projects / mesa.git / blob
? search:


7b26c954d3fdbd653791e8ba79b775cfc9752c43
[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 DBG_CHANNEL DBG_FF
28
29 #define NINE_FF_NUM_VS_CONST 256
30 #define NINE_FF_NUM_PS_CONST 24
31
32 struct fvec4
33 {
34 float x, y, z, w;
35 };
36
37 struct nine_ff_vs_key
38 {
39 union {
40 struct {
41 uint32_t position_t : 1;
42 uint32_t lighting : 1;
43 uint32_t darkness : 1; /* lighting enabled but no active lights */
44 uint32_t localviewer : 1;
45 uint32_t vertexpointsize : 1;
46 uint32_t pointscale : 1;
47 uint32_t vertexblend : 3;
48 uint32_t vertexblend_indexed : 1;
49 uint32_t vertextween : 1;
50 uint32_t mtl_diffuse : 2; /* 0 = material, 1 = color1, 2 = color2 */
51 uint32_t mtl_ambient : 2;
52 uint32_t mtl_specular : 2;
53 uint32_t mtl_emissive : 2;
54 uint32_t fog_mode : 2;
55 uint32_t fog_range : 1;
56 uint32_t color0in_one : 1;
57 uint32_t color1in_one : 1;
58 uint32_t fog : 1;
59 uint32_t specular_enable : 1;
60 uint32_t pad1 : 6;
61 uint32_t tc_dim_input: 16; /* 8 * 2 bits */
62 uint32_t pad2 : 16;
63 uint32_t tc_dim_output: 24; /* 8 * 3 bits */
64 uint32_t pad3 : 8;
65 uint32_t tc_gen : 24; /* 8 * 3 bits */
66 uint32_t pad4 : 8;
67 uint32_t tc_idx : 24;
68 uint32_t pad5 : 8;
69 uint32_t passthrough;
70 };
71 uint64_t value64[3]; /* don't forget to resize VertexShader9.ff_key */
72 uint32_t value32[6];
73 };
74 };
75
76 /* Texture stage state:
77 *
78 * COLOROP D3DTOP 5 bit
79 * ALPHAOP D3DTOP 5 bit
80 * COLORARG0 D3DTA 3 bit
81 * COLORARG1 D3DTA 3 bit
82 * COLORARG2 D3DTA 3 bit
83 * ALPHAARG0 D3DTA 3 bit
84 * ALPHAARG1 D3DTA 3 bit
85 * ALPHAARG2 D3DTA 3 bit
86 * RESULTARG D3DTA 1 bit (CURRENT:0 or TEMP:1)
87 * TEXCOORDINDEX 0 - 7 3 bit
88 * ===========================
89 * 32 bit per stage
90 */
91 struct nine_ff_ps_key
92 {
93 union {
94 struct {
95 struct {
96 uint32_t colorop : 5;
97 uint32_t alphaop : 5;
98 uint32_t colorarg0 : 3;
99 uint32_t colorarg1 : 3;
100 uint32_t colorarg2 : 3;
101 uint32_t alphaarg0 : 3;
102 uint32_t alphaarg1 : 3;
103 uint32_t alphaarg2 : 3;
104 uint32_t resultarg : 1; /* CURRENT:0 or TEMP:1 */
105 uint32_t textarget : 2; /* 1D/2D/3D/CUBE */
106 uint32_t pad : 1;
107 /* that's 32 bit exactly */
108 } ts[8];
109 uint32_t projected : 16;
110 uint32_t fog : 1; /* for vFog coming from VS */
111 uint32_t fog_mode : 2;
112 uint32_t specular : 1;
113 uint32_t pad1 : 12; /* 9 32-bit words with this */
114 uint8_t colorarg_b4[3];
115 uint8_t colorarg_b5[3];
116 uint8_t alphaarg_b4[3]; /* 11 32-bit words plus a byte */
117 uint8_t pad2[3];
118 };
119 uint64_t value64[6]; /* don't forget to resize PixelShader9.ff_key */
120 uint32_t value32[12];
121 };
122 };
123
124 static unsigned nine_ff_vs_key_hash(void *key)
125 {
126 struct nine_ff_vs_key *vs = key;
127 unsigned i;
128 uint32_t hash = vs->value32[0];
129 for (i = 1; i < Elements(vs->value32); ++i)
130 hash ^= vs->value32[i];
131 return hash;
132 }
133 static int nine_ff_vs_key_comp(void *key1, void *key2)
134 {
135 struct nine_ff_vs_key *a = (struct nine_ff_vs_key *)key1;
136 struct nine_ff_vs_key *b = (struct nine_ff_vs_key *)key2;
137
138 return memcmp(a->value64, b->value64, sizeof(a->value64));
139 }
140 static unsigned nine_ff_ps_key_hash(void *key)
141 {
142 struct nine_ff_ps_key *ps = key;
143 unsigned i;
144 uint32_t hash = ps->value32[0];
145 for (i = 1; i < Elements(ps->value32); ++i)
146 hash ^= ps->value32[i];
147 return hash;
148 }
149 static int nine_ff_ps_key_comp(void *key1, void *key2)
150 {
151 struct nine_ff_ps_key *a = (struct nine_ff_ps_key *)key1;
152 struct nine_ff_ps_key *b = (struct nine_ff_ps_key *)key2;
153
154 return memcmp(a->value64, b->value64, sizeof(a->value64));
155 }
156 static unsigned nine_ff_fvf_key_hash(void *key)
157 {
158 return *(DWORD *)key;
159 }
160 static int nine_ff_fvf_key_comp(void *key1, void *key2)
161 {
162 return *(DWORD *)key1 != *(DWORD *)key2;
163 }
164
165 static void nine_ff_prune_vs(struct NineDevice9 *);
166 static void nine_ff_prune_ps(struct NineDevice9 *);
167
168 static void nine_ureg_tgsi_dump(struct ureg_program *ureg, boolean override)
169 {
170 if (debug_get_bool_option("NINE_FF_DUMP", FALSE) || override) {
171 unsigned count;
172 const struct tgsi_token *toks = ureg_get_tokens(ureg, &count);
173 tgsi_dump(toks, 0);
174 ureg_free_tokens(toks);
175 }
176 }
177
178 #define _X(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_X)
179 #define _Y(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_Y)
180 #define _Z(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_Z)
181 #define _W(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_W)
182
183 #define _XXXX(r) ureg_scalar(r, TGSI_SWIZZLE_X)
184 #define _YYYY(r) ureg_scalar(r, TGSI_SWIZZLE_Y)
185 #define _ZZZZ(r) ureg_scalar(r, TGSI_SWIZZLE_Z)
186 #define _WWWW(r) ureg_scalar(r, TGSI_SWIZZLE_W)
187
188 #define _XYZW(r) (r)
189
190 /* AL should contain base address of lights table. */
191 #define LIGHT_CONST(i) \
192 ureg_src_indirect(ureg_DECL_constant(ureg, i), _X(AL))
193
194 #define MATERIAL_CONST(i) \
195 ureg_DECL_constant(ureg, 19 + (i))
196
197 #define _CONST(n) ureg_DECL_constant(ureg, n)
198
199 /* VS FF constants layout:
200 *
201 * CONST[ 0.. 3] D3DTS_WORLD * D3DTS_VIEW * D3DTS_PROJECTION
202 * CONST[ 4.. 7] D3DTS_WORLD * D3DTS_VIEW
203 * CONST[ 8..11] D3DTS_VIEW * D3DTS_PROJECTION
204 * CONST[12..15] D3DTS_VIEW
205 * CONST[16..18] Normal matrix
206 *
207 * CONST[19] MATERIAL.Emissive + Material.Ambient * RS.Ambient
208 * CONST[20] MATERIAL.Diffuse
209 * CONST[21] MATERIAL.Ambient
210 * CONST[22] MATERIAL.Specular
211 * CONST[23].x___ MATERIAL.Power
212 * CONST[24] MATERIAL.Emissive
213 * CONST[25] RS.Ambient
214 *
215 * CONST[26].x___ RS.PointSizeMin
216 * CONST[26]._y__ RS.PointSizeMax
217 * CONST[26].__z_ RS.PointSize
218 * CONST[26].___w RS.PointScaleA
219 * CONST[27].x___ RS.PointScaleB
220 * CONST[27]._y__ RS.PointScaleC
221 *
222 * CONST[28].x___ RS.FogEnd
223 * CONST[28]._y__ 1.0f / (RS.FogEnd - RS.FogStart)
224 * CONST[28].__z_ RS.FogDensity
225
226 * CONST[30].x___ TWEENFACTOR
227 *
228 * CONST[32].x___ LIGHT[0].Type
229 * CONST[32]._yzw LIGHT[0].Attenuation0,1,2
230 * CONST[33] LIGHT[0].Diffuse
231 * CONST[34] LIGHT[0].Specular
232 * CONST[35] LIGHT[0].Ambient
233 * CONST[36].xyz_ LIGHT[0].Position
234 * CONST[36].___w LIGHT[0].Range
235 * CONST[37].xyz_ LIGHT[0].Direction
236 * CONST[37].___w LIGHT[0].Falloff
237 * CONST[38].x___ cos(LIGHT[0].Theta / 2)
238 * CONST[38]._y__ cos(LIGHT[0].Phi / 2)
239 * CONST[38].__z_ 1.0f / (cos(LIGHT[0].Theta / 2) - cos(Light[0].Phi / 2))
240 * CONST[39].xyz_ LIGHT[0].HalfVector (for directional lights)
241 * CONST[39].___w 1 if this is the last active light, 0 if not
242 * CONST[40] LIGHT[1]
243 * CONST[48] LIGHT[2]
244 * CONST[56] LIGHT[3]
245 * CONST[64] LIGHT[4]
246 * CONST[72] LIGHT[5]
247 * CONST[80] LIGHT[6]
248 * CONST[88] LIGHT[7]
249 * NOTE: no lighting code is generated if there are no active lights
250 *
251 * CONST[100].x___ Viewport 2/width
252 * CONST[100]._y__ Viewport 2/height
253 * CONST[100].__z_ Viewport 1/(zmax - zmin)
254 * CONST[101].x___ Viewport x0
255 * CONST[101]._y__ Viewport y0
256 * CONST[101].__z_ Viewport z0
257 *
258 * CONST[128..131] D3DTS_TEXTURE0
259 * CONST[132..135] D3DTS_TEXTURE1
260 * CONST[136..139] D3DTS_TEXTURE2
261 * CONST[140..143] D3DTS_TEXTURE3
262 * CONST[144..147] D3DTS_TEXTURE4
263 * CONST[148..151] D3DTS_TEXTURE5
264 * CONST[152..155] D3DTS_TEXTURE6
265 * CONST[156..159] D3DTS_TEXTURE7
266 *
267 * CONST[224] D3DTS_WORLDMATRIX[0]
268 * CONST[228] D3DTS_WORLDMATRIX[1]
269 * ...
