projects / mesa.git / blob
? search:


103305037a8bb2186918ba2a1d472e67060146c9
[mesa.git] / src / gallium / state_trackers / nine / nine_ff.c
1
2 /* FF is big and ugly so feel free to write lines as long as you like.
3 * Aieeeeeeeee !
4 *
5 * Let me make that clearer:
6 * Aieeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee ! !! !!!
7 */
8
9 #include "device9.h"
10 #include "basetexture9.h"
11 #include "vertexdeclaration9.h"
12 #include "vertexshader9.h"
13 #include "pixelshader9.h"
14 #include "nine_ff.h"
15 #include "nine_defines.h"
16 #include "nine_helpers.h"
17 #include "nine_pipe.h"
18 #include "nine_dump.h"
19
20 #include "pipe/p_context.h"
21 #include "tgsi/tgsi_ureg.h"
22 #include "tgsi/tgsi_dump.h"
23 #include "util/u_box.h"
24 #include "util/u_hash_table.h"
25 #include "util/u_upload_mgr.h"
26
27 #define NINE_TGSI_LAZY_DEVS 1
28
29 #define DBG_CHANNEL DBG_FF
30
31 #define NINE_FF_NUM_VS_CONST 256
32 #define NINE_FF_NUM_PS_CONST 24
33
34 struct fvec4
35 {
36 float x, y, z, w;
37 };
38
39 struct nine_ff_vs_key
40 {
41 union {
42 struct {
43 uint32_t position_t : 1;
44 uint32_t lighting : 1;
45 uint32_t darkness : 1; /* lighting enabled but no active lights */
46 uint32_t localviewer : 1;
47 uint32_t vertexpointsize : 1;
48 uint32_t pointscale : 1;
49 uint32_t vertexblend : 3;
50 uint32_t vertexblend_indexed : 1;
51 uint32_t vertextween : 1;
52 uint32_t mtl_diffuse : 2; /* 0 = material, 1 = color1, 2 = color2 */
53 uint32_t mtl_ambient : 2;
54 uint32_t mtl_specular : 2;
55 uint32_t mtl_emissive : 2;
56 uint32_t fog_mode : 2;
57 uint32_t fog_range : 1;
58 uint32_t color0in_one : 1;
59 uint32_t color1in_one : 1;
60 uint32_t fog : 1;
61 uint32_t pad1 : 7;
62 uint32_t tc_dim_input: 16; /* 8 * 2 bits */
63 uint32_t pad2 : 16;
64 uint32_t tc_dim_output: 24; /* 8 * 3 bits */
65 uint32_t pad3 : 8;
66 uint32_t tc_gen : 24; /* 8 * 3 bits */
67 uint32_t pad4 : 8;
68 uint32_t tc_idx : 24;
69 uint32_t pad5 : 8;
70 uint32_t pad6;
71 };
72 uint64_t value64[3]; /* don't forget to resize VertexShader9.ff_key */
73 uint32_t value32[6];
74 };
75 };
76
77 /* Texture stage state:
78 *
79 * COLOROP D3DTOP 5 bit
80 * ALPHAOP D3DTOP 5 bit
81 * COLORARG0 D3DTA 3 bit
82 * COLORARG1 D3DTA 3 bit
83 * COLORARG2 D3DTA 3 bit
84 * ALPHAARG0 D3DTA 3 bit
85 * ALPHAARG1 D3DTA 3 bit
86 * ALPHAARG2 D3DTA 3 bit
87 * RESULTARG D3DTA 1 bit (CURRENT:0 or TEMP:1)
88 * TEXCOORDINDEX 0 - 7 3 bit
89 * ===========================
90 * 32 bit per stage
91 */
92 struct nine_ff_ps_key
93 {
94 union {
95 struct {
96 struct {
97 uint32_t colorop : 5;
98 uint32_t alphaop : 5;
99 uint32_t colorarg0 : 3;
100 uint32_t colorarg1 : 3;
101 uint32_t colorarg2 : 3;
102 uint32_t alphaarg0 : 3;
103 uint32_t alphaarg1 : 3;
104 uint32_t alphaarg2 : 3;
105 uint32_t resultarg : 1; /* CURRENT:0 or TEMP:1 */
106 uint32_t textarget : 2; /* 1D/2D/3D/CUBE */
107 uint32_t pad : 1;
108 /* that's 32 bit exactly */
109 } ts[8];
110 uint32_t projected : 16;
111 uint32_t fog : 1; /* for vFog coming from VS */
112 uint32_t fog_mode : 2;
113 uint32_t specular : 1;
114 uint32_t pad1 : 12; /* 9 32-bit words with this */
115 uint8_t colorarg_b4[3];
116 uint8_t colorarg_b5[3];
117 uint8_t alphaarg_b4[3]; /* 11 32-bit words plus a byte */
118 uint8_t pad2[3];
119 };
120 uint64_t value64[6]; /* don't forget to resize PixelShader9.ff_key */
121 uint32_t value32[12];
122 };
123 };
124
125 static unsigned nine_ff_vs_key_hash(void *key)
126 {
127 struct nine_ff_vs_key *vs = key;
128 unsigned i;
129 uint32_t hash = vs->value32[0];
130 for (i = 1; i < Elements(vs->value32); ++i)
131 hash ^= vs->value32[i];
132 return hash;
133 }
134 static int nine_ff_vs_key_comp(void *key1, void *key2)
135 {
136 struct nine_ff_vs_key *a = (struct nine_ff_vs_key *)key1;
137 struct nine_ff_vs_key *b = (struct nine_ff_vs_key *)key2;
138
139 return memcmp(a->value64, b->value64, sizeof(a->value64));
140 }
141 static unsigned nine_ff_ps_key_hash(void *key)
142 {
143 struct nine_ff_ps_key *ps = key;
144 unsigned i;
145 uint32_t hash = ps->value32[0];
146 for (i = 1; i < Elements(ps->value32); ++i)
147 hash ^= ps->value32[i];
148 return hash;
149 }
150 static int nine_ff_ps_key_comp(void *key1, void *key2)
151 {
152 struct nine_ff_ps_key *a = (struct nine_ff_ps_key *)key1;
153 struct nine_ff_ps_key *b = (struct nine_ff_ps_key *)key2;
154
155 return memcmp(a->value64, b->value64, sizeof(a->value64));
156 }
157 static unsigned nine_ff_fvf_key_hash(void *key)
158 {
159 return *(DWORD *)key;
160 }
161 static int nine_ff_fvf_key_comp(void *key1, void *key2)
162 {
163 return *(DWORD *)key1 != *(DWORD *)key2;
164 }
165
166 static void nine_ff_prune_vs(struct NineDevice9 *);
167 static void nine_ff_prune_ps(struct NineDevice9 *);
168
169 static void nine_ureg_tgsi_dump(struct ureg_program *ureg, boolean override)
170 {
171 if (debug_get_bool_option("NINE_FF_DUMP", FALSE) || override) {
172 unsigned count;
173 const struct tgsi_token *toks = ureg_get_tokens(ureg, &count);
174 tgsi_dump(toks, 0);
175 ureg_free_tokens(toks);
176 }
177 }
178
179 #define _X(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_X)
180 #define _Y(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_Y)
181 #define _Z(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_Z)
182 #define _W(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_W)
183
184 #define _XXXX(r) ureg_scalar(r, TGSI_SWIZZLE_X)
185 #define _YYYY(r) ureg_scalar(r, TGSI_SWIZZLE_Y)
186 #define _ZZZZ(r) ureg_scalar(r, TGSI_SWIZZLE_Z)
187 #define _WWWW(r) ureg_scalar(r, TGSI_SWIZZLE_W)
188
189 #define _XYZW(r) (r)
190
191 /* AL should contain base address of lights table. */
192 #define LIGHT_CONST(i) \
193 ureg_src_indirect(ureg_DECL_constant(ureg, i), _X(AL))
194
195 #define MATERIAL_CONST(i) \
196 ureg_DECL_constant(ureg, 19 + (i))
197
198 #define _CONST(n) ureg_DECL_constant(ureg, n)
199
200 /* VS FF constants layout:
201 *
202 * CONST[ 0.. 3] D3DTS_WORLD * D3DTS_VIEW * D3DTS_PROJECTION
203 * CONST[ 4.. 7] D3DTS_WORLD * D3DTS_VIEW
204 * CONST[ 8..11] D3DTS_VIEW * D3DTS_PROJECTION
205 * CONST[12..15] D3DTS_VIEW
206 * CONST[16..18] Normal matrix
207 *
208 * CONST[19] MATERIAL.Emissive + Material.Ambient * RS.Ambient
209 * CONST[20] MATERIAL.Diffuse
210 * CONST[21] MATERIAL.Ambient
211 * CONST[22] MATERIAL.Specular
212 * CONST[23].x___ MATERIAL.Power
213 * CONST[24] MATERIAL.Emissive
214 * CONST[25] RS.Ambient
215 *
216 * CONST[26].x___ RS.PointSizeMin
217 * CONST[26]._y__ RS.PointSizeMax
218 * CONST[26].__z_ RS.PointSize
219 * CONST[26].___w RS.PointScaleA
220 * CONST[27].x___ RS.PointScaleB
221 * CONST[27]._y__ RS.PointScaleC
222 *
223 * CONST[28].x___ RS.FogEnd
224 * CONST[28]._y__ 1.0f / (RS.FogEnd - RS.FogStart)
225 * CONST[28].__z_ RS.FogDensity
226
227 * CONST[30].x___ TWEENFACTOR
228 *
229 * CONST[32].x___ LIGHT[0].Type
230 * CONST[32]._yzw LIGHT[0].Attenuation0,1,2
231 * CONST[33] LIGHT[0].Diffuse
232 * CONST[34] LIGHT[0].Specular
233 * CONST[35] LIGHT[0].Ambient
234 * CONST[36].xyz_ LIGHT[0].Position
235 * CONST[36].___w LIGHT[0].Range
236 * CONST[37].xyz_ LIGHT[0].Direction
237 * CONST[37].___w LIGHT[0].Falloff
238 * CONST[38].x___ cos(LIGHT[0].Theta / 2)
239 * CONST[38]._y__ cos(LIGHT[0].Phi / 2)
240 * CONST[38].__z_ 1.0f / (cos(LIGHT[0].Theta / 2) - cos(Light[0].Phi / 2))
241 * CONST[39].xyz_ LIGHT[0].HalfVector (for directional lights)
242 * CONST[39].___w 1 if this is the last active light, 0 if not
243 * CONST[40] LIGHT[1]
244 * CONST[48] LIGHT[2]
245 * CONST[56] LIGHT[3]
246 * CONST[64] LIGHT[4]
247 * CONST[72] LIGHT[5]
248 * CONST[80] LIGHT[6]
249 * CONST[88] LIGHT[7]
250 * NOTE: no lighting code is generated if there are no active lights
251 *
252 * CONST[100].x___ Viewport 2/width
253 * CONST[100]._y__ Viewport 2/height
254 * CONST[100].__z_ Viewport 1/(zmax - zmin)
255 * CONST[101].x___ Viewport x0
256 * CONST[101]._y__ Viewport y0
257 * CONST[101].__z_ Viewport z0
258 *
259 * CONST[128..131] D3DTS_TEXTURE0
260 * CONST[132..135] D3DTS_TEXTURE1
261 * CONST[136..139] D3DTS_TEXTURE2
262 * CONST[140..143] D3DTS_TEXTURE3
263 * CONST[144..147] D3DTS_TEXTURE4
264 * CONST[148..151] D3DTS_TEXTURE5
265 * CONST[152..155] D3DTS_TEXTURE6
266 * CONST[156..159] D3DTS_TEXTURE7
267 *
268 * CONST[224] D3DTS_WORLDMATRIX[0]
269 * CONST[228] D3DTS_WORLDMATRIX[1]
270 * ...
