Merge remote branch 'origin/master' into pipe-video
[mesa.git] / src / gallium / drivers / nvfx / nvfx_state_fb.c
1 #include "nvfx_context.h"
2 #include "nvfx_resource.h"
3 #include "util/u_format.h"
4
5 static inline boolean
6 nvfx_surface_linear_renderable(struct pipe_surface* surf)
7 {
8 /* TODO: precompute this in nvfx_surface creation */
9 return (surf->texture->flags & NVFX_RESOURCE_FLAG_LINEAR)
10 && !(((struct nvfx_surface*)surf)->offset & 63)
11 && !(((struct nvfx_surface*)surf)->pitch & 63);
12 }
13
14 static inline boolean
15 nvfx_surface_swizzled_renderable(struct pipe_framebuffer_state* fb, struct pipe_surface* surf)
16 {
17 /* TODO: precompute this in nvfx_surface creation */
18 return !((struct nvfx_miptree*)surf->texture)->linear_pitch
19 && (surf->texture->target != PIPE_TEXTURE_3D || u_minify(surf->texture->depth0, surf->u.tex.level) <= 1)
20 && !(((struct nvfx_surface*)surf)->offset & 127)
21 && (surf->width == fb->width)
22 && (surf->height == fb->height)
23 && !((struct nvfx_surface*)surf)->temp
24 && (surf->format == PIPE_FORMAT_B8G8R8A8_UNORM || surf->format == PIPE_FORMAT_B8G8R8X8_UNORM || surf->format == PIPE_FORMAT_B5G6R5_UNORM);
25 }
26
27 static boolean
28 nvfx_surface_get_render_target(struct pipe_surface* surf, int all_swizzled, struct nvfx_render_target* target)
29 {
30 struct nvfx_surface* ns = (struct nvfx_surface*)surf;
31 if(!ns->temp)
32 {
33 target->bo = ((struct nvfx_miptree*)surf->texture)->base.bo;
34 target->offset = ns->offset;
35 target->pitch = align(ns->pitch, 64);
36 assert(target->pitch);
37 return FALSE;
38 }
39 else
40 {
41 target->offset = 0;
42 target->pitch = ns->temp->linear_pitch;
43 target->bo = ns->temp->base.bo;
44 assert(target->pitch);
45 return TRUE;
46 }
47 }
48
49 int
50 nvfx_framebuffer_prepare(struct nvfx_context *nvfx)
51 {
52 struct pipe_framebuffer_state *fb = &nvfx->framebuffer;
53 int i, color_format = 0, zeta_format = 0;
54 int all_swizzled = 1;
55
56 if(!nvfx->is_nv4x)
57 assert(fb->nr_cbufs <= 1);
58 else
59 assert(fb->nr_cbufs <= 4);
60
61 for (i = 0; i < fb->nr_cbufs; i++) {
62 if (color_format) {
63 if(color_format != fb->cbufs[i]->format)
64 return -1;
65 } else
66 color_format = fb->cbufs[i]->format;
67
68 if(!nvfx_surface_swizzled_renderable(fb, fb->cbufs[i]))
69 all_swizzled = 0;
70 }
71
72 if (fb->zsbuf) {
73 /* TODO: return FALSE if we have a format not supporting a depth buffer (e.g. r8); currently those are not supported at all */
74 if(!nvfx_surface_swizzled_renderable(fb, fb->zsbuf))
75 all_swizzled = 0;
76
77 if(all_swizzled && util_format_get_blocksize(color_format) != util_format_get_blocksize(zeta_format))
78 all_swizzled = 0;
79 }
80
81 for (i = 0; i < fb->nr_cbufs; i++) {