270 * CONST[252] D3DTS_WORLDMATRIX[7]
271 */
272 struct vs_build_ctx
273 {
274 struct ureg_program *ureg;
275 const struct nine_ff_vs_key *key;
276
277 uint16_t input[PIPE_MAX_ATTRIBS];
278 unsigned num_inputs;
279
280 struct ureg_src aVtx;
281 struct ureg_src aNrm;
282 struct ureg_src aCol[2];
283 struct ureg_src aTex[8];
284 struct ureg_src aPsz;
285 struct ureg_src aInd;
286 struct ureg_src aWgt;
287
288 struct ureg_src aVtx1; /* tweening */
289 struct ureg_src aNrm1;
290
291 struct ureg_src mtlA;
292 struct ureg_src mtlD;
293 struct ureg_src mtlS;
294 struct ureg_src mtlE;
295 };
296
297 static inline unsigned
298 get_texcoord_sn(struct pipe_screen *screen)
299 {
300 if (screen->get_param(screen, PIPE_CAP_TGSI_TEXCOORD))
301 return TGSI_SEMANTIC_TEXCOORD;
302 return TGSI_SEMANTIC_GENERIC;
303 }
304
305 static inline struct ureg_src
306 build_vs_add_input(struct vs_build_ctx *vs, uint16_t ndecl)
307 {
308 const unsigned i = vs->num_inputs++;
309 assert(i < PIPE_MAX_ATTRIBS);
310 vs->input[i] = ndecl;
311 return ureg_DECL_vs_input(vs->ureg, i);
312 }
313
314 /* NOTE: dst may alias src */
315 static inline void
316 ureg_normalize3(struct ureg_program *ureg,
317 struct ureg_dst dst, struct ureg_src src,
318 struct ureg_dst tmp)
319 {
320 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
321
322 ureg_DP3(ureg, tmp_x, src, src);
323 ureg_RSQ(ureg, tmp_x, _X(tmp));
324 ureg_MUL(ureg, dst, src, _X(tmp));
325 }
326
327 static void *
328 nine_ff_build_vs(struct NineDevice9 *device, struct vs_build_ctx *vs)
329 {
330 const struct nine_ff_vs_key *key = vs->key;
331 struct ureg_program *ureg = ureg_create(PIPE_SHADER_VERTEX);
332 struct ureg_dst oPos, oCol[2], oPsz, oFog;
333 struct ureg_dst rVtx, rNrm;
334 struct ureg_dst r[8];
335 struct ureg_dst AR;
336 struct ureg_dst tmp, tmp_x, tmp_y, tmp_z;
337 unsigned i, c;
338 unsigned label[32], l = 0;
339 unsigned num_r = 8;
340 boolean need_rNrm = key->lighting || key->pointscale || key->passthrough & (1 << NINE_DECLUSAGE_NORMAL);
341 boolean need_rVtx = key->lighting || key->fog_mode;
342 const unsigned texcoord_sn = get_texcoord_sn(device->screen);
343
344 vs->ureg = ureg;
345
346 /* Check which inputs we should transform. */
347 for (i = 0; i < 8 * 3; i += 3) {
348 switch ((key->tc_gen >> i) & 0x3) {
349 case NINED3DTSS_TCI_CAMERASPACENORMAL:
350 need_rNrm = TRUE;
351 break;
352 case NINED3DTSS_TCI_CAMERASPACEPOSITION:
353 need_rVtx = TRUE;
354 break;
355 case NINED3DTSS_TCI_CAMERASPACEREFLECTIONVECTOR:
356 need_rVtx = need_rNrm = TRUE;
357 break;
358 default:
359 break;
360 }
361 }
362
363 /* Declare and record used inputs (needed for linkage with vertex format):
364 * (texture coordinates handled later)
365 */
366 vs->aVtx = build_vs_add_input(vs,
367 key->position_t ? NINE_DECLUSAGE_POSITIONT : NINE_DECLUSAGE_POSITION);
368
369 if (need_rNrm)
370 vs->aNrm = build_vs_add_input(vs, NINE_DECLUSAGE_NORMAL);
371
372 vs->aCol[0] = ureg_imm1f(ureg, 1.0f);
373 vs->aCol[1] = ureg_imm1f(ureg, 1.0f);
374
375 if (key->lighting || key->darkness) {
376 const unsigned mask = key->mtl_diffuse | key->mtl_specular |
377 key->mtl_ambient | key->mtl_emissive;
378 if ((mask & 0x1) && !key->color0in_one)
379 vs->aCol[0] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 0));
380 if ((mask & 0x2) && !key->color1in_one)
381 vs->aCol[1] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 1));
382
383 vs->mtlD = MATERIAL_CONST(1);
384 vs->mtlA = MATERIAL_CONST(2);
385 vs->mtlS = MATERIAL_CONST(3);
386 vs->mtlE = MATERIAL_CONST(5);
387 if (key->mtl_diffuse == 1) vs->mtlD = vs->aCol[0]; else
388 if (key->mtl_diffuse == 2) vs->mtlD = vs->aCol[1];
389 if (key->mtl_ambient == 1) vs->mtlA = vs->aCol[0]; else
390 if (key->mtl_ambient == 2) vs->mtlA = vs->aCol[1];
391 if (key->mtl_specular == 1) vs->mtlS = vs->aCol[0]; else
392 if (key->mtl_specular == 2) vs->mtlS = vs->aCol[1];
393 if (key->mtl_emissive == 1) vs->mtlE = vs->aCol[0]; else
394 if (key->mtl_emissive == 2) vs->mtlE = vs->aCol[1];
395 } else {
396 if (!key->color0in_one) vs->aCol[0] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 0));
397 if (!key->color1in_one) vs->aCol[1] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 1));
398 }
399
400 if (key->vertexpointsize)
401 vs->aPsz = build_vs_add_input(vs, NINE_DECLUSAGE_PSIZE);
402
403 if (key->vertexblend_indexed || key->passthrough & (1 << NINE_DECLUSAGE_BLENDINDICES))
404 vs->aInd = build_vs_add_input(vs, NINE_DECLUSAGE_BLENDINDICES);
405 if (key->vertexblend || key->passthrough & (1 << NINE_DECLUSAGE_BLENDWEIGHT))
406 vs->aWgt = build_vs_add_input(vs, NINE_DECLUSAGE_BLENDWEIGHT);
407 if (key->vertextween) {
408 vs->aVtx1 = build_vs_add_input(vs, NINE_DECLUSAGE_i(POSITION,1));
409 vs->aNrm1 = build_vs_add_input(vs, NINE_DECLUSAGE_i(NORMAL,1));
410 }
411
412 /* Declare outputs:
413 */
414 oPos = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0); /* HPOS */
415 oCol[0] = ureg_saturate(ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0));
416 oCol[1] = ureg_saturate(ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 1));
417 if (key->fog || key->passthrough & (1 << NINE_DECLUSAGE_FOG)) {
418 oFog = ureg_DECL_output(ureg, TGSI_SEMANTIC_FOG, 0);
419 oFog = ureg_writemask(oFog, TGSI_WRITEMASK_X);
420 }
421
422 if (key->vertexpointsize || key->pointscale) {
423 oPsz = ureg_DECL_output_masked(ureg, TGSI_SEMANTIC_PSIZE, 0,
424 TGSI_WRITEMASK_X, 0, 1);
425 oPsz = ureg_writemask(oPsz, TGSI_WRITEMASK_X);
426 }
427
428 /* Declare TEMPs:
429 */
430 for (i = 0; i < num_r; ++i)
431 r[i] = ureg_DECL_local_temporary(ureg);
432 tmp = r[0];
433 tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
434 tmp_y = ureg_writemask(tmp, TGSI_WRITEMASK_Y);
435 tmp_z = ureg_writemask(tmp, TGSI_WRITEMASK_Z);
436 if (key->lighting || key->vertexblend)
437 AR = ureg_DECL_address(ureg);
438
439 rVtx = ureg_writemask(r[1], TGSI_WRITEMASK_XYZ);
440 rNrm = ureg_writemask(r[2], TGSI_WRITEMASK_XYZ);
441
442 /* === Vertex transformation / vertex blending:
443 */
444 if (key->vertextween) {
445 assert(!key->vertexblend);
446 ureg_LRP(ureg, r[2], _XXXX(_CONST(30)), vs->aVtx, vs->aVtx1);
447 if (need_rNrm)
448 ureg_LRP(ureg, r[3], _XXXX(_CONST(30)), vs->aNrm, vs->aNrm1);
449 vs->aVtx = ureg_src(r[2]);
450 vs->aNrm = ureg_src(r[3]);
451 }
452
453 if (key->vertexblend) {
454 struct ureg_src cWM[4];
455
456 for (i = 224; i <= 255; ++i)
457 ureg_DECL_constant(ureg, i);
458
459 /* translate world matrix index to constant file index */
460 if (key->vertexblend_indexed) {
461 ureg_MAD(ureg, tmp, vs->aInd, ureg_imm1f(ureg, 4.0f), ureg_imm1f(ureg, 224.0f));
462 ureg_ARL(ureg, AR, ureg_src(tmp));
463 }
464