271 * CONST[252] D3DTS_WORLDMATRIX[7]
272 */
273 struct vs_build_ctx
274 {
275 struct ureg_program *ureg;
276 const struct nine_ff_vs_key *key;
277
278 uint16_t input[PIPE_MAX_ATTRIBS];
279 unsigned num_inputs;
280
281 struct ureg_src aVtx;
282 struct ureg_src aNrm;
283 struct ureg_src aCol[2];
284 struct ureg_src aTex[8];
285 struct ureg_src aPsz;
286 struct ureg_src aInd;
287 struct ureg_src aWgt;
288
289 struct ureg_src aVtx1; /* tweening */
290 struct ureg_src aNrm1;
291
292 struct ureg_src mtlA;
293 struct ureg_src mtlD;
294 struct ureg_src mtlS;
295 struct ureg_src mtlE;
296 };
297
298 static inline unsigned
299 get_texcoord_sn(struct pipe_screen *screen)
300 {
301 if (screen->get_param(screen, PIPE_CAP_TGSI_TEXCOORD))
302 return TGSI_SEMANTIC_TEXCOORD;
303 return TGSI_SEMANTIC_GENERIC;
304 }
305
306 static inline struct ureg_src
307 build_vs_add_input(struct vs_build_ctx *vs, uint16_t ndecl)
308 {
309 const unsigned i = vs->num_inputs++;
310 assert(i < PIPE_MAX_ATTRIBS);
311 vs->input[i] = ndecl;
312 return ureg_DECL_vs_input(vs->ureg, i);
313 }
314
315 /* NOTE: dst may alias src */
316 static inline void
317 ureg_normalize3(struct ureg_program *ureg,
318 struct ureg_dst dst, struct ureg_src src,
319 struct ureg_dst tmp)
320 {
321 #ifdef NINE_TGSI_LAZY_DEVS
322 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
323
324 ureg_DP3(ureg, tmp_x, src, src);
325 ureg_RSQ(ureg, tmp_x, _X(tmp));
326 ureg_MUL(ureg, dst, src, _X(tmp));
327 #else
328 ureg_NRM(ureg, dst, src);
329 #endif
330 }
331
332 static void *
333 nine_ff_build_vs(struct NineDevice9 *device, struct vs_build_ctx *vs)
334 {
335 const struct nine_ff_vs_key *key = vs->key;
336 struct ureg_program *ureg = ureg_create(TGSI_PROCESSOR_VERTEX);
337 struct ureg_dst oPos, oCol[2], oPsz, oFog;
338 struct ureg_dst rVtx, rNrm;
339 struct ureg_dst r[8];
340 struct ureg_dst AR;
341 struct ureg_dst tmp, tmp_x, tmp_y, tmp_z;
342 unsigned i, c;
343 unsigned label[32], l = 0;
344 unsigned num_r = 8;
345 boolean need_rNrm = key->lighting || key->pointscale;
346 boolean need_rVtx = key->lighting || key->fog_mode;
347 const unsigned texcoord_sn = get_texcoord_sn(device->screen);
348
349 vs->ureg = ureg;
350
351 /* Check which inputs we should transform. */
352 for (i = 0; i < 8 * 3; i += 3) {
353 switch ((key->tc_gen >> i) & 0x3) {
354 case NINED3DTSS_TCI_CAMERASPACENORMAL:
355 need_rNrm = TRUE;
356 break;
357 case NINED3DTSS_TCI_CAMERASPACEPOSITION:
358 need_rVtx = TRUE;
359 break;
360 case NINED3DTSS_TCI_CAMERASPACEREFLECTIONVECTOR:
361 need_rVtx = need_rNrm = TRUE;
362 break;
363 default:
364 break;
365 }
366 }
367
368 /* Declare and record used inputs (needed for linkage with vertex format):
369 * (texture coordinates handled later)
370 */
371 vs->aVtx = build_vs_add_input(vs,
372 key->position_t ? NINE_DECLUSAGE_POSITIONT : NINE_DECLUSAGE_POSITION);
373
374 if (need_rNrm)
375 vs->aNrm = build_vs_add_input(vs, NINE_DECLUSAGE_NORMAL);
376
377 vs->aCol[0] = ureg_imm1f(ureg, 1.0f);
378 vs->aCol[1] = ureg_imm1f(ureg, 1.0f);
379
380 if (key->lighting || key->darkness) {
381 const unsigned mask = key->mtl_diffuse | key->mtl_specular |
382 key->mtl_ambient | key->mtl_emissive;
383 if ((mask & 0x1) && !key->color0in_one)
384 vs->aCol[0] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 0));
385 if ((mask & 0x2) && !key->color1in_one)
386 vs->aCol[1] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 1));
387
388 vs->mtlD = MATERIAL_CONST(1);
389 vs->mtlA = MATERIAL_CONST(2);
390 vs->mtlS = MATERIAL_CONST(3);
391 vs->mtlE = MATERIAL_CONST(5);
392 if (key->mtl_diffuse == 1) vs->mtlD = vs->aCol[0]; else
393 if (key->mtl_diffuse == 2) vs->mtlD = vs->aCol[1];
394 if (key->mtl_ambient == 1) vs->mtlA = vs->aCol[0]; else
395 if (key->mtl_ambient == 2) vs->mtlA = vs->aCol[1];
396 if (key->mtl_specular == 1) vs->mtlS = vs->aCol[0]; else
397 if (key->mtl_specular == 2) vs->mtlS = vs->aCol[1];
398 if (key->mtl_emissive == 1) vs->mtlE = vs->aCol[0]; else
399 if (key->mtl_emissive == 2) vs->mtlE = vs->aCol[1];
400 } else {
401 if (!key->color0in_one) vs->aCol[0] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 0));
402 if (!key->color1in_one) vs->aCol[1] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 1));
403 }
404
405 if (key->vertexpointsize)
406 vs->aPsz = build_vs_add_input(vs, NINE_DECLUSAGE_PSIZE);
407
408 if (key->vertexblend_indexed)
409 vs->aInd = build_vs_add_input(vs, NINE_DECLUSAGE_BLENDINDICES);
410 if (key->vertexblend)
411 vs->aWgt = build_vs_add_input(vs, NINE_DECLUSAGE_BLENDWEIGHT);
412 if (key->vertextween) {
413 vs->aVtx1 = build_vs_add_input(vs, NINE_DECLUSAGE_i(POSITION,1));
414 vs->aNrm1 = build_vs_add_input(vs, NINE_DECLUSAGE_i(NORMAL,1));
415 }
416
417 /* Declare outputs:
418 */
419 oPos = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0); /* HPOS */
420 oCol[0] = ureg_saturate(ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0));
421 oCol[1] = ureg_saturate(ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 1));
422 if (key->fog) {
423 oFog = ureg_DECL_output(ureg, TGSI_SEMANTIC_FOG, 0);
424 oFog = ureg_writemask(oFog, TGSI_WRITEMASK_X);
425 }
426
427 if (key->vertexpointsize || key->pointscale) {
428 oPsz = ureg_DECL_output_masked(ureg, TGSI_SEMANTIC_PSIZE, 0,
429 TGSI_WRITEMASK_X, 0, 1);
430 oPsz = ureg_writemask(oPsz, TGSI_WRITEMASK_X);
431 }
432
433 /* Declare TEMPs:
434 */
435 for (i = 0; i < num_r; ++i)
436 r[i] = ureg_DECL_local_temporary(ureg);
437 tmp = r[0];
438 tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
439 tmp_y = ureg_writemask(tmp, TGSI_WRITEMASK_Y);
440 tmp_z = ureg_writemask(tmp, TGSI_WRITEMASK_Z);
441 if (key->lighting || key->vertexblend)
442 AR = ureg_DECL_address(ureg);
443
444 rVtx = ureg_writemask(r[1], TGSI_WRITEMASK_XYZ);
445 rNrm = ureg_writemask(r[2], TGSI_WRITEMASK_XYZ);
446
447 /* === Vertex transformation / vertex blending:
448 */
449 if (key->vertextween) {
450 assert(!key->vertexblend);
451 ureg_LRP(ureg, r[2], _XXXX(_CONST(30)), vs->aVtx, vs->aVtx1);
452 if (need_rNrm)
453 ureg_LRP(ureg, r[3], _XXXX(_CONST(30)), vs->aNrm, vs->aNrm1);
454 vs->aVtx = ureg_src(r[2]);
455 vs->aNrm = ureg_src(r[3]);
456 }
457
458 if (key->vertexblend) {
459 struct ureg_src cWM[4];
460
461 for (i = 224; i <= 255; ++i)
462 ureg_DECL_constant(ureg, i);
463
464 /* translate world matrix index to constant file index */
465 if (key->vertexblend_indexed) {
466 ureg_MAD(ureg, tmp, vs->aInd, ureg_imm1f(ureg, 4.0f), ureg_imm1f(ureg, 224.0f));
467 ureg_ARL(ureg, AR, ureg_src(tmp));
468 }
469 for (i = 0; i < key->vertexblend; ++i) {
470 for (c = 0; c < 4; ++c) {
471 cWM[c] = ureg_src_register(TGSI_FILE_CONSTANT, (224 + i * 4) * !key->vertexblend_indexed + c);
472 if (key->vertexblend_indexed)
473 cWM[c] = ureg_src_indirect(cWM[c], ureg_scalar(ureg_src(AR), i));
474 }
475 /* multiply by WORLD(index) */
476 ureg_MUL(ureg, r[0], _XXXX(vs->aVtx), cWM[0]);
477 ureg_MAD(ureg, r[0], _YYYY(vs->aVtx), cWM[1], ureg_src(r[0]));
478 ureg_MAD(ureg, r[0], _ZZZZ(vs->aVtx), cWM[2], ureg_src(r[0]));
479 ureg_MAD(ureg, r[0], _WWWW(vs->aVtx), cWM[3], ureg_src(r[0]));
480
481 /* accumulate weighted position value */
482 if (i)
483 ureg_MAD(ureg, r[2], ureg_src(r[0]), ureg_scalar(vs->aWgt, i), ureg_src(r[2]));
484 else
485 ureg_MUL(ureg, r[2], ureg_src(r[0]), ureg_scalar(vs->aWgt, 0));
486 }
487 /* multiply by VIEW_PROJ */
488 ureg_MUL(ureg, r[0], _X(r[2]), _CONST(8));
489 ureg_MAD(ureg, r[0], _Y(r[2]), _CONST(9), ureg_src(r[0]));
490 ureg_MAD(ureg, r[0], _Z(r[2]), _CONST(10), ureg_src(r[0]));
491 ureg_MAD(ureg, oPos, _W(r[2]), _CONST(11), ureg_src(r[0]));
492
493 if (need_rVtx)
494 vs->aVtx = ureg_src(r[2]);
495 } else
496 if (key->position_t && device->driver_caps.window_space_position_support) {
497 ureg_MOV(ureg, oPos, vs->aVtx);
498 } else if (key->position_t) {
499 /* vs->aVtx contains the coordinates buffer wise.
500 * later in the pipeline, clipping, viewport and division
501 * by w (rhw = 1/w) are going to be applied, so do the reverse
502 * of these transformations (except clipping) to have the good
503 * position at the end.*/
504 ureg_MOV(ureg, tmp, vs->aVtx);
505 /* X from [X_min, X_min + width] to [-1, 1], same for Y. Z to [0, 1] */
506 ureg_SUB(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _CONST(101));
507 ureg_MUL(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _CONST(100));
508 ureg_SUB(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XY), ureg_src(tmp), ureg_imm1f(ureg, 1.0f));
509 /* Y needs to be reversed */
510 ureg_MOV(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_Y), ureg_negate(ureg_src(tmp)));
511 /* inverse rhw */
512 ureg_RCP(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_W), _W(tmp));
513 /* multiply X, Y, Z by w */
514 ureg_MUL(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _W(tmp));
515 ureg_MOV(ureg, oPos, ureg_src(tmp));
516 } else {
517 /* position = vertex * WORLD_VIEW_PROJ */
518 ureg_MUL(ureg, r[0], _XXXX(vs->aVtx), _CONST(0));
519 ureg_MAD(ureg, r[0], _YYYY(vs->aVtx), _CONST(1), ureg_src(r[0]));
520 ureg_MAD(ureg, r[0], _ZZZZ(vs->aVtx), _CONST(2), ureg_src(r[0]));
521 ureg_MAD(ureg, oPos, _WWWW(vs->aVtx), _CONST(3), ureg_src(r[0]));
522 }
523
524 if (need_rVtx) {
525 ureg_MUL(ureg, rVtx, _XXXX(vs->aVtx), _CONST(4));
526 ureg_MAD(ureg, rVtx, _YYYY(vs->aVtx), _CONST(5), ureg_src(rVtx));
527 ureg_MAD(ureg, rVtx, _ZZZZ(vs->aVtx), _CONST(6), ureg_src(rVtx));
528 ureg_MAD(ureg, rVtx, _WWWW(vs->aVtx), _CONST(7), ureg_src(rVtx));
529 }
530 if (need_rNrm) {
531 ureg_MUL(ureg, rNrm, _XXXX(vs->aNrm), _CONST(16));
532 ureg_MAD(ureg, rNrm, _YYYY(vs->aNrm), _CONST(17), ureg_src(rNrm));
533 ureg_MAD(ureg, rNrm, _ZZZZ(vs->aNrm), _CONST(18), ureg_src(rNrm));
534 ureg_normalize3(ureg, rNrm, ureg_src(rNrm), tmp);
535 }
536 /* NOTE: don't use vs->aVtx, vs->aNrm after this line */
537
538 /* === Process point size:
539 */
540 if (key->vertexpointsize) {
541 struct ureg_src cPsz1 = ureg_DECL_constant(ureg, 26);
542 #ifdef NINE_TGSI_LAZY_DEVS
543 struct ureg_dst tmp_clamp = ureg_DECL_temporary(ureg);
544
545 ureg_MAX(ureg, tmp_clamp, vs->aPsz, _XXXX(cPsz1));
546 ureg_MIN(ureg, oPsz, ureg_src(tmp_clamp), _YYYY(cPsz1));
547 ureg_release_temporary(ureg, tmp_clamp);
548 #else
549 ureg_CLAMP(ureg, oPsz, vs->aPsz, _XXXX(cPsz1), _YYYY(cPsz1));
550 #endif
551 } else if (key->pointscale) {
552 struct ureg_src cPsz1 = ureg_DECL_constant(ureg, 26);
553 struct ureg_src cPsz2 = ureg_DECL_constant(ureg, 27);
554
555 ureg_DP3(ureg, tmp_x, ureg_src(r[1]), ureg_src(r[1]));
556 ureg_SQRT(ureg, tmp_y, _X(tmp));
557 ureg_MAD(ureg, tmp_x, _Y(tmp), _YYYY(cPsz2), _XXXX(cPsz2));
558 ureg_MAD(ureg, tmp_x, _Y(tmp), _X(tmp), _WWWW(cPsz1));
559 ureg_RCP(ureg, tmp_x, ureg_src(tmp));
560 ureg_MUL(ureg, tmp_x, ureg_src(tmp), _ZZZZ(cPsz1));
561 #ifdef NINE_TGSI_LAZY_DEVS
562 struct ureg_dst tmp_clamp = ureg_DECL_temporary(ureg);
563
564 ureg_MAX(ureg, tmp_clamp, _X(tmp), _XXXX(cPsz1));
565 ureg_MIN(ureg, oPsz, ureg_src(tmp_clamp), _YYYY(cPsz1));
566 ureg_release_temporary(ureg, tmp_clamp);
567 #else
568 ureg_CLAMP(ureg, oPsz, _X(tmp), _XXXX(cPsz1), _YYYY(cPsz1));
569 #endif
570 }
571
572 for (i = 0; i < 8; ++i) {
573 struct ureg_dst oTex, input_coord, transformed, t;
574 unsigned c, writemask;
575 const unsigned tci = (key->tc_gen >> (i * 3)) & 0x7;
576 const unsigned idx = (key->tc_idx >> (i * 3)) & 0x7;
577 unsigned dim_input = 1 + ((key->tc_dim_input >> (i * 2)) & 0x3);
578 const unsigned dim_output = (key->tc_dim_output >> (i * 3)) & 0x7;
579
580 /* No texture output of index s */
581 if (tci == NINED3DTSS_TCI_DISABLE)
582 continue;
583 oTex = ureg_DECL_output(ureg, texcoord_sn, i);
584 input_coord = r[5];
585 transformed = r[6];
586
587 /* Get the coordinate */
588 switch (tci) {
589 case NINED3DTSS_TCI_PASSTHRU:
590 /* NINED3DTSS_TCI_PASSTHRU => Use texcoord coming from index idx *
591 * Else the idx is used only to determine wrapping mode. */
592 vs->aTex[idx] = build_vs_add_input(vs, NINE_DECLUSAGE_i(TEXCOORD,idx));
593 ureg_MOV(ureg, input_coord, vs->aTex[idx]);
594 break;
595 case NINED3DTSS_TCI_CAMERASPACENORMAL:
596 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XYZ), ureg_src(rNrm));
597 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
598 dim_input = 4;
599 break;
600 case NINED3DTSS_TCI_CAMERASPACEPOSITION:
601 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XYZ), ureg_src(rVtx));
602 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
603 dim_input = 4;
604 break;
605 case NINED3DTSS_TCI_CAMERASPACEREFLECTIONVECTOR:
606 tmp.WriteMask = TGSI_WRITEMASK_XYZ;
607 ureg_DP3(ureg, tmp_x, ureg_src(rVtx), ureg_src(rNrm));
608 ureg_MUL(ureg, tmp, ureg_src(rNrm), _X(tmp));
609 ureg_ADD(ureg, tmp, ureg_src(tmp), ureg_src(tmp));
610 ureg_SUB(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XYZ), ureg_src(rVtx), ureg_src(tmp));
611 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
612 dim_input = 4;
613 tmp.WriteMask = TGSI_WRITEMASK_XYZW;
614 break;
615 case NINED3DTSS_TCI_SPHEREMAP:
616 assert(!"TODO");
617 break;
618 default:
619 assert(0);
620 break;
621 }
622
623 /* Apply the transformation */
624 /* dim_output == 0 => do not transform the components.
