r300g: do not clear with blitter if we clear just the ZMask RAM
[mesa.git] / src / gallium / drivers / r300 / r300_blit.c
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_hyperz.h"
26 #include "r300_texture.h"
27 #include "r300_winsys.h"
28
29 #include "util/u_format.h"
30 #include "util/u_pack_color.h"
31
32 enum r300_blitter_op /* bitmask */
33 {
34 R300_CLEAR = 1,
35 R300_CLEAR_SURFACE = 2,
36 R300_COPY = 4
37 };
38
39 static void r300_blitter_begin(struct r300_context* r300, enum r300_blitter_op op)
40 {
41 if (r300->query_current) {
42 r300->blitter_saved_query = r300->query_current;
43 r300_stop_query(r300);
44 }
45
46 /* Yeah we have to save all those states to ensure the blitter operation
47 * is really transparent. The states will be restored by the blitter once
48 * copying is done. */
49 util_blitter_save_blend(r300->blitter, r300->blend_state.state);
50 util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state.state);
51 util_blitter_save_stencil_ref(r300->blitter, &(r300->stencil_ref));
52 util_blitter_save_rasterizer(r300->blitter, r300->rs_state.state);
53 util_blitter_save_fragment_shader(r300->blitter, r300->fs.state);
54 util_blitter_save_vertex_shader(r300->blitter, r300->vs_state.state);
55 util_blitter_save_viewport(r300->blitter, &r300->viewport);
56 util_blitter_save_clip(r300->blitter, (struct pipe_clip_state*)r300->clip_state.state);
57 util_blitter_save_vertex_elements(r300->blitter, r300->velems);
58 util_blitter_save_vertex_buffers(r300->blitter, r300->vertex_buffer_count,
59 r300->vertex_buffer);
60
61 if (op & (R300_CLEAR_SURFACE | R300_COPY))
62 util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state);
63
64 if (op & R300_COPY) {
65 struct r300_textures_state* state =
66 (struct r300_textures_state*)r300->textures_state.state;
67
68 util_blitter_save_fragment_sampler_states(
69 r300->blitter, state->sampler_state_count,
70 (void**)state->sampler_states);
71
72 util_blitter_save_fragment_sampler_views(
73 r300->blitter, state->sampler_view_count,
74 (struct pipe_sampler_view**)state->sampler_views);
75 }
76 }
77
78 static void r300_blitter_end(struct r300_context *r300)
79 {
80 if (r300->blitter_saved_query) {
81 r300_resume_query(r300, r300->blitter_saved_query);
82 r300->blitter_saved_query = NULL;
83 }
84 }
85
86 static uint32_t r300_depth_clear_cb_value(enum pipe_format format,
87 const float* rgba)
88 {
89 union util_color uc;
90 util_pack_color(rgba, format, &uc);
91
92 if (util_format_get_blocksizebits(format) == 32)
93 return uc.ui;
94 else
95 return uc.us | (uc.us << 16);
96 }
97
98 static boolean r300_cbzb_clear_allowed(struct r300_context *r300,
99 unsigned clear_buffers)
100 {
101 struct pipe_framebuffer_state *fb =
102 (struct pipe_framebuffer_state*)r300->fb_state.state;
103
104 /* Only color clear allowed, and only one colorbuffer. */
105 if (clear_buffers != PIPE_CLEAR_COLOR || fb->nr_cbufs != 1)
106 return FALSE;
107
108 return r300_surface(fb->cbufs[0])->cbzb_allowed;
109 }
110
111 static uint32_t r300_depth_clear_value(enum pipe_format format,
112 double depth, unsigned stencil)
113 {
114 switch (format) {
115 case PIPE_FORMAT_Z16_UNORM:
116 case PIPE_FORMAT_X8Z24_UNORM:
117 return util_pack_z(format, depth);
118
119 case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
120 return util_pack_z_stencil(format, depth, stencil);
121
122 default:
123 assert(0);
124 return 0;
125 }
126 }
127
128 /* Clear currently bound buffers. */
129 static void r300_clear(struct pipe_context* pipe,
130 unsigned buffers,
131 const float* rgba,
132 double depth,
133 unsigned stencil)
134 {
135 /* My notes about fastfill:
136 *
137 * 1) Only the zbuffer is cleared.
