src_templ.format, src->target,
579                                       src->nr_samples, src->nr_storage_samples,
580                                       PIPE_BIND_SAMPLER_VIEW) ||
581          !screen->is_format_supported(screen, dst_templ.format, dst->target,
582                                       dst->nr_samples, dst->nr_storage_samples,
583                                       PIPE_BIND_RENDER_TARGET))) {
584         switch (util_format_get_blocksize(dst_templ.format)) {
585             case 1:
586                 dst_templ.format = PIPE_FORMAT_I8_UNORM;
587                 break;
588             case 2:
589                 dst_templ.format = PIPE_FORMAT_B4G4R4A4_UNORM;
590                 break;
591             case 4:
592                 dst_templ.format = PIPE_FORMAT_B8G8R8A8_UNORM;
593                 break;
594             case 8:
595                 dst_templ.format = PIPE_FORMAT_R16G16B16A16_UNORM;
596                 break;
597             default:
598                 debug_printf("r300: copy_region: Unhandled format: %s. Falling back to software.\n"
599                              "r300: copy_region: Software fallback doesn't work for tiled textures.\n",
600                              util_format_short_name(dst_templ.format));
601         }
602         src_templ.format = dst_templ.format;
603     }
604 
605     /* Handle compressed formats. */
606     if (layout == UTIL_FORMAT_LAYOUT_S3TC ||
607         layout == UTIL_FORMAT_LAYOUT_RGTC) {
608         assert(src_templ.format == dst_templ.format);
609 
610         box = *src_box;
611         src_box = &box;
612 
613         dst_width0 = align(dst_width0, 4);
614         dst_height0 = align(dst_height0, 4);
615         src_width0 = align(src_width0, 4);
616         src_height0 = align(src_height0, 4);
617         box.width = align(box.width, 4);
618         box.height = align(box.height, 4);
619 
620         switch (util_format_get_blocksize(dst_templ.format)) {
621         case 8:
622             /* one 4x4 pixel block has 8 bytes.
623              * we set 1 pixel = 4 bytes ===> 1 block corrensponds to 2 pixels. */
624             dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
625             dst_width0 = dst_width0 / 2;
626             src_width0 = src_width0 / 2;
627             dstx /= 2;
628             box.x /= 2;
629             box.width /= 2;
630             break;
631         case 16:
632             /* one 4x4 pixel block has 16 bytes.
633              * we set 1 pixel = 4 bytes ===> 1 block corresponds to 4 pixels. */
634             dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
635             break;
636         }
637         src_templ.format = dst_templ.format;
638 
639         dst_height0 = dst_height0 / 4;
640         src_height0 = src_height0 / 4;
641         dsty /= 4;
642         box.y /= 4;
643         box.height /= 4;
644     }
645 
646     /* Fallback for textures. */
647     if (!screen->is_format_supported(screen, dst_templ.format,
648                                      dst->target, dst->nr_samples,
649                                      dst->nr_storage_samples,
650                                      PIPE_BIND_RENDER_TARGET) ||
651 	!screen->is_format_supported(screen, src_templ.format,
652                                      src->target, src->nr_samples,
653                                      src->nr_storage_samples,
654                                      PIPE_BIND_SAMPLER_VIEW)) {
655         assert(0 && "this shouldn't happen, update r300_is_blit_supported");
656         util_resource_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
657                                   src, src_level, src_box);
658         return;
659     }
660 
661     /* Decompress ZMASK. */
662     if (r300->zmask_in_use && !r300->locked_zbuffer) {
663         if (fb->zsbuf->texture == src ||
664             fb->zsbuf->texture == dst) {
665             r300_decompress_zmask(r300);
666         }
667     }
668 
669     dst_view = r300_create_surface_custom(pipe, dst, &dst_templ, dst_width0, dst_height0);
670     src_view = r300_create_sampler_view_custom(pipe, src, &src_templ, src_width0, src_height0);
671 
672     u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
673              abs(src_box->depth), &dstbox);
674 
675     r300_blitter_begin(r300, R300_COPY);
676     util_blitter_blit_generic(r300->blitter, dst_view, &dstbox,
677                               src_view, src_box, src_width0, src_height0,
