2 * Copyright 2009 Marek Olšák <maraeo@gmail.com>
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:
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
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. */
23 #include "r300_context.h"
24 #include "r300_emit.h"
25 #include "r300_texture.h"
27 #include "util/u_format.h"
28 #include "util/u_pack_color.h"
29 #include "util/u_surface.h"
31 enum r300_blitter_op /* bitmask */
34 R300_SAVE_TEXTURES = 2,
35 R300_SAVE_FRAMEBUFFER = 4,
36 R300_IGNORE_RENDER_COND = 8,
38 R300_CLEAR = R300_STOP_QUERY,
40 R300_CLEAR_SURFACE = R300_STOP_QUERY | R300_SAVE_FRAMEBUFFER,
42 R300_COPY = R300_STOP_QUERY | R300_SAVE_FRAMEBUFFER |
43 R300_SAVE_TEXTURES | R300_IGNORE_RENDER_COND,
45 R300_DECOMPRESS = R300_STOP_QUERY | R300_IGNORE_RENDER_COND,
48 static void r300_blitter_begin(struct r300_context* r300, enum r300_blitter_op op)
50 if ((op & R300_STOP_QUERY) && r300->query_current) {
51 r300->blitter_saved_query = r300->query_current;
52 r300_stop_query(r300);
55 /* Yeah we have to save all those states to ensure the blitter operation
56 * is really transparent. The states will be restored by the blitter once
58 util_blitter_save_blend(r300->blitter, r300->blend_state.state);
59 util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state.state);
60 util_blitter_save_stencil_ref(r300->blitter, &(r300->stencil_ref));
61 util_blitter_save_rasterizer(r300->blitter, r300->rs_state.state);
62 util_blitter_save_fragment_shader(r300->blitter, r300->fs.state);
63 util_blitter_save_vertex_shader(r300->blitter, r300->vs_state.state);
64 util_blitter_save_viewport(r300->blitter, &r300->viewport);
66 util_blitter_save_vertex_buffers(r300->blitter, r300->vbuf_mgr->nr_vertex_buffers,
67 r300->vbuf_mgr->vertex_buffer);
68 util_blitter_save_vertex_elements(r300->blitter, r300->vbuf_mgr->vertex_elements);
70 util_blitter_save_vertex_buffers(r300->blitter, r300->nr_vertex_buffers,
72 util_blitter_save_vertex_elements(r300->blitter, r300->velems);
75 if (op & R300_SAVE_FRAMEBUFFER) {
76 util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state);
79 if (op & R300_SAVE_TEXTURES) {
80 struct r300_textures_state* state =
81 (struct r300_textures_state*)r300->textures_state.state;
83 util_blitter_save_fragment_sampler_states(
84 r300->blitter, state->sampler_state_count,
85 (void**)state->sampler_states);
87 util_blitter_save_fragment_sampler_views(
88 r300->blitter, state->sampler_view_count,
89 (struct pipe_sampler_view**)state->sampler_views);
92 if (op & R300_IGNORE_RENDER_COND) {
94 r300->blitter_saved_skip_rendering = r300->skip_rendering+1;
95 r300->skip_rendering = FALSE;
97 r300->blitter_saved_skip_rendering = 0;
101 static void r300_blitter_end(struct r300_context *r300)
103 if (r300->blitter_saved_query) {
104 r300_resume_query(r300, r300->blitter_saved_query);
105 r300->blitter_saved_query = NULL;
108 if (r300->blitter_saved_skip_rendering) {
109 /* Restore the flag. */
110 r300->skip_rendering = r300->blitter_saved_skip_rendering-1;
114 static uint32_t r300_depth_clear_cb_value(enum pipe_format format,
118 util_pack_color(rgba, format, &uc);
120 if (util_format_get_blocksizebits(format) == 32)
123 return uc.us | (uc.us << 16);
126 static boolean r300_cbzb_clear_allowed(struct r300_context *r300,
127 unsigned clear_buffers)
129 struct pipe_framebuffer_state *fb =
130 (struct pipe_framebuffer_state*)r300->fb_state.state;
132 /* Only color clear allowed, and only one colorbuffer. */