138 *
139 * 2) The zbuffer must be micro-tiled and whole microtiles must be
140 * written. If microtiling is disabled, it locks up.
141 *
142 * 3) There is Z Mask RAM which contains a compressed zbuffer and
143 * it interacts with fastfill. We should figure out how to use it
144 * to get more performance.
145 * This is what we know about the Z Mask:
146 *
147 * Each dword of the Z Mask contains compression information
148 * for 16 4x4 pixel blocks, that is 2 bits for each block.
149 * On chips with 2 Z pipes, every other dword maps to a different
150 * pipe.
151 *
152 * 4) ZB_DEPTHCLEARVALUE is used to clear the zbuffer and the Z Mask must
153 * be equal to 0. (clear the Z Mask RAM with zeros)
154 *
155 * 5) For 16-bit zbuffer, compression causes a hung with one or
156 * two samples and should not be used.
157 *
158 * 6) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears
159 * to avoid needless decompression.
160 *
161 * 7) Fastfill must not be used if reading of compressed Z data is disabled
162 * and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE),
163 * i.e. it cannot be used to compress the zbuffer.
164 *
165 * 8) ZB_CB_CLEAR does not interact with fastfill in any way.
166 *
167 * - Marek
168 */
169
170 struct r300_context* r300 = r300_context(pipe);
171 struct pipe_framebuffer_state *fb =
172 (struct pipe_framebuffer_state*)r300->fb_state.state;
173 struct r300_hyperz_state *hyperz =
174 (struct r300_hyperz_state*)r300->hyperz_state.state;
175 uint32_t width = fb->width;
176 uint32_t height = fb->height;
177 boolean has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
178 uint32_t hyperz_dcv = hyperz->zb_depthclearvalue;
179
180 /* Enable fast Z clear.
181 * The zbuffer must be in micro-tiled mode, otherwise it locks up. */
182 if ((buffers & PIPE_CLEAR_DEPTHSTENCIL) && has_hyperz) {
183 hyperz_dcv = hyperz->zb_depthclearvalue =
184 r300_depth_clear_value(fb->zsbuf->format, depth, stencil);
185
186 r300_mark_fb_state_dirty(r300, R300_CHANGED_ZCLEAR_FLAG);
187 if (r300_texture(fb->zsbuf->texture)->zmask_mem[fb->zsbuf->level]) {
188 r300->zmask_clear.dirty = TRUE;
189 buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
190 }
191 if (r300->hiz_enable)
192 r300->hiz_clear.dirty = TRUE;
193 }
194
195 /* Enable CBZB clear. */
196 if (r300_cbzb_clear_allowed(r300, buffers)) {
197 struct r300_surface *surf = r300_surface(fb->cbufs[0]);
198
199 hyperz->zb_depthclearvalue =
200 r300_depth_clear_cb_value(surf->base.format, rgba);
201
202 width = surf->cbzb_width;
203 height = surf->cbzb_height;
204
205 r300->cbzb_clear = TRUE;
206 r300_mark_fb_state_dirty(r300, R300_CHANGED_CBZB_FLAG);
207 }
208
209 /* Clear. */
210 if (buffers) {
211 /* Clear using the blitter. */
212 r300_blitter_begin(r300, R300_CLEAR);
213 util_blitter_clear(r300->blitter,
214 width,
215 height,
216 fb->nr_cbufs,
217 buffers, rgba, depth, stencil);
218 r300_blitter_end(r300);
219 } else if (r300->zmask_clear.dirty) {
220 /* Just clear zmask and hiz now, this does not use a standard draw
221 * procedure. */
222 unsigned dwords;
223
224 /* Calculate zmask_clear and hiz_clear atom sizes. */
225 r300_update_hyperz_state(r300);
226 dwords = r300->zmask_clear.size +
227 (r300->hiz_clear.dirty ? r300->hiz_clear.size : 0) +
228 r300_get_num_cs_end_dwords(r300);
229
230 /* Reserve CS space. */
231 if (dwords > (r300->cs->ndw - r300->cs->cdw)) {
232 r300->context.flush(&r300->context, 0, NULL);
233 }
234
235 /* Emit clear packets. */
236 r300_emit_zmask_clear(r300, r300->zmask_clear.size,
237 r300->zmask_clear.state);
238 if (r300->hiz_clear.dirty)
239 r300_emit_hiz_clear(r300, r300->hiz_clear.size,
240 r300->hiz_clear.state);
241 } else {
242 assert(0);
243 }
244
245 /* Disable CBZB clear. */
246 if (r300->cbzb_clear) {
247 r300->cbzb_clear = FALSE;
248 hyperz->zb_depthclearvalue = hyperz_dcv;
249 r300_mark_fb_state_dirty(r300, R300_CHANGED_CBZB_FLAG);
250 }
251
252 /* Enable fastfill.