678                               PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL,
679                               FALSE, FALSE);
680     r300_blitter_end(r300);
681 
682     pipe_surface_reference(&dst_view, NULL);
683     pipe_sampler_view_reference(&src_view, NULL);
684 }
685 
r300_is_simple_msaa_resolve(const struct pipe_blit_info * info)686 static boolean r300_is_simple_msaa_resolve(const struct pipe_blit_info *info)
687 {
688     unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
689     unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
690 
691     return info->src.resource->nr_samples > 1 &&
692            info->dst.resource->nr_samples <= 1 &&
693            info->dst.resource->format == info->src.resource->format &&
694            info->dst.resource->format == info->dst.format &&
695            info->src.resource->format == info->src.format &&
696            !info->scissor_enable &&
697            info->mask == PIPE_MASK_RGBA &&
698            dst_width == info->src.resource->width0 &&
699            dst_height == info->src.resource->height0 &&
700            info->dst.box.x == 0 &&
701            info->dst.box.y == 0 &&
702            info->dst.box.width == dst_width &&
703            info->dst.box.height == dst_height &&
704            info->src.box.x == 0 &&
705            info->src.box.y == 0 &&
706            info->src.box.width == dst_width &&
707            info->src.box.height == dst_height &&
708            (r300_resource(info->dst.resource)->tex.microtile != RADEON_LAYOUT_LINEAR ||
709             r300_resource(info->dst.resource)->tex.macrotile[info->dst.level] != RADEON_LAYOUT_LINEAR);
710 }
711 
r300_simple_msaa_resolve(struct pipe_context * pipe,struct pipe_resource * dst,unsigned dst_level,unsigned dst_layer,struct pipe_resource * src,enum pipe_format format)712 static void r300_simple_msaa_resolve(struct pipe_context *pipe,
713                                      struct pipe_resource *dst,
714                                      unsigned dst_level,
715                                      unsigned dst_layer,
716                                      struct pipe_resource *src,
717                                      enum pipe_format format)
718 {
719     struct r300_context *r300 = r300_context(pipe);
720     struct r300_surface *srcsurf, *dstsurf;
721     struct pipe_surface surf_tmpl;
722     struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
723 
724     memset(&surf_tmpl, 0, sizeof(surf_tmpl));
725     surf_tmpl.format = format;
726     srcsurf = r300_surface(pipe->create_surface(pipe, src, &surf_tmpl));
727 
728     surf_tmpl.format = format;
729     surf_tmpl.u.tex.level = dst_level;
730     surf_tmpl.u.tex.first_layer =
731     surf_tmpl.u.tex.last_layer = dst_layer;
732     dstsurf = r300_surface(pipe->create_surface(pipe, dst, &surf_tmpl));
733 
734     /* COLORPITCH should contain the tiling info of the resolve buffer.
735      * The tiling of the AA buffer isn't programmable anyway. */
736     srcsurf->pitch &= ~(R300_COLOR_TILE(1) | R300_COLOR_MICROTILE(3));
737     srcsurf->pitch |= dstsurf->pitch & (R300_COLOR_TILE(1) | R300_COLOR_MICROTILE(3));
738 
739     /* Enable AA resolve. */
740     aa->dest = dstsurf;
741     r300->aa_state.size = 8;
742     r300_mark_atom_dirty(r300, &r300->aa_state);
743 
744     /* Resolve the surface. */
745     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
746     util_blitter_custom_color(r300->blitter, &srcsurf->base, NULL);
747     r300_blitter_end(r300);
748 
749     /* Disable AA resolve. */
750     aa->dest = NULL;
751     r300->aa_state.size = 4;
752     r300_mark_atom_dirty(r300, &r300->aa_state);
753 
754     pipe_surface_reference((struct pipe_surface**)&srcsurf, NULL);
755     pipe_surface_reference((struct pipe_surface**)&dstsurf, NULL);
756 }
757 
r300_msaa_resolve(struct pipe_context * pipe,const struct pipe_blit_info * info)758 static void r300_msaa_resolve(struct pipe_context *pipe,
759                               const struct pipe_blit_info *info)
760 {
761     struct r300_context *r300 = r300_context(pipe);
762     struct pipe_screen *screen = pipe->screen;
763     struct pipe_resource *tmp, templ;
764     struct pipe_blit_info blit;
765 
766     assert(info->src.level == 0);
767     assert(info->src.box.z == 0);
768     assert(info->src.box.depth == 1);
769     assert(info->dst.box.depth == 1);
770 
771     if (r300_is_simple_msaa_resolve(info)) {
772         r300_simple_msaa_resolve(pipe, info->dst.resource, info->dst.level,
