nv50: move VERTEX_{START,END} to {Prepare,Done}Composite
[nouveau] / src / nv40_xv_tex.c
1 /*
2  * Copyright 2007-2008 Maarten Maathuis
3  * Copyright 2008 Stephane Marchesin
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a
6  * copy of this software and associated documentation files (the "Software"),
7  * to deal in the Software without restriction, including without limitation
8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9  * and/or sell copies of the Software, and to permit persons to whom the
10  * Software is furnished to do so, subject to the following conditions:
11  *
12  * The above copyright notice and this permission notice (including the next
13  * paragraph) shall be included in all copies or substantial portions of the
14  * Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22  * DEALINGS IN THE SOFTWARE.
23  */
24
25 #ifdef HAVE_CONFIG_H
26 #include "config.h"
27 #endif
28
29 #include "xf86xv.h"
30 #include <X11/extensions/Xv.h>
31 #include "exa.h"
32 #include "damage.h"
33 #include "dixstruct.h"
34 #include "fourcc.h"
35
36 #include "nv_include.h"
37 #include "nv_dma.h"
38
39 #include "nv30_shaders.h"
40
41 extern Atom xvSyncToVBlank, xvSetDefaults;
42
43 /*
44  * The filtering function used for video scaling. We use a cubic filter as defined in 
45  * "Reconstruction Filters in Computer Graphics"
46  * Mitchell & Netravali in SIGGRAPH '88 
47  */
48 static float filter_func(float x)
49 {
50         const double B=0.75;
51         const double C=(1.0-B)/2.0;
52         double x1=fabs(x);
53         double x2=fabs(x)*x1;
54         double x3=fabs(x)*x2;
55
56         if (fabs(x)<1.0) 
57                 return ( (12.0-9.0*B-6.0*C)*x3+(-18.0+12.0*B+6.0*C)*x2+(6.0-2.0*B) )/6.0; 
58         else 
59                 return ( (-B-6.0*C)*x3+(6.0*B+30.0*C)*x2+(-12.0*B-48.0*C)*x1+(8.0*B+24.0*C) )/6.0;
60 }
61
62 static int8_t f32tosb8(float v)
63 {
64         return (int8_t)(v*127.0);
65 }
66
67 /*
68  * Implements the filtering as described in
69  * "Fast Third-Order Texture Filtering"
70  * Sigg & Hardwiger in GPU Gems 2
71  */
72 #define TABLE_SIZE 512
73 static void compute_filter_table(int8_t *t) {
74         int i;
75         float x;
76         for(i=0;i<TABLE_SIZE;i++) {
77                 x=(i+0.5)/TABLE_SIZE;
78
79                 float w0=filter_func(x+1.0);
80                 float w1=filter_func(x);
81                 float w2=filter_func(x-1.0);
82                 float w3=filter_func(x-2.0);
83
84                 t[4*i+2]=f32tosb8(1.0+x-w1/(w0+w1));
85                 t[4*i+1]=f32tosb8(1.0-x+w3/(w2+w3));
86                 t[4*i+0]=f32tosb8(w0+w1);
87                 t[4*i+3]=f32tosb8(0.0);
88         }
89 }
90
91 static struct nouveau_bo *table_mem = NULL;
92 static void
93 NV40_LoadFilterTable(ScrnInfoPtr pScrn)
94 {
95         NVPtr pNv = NVPTR(pScrn);
96
97         if (!table_mem) {
98                 if (nouveau_bo_new(pNv->dev, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART,
99                                 0, TABLE_SIZE*sizeof(float)*4, &table_mem)) {
100                         xf86DrvMsg(pScrn->scrnIndex, X_ERROR,
101                                 "Couldn't alloc filter table!\n");
102                         return;
103                 }
104