253 *
254 * If we cleared the zmask, it's dirty now. The Hyper-Z state update
255 * looks for a dirty zmask and enables fastfill accordingly. */
256 if (fb->zsbuf &&
257 r300_texture(fb->zsbuf->texture)->dirty_zmask[fb->zsbuf->level]) {
258 r300->hyperz_state.dirty = TRUE;
259 }
260
261 /* XXX this flush "fixes" a hardlock in the cubestorm xscreensaver */
262 if (r300->flush_counter == 0)
263 pipe->flush(pipe, 0, NULL);
264 }
265
266 /* Clear a region of a color surface to a constant value. */
267 static void r300_clear_render_target(struct pipe_context *pipe,
268 struct pipe_surface *dst,
269 const float *rgba,
270 unsigned dstx, unsigned dsty,
271 unsigned width, unsigned height)
272 {
273 struct r300_context *r300 = r300_context(pipe);
274
275 r300_blitter_begin(r300, R300_CLEAR_SURFACE);
276 util_blitter_clear_render_target(r300->blitter, dst, rgba,
277 dstx, dsty, width, height);
278 r300_blitter_end(r300);
279 }
280
281 /* Clear a region of a depth stencil surface. */
282 static void r300_clear_depth_stencil(struct pipe_context *pipe,
283 struct pipe_surface *dst,
284 unsigned clear_flags,
285 double depth,
286 unsigned stencil,
287 unsigned dstx, unsigned dsty,
288 unsigned width, unsigned height)
289 {
290 struct r300_context *r300 = r300_context(pipe);
291
292 r300_blitter_begin(r300, R300_CLEAR_SURFACE);
293 util_blitter_clear_depth_stencil(r300->blitter, dst, clear_flags, depth, stencil,
294 dstx, dsty, width, height);
295 r300_blitter_end(r300);
296 }
297
298 /* Flush a depth stencil buffer. */
299 void r300_flush_depth_stencil(struct pipe_context *pipe,
300 struct pipe_resource *dst,
301 struct pipe_subresource subdst,
302 unsigned zslice)
303 {
304 struct r300_context *r300 = r300_context(pipe);
305 struct pipe_surface *dstsurf;
306 struct r300_texture *tex = r300_texture(dst);
307
308 if (!tex->zmask_mem[subdst.level])
309 return;
310 if (!tex->dirty_zmask[subdst.level])
311 return;
312
313 dstsurf = pipe->screen->get_tex_surface(pipe->screen, dst,
314 subdst.face, subdst.level, zslice,
315 PIPE_BIND_DEPTH_STENCIL);
316 r300->z_decomp_rd = TRUE;
317 r300_blitter_begin(r300, R300_CLEAR_SURFACE);
318 util_blitter_flush_depth_stencil(r300->blitter, dstsurf);
319 r300_blitter_end(r300);
320 r300->z_decomp_rd = FALSE;
321
322 tex->dirty_zmask[subdst.level] = FALSE;
323 }
324
325 /* Copy a block of pixels from one surface to another using HW. */
326 static void r300_hw_copy_region(struct pipe_context* pipe,
327 struct pipe_resource *dst,
328 struct pipe_subresource subdst,
329 unsigned dstx, unsigned dsty, unsigned dstz,
330 struct pipe_resource *src,
331 struct pipe_subresource subsrc,
332 unsigned srcx, unsigned srcy, unsigned srcz,