625 * XYZRHW also disables transformation */
626 if (!dim_output || key->position_t) {
627 transformed = input_coord;
628 writemask = TGSI_WRITEMASK_XYZW;
629 } else {
630 for (c = 0; c < dim_output; c++) {
631 t = ureg_writemask(transformed, 1 << c);
632 switch (dim_input) {
633 /* dim_input = 1 2 3: -> we add trailing 1 to input*/
634 case 1: ureg_MAD(ureg, t, _X(input_coord), _XXXX(_CONST(128 + i * 4 + c)), _YYYY(_CONST(128 + i * 4 + c)));
635 break;
636 case 2: ureg_DP2(ureg, t, ureg_src(input_coord), _CONST(128 + i * 4 + c));
637 ureg_ADD(ureg, t, ureg_src(transformed), _ZZZZ(_CONST(128 + i * 4 + c)));
638 break;
639 case 3: ureg_DP3(ureg, t, ureg_src(input_coord), _CONST(128 + i * 4 + c));
640 ureg_ADD(ureg, t, ureg_src(transformed), _WWWW(_CONST(128 + i * 4 + c)));
641 break;
642 case 4: ureg_DP4(ureg, t, ureg_src(input_coord), _CONST(128 + i * 4 + c)); break;
643 default:
644 assert(0);
645 }
646 }
647 writemask = (1 << dim_output) - 1;
648 }
649
650 ureg_MOV(ureg, ureg_writemask(oTex, writemask), ureg_src(transformed));
651 }
652
653 /* === Lighting:
654 *
655 * DIRECTIONAL: Light at infinite distance, parallel rays, no attenuation.
656 * POINT: Finite distance to scene, divergent rays, isotropic, attenuation.
657 * SPOT: Finite distance, divergent rays, angular dependence, attenuation.
658 *
659 * vec3 normal = normalize(in.Normal * NormalMatrix);
660 * vec3 hitDir = light.direction;
661 * float atten = 1.0;
662 *
663 * if (light.type != DIRECTIONAL)
664 * {
665 * vec3 hitVec = light.position - eyeVertex;
666 * float d = length(hitVec);
667 * hitDir = hitVec / d;
668 * atten = 1 / ((light.atten2 * d + light.atten1) * d + light.atten0);
669 * }
670 *
671 * if (light.type == SPOTLIGHT)
672 * {
673 * float rho = dp3(-hitVec, light.direction);
674 * if (rho < cos(light.phi / 2))
675 * atten = 0;
676 * if (rho < cos(light.theta / 2))
677 * atten *= pow(some_func(rho), light.falloff);
678 * }
679 *
680 * float nDotHit = dp3_sat(normal, hitVec);
681 * float powFact = 0.0;
682 *
683 * if (nDotHit > 0.0)
684 * {
685 * vec3 midVec = normalize(hitDir + eye);
686 * float nDotMid = dp3_sat(normal, midVec);
687 * pFact = pow(nDotMid, material.power);
688 * }
689 *
690 * ambient += light.ambient * atten;
691 * diffuse += light.diffuse * atten * nDotHit;
692 * specular += light.specular * atten * powFact;
693 */
694 if (key->lighting) {
695 struct ureg_dst rAtt = ureg_writemask(r[1], TGSI_WRITEMASK_W);
696 struct ureg_dst rHit = ureg_writemask(r[3], TGSI_WRITEMASK_XYZ);
697 struct ureg_dst rMid = ureg_writemask(r[4], TGSI_WRITEMASK_XYZ);
698
699 struct ureg_dst rCtr = ureg_writemask(r[2], TGSI_WRITEMASK_W);
700
701 struct ureg_dst AL = ureg_writemask(AR, TGSI_WRITEMASK_X);
702
703 /* Light.*.Alpha is not used. */
704 struct ureg_dst rD = ureg_writemask(r[5], TGSI_WRITEMASK_XYZ);
705 struct ureg_dst rA = ureg_writemask(r[6], TGSI_WRITEMASK_XYZ);
706 struct ureg_dst rS = ureg_writemask(r[7], TGSI_WRITEMASK_XYZ);
707
708 struct ureg_src mtlP = _XXXX(MATERIAL_CONST(4));
709
710 struct ureg_src cLKind = _XXXX(LIGHT_CONST(0));
711 struct ureg_src cLAtt0 = _YYYY(LIGHT_CONST(0));
712 struct ureg_src cLAtt1 = _ZZZZ(LIGHT_CONST(0));
713 struct ureg_src cLAtt2 = _WWWW(LIGHT_CONST(0));
714 struct ureg_src cLColD = _XYZW(LIGHT_CONST(1));
715 struct ureg_src cLColS = _XYZW(LIGHT_CONST(2));
716 struct ureg_src cLColA = _XYZW(LIGHT_CONST(3));
717 struct ureg_src cLPos = _XYZW(LIGHT_CONST(4));
718 struct ureg_src cLRng = _WWWW(LIGHT_CONST(4));
719 struct ureg_src cLDir = _XYZW(LIGHT_CONST(5));
720 struct ureg_src cLFOff = _WWWW(LIGHT_CONST(5));
721 struct ureg_src cLTht = _XXXX(LIGHT_CONST(6));
722 struct ureg_src cLPhi = _YYYY(LIGHT_CONST(6));
723 struct ureg_src cLSDiv = _ZZZZ(LIGHT_CONST(6));
724 struct ureg_src cLLast = _WWWW(LIGHT_CONST(7));
725
726 const unsigned loop_label = l++;
727
728 ureg_MOV(ureg, rCtr, ureg_imm1f(ureg, 32.0f)); /* &lightconst(0) */
729 ureg_MOV(ureg, rD, ureg_imm1f(ureg, 0.0f));
730 ureg_MOV(ureg, rA, ureg_imm1f(ureg, 0.0f));
731 ureg_MOV(ureg, rS, ureg_imm1f(ureg, 0.0f));
732 rD = ureg_saturate(rD);
733 rA = ureg_saturate(rA);
734 rS = ureg_saturate(rS);
735
736
737 /* loop management */
738 ureg_BGNLOOP(ureg, &label[loop_label]);
739 ureg_ARL(ureg, AL, _W(rCtr));
740
741 /* if (not DIRECTIONAL light): */
742 ureg_SNE(ureg, tmp_x, cLKind, ureg_imm1f(ureg, D3DLIGHT_DIRECTIONAL));
743 ureg_MOV(ureg, rHit, ureg_negate(cLDir));
744 ureg_MOV(ureg, rAtt, ureg_imm1f(ureg, 1.0f));
745 ureg_IF(ureg, _X(tmp), &label[l++]);
746 {
747 /* hitDir = light.position - eyeVtx
748 * d = length(hitDir)
749 * hitDir /= d
750 */
751 ureg_SUB(ureg, rHit, cLPos, ureg_src(rVtx));
752 ureg_DP3(ureg, tmp_x, ureg_src(rHit), ureg_src(rHit));
753 ureg_RSQ(ureg, tmp_y, _X(tmp));
754 ureg_MUL(ureg, rHit, ureg_src(rHit), _Y(tmp)); /* normalize */
755 ureg_MUL(ureg, tmp_x, _X(tmp), _Y(tmp)); /* length */
756
757 /* att = 1.0 / (light.att0 + (light.att1 + light.att2 * d) * d) */
758 ureg_MAD(ureg, rAtt, _X(tmp), cLAtt2, cLAtt1);
759 ureg_MAD(ureg, rAtt, _X(tmp), _W(rAtt), cLAtt0);
760 ureg_RCP(ureg, rAtt, _W(rAtt));
761 /* cut-off if distance exceeds Light.Range */
762 ureg_SLT(ureg, tmp_x, _X(tmp), cLRng);
763 ureg_MUL(ureg, rAtt, _W(rAtt), _X(tmp));
764 }
765 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
766 ureg_ENDIF(ureg);
767
768 /* if (SPOT light) */
769 ureg_SEQ(ureg, tmp_x, cLKind, ureg_imm1f(ureg, D3DLIGHT_SPOT));
770 ureg_IF(ureg, _X(tmp), &label[l++]);
771 {
772 /* rho = dp3(-hitDir, light.spotDir)
773 *
774 * if (rho > light.ctht2) NOTE: 0 <= phi <= pi, 0 <= theta <= phi
775 * spotAtt = 1
776 * else
777 * if (rho <= light.cphi2)
778 * spotAtt = 0
779 * else
780 * spotAtt = (rho - light.cphi2) / (light.ctht2 - light.cphi2) ^ light.falloff
781 */
782 ureg_DP3(ureg, tmp_y, ureg_negate(ureg_src(rHit)), cLDir); /* rho */
783 ureg_SUB(ureg, tmp_x, _Y(tmp), cLPhi);
784 ureg_MUL(ureg, tmp_x, _X(tmp), cLSDiv);
785 ureg_POW(ureg, tmp_x, _X(tmp), cLFOff); /* spotAtten */
786 ureg_SGE(ureg, tmp_z, _Y(tmp), cLTht); /* if inside theta && phi */
787 ureg_SGE(ureg, tmp_y, _Y(tmp), cLPhi); /* if inside phi */
788 ureg_MAD(ureg, ureg_saturate(tmp_x), _X(tmp), _Y(tmp), _Z(tmp));
789 ureg_MUL(ureg, rAtt, _W(rAtt), _X(tmp));
790 }
791 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
792 ureg_ENDIF(ureg);
793
794 /* directional factors, let's not use LIT because of clarity */
795 ureg_DP3(ureg, ureg_saturate(tmp_x), ureg_src(rNrm), ureg_src(rHit));
796 ureg_MOV(ureg, tmp_y, ureg_imm1f(ureg, 0.0f));
797 ureg_IF(ureg, _X(tmp), &label[l++]);
798 {
799 /* midVec = normalize(hitDir + eyeDir) */
800 if (key->localviewer) {
801 ureg_normalize3(ureg, rMid, ureg_src(rVtx), tmp);
802 ureg_ADD(ureg, rMid, ureg_src(rHit), ureg_negate(ureg_src(rMid)));
803 } else {
804 ureg_ADD(ureg, rMid, ureg_src(rHit), ureg_imm3f(ureg, 0.0f, 0.0f, 1.0f));
805 }
806 ureg_normalize3(ureg, rMid, ureg_src(rMid), tmp);
807 ureg_DP3(ureg, ureg_saturate(tmp_y), ureg_src(rNrm), ureg_src(rMid));
808 ureg_POW(ureg, tmp_y, _Y(tmp), mtlP);
809
810 ureg_MUL(ureg, tmp_x, _W(rAtt), _X(tmp)); /* dp3(normal,hitDir) * att */
811 ureg_MUL(ureg, tmp_y, _W(rAtt), _Y(tmp)); /* power factor * att */
812 ureg_MAD(ureg, rD, cLColD, _X(tmp), ureg_src(rD)); /* accumulate diffuse */
813 ureg_MAD(ureg, rS, cLColS, _Y(tmp), ureg_src(rS)); /* accumulate specular */
814 }
815 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
816 ureg_ENDIF(ureg);
817
818 ureg_MAD(ureg, rA, cLColA, _W(rAtt), ureg_src(rA)); /* accumulate ambient */
819
820 /* break if this was the last light */
821 ureg_IF(ureg, cLLast, &label[l++]);
822 ureg_BRK(ureg);
823 ureg_ENDIF(ureg);
824 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
825
826 ureg_ADD(ureg, rCtr, _W(rCtr), ureg_imm1f(ureg, 8.0f));
827 ureg_fixup_label(ureg, label[loop_label], ureg_get_instruction_number(ureg));
828 ureg_ENDLOOP(ureg, &label[loop_label]);
829
830 /* Set alpha factors of illumination to 1.0 for the multiplications. */
831 rD.WriteMask = TGSI_WRITEMASK_W; rD.Saturate = 0;
832 rS.WriteMask = TGSI_WRITEMASK_W; rS.Saturate = 0;
833 rA.WriteMask = TGSI_WRITEMASK_W; rA.Saturate = 0;
834 ureg_MOV(ureg, rD, ureg_imm1f(ureg, 1.0f));
835 ureg_MOV(ureg, rS, ureg_imm1f(ureg, 1.0f));
836
837 /* Apply to material:
838 *
839 * oCol[0] = (material.emissive + material.ambient * rs.ambient) +
840 * material.ambient * ambient +
841 * material.diffuse * diffuse +
842 * oCol[1] = material.specular * specular;
843 */
844 if (key->mtl_emissive == 0 && key->mtl_ambient == 0) {
845 ureg_MOV(ureg, rA, ureg_imm1f(ureg, 1.0f));
846 ureg_MAD(ureg, tmp, ureg_src(rA), vs->mtlA, _CONST(19));
847 } else {
848 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(rA), _CONST(25));
849 ureg_MAD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), vs->mtlA, ureg_src(tmp), vs->mtlE);
850 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_W ), vs->mtlA, vs->mtlE);
851 }
852 ureg_MAD(ureg, oCol[0], ureg_src(rD), vs->mtlD, ureg_src(tmp));
853 ureg_MUL(ureg, oCol[1], ureg_src(rS), vs->mtlS);
854 } else
855 /* COLOR */
856 if (key->darkness) {
857 if (key->mtl_emissive == 0 && key->mtl_ambient == 0) {
858 ureg_MAD(ureg, oCol[0], vs->mtlD, ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 1.0f), _CONST(19));
859 } else {
860 ureg_MAD(ureg, ureg_writemask(oCol[0], TGSI_WRITEMASK_XYZ), vs->mtlA, _CONST(25), vs->mtlE);
861 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_W), vs->mtlA, vs->mtlE);
862 ureg_ADD(ureg, ureg_writemask(oCol[0], TGSI_WRITEMASK_W), vs->mtlD, _W(tmp));
863 }
864 ureg_MUL(ureg, oCol[1], ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 1.0f), vs->mtlS);
865 } else {
866 ureg_MOV(ureg, oCol[0], vs->aCol[0]);
867 ureg_MOV(ureg, oCol[1], vs->aCol[1]);
868 }
869
870 /* === Process fog.