105                 if (nouveau_bo_map(table_mem, NOUVEAU_BO_RDWR)) {
106                         xf86DrvMsg(pScrn->scrnIndex, X_ERROR,
107                                    "Couldn't map filter table!\n");
108                         return;
109                 }
110
111                 int8_t *t=table_mem->map;
112                 compute_filter_table(t);
113         }
114 }
115
116 #define SWIZZLE(ts0x,ts0y,ts0z,ts0w,ts1x,ts1y,ts1z,ts1w)                                \
117         (                                                                               \
118         NV40TCL_TEX_SWIZZLE_S0_X_##ts0x | NV40TCL_TEX_SWIZZLE_S0_Y_##ts0y       |       \
119         NV40TCL_TEX_SWIZZLE_S0_Z_##ts0z | NV40TCL_TEX_SWIZZLE_S0_W_##ts0w       |       \
120         NV40TCL_TEX_SWIZZLE_S1_X_##ts1x | NV40TCL_TEX_SWIZZLE_S1_Y_##ts1y       |       \
121         NV40TCL_TEX_SWIZZLE_S1_Z_##ts1z | NV40TCL_TEX_SWIZZLE_S1_W_##ts1w               \
122         )
123
124 /*
125  * Texture 0 : filter table
126  * Texture 1 : Y data
127  * Texture 2 : UV data
128  */
129 static Bool
130 NV40VideoTexture(ScrnInfoPtr pScrn, struct nouveau_bo *src, int offset,
131                  uint16_t width, uint16_t height, uint16_t src_pitch, int unit)
132 {
133         NVPtr pNv = NVPTR(pScrn);
134         struct nouveau_channel *chan = pNv->chan;
135         struct nouveau_grobj *curie = pNv->Nv3D;
136
137         uint32_t card_fmt = 0;
138         uint32_t card_swz = 0;
139
140         switch(unit) {
141         case 0:
142                 card_fmt = NV40TCL_TEX_FORMAT_FORMAT_A8R8G8B8;
143                 card_swz = SWIZZLE(S1, S1, S1, S1, X, Y, Z, W);
144                 break;
145         case 1:
146                 card_fmt = NV40TCL_TEX_FORMAT_FORMAT_L8;
147                 card_swz = SWIZZLE(S1, S1, S1, S1, X, X, X, X);
148                 break;
149         case 2:
150                 card_fmt = NV40TCL_TEX_FORMAT_FORMAT_A8L8;
151 #if X_BYTE_ORDER == X_BIG_ENDIAN
152                 card_swz = SWIZZLE(S1, S1, S1, S1, Z, W, X, Y); /* x = V, y = U */
153 #else
154                 card_swz = SWIZZLE(S1, S1, S1, S1, W, Z, Y, X); /* x = V, y = U */
155 #endif
156                 break;
157         }
158
159         BEGIN_RING(chan, curie, NV40TCL_TEX_OFFSET(unit), 8);
160         OUT_RELOCl(chan, src, offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
161         if (unit==0) {
162                 OUT_RELOCd(chan, pNv->FB, card_fmt | 
163                                  NV40TCL_TEX_FORMAT_DIMS_1D | 0x8000 |
164                                  NV40TCL_TEX_FORMAT_NO_BORDER |
165                                  (1 << NV40TCL_TEX_FORMAT_MIPMAP_COUNT_SHIFT),
166                                  NOUVEAU_BO_VRAM | NOUVEAU_BO_RD,
167                                  NV40TCL_TEX_FORMAT_DMA0, 0);
168                 OUT_RING  (chan, NV40TCL_TEX_WRAP_S_REPEAT |
169                                  NV40TCL_TEX_WRAP_T_CLAMP_TO_EDGE |
170                                  NV40TCL_TEX_WRAP_R_CLAMP_TO_EDGE);
171         } else {
172                 OUT_RELOCd(chan, pNv->FB, card_fmt | NV40TCL_TEX_FORMAT_LINEAR |
173                                  NV40TCL_TEX_FORMAT_RECT | 0x8000 |
174                                  NV40TCL_TEX_FORMAT_DIMS_2D |
175                                  NV40TCL_TEX_FORMAT_NO_BORDER |