465 ureg_MOV(ureg, r[2], ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 0.0f));
466 ureg_MOV(ureg, r[3], ureg_imm4f(ureg, 1.0f, 1.0f, 1.0f, 1.0f));
467
468 for (i = 0; i < key->vertexblend; ++i) {
469 for (c = 0; c < 4; ++c) {
470 cWM[c] = ureg_src_register(TGSI_FILE_CONSTANT, (224 + i * 4) * !key->vertexblend_indexed + c);
471 if (key->vertexblend_indexed)
472 cWM[c] = ureg_src_indirect(cWM[c], ureg_scalar(ureg_src(AR), i));
473 }
474 /* multiply by WORLD(index) */
475 ureg_MUL(ureg, tmp, _XXXX(vs->aVtx), cWM[0]);
476 ureg_MAD(ureg, tmp, _YYYY(vs->aVtx), cWM[1], ureg_src(tmp));
477 ureg_MAD(ureg, tmp, _ZZZZ(vs->aVtx), cWM[2], ureg_src(tmp));
478 ureg_MAD(ureg, tmp, _WWWW(vs->aVtx), cWM[3], ureg_src(tmp));
479
480 if (i < (key->vertexblend - 1)) {
481 /* accumulate weighted position value */
482 ureg_MAD(ureg, r[2], ureg_src(tmp), ureg_scalar(vs->aWgt, i), ureg_src(r[2]));
483 /* subtract weighted position value for last value */
484 ureg_SUB(ureg, r[3], ureg_src(r[3]), ureg_scalar(vs->aWgt, i));
485 }
486 }
487
488 /* the last weighted position is always 1 - sum_of_previous_weights */
489 ureg_MAD(ureg, r[2], ureg_src(tmp), ureg_scalar(ureg_src(r[3]), key->vertexblend - 1), ureg_src(r[2]));
490
491 /* multiply by VIEW_PROJ */
492 ureg_MUL(ureg, tmp, _X(r[2]), _CONST(8));
493 ureg_MAD(ureg, tmp, _Y(r[2]), _CONST(9), ureg_src(tmp));
494 ureg_MAD(ureg, tmp, _Z(r[2]), _CONST(10), ureg_src(tmp));
495 ureg_MAD(ureg, oPos, _W(r[2]), _CONST(11), ureg_src(tmp));
496
497 if (need_rVtx)
498 vs->aVtx = ureg_src(r[2]);
499 } else
500 if (key->position_t && device->driver_caps.window_space_position_support) {
501 ureg_MOV(ureg, oPos, vs->aVtx);
502 } else if (key->position_t) {
503 /* vs->aVtx contains the coordinates buffer wise.
504 * later in the pipeline, clipping, viewport and division
505 * by w (rhw = 1/w) are going to be applied, so do the reverse
506 * of these transformations (except clipping) to have the good
507 * position at the end.*/
508 ureg_MOV(ureg, tmp, vs->aVtx);
509 /* X from [X_min, X_min + width] to [-1, 1], same for Y. Z to [0, 1] */
510 ureg_SUB(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _CONST(101));
511 ureg_MUL(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _CONST(100));
512 ureg_SUB(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XY), ureg_src(tmp), ureg_imm1f(ureg, 1.0f));
513 /* Y needs to be reversed */
514 ureg_MOV(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_Y), ureg_negate(ureg_src(tmp)));
515 /* inverse rhw */
516 ureg_RCP(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_W), _W(tmp));
517 /* multiply X, Y, Z by w */
518 ureg_MUL(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _W(tmp));
519 ureg_MOV(ureg, oPos, ureg_src(tmp));
520 } else {
521 /* position = vertex * WORLD_VIEW_PROJ */
522 ureg_MUL(ureg, tmp, _XXXX(vs->aVtx), _CONST(0));
523 ureg_MAD(ureg, tmp, _YYYY(vs->aVtx), _CONST(1), ureg_src(tmp));
524 ureg_MAD(ureg, tmp, _ZZZZ(vs->aVtx), _CONST(2), ureg_src(tmp));
525 ureg_MAD(ureg, oPos, _WWWW(vs->aVtx), _CONST(3), ureg_src(tmp));
526 }
527
528 if (need_rVtx) {
529 ureg_MUL(ureg, rVtx, _XXXX(vs->aVtx), _CONST(4));
530 ureg_MAD(ureg, rVtx, _YYYY(vs->aVtx), _CONST(5), ureg_src(rVtx));
531 ureg_MAD(ureg, rVtx, _ZZZZ(vs->aVtx), _CONST(6), ureg_src(rVtx));
532 ureg_MAD(ureg, rVtx, _WWWW(vs->aVtx), _CONST(7), ureg_src(rVtx));
533 }
534 if (need_rNrm) {
535 ureg_MUL(ureg, rNrm, _XXXX(vs->aNrm), _CONST(16));
536 ureg_MAD(ureg, rNrm, _YYYY(vs->aNrm), _CONST(17), ureg_src(rNrm));
537 ureg_MAD(ureg, rNrm, _ZZZZ(vs->aNrm), _CONST(18), ureg_src(rNrm));
538 ureg_normalize3(ureg, rNrm, ureg_src(rNrm), tmp);
539 }
540 /* NOTE: don't use vs->aVtx, vs->aNrm after this line */
541
542 /* === Process point size:
543 */
544 if (key->vertexpointsize) {
545 struct ureg_src cPsz1 = ureg_DECL_constant(ureg, 26);
546 ureg_MAX(ureg, tmp_x, _XXXX(vs->aPsz), _XXXX(cPsz1));
547 ureg_MIN(ureg, oPsz, _X(tmp), _YYYY(cPsz1));
548 } else if (key->pointscale) {
549 struct ureg_src cPsz1 = ureg_DECL_constant(ureg, 26);
550 struct ureg_src cPsz2 = ureg_DECL_constant(ureg, 27);
551
552 ureg_DP3(ureg, tmp_x, ureg_src(r[1]), ureg_src(r[1]));
553 ureg_RSQ(ureg, tmp_y, _X(tmp));
554 ureg_MUL(ureg, tmp_y, _Y(tmp), _X(tmp));
555 ureg_CMP(ureg, tmp_y, ureg_negate(_Y(tmp)), _Y(tmp), ureg_imm1f(ureg, 0.0f));
556 ureg_MAD(ureg, tmp_x, _Y(tmp), _YYYY(cPsz2), _XXXX(cPsz2));
557 ureg_MAD(ureg, tmp_x, _Y(tmp), _X(tmp), _WWWW(cPsz1));
558 ureg_RCP(ureg, tmp_x, ureg_src(tmp));
559 ureg_MUL(ureg, tmp_x, ureg_src(tmp), _ZZZZ(cPsz1));
560 ureg_MAX(ureg, tmp_x, _X(tmp), _XXXX(cPsz1));
561 ureg_MIN(ureg, oPsz, _X(tmp), _YYYY(cPsz1));
562 }
563
564 for (i = 0; i < 8; ++i) {
565 struct ureg_dst oTex, input_coord, transformed, t;
566 unsigned c, writemask;
567 const unsigned tci = (key->tc_gen >> (i * 3)) & 0x7;
568 const unsigned idx = (key->tc_idx >> (i * 3)) & 0x7;
569 unsigned dim_input = 1 + ((key->tc_dim_input >> (i * 2)) & 0x3);
570 const unsigned dim_output = (key->tc_dim_output >> (i * 3)) & 0x7;
571
572 /* No texture output of index s */
573 if (tci == NINED3DTSS_TCI_DISABLE)
574 continue;
575 oTex = ureg_DECL_output(ureg, texcoord_sn, i);
576 input_coord = r[5];
577 transformed = r[6];
578
579 /* Get the coordinate */
580 switch (tci) {
581 case NINED3DTSS_TCI_PASSTHRU:
582 /* NINED3DTSS_TCI_PASSTHRU => Use texcoord coming from index idx *
583 * Else the idx is used only to determine wrapping mode. */
584 vs->aTex[idx] = build_vs_add_input(vs, NINE_DECLUSAGE_i(TEXCOORD,idx));
585 ureg_MOV(ureg, input_coord, vs->aTex[idx]);
586 break;
587 case NINED3DTSS_TCI_CAMERASPACENORMAL:
588 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XYZ), ureg_src(rNrm));
589 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
590 dim_input = 4;
591 break;
592 case NINED3DTSS_TCI_CAMERASPACEPOSITION:
593 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XYZ), ureg_src(rVtx));
594 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
595 dim_input = 4;
596 break;
597 case NINED3DTSS_TCI_CAMERASPACEREFLECTIONVECTOR:
598 tmp.WriteMask = TGSI_WRITEMASK_XYZ;
599 ureg_DP3(ureg, tmp_x, ureg_src(rVtx), ureg_src(rNrm));
600 ureg_MUL(ureg, tmp, ureg_src(rNrm), _X(tmp));
601 ureg_ADD(ureg, tmp, ureg_src(tmp), ureg_src(tmp));
602 ureg_SUB(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XYZ), ureg_src(rVtx), ureg_src(tmp));
603 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
604 dim_input = 4;
605 tmp.WriteMask = TGSI_WRITEMASK_XYZW;
606 break;
607 case NINED3DTSS_TCI_SPHEREMAP:
608 assert(!"TODO");
609 break;
610 default:
611 assert(0);
612 break;
613 }
614
615 /* Apply the transformation */
616 /* dim_output == 0 => do not transform the components.