871 *
872 * exp(x) = ex2(log2(e) * x)
873 */
874 if (key->fog_mode) {
875 if (key->position_t) {
876 ureg_MOV(ureg, ureg_saturate(tmp_x), ureg_scalar(vs->aCol[1], TGSI_SWIZZLE_W));
877 } else
878 if (key->fog_range) {
879 ureg_DP3(ureg, tmp_x, ureg_src(rVtx), ureg_src(rVtx));
880 ureg_RSQ(ureg, tmp_z, _X(tmp));
881 ureg_MUL(ureg, tmp_z, _Z(tmp), _X(tmp));
882 } else {
883 ureg_MOV(ureg, tmp_z, ureg_abs(_Z(rVtx)));
884 }
885
886 if (key->fog_mode == D3DFOG_EXP) {
887 ureg_MUL(ureg, tmp_x, _Z(tmp), _ZZZZ(_CONST(28)));
888 ureg_MUL(ureg, tmp_x, _X(tmp), ureg_imm1f(ureg, -1.442695f));
889 ureg_EX2(ureg, tmp_x, _X(tmp));
890 } else
891 if (key->fog_mode == D3DFOG_EXP2) {
892 ureg_MUL(ureg, tmp_x, _Z(tmp), _ZZZZ(_CONST(28)));
893 ureg_MUL(ureg, tmp_x, _X(tmp), _X(tmp));
894 ureg_MUL(ureg, tmp_x, _X(tmp), ureg_imm1f(ureg, -1.442695f));
895 ureg_EX2(ureg, tmp_x, _X(tmp));
896 } else
897 if (key->fog_mode == D3DFOG_LINEAR && !key->position_t) {
898 ureg_SUB(ureg, tmp_x, _XXXX(_CONST(28)), _Z(tmp));
899 ureg_MUL(ureg, ureg_saturate(tmp_x), _X(tmp), _YYYY(_CONST(28)));
900 }
901 ureg_MOV(ureg, oFog, _X(tmp));
902 } else if (key->fog) {
903 ureg_MOV(ureg, oFog, ureg_scalar(vs->aCol[1], TGSI_SWIZZLE_W));
904 }
905
906 if (key->position_t && device->driver_caps.window_space_position_support)
907 ureg_property(ureg, TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION, TRUE);
908
909 ureg_END(ureg);
910 nine_ureg_tgsi_dump(ureg, FALSE);
911 return ureg_create_shader_and_destroy(ureg, device->pipe);
912 }
913
914 /* PS FF constants layout:
915 *
916 * CONST[ 0.. 7] stage[i].D3DTSS_CONSTANT
917 * CONST[ 8..15].x___ stage[i].D3DTSS_BUMPENVMAT00
918 * CONST[ 8..15]._y__ stage[i].D3DTSS_BUMPENVMAT01
919 * CONST[ 8..15].__z_ stage[i].D3DTSS_BUMPENVMAT10
920 * CONST[ 8..15].___w stage[i].D3DTSS_BUMPENVMAT11
921 * CONST[16..19].x_z_ stage[i].D3DTSS_BUMPENVLSCALE
922 * CONST[17..19]._y_w stage[i].D3DTSS_BUMPENVLOFFSET
923 *
924 * CONST[20] D3DRS_TEXTUREFACTOR
925 * CONST[21] D3DRS_FOGCOLOR
926 * CONST[22].x___ RS.FogEnd
927 * CONST[22]._y__ 1.0f / (RS.FogEnd - RS.FogStart)
928 * CONST[22].__z_ RS.FogDensity
929 */
930 struct ps_build_ctx
931 {
932 struct ureg_program *ureg;
933
934 struct ureg_src vC[2]; /* DIFFUSE, SPECULAR */
935 struct ureg_src vT[8]; /* TEXCOORD[i] */
936 struct ureg_dst r[6]; /* TEMPs */
937 struct ureg_dst rCur; /* D3DTA_CURRENT */
938 struct ureg_dst rMod;
939 struct ureg_src rCurSrc;
940 struct ureg_dst rTmp; /* D3DTA_TEMP */
941 struct ureg_src rTmpSrc;
942 struct ureg_dst rTex;
943 struct ureg_src rTexSrc;
944 struct ureg_src cBEM[8];
945 struct ureg_src s[8];
946
947 struct {
948 unsigned index;
949 unsigned index_pre_mod;
950 unsigned num_regs;
951 } stage;
952 };
953
954 static struct ureg_src
955 ps_get_ts_arg(struct ps_build_ctx *ps, unsigned ta)
956 {
957 struct ureg_src reg;
958
959 switch (ta & D3DTA_SELECTMASK) {
960 case D3DTA_CONSTANT:
961 reg = ureg_DECL_constant(ps->ureg, ps->stage.index);
962 break;
963 case D3DTA_CURRENT:
964 reg = (ps->stage.index == ps->stage.index_pre_mod) ? ureg_src(ps->rMod) : ps->rCurSrc;
965 break;
966 case D3DTA_DIFFUSE:
967 reg = ureg_DECL_fs_input(ps->ureg, TGSI_SEMANTIC_COLOR, 0, TGSI_INTERPOLATE_PERSPECTIVE);
968 break;
969 case D3DTA_SPECULAR:
970 reg = ureg_DECL_fs_input(ps->ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_PERSPECTIVE);
971 break;
972 case D3DTA_TEMP:
973 reg = ps->rTmpSrc;
974 break;
975 case D3DTA_TEXTURE:
976 reg = ps->rTexSrc;
977 break;
978 case D3DTA_TFACTOR:
979 reg = ureg_DECL_constant(ps->ureg, 20);
980 break;
981 default:
982 assert(0);
983 reg = ureg_src_undef();
984 break;
985 }
986 if (ta & D3DTA_COMPLEMENT) {
987 struct ureg_dst dst = ps->r[ps->stage.num_regs++];
988 ureg_SUB(ps->ureg, dst, ureg_imm1f(ps->ureg, 1.0f), reg);
989 reg = ureg_src(dst);
990 }
991 if (ta & D3DTA_ALPHAREPLICATE)
992 reg = _WWWW(reg);
993 return reg;
994 }
995
996 static struct ureg_dst
997 ps_get_ts_dst(struct ps_build_ctx *ps, unsigned ta)
998 {
999 assert(!(ta & (D3DTA_COMPLEMENT | D3DTA_ALPHAREPLICATE)));
1000
1001 switch (ta & D3DTA_SELECTMASK) {
1002 case D3DTA_CURRENT:
1003 return ps->rCur;
1004 case D3DTA_TEMP:
1005 return ps->rTmp;
1006 default:
1007 assert(0);
1008 return ureg_dst_undef();
1009 }
1010 }
1011
1012 static uint8_t ps_d3dtop_args_mask(D3DTEXTUREOP top)
1013 {
1014 switch (top) {
1015 case D3DTOP_DISABLE:
1016 return 0x0;
1017 case D3DTOP_SELECTARG1:
1018 case D3DTOP_PREMODULATE:
1019 return 0x2;
1020 case D3DTOP_SELECTARG2:
1021 return 0x4;
1022 case D3DTOP_MULTIPLYADD:
1023 case D3DTOP_LERP:
1024 return 0x7;
1025 default:
1026 return 0x6;
1027 }
1028 }
1029
1030 static inline boolean
1031 is_MOV_no_op(struct ureg_dst dst, struct ureg_src src)
1032 {
1033 return !dst.WriteMask ||
1034 (dst.File == src.File &&
1035 dst.Index == src.Index &&
1036 !dst.Indirect &&
1037 !dst.Saturate &&
1038 !src.Indirect &&
1039 !src.Negate &&
1040 !src.Absolute &&
1041 (!(dst.WriteMask & TGSI_WRITEMASK_X) || (src.SwizzleX == TGSI_SWIZZLE_X)) &&
1042 (!(dst.WriteMask & TGSI_WRITEMASK_Y) || (src.SwizzleY == TGSI_SWIZZLE_Y)) &&
1043 (!(dst.WriteMask & TGSI_WRITEMASK_Z) || (src.SwizzleZ == TGSI_SWIZZLE_Z)) &&
1044 (!(dst.WriteMask & TGSI_WRITEMASK_W) || (src.SwizzleW == TGSI_SWIZZLE_W)));
1045
1046 }
1047
1048 static void
1049 ps_do_ts_op(struct ps_build_ctx *ps, unsigned top, struct ureg_dst dst, struct ureg_src *arg)
1050 {
1051 struct ureg_program *ureg = ps->ureg;
1052 struct ureg_dst tmp = ps->r[ps->stage.num_regs];
1053 struct ureg_dst tmp2 = ps->r[ps->stage.num_regs+1];
1054 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
1055
1056 tmp.WriteMask = dst.WriteMask;
1057
1058 if (top != D3DTOP_SELECTARG1 && top != D3DTOP_SELECTARG2 &&
1059 top != D3DTOP_MODULATE && top != D3DTOP_PREMODULATE &&
1060 top != D3DTOP_BLENDDIFFUSEALPHA && top != D3DTOP_BLENDTEXTUREALPHA &&
1061 top != D3DTOP_BLENDFACTORALPHA && top != D3DTOP_BLENDCURRENTALPHA &&
1062 top != D3DTOP_BUMPENVMAP && top != D3DTOP_BUMPENVMAPLUMINANCE &&
1063 top != D3DTOP_LERP)
1064 dst = ureg_saturate(dst);
1065
1066 switch (top) {
1067 case D3DTOP_SELECTARG1:
1068 if (!is_MOV_no_op(dst, arg[1]))
1069 ureg_MOV(ureg, dst, arg[1]);
1070 break;
1071 case D3DTOP_SELECTARG2:
1072 if (!is_MOV_no_op(dst, arg[2]))
1073 ureg_MOV(ureg, dst, arg[2]);
1074 break;
1075 case D3DTOP_MODULATE:
1076 ureg_MUL(ureg, dst, arg[1], arg[2]);
1077 break;
1078 case D3DTOP_MODULATE2X:
1079 ureg_MUL(ureg, tmp, arg[1], arg[2]);
1080 ureg_ADD(ureg, dst, ureg_src(tmp), ureg_src(tmp));
1081 break;
1082 case D3DTOP_MODULATE4X:
1083 ureg_MUL(ureg, tmp, arg[1], arg[2]);
1084 ureg_MUL(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 4.0f));
1085 break;
1086 case D3DTOP_ADD:
1087 ureg_ADD(ureg, dst, arg[1], arg[2]);
1088 break;
1089 case D3DTOP_ADDSIGNED:
1090 ureg_ADD(ureg, tmp, arg[1], arg[2]);
1091 ureg_SUB(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 0.5f));
1092 break;
1093 case D3DTOP_ADDSIGNED2X:
1094 ureg_ADD(ureg, tmp, arg[1], arg[2]);
1095 ureg_MAD(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 2.0f), ureg_imm1f(ureg, -1.0f));
1096 break;
1097 case D3DTOP_SUBTRACT:
1098 ureg_SUB(ureg, dst, arg[1], arg[2]);
1099 break;
1100 case D3DTOP_ADDSMOOTH:
1101 ureg_SUB(ureg, tmp, ureg_imm1f(ureg, 1.0f), arg[1]);
1102 ureg_MAD(ureg, dst, ureg_src(tmp), arg[2], arg[1]);
1103 break;
1104 case D3DTOP_BLENDDIFFUSEALPHA:
1105 ureg_LRP(ureg, dst, _WWWW(ps->vC[0]), arg[1], arg[2]);
1106 break;
1107 case D3DTOP_BLENDTEXTUREALPHA:
1108 /* XXX: alpha taken from previous stage, texture or result ? */
1109 ureg_LRP(ureg, dst, _W(ps->rTex), arg[1], arg[2]);
1110 break;
1111 case D3DTOP_BLENDFACTORALPHA:
1112 ureg_LRP(ureg, dst, _WWWW(_CONST(20)), arg[1], arg[2]);
1113 break;
1114 case D3DTOP_BLENDTEXTUREALPHAPM:
1115 ureg_SUB(ureg, tmp_x, ureg_imm1f(ureg, 1.0f), _W(ps->rTex));
1116 ureg_MAD(ureg, dst, arg[2], _X(tmp), arg[1]);
1117 break;
1118 case D3DTOP_BLENDCURRENTALPHA:
1119 ureg_LRP(ureg, dst, _WWWW(ps->rCurSrc), arg[1], arg[2]);
1120 break;
1121 case D3DTOP_PREMODULATE:
1122 ureg_MOV(ureg, dst, arg[1]);
1123 ps->stage.index_pre_mod = ps->stage.index + 1;
1124 break;
1125 case D3DTOP_MODULATEALPHA_ADDCOLOR:
1126 ureg_MAD(ureg, dst, _WWWW(arg[1]), arg[2], arg[1]);
1127 break;
1128 case D3DTOP_MODULATECOLOR_ADDALPHA:
1129 ureg_MAD(ureg, dst, arg[1], arg[2], _WWWW(arg[1]));
1130 break;
1131 case D3DTOP_MODULATEINVALPHA_ADDCOLOR:
1132 ureg_SUB(ureg, tmp_x, ureg_imm1f(ureg, 1.0f), _WWWW(arg[1]));
1133 ureg_MAD(ureg, dst, _X(tmp), arg[2], arg[1]);
1134 break;
1135 case D3DTOP_MODULATEINVCOLOR_ADDALPHA:
1136 ureg_SUB(ureg, tmp, ureg_imm1f(ureg, 1.0f), arg[1]);
1137 ureg_MAD(ureg, dst, ureg_src(tmp), arg[2], _WWWW(arg[1]));
1138 break;
1139 case D3DTOP_BUMPENVMAP:
1140 break;
1141 case D3DTOP_BUMPENVMAPLUMINANCE:
1142 break;
1143 case D3DTOP_DOTPRODUCT3:
1144 ureg_SUB(ureg, tmp, arg[1], ureg_imm4f(ureg,0.5,0.5,0.5,0.5));
1145 ureg_SUB(ureg, tmp2, arg[2] , ureg_imm4f(ureg,0.5,0.5,0.5,0.5));
1146 ureg_DP3(ureg, tmp, ureg_src(tmp), ureg_src(tmp2));
1147 ureg_MUL(ureg, ureg_saturate(dst), ureg_src(tmp), ureg_imm4f(ureg,4.0,4.0,4.0,4.0));
1148 break;
1149 case D3DTOP_MULTIPLYADD:
1150 ureg_MAD(ureg, dst, arg[1], arg[2], arg[0]);
1151 break;
1152 case D3DTOP_LERP:
1153 ureg_LRP(ureg, dst, arg[0], arg[1], arg[2]);
1154 break;
1155 case D3DTOP_DISABLE:
1156 /* no-op ? */
1157 break;
1158 default:
1159 assert(!"invalid D3DTOP");
1160 break;
1161 }
1162 }
1163
1164 static void *
1165 nine_ff_build_ps(struct NineDevice9 *device, struct nine_ff_ps_key *key)
1166 {
1167 struct ps_build_ctx ps;
1168 struct ureg_program *ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT);
1169 struct ureg_dst oCol;
1170 unsigned i, s;
1171 const unsigned texcoord_sn = get_texcoord_sn(device->screen);
1172
1173 memset(&ps, 0, sizeof(ps));
1174 ps.ureg = ureg;
1175 ps.stage.index_pre_mod = -1;
1176
1177 ps.vC[0] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 0, TGSI_INTERPOLATE_PERSPECTIVE);
1178
1179 /* Declare all TEMPs we might need, serious drivers have a register allocator. */
1180 for (i = 0; i < Elements(ps.r); ++i)
1181 ps.r[i] = ureg_DECL_local_temporary(ureg);
1182 ps.rCur = ps.r[0];
1183 ps.rTmp = ps.r[1];
1184 ps.rTex = ps.r[2];
1185 ps.rCurSrc = ureg_src(ps.rCur);
1186 ps.rTmpSrc = ureg_src(ps.rTmp);
1187 ps.rTexSrc = ureg_src(ps.rTex);
1188
1189 for (s = 0; s < 8; ++s) {
1190 ps.s[s] = ureg_src_undef();
1191
1192 if (key->ts[s].colorop != D3DTOP_DISABLE) {
1193 if (key->ts[s].colorarg0 == D3DTA_SPECULAR ||
1194 key->ts[s].colorarg1 == D3DTA_SPECULAR ||
1195 key->ts[s].colorarg2 == D3DTA_SPECULAR)
1196 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_PERSPECTIVE);
1197
1198 if (key->ts[s].colorarg0 == D3DTA_TEXTURE ||
1199 key->ts[s].colorarg1 == D3DTA_TEXTURE ||
1200 key->ts[s].colorarg2 == D3DTA_TEXTURE) {
1201 ps.s[s] = ureg_DECL_sampler(ureg, s);
1202 ps.vT[s] = ureg_DECL_fs_input(ureg, texcoord_sn, s, TGSI_INTERPOLATE_PERSPECTIVE);
1203 }
1204 if (s && (key->ts[s - 1].colorop == D3DTOP_PREMODULATE ||
1205 key->ts[s - 1].alphaop == D3DTOP_PREMODULATE))
1206 ps.s[s] = ureg_DECL_sampler(ureg, s);
1207 }
1208
1209 if (key->ts[s].alphaop != D3DTOP_DISABLE) {
1210 if (key->ts[s].alphaarg0 == D3DTA_SPECULAR ||
1211 key->ts[s].alphaarg1 == D3DTA_SPECULAR ||
1212 key->ts[s].alphaarg2 == D3DTA_SPECULAR)
1213 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_PERSPECTIVE);
1214
1215 if (key->ts[s].alphaarg0 == D3DTA_TEXTURE ||
1216 key->ts[s].alphaarg1 == D3DTA_TEXTURE ||
1217 key->ts[s].alphaarg2 == D3DTA_TEXTURE) {
1218 ps.s[s] = ureg_DECL_sampler(ureg, s);
1219 ps.vT[s] = ureg_DECL_fs_input(ureg, texcoord_sn, s, TGSI_INTERPOLATE_PERSPECTIVE);
1220 }
1221 }
1222 }
1223 if (key->specular)
1224 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_PERSPECTIVE);
1225
1226 oCol = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
1227
1228 if (key->ts[0].colorop == D3DTOP_DISABLE &&
1229 key->ts[0].alphaop == D3DTOP_DISABLE)
1230 ureg_MOV(ureg, ps.rCur, ps.vC[0]);
1231 /* Or is it undefined then ? */
1232
1233 /* Run stages.