176                                  (1 << NV40TCL_TEX_FORMAT_MIPMAP_COUNT_SHIFT),
177                                  NOUVEAU_BO_VRAM | NOUVEAU_BO_RD,
178                                  NV40TCL_TEX_FORMAT_DMA0, 0);
179                 OUT_RING  (chan, NV40TCL_TEX_WRAP_S_CLAMP_TO_EDGE |
180                                  NV40TCL_TEX_WRAP_T_CLAMP_TO_EDGE |
181                                  NV40TCL_TEX_WRAP_R_CLAMP_TO_EDGE);
182         }
183
184         OUT_RING  (chan, NV40TCL_TEX_ENABLE_ENABLE);
185         OUT_RING  (chan, card_swz);
186         if (unit == 0)
187                 OUT_RING  (chan, NV40TCL_TEX_FILTER_SIGNED_ALPHA |
188                                  NV40TCL_TEX_FILTER_SIGNED_RED |
189                                  NV40TCL_TEX_FILTER_SIGNED_GREEN |
190                                  NV40TCL_TEX_FILTER_SIGNED_BLUE |
191                                  NV40TCL_TEX_FILTER_MIN_LINEAR |
192                                  NV40TCL_TEX_FILTER_MAG_LINEAR | 0x3fd6);
193         else
194                 OUT_RING  (chan, NV40TCL_TEX_FILTER_MIN_LINEAR |
195                                  NV40TCL_TEX_FILTER_MAG_LINEAR | 0x3fd6);
196         OUT_RING  (chan, (width << 16) | height);
197         OUT_RING  (chan, 0); /* border ARGB */
198
199         BEGIN_RING(chan, curie, NV40TCL_TEX_SIZE1(unit), 1);
200         OUT_RING  (chan, (1 << NV40TCL_TEX_SIZE1_DEPTH_SHIFT) |
201                          (uint16_t) src_pitch);
202
203         return TRUE;
204 }
205
206 Bool
207 NV40GetSurfaceFormat(PixmapPtr ppix, int *fmt_ret)
208 {
209         switch (ppix->drawable.bitsPerPixel) {
210         case 32:
211                 *fmt_ret = NV40TCL_RT_FORMAT_COLOR_A8R8G8B8;
212                 break;
213         case 24:
214                 *fmt_ret = NV40TCL_RT_FORMAT_COLOR_X8R8G8B8;
215                 break;
216         case 16:
217                 *fmt_ret = NV40TCL_RT_FORMAT_COLOR_R5G6B5;
218                 break;
219         case 8:
220                 *fmt_ret = NV40TCL_RT_FORMAT_COLOR_B8;
221                 break;
222         default:
223                 return FALSE;
224         }
225
226         return TRUE;
227 }
228
229 void
230 NV40StopTexturedVideo(ScrnInfoPtr pScrn, pointer data, Bool Exit)
231 {
232 }
233
234 #define VERTEX_OUT(sx,sy,dx,dy) do {                                           \
235         BEGIN_RING(chan, curie, NV40TCL_VTX_ATTR_2F_X(8), 4);                  \
236         OUT_RINGf (chan, (sx)); OUT_RINGf (chan, (sy));                        \
237         OUT_RINGf (chan, (sx)/2.0); OUT_RINGf (chan, (sy)/2.0);                \
238         BEGIN_RING(chan, curie, NV40TCL_VTX_ATTR_2I(0), 1);                    \
239         OUT_RING  (chan, ((dy)<<16)|(dx));                                     \
240 } while(0)
241
242 int
243 NV40PutTextureImage(ScrnInfoPtr pScrn,
244                     struct nouveau_bo *src, int src_offset, int src_offset2,
245                     int id, int src_pitch, BoxPtr dstBox,
246                     int x1, int y1, int x2, int y2,
247                     uint16_t width, uint16_t height,
248                     uint16_t src_w, uint16_t src_h,
249                     uint16_t drw_w, uint16_t drw_h,
250                     RegionPtr clipBoxes, PixmapPtr ppix,