82 if(!((struct nvfx_surface*)fb->cbufs[i])->temp && !all_swizzled && !nvfx_surface_linear_renderable(fb->cbufs[i]))
83 nvfx_surface_create_temp(&nvfx->pipe, fb->cbufs[i]);
84 }
85
86 if(fb->zsbuf) {
87 if(!((struct nvfx_surface*)fb->zsbuf)->temp && !all_swizzled && !nvfx_surface_linear_renderable(fb->zsbuf))
88 nvfx_surface_create_temp(&nvfx->pipe, fb->zsbuf);
89 }
90
91 return all_swizzled;
92 }
93
94 void
95 nvfx_framebuffer_validate(struct nvfx_context *nvfx, unsigned prepare_result)
96 {
97 struct pipe_framebuffer_state *fb = &nvfx->framebuffer;
98 struct nouveau_channel *chan = nvfx->screen->base.channel;
99 uint32_t rt_enable, rt_format;
100 int i;
101 unsigned rt_flags = NOUVEAU_BO_RDWR | NOUVEAU_BO_VRAM;
102 unsigned w = fb->width;
103 unsigned h = fb->height;
104
105 rt_enable = (NV30_3D_RT_ENABLE_COLOR0 << fb->nr_cbufs) - 1;
106 if (rt_enable & (NV30_3D_RT_ENABLE_COLOR1 |
107 NV40_3D_RT_ENABLE_COLOR2 | NV40_3D_RT_ENABLE_COLOR3))
108 rt_enable |= NV30_3D_RT_ENABLE_MRT;
109
110 nvfx->state.render_temps = 0;
111
112 for (i = 0; i < fb->nr_cbufs; i++)
113 nvfx->state.render_temps |= nvfx_surface_get_render_target(fb->cbufs[i], prepare_result, &nvfx->hw_rt[i]) << i;
114
115 for(; i < 4; ++i)
116 nvfx->hw_rt[i].bo = NULL;
117
118 nvfx->hw_zeta.bo = NULL;
119
120 if (fb->zsbuf) {
121 nvfx->state.render_temps |= nvfx_surface_get_render_target(fb->zsbuf, prepare_result, &nvfx->hw_zeta) << 7;
122
123 assert(util_format_get_stride(fb->zsbuf->format, fb->width) <= nvfx->hw_zeta.pitch);
124 assert(nvfx->hw_zeta.offset + nvfx->hw_zeta.pitch * fb->height <= nvfx->hw_zeta.bo->size);
125 }
126
127 if (prepare_result) {
128 assert(!(fb->width & (fb->width - 1)) && !(fb->height & (fb->height - 1)));
129
130 rt_format = NV30_3D_RT_FORMAT_TYPE_SWIZZLED |
131 (util_logbase2(fb->width) << NV30_3D_RT_FORMAT_LOG2_WIDTH__SHIFT) |
132 (util_logbase2(fb->height) << NV30_3D_RT_FORMAT_LOG2_HEIGHT__SHIFT);
133 } else
134 rt_format = NV30_3D_RT_FORMAT_TYPE_LINEAR;
135
136 if(fb->nr_cbufs > 0) {
137 switch (fb->cbufs[0]->format) {
138 case PIPE_FORMAT_B8G8R8X8_UNORM:
139 rt_format |= NV30_3D_RT_FORMAT_COLOR_X8R8G8B8;
140 break;
141 case PIPE_FORMAT_B8G8R8A8_UNORM:
142 case 0:
143 rt_format |= NV30_3D_RT_FORMAT_COLOR_A8R8G8B8;
144 break;
145 case PIPE_FORMAT_R8G8B8X8_UNORM:
146 rt_format |= NV30_3D_RT_FORMAT_COLOR_X8B8G8R8;
147 break;
148 case PIPE_FORMAT_R8G8B8A8_UNORM:
149 rt_format |= NV30_3D_RT_FORMAT_COLOR_A8B8G8R8;
150 break;
151 case PIPE_FORMAT_B5G6R5_UNORM:
152 rt_format |= NV30_3D_RT_FORMAT_COLOR_R5G6B5;
153 break;
154 case PIPE_FORMAT_R32G32B32A32_FLOAT:
155 rt_format |= NV30_3D_RT_FORMAT_COLOR_A32B32G32R32_FLOAT;
156 break;
157 case PIPE_FORMAT_R16G16B16A16_FLOAT:
158 rt_format |= NV30_3D_RT_FORMAT_COLOR_A16B16G16R16_FLOAT;
