projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
st/nine: Replace r[0] with tmp
[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 specular_enable : 1;
62 uint32_t pad1 : 6;
63 uint32_t tc_dim_input: 16; /* 8 * 2 bits */
64 uint32_t pad2 : 16;
65 uint32_t tc_dim_output: 24; /* 8 * 3 bits */
66 uint32_t pad3 : 8;
67 uint32_t tc_gen : 24; /* 8 * 3 bits */
68 uint32_t pad4 : 8;
69 uint32_t tc_idx : 24;
70 uint32_t pad5 : 8;
71 uint32_t passthrough;
72 };
73 uint64_t value64[3]; /* don't forget to resize VertexShader9.ff_key */
74 uint32_t value32[6];
75 };
76 };
77
78 /* Texture stage state:
79 *
80 * COLOROP D3DTOP 5 bit
81 * ALPHAOP D3DTOP 5 bit
82 * COLORARG0 D3DTA 3 bit
83 * COLORARG1 D3DTA 3 bit
84 * COLORARG2 D3DTA 3 bit
85 * ALPHAARG0 D3DTA 3 bit
86 * ALPHAARG1 D3DTA 3 bit
87 * ALPHAARG2 D3DTA 3 bit
88 * RESULTARG D3DTA 1 bit (CURRENT:0 or TEMP:1)
89 * TEXCOORDINDEX 0 - 7 3 bit
90 * ===========================
91 * 32 bit per stage
92 */
93 struct nine_ff_ps_key
94 {
95 union {
96 struct {
97 struct {
98 uint32_t colorop : 5;
99 uint32_t alphaop : 5;
100 uint32_t colorarg0 : 3;
101 uint32_t colorarg1 : 3;
102 uint32_t colorarg2 : 3;
103 uint32_t alphaarg0 : 3;
104 uint32_t alphaarg1 : 3;
105 uint32_t alphaarg2 : 3;
106 uint32_t resultarg : 1; /* CURRENT:0 or TEMP:1 */
107 uint32_t textarget : 2; /* 1D/2D/3D/CUBE */
108 uint32_t pad : 1;
109 /* that's 32 bit exactly */
110 } ts[8];
111 uint32_t projected : 16;
112 uint32_t fog : 1; /* for vFog coming from VS */
113 uint32_t fog_mode : 2;
114 uint32_t specular : 1;
115 uint32_t pad1 : 12; /* 9 32-bit words with this */
116 uint8_t colorarg_b4[3];
117 uint8_t colorarg_b5[3];
118 uint8_t alphaarg_b4[3]; /* 11 32-bit words plus a byte */
119 uint8_t pad2[3];
120 };
121 uint64_t value64[6]; /* don't forget to resize PixelShader9.ff_key */
122 uint32_t value32[12];
123 };
124 };
125
126 static unsigned nine_ff_vs_key_hash(void *key)
127 {
128 struct nine_ff_vs_key *vs = key;
129 unsigned i;
130 uint32_t hash = vs->value32[0];
131 for (i = 1; i < Elements(vs->value32); ++i)
132 hash ^= vs->value32[i];
133 return hash;
134 }
135 static int nine_ff_vs_key_comp(void *key1, void *key2)
136 {
137 struct nine_ff_vs_key *a = (struct nine_ff_vs_key *)key1;
138 struct nine_ff_vs_key *b = (struct nine_ff_vs_key *)key2;
139
140 return memcmp(a->value64, b->value64, sizeof(a->value64));
141 }
142 static unsigned nine_ff_ps_key_hash(void *key)
143 {
144 struct nine_ff_ps_key *ps = key;
145 unsigned i;
146 uint32_t hash = ps->value32[0];
147 for (i = 1; i < Elements(ps->value32); ++i)
148 hash ^= ps->value32[i];
149 return hash;
150 }
151 static int nine_ff_ps_key_comp(void *key1, void *key2)
152 {
153 struct nine_ff_ps_key *a = (struct nine_ff_ps_key *)key1;
154 struct nine_ff_ps_key *b = (struct nine_ff_ps_key *)key2;
155
156 return memcmp(a->value64, b->value64, sizeof(a->value64));
157 }
158 static unsigned nine_ff_fvf_key_hash(void *key)
159 {
160 return *(DWORD *)key;
161 }
162 static int nine_ff_fvf_key_comp(void *key1, void *key2)
163 {
164 return *(DWORD *)key1 != *(DWORD *)key2;
165 }
166
167 static void nine_ff_prune_vs(struct NineDevice9 *);
168 static void nine_ff_prune_ps(struct NineDevice9 *);
169
170 static void nine_ureg_tgsi_dump(struct ureg_program *ureg, boolean override)
171 {
172 if (debug_get_bool_option("NINE_FF_DUMP", FALSE) || override) {
173 unsigned count;
174 const struct tgsi_token *toks = ureg_get_tokens(ureg, &count);
175 tgsi_dump(toks, 0);
176 ureg_free_tokens(toks);
177 }
178 }
179
180 #define _X(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_X)
181 #define _Y(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_Y)
182 #define _Z(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_Z)
183 #define _W(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_W)
184
185 #define _XXXX(r) ureg_scalar(r, TGSI_SWIZZLE_X)
186 #define _YYYY(r) ureg_scalar(r, TGSI_SWIZZLE_Y)
187 #define _ZZZZ(r) ureg_scalar(r, TGSI_SWIZZLE_Z)
188 #define _WWWW(r) ureg_scalar(r, TGSI_SWIZZLE_W)
189
190 #define _XYZW(r) (r)
191
192 /* AL should contain base address of lights table. */
193 #define LIGHT_CONST(i) \
194 ureg_src_indirect(ureg_DECL_constant(ureg, i), _X(AL))
195
196 #define MATERIAL_CONST(i) \
197 ureg_DECL_constant(ureg, 19 + (i))
198
199 #define _CONST(n) ureg_DECL_constant(ureg, n)
200
201 /* VS FF constants layout:
202 *
203 * CONST[ 0.. 3] D3DTS_WORLD * D3DTS_VIEW * D3DTS_PROJECTION
204 * CONST[ 4.. 7] D3DTS_WORLD * D3DTS_VIEW
205 * CONST[ 8..11] D3DTS_VIEW * D3DTS_PROJECTION
206 * CONST[12..15] D3DTS_VIEW
207 * CONST[16..18] Normal matrix
208 *
209 * CONST[19] MATERIAL.Emissive + Material.Ambient * RS.Ambient
210 * CONST[20] MATERIAL.Diffuse
211 * CONST[21] MATERIAL.Ambient
212 * CONST[22] MATERIAL.Specular
213 * CONST[23].x___ MATERIAL.Power
214 * CONST[24] MATERIAL.Emissive
215 * CONST[25] RS.Ambient
216 *
217 * CONST[26].x___ RS.PointSizeMin
218 * CONST[26]._y__ RS.PointSizeMax
219 * CONST[26].__z_ RS.PointSize
220 * CONST[26].___w RS.PointScaleA
221 * CONST[27].x___ RS.PointScaleB
222 * CONST[27]._y__ RS.PointScaleC
223 *
224 * CONST[28].x___ RS.FogEnd
225 * CONST[28]._y__ 1.0f / (RS.FogEnd - RS.FogStart)
226 * CONST[28].__z_ RS.FogDensity
227
228 * CONST[30].x___ TWEENFACTOR
229 *
230 * CONST[32].x___ LIGHT[0].Type
231 * CONST[32]._yzw LIGHT[0].Attenuation0,1,2
232 * CONST[33] LIGHT[0].Diffuse
233 * CONST[34] LIGHT[0].Specular
234 * CONST[35] LIGHT[0].Ambient
235 * CONST[36].xyz_ LIGHT[0].Position
236 * CONST[36].___w LIGHT[0].Range
237 * CONST[37].xyz_ LIGHT[0].Direction
238 * CONST[37].___w LIGHT[0].Falloff
239 * CONST[38].x___ cos(LIGHT[0].Theta / 2)
240 * CONST[38]._y__ cos(LIGHT[0].Phi / 2)
241 * CONST[38].__z_ 1.0f / (cos(LIGHT[0].Theta / 2) - cos(Light[0].Phi / 2))
242 * CONST[39].xyz_ LIGHT[0].HalfVector (for directional lights)
243 * CONST[39].___w 1 if this is the last active light, 0 if not
244 * CONST[40] LIGHT[1]
245 * CONST[48] LIGHT[2]
246 * CONST[56] LIGHT[3]
247 * CONST[64] LIGHT[4]
248 * CONST[72] LIGHT[5]
249 * CONST[80] LIGHT[6]
250 * CONST[88] LIGHT[7]
251 * NOTE: no lighting code is generated if there are no active lights
252 *
253 * CONST[100].x___ Viewport 2/width
254 * CONST[100]._y__ Viewport 2/height
255 * CONST[100].__z_ Viewport 1/(zmax - zmin)
256 * CONST[101].x___ Viewport x0
257 * CONST[101]._y__ Viewport y0
258 * CONST[101].__z_ Viewport z0
259 *
260 * CONST[128..131] D3DTS_TEXTURE0
261 * CONST[132..135] D3DTS_TEXTURE1
262 * CONST[136..139] D3DTS_TEXTURE2
263 * CONST[140..143] D3DTS_TEXTURE3
264 * CONST[144..147] D3DTS_TEXTURE4
265 * CONST[148..151] D3DTS_TEXTURE5
266 * CONST[152..155] D3DTS_TEXTURE6
267 * CONST[156..159] D3DTS_TEXTURE7
268 *
269 * CONST[224] D3DTS_WORLDMATRIX[0]
270 * CONST[228] D3DTS_WORLDMATRIX[1]
271 * ...
272 * CONST[252] D3DTS_WORLDMATRIX[7]
273 */
274 struct vs_build_ctx
275 {
276 struct ureg_program *ureg;
277 const struct nine_ff_vs_key *key;
278
279 uint16_t input[PIPE_MAX_ATTRIBS];
280 unsigned num_inputs;
281
282 struct ureg_src aVtx;
283 struct ureg_src aNrm;
284 struct ureg_src aCol[2];
285 struct ureg_src aTex[8];
286 struct ureg_src aPsz;
287 struct ureg_src aInd;
288 struct ureg_src aWgt;
289
290 struct ureg_src aVtx1; /* tweening */
291 struct ureg_src aNrm1;
292
293 struct ureg_src mtlA;
294 struct ureg_src mtlD;
295 struct ureg_src mtlS;
296 struct ureg_src mtlE;
297 };
298
299 static inline unsigned
300 get_texcoord_sn(struct pipe_screen *screen)
301 {
302 if (screen->get_param(screen, PIPE_CAP_TGSI_TEXCOORD))
303 return TGSI_SEMANTIC_TEXCOORD;
304 return TGSI_SEMANTIC_GENERIC;
305 }
306
307 static inline struct ureg_src
308 build_vs_add_input(struct vs_build_ctx *vs, uint16_t ndecl)
309 {
310 const unsigned i = vs->num_inputs++;
311 assert(i < PIPE_MAX_ATTRIBS);
312 vs->input[i] = ndecl;
313 return ureg_DECL_vs_input(vs->ureg, i);
314 }
315
316 /* NOTE: dst may alias src */
317 static inline void
318 ureg_normalize3(struct ureg_program *ureg,
319 struct ureg_dst dst, struct ureg_src src,
320 struct ureg_dst tmp)
321 {
322 #ifdef NINE_TGSI_LAZY_DEVS
323 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
324
325 ureg_DP3(ureg, tmp_x, src, src);
326 ureg_RSQ(ureg, tmp_x, _X(tmp));
327 ureg_MUL(ureg, dst, src, _X(tmp));
328 #else
329 ureg_NRM(ureg, dst, src);
330 #endif
331 }
332
333 static void *
334 nine_ff_build_vs(struct NineDevice9 *device, struct vs_build_ctx *vs)
335 {
336 const struct nine_ff_vs_key *key = vs->key;
337 struct ureg_program *ureg = ureg_create(TGSI_PROCESSOR_VERTEX);
338 struct ureg_dst oPos, oCol[2], oPsz, oFog;
339 struct ureg_dst rVtx, rNrm;
340 struct ureg_dst r[8];
341 struct ureg_dst AR;
342 struct ureg_dst tmp, tmp_x, tmp_y, tmp_z;
343 unsigned i, c;
344 unsigned label[32], l = 0;
345 unsigned num_r = 8;
346 boolean need_rNrm = key->lighting || key->pointscale || key->passthrough & (1 << NINE_DECLUSAGE_NORMAL);
347 boolean need_rVtx = key->lighting || key->fog_mode;
348 const unsigned texcoord_sn = get_texcoord_sn(device->screen);
349
350 vs->ureg = ureg;
351
352 /* Check which inputs we should transform. */
353 for (i = 0; i < 8 * 3; i += 3) {
354 switch ((key->tc_gen >> i) & 0x3) {
355 case NINED3DTSS_TCI_CAMERASPACENORMAL:
356 need_rNrm = TRUE;
357 break;
358 case NINED3DTSS_TCI_CAMERASPACEPOSITION:
359 need_rVtx = TRUE;
360 break;
361 case NINED3DTSS_TCI_CAMERASPACEREFLECTIONVECTOR:
362 need_rVtx = need_rNrm = TRUE;
363 break;
364 default:
365 break;
366 }
367 }
368
369 /* Declare and record used inputs (needed for linkage with vertex format):
370 * (texture coordinates handled later)
371 */
372 vs->aVtx = build_vs_add_input(vs,
373 key->position_t ? NINE_DECLUSAGE_POSITIONT : NINE_DECLUSAGE_POSITION);
374
375 if (need_rNrm)
376 vs->aNrm = build_vs_add_input(vs, NINE_DECLUSAGE_NORMAL);
377
378 vs->aCol[0] = ureg_imm1f(ureg, 1.0f);
379 vs->aCol[1] = ureg_imm1f(ureg, 1.0f);
380
381 if (key->lighting || key->darkness) {
382 const unsigned mask = key->mtl_diffuse | key->mtl_specular |
383 key->mtl_ambient | key->mtl_emissive;
384 if ((mask & 0x1) && !key->color0in_one)
385 vs->aCol[0] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 0));
386 if ((mask & 0x2) && !key->color1in_one)
387 vs->aCol[1] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 1));
388
389 vs->mtlD = MATERIAL_CONST(1);
390 vs->mtlA = MATERIAL_CONST(2);
391 vs->mtlS = MATERIAL_CONST(3);
392 vs->mtlE = MATERIAL_CONST(5);
393 if (key->mtl_diffuse == 1) vs->mtlD = vs->aCol[0]; else
394 if (key->mtl_diffuse == 2) vs->mtlD = vs->aCol[1];
395 if (key->mtl_ambient == 1) vs->mtlA = vs->aCol[0]; else
396 if (key->mtl_ambient == 2) vs->mtlA = vs->aCol[1];
397 if (key->mtl_specular == 1) vs->mtlS = vs->aCol[0]; else
398 if (key->mtl_specular == 2) vs->mtlS = vs->aCol[1];