333 unsigned width, unsigned height)
334 {
335 struct r300_context* r300 = r300_context(pipe);
336
337 r300_blitter_begin(r300, R300_COPY);
338 util_blitter_copy_region(r300->blitter, dst, subdst, dstx, dsty, dstz,
339 src, subsrc, srcx, srcy, srcz, width, height,
340 TRUE);
341 r300_blitter_end(r300);
342 }
343
344 /* Copy a block of pixels from one surface to another. */
345 static void r300_resource_copy_region(struct pipe_context *pipe,
346 struct pipe_resource *dst,
347 struct pipe_subresource subdst,
348 unsigned dstx, unsigned dsty, unsigned dstz,
349 struct pipe_resource *src,
350 struct pipe_subresource subsrc,
351 unsigned srcx, unsigned srcy, unsigned srcz,
352 unsigned width, unsigned height)
353 {
354 enum pipe_format old_format = dst->format;
355 enum pipe_format new_format = old_format;
356 boolean is_depth;
357 if (!pipe->screen->is_format_supported(pipe->screen,
358 old_format, src->target,
359 src->nr_samples,
360 PIPE_BIND_RENDER_TARGET |
361 PIPE_BIND_SAMPLER_VIEW, 0) &&
362 util_format_is_plain(old_format)) {
363 switch (util_format_get_blocksize(old_format)) {
364 case 1:
365 new_format = PIPE_FORMAT_I8_UNORM;
366 break;
367 case 2:
368 new_format = PIPE_FORMAT_B4G4R4A4_UNORM;
369 break;
370 case 4:
371 new_format = PIPE_FORMAT_B8G8R8A8_UNORM;
372 break;
373 case 8:
374 new_format = PIPE_FORMAT_R16G16B16A16_UNORM;
375 break;
376 default:
377 debug_printf("r300: surface_copy: Unhandled format: %s. Falling back to software.\n"
378 "r300: surface_copy: Software fallback doesn't work for tiled textures.\n",
379 util_format_short_name(old_format));
380 }
381 }
382
383 is_depth = util_format_get_component_bits(src->format, UTIL_FORMAT_COLORSPACE_ZS, 0) != 0;
384 if (is_depth) {
385 r300_flush_depth_stencil(pipe, src, subsrc, srcz);
386 }
387 if (old_format != new_format) {
388 dst->format = new_format;
389 src->format = new_format;
390
391 r300_texture_reinterpret_format(pipe->screen,
392 dst, new_format);
393 r300_texture_reinterpret_format(pipe->screen,
394 src, new_format);
395 }
396
397 r300_hw_copy_region(pipe, dst, subdst, dstx, dsty, dstz,
398 src, subsrc, srcx, srcy, srcz, width, height);
399
400 if (old_format != new_format) {
401 dst->format = old_format;
402 src->format = old_format;
403
404 r300_texture_reinterpret_format(pipe->screen,
405 dst, old_format);
406 r300_texture_reinterpret_format(pipe->screen,
407 src, old_format);
408 }
409 }
410
411 void r300_init_blit_functions(struct r300_context *r300)
412 {
413 r300->context.clear = r300_clear;
414 r300->context.clear_render_target = r300_clear_render_target;
415 r300->context.clear_depth_stencil = r300_clear_depth_stencil;
416 r300->context.resource_copy_region = r300_resource_copy_region;
417 }