251                     NVPortPrivPtr pPriv)
252 {
253         NVPtr pNv = NVPTR(pScrn);
254         struct nouveau_channel *chan = pNv->chan;
255         struct nouveau_grobj *curie = pNv->Nv3D;
256         Bool redirected = FALSE;
257         float X1, X2, Y1, Y2;
258         BoxPtr pbox;
259         int nbox;
260         int dst_format = 0;
261
262         if (drw_w > 4096 || drw_h > 4096) {
263                 xf86DrvMsg(pScrn->scrnIndex, X_ERROR,
264                         "XV: Draw size too large.\n");
265                 return BadAlloc;
266         }
267
268         if (!NV40GetSurfaceFormat(ppix, &dst_format)) {
269                 ErrorF("No surface format, bad.\n");
270         }
271
272 #ifdef COMPOSITE
273         if (!NVExaPixmapIsOnscreen(ppix))
274                 redirected = TRUE;
275 #endif
276
277         pbox = REGION_RECTS(clipBoxes);
278         nbox = REGION_NUM_RECTS(clipBoxes);
279
280         /* Disable blending */
281         BEGIN_RING(chan, curie, NV40TCL_BLEND_ENABLE, 1);
282         OUT_RING  (chan, 0);
283
284         /* Setup surface */
285         BEGIN_RING(chan, curie, NV40TCL_RT_FORMAT, 3);
286         OUT_RING  (chan, NV40TCL_RT_FORMAT_TYPE_LINEAR |
287                          NV40TCL_RT_FORMAT_ZETA_Z24S8 | dst_format);
288         OUT_RING  (chan, exaGetPixmapPitch(ppix));
289         OUT_PIXMAPl(chan, ppix, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
290
291         NV40_LoadFilterTable(pScrn);
292
293         NV40VideoTexture(pScrn, table_mem, 0, TABLE_SIZE, 1, 0 , 0);
294         NV40VideoTexture(pScrn, src, src_offset, src_w, src_h, src_pitch, 1);
295         /* We've got NV12 format, which means half width and half height texture of chroma channels. */
296         NV40VideoTexture(pScrn, src, src_offset2, src_w/2, src_h/2, src_pitch, 2);
297
298         NV40_LoadVtxProg(pScrn, &nv40_vp_video);
299         if (pPriv->bicubic)
300                 NV40_LoadFragProg(pScrn, &nv40_fp_yv12_bicubic);
301         else
302                 NV40_LoadFragProg(pScrn, &nv30_fp_yv12_bilinear);
303
304         /* Appears to be some kind of cache flush, needed here at least
305          * sometimes.. funky text rendering otherwise :)
306          */
307         BEGIN_RING(chan, curie, NV40TCL_TEX_CACHE_CTL, 1);
308         OUT_RING  (chan, 2);
309         BEGIN_RING(chan, curie, NV40TCL_TEX_CACHE_CTL, 1);
310         OUT_RING  (chan, 1);
311
312         /* Just before rendering we wait for vblank in the non-composited case. */
313         if (pPriv->SyncToVBlank && !redirected) {
314                 uint8_t crtcs = nv_window_belongs_to_crtc(pScrn, dstBox->x1, dstBox->y1,
315                         dstBox->x2 - dstBox->x1, dstBox->y2 - dstBox->y1);
316
317                 FIRE_RING (chan);
318                 if (crtcs & 0x1)
319                         NVWaitVSync(pScrn, 0);
320                 else if (crtcs & 0x2)
321                         NVWaitVSync(pScrn, 1);
322         }
323
324         /* These are fixed point values in the 16.16 format. */
325         X1 = (float)(x1>>16)+(float)(x1&0xFFFF)/(float)0x10000;
326         Y1 = (float)(y1>>16)+(float)(y1&0xFFFF)/(float)0x10000;
327         X2 = (float)(x2>>16)+(float)(x2&0xFFFF)/(float)0x10000;