133 if (clear_buffers != PIPE_CLEAR_COLOR || fb->nr_cbufs != 1)
136 return r300_surface(fb->cbufs[0])->cbzb_allowed;
139 static boolean r300_fast_zclear_allowed(struct r300_context *r300)
141 struct pipe_framebuffer_state *fb =
142 (struct pipe_framebuffer_state*)r300->fb_state.state;
144 return r300_resource(fb->zsbuf->texture)->tex.zmask_dwords[fb->zsbuf->u.tex.level] != 0;
147 static boolean r300_hiz_clear_allowed(struct r300_context *r300)
149 struct pipe_framebuffer_state *fb =
150 (struct pipe_framebuffer_state*)r300->fb_state.state;
152 return r300_resource(fb->zsbuf->texture)->tex.hiz_dwords[fb->zsbuf->u.tex.level] != 0;
155 static uint32_t r300_depth_clear_value(enum pipe_format format,
156 double depth, unsigned stencil)
159 case PIPE_FORMAT_Z16_UNORM:
160 case PIPE_FORMAT_X8Z24_UNORM:
161 return util_pack_z(format, depth);
163 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
164 return util_pack_z_stencil(format, depth, stencil);
172 static uint32_t r300_hiz_clear_value(double depth)
174 uint32_t r = (uint32_t)(CLAMP(depth, 0, 1) * 255.5);
176 return r | (r << 8) | (r << 16) | (r << 24);
179 /* Clear currently bound buffers. */
180 static void r300_clear(struct pipe_context* pipe,
182 const union pipe_color_union *color,
186 /* My notes about Zbuffer compression:
188 * 1) The zbuffer must be micro-tiled and whole microtiles must be
189 * written if compression is enabled. If microtiling is disabled,
192 * 2) There is ZMASK RAM which contains a compressed zbuffer.
193 * Each dword of the Z Mask contains compression information
194 * for 16 4x4 pixel tiles, that is 2 bits for each tile.
195 * On chips with 2 Z pipes, every other dword maps to a different
196 * pipe. On newer chipsets, there is a new compression mode
197 * with 8x8 pixel tiles per 2 bits.
199 * 3) The FASTFILL bit has nothing to do with filling. It only tells hw
200 * it should look in the ZMASK RAM first before fetching from a real
203 * 4) If a pixel is in a cleared state, ZB_DEPTHCLEARVALUE is returned
204 * during zbuffer reads instead of the value that is actually stored
205 * in the zbuffer memory. A pixel is in a cleared state when its ZMASK
206 * is equal to 0. Therefore, if you clear ZMASK with zeros, you may
207 * leave the zbuffer memory uninitialized, but then you must enable
208 * compression, so that the ZMASK RAM is actually used.
210 * 5) Each 4x4 (or 8x8) tile is automatically decompressed and recompressed
211 * during zbuffer updates. A special decompressing operation should be
212 * used to fully decompress a zbuffer, which basically just stores all
213 * compressed tiles in ZMASK to the zbuffer memory.
215 * 6) For a 16-bit zbuffer, compression causes a hung with one or
216 * two samples and should not be used.
218 * 7) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears
219 * to avoid needless decompression.
221 * 8) Fastfill must not be used if reading of compressed Z data is disabled
222 * and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE),
223 * i.e. it cannot be used to compress the zbuffer.
225 * 9) ZB_CB_CLEAR does not interact with zbuffer compression in any way.
230 struct r300_context* r300 = r300_context(pipe);
231 struct pipe_framebuffer_state *fb =
232 (struct pipe_framebuffer_state*)r300->fb_state.state;
233 struct r300_hyperz_state *hyperz =
234 (struct r300_hyperz_state*)r300->hyperz_state.state;
235 uint32_t width = fb->width;
236 uint32_t height = fb->height;
237 uint32_t hyperz_dcv = hyperz->zb_depthclearvalue;
239 /* Enable fast Z clear.