1234 */
1235 for (s = 0; s < 8; ++s) {
1236 unsigned colorarg[3];
1237 unsigned alphaarg[3];
1238 const uint8_t used_c = ps_d3dtop_args_mask(key->ts[s].colorop);
1239 const uint8_t used_a = ps_d3dtop_args_mask(key->ts[s].alphaop);
1240 struct ureg_dst dst;
1241 struct ureg_src arg[3];
1242
1243 if (key->ts[s].colorop == D3DTOP_DISABLE &&
1244 key->ts[s].alphaop == D3DTOP_DISABLE)
1245 continue;
1246 ps.stage.index = s;
1247 ps.stage.num_regs = 3;
1248
1249 DBG("STAGE[%u]: colorop=%s alphaop=%s\n", s,
1250 nine_D3DTOP_to_str(key->ts[s].colorop),
1251 nine_D3DTOP_to_str(key->ts[s].alphaop));
1252
1253 if (!ureg_src_is_undef(ps.s[s])) {
1254 unsigned target;
1255 switch (key->ts[s].textarget) {
1256 case 0: target = TGSI_TEXTURE_1D; break;
1257 case 1: target = TGSI_TEXTURE_2D; break;
1258 case 2: target = TGSI_TEXTURE_3D; break;
1259 case 3: target = TGSI_TEXTURE_CUBE; break;
1260 /* this is a 2 bit bitfield, do I really need a default case ? */
1261 }
1262
1263 /* sample the texture */
1264 if (key->ts[s].colorop == D3DTOP_BUMPENVMAP ||
1265 key->ts[s].colorop == D3DTOP_BUMPENVMAPLUMINANCE) {
1266 }
1267 if (key->projected & (3 << (s *2))) {
1268 unsigned dim = 1 + ((key->projected >> (2 * s)) & 3);
1269 if (dim == 4)
1270 ureg_TXP(ureg, ps.rTex, target, ps.vT[s], ps.s[s]);
1271 else {
1272 ureg_RCP(ureg, ureg_writemask(ps.rTmp, TGSI_WRITEMASK_X), ureg_scalar(ps.vT[s], dim-1));
1273 ureg_MUL(ureg, ps.rTmp, _XXXX(ps.rTmpSrc), ps.vT[s]);
1274 ureg_TEX(ureg, ps.rTex, target, ps.rTmpSrc, ps.s[s]);
1275 }
1276 } else {
1277 ureg_TEX(ureg, ps.rTex, target, ps.vT[s], ps.s[s]);
1278 }
1279 }
1280
1281 if (s == 0 &&
1282 (key->ts[0].resultarg != 0 /* not current */ ||
1283 key->ts[0].colorop == D3DTOP_DISABLE ||
1284 key->ts[0].alphaop == D3DTOP_DISABLE ||
1285 key->ts[0].colorop == D3DTOP_BLENDCURRENTALPHA ||
1286 key->ts[0].alphaop == D3DTOP_BLENDCURRENTALPHA ||
1287 key->ts[0].colorarg0 == D3DTA_CURRENT ||
1288 key->ts[0].colorarg1 == D3DTA_CURRENT ||
1289 key->ts[0].colorarg2 == D3DTA_CURRENT ||
1290 key->ts[0].alphaarg0 == D3DTA_CURRENT ||
1291 key->ts[0].alphaarg1 == D3DTA_CURRENT ||
1292 key->ts[0].alphaarg2 == D3DTA_CURRENT)
1293 ) {
1294 /* Initialize D3DTA_CURRENT.
1295 * (Yes we can do this before the loop but not until
1296 * NVE4 has an instruction scheduling pass.)
1297 */
1298 ureg_MOV(ureg, ps.rCur, ps.vC[0]);
1299 }
1300
1301 dst = ps_get_ts_dst(&ps, key->ts[s].resultarg ? D3DTA_TEMP : D3DTA_CURRENT);
1302
1303 if (ps.stage.index_pre_mod == ps.stage.index) {
1304 ps.rMod = ps.r[ps.stage.num_regs++];
1305 ureg_MUL(ureg, ps.rMod, ps.rCurSrc, ps.rTexSrc);
1306 }
1307
1308 colorarg[0] = (key->ts[s].colorarg0 | ((key->colorarg_b4[0] >> s) << 4) | ((key->colorarg_b5[0] >> s) << 5)) & 0x3f;
1309 colorarg[1] = (key->ts[s].colorarg1 | ((key->colorarg_b4[1] >> s) << 4) | ((key->colorarg_b5[1] >> s) << 5)) & 0x3f;
1310 colorarg[2] = (key->ts[s].colorarg2 | ((key->colorarg_b4[2] >> s) << 4) | ((key->colorarg_b5[2] >> s) << 5)) & 0x3f;
1311 alphaarg[0] = (key->ts[s].alphaarg0 | ((key->alphaarg_b4[0] >> s) << 4)) & 0x1f;
1312 alphaarg[1] = (key->ts[s].alphaarg1 | ((key->alphaarg_b4[1] >> s) << 4)) & 0x1f;
1313 alphaarg[2] = (key->ts[s].alphaarg2 | ((key->alphaarg_b4[2] >> s) << 4)) & 0x1f;
1314
1315 if (key->ts[s].colorop != key->ts[s].alphaop ||
1316 colorarg[0] != alphaarg[0] ||
1317 colorarg[1] != alphaarg[1] ||
1318 colorarg[2] != alphaarg[2])
1319 dst.WriteMask = TGSI_WRITEMASK_XYZ;
1320
1321 if (used_c & 0x1) arg[0] = ps_get_ts_arg(&ps, colorarg[0]);
1322 if (used_c & 0x2) arg[1] = ps_get_ts_arg(&ps, colorarg[1]);
1323 if (used_c & 0x4) arg[2] = ps_get_ts_arg(&ps, colorarg[2]);
1324 ps_do_ts_op(&ps, key->ts[s].colorop, dst, arg);
1325
1326 if (dst.WriteMask != TGSI_WRITEMASK_XYZW) {
1327 dst.WriteMask = TGSI_WRITEMASK_W;
1328
1329 if (used_a & 0x1) arg[0] = ps_get_ts_arg(&ps, alphaarg[0]);
1330 if (used_a & 0x2) arg[1] = ps_get_ts_arg(&ps, alphaarg[1]);
1331 if (used_a & 0x4) arg[2] = ps_get_ts_arg(&ps, alphaarg[2]);
1332 ps_do_ts_op(&ps, key->ts[s].alphaop, dst, arg);
1333 }
1334 }
1335
1336 if (key->specular)
1337 ureg_ADD(ureg, ps.rCur, ps.rCurSrc, ps.vC[1]);
1338
1339 /* Fog.
1340 */
1341 if (key->fog_mode) {
1342 struct ureg_src vPos = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_POSITION, 0, TGSI_INTERPOLATE_LINEAR);
1343 struct ureg_dst rFog = ureg_writemask(ps.rTmp, TGSI_WRITEMASK_X);
1344 if (key->fog_mode == D3DFOG_EXP) {
1345 ureg_MUL(ureg, rFog, _ZZZZ(vPos), _ZZZZ(_CONST(22)));
1346 ureg_MUL(ureg, rFog, _X(rFog), ureg_imm1f(ureg, -1.442695f));
1347 ureg_EX2(ureg, rFog, _X(rFog));
1348 } else
1349 if (key->fog_mode == D3DFOG_EXP2) {
1350 ureg_MUL(ureg, rFog, _ZZZZ(vPos), _ZZZZ(_CONST(22)));
1351 ureg_MUL(ureg, rFog, _X(rFog), _X(rFog));
1352 ureg_MUL(ureg, rFog, _X(rFog), ureg_imm1f(ureg, -1.442695f));
1353 ureg_EX2(ureg, rFog, _X(rFog));
1354 } else
1355 if (key->fog_mode == D3DFOG_LINEAR) {
1356 ureg_SUB(ureg, rFog, _XXXX(_CONST(22)), _ZZZZ(vPos));
1357 ureg_MUL(ureg, ureg_saturate(rFog), _X(rFog), _YYYY(_CONST(22)));
1358 }
1359 ureg_LRP(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_XYZ), _X(rFog), ps.rCurSrc, _CONST(21));
1360 ureg_MOV(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_W), ps.rCurSrc);
1361 } else
1362 if (key->fog) {
1363 struct ureg_src vFog = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_FOG, 0, TGSI_INTERPOLATE_PERSPECTIVE);
1364 ureg_LRP(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_XYZ), _XXXX(vFog), ps.rCurSrc, _CONST(21));
1365 ureg_MOV(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_W), ps.rCurSrc);
1366 } else {
1367 ureg_MOV(ureg, oCol, ps.rCurSrc);
1368 }
1369
1370 ureg_END(ureg);
1371 nine_ureg_tgsi_dump(ureg, FALSE);
1372 return ureg_create_shader_and_destroy(ureg, device->pipe);
1373 }
1374
1375 static struct NineVertexShader9 *
1376 nine_ff_get_vs(struct NineDevice9 *device)
1377 {
1378 const struct nine_state *state = &device->state;
1379 struct NineVertexShader9 *vs;
1380 enum pipe_error err;
1381 struct vs_build_ctx bld;
1382 struct nine_ff_vs_key key;
1383 unsigned s, i;
1384 char input_texture_coord[8];
1385
1386 assert(sizeof(key) <= sizeof(key.value32));
1387
1388 memset(&key, 0, sizeof(key));
1389 memset(&bld, 0, sizeof(bld));
1390 memset(&input_texture_coord, 0, sizeof(input_texture_coord));
1391
1392 bld.key = &key;
1393
1394 /* FIXME: this shouldn't be NULL, but it is on init */
1395 if (state->vdecl) {
1396 key.color0in_one = 1;
1397 key.color1in_one = 1;
1398 for (i = 0; i < state->vdecl->nelems; i++) {
1399 uint16_t usage = state->vdecl->usage_map[i];
1400 if (usage == NINE_DECLUSAGE_POSITIONT)
1401 key.position_t = 1;
1402 else if (usage == NINE_DECLUSAGE_i(COLOR, 0))
1403 key.color0in_one = 0;
1404 else if (usage == NINE_DECLUSAGE_i(COLOR, 1))
1405 key.color1in_one = 0;
1406 else if (usage == NINE_DECLUSAGE_PSIZE)
1407 key.vertexpointsize = 1;
1408 else if (usage % NINE_DECLUSAGE_COUNT == NINE_DECLUSAGE_TEXCOORD) {
1409 s = usage / NINE_DECLUSAGE_COUNT;
1410 if (s < 8)
1411 input_texture_coord[s] = nine_decltype_get_dim(state->vdecl->decls[i].Type);
1412 else
1413 DBG("FF given texture coordinate >= 8. Ignoring\n");
1414 }
1415 }
1416 }
1417 if (!key.vertexpointsize)
1418 key.pointscale = !!state->rs[D3DRS_POINTSCALEENABLE];
1419
1420 key.lighting = !!state->rs[D3DRS_LIGHTING] && state->ff.num_lights_active;
1421 key.darkness = !!state->rs[D3DRS_LIGHTING] && !state->ff.num_lights_active;
1422 if (key.position_t) {
1423 key.darkness = 0; /* |= key.lighting; */ /* XXX ? */
1424 key.lighting = 0;
1425 }
1426 if ((key.lighting | key.darkness) && state->rs[D3DRS_COLORVERTEX]) {
1427 key.mtl_diffuse = state->rs[D3DRS_DIFFUSEMATERIALSOURCE];
1428 key.mtl_ambient = state->rs[D3DRS_AMBIENTMATERIALSOURCE];
1429 key.mtl_specular = state->rs[D3DRS_SPECULARMATERIALSOURCE];
1430 key.mtl_emissive = state->rs[D3DRS_EMISSIVEMATERIALSOURCE];
1431 }
1432 key.fog = !!state->rs[D3DRS_FOGENABLE];
1433 key.fog_mode = state->rs[D3DRS_FOGENABLE] ? state->rs[D3DRS_FOGVERTEXMODE] : 0;
1434 if (key.fog_mode)
1435 key.fog_range = !key.position_t && state->rs[D3DRS_RANGEFOGENABLE];
1436
1437 if (state->rs[D3DRS_VERTEXBLEND] != D3DVBF_DISABLE) {
1438 key.vertexblend_indexed = !!state->rs[D3DRS_INDEXEDVERTEXBLENDENABLE];
1439
1440 switch (state->rs[D3DRS_VERTEXBLEND]) {
1441 case D3DVBF_0WEIGHTS: key.vertexblend = key.vertexblend_indexed; break;
1442 case D3DVBF_1WEIGHTS: key.vertexblend = 2; break;
1443 case D3DVBF_2WEIGHTS: key.vertexblend = 3; break;
1444 case D3DVBF_3WEIGHTS: key.vertexblend = 4; break;
1445 case D3DVBF_TWEENING: key.vertextween = 1; break;
1446 default:
1447 assert(!"invalid D3DVBF");
1448 break;
1449 }
1450 }
1451
1452 for (s = 0; s < 8; ++s) {
1453 unsigned gen = (state->ff.tex_stage[s][D3DTSS_TEXCOORDINDEX] >> 16) + 1;