773                                  info->dst.box.z, info->src.resource,
774                                  info->src.format);
775         return;
776     }
777 
778     /* resolve into a temporary texture, then blit */
779     memset(&templ, 0, sizeof(templ));
780     templ.target = PIPE_TEXTURE_2D;
781     templ.format = info->src.resource->format;
782     templ.width0 = info->src.resource->width0;
783     templ.height0 = info->src.resource->height0;
784     templ.depth0 = 1;
785     templ.array_size = 1;
786     templ.usage = PIPE_USAGE_DEFAULT;
787     templ.flags = R300_RESOURCE_FORCE_MICROTILING;
788 
789     tmp = screen->resource_create(screen, &templ);
790 
791     /* resolve */
792     r300_simple_msaa_resolve(pipe, tmp, 0, 0, info->src.resource,
793                              info->src.format);
794 
795     /* blit */
796     blit = *info;
797     blit.src.resource = tmp;
798     blit.src.box.z = 0;
799 
800     r300_blitter_begin(r300, R300_BLIT | R300_IGNORE_RENDER_COND);
801     util_blitter_blit(r300->blitter, &blit);
802     r300_blitter_end(r300);
803 
804     pipe_resource_reference(&tmp, NULL);
805 }
806 
r300_blit(struct pipe_context * pipe,const struct pipe_blit_info * blit)807 static void r300_blit(struct pipe_context *pipe,
808                       const struct pipe_blit_info *blit)
809 {
810     struct r300_context *r300 = r300_context(pipe);
811     struct pipe_framebuffer_state *fb =
812         (struct pipe_framebuffer_state*)r300->fb_state.state;
813     struct pipe_blit_info info = *blit;
814 
815     /* The driver supports sRGB textures but not framebuffers. Blitting
816      * from sRGB to sRGB should be the same as blitting from linear
817      * to linear, so use that, This avoids incorrect linearization.
818      */
819     if (util_format_is_srgb(info.src.format)) {
820       info.src.format = util_format_linear(info.src.format);
821       info.dst.format = util_format_linear(info.dst.format);
822     }
823 
824     /* MSAA resolve. */
825     if (info.src.resource->nr_samples > 1 &&
826         !util_format_is_depth_or_stencil(info.src.resource->format)) {
827         r300_msaa_resolve(pipe, &info);
828         return;
829     }
830 
831     /* Can't read MSAA textures. */
832     if (info.src.resource->nr_samples > 1) {
833         return;
834     }
835 
836     /* Blit a combined depth-stencil resource as color.
837      * S8Z24 is the only supported stencil format. */
838     if ((info.mask & PIPE_MASK_S) &&
839         info.src.format == PIPE_FORMAT_S8_UINT_Z24_UNORM &&
840         info.dst.format == PIPE_FORMAT_S8_UINT_Z24_UNORM) {
841         if (info.dst.resource->nr_samples > 1) {
842             /* Cannot do that with MSAA buffers. */
843             info.mask &= ~PIPE_MASK_S;
844             if (!(info.mask & PIPE_MASK_Z)) {
845                 return;
846             }
847         } else {
848             /* Single-sample buffer. */
849             info.src.format = PIPE_FORMAT_B8G8R8A8_UNORM;
850             info.dst.format = PIPE_FORMAT_B8G8R8A8_UNORM;
851             if (info.mask & PIPE_MASK_Z) {
852                 info.mask = PIPE_MASK_RGBA; /* depth+stencil */
853             } else {
854                 info.mask = PIPE_MASK_B; /* stencil only */
855             }
856         }
857     }
858 
859     /* Decompress ZMASK. */
860     if (r300->zmask_in_use && !r300->locked_zbuffer) {
861         if (fb->zsbuf->texture == info.src.resource ||
862             fb->zsbuf->texture == info.dst.resource) {
863             r300_decompress_zmask(r300);
864         }
865     }
866 
867     r300_blitter_begin(r300, R300_BLIT |
868 		       (info.render_condition_enable ? 0 : R300_IGNORE_RENDER_COND));
869     util_blitter_blit(r300->blitter, &info);
870     r300_blitter_end(r300);
871 }
872 
r300_flush_resource(struct pipe_context * ctx,struct pipe_resource * resource)873 static void r300_flush_resource(struct pipe_context *ctx,
874 				struct pipe_resource *resource)
875 {
876 }
877 
r300_init_blit_functions(struct r300_context * r300)878 void r300_init_blit_functions(struct r300_context *r300)
879 {
880     r300->context.clear = r300_clear;
881     r300->context.clear_render_target = r300_clear_render_target;
882     r300->context.clear_depth_stencil = r300_clear_depth_stencil;
883     r300->context.resource_copy_region = r300_resource_copy_region;
884     r300->context.blit = r300_blit;
885     r300->context.flush_resource = r300_flush_resource;
886 }
887