240 * The zbuffer must be in micro-tiled mode, otherwise it locks up. */
241 if (buffers & PIPE_CLEAR_DEPTHSTENCIL) {
242 boolean zmask_clear, hiz_clear;
244 zmask_clear = r300_fast_zclear_allowed(r300);
245 hiz_clear = r300_hiz_clear_allowed(r300);
247 /* If we need Hyper-Z. */
248 if (zmask_clear || hiz_clear) {
249 r300->num_z_clears++;
251 /* Try to obtain the access to Hyper-Z buffers if we don't have one. */
252 if (!r300->hyperz_enabled) {
253 r300->hyperz_enabled =
254 r300->rws->cs_request_feature(r300->cs,
255 RADEON_FID_R300_HYPERZ_ACCESS,
257 if (r300->hyperz_enabled) {
258 /* Need to emit HyperZ buffer regs for the first time. */
259 r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
263 /* Setup Hyper-Z clears. */
264 if (r300->hyperz_enabled) {
265 DBG(r300, DBG_HYPERZ, "r300: Clear memory: %s%s\n",
266 zmask_clear ? "ZMASK " : "", hiz_clear ? "HIZ" : "");
269 hyperz_dcv = hyperz->zb_depthclearvalue =
270 r300_depth_clear_value(fb->zsbuf->format, depth, stencil);
272 r300_mark_atom_dirty(r300, &r300->zmask_clear);
273 buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
277 r300->hiz_clear_value = r300_hiz_clear_value(depth);
278 r300_mark_atom_dirty(r300, &r300->hiz_clear);
284 /* Enable CBZB clear. */
285 if (r300_cbzb_clear_allowed(r300, buffers)) {
286 struct r300_surface *surf = r300_surface(fb->cbufs[0]);
288 hyperz->zb_depthclearvalue =
289 r300_depth_clear_cb_value(surf->base.format, color->f);
291 width = surf->cbzb_width;
292 height = surf->cbzb_height;
294 r300->cbzb_clear = TRUE;
295 r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
300 enum pipe_format cformat = fb->nr_cbufs ? fb->cbufs[0]->format : PIPE_FORMAT_NONE;
301 /* Clear using the blitter. */
302 r300_blitter_begin(r300, R300_CLEAR);
303 util_blitter_clear(r300->blitter,
307 buffers, cformat, color, depth, stencil);
308 r300_blitter_end(r300);
309 } else if (r300->zmask_clear.dirty || r300->hiz_clear.dirty) {
310 /* Just clear zmask and hiz now, this does not use the standard draw
312 /* Calculate zmask_clear and hiz_clear atom sizes. */
314 (r300->zmask_clear.dirty ? r300->zmask_clear.size : 0) +
315 (r300->hiz_clear.dirty ? r300->hiz_clear.size : 0) +
316 r300_get_num_cs_end_dwords(r300);
318 /* Reserve CS space. */
319 if (dwords > (RADEON_MAX_CMDBUF_DWORDS - r300->cs->cdw)) {
320 r300_flush(&r300->context, RADEON_FLUSH_ASYNC, NULL);
323 /* Emit clear packets. */
324 if (r300->zmask_clear.dirty) {
325 r300_emit_zmask_clear(r300, r300->zmask_clear.size,
326 r300->zmask_clear.state);
327 r300->zmask_clear.dirty = FALSE;
329 if (r300->hiz_clear.dirty) {
330 r300_emit_hiz_clear(r300, r300->hiz_clear.size,
331 r300->hiz_clear.state);
332 r300->hiz_clear.dirty = FALSE;
338 /* Disable CBZB clear. */
339 if (r300->cbzb_clear) {
340 r300->cbzb_clear = FALSE;
341 hyperz->zb_depthclearvalue = hyperz_dcv;
342 r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
345 /* Enable fastfill and/or hiz.