399 if (key->mtl_emissive == 1) vs->mtlE = vs->aCol[0]; else
400 if (key->mtl_emissive == 2) vs->mtlE = vs->aCol[1];
401 } else {
402 if (!key->color0in_one) vs->aCol[0] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 0));
403 if (!key->color1in_one) vs->aCol[1] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 1));
404 }
405
406 if (key->vertexpointsize)
407 vs->aPsz = build_vs_add_input(vs, NINE_DECLUSAGE_PSIZE);
408
409 if (key->vertexblend_indexed || key->passthrough & (1 << NINE_DECLUSAGE_BLENDINDICES))
410 vs->aInd = build_vs_add_input(vs, NINE_DECLUSAGE_BLENDINDICES);
411 if (key->vertexblend || key->passthrough & (1 << NINE_DECLUSAGE_BLENDWEIGHT))
412 vs->aWgt = build_vs_add_input(vs, NINE_DECLUSAGE_BLENDWEIGHT);
413 if (key->vertextween) {
414 vs->aVtx1 = build_vs_add_input(vs, NINE_DECLUSAGE_i(POSITION,1));
415 vs->aNrm1 = build_vs_add_input(vs, NINE_DECLUSAGE_i(NORMAL,1));
416 }
417
418 /* Declare outputs:
419 */
420 oPos = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0); /* HPOS */
421 oCol[0] = ureg_saturate(ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0));
422 oCol[1] = ureg_saturate(ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 1));
423 if (key->fog || key->passthrough & (1 << NINE_DECLUSAGE_FOG)) {
424 oFog = ureg_DECL_output(ureg, TGSI_SEMANTIC_FOG, 0);
425 oFog = ureg_writemask(oFog, TGSI_WRITEMASK_X);
426 }
427
428 if (key->vertexpointsize || key->pointscale) {
429 oPsz = ureg_DECL_output_masked(ureg, TGSI_SEMANTIC_PSIZE, 0,
430 TGSI_WRITEMASK_X, 0, 1);
431 oPsz = ureg_writemask(oPsz, TGSI_WRITEMASK_X);
432 }
433
434 /* Declare TEMPs:
435 */
436 for (i = 0; i < num_r; ++i)
437 r[i] = ureg_DECL_local_temporary(ureg);
438 tmp = r[0];
439 tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
440 tmp_y = ureg_writemask(tmp, TGSI_WRITEMASK_Y);
441 tmp_z = ureg_writemask(tmp, TGSI_WRITEMASK_Z);
442 if (key->lighting || key->vertexblend)
443 AR = ureg_DECL_address(ureg);
444
445 rVtx = ureg_writemask(r[1], TGSI_WRITEMASK_XYZ);
446 rNrm = ureg_writemask(r[2], TGSI_WRITEMASK_XYZ);
447
448 /* === Vertex transformation / vertex blending:
449 */
450 if (key->vertextween) {
451 assert(!key->vertexblend);
452 ureg_LRP(ureg, r[2], _XXXX(_CONST(30)), vs->aVtx, vs->aVtx1);
453 if (need_rNrm)
454 ureg_LRP(ureg, r[3], _XXXX(_CONST(30)), vs->aNrm, vs->aNrm1);
455 vs->aVtx = ureg_src(r[2]);
456 vs->aNrm = ureg_src(r[3]);
457 }
458
459 if (key->vertexblend) {
460 struct ureg_src cWM[4];
461
462 for (i = 224; i <= 255; ++i)
463 ureg_DECL_constant(ureg, i);
464
465 /* translate world matrix index to constant file index */
466 if (key->vertexblend_indexed) {
467 ureg_MAD(ureg, tmp, vs->aInd, ureg_imm1f(ureg, 4.0f), ureg_imm1f(ureg, 224.0f));
468 ureg_ARL(ureg, AR, ureg_src(tmp));
469 }
470 for (i = 0; i < key->vertexblend; ++i) {
471 for (c = 0; c < 4; ++c) {
472 cWM[c] = ureg_src_register(TGSI_FILE_CONSTANT, (224 + i * 4) * !key->vertexblend_indexed + c);
473 if (key->vertexblend_indexed)
474 cWM[c] = ureg_src_indirect(cWM[c], ureg_scalar(ureg_src(AR), i));
475 }
476 /* multiply by WORLD(index) */
477 ureg_MUL(ureg, tmp, _XXXX(vs->aVtx), cWM[0]);
478 ureg_MAD(ureg, tmp, _YYYY(vs->aVtx), cWM[1], ureg_src(tmp));
479 ureg_MAD(ureg, tmp, _ZZZZ(vs->aVtx), cWM[2], ureg_src(tmp));
480 ureg_MAD(ureg, tmp, _WWWW(vs->aVtx), cWM[3], ureg_src(tmp));
481
482 /* accumulate weighted position value */
483 if (i)
484 ureg_MAD(ureg, r[2], ureg_src(tmp), ureg_scalar(vs->aWgt, i), ureg_src(r[2]));
485 else
486 ureg_MUL(ureg, r[2], ureg_src(tmp), ureg_scalar(vs->aWgt, 0));
487 }
488 /* multiply by VIEW_PROJ */
489 ureg_MUL(ureg, tmp, _X(r[2]), _CONST(8));
490 ureg_MAD(ureg, tmp, _Y(r[2]), _CONST(9), ureg_src(tmp));
491 ureg_MAD(ureg, tmp, _Z(r[2]), _CONST(10), ureg_src(tmp));
492 ureg_MAD(ureg, oPos, _W(r[2]), _CONST(11), ureg_src(tmp));
493
494 if (need_rVtx)
495 vs->aVtx = ureg_src(r[2]);
496 } else
497 if (key->position_t && device->driver_caps.window_space_position_support) {
498 ureg_MOV(ureg, oPos, vs->aVtx);
499 } else if (key->position_t) {
500 /* vs->aVtx contains the coordinates buffer wise.
501 * later in the pipeline, clipping, viewport and division
502 * by w (rhw = 1/w) are going to be applied, so do the reverse
503 * of these transformations (except clipping) to have the good
504 * position at the end.*/
505 ureg_MOV(ureg, tmp, vs->aVtx);
506 /* X from [X_min, X_min + width] to [-1, 1], same for Y. Z to [0, 1] */
507 ureg_SUB(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _CONST(101));
508 ureg_MUL(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _CONST(100));
509 ureg_SUB(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XY), ureg_src(tmp), ureg_imm1f(ureg, 1.0f));
510 /* Y needs to be reversed */
511 ureg_MOV(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_Y), ureg_negate(ureg_src(tmp)));
512 /* inverse rhw */
513 ureg_RCP(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_W), _W(tmp));
514 /* multiply X, Y, Z by w */
515 ureg_MUL(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _W(tmp));
516 ureg_MOV(ureg, oPos, ureg_src(tmp));
517 } else {
518 /* position = vertex * WORLD_VIEW_PROJ */
519 ureg_MUL(ureg, tmp, _XXXX(vs->aVtx), _CONST(0));
520 ureg_MAD(ureg, tmp, _YYYY(vs->aVtx), _CONST(1), ureg_src(tmp));
521 ureg_MAD(ureg, tmp, _ZZZZ(vs->aVtx), _CONST(2), ureg_src(tmp));
522 ureg_MAD(ureg, oPos, _WWWW(vs->aVtx), _CONST(3), ureg_src(tmp));
523 }
524
525 if (need_rVtx) {
526 ureg_MUL(ureg, rVtx, _XXXX(vs->aVtx), _CONST(4));
527 ureg_MAD(ureg, rVtx, _YYYY(vs->aVtx), _CONST(5), ureg_src(rVtx));
528 ureg_MAD(ureg, rVtx, _ZZZZ(vs->aVtx), _CONST(6), ureg_src(rVtx));
529 ureg_MAD(ureg, rVtx, _WWWW(vs->aVtx), _CONST(7), ureg_src(rVtx));
530 }
531 if (need_rNrm) {
532 ureg_MUL(ureg, rNrm, _XXXX(vs->aNrm), _CONST(16));
533 ureg_MAD(ureg, rNrm, _YYYY(vs->aNrm), _CONST(17), ureg_src(rNrm));
534 ureg_MAD(ureg, rNrm, _ZZZZ(vs->aNrm), _CONST(18), ureg_src(rNrm));
535 ureg_normalize3(ureg, rNrm, ureg_src(rNrm), tmp);
536 }
537 /* NOTE: don't use vs->aVtx, vs->aNrm after this line */
538
539 /* === Process point size:
540 */
541 if (key->vertexpointsize) {
542 struct ureg_src cPsz1 = ureg_DECL_constant(ureg, 26);
543 #ifdef NINE_TGSI_LAZY_DEVS
544 struct ureg_dst tmp_clamp = ureg_DECL_temporary(ureg);
545
546 ureg_MAX(ureg, tmp_clamp, vs->aPsz, _XXXX(cPsz1));
547 ureg_MIN(ureg, oPsz, ureg_src(tmp_clamp), _YYYY(cPsz1));
548 ureg_release_temporary(ureg, tmp_clamp);
549 #else
550 ureg_CLAMP(ureg, oPsz, vs->aPsz, _XXXX(cPsz1), _YYYY(cPsz1));
551 #endif
552 } else if (key->pointscale) {
553 struct ureg_src cPsz1 = ureg_DECL_constant(ureg, 26);
554 struct ureg_src cPsz2 = ureg_DECL_constant(ureg, 27);
555
556 ureg_DP3(ureg, tmp_x, ureg_src(r[1]), ureg_src(r[1]));
557 ureg_SQRT(ureg, tmp_y, _X(tmp));
558 ureg_MAD(ureg, tmp_x, _Y(tmp), _YYYY(cPsz2), _XXXX(cPsz2));
559 ureg_MAD(ureg, tmp_x, _Y(tmp), _X(tmp), _WWWW(cPsz1));
560 ureg_RCP(ureg, tmp_x, ureg_src(tmp));
561 ureg_MUL(ureg, tmp_x, ureg_src(tmp), _ZZZZ(cPsz1));
562 #ifdef NINE_TGSI_LAZY_DEVS
563 struct ureg_dst tmp_clamp = ureg_DECL_temporary(ureg);
564
565 ureg_MAX(ureg, tmp_clamp, _X(tmp), _XXXX(cPsz1));
566 ureg_MIN(ureg, oPsz, ureg_src(tmp_clamp), _YYYY(cPsz1));
567 ureg_release_temporary(ureg, tmp_clamp);
568 #else
569 ureg_CLAMP(ureg, oPsz, _X(tmp), _XXXX(cPsz1), _YYYY(cPsz1));
570 #endif
571 }
572
573 for (i = 0; i < 8; ++i) {
574 struct ureg_dst oTex, input_coord, transformed, t;
575 unsigned c, writemask;
576 const unsigned tci = (key->tc_gen >> (i * 3)) & 0x7;
577 const unsigned idx = (key->tc_idx >> (i * 3)) & 0x7;
578 unsigned dim_input = 1 + ((key->tc_dim_input >> (i * 2)) & 0x3);
579 const unsigned dim_output = (key->tc_dim_output >> (i * 3)) & 0x7;
580
581 /* No texture output of index s */
582 if (tci == NINED3DTSS_TCI_DISABLE)
583 continue;
584 oTex = ureg_DECL_output(ureg, texcoord_sn, i);
585 input_coord = r[5];
586 transformed = r[6];
587
588 /* Get the coordinate */
589 switch (tci) {
590 case NINED3DTSS_TCI_PASSTHRU:
591 /* NINED3DTSS_TCI_PASSTHRU => Use texcoord coming from index idx *
592 * Else the idx is used only to determine wrapping mode. */
593 vs->aTex[idx] = build_vs_add_input(vs, NINE_DECLUSAGE_i(TEXCOORD,idx));
594 ureg_MOV(ureg, input_coord, vs->aTex[idx]);
595 break;
596 case NINED3DTSS_TCI_CAMERASPACENORMAL:
597 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XYZ), ureg_src(rNrm));
598 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
599 dim_input = 4;
600 break;
601 case NINED3DTSS_TCI_CAMERASPACEPOSITION:
602 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XYZ), ureg_src(rVtx));
603 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
604 dim_input = 4;
605 break;
606 case NINED3DTSS_TCI_CAMERASPACEREFLECTIONVECTOR:
607 tmp.WriteMask = TGSI_WRITEMASK_XYZ;
608 ureg_DP3(ureg, tmp_x, ureg_src(rVtx), ureg_src(rNrm));
609 ureg_MUL(ureg, tmp, ureg_src(rNrm), _X(tmp));
610 ureg_ADD(ureg, tmp, ureg_src(tmp), ureg_src(tmp));
611 ureg_SUB(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XYZ), ureg_src(rVtx), ureg_src(tmp));
612 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
613 dim_input = 4;
614 tmp.WriteMask = TGSI_WRITEMASK_XYZW;
615 break;
616 case NINED3DTSS_TCI_SPHEREMAP:
617 assert(!"TODO");
618 break;
619 default:
620 assert(0);
621 break;
622 }
623
624 /* Apply the transformation */
625 /* dim_output == 0 => do not transform the components.
626 * XYZRHW also disables transformation */
627 if (!dim_output || key->position_t) {
628 transformed = input_coord;
629 writemask = TGSI_WRITEMASK_XYZW;
630 } else {
631 for (c = 0; c < dim_output; c++) {
632 t = ureg_writemask(transformed, 1 << c);
633 switch (dim_input) {
634 /* dim_input = 1 2 3: -> we add trailing 1 to input*/
635 case 1: ureg_MAD(ureg, t, _X(input_coord), _XXXX(_CONST(128 + i * 4 + c)), _YYYY(_CONST(128 + i * 4 + c)));
636 break;
637 case 2: ureg_DP2(ureg, t, ureg_src(input_coord), _CONST(128 + i * 4 + c));
638 ureg_ADD(ureg, t, ureg_src(transformed), _ZZZZ(_CONST(128 + i * 4 + c)));
639 break;
640 case 3: ureg_DP3(ureg, t, ureg_src(input_coord), _CONST(128 + i * 4 + c));
641 ureg_ADD(ureg, t, ureg_src(transformed), _WWWW(_CONST(128 + i * 4 + c)));
642 break;
643 case 4: ureg_DP4(ureg, t, ureg_src(input_coord), _CONST(128 + i * 4 + c)); break;
644 default:
645 assert(0);
646 }
647 }
648 writemask = (1 << dim_output) - 1;
649 }
650
651 ureg_MOV(ureg, ureg_writemask(oTex, writemask), ureg_src(transformed));
652 }
653
654 /* === Lighting:
655 *
656 * DIRECTIONAL: Light at infinite distance, parallel rays, no attenuation.
657 * POINT: Finite distance to scene, divergent rays, isotropic, attenuation.
658 * SPOT: Finite distance, divergent rays, angular dependence, attenuation.