617 * XYZRHW also disables transformation */
618 if (!dim_output || key->position_t) {
619 transformed = input_coord;
620 writemask = TGSI_WRITEMASK_XYZW;
621 } else {
622 for (c = 0; c < dim_output; c++) {
623 t = ureg_writemask(transformed, 1 << c);
624 switch (dim_input) {
625 /* dim_input = 1 2 3: -> we add trailing 1 to input*/
626 case 1: ureg_MAD(ureg, t, _X(input_coord), _XXXX(_CONST(128 + i * 4 + c)), _YYYY(_CONST(128 + i * 4 + c)));
627 break;
628 case 2: ureg_DP2(ureg, t, ureg_src(input_coord), _CONST(128 + i * 4 + c));
629 ureg_ADD(ureg, t, ureg_src(transformed), _ZZZZ(_CONST(128 + i * 4 + c)));
630 break;
631 case 3: ureg_DP3(ureg, t, ureg_src(input_coord), _CONST(128 + i * 4 + c));
632 ureg_ADD(ureg, t, ureg_src(transformed), _WWWW(_CONST(128 + i * 4 + c)));
633 break;
634 case 4: ureg_DP4(ureg, t, ureg_src(input_coord), _CONST(128 + i * 4 + c)); break;
635 default:
636 assert(0);
637 }
638 }
639 writemask = (1 << dim_output) - 1;
640 }
641
642 ureg_MOV(ureg, ureg_writemask(oTex, writemask), ureg_src(transformed));
643 }
644
645 /* === Lighting:
646 *
647 * DIRECTIONAL: Light at infinite distance, parallel rays, no attenuation.
648 * POINT: Finite distance to scene, divergent rays, isotropic, attenuation.
649 * SPOT: Finite distance, divergent rays, angular dependence, attenuation.
650 *
651 * vec3 normal = normalize(in.Normal * NormalMatrix);
652 * vec3 hitDir = light.direction;
653 * float atten = 1.0;
654 *
655 * if (light.type != DIRECTIONAL)
656 * {
657 * vec3 hitVec = light.position - eyeVertex;
658 * float d = length(hitVec);
659 * hitDir = hitVec / d;
660 * atten = 1 / ((light.atten2 * d + light.atten1) * d + light.atten0);
661 * }
662 *
663 * if (light.type == SPOTLIGHT)
664 * {
665 * float rho = dp3(-hitVec, light.direction);
666 * if (rho < cos(light.phi / 2))
667 * atten = 0;
668 * if (rho < cos(light.theta / 2))
669 * atten *= pow(some_func(rho), light.falloff);
670 * }
671 *
672 * float nDotHit = dp3_sat(normal, hitVec);
673 * float powFact = 0.0;
674 *
675 * if (nDotHit > 0.0)
676 * {
677 * vec3 midVec = normalize(hitDir + eye);
678 * float nDotMid = dp3_sat(normal, midVec);
679 * pFact = pow(nDotMid, material.power);
680 * }
681 *
682 * ambient += light.ambient * atten;
683 * diffuse += light.diffuse * atten * nDotHit;
684 * specular += light.specular * atten * powFact;
685 */
686 if (key->lighting) {
687 struct ureg_dst rAtt = ureg_writemask(r[1], TGSI_WRITEMASK_W);
688 struct ureg_dst rHit = ureg_writemask(r[3], TGSI_WRITEMASK_XYZ);
689 struct ureg_dst rMid = ureg_writemask(r[4], TGSI_WRITEMASK_XYZ);
690
691 struct ureg_dst rCtr = ureg_writemask(r[2], TGSI_WRITEMASK_W);
692
693 struct ureg_dst AL = ureg_writemask(AR, TGSI_WRITEMASK_X);
694
695 /* Light.*.Alpha is not used. */
696 struct ureg_dst rD = ureg_writemask(r[5], TGSI_WRITEMASK_XYZ);
697 struct ureg_dst rA = ureg_writemask(r[6], TGSI_WRITEMASK_XYZ);
698 struct ureg_dst rS = ureg_writemask(r[7], TGSI_WRITEMASK_XYZ);
699
700 struct ureg_src mtlP = _XXXX(MATERIAL_CONST(4));
701
702 struct ureg_src cLKind = _XXXX(LIGHT_CONST(0));
703 struct ureg_src cLAtt0 = _YYYY(LIGHT_CONST(0));
704 struct ureg_src cLAtt1 = _ZZZZ(LIGHT_CONST(0));
705 struct ureg_src cLAtt2 = _WWWW(LIGHT_CONST(0));
706 struct ureg_src cLColD = _XYZW(LIGHT_CONST(1));
707 struct ureg_src cLColS = _XYZW(LIGHT_CONST(2));
708 struct ureg_src cLColA = _XYZW(LIGHT_CONST(3));
709 struct ureg_src cLPos = _XYZW(LIGHT_CONST(4));
710 struct ureg_src cLRng = _WWWW(LIGHT_CONST(4));
711 struct ureg_src cLDir = _XYZW(LIGHT_CONST(5));
712 struct ureg_src cLFOff = _WWWW(LIGHT_CONST(5));
713 struct ureg_src cLTht = _XXXX(LIGHT_CONST(6));
714 struct ureg_src cLPhi = _YYYY(LIGHT_CONST(6));
715 struct ureg_src cLSDiv = _ZZZZ(LIGHT_CONST(6));
716 struct ureg_src cLLast = _WWWW(LIGHT_CONST(7));
717
718 const unsigned loop_label = l++;
719
720 ureg_MOV(ureg, rCtr, ureg_imm1f(ureg, 32.0f)); /* &lightconst(0) */
721 ureg_MOV(ureg, rD, ureg_imm1f(ureg, 0.0f));
722 ureg_MOV(ureg, rA, ureg_imm1f(ureg, 0.0f));
723 ureg_MOV(ureg, rS, ureg_imm1f(ureg, 0.0f));
724 rD = ureg_saturate(rD);
725 rA = ureg_saturate(rA);
726 rS = ureg_saturate(rS);
727
728
729 /* loop management */
730 ureg_BGNLOOP(ureg, &label[loop_label]);
731 ureg_ARL(ureg, AL, _W(rCtr));
732
733 /* if (not DIRECTIONAL light): */
734 ureg_SNE(ureg, tmp_x, cLKind, ureg_imm1f(ureg, D3DLIGHT_DIRECTIONAL));
735 ureg_MOV(ureg, rHit, ureg_negate(cLDir));
736 ureg_MOV(ureg, rAtt, ureg_imm1f(ureg, 1.0f));
737 ureg_IF(ureg, _X(tmp), &label[l++]);
738 {
739 /* hitDir = light.position - eyeVtx
740 * d = length(hitDir)
741 */
742 ureg_SUB(ureg, rHit, cLPos, ureg_src(rVtx));
743 ureg_DP3(ureg, tmp_x, ureg_src(rHit), ureg_src(rHit));
744 ureg_RSQ(ureg, tmp_y, _X(tmp));
745 ureg_MUL(ureg, tmp_x, _X(tmp), _Y(tmp)); /* length */
746
747 /* att = 1.0 / (light.att0 + (light.att1 + light.att2 * d) * d) */
748 ureg_MAD(ureg, rAtt, _X(tmp), cLAtt2, cLAtt1);
749 ureg_MAD(ureg, rAtt, _X(tmp), _W(rAtt), cLAtt0);
750 ureg_RCP(ureg, rAtt, _W(rAtt));
751 /* cut-off if distance exceeds Light.Range */
752 ureg_SLT(ureg, tmp_x, _X(tmp), cLRng);
753 ureg_MUL(ureg, rAtt, _W(rAtt), _X(tmp));
754 }
755 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
756 ureg_ENDIF(ureg);
757
758 /* normalize hitDir */
759 ureg_normalize3(ureg, rHit, ureg_src(rHit), tmp);
760
761 /* if (SPOT light) */
762 ureg_SEQ(ureg, tmp_x, cLKind, ureg_imm1f(ureg, D3DLIGHT_SPOT));
763 ureg_IF(ureg, _X(tmp), &label[l++]);
764 {
765 /* rho = dp3(-hitDir, light.spotDir)
766 *
767 * if (rho > light.ctht2) NOTE: 0 <= phi <= pi, 0 <= theta <= phi
768 * spotAtt = 1
769 * else
770 * if (rho <= light.cphi2)
771 * spotAtt = 0
772 * else
773 * spotAtt = (rho - light.cphi2) / (light.ctht2 - light.cphi2) ^ light.falloff
774 */
775 ureg_DP3(ureg, tmp_y, ureg_negate(ureg_src(rHit)), cLDir); /* rho */
776 ureg_SUB(ureg, tmp_x, _Y(tmp), cLPhi);
777 ureg_MUL(ureg, tmp_x, _X(tmp), cLSDiv);
778 ureg_POW(ureg, tmp_x, _X(tmp), cLFOff); /* spotAtten */
779 ureg_SGE(ureg, tmp_z, _Y(tmp), cLTht); /* if inside theta && phi */