347 * If we cleared zmask/hiz, it's in use now. The Hyper-Z state update
348 * looks if zmask/hiz is in use and programs hardware accordingly. */
349 if (r300->zmask_in_use || r300->hiz_in_use) {
350 r300_mark_atom_dirty(r300, &r300->hyperz_state);
354 /* Clear a region of a color surface to a constant value. */
355 static void r300_clear_render_target(struct pipe_context *pipe,
356 struct pipe_surface *dst,
357 const union pipe_color_union *color,
358 unsigned dstx, unsigned dsty,
359 unsigned width, unsigned height)
361 struct r300_context *r300 = r300_context(pipe);
363 r300_blitter_begin(r300, R300_CLEAR_SURFACE);
364 util_blitter_clear_render_target(r300->blitter, dst, color,
365 dstx, dsty, width, height);
366 r300_blitter_end(r300);
369 /* Clear a region of a depth stencil surface. */
370 static void r300_clear_depth_stencil(struct pipe_context *pipe,
371 struct pipe_surface *dst,
372 unsigned clear_flags,
375 unsigned dstx, unsigned dsty,
376 unsigned width, unsigned height)
378 struct r300_context *r300 = r300_context(pipe);
379 struct pipe_framebuffer_state *fb =
380 (struct pipe_framebuffer_state*)r300->fb_state.state;
382 if (r300->zmask_in_use && !r300->locked_zbuffer) {
383 if (fb->zsbuf->texture == dst->texture) {
384 r300_decompress_zmask(r300);
388 /* XXX Do not decompress ZMask of the currently-set zbuffer. */
389 r300_blitter_begin(r300, R300_CLEAR_SURFACE);
390 util_blitter_clear_depth_stencil(r300->blitter, dst, clear_flags, depth, stencil,
391 dstx, dsty, width, height);
392 r300_blitter_end(r300);
395 void r300_decompress_zmask(struct r300_context *r300)
397 struct pipe_framebuffer_state *fb =
398 (struct pipe_framebuffer_state*)r300->fb_state.state;
400 if (!r300->zmask_in_use || r300->locked_zbuffer)
403 r300->zmask_decompress = TRUE;
404 r300_mark_atom_dirty(r300, &r300->hyperz_state);
406 r300_blitter_begin(r300, R300_DECOMPRESS);
407 util_blitter_clear_depth_custom(r300->blitter, fb->width, fb->height, 0,
408 r300->dsa_decompress_zmask);
409 r300_blitter_end(r300);
411 r300->zmask_decompress = FALSE;
412 r300->zmask_in_use = FALSE;
413 r300_mark_atom_dirty(r300, &r300->hyperz_state);
416 void r300_decompress_zmask_locked_unsafe(struct r300_context *r300)
418 struct pipe_framebuffer_state fb;
420 memset(&fb, 0, sizeof(fb));
421 fb.width = r300->locked_zbuffer->width;
422 fb.height = r300->locked_zbuffer->height;
423 fb.zsbuf = r300->locked_zbuffer;
425 r300->context.set_framebuffer_state(&r300->context, &fb);
426 r300_decompress_zmask(r300);
429 void r300_decompress_zmask_locked(struct r300_context *r300)
431 struct pipe_framebuffer_state saved_fb;
433 memset(&saved_fb, 0, sizeof(saved_fb));
434 util_copy_framebuffer_state(&saved_fb, r300->fb_state.state);
435 r300_decompress_zmask_locked_unsafe(r300);
436 r300->context.set_framebuffer_state(&r300->context, &saved_fb);
437 util_unreference_framebuffer_state(&saved_fb);
439 pipe_surface_reference(&r300->locked_zbuffer, NULL);
442 bool r300_is_blit_supported(enum pipe_format format)
444 const struct util_format_description *desc =
445 util_format_description(format);
447 return desc->layout == UTIL_FORMAT_LAYOUT_PLAIN ||
448 desc->layout == UTIL_FORMAT_LAYOUT_S3TC ||
449 desc->layout == UTIL_FORMAT_LAYOUT_RGTC;
452 /* Copy a block of pixels from one surface to another. */
453 static void r300_resource_copy_region(struct pipe_context *pipe,
454 struct pipe_resource *dst,
456 unsigned dstx, unsigned dsty, unsigned dstz,
457 struct pipe_resource *src,
459 const struct pipe_box *src_box)
461 struct pipe_screen *screen = pipe->screen;
462 struct r300_context *r300 = r300_context(pipe);
463 struct pipe_framebuffer_state *fb =
464 (struct pipe_framebuffer_state*)r300->fb_state.state;
465 unsigned src_width0 = r300_resource(src)->tex.width0;
466 unsigned src_height0 = r300_resource(src)->tex.height0;
467 unsigned dst_width0 = r300_resource(dst)->tex.width0;