659 *
660 * vec3 normal = normalize(in.Normal * NormalMatrix);
661 * vec3 hitDir = light.direction;
662 * float atten = 1.0;
663 *
664 * if (light.type != DIRECTIONAL)
665 * {
666 * vec3 hitVec = light.position - eyeVertex;
667 * float d = length(hitVec);
668 * hitDir = hitVec / d;
669 * atten = 1 / ((light.atten2 * d + light.atten1) * d + light.atten0);
670 * }
671 *
672 * if (light.type == SPOTLIGHT)
673 * {
674 * float rho = dp3(-hitVec, light.direction);
675 * if (rho < cos(light.phi / 2))
676 * atten = 0;
677 * if (rho < cos(light.theta / 2))
678 * atten *= pow(some_func(rho), light.falloff);
679 * }
680 *
681 * float nDotHit = dp3_sat(normal, hitVec);
682 * float powFact = 0.0;
683 *
684 * if (nDotHit > 0.0)
685 * {
686 * vec3 midVec = normalize(hitDir + eye);
687 * float nDotMid = dp3_sat(normal, midVec);
688 * pFact = pow(nDotMid, material.power);
689 * }
690 *
691 * ambient += light.ambient * atten;
692 * diffuse += light.diffuse * atten * nDotHit;
693 * specular += light.specular * atten * powFact;
694 */
695 if (key->lighting) {
696 struct ureg_dst rAtt = ureg_writemask(r[1], TGSI_WRITEMASK_W);
697 struct ureg_dst rHit = ureg_writemask(r[3], TGSI_WRITEMASK_XYZ);
698 struct ureg_dst rMid = ureg_writemask(r[4], TGSI_WRITEMASK_XYZ);
699
700 struct ureg_dst rCtr = ureg_writemask(r[2], TGSI_WRITEMASK_W);
701
702 struct ureg_dst AL = ureg_writemask(AR, TGSI_WRITEMASK_X);
703
704 /* Light.*.Alpha is not used. */
705 struct ureg_dst rD = ureg_writemask(r[5], TGSI_WRITEMASK_XYZ);
706 struct ureg_dst rA = ureg_writemask(r[6], TGSI_WRITEMASK_XYZ);
707 struct ureg_dst rS = ureg_writemask(r[7], TGSI_WRITEMASK_XYZ);
708
709 struct ureg_src mtlP = _XXXX(MATERIAL_CONST(4));
710
711 struct ureg_src cLKind = _XXXX(LIGHT_CONST(0));
712 struct ureg_src cLAtt0 = _YYYY(LIGHT_CONST(0));
713 struct ureg_src cLAtt1 = _ZZZZ(LIGHT_CONST(0));
714 struct ureg_src cLAtt2 = _WWWW(LIGHT_CONST(0));
715 struct ureg_src cLColD = _XYZW(LIGHT_CONST(1));
716 struct ureg_src cLColS = _XYZW(LIGHT_CONST(2));
717 struct ureg_src cLColA = _XYZW(LIGHT_CONST(3));
718 struct ureg_src cLPos = _XYZW(LIGHT_CONST(4));
719 struct ureg_src cLRng = _WWWW(LIGHT_CONST(4));
720 struct ureg_src cLDir = _XYZW(LIGHT_CONST(5));
721 struct ureg_src cLFOff = _WWWW(LIGHT_CONST(5));
722 struct ureg_src cLTht = _XXXX(LIGHT_CONST(6));
723 struct ureg_src cLPhi = _YYYY(LIGHT_CONST(6));
724 struct ureg_src cLSDiv = _ZZZZ(LIGHT_CONST(6));
725 struct ureg_src cLLast = _WWWW(LIGHT_CONST(7));
726
727 const unsigned loop_label = l++;
728
729 ureg_MOV(ureg, rCtr, ureg_imm1f(ureg, 32.0f)); /* &lightconst(0) */
730 ureg_MOV(ureg, rD, ureg_imm1f(ureg, 0.0f));
731 ureg_MOV(ureg, rA, ureg_imm1f(ureg, 0.0f));
732 ureg_MOV(ureg, rS, ureg_imm1f(ureg, 0.0f));
733 rD = ureg_saturate(rD);
734 rA = ureg_saturate(rA);
735 rS = ureg_saturate(rS);
736
737
738 /* loop management */
739 ureg_BGNLOOP(ureg, &label[loop_label]);
740 ureg_ARL(ureg, AL, _W(rCtr));
741
742 /* if (not DIRECTIONAL light): */
743 ureg_SNE(ureg, tmp_x, cLKind, ureg_imm1f(ureg, D3DLIGHT_DIRECTIONAL));
744 ureg_MOV(ureg, rHit, ureg_negate(cLDir));
745 ureg_MOV(ureg, rAtt, ureg_imm1f(ureg, 1.0f));
746 ureg_IF(ureg, _X(tmp), &label[l++]);
747 {
748 /* hitDir = light.position - eyeVtx
749 * d = length(hitDir)
750 * hitDir /= d
751 */
752 ureg_SUB(ureg, rHit, cLPos, ureg_src(rVtx));
753 ureg_DP3(ureg, tmp_x, ureg_src(rHit), ureg_src(rHit));
754 ureg_RSQ(ureg, tmp_y, _X(tmp));
755 ureg_MUL(ureg, rHit, ureg_src(rHit), _Y(tmp)); /* normalize */
756 ureg_MUL(ureg, tmp_x, _X(tmp), _Y(tmp)); /* length */
757
758 /* att = 1.0 / (light.att0 + (light.att1 + light.att2 * d) * d) */
759 ureg_MAD(ureg, rAtt, _X(tmp), cLAtt2, cLAtt1);
760 ureg_MAD(ureg, rAtt, _X(tmp), _W(rAtt), cLAtt0);
761 ureg_RCP(ureg, rAtt, _W(rAtt));
762 /* cut-off if distance exceeds Light.Range */
763 ureg_SLT(ureg, tmp_x, _X(tmp), cLRng);
764 ureg_MUL(ureg, rAtt, _W(rAtt), _X(tmp));
765 }
766 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
767 ureg_ENDIF(ureg);
768
769 /* if (SPOT light) */
770 ureg_SEQ(ureg, tmp_x, cLKind, ureg_imm1f(ureg, D3DLIGHT_SPOT));
771 ureg_IF(ureg, _X(tmp), &label[l++]);
772 {
773 /* rho = dp3(-hitDir, light.spotDir)
774 *
775 * if (rho > light.ctht2) NOTE: 0 <= phi <= pi, 0 <= theta <= phi
776 * spotAtt = 1
777 * else
778 * if (rho <= light.cphi2)
779 * spotAtt = 0
780 * else
781 * spotAtt = (rho - light.cphi2) / (light.ctht2 - light.cphi2) ^ light.falloff
782 */
783 ureg_DP3(ureg, tmp_y, ureg_negate(ureg_src(rHit)), cLDir); /* rho */
784 ureg_SUB(ureg, tmp_x, _Y(tmp), cLPhi);
785 ureg_MUL(ureg, tmp_x, _X(tmp), cLSDiv);
786 ureg_POW(ureg, tmp_x, _X(tmp), cLFOff); /* spotAtten */
787 ureg_SGE(ureg, tmp_z, _Y(tmp), cLTht); /* if inside theta && phi */
788 ureg_SGE(ureg, tmp_y, _Y(tmp), cLPhi); /* if inside phi */
789 ureg_MAD(ureg, ureg_saturate(tmp_x), _X(tmp), _Y(tmp), _Z(tmp));
790 ureg_MUL(ureg, rAtt, _W(rAtt), _X(tmp));
791 }
792 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
793 ureg_ENDIF(ureg);
794
795 /* directional factors, let's not use LIT because of clarity */
796 ureg_DP3(ureg, ureg_saturate(tmp_x), ureg_src(rNrm), ureg_src(rHit));
797 ureg_MOV(ureg, tmp_y, ureg_imm1f(ureg, 0.0f));
798 ureg_IF(ureg, _X(tmp), &label[l++]);
799 {
800 /* midVec = normalize(hitDir + eyeDir) */
801 if (key->localviewer) {
802 ureg_normalize3(ureg, rMid, ureg_src(rVtx), tmp);
803 ureg_SUB(ureg, rMid, ureg_src(rHit), ureg_src(rMid));
804 } else {
805 ureg_SUB(ureg, rMid, ureg_src(rHit), ureg_imm3f(ureg, 0.0f, 0.0f, 1.0f));
806 }
807 ureg_normalize3(ureg, rMid, ureg_src(rMid), tmp);
808 ureg_DP3(ureg, ureg_saturate(tmp_y), ureg_src(rNrm), ureg_src(rMid));
809 ureg_POW(ureg, tmp_y, _Y(tmp), mtlP);
810
811 ureg_MUL(ureg, tmp_x, _W(rAtt), _X(tmp)); /* dp3(normal,hitDir) * att */
812 ureg_MUL(ureg, tmp_y, _W(rAtt), _Y(tmp)); /* power factor * att */
813 ureg_MAD(ureg, rD, cLColD, _X(tmp), ureg_src(rD)); /* accumulate diffuse */
814 ureg_MAD(ureg, rS, cLColS, _Y(tmp), ureg_src(rS)); /* accumulate specular */
815 }
816 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
817 ureg_ENDIF(ureg);
818
819 ureg_MAD(ureg, rA, cLColA, _W(rAtt), ureg_src(rA)); /* accumulate ambient */
820
821 /* break if this was the last light */
822 ureg_IF(ureg, cLLast, &label[l++]);
823 ureg_BRK(ureg);
824 ureg_ENDIF(ureg);
825 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
826
827 ureg_ADD(ureg, rCtr, _W(rCtr), ureg_imm1f(ureg, 8.0f));
828 ureg_fixup_label(ureg, label[loop_label], ureg_get_instruction_number(ureg));
829 ureg_ENDLOOP(ureg, &label[loop_label]);
830
831 /* Set alpha factors of illumination to 1.0 for the multiplications. */
832 rD.WriteMask = TGSI_WRITEMASK_W; rD.Saturate = 0;
833 rS.WriteMask = TGSI_WRITEMASK_W; rS.Saturate = 0;
834 rA.WriteMask = TGSI_WRITEMASK_W; rA.Saturate = 0;
835 ureg_MOV(ureg, rD, ureg_imm1f(ureg, 1.0f));
836 ureg_MOV(ureg, rS, ureg_imm1f(ureg, 1.0f));
837
838 /* Apply to material:
839 *
840 * oCol[0] = (material.emissive + material.ambient * rs.ambient) +
841 * material.ambient * ambient +
842 * material.diffuse * diffuse +
843 * oCol[1] = material.specular * specular;
844 */
845 if (key->mtl_emissive == 0 && key->mtl_ambient == 0) {
846 ureg_MOV(ureg, rA, ureg_imm1f(ureg, 1.0f));
847 ureg_MAD(ureg, tmp, ureg_src(rA), vs->mtlA, _CONST(19));
848 } else {
849 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(rA), _CONST(25));
850 ureg_MAD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), vs->mtlA, ureg_src(tmp), vs->mtlE);
851 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_W ), vs->mtlA, vs->mtlE);
852 }
853
854 if (key->specular_enable) {
855 /* add oCol[1] to oCol[0] */
856 ureg_MAD(ureg, tmp, ureg_src(rD), vs->mtlD, ureg_src(tmp));
857 ureg_MAD(ureg, oCol[0], ureg_src(rS), vs->mtlS, ureg_src(tmp));
858 } else {
859 ureg_MAD(ureg, oCol[0], ureg_src(rD), vs->mtlD, ureg_src(tmp));
860 }
861 ureg_MUL(ureg, oCol[1], ureg_src(rS), vs->mtlS);
862 } else
863 /* COLOR */
864 if (key->darkness) {
865 if (key->mtl_emissive == 0 && key->mtl_ambient == 0) {
866 ureg_MAD(ureg, oCol[0], vs->mtlD, ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 1.0f), _CONST(19));
867 } else {
868 ureg_MAD(ureg, ureg_writemask(oCol[0], TGSI_WRITEMASK_XYZ), vs->mtlA, _CONST(25), vs->mtlE);
869 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_W), vs->mtlA, vs->mtlE);
870 ureg_ADD(ureg, ureg_writemask(oCol[0], TGSI_WRITEMASK_W), vs->mtlD, _W(tmp));
871 }
872 ureg_MUL(ureg, oCol[1], ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 1.0f), vs->mtlS);
873 } else {
874 ureg_MOV(ureg, oCol[0], vs->aCol[0]);
875 ureg_MOV(ureg, oCol[1], vs->aCol[1]);
876 }
877
878 /* === Process fog.