780 ureg_SGE(ureg, tmp_y, _Y(tmp), cLPhi); /* if inside phi */
781 ureg_MAD(ureg, ureg_saturate(tmp_x), _X(tmp), _Y(tmp), _Z(tmp));
782 ureg_MUL(ureg, rAtt, _W(rAtt), _X(tmp));
783 }
784 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
785 ureg_ENDIF(ureg);
786
787 /* directional factors, let's not use LIT because of clarity */
788 ureg_DP3(ureg, ureg_saturate(tmp_x), ureg_src(rNrm), ureg_src(rHit));
789 ureg_MOV(ureg, tmp_y, ureg_imm1f(ureg, 0.0f));
790 ureg_IF(ureg, _X(tmp), &label[l++]);
791 {
792 /* midVec = normalize(hitDir + eyeDir) */
793 if (key->localviewer) {
794 ureg_normalize3(ureg, rMid, ureg_src(rVtx), tmp);
795 ureg_SUB(ureg, rMid, ureg_src(rHit), ureg_src(rMid));
796 } else {
797 ureg_SUB(ureg, rMid, ureg_src(rHit), ureg_imm3f(ureg, 0.0f, 0.0f, 1.0f));
798 }
799 ureg_normalize3(ureg, rMid, ureg_src(rMid), tmp);
800 ureg_DP3(ureg, ureg_saturate(tmp_y), ureg_src(rNrm), ureg_src(rMid));
801 ureg_POW(ureg, tmp_y, _Y(tmp), mtlP);
802
803 ureg_MUL(ureg, tmp_x, _W(rAtt), _X(tmp)); /* dp3(normal,hitDir) * att */
804 ureg_MUL(ureg, tmp_y, _W(rAtt), _Y(tmp)); /* power factor * att */
805 ureg_MAD(ureg, rD, cLColD, _X(tmp), ureg_src(rD)); /* accumulate diffuse */
806 ureg_MAD(ureg, rS, cLColS, _Y(tmp), ureg_src(rS)); /* accumulate specular */
807 }
808 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
809 ureg_ENDIF(ureg);
810
811 ureg_MAD(ureg, rA, cLColA, _W(rAtt), ureg_src(rA)); /* accumulate ambient */
812
813 /* break if this was the last light */
814 ureg_IF(ureg, cLLast, &label[l++]);
815 ureg_BRK(ureg);
816 ureg_ENDIF(ureg);
817 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
818
819 ureg_ADD(ureg, rCtr, _W(rCtr), ureg_imm1f(ureg, 8.0f));
820 ureg_fixup_label(ureg, label[loop_label], ureg_get_instruction_number(ureg));
821 ureg_ENDLOOP(ureg, &label[loop_label]);
822
823 /* Set alpha factors of illumination to 1.0 for the multiplications. */
824 rD.WriteMask = TGSI_WRITEMASK_W; rD.Saturate = 0;
825 rS.WriteMask = TGSI_WRITEMASK_W; rS.Saturate = 0;
826 rA.WriteMask = TGSI_WRITEMASK_W; rA.Saturate = 0;
827 ureg_MOV(ureg, rD, ureg_imm1f(ureg, 1.0f));
828 ureg_MOV(ureg, rS, ureg_imm1f(ureg, 1.0f));
829
830 /* Apply to material:
831 *
832 * oCol[0] = (material.emissive + material.ambient * rs.ambient) +
833 * material.ambient * ambient +
834 * material.diffuse * diffuse +
835 * oCol[1] = material.specular * specular;
836 */
837 if (key->mtl_emissive == 0 && key->mtl_ambient == 0) {
838 ureg_MOV(ureg, rA, ureg_imm1f(ureg, 1.0f));
839 ureg_MAD(ureg, tmp, ureg_src(rA), vs->mtlA, _CONST(19));
840 } else {
841 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(rA), _CONST(25));
842 ureg_MAD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), vs->mtlA, ureg_src(tmp), vs->mtlE);
843 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_W ), vs->mtlA, vs->mtlE);
844 }
845
846 if (key->specular_enable) {
847 /* add oCol[1] to oCol[0] */
848 ureg_MAD(ureg, tmp, ureg_src(rD), vs->mtlD, ureg_src(tmp));
849 ureg_MAD(ureg, oCol[0], ureg_src(rS), vs->mtlS, ureg_src(tmp));
850 } else {
851 ureg_MAD(ureg, oCol[0], ureg_src(rD), vs->mtlD, ureg_src(tmp));
852 }
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 && !(key->passthrough & (1 << NINE_DECLUSAGE_FOG))) {
903 ureg_MOV(ureg, oFog, ureg_scalar(vs->aCol[1], TGSI_SWIZZLE_W));
904 }
905
906 if (key->passthrough & (1 << NINE_DECLUSAGE_BLENDWEIGHT)) {
907 struct ureg_src input;
908 struct ureg_dst output;
909 input = vs->aWgt;
910 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 18);
911 ureg_MOV(ureg, output, input);
912 }
913 if (key->passthrough & (1 << NINE_DECLUSAGE_BLENDINDICES)) {
914 struct ureg_src input;
915 struct ureg_dst output;
916 input = vs->aInd;
917 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 19);
918 ureg_MOV(ureg, output, input);
919 }
920 if (key->passthrough & (1 << NINE_DECLUSAGE_NORMAL)) {
921 struct ureg_src input;
922 struct ureg_dst output;
923 input = vs->aNrm;
924 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 20);
925 ureg_MOV(ureg, output, input);
926 }
927 if (key->passthrough & (1 << NINE_DECLUSAGE_TANGENT)) {
928 struct ureg_src input;
929 struct ureg_dst output;
930 input = build_vs_add_input(vs, NINE_DECLUSAGE_TANGENT);
931 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 21);
932 ureg_MOV(ureg, output, input);
933 }
934 if (key->passthrough & (1 << NINE_DECLUSAGE_BINORMAL)) {
935 struct ureg_src input;
936 struct ureg_dst output;
937 input = build_vs_add_input(vs, NINE_DECLUSAGE_BINORMAL);
938 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 22);
939 ureg_MOV(ureg, output, input);
940 }
941 if (key->passthrough & (1 << NINE_DECLUSAGE_FOG)) {
942 struct ureg_src input;
943 struct ureg_dst output;
944 input = build_vs_add_input(vs, NINE_DECLUSAGE_FOG);
945 input = ureg_scalar(input, TGSI_SWIZZLE_X);
946 output = oFog;
947 ureg_MOV(ureg, output, input);
948 }
949 if (key->passthrough & (1 << NINE_DECLUSAGE_DEPTH)) {
950 (void) 0; /* TODO: replace z of position output ? */
951 }
952
953
954 if (key->position_t && device->driver_caps.window_space_position_support)
955 ureg_property(ureg, TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION, TRUE);
956
957 ureg_END(ureg);
958 nine_ureg_tgsi_dump(ureg, FALSE);
959 return ureg_create_shader_and_destroy(ureg, device->pipe);
960 }
961
962 /* PS FF constants layout:
963 *
964 * CONST[ 0.. 7] stage[i].D3DTSS_CONSTANT
965 * CONST[ 8..15].x___ stage[i].D3DTSS_BUMPENVMAT00
966 * CONST[ 8..15]._y__ stage[i].D3DTSS_BUMPENVMAT01
967 * CONST[ 8..15].__z_ stage[i].D3DTSS_BUMPENVMAT10
968 * CONST[ 8..15].___w stage[i].D3DTSS_BUMPENVMAT11
969 * CONST[16..19].x_z_ stage[i].D3DTSS_BUMPENVLSCALE
970 * CONST[17..19]._y_w stage[i].D3DTSS_BUMPENVLOFFSET
971 *
972 * CONST[20] D3DRS_TEXTUREFACTOR
973 * CONST[21] D3DRS_FOGCOLOR
974 * CONST[22].x___ RS.FogEnd
975 * CONST[22]._y__ 1.0f / (RS.FogEnd - RS.FogStart)
976 * CONST[22].__z_ RS.FogDensity
977 */
978 struct ps_build_ctx
979 {
980 struct ureg_program *ureg;
981
982 struct ureg_src vC[2]; /* DIFFUSE, SPECULAR */