468 unsigned dst_height0 = r300_resource(dst)->tex.height0;
471 struct pipe_sampler_view src_templ, *src_view;
472 struct pipe_surface dst_templ, *dst_view;
474 /* Fallback for buffers. */
475 if ((dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) ||
476 !r300_is_blit_supported(dst->format)) {
477 util_resource_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
478 src, src_level, src_box);
482 /* The code below changes the texture format so that the copy can be done
483 * on hardware. E.g. depth-stencil surfaces are copied as RGBA
486 util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz, src_box);
487 util_blitter_default_src_texture(&src_templ, src, src_level);
489 layout = util_format_description(dst_templ.format)->layout;
491 /* Handle non-renderable plain formats. */
492 if (layout == UTIL_FORMAT_LAYOUT_PLAIN &&
493 (!screen->is_format_supported(screen, src_templ.format, src->target,
495 PIPE_BIND_SAMPLER_VIEW) ||
496 !screen->is_format_supported(screen, dst_templ.format, dst->target,
498 PIPE_BIND_RENDER_TARGET))) {
499 switch (util_format_get_blocksize(dst_templ.format)) {
501 dst_templ.format = PIPE_FORMAT_I8_UNORM;
504 dst_templ.format = PIPE_FORMAT_B4G4R4A4_UNORM;
507 dst_templ.format = PIPE_FORMAT_B8G8R8A8_UNORM;
510 dst_templ.format = PIPE_FORMAT_R16G16B16A16_UNORM;
513 debug_printf("r300: copy_region: Unhandled format: %s. Falling back to software.\n"
514 "r300: copy_region: Software fallback doesn't work for tiled textures.\n",
515 util_format_short_name(dst_templ.format));
517 src_templ.format = dst_templ.format;
520 /* Handle compressed formats. */
521 if (layout == UTIL_FORMAT_LAYOUT_S3TC ||
522 layout == UTIL_FORMAT_LAYOUT_RGTC) {
523 assert(src_templ.format == dst_templ.format);
528 dst_width0 = align(dst_width0, 4);
529 dst_height0 = align(dst_height0, 4);
530 src_width0 = align(src_width0, 4);
531 src_height0 = align(src_height0, 4);
532 box.width = align(box.width, 4);
533 box.height = align(box.height, 4);
535 switch (util_format_get_blocksize(dst_templ.format)) {
537 /* one 4x4 pixel block has 8 bytes.
538 * we set 1 pixel = 4 bytes ===> 1 block corrensponds to 2 pixels. */
539 dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
540 dst_width0 = dst_width0 / 2;
541 src_width0 = src_width0 / 2;
547 /* one 4x4 pixel block has 16 bytes.
548 * we set 1 pixel = 4 bytes ===> 1 block corresponds to 4 pixels. */
549 dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
552 src_templ.format = dst_templ.format;
554 dst_height0 = dst_height0 / 4;
555 src_height0 = src_height0 / 4;
561 /* Fallback for textures. */
562 if (!screen->is_format_supported(screen, dst_templ.format,
563 dst->target, dst->nr_samples,
564 PIPE_BIND_RENDER_TARGET) ||
565 !screen->is_format_supported(screen, src_templ.format,
566 src->target, src->nr_samples,
567 PIPE_BIND_SAMPLER_VIEW)) {
568 assert(0 && "this shouldn't happen, update r300_is_blit_supported");
569 util_resource_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
570 src, src_level, src_box);
574 /* Decompress ZMASK. */
575 if (r300->zmask_in_use && !r300->locked_zbuffer) {
576 if (fb->zsbuf->texture == src ||
577 fb->zsbuf->texture == dst) {
578 r300_decompress_zmask(r300);
582 dst_view = r300_create_surface_custom(pipe, dst, &dst_templ, dst_width0, dst_height0);
583 src_view = r300_create_sampler_view_custom(pipe, src, &src_templ, src_width0, src_height0);
585 r300_blitter_begin(r300, R300_COPY);
586 util_blitter_copy_texture_view(r300->blitter, dst_view, dstx, dsty,
588 src_width0, src_height0);
589 r300_blitter_end(r300);
591 pipe_surface_reference(&dst_view, NULL);
592 pipe_sampler_view_reference(&src_view, NULL);
595 void r300_init_blit_functions(struct r300_context *r300)
597 r300->context.clear = r300_clear;
598 r300->context.clear_render_target = r300_clear_render_target;
599 r300->context.clear_depth_stencil = r300_clear_depth_stencil;
600 r300->context.resource_copy_region = r300_resource_copy_region;