879 *
880 * exp(x) = ex2(log2(e) * x)
881 */
882 if (key->fog_mode) {
883 if (key->position_t) {
884 ureg_MOV(ureg, ureg_saturate(tmp_x), ureg_scalar(vs->aCol[1], TGSI_SWIZZLE_W));
885 } else
886 if (key->fog_range) {
887 ureg_DP3(ureg, tmp_x, ureg_src(rVtx), ureg_src(rVtx));
888 ureg_RSQ(ureg, tmp_z, _X(tmp));
889 ureg_MUL(ureg, tmp_z, _Z(tmp), _X(tmp));
890 } else {
891 ureg_MOV(ureg, tmp_z, ureg_abs(_Z(rVtx)));
892 }
893
894 if (key->fog_mode == D3DFOG_EXP) {
895 ureg_MUL(ureg, tmp_x, _Z(tmp), _ZZZZ(_CONST(28)));
896 ureg_MUL(ureg, tmp_x, _X(tmp), ureg_imm1f(ureg, -1.442695f));
897 ureg_EX2(ureg, tmp_x, _X(tmp));
898 } else
899 if (key->fog_mode == D3DFOG_EXP2) {
900 ureg_MUL(ureg, tmp_x, _Z(tmp), _ZZZZ(_CONST(28)));
901 ureg_MUL(ureg, tmp_x, _X(tmp), _X(tmp));
902 ureg_MUL(ureg, tmp_x, _X(tmp), ureg_imm1f(ureg, -1.442695f));
903 ureg_EX2(ureg, tmp_x, _X(tmp));
904 } else
905 if (key->fog_mode == D3DFOG_LINEAR && !key->position_t) {
906 ureg_SUB(ureg, tmp_x, _XXXX(_CONST(28)), _Z(tmp));
907 ureg_MUL(ureg, ureg_saturate(tmp_x), _X(tmp), _YYYY(_CONST(28)));
908 }
909 ureg_MOV(ureg, oFog, _X(tmp));
910 } else if (key->fog && !(key->passthrough & (1 << NINE_DECLUSAGE_FOG))) {
911 ureg_MOV(ureg, oFog, ureg_scalar(vs->aCol[1], TGSI_SWIZZLE_W));
912 }
913
914 if (key->passthrough & (1 << NINE_DECLUSAGE_BLENDWEIGHT)) {
915 struct ureg_src input;
916 struct ureg_dst output;
917 input = vs->aWgt;
918 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 18);
919 ureg_MOV(ureg, output, input);
920 }
921 if (key->passthrough & (1 << NINE_DECLUSAGE_BLENDINDICES)) {
922 struct ureg_src input;
923 struct ureg_dst output;
924 input = vs->aInd;
925 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 19);
926 ureg_MOV(ureg, output, input);
927 }
928 if (key->passthrough & (1 << NINE_DECLUSAGE_NORMAL)) {
929 struct ureg_src input;
930 struct ureg_dst output;
931 input = vs->aNrm;
932 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 20);
933 ureg_MOV(ureg, output, input);
934 }
935 if (key->passthrough & (1 << NINE_DECLUSAGE_TANGENT)) {
936 struct ureg_src input;
937 struct ureg_dst output;
938 input = build_vs_add_input(vs, NINE_DECLUSAGE_TANGENT);
939 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 21);
940 ureg_MOV(ureg, output, input);
941 }
942 if (key->passthrough & (1 << NINE_DECLUSAGE_BINORMAL)) {
943 struct ureg_src input;
944 struct ureg_dst output;
945 input = build_vs_add_input(vs, NINE_DECLUSAGE_BINORMAL);
946 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 22);
947 ureg_MOV(ureg, output, input);
948 }
949 if (key->passthrough & (1 << NINE_DECLUSAGE_FOG)) {
950 struct ureg_src input;
951 struct ureg_dst output;
952 input = build_vs_add_input(vs, NINE_DECLUSAGE_FOG);
953 input = ureg_scalar(input, TGSI_SWIZZLE_X);
954 output = oFog;
955 ureg_MOV(ureg, output, input);
956 }
957 if (key->passthrough & (1 << NINE_DECLUSAGE_DEPTH)) {
958 (void) 0; /* TODO: replace z of position output ? */
959 }
960
961
962 if (key->position_t && device->driver_caps.window_space_position_support)
963 ureg_property(ureg, TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION, TRUE);
964
965 ureg_END(ureg);
966 nine_ureg_tgsi_dump(ureg, FALSE);
967 return ureg_create_shader_and_destroy(ureg, device->pipe);
968 }
969
970 /* PS FF constants layout:
971 *
972 * CONST[ 0.. 7] stage[i].D3DTSS_CONSTANT
973 * CONST[ 8..15].x___ stage[i].D3DTSS_BUMPENVMAT00
974 * CONST[ 8..15]._y__ stage[i].D3DTSS_BUMPENVMAT01
975 * CONST[ 8..15].__z_ stage[i].D3DTSS_BUMPENVMAT10
976 * CONST[ 8..15].___w stage[i].D3DTSS_BUMPENVMAT11
977 * CONST[16..19].x_z_ stage[i].D3DTSS_BUMPENVLSCALE
978 * CONST[17..19]._y_w stage[i].D3DTSS_BUMPENVLOFFSET
979 *
980 * CONST[20] D3DRS_TEXTUREFACTOR
981 * CONST[21] D3DRS_FOGCOLOR
982 * CONST[22].x___ RS.FogEnd
983 * CONST[22]._y__ 1.0f / (RS.FogEnd - RS.FogStart)
984 * CONST[22].__z_ RS.FogDensity
985 */
986 struct ps_build_ctx
987 {
988 struct ureg_program *ureg;
989
990 struct ureg_src vC[2]; /* DIFFUSE, SPECULAR */
991 struct ureg_src vT[8]; /* TEXCOORD[i] */
992 struct ureg_dst r[6]; /* TEMPs */
993 struct ureg_dst rCur; /* D3DTA_CURRENT */
994 struct ureg_dst rMod;
995 struct ureg_src rCurSrc;
996 struct ureg_dst rTmp; /* D3DTA_TEMP */
997 struct ureg_src rTmpSrc;
998 struct ureg_dst rTex;
999 struct ureg_src rTexSrc;
1000 struct ureg_src cBEM[8];
1001 struct ureg_src s[8];
1002
1003 struct {
1004 unsigned index;
1005 unsigned index_pre_mod;
1006 unsigned num_regs;
1007 } stage;
1008 };
1009
1010 static struct ureg_src
1011 ps_get_ts_arg(struct ps_build_ctx *ps, unsigned ta)
1012 {
1013 struct ureg_src reg;
1014
1015 switch (ta & D3DTA_SELECTMASK) {
1016 case D3DTA_CONSTANT:
1017 reg = ureg_DECL_constant(ps->ureg, ps->stage.index);
1018 break;
1019 case D3DTA_CURRENT:
1020 reg = (ps->stage.index == ps->stage.index_pre_mod) ? ureg_src(ps->rMod) : ps->rCurSrc;
1021 break;
1022 case D3DTA_DIFFUSE:
1023 reg = ureg_DECL_fs_input(ps->ureg, TGSI_SEMANTIC_COLOR, 0, TGSI_INTERPOLATE_PERSPECTIVE);
1024 break;
1025 case D3DTA_SPECULAR:
1026 reg = ureg_DECL_fs_input(ps->ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_PERSPECTIVE);
1027 break;
1028 case D3DTA_TEMP:
1029 reg = ps->rTmpSrc;
1030 break;
1031 case D3DTA_TEXTURE:
1032 reg = ps->rTexSrc;
1033 break;
1034 case D3DTA_TFACTOR:
1035 reg = ureg_DECL_constant(ps->ureg, 20);
1036 break;
1037 default:
1038 assert(0);
1039 reg = ureg_src_undef();
1040 break;
1041 }
1042 if (ta & D3DTA_COMPLEMENT) {
1043 struct ureg_dst dst = ps->r[ps->stage.num_regs++];
1044 ureg_SUB(ps->ureg, dst, ureg_imm1f(ps->ureg, 1.0f), reg);
1045 reg = ureg_src(dst);
1046 }
1047 if (ta & D3DTA_ALPHAREPLICATE)
1048 reg = _WWWW(reg);
1049 return reg;
1050 }
1051
1052 static struct ureg_dst
1053 ps_get_ts_dst(struct ps_build_ctx *ps, unsigned ta)
1054 {
1055 assert(!(ta & (D3DTA_COMPLEMENT | D3DTA_ALPHAREPLICATE)));
1056
1057 switch (ta & D3DTA_SELECTMASK) {
1058 case D3DTA_CURRENT:
1059 return ps->rCur;
1060 case D3DTA_TEMP:
1061 return ps->rTmp;
1062 default:
1063 assert(0);
1064 return ureg_dst_undef();
1065 }
1066 }
1067
1068 static uint8_t ps_d3dtop_args_mask(D3DTEXTUREOP top)
1069 {
1070 switch (top) {
1071 case D3DTOP_DISABLE:
1072 return 0x0;
1073 case D3DTOP_SELECTARG1:
1074 case D3DTOP_PREMODULATE:
1075 return 0x2;
1076 case D3DTOP_SELECTARG2:
1077 return 0x4;
1078 case D3DTOP_MULTIPLYADD:
1079 case D3DTOP_LERP:
1080 return 0x7;
1081 default:
1082 return 0x6;
1083 }
1084 }
1085
1086 static inline boolean
1087 is_MOV_no_op(struct ureg_dst dst, struct ureg_src src)
1088 {
1089 return !dst.WriteMask ||
1090 (dst.File == src.File &&
1091 dst.Index == src.Index &&
1092 !dst.Indirect &&
1093 !dst.Saturate &&
1094 !src.Indirect &&
1095 !src.Negate &&
1096 !src.Absolute &&
1097 (!(dst.WriteMask & TGSI_WRITEMASK_X) || (src.SwizzleX == TGSI_SWIZZLE_X)) &&
1098 (!(dst.WriteMask & TGSI_WRITEMASK_Y) || (src.SwizzleY == TGSI_SWIZZLE_Y)) &&
1099 (!(dst.WriteMask & TGSI_WRITEMASK_Z) || (src.SwizzleZ == TGSI_SWIZZLE_Z)) &&
1100 (!(dst.WriteMask & TGSI_WRITEMASK_W) || (src.SwizzleW == TGSI_SWIZZLE_W)));
1101
1102 }
1103
1104 static void
1105 ps_do_ts_op(struct ps_build_ctx *ps, unsigned top, struct ureg_dst dst, struct ureg_src *arg)
1106 {
1107 struct ureg_program *ureg = ps->ureg;
1108 struct ureg_dst tmp = ps->r[ps->stage.num_regs];
1109 struct ureg_dst tmp2 = ps->r[ps->stage.num_regs+1];
1110 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
1111
1112 tmp.WriteMask = dst.WriteMask;
1113
1114 if (top != D3DTOP_SELECTARG1 && top != D3DTOP_SELECTARG2 &&
1115 top != D3DTOP_MODULATE && top != D3DTOP_PREMODULATE &&
1116 top != D3DTOP_BLENDDIFFUSEALPHA && top != D3DTOP_BLENDTEXTUREALPHA &&
1117 top != D3DTOP_BLENDFACTORALPHA && top != D3DTOP_BLENDCURRENTALPHA &&
1118 top != D3DTOP_BUMPENVMAP && top != D3DTOP_BUMPENVMAPLUMINANCE &&
1119 top != D3DTOP_LERP)
1120 dst = ureg_saturate(dst);
1121
1122 switch (top) {
1123 case D3DTOP_SELECTARG1:
1124 if (!is_MOV_no_op(dst, arg[1]))
1125 ureg_MOV(ureg, dst, arg[1]);
1126 break;
1127 case D3DTOP_SELECTARG2:
1128 if (!is_MOV_no_op(dst, arg[2]))
1129 ureg_MOV(ureg, dst, arg[2]);
1130 break;
1131 case D3DTOP_MODULATE:
1132 ureg_MUL(ureg, dst, arg[1], arg[2]);
1133 break;
1134 case D3DTOP_MODULATE2X:
1135 ureg_MUL(ureg, tmp, arg[1], arg[2]);
1136 ureg_ADD(ureg, dst, ureg_src(tmp), ureg_src(tmp));
1137 break;
1138 case D3DTOP_MODULATE4X:
1139 ureg_MUL(ureg, tmp, arg[1], arg[2]);
1140 ureg_MUL(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 4.0f));
1141 break;
1142 case D3DTOP_ADD:
1143 ureg_ADD(ureg, dst, arg[1], arg[2]);
1144 break;
1145 case D3DTOP_ADDSIGNED:
1146 ureg_ADD(ureg, tmp, arg[1], arg[2]);
1147 ureg_SUB(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 0.5f));
1148 break;
1149 case D3DTOP_ADDSIGNED2X:
1150 ureg_ADD(ureg, tmp, arg[1], arg[2]);
1151 ureg_MAD(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 2.0f), ureg_imm1f(ureg, -1.0f));
1152 break;
1153 case D3DTOP_SUBTRACT:
1154 ureg_SUB(ureg, dst, arg[1], arg[2]);
1155 break;
1156 case D3DTOP_ADDSMOOTH:
1157 ureg_SUB(ureg, tmp, ureg_imm1f(ureg, 1.0f), arg[1]);
1158 ureg_MAD(ureg, dst, ureg_src(tmp), arg[2], arg[1]);
1159 break;
1160 case D3DTOP_BLENDDIFFUSEALPHA:
1161 ureg_LRP(ureg, dst, _WWWW(ps->vC[0]), arg[1], arg[2]);
1162 break;
1163 case D3DTOP_BLENDTEXTUREALPHA:
1164 /* XXX: alpha taken from previous stage, texture or result ? */
1165 ureg_LRP(ureg, dst, _W(ps->rTex), arg[1], arg[2]);
1166 break;
1167 case D3DTOP_BLENDFACTORALPHA:
1168 ureg_LRP(ureg, dst, _WWWW(_CONST(20)), arg[1], arg[2]);
1169 break;
1170 case D3DTOP_BLENDTEXTUREALPHAPM:
1171 ureg_SUB(ureg, tmp_x, ureg_imm1f(ureg, 1.0f), _W(ps->rTex));
1172 ureg_MAD(ureg, dst, arg[2], _X(tmp), arg[1]);
1173 break;
1174 case D3DTOP_BLENDCURRENTALPHA:
1175 ureg_LRP(ureg, dst, _WWWW(ps->rCurSrc), arg[1], arg[2]);
1176 break;
1177 case D3DTOP_PREMODULATE:
1178 ureg_MOV(ureg, dst, arg[1]);
1179 ps->stage.index_pre_mod = ps->stage.index + 1;
1180 break;
1181 case D3DTOP_MODULATEALPHA_ADDCOLOR:
1182 ureg_MAD(ureg, dst, _WWWW(arg[1]), arg[2], arg[1]);
1183 break;
1184 case D3DTOP_MODULATECOLOR_ADDALPHA:
1185 ureg_MAD(ureg, dst, arg[1], arg[2], _WWWW(arg[1]));
1186 break;
1187 case D3DTOP_MODULATEINVALPHA_ADDCOLOR:
1188 ureg_SUB(ureg, tmp_x, ureg_imm1f(ureg, 1.0f), _WWWW(arg[1]));
1189 ureg_MAD(ureg, dst, _X(tmp), arg[2], arg[1]);
1190 break;
1191 case D3DTOP_MODULATEINVCOLOR_ADDALPHA:
1192 ureg_SUB(ureg, tmp, ureg_imm1f(ureg, 1.0f), arg[1]);
1193 ureg_MAD(ureg, dst, ureg_src(tmp), arg[2], _WWWW(arg[1]));
1194 break;
1195 case D3DTOP_BUMPENVMAP:
1196 break;
1197 case D3DTOP_BUMPENVMAPLUMINANCE:
1198 break;
1199 case D3DTOP_DOTPRODUCT3:
1200 ureg_SUB(ureg, tmp, arg[1], ureg_imm4f(ureg,0.5,0.5,0.5,0.5));
1201 ureg_SUB(ureg, tmp2, arg[2] , ureg_imm4f(ureg,0.5,0.5,0.5,0.5));
1202 ureg_DP3(ureg, tmp, ureg_src(tmp), ureg_src(tmp2));
1203 ureg_MUL(ureg, ureg_saturate(dst), ureg_src(tmp), ureg_imm4f(ureg,4.0,4.0,4.0,4.0));
1204 break;
1205 case D3DTOP_MULTIPLYADD:
1206 ureg_MAD(ureg, dst, arg[1], arg[2], arg[0]);
1207 break;
1208 case D3DTOP_LERP:
1209 ureg_LRP(ureg, dst, arg[0], arg[1], arg[2]);
1210 break;
1211 case D3DTOP_DISABLE:
1212 /* no-op ? */
1213 break;
1214 default:
1215 assert(!"invalid D3DTOP");
1216 break;
1217 }
1218 }
1219
1220 static void *
1221 nine_ff_build_ps(struct NineDevice9 *device, struct nine_ff_ps_key *key)
1222 {
1223 struct ps_build_ctx ps;
1224 struct ureg_program *ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT);
1225 struct ureg_dst oCol;
1226 unsigned i, s;
1227 const unsigned texcoord_sn = get_texcoord_sn(device->screen);
1228
1229 memset(&ps, 0, sizeof(ps));
1230 ps.ureg = ureg;
1231 ps.stage.index_pre_mod = -1;
1232
1233 ps.vC[0] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 0, TGSI_INTERPOLATE_PERSPECTIVE);
1234
1235 /* Declare all TEMPs we might need, serious drivers have a register allocator. */
1236 for (i = 0; i < Elements(ps.r); ++i)
1237 ps.r[i] = ureg_DECL_local_temporary(ureg);
1238 ps.rCur = ps.r[0];
1239 ps.rTmp = ps.r[1];
1240 ps.rTex = ps.r[2];
1241 ps.rCurSrc = ureg_src(ps.rCur);
1242 ps.rTmpSrc = ureg_src(ps.rTmp);
1243 ps.rTexSrc = ureg_src(ps.rTex);
1244
1245 for (s = 0; s < 8; ++s) {
1246 ps.s[s] = ureg_src_undef();
1247
1248 if (key->ts[s].colorop != D3DTOP_DISABLE) {
1249 if (key->ts[s].colorarg0 == D3DTA_SPECULAR ||
1250 key->ts[s].colorarg1 == D3DTA_SPECULAR ||
1251 key->ts[s].colorarg2 == D3DTA_SPECULAR)
1252 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_PERSPECTIVE);
1253
1254 if (key->ts[s].colorarg0 == D3DTA_TEXTURE ||
1255 key->ts[s].colorarg1 == D3DTA_TEXTURE ||
1256 key->ts[s].colorarg2 == D3DTA_TEXTURE) {
1257 ps.s[s] = ureg_DECL_sampler(ureg, s);
1258 ps.vT[s] = ureg_DECL_fs_input(ureg, texcoord_sn, s, TGSI_INTERPOLATE_PERSPECTIVE);
1259 }
1260 if (s && (key->ts[s - 1].colorop == D3DTOP_PREMODULATE ||
1261 key->ts[s - 1].alphaop == D3DTOP_PREMODULATE))
1262 ps.s[s] = ureg_DECL_sampler(ureg, s);
1263 }
1264
1265 if (key->ts[s].alphaop != D3DTOP_DISABLE) {
1266 if (key->ts[s].alphaarg0 == D3DTA_SPECULAR ||
1267 key->ts[s].alphaarg1 == D3DTA_SPECULAR ||
1268 key->ts[s].alphaarg2 == D3DTA_SPECULAR)
1269 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_PERSPECTIVE);
1270
1271 if (key->ts[s].alphaarg0 == D3DTA_TEXTURE ||
1272 key->ts[s].alphaarg1 == D3DTA_TEXTURE ||
1273 key->ts[s].alphaarg2 == D3DTA_TEXTURE) {
1274 ps.s[s] = ureg_DECL_sampler(ureg, s);
1275 ps.vT[s] = ureg_DECL_fs_input(ureg, texcoord_sn, s, TGSI_INTERPOLATE_PERSPECTIVE);
1276 }
1277 }
1278 }
1279 if (key->specular)
1280 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_PERSPECTIVE);
1281
1282 oCol = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
1283
1284 if (key->ts[0].colorop == D3DTOP_DISABLE &&
1285 key->ts[0].alphaop == D3DTOP_DISABLE)
1286 ureg_MOV(ureg, ps.rCur, ps.vC[0]);
1287 /* Or is it undefined then ? */
1288
1289 /* Run stages.