983 struct ureg_src vT[8]; /* TEXCOORD[i] */
984 struct ureg_dst r[6]; /* TEMPs */
985 struct ureg_dst rCur; /* D3DTA_CURRENT */
986 struct ureg_dst rMod;
987 struct ureg_src rCurSrc;
988 struct ureg_dst rTmp; /* D3DTA_TEMP */
989 struct ureg_src rTmpSrc;
990 struct ureg_dst rTex;
991 struct ureg_src rTexSrc;
992 struct ureg_src cBEM[8];
993 struct ureg_src s[8];
994
995 struct {
996 unsigned index;
997 unsigned index_pre_mod;
998 unsigned num_regs;
999 } stage;
1000 };
1001
1002 static struct ureg_src
1003 ps_get_ts_arg(struct ps_build_ctx *ps, unsigned ta)
1004 {
1005 struct ureg_src reg;
1006
1007 switch (ta & D3DTA_SELECTMASK) {
1008 case D3DTA_CONSTANT:
1009 reg = ureg_DECL_constant(ps->ureg, ps->stage.index);
1010 break;
1011 case D3DTA_CURRENT:
1012 reg = (ps->stage.index == ps->stage.index_pre_mod) ? ureg_src(ps->rMod) : ps->rCurSrc;
1013 break;
1014 case D3DTA_DIFFUSE:
1015 reg = ureg_DECL_fs_input(ps->ureg, TGSI_SEMANTIC_COLOR, 0, TGSI_INTERPOLATE_COLOR);
1016 break;
1017 case D3DTA_SPECULAR:
1018 reg = ureg_DECL_fs_input(ps->ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_COLOR);
1019 break;
1020 case D3DTA_TEMP:
1021 reg = ps->rTmpSrc;
1022 break;
1023 case D3DTA_TEXTURE:
1024 reg = ps->rTexSrc;
1025 break;
1026 case D3DTA_TFACTOR:
1027 reg = ureg_DECL_constant(ps->ureg, 20);
1028 break;
1029 default:
1030 assert(0);
1031 reg = ureg_src_undef();
1032 break;
1033 }
1034 if (ta & D3DTA_COMPLEMENT) {
1035 struct ureg_dst dst = ps->r[ps->stage.num_regs++];
1036 ureg_SUB(ps->ureg, dst, ureg_imm1f(ps->ureg, 1.0f), reg);
1037 reg = ureg_src(dst);
1038 }
1039 if (ta & D3DTA_ALPHAREPLICATE)
1040 reg = _WWWW(reg);
1041 return reg;
1042 }
1043
1044 static struct ureg_dst
1045 ps_get_ts_dst(struct ps_build_ctx *ps, unsigned ta)
1046 {
1047 assert(!(ta & (D3DTA_COMPLEMENT | D3DTA_ALPHAREPLICATE)));
1048
1049 switch (ta & D3DTA_SELECTMASK) {
1050 case D3DTA_CURRENT:
1051 return ps->rCur;
1052 case D3DTA_TEMP:
1053 return ps->rTmp;
1054 default:
1055 assert(0);
1056 return ureg_dst_undef();
1057 }
1058 }
1059
1060 static uint8_t ps_d3dtop_args_mask(D3DTEXTUREOP top)
1061 {
1062 switch (top) {
1063 case D3DTOP_DISABLE:
1064 return 0x0;
1065 case D3DTOP_SELECTARG1:
1066 case D3DTOP_PREMODULATE:
1067 return 0x2;
1068 case D3DTOP_SELECTARG2:
1069 return 0x4;
1070 case D3DTOP_MULTIPLYADD:
1071 case D3DTOP_LERP:
1072 return 0x7;
1073 default:
1074 return 0x6;
1075 }
1076 }
1077
1078 static inline boolean
1079 is_MOV_no_op(struct ureg_dst dst, struct ureg_src src)
1080 {
1081 return !dst.WriteMask ||
1082 (dst.File == src.File &&
1083 dst.Index == src.Index &&
1084 !dst.Indirect &&
1085 !dst.Saturate &&
1086 !src.Indirect &&
1087 !src.Negate &&
1088 !src.Absolute &&
1089 (!(dst.WriteMask & TGSI_WRITEMASK_X) || (src.SwizzleX == TGSI_SWIZZLE_X)) &&
1090 (!(dst.WriteMask & TGSI_WRITEMASK_Y) || (src.SwizzleY == TGSI_SWIZZLE_Y)) &&
1091 (!(dst.WriteMask & TGSI_WRITEMASK_Z) || (src.SwizzleZ == TGSI_SWIZZLE_Z)) &&
1092 (!(dst.WriteMask & TGSI_WRITEMASK_W) || (src.SwizzleW == TGSI_SWIZZLE_W)));
1093
1094 }
1095
1096 static void
1097 ps_do_ts_op(struct ps_build_ctx *ps, unsigned top, struct ureg_dst dst, struct ureg_src *arg)
1098 {
1099 struct ureg_program *ureg = ps->ureg;
1100 struct ureg_dst tmp = ps->r[ps->stage.num_regs];
1101 struct ureg_dst tmp2 = ps->r[ps->stage.num_regs+1];
1102 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
1103
1104 tmp.WriteMask = dst.WriteMask;
1105
1106 if (top != D3DTOP_SELECTARG1 && top != D3DTOP_SELECTARG2 &&
1107 top != D3DTOP_MODULATE && top != D3DTOP_PREMODULATE &&
1108 top != D3DTOP_BLENDDIFFUSEALPHA && top != D3DTOP_BLENDTEXTUREALPHA &&
1109 top != D3DTOP_BLENDFACTORALPHA && top != D3DTOP_BLENDCURRENTALPHA &&
1110 top != D3DTOP_BUMPENVMAP && top != D3DTOP_BUMPENVMAPLUMINANCE &&
1111 top != D3DTOP_LERP)
1112 dst = ureg_saturate(dst);
1113
1114 switch (top) {
1115 case D3DTOP_SELECTARG1:
1116 if (!is_MOV_no_op(dst, arg[1]))
1117 ureg_MOV(ureg, dst, arg[1]);
1118 break;
1119 case D3DTOP_SELECTARG2:
1120 if (!is_MOV_no_op(dst, arg[2]))
1121 ureg_MOV(ureg, dst, arg[2]);
1122 break;
1123 case D3DTOP_MODULATE:
1124 ureg_MUL(ureg, dst, arg[1], arg[2]);
1125 break;
1126 case D3DTOP_MODULATE2X:
1127 ureg_MUL(ureg, tmp, arg[1], arg[2]);
1128 ureg_ADD(ureg, dst, ureg_src(tmp), ureg_src(tmp));
1129 break;
1130 case D3DTOP_MODULATE4X:
1131 ureg_MUL(ureg, tmp, arg[1], arg[2]);
1132 ureg_MUL(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 4.0f));
1133 break;
1134 case D3DTOP_ADD:
1135 ureg_ADD(ureg, dst, arg[1], arg[2]);
1136 break;
1137 case D3DTOP_ADDSIGNED:
1138 ureg_ADD(ureg, tmp, arg[1], arg[2]);
1139 ureg_SUB(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 0.5f));
1140 break;
1141 case D3DTOP_ADDSIGNED2X:
1142 ureg_ADD(ureg, tmp, arg[1], arg[2]);
1143 ureg_MAD(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 2.0f), ureg_imm1f(ureg, -1.0f));
1144 break;
1145 case D3DTOP_SUBTRACT:
1146 ureg_SUB(ureg, dst, arg[1], arg[2]);
1147 break;
1148 case D3DTOP_ADDSMOOTH:
1149 ureg_SUB(ureg, tmp, ureg_imm1f(ureg, 1.0f), arg[1]);
1150 ureg_MAD(ureg, dst, ureg_src(tmp), arg[2], arg[1]);
1151 break;
1152 case D3DTOP_BLENDDIFFUSEALPHA:
1153 ureg_LRP(ureg, dst, _WWWW(ps->vC[0]), arg[1], arg[2]);
1154 break;
1155 case D3DTOP_BLENDTEXTUREALPHA:
1156 /* XXX: alpha taken from previous stage, texture or result ? */
1157 ureg_LRP(ureg, dst, _W(ps->rTex), arg[1], arg[2]);
1158 break;
1159 case D3DTOP_BLENDFACTORALPHA:
1160 ureg_LRP(ureg, dst, _WWWW(_CONST(20)), arg[1], arg[2]);
1161 break;
1162 case D3DTOP_BLENDTEXTUREALPHAPM:
1163 ureg_SUB(ureg, tmp_x, ureg_imm1f(ureg, 1.0f), _W(ps->rTex));
1164 ureg_MAD(ureg, dst, arg[2], _X(tmp), arg[1]);
1165 break;
1166 case D3DTOP_BLENDCURRENTALPHA:
1167 ureg_LRP(ureg, dst, _WWWW(ps->rCurSrc), arg[1], arg[2]);
1168 break;
1169 case D3DTOP_PREMODULATE:
1170 ureg_MOV(ureg, dst, arg[1]);
1171 ps->stage.index_pre_mod = ps->stage.index + 1;
1172 break;
1173 case D3DTOP_MODULATEALPHA_ADDCOLOR:
1174 ureg_MAD(ureg, dst, _WWWW(arg[1]), arg[2], arg[1]);
1175 break;
1176 case D3DTOP_MODULATECOLOR_ADDALPHA:
1177 ureg_MAD(ureg, dst, arg[1], arg[2], _WWWW(arg[1]));
1178 break;
1179 case D3DTOP_MODULATEINVALPHA_ADDCOLOR:
1180 ureg_SUB(ureg, tmp_x, ureg_imm1f(ureg, 1.0f), _WWWW(arg[1]));
1181 ureg_MAD(ureg, dst, _X(tmp), arg[2], arg[1]);
1182 break;
1183 case D3DTOP_MODULATEINVCOLOR_ADDALPHA:
1184 ureg_SUB(ureg, tmp, ureg_imm1f(ureg, 1.0f), arg[1]);
1185 ureg_MAD(ureg, dst, ureg_src(tmp), arg[2], _WWWW(arg[1]));
1186 break;
1187 case D3DTOP_BUMPENVMAP:
1188 break;
1189 case D3DTOP_BUMPENVMAPLUMINANCE:
1190 break;
1191 case D3DTOP_DOTPRODUCT3:
1192 ureg_SUB(ureg, tmp, arg[1], ureg_imm4f(ureg,0.5,0.5,0.5,0.5));
1193 ureg_SUB(ureg, tmp2, arg[2] , ureg_imm4f(ureg,0.5,0.5,0.5,0.5));
1194 ureg_DP3(ureg, tmp, ureg_src(tmp), ureg_src(tmp2));
1195 ureg_MUL(ureg, ureg_saturate(dst), ureg_src(tmp), ureg_imm4f(ureg,4.0,4.0,4.0,4.0));
1196 break;
1197 case D3DTOP_MULTIPLYADD:
1198 ureg_MAD(ureg, dst, arg[1], arg[2], arg[0]);
1199 break;
1200 case D3DTOP_LERP:
1201 ureg_LRP(ureg, dst, arg[0], arg[1], arg[2]);
1202 break;
1203 case D3DTOP_DISABLE:
1204 /* no-op ? */
1205 break;
1206 default:
1207 assert(!"invalid D3DTOP");
1208 break;
1209 }
1210 }
1211
1212 static void *
1213 nine_ff_build_ps(struct NineDevice9 *device, struct nine_ff_ps_key *key)
1214 {
1215 struct ps_build_ctx ps;
1216 struct ureg_program *ureg = ureg_create(PIPE_SHADER_FRAGMENT);
1217 struct ureg_dst oCol;
1218 unsigned i, s;
1219 const unsigned texcoord_sn = get_texcoord_sn(device->screen);
1220
1221 memset(&ps, 0, sizeof(ps));
1222 ps.ureg = ureg;
1223 ps.stage.index_pre_mod = -1;
1224
1225 ps.vC[0] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 0, TGSI_INTERPOLATE_COLOR);
1226
1227 /* Declare all TEMPs we might need, serious drivers have a register allocator. */
1228 for (i = 0; i < Elements(ps.r); ++i)
1229 ps.r[i] = ureg_DECL_local_temporary(ureg);
1230 ps.rCur = ps.r[0];
1231 ps.rTmp = ps.r[1];
1232 ps.rTex = ps.r[2];
1233 ps.rCurSrc = ureg_src(ps.rCur);
1234 ps.rTmpSrc = ureg_src(ps.rTmp);
1235 ps.rTexSrc = ureg_src(ps.rTex);
1236
1237 for (s = 0; s < 8; ++s) {
1238 ps.s[s] = ureg_src_undef();
1239
1240 if (key->ts[s].colorop != D3DTOP_DISABLE) {
1241 if (key->ts[s].colorarg0 == D3DTA_SPECULAR ||
1242 key->ts[s].colorarg1 == D3DTA_SPECULAR ||
1243 key->ts[s].colorarg2 == D3DTA_SPECULAR)
1244 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_COLOR);
1245
1246 if (key->ts[s].colorarg0 == D3DTA_TEXTURE ||
1247 key->ts[s].colorarg1 == D3DTA_TEXTURE ||
1248 key->ts[s].colorarg2 == D3DTA_TEXTURE) {
1249 ps.s[s] = ureg_DECL_sampler(ureg, s);
1250 ps.vT[s] = ureg_DECL_fs_input(ureg, texcoord_sn, s, TGSI_INTERPOLATE_PERSPECTIVE);
1251 }
1252 if (s && (key->ts[s - 1].colorop == D3DTOP_PREMODULATE ||
1253 key->ts[s - 1].alphaop == D3DTOP_PREMODULATE))
1254 ps.s[s] = ureg_DECL_sampler(ureg, s);
1255 }
1256
1257 if (key->ts[s].alphaop != D3DTOP_DISABLE) {
1258 if (key->ts[s].alphaarg0 == D3DTA_SPECULAR ||
1259 key->ts[s].alphaarg1 == D3DTA_SPECULAR ||
1260 key->ts[s].alphaarg2 == D3DTA_SPECULAR)
1261 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_COLOR);
1262
1263 if (key->ts[s].alphaarg0 == D3DTA_TEXTURE ||
1264 key->ts[s].alphaarg1 == D3DTA_TEXTURE ||
1265 key->ts[s].alphaarg2 == D3DTA_TEXTURE) {
1266 ps.s[s] = ureg_DECL_sampler(ureg, s);
1267 ps.vT[s] = ureg_DECL_fs_input(ureg, texcoord_sn, s, TGSI_INTERPOLATE_PERSPECTIVE);
1268 }
1269 }
1270 }
1271 if (key->specular)
1272 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_COLOR);
1273
1274 oCol = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
1275
1276 if (key->ts[0].colorop == D3DTOP_DISABLE &&
1277 key->ts[0].alphaop == D3DTOP_DISABLE)
1278 ureg_MOV(ureg, ps.rCur, ps.vC[0]);
1279 /* Or is it undefined then ? */
1280
1281 /* Run stages.
1282 */
1283 for (s = 0; s < 8; ++s) {
1284 unsigned colorarg[3];
1285 unsigned alphaarg[3];
1286 const uint8_t used_c = ps_d3dtop_args_mask(key->ts[s].colorop);
1287 const uint8_t used_a = ps_d3dtop_args_mask(key->ts[s].alphaop);
1288 struct ureg_dst dst;
1289 struct ureg_src arg[3];
1290
1291 if (key->ts[s].colorop == D3DTOP_DISABLE &&
1292 key->ts[s].alphaop == D3DTOP_DISABLE)
1293 continue;
1294 ps.stage.index = s;
1295 ps.stage.num_regs = 3;
1296
1297 DBG("STAGE[%u]: colorop=%s alphaop=%s\n", s,
1298 nine_D3DTOP_to_str(key->ts[s].colorop),
1299 nine_D3DTOP_to_str(key->ts[s].alphaop));
1300
1301 if (!ureg_src_is_undef(ps.s[s])) {
1302 unsigned target;
1303 switch (key->ts[s].textarget) {
1304 case 0: target = TGSI_TEXTURE_1D; break;
1305 case 1: target = TGSI_TEXTURE_2D; break;
1306 case 2: target = TGSI_TEXTURE_3D; break;
1307 case 3: target = TGSI_TEXTURE_CUBE; break;
1308 /* this is a 2 bit bitfield, do I really need a default case ? */
1309 }
1310
1311 /* sample the texture */
1312 if (key->ts[s].colorop == D3DTOP_BUMPENVMAP ||
1313 key->ts[s].colorop == D3DTOP_BUMPENVMAPLUMINANCE) {
1314 }
1315 if (key->projected & (3 << (s *2))) {
1316 unsigned dim = 1 + ((key->projected >> (2 * s)) & 3);
1317 if (dim == 4)
1318 ureg_TXP(ureg, ps.rTex, target, ps.vT[s], ps.s[s]);
1319 else {
1320 ureg_RCP(ureg, ureg_writemask(ps.rTmp, TGSI_WRITEMASK_X), ureg_scalar(ps.vT[s], dim-1));
1321 ureg_MUL(ureg, ps.rTmp, _XXXX(ps.rTmpSrc), ps.vT[s]);
1322 ureg_TEX(ureg, ps.rTex, target, ps.rTmpSrc, ps.s[s]);
1323 }
1324 } else {
1325 ureg_TEX(ureg, ps.rTex, target, ps.vT[s], ps.s[s]);
1326 }
1327 }
1328
1329 if (s == 0 &&
1330 (key->ts[0].resultarg != 0 /* not current */ ||
1331 key->ts[0].colorop == D3DTOP_DISABLE ||
1332 key->ts[0].alphaop == D3DTOP_DISABLE ||
1333 key->ts[0].colorop == D3DTOP_BLENDCURRENTALPHA ||
1334 key->ts[0].alphaop == D3DTOP_BLENDCURRENTALPHA ||
1335 key->ts[0].colorarg0 == D3DTA_CURRENT ||
1336 key->ts[0].colorarg1 == D3DTA_CURRENT ||
1337 key->ts[0].colorarg2 == D3DTA_CURRENT ||
1338 key->ts[0].alphaarg0 == D3DTA_CURRENT ||
1339 key->ts[0].alphaarg1 == D3DTA_CURRENT ||
1340 key->ts[0].alphaarg2 == D3DTA_CURRENT)
1341 ) {
1342 /* Initialize D3DTA_CURRENT.
1343 * (Yes we can do this before the loop but not until
1344 * NVE4 has an instruction scheduling pass.)