328         Y2 = (float)(y2>>16)+(float)(y2&0xFFFF)/(float)0x10000;
329
330         BEGIN_RING(chan, curie, NV40TCL_BEGIN_END, 1);
331         OUT_RING  (chan, NV40TCL_BEGIN_END_TRIANGLES);
332
333         while(nbox--) {
334                 float tx1=X1+(float)(pbox->x1 - dstBox->x1)*(X2-X1)/(float)(drw_w);
335                 float tx2=X1+(float)(pbox->x2 - dstBox->x1)*(src_w)/(float)(drw_w);
336                 float ty1=Y1+(float)(pbox->y1 - dstBox->y1)*(Y2-Y1)/(float)(drw_h);
337                 float ty2=Y1+(float)(pbox->y2 - dstBox->y1)*(src_h)/(float)(drw_h);
338                 int sx1=pbox->x1;
339                 int sx2=pbox->x2;
340                 int sy1=pbox->y1;
341                 int sy2=pbox->y2;
342
343                 BEGIN_RING(chan, curie, NV40TCL_SCISSOR_HORIZ, 2);
344                 OUT_RING  (chan, (sx2 << 16) | 0);
345                 OUT_RING  (chan, (sy2 << 16) | 0);
346
347                 VERTEX_OUT(tx1, ty1, sx1, sy1);
348                 VERTEX_OUT(tx2+(tx2-tx1), ty1, sx2+(sx2-sx1), sy1);
349                 VERTEX_OUT(tx1, ty2+(ty2-ty1), sx1, sy2+(sy2-sy1));
350
351                 pbox++;
352         }
353
354         BEGIN_RING(chan, curie, NV40TCL_BEGIN_END, 1);
355         OUT_RING  (chan, NV40TCL_BEGIN_END_STOP);
356
357         FIRE_RING (chan);
358
359         return Success;
360 }
361
362 /**
363  * NV40SetTexturePortAttribute
364  * sets the attribute "attribute" of port "data" to value "value"
365  * supported attributes:
366  * Sync to vblank.
367  * 
368  * @param pScrenInfo
369  * @param attribute attribute to set
370  * @param value value to which attribute is to be set
371  * @param data port from which the attribute is to be set
372  * 
373  * @return Success, if setting is successful
374  * BadValue/BadMatch, if value/attribute are invalid
375  */
376 int
377 NV40SetTexturePortAttribute(ScrnInfoPtr pScrn, Atom attribute,
378                        INT32 value, pointer data)
379 {
380         NVPortPrivPtr pPriv = (NVPortPrivPtr)data;
381         NVPtr           pNv = NVPTR(pScrn);
382
383         if ((attribute == xvSyncToVBlank) && pNv->WaitVSyncPossible) {
384                 if ((value < 0) || (value > 1))
385                         return BadValue;
386                 pPriv->SyncToVBlank = value;
387         } else
388         if (attribute == xvSetDefaults) {
389                 pPriv->SyncToVBlank = pNv->WaitVSyncPossible;
390         } else
391                 return BadMatch;
392
393         return Success;
394 }
395
396 /**
397  * NV40GetTexturePortAttribute
398  * reads the value of attribute "attribute" from port "data" into INT32 "*value"
399  * Sync to vblank.
400  * 
401  * @param pScrn unused
402  * @param attribute attribute to be read
403  * @param value value of attribute will be stored here
404  * @param data port from which attribute will be read
405  * @return Success, if queried attribute exists
406  */
407 int
408 NV40GetTexturePortAttribute(ScrnInfoPtr pScrn, Atom attribute,
409                        INT32 *value, pointer data)
410 {
411         NVPortPrivPtr pPriv = (NVPortPrivPtr)data;
412
413         if(attribute == xvSyncToVBlank)
414                 *value = (pPriv->SyncToVBlank) ? 1 : 0;
415         else
416                 return BadMatch;
417
418         return Success;
419 }
420