1290 */
1291 for (s = 0; s < 8; ++s) {
1292 unsigned colorarg[3];
1293 unsigned alphaarg[3];
1294 const uint8_t used_c = ps_d3dtop_args_mask(key->ts[s].colorop);
1295 const uint8_t used_a = ps_d3dtop_args_mask(key->ts[s].alphaop);
1296 struct ureg_dst dst;
1297 struct ureg_src arg[3];
1298
1299 if (key->ts[s].colorop == D3DTOP_DISABLE &&
1300 key->ts[s].alphaop == D3DTOP_DISABLE)
1301 continue;
1302 ps.stage.index = s;
1303 ps.stage.num_regs = 3;
1304
1305 DBG("STAGE[%u]: colorop=%s alphaop=%s\n", s,
1306 nine_D3DTOP_to_str(key->ts[s].colorop),
1307 nine_D3DTOP_to_str(key->ts[s].alphaop));
1308
1309 if (!ureg_src_is_undef(ps.s[s])) {
1310 unsigned target;
1311 switch (key->ts[s].textarget) {
1312 case 0: target = TGSI_TEXTURE_1D; break;
1313 case 1: target = TGSI_TEXTURE_2D; break;
1314 case 2: target = TGSI_TEXTURE_3D; break;
1315 case 3: target = TGSI_TEXTURE_CUBE; break;
1316 /* this is a 2 bit bitfield, do I really need a default case ? */
1317 }
1318
1319 /* sample the texture */
1320 if (key->ts[s].colorop == D3DTOP_BUMPENVMAP ||
1321 key->ts[s].colorop == D3DTOP_BUMPENVMAPLUMINANCE) {
1322 }
1323 if (key->projected & (3 << (s *2))) {
1324 unsigned dim = 1 + ((key->projected >> (2 * s)) & 3);
1325 if (dim == 4)
1326 ureg_TXP(ureg, ps.rTex, target, ps.vT[s], ps.s[s]);
1327 else {
1328 ureg_RCP(ureg, ureg_writemask(ps.rTmp, TGSI_WRITEMASK_X), ureg_scalar(ps.vT[s], dim-1));
1329 ureg_MUL(ureg, ps.rTmp, _XXXX(ps.rTmpSrc), ps.vT[s]);
1330 ureg_TEX(ureg, ps.rTex, target, ps.rTmpSrc, ps.s[s]);
1331 }
1332 } else {
1333 ureg_TEX(ureg, ps.rTex, target, ps.vT[s], ps.s[s]);
1334 }
1335 }
1336
1337 if (s == 0 &&
1338 (key->ts[0].resultarg != 0 /* not current */ ||
1339 key->ts[0].colorop == D3DTOP_DISABLE ||
1340 key->ts[0].alphaop == D3DTOP_DISABLE ||
1341 key->ts[0].colorop == D3DTOP_BLENDCURRENTALPHA ||
1342 key->ts[0].alphaop == D3DTOP_BLENDCURRENTALPHA ||
1343 key->ts[0].colorarg0 == D3DTA_CURRENT ||
1344 key->ts[0].colorarg1 == D3DTA_CURRENT ||
1345 key->ts[0].colorarg2 == D3DTA_CURRENT ||
1346 key->ts[0].alphaarg0 == D3DTA_CURRENT ||
1347 key->ts[0].alphaarg1 == D3DTA_CURRENT ||
1348 key->ts[0].alphaarg2 == D3DTA_CURRENT)
1349 ) {
1350 /* Initialize D3DTA_CURRENT.
1351 * (Yes we can do this before the loop but not until
1352 * NVE4 has an instruction scheduling pass.)
1353 */
1354 ureg_MOV(ureg, ps.rCur, ps.vC[0]);
1355 }
1356
1357 dst = ps_get_ts_dst(&ps, key->ts[s].resultarg ? D3DTA_TEMP : D3DTA_CURRENT);
1358
1359 if (ps.stage.index_pre_mod == ps.stage.index) {
1360 ps.rMod = ps.r[ps.stage.num_regs++];
1361 ureg_MUL(ureg, ps.rMod, ps.rCurSrc, ps.rTexSrc);
1362 }
1363
1364 colorarg[0] = (key->ts[s].colorarg0 | ((key->colorarg_b4[0] >> s) << 4) | ((key->colorarg_b5[0] >> s) << 5)) & 0x3f;
1365 colorarg[1] = (key->ts[s].colorarg1 | ((key->colorarg_b4[1] >> s) << 4) | ((key->colorarg_b5[1] >> s) << 5)) & 0x3f;
1366 colorarg[2] = (key->ts[s].colorarg2 | ((key->colorarg_b4[2] >> s) << 4) | ((key->colorarg_b5[2] >> s) << 5)) & 0x3f;
1367 alphaarg[0] = (key->ts[s].alphaarg0 | ((key->alphaarg_b4[0] >> s) << 4)) & 0x1f;
1368 alphaarg[1] = (key->ts[s].alphaarg1 | ((key->alphaarg_b4[1] >> s) << 4)) & 0x1f;
1369 alphaarg[2] = (key->ts[s].alphaarg2 | ((key->alphaarg_b4[2] >> s) << 4)) & 0x1f;
1370
1371 if (key->ts[s].colorop != key->ts[s].alphaop ||
1372 colorarg[0] != alphaarg[0] ||
1373 colorarg[1] != alphaarg[1] ||
1374 colorarg[2] != alphaarg[2])
1375 dst.WriteMask = TGSI_WRITEMASK_XYZ;
1376
1377 /* Special DOTPRODUCT behaviour (see wine tests) */
1378 if (key->ts[s].colorop == D3DTOP_DOTPRODUCT3)
1379 dst.WriteMask = TGSI_WRITEMASK_XYZW;
1380
1381 if (used_c & 0x1) arg[0] = ps_get_ts_arg(&ps, colorarg[0]);
1382 if (used_c & 0x2) arg[1] = ps_get_ts_arg(&ps, colorarg[1]);
1383 if (used_c & 0x4) arg[2] = ps_get_ts_arg(&ps, colorarg[2]);
1384 ps_do_ts_op(&ps, key->ts[s].colorop, dst, arg);
1385
1386 if (dst.WriteMask != TGSI_WRITEMASK_XYZW) {
1387 dst.WriteMask = TGSI_WRITEMASK_W;
1388
1389 if (used_a & 0x1) arg[0] = ps_get_ts_arg(&ps, alphaarg[0]);
1390 if (used_a & 0x2) arg[1] = ps_get_ts_arg(&ps, alphaarg[1]);
1391 if (used_a & 0x4) arg[2] = ps_get_ts_arg(&ps, alphaarg[2]);
1392 ps_do_ts_op(&ps, key->ts[s].alphaop, dst, arg);
1393 }
1394 }
1395
1396 if (key->specular)
1397 ureg_ADD(ureg, ps.rCur, ps.rCurSrc, ps.vC[1]);
1398
1399 /* Fog.
1400 */
1401 if (key->fog_mode) {
1402 struct ureg_src vPos;
1403 if (device->screen->get_param(device->screen,
1404 PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL)) {
1405 vPos = ureg_DECL_system_value(ureg, TGSI_SEMANTIC_POSITION, 0);
1406 } else {
1407 vPos = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_POSITION, 0,
1408 TGSI_INTERPOLATE_LINEAR);
1409 }
1410
1411 struct ureg_dst rFog = ureg_writemask(ps.rTmp, TGSI_WRITEMASK_X);
1412 if (key->fog_mode == D3DFOG_EXP) {
1413 ureg_MUL(ureg, rFog, _ZZZZ(vPos), _ZZZZ(_CONST(22)));
1414 ureg_MUL(ureg, rFog, _X(rFog), ureg_imm1f(ureg, -1.442695f));
1415 ureg_EX2(ureg, rFog, _X(rFog));
1416 } else
1417 if (key->fog_mode == D3DFOG_EXP2) {
1418 ureg_MUL(ureg, rFog, _ZZZZ(vPos), _ZZZZ(_CONST(22)));
1419 ureg_MUL(ureg, rFog, _X(rFog), _X(rFog));
1420 ureg_MUL(ureg, rFog, _X(rFog), ureg_imm1f(ureg, -1.442695f));
1421 ureg_EX2(ureg, rFog, _X(rFog));
1422 } else
1423 if (key->fog_mode == D3DFOG_LINEAR) {
1424 ureg_SUB(ureg, rFog, _XXXX(_CONST(22)), _ZZZZ(vPos));
1425 ureg_MUL(ureg, ureg_saturate(rFog), _X(rFog), _YYYY(_CONST(22)));
1426 }
1427 ureg_LRP(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_XYZ), _X(rFog), ps.rCurSrc, _CONST(21));
1428 ureg_MOV(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_W), ps.rCurSrc);
1429 } else
1430 if (key->fog) {
1431 struct ureg_src vFog = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_FOG, 0, TGSI_INTERPOLATE_PERSPECTIVE);
1432 ureg_LRP(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_XYZ), _XXXX(vFog), ps.rCurSrc, _CONST(21));
1433 ureg_MOV(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_W), ps.rCurSrc);
1434 } else {
1435 ureg_MOV(ureg, oCol, ps.rCurSrc);
1436 }
1437
1438 ureg_END(ureg);
1439 nine_ureg_tgsi_dump(ureg, FALSE);
1440 return ureg_create_shader_and_destroy(ureg, device->pipe);
1441 }
1442
1443 static struct NineVertexShader9 *
1444 nine_ff_get_vs(struct NineDevice9 *device)
1445 {
1446 const struct nine_state *state = &device->state;
1447 struct NineVertexShader9 *vs;
1448 enum pipe_error err;
1449 struct vs_build_ctx bld;
1450 struct nine_ff_vs_key key;
1451 unsigned s, i;
1452 char input_texture_coord[8];
1453
1454 assert(sizeof(key) <= sizeof(key.value32));
1455
1456 memset(&key, 0, sizeof(key));
1457 memset(&bld, 0, sizeof(bld));
1458 memset(&input_texture_coord, 0, sizeof(input_texture_coord));
1459
1460 bld.key = &key;
1461
1462 /* FIXME: this shouldn't be NULL, but it is on init */
1463 if (state->vdecl) {
1464 key.color0in_one = 1;
1465 key.color1in_one = 1;
1466 for (i = 0; i < state->vdecl->nelems; i++) {
1467 uint16_t usage = state->vdecl->usage_map[i];
1468 if (usage == NINE_DECLUSAGE_POSITIONT)
1469 key.position_t = 1;
1470 else if (usage == NINE_DECLUSAGE_i(COLOR, 0))
1471 key.color0in_one = 0;
1472 else if (usage == NINE_DECLUSAGE_i(COLOR, 1))
1473 key.color1in_one = 0;
1474 else if (usage == NINE_DECLUSAGE_PSIZE)
1475 key.vertexpointsize = 1;
1476 else if (usage % NINE_DECLUSAGE_COUNT == NINE_DECLUSAGE_TEXCOORD) {
1477 s = usage / NINE_DECLUSAGE_COUNT;
1478 if (s < 8)
1479 input_texture_coord[s] = nine_decltype_get_dim(state->vdecl->decls[i].Type);
1480 else
1481 DBG("FF given texture coordinate >= 8. Ignoring\n");
1482 } else if (usage < NINE_DECLUSAGE_NONE)
1483 key.passthrough |= 1 << usage;
1484 }
1485 }
1486 /* ff vs + ps 3.0: some elements are passed to the ps (wine test).