1454 unsigned dim;
1455
1456 if (key.position_t && gen > NINED3DTSS_TCI_PASSTHRU)
1457 gen = NINED3DTSS_TCI_PASSTHRU;
1458
1459 if (!input_texture_coord[s] && gen == NINED3DTSS_TCI_PASSTHRU)
1460 gen = NINED3DTSS_TCI_DISABLE;
1461
1462 key.tc_gen |= gen << (s * 3);
1463 key.tc_idx |= (state->ff.tex_stage[s][D3DTSS_TEXCOORDINDEX] & 7) << (s * 3);
1464 key.tc_dim_input |= ((input_texture_coord[s]-1) & 0x3) << (s * 2);
1465
1466 dim = state->ff.tex_stage[s][D3DTSS_TEXTURETRANSFORMFLAGS] & 0x7;
1467 if (dim > 4)
1468 dim = input_texture_coord[s];
1469 if (dim == 1) /* NV behaviour */
1470 dim = 0;
1471 key.tc_dim_output |= dim << (s * 3);
1472 }
1473
1474 vs = util_hash_table_get(device->ff.ht_vs, &key);
1475 if (vs)
1476 return vs;
1477 NineVertexShader9_new(device, &vs, NULL, nine_ff_build_vs(device, &bld));
1478
1479 nine_ff_prune_vs(device);
1480 if (vs) {
1481 unsigned n;
1482
1483 memcpy(&vs->ff_key, &key, sizeof(vs->ff_key));
1484
1485 err = util_hash_table_set(device->ff.ht_vs, &vs->ff_key, vs);
1486 assert(err == PIPE_OK);
1487 device->ff.num_vs++;
1488 NineUnknown_ConvertRefToBind(NineUnknown(vs));
1489
1490 vs->num_inputs = bld.num_inputs;
1491 for (n = 0; n < bld.num_inputs; ++n)
1492 vs->input_map[n].ndecl = bld.input[n];
1493
1494 vs->position_t = key.position_t;
1495 vs->point_size = key.vertexpointsize | key.pointscale;
1496 }
1497 return vs;
1498 }
1499
1500 static struct NinePixelShader9 *
1501 nine_ff_get_ps(struct NineDevice9 *device)
1502 {
1503 struct nine_state *state = &device->state;
1504 struct NinePixelShader9 *ps;
1505 enum pipe_error err;
1506 struct nine_ff_ps_key key;
1507 unsigned s;
1508 uint8_t sampler_mask = 0;
1509
1510 assert(sizeof(key) <= sizeof(key.value32));
1511
1512 memset(&key, 0, sizeof(key));
1513 for (s = 0; s < 8; ++s) {
1514 key.ts[s].colorop = state->ff.tex_stage[s][D3DTSS_COLOROP];
1515 key.ts[s].alphaop = state->ff.tex_stage[s][D3DTSS_ALPHAOP];
1516 /* MSDN says D3DTOP_DISABLE disables this and all subsequent stages. */
1517 /* ALPHAOP cannot be disabled if COLOROP is enabled. */
1518 if (key.ts[s].colorop == D3DTOP_DISABLE) {
1519 key.ts[s].alphaop = D3DTOP_DISABLE; /* DISABLE == 1, avoid degenerate keys */
1520 break;
1521 }
1522
1523 if (!state->texture[s] &&
1524 state->ff.tex_stage[s][D3DTSS_COLORARG1] == D3DTA_TEXTURE) {
1525 /* This should also disable the stage. */
1526 key.ts[s].colorop = key.ts[s].alphaop = D3DTOP_DISABLE;
1527 break;
1528 }
1529
1530 if (state->ff.tex_stage[s][D3DTSS_COLORARG1] == D3DTA_TEXTURE)
1531 sampler_mask |= (1 << s);
1532
1533 if (key.ts[s].colorop != D3DTOP_DISABLE) {
1534 uint8_t used_c = ps_d3dtop_args_mask(key.ts[s].colorop);
1535 if (used_c & 0x1) key.ts[s].colorarg0 = state->ff.tex_stage[s][D3DTSS_COLORARG0];
1536 if (used_c & 0x2) key.ts[s].colorarg1 = state->ff.tex_stage[s][D3DTSS_COLORARG1];
1537 if (used_c & 0x4) key.ts[s].colorarg2 = state->ff.tex_stage[s][D3DTSS_COLORARG2];
1538 if (used_c & 0x1) key.colorarg_b4[0] |= (state->ff.tex_stage[s][D3DTSS_COLORARG0] >> 4) << s;
1539 if (used_c & 0x1) key.colorarg_b5[0] |= (state->ff.tex_stage[s][D3DTSS_COLORARG0] >> 5) << s;
1540 if (used_c & 0x2) key.colorarg_b4[1] |= (state->ff.tex_stage[s][D3DTSS_COLORARG1] >> 4) << s;
1541 if (used_c & 0x2) key.colorarg_b5[1] |= (state->ff.tex_stage[s][D3DTSS_COLORARG1] >> 5) << s;
1542 if (used_c & 0x4) key.colorarg_b4[2] |= (state->ff.tex_stage[s][D3DTSS_COLORARG2] >> 4) << s;
1543 if (used_c & 0x4) key.colorarg_b5[2] |= (state->ff.tex_stage[s][D3DTSS_COLORARG2] >> 5) << s;
1544 }
1545 if (key.ts[s].alphaop != D3DTOP_DISABLE) {
1546 uint8_t used_a = ps_d3dtop_args_mask(key.ts[s].alphaop);
1547 if (used_a & 0x1) key.ts[s].alphaarg0 = state->ff.tex_stage[s][D3DTSS_ALPHAARG0];
1548 if (used_a & 0x2) key.ts[s].alphaarg1 = state->ff.tex_stage[s][D3DTSS_ALPHAARG1];
1549 if (used_a & 0x4) key.ts[s].alphaarg2 = state->ff.tex_stage[s][D3DTSS_ALPHAARG2];
1550 if (used_a & 0x1) key.alphaarg_b4[0] |= (state->ff.tex_stage[s][D3DTSS_ALPHAARG0] >> 4) << s;
1551 if (used_a & 0x2) key.alphaarg_b4[1] |= (state->ff.tex_stage[s][D3DTSS_ALPHAARG1] >> 4) << s;
1552 if (used_a & 0x4) key.alphaarg_b4[2] |= (state->ff.tex_stage[s][D3DTSS_ALPHAARG2] >> 4) << s;
1553 }
1554 key.ts[s].resultarg = state->ff.tex_stage[s][D3DTSS_RESULTARG] == D3DTA_TEMP;
1555
1556 if (state->texture[s]) {
1557 switch (state->texture[s]->base.type) {
1558 case D3DRTYPE_TEXTURE: key.ts[s].textarget = 1; break;
1559 case D3DRTYPE_VOLUMETEXTURE: key.ts[s].textarget = 2; break;
1560 case D3DRTYPE_CUBETEXTURE: key.ts[s].textarget = 3; break;
1561 default:
1562 assert(!"unexpected texture type");
1563 break;
1564 }
1565 } else {
1566 key.ts[s].textarget = 1;
1567 }
1568 }
1569
1570 key.projected = nine_ff_get_projected_key(state);
1571
1572 for (; s < 8; ++s)
1573 key.ts[s].colorop = key.ts[s].alphaop = D3DTOP_DISABLE;
1574 if (state->rs[D3DRS_FOGENABLE])
1575 key.fog_mode = state->rs[D3DRS_FOGTABLEMODE];
1576 key.fog = !!state->rs[D3DRS_FOGENABLE];
1577
1578 ps = util_hash_table_get(device->ff.ht_ps, &key);
1579 if (ps)
1580 return ps;
1581 NinePixelShader9_new(device, &ps, NULL, nine_ff_build_ps(device, &key));
1582
1583 nine_ff_prune_ps(device);
1584 if (ps) {
1585 memcpy(&ps->ff_key, &key, sizeof(ps->ff_key));
1586
1587 err = util_hash_table_set(device->ff.ht_ps, &ps->ff_key, ps);
1588 assert(err == PIPE_OK);
1589 device->ff.num_ps++;
1590 NineUnknown_ConvertRefToBind(NineUnknown(ps));
1591
1592 ps->rt_mask = 0x1;
1593 ps->sampler_mask = sampler_mask;
1594 }
1595 return ps;
1596 }
1597
1598 #define GET_D3DTS(n) nine_state_access_transform(state, D3DTS_##n, FALSE)
1599 #define IS_D3DTS_DIRTY(s,n) ((s)->ff.changed.transform[(D3DTS_##n) / 32] & (1 << ((D3DTS_##n) % 32)))
1600 static void
1601 nine_ff_load_vs_transforms(struct NineDevice9 *device)
1602 {
1603 struct nine_state *state = &device->state;
1604 D3DMATRIX T;
1605 D3DMATRIX *M = (D3DMATRIX *)device->ff.vs_const;
1606 unsigned i;
1607
1608 /* TODO: make this nicer, and only upload the ones we need */
1609 /* TODO: use ff.vs_const as storage of W, V, P matrices */
1610
1611 if (IS_D3DTS_DIRTY(state, WORLD) ||
1612 IS_D3DTS_DIRTY(state, VIEW) ||
1613 IS_D3DTS_DIRTY(state, PROJECTION)) {
1614 /* WVP, WV matrices */
1615 nine_d3d_matrix_matrix_mul(&M[1], GET_D3DTS(WORLD), GET_D3DTS(VIEW));
1616 nine_d3d_matrix_matrix_mul(&M[0], &M[1], GET_D3DTS(PROJECTION));
1617
1618 /* normal matrix == transpose(inverse(WV)) */
1619 nine_d3d_matrix_inverse_3x3(&T, &M[1]);
1620 nine_d3d_matrix_transpose(&M[4], &T);
1621
1622 /* VP matrix */
1623 nine_d3d_matrix_matrix_mul(&M[2], GET_D3DTS(VIEW), GET_D3DTS(PROJECTION));
1624
1625 /* V and W matrix */
1626 M[3] = *GET_D3DTS(VIEW);
1627 M[56] = *GET_D3DTS(WORLD);
1628 }
1629
1630 if (state->rs[D3DRS_VERTEXBLEND] != D3DVBF_DISABLE) {
1631 /* load other world matrices */
1632 for (i = 1; i <= 7; ++i)
1633 M[56 + i] = *GET_D3DTS(WORLDMATRIX(i));
1634 }
1635
1636 device->ff.vs_const[30 * 4] = asfloat(state->rs[D3DRS_TWEENFACTOR]);
1637 }
1638
1639 static void
1640 nine_ff_load_lights(struct NineDevice9 *device)
1641 {
1642 struct nine_state *state = &device->state;
1643 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1644 unsigned l;
1645
1646 if (state->changed.group & NINE_STATE_FF_MATERIAL) {
1647 const D3DMATERIAL9 *mtl = &state->ff.material;
1648
1649 memcpy(&dst[20], &mtl->Diffuse, 4 * sizeof(float));
1650 memcpy(&dst[21], &mtl->Ambient, 4 * sizeof(float));
1651 memcpy(&dst[22], &mtl->Specular, 4 * sizeof(float));
1652 dst[23].x = mtl->Power;
1653 memcpy(&dst[24], &mtl->Emissive, 4 * sizeof(float));
1654 d3dcolor_to_rgba(&dst[25].x, state->rs[D3DRS_AMBIENT]);
1655 dst[19].x = dst[25].x * mtl->Ambient.r + mtl->Emissive.r;
1656 dst[19].y = dst[25].y * mtl->Ambient.g + mtl->Emissive.g;
1657 dst[19].z = dst[25].z * mtl->Ambient.b + mtl->Emissive.b;
1658 dst[19].w = mtl->Ambient.a + mtl->Emissive.a;
1659 }
1660
1661 if (!(state->changed.group & NINE_STATE_FF_LIGHTING))
1662 return;
1663
1664 for (l = 0; l < state->ff.num_lights_active; ++l) {
1665 const D3DLIGHT9 *light = &state->ff.light[state->ff.active_light[l]];
1666
1667 dst[32 + l * 8].x = light->Type;
1668 dst[32 + l * 8].y = light->Attenuation0;
1669 dst[32 + l * 8].z = light->Attenuation1;
1670 dst[32 + l * 8].w = light->Attenuation2;
1671 memcpy(&dst[33 + l * 8].x, &light->Diffuse, sizeof(light->Diffuse));
1672 memcpy(&dst[34 + l * 8].x, &light->Specular, sizeof(light->Specular));
1673 memcpy(&dst[35 + l * 8].x, &light->Ambient, sizeof(light->Ambient));
1674 nine_d3d_vector4_matrix_mul((D3DVECTOR *)&dst[36 + l * 8].x, &light->Position, GET_D3DTS(VIEW));
1675 nine_d3d_vector3_matrix_mul((D3DVECTOR *)&dst[37 + l * 8].x, &light->Direction, GET_D3DTS(VIEW));
1676 dst[36 + l * 8].w = light->Type == D3DLIGHT_DIRECTIONAL ? 1e9f : light->Range;