159 break;
160 default:
161 assert(0);
162 }
163 } else if(fb->zsbuf && util_format_get_blocksize(fb->zsbuf->format) == 2)
164 rt_format |= NV30_3D_RT_FORMAT_COLOR_R5G6B5;
165 else
166 rt_format |= NV30_3D_RT_FORMAT_COLOR_A8R8G8B8;
167
168 if(fb->zsbuf) {
169 switch (fb->zsbuf->format) {
170 case PIPE_FORMAT_Z16_UNORM:
171 rt_format |= NV30_3D_RT_FORMAT_ZETA_Z16;
172 break;
173 case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
174 case PIPE_FORMAT_X8Z24_UNORM:
175 case 0:
176 rt_format |= NV30_3D_RT_FORMAT_ZETA_Z24S8;
177 break;
178 default:
179 assert(0);
180 }
181 } else if(fb->nr_cbufs && util_format_get_blocksize(fb->cbufs[0]->format) == 2)
182 rt_format |= NV30_3D_RT_FORMAT_ZETA_Z16;
183 else
184 rt_format |= NV30_3D_RT_FORMAT_ZETA_Z24S8;
185
186 MARK_RING(chan, 42, 10);
187
188 if ((rt_enable & NV30_3D_RT_ENABLE_COLOR0) || fb->zsbuf) {
189 struct nvfx_render_target *rt0 = &nvfx->hw_rt[0];
190 uint32_t pitch;
191
192 if(!(rt_enable & NV30_3D_RT_ENABLE_COLOR0))
193 rt0 = &nvfx->hw_zeta;
194
195 pitch = rt0->pitch;
196
197 if(!nvfx->is_nv4x)
198 {
199 if (nvfx->hw_zeta.bo)
200 pitch |= (nvfx->hw_zeta.pitch << 16);
201 else
202 pitch |= (pitch << 16);
203 }
204
205 //printf("rendering to bo %p [%i] at offset %i with pitch %i\n", rt0->bo, rt0->bo->handle, rt0->offset, pitch);
206
207 OUT_RING(chan, RING_3D(NV30_3D_DMA_COLOR0, 1));
208 OUT_RELOC(chan, rt0->bo, 0,
209 rt_flags | NOUVEAU_BO_OR,
210 chan->vram->handle, chan->gart->handle);
211 OUT_RING(chan, RING_3D(NV30_3D_COLOR0_PITCH, 2));
212 OUT_RING(chan, pitch);
213 OUT_RELOC(chan, rt0->bo,
214 rt0->offset, rt_flags | NOUVEAU_BO_LOW,
215 0, 0);
216 }
217
218 if (rt_enable & NV30_3D_RT_ENABLE_COLOR1) {
219 OUT_RING(chan, RING_3D(NV30_3D_DMA_COLOR1, 1));
220 OUT_RELOC(chan, nvfx->hw_rt[1].bo, 0,
221 rt_flags | NOUVEAU_BO_OR,
222 chan->vram->handle, chan->gart->handle);
223 OUT_RING(chan, RING_3D(NV30_3D_COLOR1_OFFSET, 2));
224 OUT_RELOC(chan, nvfx->hw_rt[1].bo,
225 nvfx->hw_rt[1].offset, rt_flags | NOUVEAU_BO_LOW,
226 0, 0);
227 OUT_RING(chan, nvfx->hw_rt[1].pitch);
228 }
229
230 if(nvfx->is_nv4x)
231 {
232 if (rt_enable & NV40_3D_RT_ENABLE_COLOR2) {
233 OUT_RING(chan, RING_3D(NV40_3D_DMA_COLOR2, 1));
234 OUT_RELOC(chan, nvfx->hw_rt[2].bo, 0,
235 rt_flags | NOUVEAU_BO_OR,
236 chan->vram->handle, chan->gart->handle);
237 OUT_RING(chan, RING_3D(NV40_3D_COLOR2_OFFSET, 1));
238 OUT_RELOC(chan, nvfx->hw_rt[2].bo,
239 nvfx->hw_rt[2].offset, rt_flags | NOUVEAU_BO_LOW,
240 0, 0);
241 OUT_RING(chan, RING_3D(NV40_3D_COLOR2_PITCH, 1));
242 OUT_RING(chan, nvfx->hw_rt[2].pitch);
243 }
244
245 if (rt_enable & NV40_3D_RT_ENABLE_COLOR3) {
246 OUT_RING(chan, RING_3D(NV40_3D_DMA_COLOR3, 1));
247 OUT_RELOC(chan, nvfx->hw_rt[3].bo, 0,