1487 * We do restrict to indices 0 */
1488 key.passthrough &= ~((1 << NINE_DECLUSAGE_POSITION) | (1 << NINE_DECLUSAGE_PSIZE) |
1489 (1 << NINE_DECLUSAGE_TEXCOORD) | (1 << NINE_DECLUSAGE_POSITIONT) |
1490 (1 << NINE_DECLUSAGE_TESSFACTOR) | (1 << NINE_DECLUSAGE_SAMPLE));
1491 if (!key.vertexpointsize)
1492 key.pointscale = !!state->rs[D3DRS_POINTSCALEENABLE];
1493
1494 key.lighting = !!state->rs[D3DRS_LIGHTING] && state->ff.num_lights_active;
1495 key.darkness = !!state->rs[D3DRS_LIGHTING] && !state->ff.num_lights_active;
1496 if (key.position_t) {
1497 key.darkness = 0; /* |= key.lighting; */ /* XXX ? */
1498 key.lighting = 0;
1499 }
1500 if ((key.lighting | key.darkness) && state->rs[D3DRS_COLORVERTEX]) {
1501 key.mtl_diffuse = state->rs[D3DRS_DIFFUSEMATERIALSOURCE];
1502 key.mtl_ambient = state->rs[D3DRS_AMBIENTMATERIALSOURCE];
1503 key.mtl_specular = state->rs[D3DRS_SPECULARMATERIALSOURCE];
1504 key.mtl_emissive = state->rs[D3DRS_EMISSIVEMATERIALSOURCE];
1505 }
1506 key.fog = !!state->rs[D3DRS_FOGENABLE];
1507 key.fog_mode = state->rs[D3DRS_FOGENABLE] ? state->rs[D3DRS_FOGVERTEXMODE] : 0;
1508 if (key.fog_mode)
1509 key.fog_range = !key.position_t && state->rs[D3DRS_RANGEFOGENABLE];
1510
1511 key.localviewer = !!state->rs[D3DRS_LOCALVIEWER];
1512 key.specular_enable = !!state->rs[D3DRS_SPECULARENABLE];
1513
1514 if (state->rs[D3DRS_VERTEXBLEND] != D3DVBF_DISABLE) {
1515 key.vertexblend_indexed = !!state->rs[D3DRS_INDEXEDVERTEXBLENDENABLE];
1516
1517 switch (state->rs[D3DRS_VERTEXBLEND]) {
1518 case D3DVBF_0WEIGHTS: key.vertexblend = key.vertexblend_indexed; break;
1519 case D3DVBF_1WEIGHTS: key.vertexblend = 2; break;
1520 case D3DVBF_2WEIGHTS: key.vertexblend = 3; break;
1521 case D3DVBF_3WEIGHTS: key.vertexblend = 4; break;
1522 case D3DVBF_TWEENING: key.vertextween = 1; break;
1523 default:
1524 assert(!"invalid D3DVBF");
1525 break;
1526 }
1527 }
1528
1529 for (s = 0; s < 8; ++s) {
1530 unsigned gen = (state->ff.tex_stage[s][D3DTSS_TEXCOORDINDEX] >> 16) + 1;
1531 unsigned dim;
1532
1533 if (key.position_t && gen > NINED3DTSS_TCI_PASSTHRU)
1534 gen = NINED3DTSS_TCI_PASSTHRU;
1535
1536 if (!input_texture_coord[s] && gen == NINED3DTSS_TCI_PASSTHRU)
1537 gen = NINED3DTSS_TCI_DISABLE;
1538
1539 key.tc_gen |= gen << (s * 3);
1540 key.tc_idx |= (state->ff.tex_stage[s][D3DTSS_TEXCOORDINDEX] & 7) << (s * 3);
1541 key.tc_dim_input |= ((input_texture_coord[s]-1) & 0x3) << (s * 2);
1542
1543 dim = state->ff.tex_stage[s][D3DTSS_TEXTURETRANSFORMFLAGS] & 0x7;
1544 if (dim > 4)
1545 dim = input_texture_coord[s];
1546 if (dim == 1) /* NV behaviour */
1547 dim = 0;
1548 key.tc_dim_output |= dim << (s * 3);
1549 }
1550
1551 vs = util_hash_table_get(device->ff.ht_vs, &key);
1552 if (vs)
1553 return vs;
1554 NineVertexShader9_new(device, &vs, NULL, nine_ff_build_vs(device, &bld));
1555
1556 nine_ff_prune_vs(device);
1557 if (vs) {
1558 unsigned n;
1559
1560 memcpy(&vs->ff_key, &key, sizeof(vs->ff_key));
1561
1562 err = util_hash_table_set(device->ff.ht_vs, &vs->ff_key, vs);
1563 (void)err;
1564 assert(err == PIPE_OK);
1565 device->ff.num_vs++;
1566 NineUnknown_ConvertRefToBind(NineUnknown(vs));
1567
1568 vs->num_inputs = bld.num_inputs;
1569 for (n = 0; n < bld.num_inputs; ++n)
1570 vs->input_map[n].ndecl = bld.input[n];
1571
1572 vs->position_t = key.position_t;
1573 vs->point_size = key.vertexpointsize | key.pointscale;
1574 }
1575 return vs;
1576 }
1577
1578 static struct NinePixelShader9 *
1579 nine_ff_get_ps(struct NineDevice9 *device)
1580 {
1581 struct nine_state *state = &device->state;
1582 struct NinePixelShader9 *ps;
1583 enum pipe_error err;
1584 struct nine_ff_ps_key key;
1585 unsigned s;
1586 uint8_t sampler_mask = 0;
1587
1588 assert(sizeof(key) <= sizeof(key.value32));
1589
1590 memset(&key, 0, sizeof(key));
1591 for (s = 0; s < 8; ++s) {
1592 key.ts[s].colorop = state->ff.tex_stage[s][D3DTSS_COLOROP];
1593 key.ts[s].alphaop = state->ff.tex_stage[s][D3DTSS_ALPHAOP];
1594 /* MSDN says D3DTOP_DISABLE disables this and all subsequent stages. */
1595 /* ALPHAOP cannot be disabled if COLOROP is enabled. */
1596 if (key.ts[s].colorop == D3DTOP_DISABLE) {
1597 key.ts[s].alphaop = D3DTOP_DISABLE; /* DISABLE == 1, avoid degenerate keys */
1598 break;
1599 }
1600
1601 if (!state->texture[s] &&
1602 state->ff.tex_stage[s][D3DTSS_COLORARG1] == D3DTA_TEXTURE) {
1603 /* This should also disable the stage. */
1604 key.ts[s].colorop = key.ts[s].alphaop = D3DTOP_DISABLE;
1605 break;
1606 }
1607
1608 if (state->ff.tex_stage[s][D3DTSS_COLORARG1] == D3DTA_TEXTURE)
1609 sampler_mask |= (1 << s);
1610
1611 if (key.ts[s].colorop != D3DTOP_DISABLE) {
1612 uint8_t used_c = ps_d3dtop_args_mask(key.ts[s].colorop);
1613 if (used_c & 0x1) key.ts[s].colorarg0 = state->ff.tex_stage[s][D3DTSS_COLORARG0];
1614 if (used_c & 0x2) key.ts[s].colorarg1 = state->ff.tex_stage[s][D3DTSS_COLORARG1];
1615 if (used_c & 0x4) key.ts[s].colorarg2 = state->ff.tex_stage[s][D3DTSS_COLORARG2];
1616 if (used_c & 0x1) key.colorarg_b4[0] |= (state->ff.tex_stage[s][D3DTSS_COLORARG0] >> 4) << s;
1617 if (used_c & 0x1) key.colorarg_b5[0] |= (state->ff.tex_stage[s][D3DTSS_COLORARG0] >> 5) << s;
1618 if (used_c & 0x2) key.colorarg_b4[1] |= (state->ff.tex_stage[s][D3DTSS_COLORARG1] >> 4) << s;
1619 if (used_c & 0x2) key.colorarg_b5[1] |= (state->ff.tex_stage[s][D3DTSS_COLORARG1] >> 5) << s;
1620 if (used_c & 0x4) key.colorarg_b4[2] |= (state->ff.tex_stage[s][D3DTSS_COLORARG2] >> 4) << s;
1621 if (used_c & 0x4) key.colorarg_b5[2] |= (state->ff.tex_stage[s][D3DTSS_COLORARG2] >> 5) << s;
1622 }
1623 if (key.ts[s].alphaop != D3DTOP_DISABLE) {
1624 uint8_t used_a = ps_d3dtop_args_mask(key.ts[s].alphaop);
1625 if (used_a & 0x1) key.ts[s].alphaarg0 = state->ff.tex_stage[s][D3DTSS_ALPHAARG0];
1626 if (used_a & 0x2) key.ts[s].alphaarg1 = state->ff.tex_stage[s][D3DTSS_ALPHAARG1];
1627 if (used_a & 0x4) key.ts[s].alphaarg2 = state->ff.tex_stage[s][D3DTSS_ALPHAARG2];
1628 if (used_a & 0x1) key.alphaarg_b4[0] |= (state->ff.tex_stage[s][D3DTSS_ALPHAARG0] >> 4) << s;
1629 if (used_a & 0x2) key.alphaarg_b4[1] |= (state->ff.tex_stage[s][D3DTSS_ALPHAARG1] >> 4) << s;
1630 if (used_a & 0x4) key.alphaarg_b4[2] |= (state->ff.tex_stage[s][D3DTSS_ALPHAARG2] >> 4) << s;
1631 }
1632 key.ts[s].resultarg = state->ff.tex_stage[s][D3DTSS_RESULTARG] == D3DTA_TEMP;
1633
1634 if (state->texture[s]) {
1635 switch (state->texture[s]->base.type) {
1636 case D3DRTYPE_TEXTURE: key.ts[s].textarget = 1; break;
1637 case D3DRTYPE_VOLUMETEXTURE: key.ts[s].textarget = 2; break;
1638 case D3DRTYPE_CUBETEXTURE: key.ts[s].textarget = 3; break;
1639 default:
1640 assert(!"unexpected texture type");
1641 break;
1642 }
1643 } else {
1644 key.ts[s].textarget = 1;
1645 }
1646 }
1647
1648 key.projected = nine_ff_get_projected_key(state);
1649
1650 for (; s < 8; ++s)
1651 key.ts[s].colorop = key.ts[s].alphaop = D3DTOP_DISABLE;
1652 if (state->rs[D3DRS_FOGENABLE])
1653 key.fog_mode = state->rs[D3DRS_FOGTABLEMODE];
1654 key.fog = !!state->rs[D3DRS_FOGENABLE];
1655
1656 ps = util_hash_table_get(device->ff.ht_ps, &key);
1657 if (ps)
1658 return ps;
1659 NinePixelShader9_new(device, &ps, NULL, nine_ff_build_ps(device, &key));
1660
1661 nine_ff_prune_ps(device);
1662 if (ps) {
1663 memcpy(&ps->ff_key, &key, sizeof(ps->ff_key));
1664
1665 err = util_hash_table_set(device->ff.ht_ps, &ps->ff_key, ps);
1666 (void)err;
1667 assert(err == PIPE_OK);
1668 device->ff.num_ps++;
1669 NineUnknown_ConvertRefToBind(NineUnknown(ps));
1670
1671 ps->rt_mask = 0x1;
1672 ps->sampler_mask = sampler_mask;
1673 }
1674 return ps;
1675 }
1676
1677 #define GET_D3DTS(n) nine_state_access_transform(state, D3DTS_##n, FALSE)
1678 #define IS_D3DTS_DIRTY(s,n) ((s)->ff.changed.transform[(D3DTS_##n) / 32] & (1 << ((D3DTS_##n) % 32)))
1679 static void
1680 nine_ff_load_vs_transforms(struct NineDevice9 *device)
1681 {
1682 struct nine_state *state = &device->state;
1683 D3DMATRIX T;
1684 D3DMATRIX *M = (D3DMATRIX *)device->ff.vs_const;
1685 unsigned i;
1686
1687 /* TODO: make this nicer, and only upload the ones we need */
1688 /* TODO: use ff.vs_const as storage of W, V, P matrices */
1689
1690 if (IS_D3DTS_DIRTY(state, WORLD) ||
1691 IS_D3DTS_DIRTY(state, VIEW) ||
1692 IS_D3DTS_DIRTY(state, PROJECTION)) {
1693 /* WVP, WV matrices */
1694 nine_d3d_matrix_matrix_mul(&M[1], GET_D3DTS(WORLD), GET_D3DTS(VIEW));
1695 nine_d3d_matrix_matrix_mul(&M[0], &M[1], GET_D3DTS(PROJECTION));
1696
1697 /* normal matrix == transpose(inverse(WV)) */
1698 nine_d3d_matrix_inverse_3x3(&T, &M[1]);
1699 nine_d3d_matrix_transpose(&M[4], &T);
1700
1701 /* VP matrix */
1702 nine_d3d_matrix_matrix_mul(&M[2], GET_D3DTS(VIEW), GET_D3DTS(PROJECTION));
1703
1704 /* V and W matrix */
1705 M[3] = *GET_D3DTS(VIEW);
1706 M[56] = *GET_D3DTS(WORLD);
1707 }
1708
1709 if (state->rs[D3DRS_VERTEXBLEND] != D3DVBF_DISABLE) {
1710 /* load other world matrices */
1711 for (i = 1; i <= 7; ++i)
1712 M[56 + i] = *GET_D3DTS(WORLDMATRIX(i));
1713 }
1714
1715 device->ff.vs_const[30 * 4] = asfloat(state->rs[D3DRS_TWEENFACTOR]);
1716 }
1717
1718 static void
1719 nine_ff_load_lights(struct NineDevice9 *device)
1720 {
1721 struct nine_state *state = &device->state;
1722 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1723 unsigned l;
1724
1725 if (state->changed.group & NINE_STATE_FF_MATERIAL) {
1726 const D3DMATERIAL9 *mtl = &state->ff.material;
1727
1728 memcpy(&dst[20], &mtl->Diffuse, 4 * sizeof(float));
1729 memcpy(&dst[21], &mtl->Ambient, 4 * sizeof(float));
1730 memcpy(&dst[22], &mtl->Specular, 4 * sizeof(float));
1731 dst[23].x = mtl->Power;
1732 memcpy(&dst[24], &mtl->Emissive, 4 * sizeof(float));
1733 d3dcolor_to_rgba(&dst[25].x, state->rs[D3DRS_AMBIENT]);
1734 dst[19].x = dst[25].x * mtl->Ambient.r + mtl->Emissive.r;
1735 dst[19].y = dst[25].y * mtl->Ambient.g + mtl->Emissive.g;
1736 dst[19].z = dst[25].z * mtl->Ambient.b + mtl->Emissive.b;