1677 dst[37 + l * 8].w = light->Falloff;
1678 dst[38 + l * 8].x = cosf(light->Theta * 0.5f);
1679 dst[38 + l * 8].y = cosf(light->Phi * 0.5f);
1680 dst[38 + l * 8].z = 1.0f / (dst[38 + l * 8].x - dst[38 + l * 8].y);
1681 dst[39 + l * 8].w = (l + 1) == state->ff.num_lights_active;
1682 }
1683 }
1684
1685 static void
1686 nine_ff_load_point_and_fog_params(struct NineDevice9 *device)
1687 {
1688 const struct nine_state *state = &device->state;
1689 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1690
1691 if (!(state->changed.group & NINE_STATE_FF_OTHER))
1692 return;
1693 dst[26].x = asfloat(state->rs[D3DRS_POINTSIZE_MIN]);
1694 dst[26].y = asfloat(state->rs[D3DRS_POINTSIZE_MAX]);
1695 dst[26].z = asfloat(state->rs[D3DRS_POINTSIZE]);
1696 dst[26].w = asfloat(state->rs[D3DRS_POINTSCALE_A]);
1697 dst[27].x = asfloat(state->rs[D3DRS_POINTSCALE_B]);
1698 dst[27].y = asfloat(state->rs[D3DRS_POINTSCALE_C]);
1699 dst[28].x = asfloat(state->rs[D3DRS_FOGEND]);
1700 dst[28].y = 1.0f / (asfloat(state->rs[D3DRS_FOGEND]) - asfloat(state->rs[D3DRS_FOGSTART]));
1701 if (isinf(dst[28].y))
1702 dst[28].y = 0.0f;
1703 dst[28].z = asfloat(state->rs[D3DRS_FOGDENSITY]);
1704 }
1705
1706 static void
1707 nine_ff_load_tex_matrices(struct NineDevice9 *device)
1708 {
1709 struct nine_state *state = &device->state;
1710 D3DMATRIX *M = (D3DMATRIX *)device->ff.vs_const;
1711 unsigned s;
1712
1713 if (!(state->ff.changed.transform[0] & 0xff0000))
1714 return;
1715 for (s = 0; s < 8; ++s) {
1716 if (IS_D3DTS_DIRTY(state, TEXTURE0 + s))
1717 nine_d3d_matrix_transpose(&M[32 + s], nine_state_access_transform(state, D3DTS_TEXTURE0 + s, FALSE));
1718 }
1719 }
1720
1721 static void
1722 nine_ff_load_ps_params(struct NineDevice9 *device)
1723 {
1724 const struct nine_state *state = &device->state;
1725 struct fvec4 *dst = (struct fvec4 *)device->ff.ps_const;
1726 unsigned s;
1727
1728 if (!(state->changed.group & (NINE_STATE_FF_PSSTAGES | NINE_STATE_FF_OTHER)))
1729 return;
1730
1731 for (s = 0; s < 8; ++s)
1732 d3dcolor_to_rgba(&dst[s].x, state->ff.tex_stage[s][D3DTSS_CONSTANT]);
1733
1734 for (s = 0; s < 8; ++s) {
1735 dst[8 + s].x = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT00]);
1736 dst[8 + s].y = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT01]);
1737 dst[8 + s].z = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT10]);
1738 dst[8 + s].w = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT11]);
1739 if (s & 1) {
1740 dst[8 + s / 2].z = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLSCALE]);
1741 dst[8 + s / 2].w = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLOFFSET]);
1742 } else {
1743 dst[8 + s / 2].x = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLSCALE]);
1744 dst[8 + s / 2].y = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLOFFSET]);
1745 }
1746 }
1747
1748 d3dcolor_to_rgba(&dst[20].x, state->rs[D3DRS_TEXTUREFACTOR]);
1749 d3dcolor_to_rgba(&dst[21].x, state->rs[D3DRS_FOGCOLOR]);
1750 dst[22].x = asfloat(state->rs[D3DRS_FOGEND]);
1751 dst[22].y = 1.0f / (asfloat(state->rs[D3DRS_FOGEND]) - asfloat(state->rs[D3DRS_FOGSTART]));
1752 dst[22].z = asfloat(state->rs[D3DRS_FOGDENSITY]);
1753 }
1754
1755 static void
1756 nine_ff_load_viewport_info(struct NineDevice9 *device)
1757 {
1758 D3DVIEWPORT9 *viewport = &device->state.viewport;
1759 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1760 float diffZ = viewport->MaxZ - viewport->MinZ;
1761
1762 /* Note: the other functions avoids to fill the const again if nothing changed.
1763 * But we don't have much to fill, and adding code to allow that may be complex
1764 * so just fill it always */
1765 dst[100].x = 2.0f / (float)(viewport->Width);
1766 dst[100].y = 2.0f / (float)(viewport->Height);
1767 dst[100].z = (diffZ == 0.0f) ? 0.0f : (1.0f / diffZ);
1768 dst[101].x = (float)(viewport->X);
1769 dst[101].y = (float)(viewport->Y);
1770 dst[101].z = (float)(viewport->MinZ);
1771 }
1772
1773 void
1774 nine_ff_update(struct NineDevice9 *device)
1775 {
1776 struct nine_state *state = &device->state;
1777 struct pipe_constant_buffer cb;
1778
1779 DBG("vs=%p ps=%p\n", device->state.vs, device->state.ps);
1780
1781 /* NOTE: the only reference belongs to the hash table */
1782 if (!device->state.vs) {
1783 device->ff.vs = nine_ff_get_vs(device);
1784 device->state.changed.group |= NINE_STATE_VS;
1785 }
1786 if (!device->state.ps) {
1787 device->ff.ps = nine_ff_get_ps(device);
1788 device->state.changed.group |= NINE_STATE_PS;
1789 }
1790
1791 if (!device->state.vs) {
1792 nine_ff_load_vs_transforms(device);
1793 nine_ff_load_tex_matrices(device);
1794 nine_ff_load_lights(device);
1795 nine_ff_load_point_and_fog_params(device);
1796 nine_ff_load_viewport_info(device);
1797
1798 memset(state->ff.changed.transform, 0, sizeof(state->ff.changed.transform));
1799
1800 cb.buffer_offset = 0;
1801 cb.buffer = NULL;
1802 cb.user_buffer = device->ff.vs_const;
1803 cb.buffer_size = NINE_FF_NUM_VS_CONST * 4 * sizeof(float);
1804
1805 if (!device->driver_caps.user_cbufs) {
1806 u_upload_data(device->constbuf_uploader,
1807 0,
1808 cb.buffer_size,
1809 cb.user_buffer,
1810 &cb.buffer_offset,
1811 &cb.buffer);
1812 u_upload_unmap(device->constbuf_uploader);
1813 cb.user_buffer = NULL;
1814 }
1815 state->pipe.cb_vs_ff = cb;
1816 state->commit |= NINE_STATE_COMMIT_CONST_VS;
1817 }
1818
1819 if (!device->state.ps) {
1820 nine_ff_load_ps_params(device);
1821
1822 cb.buffer_offset = 0;
1823 cb.buffer = NULL;
1824 cb.user_buffer = device->ff.ps_const;
1825 cb.buffer_size = NINE_FF_NUM_PS_CONST * 4 * sizeof(float);
1826
1827 if (!device->driver_caps.user_cbufs) {
1828 u_upload_data(device->constbuf_uploader,
1829 0,
1830 cb.buffer_size,
1831 cb.user_buffer,
1832 &cb.buffer_offset,
1833 &cb.buffer);
1834 u_upload_unmap(device->constbuf_uploader);
1835 cb.user_buffer = NULL;
1836 }
1837 state->pipe.cb_ps_ff = cb;
1838 state->commit |= NINE_STATE_COMMIT_CONST_PS;
1839 }
1840
1841 device->state.changed.group &= ~NINE_STATE_FF;
1842 }
1843
1844
1845 boolean
1846 nine_ff_init(struct NineDevice9 *device)
1847 {
1848 device->ff.ht_vs = util_hash_table_create(nine_ff_vs_key_hash,
1849 nine_ff_vs_key_comp);
1850 device->ff.ht_ps = util_hash_table_create(nine_ff_ps_key_hash,
1851 nine_ff_ps_key_comp);
1852
1853 device->ff.ht_fvf = util_hash_table_create(nine_ff_fvf_key_hash,
1854 nine_ff_fvf_key_comp);
1855
1856 device->ff.vs_const = CALLOC(NINE_FF_NUM_VS_CONST, 4 * sizeof(float));
1857 device->ff.ps_const = CALLOC(NINE_FF_NUM_PS_CONST, 4 * sizeof(float));
1858
1859 return device->ff.ht_vs && device->ff.ht_ps &&
1860 device->ff.ht_fvf &&
1861 device->ff.vs_const && device->ff.ps_const;
1862 }
1863
1864 static enum pipe_error nine_ff_ht_delete_cb(void *key, void *value, void *data)
1865 {
1866 NineUnknown_Unbind(NineUnknown(value));
1867 return PIPE_OK;
1868 }
1869
1870 void
1871 nine_ff_fini(struct NineDevice9 *device)
1872 {
1873 if (device->ff.ht_vs) {
1874 util_hash_table_foreach(device->ff.ht_vs, nine_ff_ht_delete_cb, NULL);
1875 util_hash_table_destroy(device->ff.ht_vs);
1876 }
1877 if (device->ff.ht_ps) {
1878 util_hash_table_foreach(device->ff.ht_ps, nine_ff_ht_delete_cb, NULL);
1879 util_hash_table_destroy(device->ff.ht_ps);
1880 }
1881 if (device->ff.ht_fvf) {
1882 util_hash_table_foreach(device->ff.ht_fvf, nine_ff_ht_delete_cb, NULL);
1883 util_hash_table_destroy(device->ff.ht_fvf);
1884 }
1885 device->ff.vs = NULL; /* destroyed by unbinding from hash table */
1886 device->ff.ps = NULL;
1887
1888 FREE(device->ff.vs_const);
1889 FREE(device->ff.ps_const);
1890 }
1891
1892 static void
1893 nine_ff_prune_vs(struct NineDevice9 *device)
1894 {
1895 if (device->ff.num_vs > 100) {
1896 /* could destroy the bound one here, so unbind */
1897 device->pipe->bind_vs_state(device->pipe, NULL);
1898 util_hash_table_foreach(device->ff.ht_vs, nine_ff_ht_delete_cb, NULL);
1899 util_hash_table_clear(device->ff.ht_vs);
1900 device->ff.num_vs = 0;
1901 device->state.changed.group |= NINE_STATE_VS;
1902 }
1903 }
1904 static void
1905 nine_ff_prune_ps(struct NineDevice9 *device)
1906 {
1907 if (device->ff.num_ps > 100) {
1908 /* could destroy the bound one here, so unbind */
1909 device->pipe->bind_fs_state(device->pipe, NULL);
1910 util_hash_table_foreach(device->ff.ht_ps, nine_ff_ht_delete_cb, NULL);
1911 util_hash_table_clear(device->ff.ht_ps);
1912 device->ff.num_ps = 0;
1913 device->state.changed.group |= NINE_STATE_PS;
1914 }
1915 }
1916
1917 /* ========================================================================== */
1918
1919 /* Matrix multiplication:
1920 *
1921 * in memory: 0 1 2 3 (row major)
1922 * 4 5 6 7
1923 * 8 9 a b
1924 * c d e f
1925 *
1926 * cA cB cC cD
1927 * r0 = (r0 * cA) (r0 * cB) . .
1928 * r1 = (r1 * cA) (r1 * cB)
1929 * r2 = (r2 * cA) .
1930 * r3 = (r3 * cA) .