248 rt_flags | NOUVEAU_BO_OR,
249 chan->vram->handle, chan->gart->handle);
250 OUT_RING(chan, RING_3D(NV40_3D_COLOR3_OFFSET, 1));
251 OUT_RELOC(chan, nvfx->hw_rt[3].bo,
252 nvfx->hw_rt[3].offset, rt_flags | NOUVEAU_BO_LOW,
253 0, 0);
254 OUT_RING(chan, RING_3D(NV40_3D_COLOR3_PITCH, 1));
255 OUT_RING(chan, nvfx->hw_rt[3].pitch);
256 }
257 }
258
259 if (fb->zsbuf) {
260 OUT_RING(chan, RING_3D(NV30_3D_DMA_ZETA, 1));
261 OUT_RELOC(chan, nvfx->hw_zeta.bo, 0,
262 rt_flags | NOUVEAU_BO_OR,
263 chan->vram->handle, chan->gart->handle);
264 OUT_RING(chan, RING_3D(NV30_3D_ZETA_OFFSET, 1));
265 /* TODO: reverse engineer LMA */
266 OUT_RELOC(chan, nvfx->hw_zeta.bo,
267 nvfx->hw_zeta.offset, rt_flags | NOUVEAU_BO_LOW, 0, 0);
268 if(nvfx->is_nv4x) {
269 OUT_RING(chan, RING_3D(NV40_3D_ZETA_PITCH, 1));
270 OUT_RING(chan, nvfx->hw_zeta.pitch);
271 }
272 }
273 else if(nvfx->is_nv4x) {
274 OUT_RING(chan, RING_3D(NV40_3D_ZETA_PITCH, 1));
275 OUT_RING(chan, 64);
276 }
277
278 OUT_RING(chan, RING_3D(NV30_3D_RT_ENABLE, 1));
279 OUT_RING(chan, rt_enable);
280 OUT_RING(chan, RING_3D(NV30_3D_RT_HORIZ, 3));
281 OUT_RING(chan, (w << 16) | 0);
282 OUT_RING(chan, (h << 16) | 0);
283 OUT_RING(chan, rt_format);
284 OUT_RING(chan, RING_3D(NV30_3D_VIEWPORT_HORIZ, 2));
285 OUT_RING(chan, (w << 16) | 0);
286 OUT_RING(chan, (h << 16) | 0);
287 OUT_RING(chan, RING_3D(NV30_3D_VIEWPORT_CLIP_HORIZ(0), 2));
288 OUT_RING(chan, ((w - 1) << 16) | 0);
289 OUT_RING(chan, ((h - 1) << 16) | 0);
290
291 if(!nvfx->is_nv4x) {
292 /* Wonder why this is needed, context should all be set to zero on init */
293 /* TODO: we can most likely remove this, after putting it in context init */
294 OUT_RING(chan, RING_3D(NV30_3D_VIEWPORT_TX_ORIGIN, 1));
295 OUT_RING(chan, 0);
296 }
297 nvfx->relocs_needed &=~ NVFX_RELOCATE_FRAMEBUFFER;
298 }
299
300 void
301 nvfx_framebuffer_relocate(struct nvfx_context *nvfx)
302 {
303 struct nouveau_channel *chan = nvfx->screen->base.channel;
304 unsigned rt_flags = NOUVEAU_BO_RDWR | NOUVEAU_BO_VRAM;
305 rt_flags |= NOUVEAU_BO_DUMMY;
306 MARK_RING(chan, 20, 20);
307
308 #define DO_(var, pfx, name) \
309 if(var.bo) { \
310 OUT_RELOC(chan, var.bo, RING_3D(pfx##_3D_DMA_##name, 1), rt_flags, 0, 0); \
311 OUT_RELOC(chan, var.bo, 0, \
312 rt_flags | NOUVEAU_BO_OR, \
313 chan->vram->handle, chan->gart->handle); \
314 OUT_RELOC(chan, var.bo, RING_3D(pfx##_3D_##name##_OFFSET, 1), rt_flags, 0, 0); \
315 OUT_RELOC(chan, var.bo, \
316 var.offset, rt_flags | NOUVEAU_BO_LOW, \
317 0, 0); \
318 }
319
320 #define DO(pfx, num) DO_(nvfx->hw_rt[num], pfx, COLOR##num)
321 DO(NV30, 0);
322 DO(NV30, 1);
323 DO(NV40, 2);
324 DO(NV40, 3);
325
326 DO_(nvfx->hw_zeta, NV30, ZETA);
327 nvfx->relocs_needed &=~ NVFX_RELOCATE_FRAMEBUFFER;
328 }