1737 dst[19].w = mtl->Ambient.a + mtl->Emissive.a;
1738 }
1739
1740 if (!(state->changed.group & NINE_STATE_FF_LIGHTING))
1741 return;
1742
1743 for (l = 0; l < state->ff.num_lights_active; ++l) {
1744 const D3DLIGHT9 *light = &state->ff.light[state->ff.active_light[l]];
1745
1746 dst[32 + l * 8].x = light->Type;
1747 dst[32 + l * 8].y = light->Attenuation0;
1748 dst[32 + l * 8].z = light->Attenuation1;
1749 dst[32 + l * 8].w = light->Attenuation2;
1750 memcpy(&dst[33 + l * 8].x, &light->Diffuse, sizeof(light->Diffuse));
1751 memcpy(&dst[34 + l * 8].x, &light->Specular, sizeof(light->Specular));
1752 memcpy(&dst[35 + l * 8].x, &light->Ambient, sizeof(light->Ambient));
1753 nine_d3d_vector4_matrix_mul((D3DVECTOR *)&dst[36 + l * 8].x, &light->Position, GET_D3DTS(VIEW));
1754 nine_d3d_vector3_matrix_mul((D3DVECTOR *)&dst[37 + l * 8].x, &light->Direction, GET_D3DTS(VIEW));
1755 dst[36 + l * 8].w = light->Type == D3DLIGHT_DIRECTIONAL ? 1e9f : light->Range;
1756 dst[37 + l * 8].w = light->Falloff;
1757 dst[38 + l * 8].x = cosf(light->Theta * 0.5f);
1758 dst[38 + l * 8].y = cosf(light->Phi * 0.5f);
1759 dst[38 + l * 8].z = 1.0f / (dst[38 + l * 8].x - dst[38 + l * 8].y);
1760 dst[39 + l * 8].w = (l + 1) == state->ff.num_lights_active;
1761 }
1762 }
1763
1764 static void
1765 nine_ff_load_point_and_fog_params(struct NineDevice9 *device)
1766 {
1767 const struct nine_state *state = &device->state;
1768 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1769
1770 if (!(state->changed.group & NINE_STATE_FF_OTHER))
1771 return;
1772 dst[26].x = asfloat(state->rs[D3DRS_POINTSIZE_MIN]);
1773 dst[26].y = asfloat(state->rs[D3DRS_POINTSIZE_MAX]);
1774 dst[26].z = asfloat(state->rs[D3DRS_POINTSIZE]);
1775 dst[26].w = asfloat(state->rs[D3DRS_POINTSCALE_A]);
1776 dst[27].x = asfloat(state->rs[D3DRS_POINTSCALE_B]);
1777 dst[27].y = asfloat(state->rs[D3DRS_POINTSCALE_C]);
1778 dst[28].x = asfloat(state->rs[D3DRS_FOGEND]);
1779 dst[28].y = 1.0f / (asfloat(state->rs[D3DRS_FOGEND]) - asfloat(state->rs[D3DRS_FOGSTART]));
1780 if (isinf(dst[28].y))
1781 dst[28].y = 0.0f;
1782 dst[28].z = asfloat(state->rs[D3DRS_FOGDENSITY]);
1783 }
1784
1785 static void
1786 nine_ff_load_tex_matrices(struct NineDevice9 *device)
1787 {
1788 struct nine_state *state = &device->state;
1789 D3DMATRIX *M = (D3DMATRIX *)device->ff.vs_const;
1790 unsigned s;
1791
1792 if (!(state->ff.changed.transform[0] & 0xff0000))
1793 return;
1794 for (s = 0; s < 8; ++s) {
1795 if (IS_D3DTS_DIRTY(state, TEXTURE0 + s))
1796 nine_d3d_matrix_transpose(&M[32 + s], nine_state_access_transform(state, D3DTS_TEXTURE0 + s, FALSE));
1797 }
1798 }
1799
1800 static void
1801 nine_ff_load_ps_params(struct NineDevice9 *device)
1802 {
1803 const struct nine_state *state = &device->state;
1804 struct fvec4 *dst = (struct fvec4 *)device->ff.ps_const;
1805 unsigned s;
1806
1807 if (!(state->changed.group & (NINE_STATE_FF_PSSTAGES | NINE_STATE_FF_OTHER)))
1808 return;
1809
1810 for (s = 0; s < 8; ++s)
1811 d3dcolor_to_rgba(&dst[s].x, state->ff.tex_stage[s][D3DTSS_CONSTANT]);
1812
1813 for (s = 0; s < 8; ++s) {
1814 dst[8 + s].x = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT00]);
1815 dst[8 + s].y = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT01]);
1816 dst[8 + s].z = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT10]);
1817 dst[8 + s].w = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT11]);
1818 if (s & 1) {
1819 dst[8 + s / 2].z = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLSCALE]);
1820 dst[8 + s / 2].w = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLOFFSET]);
1821 } else {
1822 dst[8 + s / 2].x = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLSCALE]);
1823 dst[8 + s / 2].y = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLOFFSET]);
1824 }
1825 }
1826
1827 d3dcolor_to_rgba(&dst[20].x, state->rs[D3DRS_TEXTUREFACTOR]);
1828 d3dcolor_to_rgba(&dst[21].x, state->rs[D3DRS_FOGCOLOR]);
1829 dst[22].x = asfloat(state->rs[D3DRS_FOGEND]);
1830 dst[22].y = 1.0f / (asfloat(state->rs[D3DRS_FOGEND]) - asfloat(state->rs[D3DRS_FOGSTART]));
1831 dst[22].z = asfloat(state->rs[D3DRS_FOGDENSITY]);
1832 }
1833
1834 static void
1835 nine_ff_load_viewport_info(struct NineDevice9 *device)
1836 {
1837 D3DVIEWPORT9 *viewport = &device->state.viewport;
1838 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1839 float diffZ = viewport->MaxZ - viewport->MinZ;
1840
1841 /* Note: the other functions avoids to fill the const again if nothing changed.
1842 * But we don't have much to fill, and adding code to allow that may be complex
1843 * so just fill it always */
1844 dst[100].x = 2.0f / (float)(viewport->Width);
1845 dst[100].y = 2.0f / (float)(viewport->Height);
1846 dst[100].z = (diffZ == 0.0f) ? 0.0f : (1.0f / diffZ);
1847 dst[101].x = (float)(viewport->X);
1848 dst[101].y = (float)(viewport->Y);
1849 dst[101].z = (float)(viewport->MinZ);
1850 }
1851
1852 void
1853 nine_ff_update(struct NineDevice9 *device)
1854 {
1855 struct nine_state *state = &device->state;
1856 struct pipe_constant_buffer cb;
1857
1858 DBG("vs=%p ps=%p\n", device->state.vs, device->state.ps);
1859
1860 /* NOTE: the only reference belongs to the hash table */
1861 if (!device->state.vs) {
1862 device->ff.vs = nine_ff_get_vs(device);
1863 device->state.changed.group |= NINE_STATE_VS;
1864 }
1865 if (!device->state.ps) {
1866 device->ff.ps = nine_ff_get_ps(device);
1867 device->state.changed.group |= NINE_STATE_PS;
1868 }
1869
1870 if (!device->state.vs) {
1871 nine_ff_load_vs_transforms(device);
1872 nine_ff_load_tex_matrices(device);
1873 nine_ff_load_lights(device);
1874 nine_ff_load_point_and_fog_params(device);
1875 nine_ff_load_viewport_info(device);
1876
1877 memset(state->ff.changed.transform, 0, sizeof(state->ff.changed.transform));
1878
1879 cb.buffer_offset = 0;
1880 cb.buffer = NULL;
1881 cb.user_buffer = device->ff.vs_const;
1882 cb.buffer_size = NINE_FF_NUM_VS_CONST * 4 * sizeof(float);
1883
1884 if (!device->driver_caps.user_cbufs) {
1885 u_upload_data(device->constbuf_uploader,
1886 0,
1887 cb.buffer_size,
1888 device->constbuf_alignment,
1889 cb.user_buffer,
1890 &cb.buffer_offset,
1891 &cb.buffer);
1892 u_upload_unmap(device->constbuf_uploader);
1893 cb.user_buffer = NULL;
1894 }
1895 state->pipe.cb_vs_ff = cb;
1896 state->commit |= NINE_STATE_COMMIT_CONST_VS;
1897 }
1898
1899 if (!device->state.ps) {
1900 nine_ff_load_ps_params(device);
1901
1902 cb.buffer_offset = 0;
1903 cb.buffer = NULL;
1904 cb.user_buffer = device->ff.ps_const;
1905 cb.buffer_size = NINE_FF_NUM_PS_CONST * 4 * sizeof(float);
1906
1907 if (!device->driver_caps.user_cbufs) {
1908 u_upload_data(device->constbuf_uploader,
1909 0,
1910 cb.buffer_size,
1911 device->constbuf_alignment,
1912 cb.user_buffer,
1913 &cb.buffer_offset,
1914 &cb.buffer);
1915 u_upload_unmap(device->constbuf_uploader);
1916 cb.user_buffer = NULL;
1917 }
1918 state->pipe.cb_ps_ff = cb;
1919 state->commit |= NINE_STATE_COMMIT_CONST_PS;
1920 }
1921
1922 device->state.changed.group &= ~NINE_STATE_FF;
1923 }
1924
1925
1926 boolean
1927 nine_ff_init(struct NineDevice9 *device)
1928 {
1929 device->ff.ht_vs = util_hash_table_create(nine_ff_vs_key_hash,
1930 nine_ff_vs_key_comp);
1931 device->ff.ht_ps = util_hash_table_create(nine_ff_ps_key_hash,
1932 nine_ff_ps_key_comp);
1933
1934 device->ff.ht_fvf = util_hash_table_create(nine_ff_fvf_key_hash,
1935 nine_ff_fvf_key_comp);
1936
1937 device->ff.vs_const = CALLOC(NINE_FF_NUM_VS_CONST, 4 * sizeof(float));
1938 device->ff.ps_const = CALLOC(NINE_FF_NUM_PS_CONST, 4 * sizeof(float));
1939
1940 return device->ff.ht_vs && device->ff.ht_ps &&
1941 device->ff.ht_fvf &&
1942 device->ff.vs_const && device->ff.ps_const;
1943 }
1944
1945 static enum pipe_error nine_ff_ht_delete_cb(void *key, void *value, void *data)
1946 {
1947 NineUnknown_Unbind(NineUnknown(value));
1948 return PIPE_OK;
1949 }
1950
1951 void
1952 nine_ff_fini(struct NineDevice9 *device)
1953 {
1954 if (device->ff.ht_vs) {
1955 util_hash_table_foreach(device->ff.ht_vs, nine_ff_ht_delete_cb, NULL);
1956 util_hash_table_destroy(device->ff.ht_vs);
1957 }
1958 if (device->ff.ht_ps) {
1959 util_hash_table_foreach(device->ff.ht_ps, nine_ff_ht_delete_cb, NULL);
1960 util_hash_table_destroy(device->ff.ht_ps);
1961 }
1962 if (device->ff.ht_fvf) {
1963 util_hash_table_foreach(device->ff.ht_fvf, nine_ff_ht_delete_cb, NULL);
1964 util_hash_table_destroy(device->ff.ht_fvf);
1965 }
1966 device->ff.vs = NULL; /* destroyed by unbinding from hash table */
1967 device->ff.ps = NULL;
1968
1969 FREE(device->ff.vs_const);
1970 FREE(device->ff.ps_const);
1971 }
1972
1973 static void
1974 nine_ff_prune_vs(struct NineDevice9 *device)
1975 {
1976 if (device->ff.num_vs > 100) {
1977 /* could destroy the bound one here, so unbind */
1978 device->pipe->bind_vs_state(device->pipe, NULL);
1979 util_hash_table_foreach(device->ff.ht_vs, nine_ff_ht_delete_cb, NULL);
1980 util_hash_table_clear(device->ff.ht_vs);
1981 device->ff.num_vs = 0;
1982 device->state.changed.group |= NINE_STATE_VS;
1983 }
1984 }
1985 static void
1986 nine_ff_prune_ps(struct NineDevice9 *device)
1987 {
1988 if (device->ff.num_ps > 100) {
1989 /* could destroy the bound one here, so unbind */
1990 device->pipe->bind_fs_state(device->pipe, NULL);
1991 util_hash_table_foreach(device->ff.ht_ps, nine_ff_ht_delete_cb, NULL);
1992 util_hash_table_clear(device->ff.ht_ps);
1993 device->ff.num_ps = 0;
1994 device->state.changed.group |= NINE_STATE_PS;
1995 }
1996 }
1997
1998 /* ========================================================================== */
1999
2000 /* Matrix multiplication:
2001 *
2002 * in memory: 0 1 2 3 (row major)
2003 * 4 5 6 7
2004 * 8 9 a b
2005 * c d e f
2006 *
2007 * cA cB cC cD
2008 * r0 = (r0 * cA) (r0 * cB) . .
2009 * r1 = (r1 * cA) (r1 * cB)
2010 * r2 = (r2 * cA) .
2011 * r3 = (r3 * cA) .