1931 *
1932 * r: (11) (12) (13) (14)
1933 * (21) (22) (23) (24)
1934 * (31) (32) (33) (34)
1935 * (41) (42) (43) (44)
1936 * l: (11 12 13 14)
1937 * (21 22 23 24)
1938 * (31 32 33 34)
1939 * (41 42 43 44)
1940 *
1941 * v: (x y z 1 )
1942 *
1943 * t.xyzw = MUL(v.xxxx, r[0]);
1944 * t.xyzw = MAD(v.yyyy, r[1], t.xyzw);
1945 * t.xyzw = MAD(v.zzzz, r[2], t.xyzw);
1946 * v.xyzw = MAD(v.wwww, r[3], t.xyzw);
1947 *
1948 * v.x = DP4(v, c[0]);
1949 * v.y = DP4(v, c[1]);
1950 * v.z = DP4(v, c[2]);
1951 * v.w = DP4(v, c[3]) = 1
1952 */
1953
1954 /*
1955 static void
1956 nine_D3DMATRIX_print(const D3DMATRIX *M)
1957 {
1958 DBG("\n(%f %f %f %f)\n"
1959 "(%f %f %f %f)\n"
1960 "(%f %f %f %f)\n"
1961 "(%f %f %f %f)\n",
1962 M->m[0][0], M->m[0][1], M->m[0][2], M->m[0][3],
1963 M->m[1][0], M->m[1][1], M->m[1][2], M->m[1][3],
1964 M->m[2][0], M->m[2][1], M->m[2][2], M->m[2][3],
1965 M->m[3][0], M->m[3][1], M->m[3][2], M->m[3][3]);
1966 }
1967 */
1968
1969 static inline float
1970 nine_DP4_row_col(const D3DMATRIX *A, int r, const D3DMATRIX *B, int c)
1971 {
1972 return A->m[r][0] * B->m[0][c] +
1973 A->m[r][1] * B->m[1][c] +
1974 A->m[r][2] * B->m[2][c] +
1975 A->m[r][3] * B->m[3][c];
1976 }
1977
1978 static inline float
1979 nine_DP4_vec_col(const D3DVECTOR *v, const D3DMATRIX *M, int c)
1980 {
1981 return v->x * M->m[0][c] +
1982 v->y * M->m[1][c] +
1983 v->z * M->m[2][c] +
1984 1.0f * M->m[3][c];
1985 }
1986
1987 static inline float
1988 nine_DP3_vec_col(const D3DVECTOR *v, const D3DMATRIX *M, int c)
1989 {
1990 return v->x * M->m[0][c] +
1991 v->y * M->m[1][c] +
1992 v->z * M->m[2][c];
1993 }
1994
1995 void
1996 nine_d3d_matrix_matrix_mul(D3DMATRIX *D, const D3DMATRIX *L, const D3DMATRIX *R)
1997 {
1998 D->_11 = nine_DP4_row_col(L, 0, R, 0);
1999 D->_12 = nine_DP4_row_col(L, 0, R, 1);
2000 D->_13 = nine_DP4_row_col(L, 0, R, 2);
2001 D->_14 = nine_DP4_row_col(L, 0, R, 3);
2002
2003 D->_21 = nine_DP4_row_col(L, 1, R, 0);
2004 D->_22 = nine_DP4_row_col(L, 1, R, 1);
2005 D->_23 = nine_DP4_row_col(L, 1, R, 2);
2006 D->_24 = nine_DP4_row_col(L, 1, R, 3);
2007
2008 D->_31 = nine_DP4_row_col(L, 2, R, 0);
2009 D->_32 = nine_DP4_row_col(L, 2, R, 1);
2010 D->_33 = nine_DP4_row_col(L, 2, R, 2);
2011 D->_34 = nine_DP4_row_col(L, 2, R, 3);
2012
2013 D->_41 = nine_DP4_row_col(L, 3, R, 0);
2014 D->_42 = nine_DP4_row_col(L, 3, R, 1);
2015 D->_43 = nine_DP4_row_col(L, 3, R, 2);
2016 D->_44 = nine_DP4_row_col(L, 3, R, 3);
2017 }
2018
2019 void
2020 nine_d3d_vector4_matrix_mul(D3DVECTOR *d, const D3DVECTOR *v, const D3DMATRIX *M)
2021 {
2022 d->x = nine_DP4_vec_col(v, M, 0);
2023 d->y = nine_DP4_vec_col(v, M, 1);
2024 d->z = nine_DP4_vec_col(v, M, 2);
2025 }
2026
2027 void
2028 nine_d3d_vector3_matrix_mul(D3DVECTOR *d, const D3DVECTOR *v, const D3DMATRIX *M)
2029 {
2030 d->x = nine_DP3_vec_col(v, M, 0);
2031 d->y = nine_DP3_vec_col(v, M, 1);
2032 d->z = nine_DP3_vec_col(v, M, 2);
2033 }
2034
2035 void
2036 nine_d3d_matrix_transpose(D3DMATRIX *D, const D3DMATRIX *M)
2037 {
2038 unsigned i, j;
2039 for (i = 0; i < 4; ++i)
2040 for (j = 0; j < 4; ++j)
2041 D->m[i][j] = M->m[j][i];
2042 }
2043
2044 #define _M_ADD_PROD_1i_2j_3k_4l(i,j,k,l) do { \
2045 float t = M->_1##i * M->_2##j * M->_3##k * M->_4##l; \
2046 if (t > 0.0f) pos += t; else neg += t; } while(0)
2047
2048 #define _M_SUB_PROD_1i_2j_3k_4l(i,j,k,l) do { \
2049 float t = M->_1##i * M->_2##j * M->_3##k * M->_4##l; \
2050 if (t > 0.0f) neg -= t; else pos -= t; } while(0)
2051 float
2052 nine_d3d_matrix_det(const D3DMATRIX *M)
2053 {
2054 float pos = 0.0f;
2055 float neg = 0.0f;
2056
2057 _M_ADD_PROD_1i_2j_3k_4l(1, 2, 3, 4);
2058 _M_ADD_PROD_1i_2j_3k_4l(1, 3, 4, 2);
2059 _M_ADD_PROD_1i_2j_3k_4l(1, 4, 2, 3);
2060
2061 _M_ADD_PROD_1i_2j_3k_4l(2, 1, 4, 3);
2062 _M_ADD_PROD_1i_2j_3k_4l(2, 3, 1, 4);
2063 _M_ADD_PROD_1i_2j_3k_4l(2, 4, 3, 1);
2064
2065 _M_ADD_PROD_1i_2j_3k_4l(3, 1, 2, 4);
2066 _M_ADD_PROD_1i_2j_3k_4l(3, 2, 4, 1);
2067 _M_ADD_PROD_1i_2j_3k_4l(3, 4, 1, 2);
2068
2069 _M_ADD_PROD_1i_2j_3k_4l(4, 1, 3, 2);
2070 _M_ADD_PROD_1i_2j_3k_4l(4, 2, 1, 3);
2071 _M_ADD_PROD_1i_2j_3k_4l(4, 3, 2, 1);
2072
2073 _M_SUB_PROD_1i_2j_3k_4l(1, 2, 4, 3);
2074 _M_SUB_PROD_1i_2j_3k_4l(1, 3, 2, 4);
2075 _M_SUB_PROD_1i_2j_3k_4l(1, 4, 3, 2);
2076
2077 _M_SUB_PROD_1i_2j_3k_4l(2, 1, 3, 4);
2078 _M_SUB_PROD_1i_2j_3k_4l(2, 3, 4, 1);
2079 _M_SUB_PROD_1i_2j_3k_4l(2, 4, 1, 3);
2080
2081 _M_SUB_PROD_1i_2j_3k_4l(3, 1, 4, 2);
2082 _M_SUB_PROD_1i_2j_3k_4l(3, 2, 1, 4);
2083 _M_SUB_PROD_1i_2j_3k_4l(3, 4, 2, 1);
2084
2085 _M_SUB_PROD_1i_2j_3k_4l(4, 1, 2, 3);
2086 _M_SUB_PROD_1i_2j_3k_4l(4, 2, 3, 1);
2087 _M_SUB_PROD_1i_2j_3k_4l(4, 3, 1, 2);
2088
2089 return pos + neg;
2090 }
2091
2092 /* XXX: Probably better to just use src/mesa/math/m_matrix.c because
2093 * I have no idea where this code came from.
2094 */
2095 void
2096 nine_d3d_matrix_inverse(D3DMATRIX *D, const D3DMATRIX *M)
2097 {
2098 int i, k;
2099 float det;
2100
2101 D->m[0][0] =
2102 M->m[1][1] * M->m[2][2] * M->m[3][3] -
2103 M->m[1][1] * M->m[3][2] * M->m[2][3] -
2104 M->m[1][2] * M->m[2][1] * M->m[3][3] +
2105 M->m[1][2] * M->m[3][1] * M->m[2][3] +
2106 M->m[1][3] * M->m[2][1] * M->m[3][2] -
2107 M->m[1][3] * M->m[3][1] * M->m[2][2];
2108
2109 D->m[0][1] =
2110 -M->m[0][1] * M->m[2][2] * M->m[3][3] +
2111 M->m[0][1] * M->m[3][2] * M->m[2][3] +
2112 M->m[0][2] * M->m[2][1] * M->m[3][3] -
2113 M->m[0][2] * M->m[3][1] * M->m[2][3] -
2114 M->m[0][3] * M->m[2][1] * M->m[3][2] +
2115 M->m[0][3] * M->m[3][1] * M->m[2][2];
2116
2117 D->m[0][2] =
2118 M->m[0][1] * M->m[1][2] * M->m[3][3] -
2119 M->m[0][1] * M->m[3][2] * M->m[1][3] -
2120 M->m[0][2] * M->m[1][1] * M->m[3][3] +
2121 M->m[0][2] * M->m[3][1] * M->m[1][3] +
2122 M->m[0][3] * M->m[1][1] * M->m[3][2] -
2123 M->m[0][3] * M->m[3][1] * M->m[1][2];
2124
2125 D->m[0][3] =
2126 -M->m[0][1] * M->m[1][2] * M->m[2][3] +
2127 M->m[0][1] * M->m[2][2] * M->m[1][3] +
2128 M->m[0][2] * M->m[1][1] * M->m[2][3] -
2129 M->m[0][2] * M->m[2][1] * M->m[1][3] -
2130 M->m[0][3] * M->m[1][1] * M->m[2][2] +
2131 M->m[0][3] * M->m[2][1] * M->m[1][2];
2132
2133 D->m[1][0] =
2134 -M->m[1][0] * M->m[2][2] * M->m[3][3] +
2135 M->m[1][0] * M->m[3][2] * M->m[2][3] +
2136 M->m[1][2] * M->m[2][0] * M->m[3][3] -
2137 M->m[1][2] * M->m[3][0] * M->m[2][3] -
2138 M->m[1][3] * M->m[2][0] * M->m[3][2] +
2139 M->m[1][3] * M->m[3][0] * M->m[2][2];
2140
2141 D->m[1][1] =
2142 M->m[0][0] * M->m[2][2] * M->m[3][3] -
2143 M->m[0][0] * M->m[3][2] * M->m[2][3] -
2144 M->m[0][2] * M->m[2][0] * M->m[3][3] +
2145 M->m[0][2] * M->m[3][0] * M->m[2][3] +
2146 M->m[0][3] * M->m[2][0] * M->m[3][2] -
2147 M->m[0][3] * M->m[3][0] * M->m[2][2];
2148
2149 D->m[1][2] =
2150 -M->m[0][0] * M->m[1][2] * M->m[3][3] +
2151 M->m[0][0] * M->m[3][2] * M->m[1][3] +
2152 M->m[0][2] * M->m[1][0] * M->m[3][3] -
2153 M->m[0][2] * M->m[3][0] * M->m[1][3] -
2154 M->m[0][3] * M->m[1][0] * M->m[3][2] +
2155 M->m[0][3] * M->m[3][0] * M->m[1][2];
2156
2157 D->m[1][3] =
2158 M->m[0][0] * M->m[1][2] * M->m[2][3] -
2159 M->m[0][0] * M->m[2][2] * M->m[1][3] -
2160 M->m[0][2] * M->m[1][0] * M->m[2][3] +
2161 M->m[0][2] * M->m[2][0] * M->m[1][3] +
2162 M->m[0][3] * M->m[1][0] * M->m[2][2] -
2163 M->m[0][3] * M->m[2][0] * M->m[1][2];
2164
2165 D->m[2][0] =
2166 M->m[1][0] * M->m[2][1] * M->m[3][3] -
2167 M->m[1][0] * M->m[3][1] * M->m[2][3] -
2168 M->m[1][1] * M->m[2][0] * M->m[3][3] +
2169 M->m[1][1] * M->m[3][0] * M->m[2][3] +
2170 M->m[1][3] * M->m[2][0] * M->m[3][1] -
2171 M->m[1][3] * M->m[3][0] * M->m[2][1];
2172
2173 D->m[2][1] =
2174 -M->m[0][0] * M->m[2][1] * M->m[3][3] +
2175 M->m[0][0] * M->m[3][1] * M->m[2][3] +
2176 M->m[0][1] * M->m[2][0] * M->m[3][3] -
2177 M->m[0][1] * M->m[3][0] * M->m[2][3] -
2178 M->m[0][3] * M->m[2][0] * M->m[3][1] +
2179 M->m[0][3] * M->m[3][0] * M->m[2][1];
2180
2181 D->m[2][2] =
2182 M->m[0][0] * M->m[1][1] * M->m[3][3] -
2183 M->m[0][0] * M->m[3][1] * M->m[1][3] -
2184 M->m[0][1] * M->m[1][0] * M->m[3][3] +
2185 M->m[0][1] * M->m[3][0] * M->m[1][3] +
2186 M->m[0][3] * M->m[1][0] * M->m[3][1] -
2187 M->m[0][3] * M->m[3][0] * M->m[1][1];
2188
2189 D->m[2][3] =
2190 -M->m[0][0] * M->m[1][1] * M->m[2][3] +
2191 M->m[0][0] * M->m[2][1] * M->m[1][3] +
2192 M->m[0][1] * M->m[1][0] * M->m[2][3] -
2193 M->m[0][1] * M->m[2][0] * M->m[1][3] -
2194 M->m[0][3] * M->m[1][0] * M->m[2][1] +
2195 M->m[0][3] * M->m[2][0] * M->m[1][1];
2196
2197 D->m[3][0] =
2198 -M->m[1][0] * M->m[2][1] * M->m[3][2] +
2199 M->m[1][0] * M->m[3][1] * M->m[2][2] +
2200 M->m[1][1] * M->m[2][0] * M->m[3][2] -
2201 M->m[1][1] * M->m[3][0] * M->m[2][2] -
2202 M->m[1][2] * M->m[2][0] * M->m[3][1] +
2203 M->m[1][2] * M->m[3][0] * M->m[2][1];
2204
2205 D->m[3][1] =
2206 M->m[0][0] * M->m[2][1] * M->m[3][2] -
2207 M->m[0][0] * M->m[3][1] * M->m[2][2] -
2208 M->m[0][1] * M->m[2][0] * M->m[3][2] +
2209 M->m[0][1] * M->m[3][0] * M->m[2][2] +
2210 M->m[0][2] * M->m[2][0] * M->m[3][1] -
2211 M->m[0][2] * M->m[3][0] * M->m[2][1];
2212
2213 D->m[3][2] =
2214 -M->m[0][0] * M->m[1][1] * M->m[3][2] +
2215 M->m[0][0] * M->m[3][1] * M->m[1][2] +
2216 M->m[0][1] * M->m[1][0] * M->m[3][2] -
2217 M->m[0][1] * M->m[3][0] * M->m[1][2] -
2218 M->m[0][2] * M->m[1][0] * M->m[3][1] +
2219 M->m[0][2] * M->m[3][0] * M->m[1][1];
2220
2221 D->m[3][3] =
2222 M->m[0][0] * M->m[1][1] * M->m[2][2] -
2223 M->m[0][0] * M->m[2][1] * M->m[1][2] -
2224 M->m[0][1] * M->m[1][0] * M->m[2][2] +
2225 M->m[0][1] * M->m[2][0] * M->m[1][2] +
2226 M->m[0][2] * M->m[1][0] * M->m[2][1] -
2227 M->m[0][2] * M->m[2][0] * M->m[1][1];
2228
2229 det =
2230 M->m[0][0] * D->m[0][0] +
2231 M->m[1][0] * D->m[0][1] +
2232 M->m[2][0] * D->m[0][2] +
2233 M->m[3][0] * D->m[0][3];
2234
2235 det = 1.0 / det;
2236
2237 for (i = 0; i < 4; i++)
2238 for (k = 0; k < 4; k++)
2239 D->m[i][k] *= det;
2240
2241 #ifdef DEBUG
2242 {
2243 D3DMATRIX I;
2244
2245 nine_d3d_matrix_matrix_mul(&I, D, M);
2246
2247 for (i = 0; i < 4; ++i)
2248 for (k = 0; k < 4; ++k)
2249 if (fabsf(I.m[i][k] - (float)(i == k)) > 1e-3)
2250 DBG("Matrix inversion check FAILED !\n");
2251 }
2252 #endif
2253 }
2254
2255 /* TODO: don't use 4x4 inverse, unless this gets all nicely inlined ? */
2256 void
2257 nine_d3d_matrix_inverse_3x3(D3DMATRIX *D, const D3DMATRIX *M)
2258 {
2259 D3DMATRIX T;
2260 unsigned i, j;
2261
2262 for (i = 0; i < 3; ++i)
2263 for (j = 0; j < 3; ++j)
2264 T.m[i][j] = M->m[i][j];
2265 for (i = 0; i < 3; ++i) {
2266 T.m[i][3] = 0.0f;
2267 T.m[3][i] = 0.0f;
2268 }
2269 T.m[3][3] = 1.0f;
2270
2271 nine_d3d_matrix_inverse(D, &T);
2272 }