2012 *
2013 * r: (11) (12) (13) (14)
2014 * (21) (22) (23) (24)
2015 * (31) (32) (33) (34)
2016 * (41) (42) (43) (44)
2017 * l: (11 12 13 14)
2018 * (21 22 23 24)
2019 * (31 32 33 34)
2020 * (41 42 43 44)
2021 *
2022 * v: (x y z 1 )
2023 *
2024 * t.xyzw = MUL(v.xxxx, r[0]);
2025 * t.xyzw = MAD(v.yyyy, r[1], t.xyzw);
2026 * t.xyzw = MAD(v.zzzz, r[2], t.xyzw);
2027 * v.xyzw = MAD(v.wwww, r[3], t.xyzw);
2028 *
2029 * v.x = DP4(v, c[0]);
2030 * v.y = DP4(v, c[1]);
2031 * v.z = DP4(v, c[2]);
2032 * v.w = DP4(v, c[3]) = 1
2033 */
2034
2035 /*
2036 static void
2037 nine_D3DMATRIX_print(const D3DMATRIX *M)
2038 {
2039 DBG("\n(%f %f %f %f)\n"
2040 "(%f %f %f %f)\n"
2041 "(%f %f %f %f)\n"
2042 "(%f %f %f %f)\n",
2043 M->m[0][0], M->m[0][1], M->m[0][2], M->m[0][3],
2044 M->m[1][0], M->m[1][1], M->m[1][2], M->m[1][3],
2045 M->m[2][0], M->m[2][1], M->m[2][2], M->m[2][3],
2046 M->m[3][0], M->m[3][1], M->m[3][2], M->m[3][3]);
2047 }
2048 */
2049
2050 static inline float
2051 nine_DP4_row_col(const D3DMATRIX *A, int r, const D3DMATRIX *B, int c)
2052 {
2053 return A->m[r][0] * B->m[0][c] +
2054 A->m[r][1] * B->m[1][c] +
2055 A->m[r][2] * B->m[2][c] +
2056 A->m[r][3] * B->m[3][c];
2057 }
2058
2059 static inline float
2060 nine_DP4_vec_col(const D3DVECTOR *v, const D3DMATRIX *M, int c)
2061 {
2062 return v->x * M->m[0][c] +
2063 v->y * M->m[1][c] +
2064 v->z * M->m[2][c] +
2065 1.0f * M->m[3][c];
2066 }
2067
2068 static inline float
2069 nine_DP3_vec_col(const D3DVECTOR *v, const D3DMATRIX *M, int c)
2070 {
2071 return v->x * M->m[0][c] +
2072 v->y * M->m[1][c] +
2073 v->z * M->m[2][c];
2074 }
2075
2076 void
2077 nine_d3d_matrix_matrix_mul(D3DMATRIX *D, const D3DMATRIX *L, const D3DMATRIX *R)
2078 {
2079 D->_11 = nine_DP4_row_col(L, 0, R, 0);
2080 D->_12 = nine_DP4_row_col(L, 0, R, 1);
2081 D->_13 = nine_DP4_row_col(L, 0, R, 2);
2082 D->_14 = nine_DP4_row_col(L, 0, R, 3);
2083
2084 D->_21 = nine_DP4_row_col(L, 1, R, 0);
2085 D->_22 = nine_DP4_row_col(L, 1, R, 1);
2086 D->_23 = nine_DP4_row_col(L, 1, R, 2);
2087 D->_24 = nine_DP4_row_col(L, 1, R, 3);
2088
2089 D->_31 = nine_DP4_row_col(L, 2, R, 0);
2090 D->_32 = nine_DP4_row_col(L, 2, R, 1);
2091 D->_33 = nine_DP4_row_col(L, 2, R, 2);
2092 D->_34 = nine_DP4_row_col(L, 2, R, 3);
2093
2094 D->_41 = nine_DP4_row_col(L, 3, R, 0);
2095 D->_42 = nine_DP4_row_col(L, 3, R, 1);
2096 D->_43 = nine_DP4_row_col(L, 3, R, 2);
2097 D->_44 = nine_DP4_row_col(L, 3, R, 3);
2098 }
2099
2100 void
2101 nine_d3d_vector4_matrix_mul(D3DVECTOR *d, const D3DVECTOR *v, const D3DMATRIX *M)
2102 {
2103 d->x = nine_DP4_vec_col(v, M, 0);
2104 d->y = nine_DP4_vec_col(v, M, 1);
2105 d->z = nine_DP4_vec_col(v, M, 2);
2106 }
2107
2108 void
2109 nine_d3d_vector3_matrix_mul(D3DVECTOR *d, const D3DVECTOR *v, const D3DMATRIX *M)
2110 {
2111 d->x = nine_DP3_vec_col(v, M, 0);
2112 d->y = nine_DP3_vec_col(v, M, 1);
2113 d->z = nine_DP3_vec_col(v, M, 2);
2114 }
2115
2116 void
2117 nine_d3d_matrix_transpose(D3DMATRIX *D, const D3DMATRIX *M)
2118 {
2119 unsigned i, j;
2120 for (i = 0; i < 4; ++i)
2121 for (j = 0; j < 4; ++j)
2122 D->m[i][j] = M->m[j][i];
2123 }
2124
2125 #define _M_ADD_PROD_1i_2j_3k_4l(i,j,k,l) do { \
2126 float t = M->_1##i * M->_2##j * M->_3##k * M->_4##l; \
2127 if (t > 0.0f) pos += t; else neg += t; } while(0)
2128
2129 #define _M_SUB_PROD_1i_2j_3k_4l(i,j,k,l) do { \
2130 float t = M->_1##i * M->_2##j * M->_3##k * M->_4##l; \
2131 if (t > 0.0f) neg -= t; else pos -= t; } while(0)
2132 float
2133 nine_d3d_matrix_det(const D3DMATRIX *M)
2134 {
2135 float pos = 0.0f;
2136 float neg = 0.0f;
2137
2138 _M_ADD_PROD_1i_2j_3k_4l(1, 2, 3, 4);
2139 _M_ADD_PROD_1i_2j_3k_4l(1, 3, 4, 2);
2140 _M_ADD_PROD_1i_2j_3k_4l(1, 4, 2, 3);
2141
2142 _M_ADD_PROD_1i_2j_3k_4l(2, 1, 4, 3);
2143 _M_ADD_PROD_1i_2j_3k_4l(2, 3, 1, 4);
2144 _M_ADD_PROD_1i_2j_3k_4l(2, 4, 3, 1);
2145
2146 _M_ADD_PROD_1i_2j_3k_4l(3, 1, 2, 4);
2147 _M_ADD_PROD_1i_2j_3k_4l(3, 2, 4, 1);
2148 _M_ADD_PROD_1i_2j_3k_4l(3, 4, 1, 2);
2149
2150 _M_ADD_PROD_1i_2j_3k_4l(4, 1, 3, 2);
2151 _M_ADD_PROD_1i_2j_3k_4l(4, 2, 1, 3);
2152 _M_ADD_PROD_1i_2j_3k_4l(4, 3, 2, 1);
2153
2154 _M_SUB_PROD_1i_2j_3k_4l(1, 2, 4, 3);
2155 _M_SUB_PROD_1i_2j_3k_4l(1, 3, 2, 4);
2156 _M_SUB_PROD_1i_2j_3k_4l(1, 4, 3, 2);
2157
2158 _M_SUB_PROD_1i_2j_3k_4l(2, 1, 3, 4);
2159 _M_SUB_PROD_1i_2j_3k_4l(2, 3, 4, 1);
2160 _M_SUB_PROD_1i_2j_3k_4l(2, 4, 1, 3);
2161
2162 _M_SUB_PROD_1i_2j_3k_4l(3, 1, 4, 2);
2163 _M_SUB_PROD_1i_2j_3k_4l(3, 2, 1, 4);
2164 _M_SUB_PROD_1i_2j_3k_4l(3, 4, 2, 1);
2165
2166 _M_SUB_PROD_1i_2j_3k_4l(4, 1, 2, 3);
2167 _M_SUB_PROD_1i_2j_3k_4l(4, 2, 3, 1);
2168 _M_SUB_PROD_1i_2j_3k_4l(4, 3, 1, 2);
2169
2170 return pos + neg;
2171 }
2172
2173 /* XXX: Probably better to just use src/mesa/math/m_matrix.c because
2174 * I have no idea where this code came from.
2175 */
2176 void
2177 nine_d3d_matrix_inverse(D3DMATRIX *D, const D3DMATRIX *M)
2178 {
2179 int i, k;
2180 float det;
2181
2182 D->m[0][0] =
2183 M->m[1][1] * M->m[2][2] * M->m[3][3] -
2184 M->m[1][1] * M->m[3][2] * M->m[2][3] -
2185 M->m[1][2] * M->m[2][1] * M->m[3][3] +
2186 M->m[1][2] * M->m[3][1] * M->m[2][3] +
2187 M->m[1][3] * M->m[2][1] * M->m[3][2] -
2188 M->m[1][3] * M->m[3][1] * M->m[2][2];
2189
2190 D->m[0][1] =
2191 -M->m[0][1] * M->m[2][2] * M->m[3][3] +
2192 M->m[0][1] * M->m[3][2] * M->m[2][3] +
2193 M->m[0][2] * M->m[2][1] * M->m[3][3] -
2194 M->m[0][2] * M->m[3][1] * M->m[2][3] -
2195 M->m[0][3] * M->m[2][1] * M->m[3][2] +
2196 M->m[0][3] * M->m[3][1] * M->m[2][2];
2197
2198 D->m[0][2] =
2199 M->m[0][1] * M->m[1][2] * M->m[3][3] -
2200 M->m[0][1] * M->m[3][2] * M->m[1][3] -
2201 M->m[0][2] * M->m[1][1] * M->m[3][3] +
2202 M->m[0][2] * M->m[3][1] * M->m[1][3] +
2203 M->m[0][3] * M->m[1][1] * M->m[3][2] -
2204 M->m[0][3] * M->m[3][1] * M->m[1][2];
2205
2206 D->m[0][3] =
2207 -M->m[0][1] * M->m[1][2] * M->m[2][3] +
2208 M->m[0][1] * M->m[2][2] * M->m[1][3] +
2209 M->m[0][2] * M->m[1][1] * M->m[2][3] -
2210 M->m[0][2] * M->m[2][1] * M->m[1][3] -
2211 M->m[0][3] * M->m[1][1] * M->m[2][2] +
2212 M->m[0][3] * M->m[2][1] * M->m[1][2];
2213
2214 D->m[1][0] =
2215 -M->m[1][0] * M->m[2][2] * M->m[3][3] +
2216 M->m[1][0] * M->m[3][2] * M->m[2][3] +
2217 M->m[1][2] * M->m[2][0] * M->m[3][3] -
2218 M->m[1][2] * M->m[3][0] * M->m[2][3] -
2219 M->m[1][3] * M->m[2][0] * M->m[3][2] +
2220 M->m[1][3] * M->m[3][0] * M->m[2][2];
2221
2222 D->m[1][1] =
2223 M->m[0][0] * M->m[2][2] * M->m[3][3] -
2224 M->m[0][0] * M->m[3][2] * M->m[2][3] -
2225 M->m[0][2] * M->m[2][0] * M->m[3][3] +
2226 M->m[0][2] * M->m[3][0] * M->m[2][3] +
2227 M->m[0][3] * M->m[2][0] * M->m[3][2] -
2228 M->m[0][3] * M->m[3][0] * M->m[2][2];
2229
2230 D->m[1][2] =
2231 -M->m[0][0] * M->m[1][2] * M->m[3][3] +
2232 M->m[0][0] * M->m[3][2] * M->m[1][3] +
2233 M->m[0][2] * M->m[1][0] * M->m[3][3] -
2234 M->m[0][2] * M->m[3][0] * M->m[1][3] -
2235 M->m[0][3] * M->m[1][0] * M->m[3][2] +
2236 M->m[0][3] * M->m[3][0] * M->m[1][2];
2237
2238 D->m[1][3] =
2239 M->m[0][0] * M->m[1][2] * M->m[2][3] -
2240 M->m[0][0] * M->m[2][2] * M->m[1][3] -
2241 M->m[0][2] * M->m[1][0] * M->m[2][3] +
2242 M->m[0][2] * M->m[2][0] * M->m[1][3] +
2243 M->m[0][3] * M->m[1][0] * M->m[2][2] -
2244 M->m[0][3] * M->m[2][0] * M->m[1][2];
2245
2246 D->m[2][0] =
2247 M->m[1][0] * M->m[2][1] * M->m[3][3] -
2248 M->m[1][0] * M->m[3][1] * M->m[2][3] -
2249 M->m[1][1] * M->m[2][0] * M->m[3][3] +
2250 M->m[1][1] * M->m[3][0] * M->m[2][3] +
2251 M->m[1][3] * M->m[2][0] * M->m[3][1] -
2252 M->m[1][3] * M->m[3][0] * M->m[2][1];
2253
2254 D->m[2][1] =
2255 -M->m[0][0] * M->m[2][1] * M->m[3][3] +
2256 M->m[0][0] * M->m[3][1] * M->m[2][3] +
2257 M->m[0][1] * M->m[2][0] * M->m[3][3] -
2258 M->m[0][1] * M->m[3][0] * M->m[2][3] -
2259 M->m[0][3] * M->m[2][0] * M->m[3][1] +
2260 M->m[0][3] * M->m[3][0] * M->m[2][1];
2261
2262 D->m[2][2] =
2263 M->m[0][0] * M->m[1][1] * M->m[3][3] -
2264 M->m[0][0] * M->m[3][1] * M->m[1][3] -
2265 M->m[0][1] * M->m[1][0] * M->m[3][3] +
2266 M->m[0][1] * M->m[3][0] * M->m[1][3] +
2267 M->m[0][3] * M->m[1][0] * M->m[3][1] -
2268 M->m[0][3] * M->m[3][0] * M->m[1][1];
2269
2270 D->m[2][3] =
2271 -M->m[0][0] * M->m[1][1] * M->m[2][3] +
2272 M->m[0][0] * M->m[2][1] * M->m[1][3] +
2273 M->m[0][1] * M->m[1][0] * M->m[2][3] -
2274 M->m[0][1] * M->m[2][0] * M->m[1][3] -
2275 M->m[0][3] * M->m[1][0] * M->m[2][1] +
2276 M->m[0][3] * M->m[2][0] * M->m[1][1];
2277
2278 D->m[3][0] =
2279 -M->m[1][0] * M->m[2][1] * M->m[3][2] +
2280 M->m[1][0] * M->m[3][1] * M->m[2][2] +
2281 M->m[1][1] * M->m[2][0] * M->m[3][2] -
2282 M->m[1][1] * M->m[3][0] * M->m[2][2] -
2283 M->m[1][2] * M->m[2][0] * M->m[3][1] +
2284 M->m[1][2] * M->m[3][0] * M->m[2][1];
2285
2286 D->m[3][1] =
2287 M->m[0][0] * M->m[2][1] * M->m[3][2] -
2288 M->m[0][0] * M->m[3][1] * M->m[2][2] -
2289 M->m[0][1] * M->m[2][0] * M->m[3][2] +
2290 M->m[0][1] * M->m[3][0] * M->m[2][2] +
2291 M->m[0][2] * M->m[2][0] * M->m[3][1] -
2292 M->m[0][2] * M->m[3][0] * M->m[2][1];
2293
2294 D->m[3][2] =
2295 -M->m[0][0] * M->m[1][1] * M->m[3][2] +
2296 M->m[0][0] * M->m[3][1] * M->m[1][2] +
2297 M->m[0][1] * M->m[1][0] * M->m[3][2] -
2298 M->m[0][1] * M->m[3][0] * M->m[1][2] -
2299 M->m[0][2] * M->m[1][0] * M->m[3][1] +
2300 M->m[0][2] * M->m[3][0] * M->m[1][1];
2301
2302 D->m[3][3] =
2303 M->m[0][0] * M->m[1][1] * M->m[2][2] -
2304 M->m[0][0] * M->m[2][1] * M->m[1][2] -
2305 M->m[0][1] * M->m[1][0] * M->m[2][2] +
2306 M->m[0][1] * M->m[2][0] * M->m[1][2] +
2307 M->m[0][2] * M->m[1][0] * M->m[2][1] -
2308 M->m[0][2] * M->m[2][0] * M->m[1][1];
2309
2310 det =
2311 M->m[0][0] * D->m[0][0] +
2312 M->m[1][0] * D->m[0][1] +
2313 M->m[2][0] * D->m[0][2] +
2314 M->m[3][0] * D->m[0][3];
2315
2316 det = 1.0 / det;
2317
2318 for (i = 0; i < 4; i++)
2319 for (k = 0; k < 4; k++)
2320 D->m[i][k] *= det;
2321
2322 #ifdef DEBUG
2323 {
2324 D3DMATRIX I;
2325
2326 nine_d3d_matrix_matrix_mul(&I, D, M);
2327
2328 for (i = 0; i < 4; ++i)
2329 for (k = 0; k < 4; ++k)
2330 if (fabsf(I.m[i][k] - (float)(i == k)) > 1e-3)
2331 DBG("Matrix inversion check FAILED !\n");
2332 }
2333 #endif
2334 }
2335
2336 /* TODO: don't use 4x4 inverse, unless this gets all nicely inlined ? */
2337 void
2338 nine_d3d_matrix_inverse_3x3(D3DMATRIX *D, const D3DMATRIX *M)
2339 {
2340 D3DMATRIX T;
2341 unsigned i, j;
2342
2343 for (i = 0; i < 3; ++i)
2344 for (j = 0; j < 3; ++j)
2345 T.m[i][j] = M->m[i][j];
2346 for (i = 0; i < 3; ++i) {
2347 T.m[i][3] = 0.0f;
2348 T.m[3][i] = 0.0f;
2349 }
2350 T.m[3][3] = 1.0f;
2351
2352 nine_d3d_matrix_inverse(D, &T);
2353 }