1345 */
1346 ureg_MOV(ureg, ps.rCur, ps.vC[0]);
1347 }
1348
1349 dst = ps_get_ts_dst(&ps, key->ts[s].resultarg ? D3DTA_TEMP : D3DTA_CURRENT);
1350
1351 if (ps.stage.index_pre_mod == ps.stage.index) {
1352 ps.rMod = ps.r[ps.stage.num_regs++];
1353 ureg_MUL(ureg, ps.rMod, ps.rCurSrc, ps.rTexSrc);
1354 }
1355
1356 colorarg[0] = (key->ts[s].colorarg0 | ((key->colorarg_b4[0] >> s) << 4) | ((key->colorarg_b5[0] >> s) << 5)) & 0x3f;
1357 colorarg[1] = (key->ts[s].colorarg1 | ((key->colorarg_b4[1] >> s) << 4) | ((key->colorarg_b5[1] >> s) << 5)) & 0x3f;
1358 colorarg[2] = (key->ts[s].colorarg2 | ((key->colorarg_b4[2] >> s) << 4) | ((key->colorarg_b5[2] >> s) << 5)) & 0x3f;
1359 alphaarg[0] = (key->ts[s].alphaarg0 | ((key->alphaarg_b4[0] >> s) << 4)) & 0x1f;
1360 alphaarg[1] = (key->ts[s].alphaarg1 | ((key->alphaarg_b4[1] >> s) << 4)) & 0x1f;
1361 alphaarg[2] = (key->ts[s].alphaarg2 | ((key->alphaarg_b4[2] >> s) << 4)) & 0x1f;
1362
1363 if (key->ts[s].colorop != key->ts[s].alphaop ||
1364 colorarg[0] != alphaarg[0] ||
1365 colorarg[1] != alphaarg[1] ||
1366 colorarg[2] != alphaarg[2])
1367 dst.WriteMask = TGSI_WRITEMASK_XYZ;
1368
1369 /* Special DOTPRODUCT behaviour (see wine tests) */
1370 if (key->ts[s].colorop == D3DTOP_DOTPRODUCT3)
1371 dst.WriteMask = TGSI_WRITEMASK_XYZW;
1372
1373 if (used_c & 0x1) arg[0] = ps_get_ts_arg(&ps, colorarg[0]);
1374 if (used_c & 0x2) arg[1] = ps_get_ts_arg(&ps, colorarg[1]);
1375 if (used_c & 0x4) arg[2] = ps_get_ts_arg(&ps, colorarg[2]);
1376 ps_do_ts_op(&ps, key->ts[s].colorop, dst, arg);
1377
1378 if (dst.WriteMask != TGSI_WRITEMASK_XYZW) {
1379 dst.WriteMask = TGSI_WRITEMASK_W;
1380
1381 if (used_a & 0x1) arg[0] = ps_get_ts_arg(&ps, alphaarg[0]);
1382 if (used_a & 0x2) arg[1] = ps_get_ts_arg(&ps, alphaarg[1]);
1383 if (used_a & 0x4) arg[2] = ps_get_ts_arg(&ps, alphaarg[2]);
1384 ps_do_ts_op(&ps, key->ts[s].alphaop, dst, arg);
1385 }
1386 }
1387
1388 if (key->specular)
1389 ureg_ADD(ureg, ps.rCur, ps.rCurSrc, ps.vC[1]);
1390
1391 /* Fog.
1392 */
1393 if (key->fog_mode) {
1394 struct ureg_src vPos;
1395 if (device->screen->get_param(device->screen,
1396 PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL)) {
1397 vPos = ureg_DECL_system_value(ureg, TGSI_SEMANTIC_POSITION, 0);
1398 } else {
1399 vPos = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_POSITION, 0,
1400 TGSI_INTERPOLATE_LINEAR);
1401 }
1402
1403 struct ureg_dst rFog = ureg_writemask(ps.rTmp, TGSI_WRITEMASK_X);
1404 if (key->fog_mode == D3DFOG_EXP) {
1405 ureg_MUL(ureg, rFog, _ZZZZ(vPos), _ZZZZ(_CONST(22)));
1406 ureg_MUL(ureg, rFog, _X(rFog), ureg_imm1f(ureg, -1.442695f));
1407 ureg_EX2(ureg, rFog, _X(rFog));
1408 } else
1409 if (key->fog_mode == D3DFOG_EXP2) {
1410 ureg_MUL(ureg, rFog, _ZZZZ(vPos), _ZZZZ(_CONST(22)));
1411 ureg_MUL(ureg, rFog, _X(rFog), _X(rFog));
1412 ureg_MUL(ureg, rFog, _X(rFog), ureg_imm1f(ureg, -1.442695f));
1413 ureg_EX2(ureg, rFog, _X(rFog));
1414 } else
1415 if (key->fog_mode == D3DFOG_LINEAR) {
1416 ureg_SUB(ureg, rFog, _XXXX(_CONST(22)), _ZZZZ(vPos));
1417 ureg_MUL(ureg, ureg_saturate(rFog), _X(rFog), _YYYY(_CONST(22)));
1418 }
1419 ureg_LRP(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_XYZ), _X(rFog), ps.rCurSrc, _CONST(21));
1420 ureg_MOV(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_W), ps.rCurSrc);
1421 } else
1422 if (key->fog) {
1423 struct ureg_src vFog = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_FOG, 0, TGSI_INTERPOLATE_PERSPECTIVE);
1424 ureg_LRP(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_XYZ), _XXXX(vFog), ps.rCurSrc, _CONST(21));
1425 ureg_MOV(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_W), ps.rCurSrc);
1426 } else {
1427 ureg_MOV(ureg, oCol, ps.rCurSrc);
1428 }
1429
1430 ureg_END(ureg);
1431 nine_ureg_tgsi_dump(ureg, FALSE);
1432 return ureg_create_shader_and_destroy(ureg, device->pipe);
1433 }
1434
1435 static struct NineVertexShader9 *
1436 nine_ff_get_vs(struct NineDevice9 *device)
1437 {
1438 const struct nine_state *state = &device->state;
1439 struct NineVertexShader9 *vs;
1440 enum pipe_error err;
1441 struct vs_build_ctx bld;
1442 struct nine_ff_vs_key key;
1443 unsigned s, i;
1444 char input_texture_coord[8];
1445
1446 assert(sizeof(key) <= sizeof(key.value32));
1447
1448 memset(&key, 0, sizeof(key));
1449 memset(&bld, 0, sizeof(bld));
1450 memset(&input_texture_coord, 0, sizeof(input_texture_coord));
1451
1452 bld.key = &key;
1453
1454 /* FIXME: this shouldn't be NULL, but it is on init */
1455 if (state->vdecl) {
1456 key.color0in_one = 1;
1457 key.color1in_one = 1;
1458 for (i = 0; i < state->vdecl->nelems; i++) {
1459 uint16_t usage = state->vdecl->usage_map[i];
1460 if (usage == NINE_DECLUSAGE_POSITIONT)
1461 key.position_t = 1;
1462 else if (usage == NINE_DECLUSAGE_i(COLOR, 0))
1463 key.color0in_one = 0;
1464 else if (usage == NINE_DECLUSAGE_i(COLOR, 1))
1465 key.color1in_one = 0;
1466 else if (usage == NINE_DECLUSAGE_PSIZE)
1467 key.vertexpointsize = 1;
1468 else if (usage % NINE_DECLUSAGE_COUNT == NINE_DECLUSAGE_TEXCOORD) {
1469 s = usage / NINE_DECLUSAGE_COUNT;
1470 if (s < 8)
1471 input_texture_coord[s] = nine_decltype_get_dim(state->vdecl->decls[i].Type);
1472 else
1473 DBG("FF given texture coordinate >= 8. Ignoring\n");
1474 } else if (usage < NINE_DECLUSAGE_NONE)
1475 key.passthrough |= 1 << usage;
1476 }
1477 }
1478 /* ff vs + ps 3.0: some elements are passed to the ps (wine test).
1479 * We do restrict to indices 0 */
1480 key.passthrough &= ~((1 << NINE_DECLUSAGE_POSITION) | (1 << NINE_DECLUSAGE_PSIZE) |
1481 (1 << NINE_DECLUSAGE_TEXCOORD) | (1 << NINE_DECLUSAGE_POSITIONT) |
1482 (1 << NINE_DECLUSAGE_TESSFACTOR) | (1 << NINE_DECLUSAGE_SAMPLE));
1483 if (!key.vertexpointsize)
1484 key.pointscale = !!state->rs[D3DRS_POINTSCALEENABLE];
1485
1486 key.lighting = !!state->rs[D3DRS_LIGHTING] && state->ff.num_lights_active;
1487 key.darkness = !!state->rs[D3DRS_LIGHTING] && !state->ff.num_lights_active;
1488 if (key.position_t) {
1489 key.darkness = 0; /* |= key.lighting; */ /* XXX ? */
1490 key.lighting = 0;
1491 }
1492 if ((key.lighting | key.darkness) && state->rs[D3DRS_COLORVERTEX]) {
1493 key.mtl_diffuse = state->rs[D3DRS_DIFFUSEMATERIALSOURCE];
1494 key.mtl_ambient = state->rs[D3DRS_AMBIENTMATERIALSOURCE];
1495 key.mtl_specular = state->rs[D3DRS_SPECULARMATERIALSOURCE];
1496 key.mtl_emissive = state->rs[D3DRS_EMISSIVEMATERIALSOURCE];
1497 }
1498 key.fog = !!state->rs[D3DRS_FOGENABLE];
1499 key.fog_mode = state->rs[D3DRS_FOGENABLE] ? state->rs[D3DRS_FOGVERTEXMODE] : 0;
1500 if (key.fog_mode)
1501 key.fog_range = !key.position_t && state->rs[D3DRS_RANGEFOGENABLE];
1502
1503 key.localviewer = !!state->rs[D3DRS_LOCALVIEWER];
1504 key.specular_enable = !!state->rs[D3DRS_SPECULARENABLE];
1505
1506 if (state->rs[D3DRS_VERTEXBLEND] != D3DVBF_DISABLE) {
1507 key.vertexblend_indexed = !!state->rs[D3DRS_INDEXEDVERTEXBLENDENABLE];
1508
1509 switch (state->rs[D3DRS_VERTEXBLEND]) {
1510 case D3DVBF_0WEIGHTS: key.vertexblend = key.vertexblend_indexed; break;
1511 case D3DVBF_1WEIGHTS: key.vertexblend = 2; break;
1512 case D3DVBF_2WEIGHTS: key.vertexblend = 3; break;
1513 case D3DVBF_3WEIGHTS: key.vertexblend = 4; break;
1514 case D3DVBF_TWEENING: key.vertextween = 1; break;
1515 default:
1516 assert(!"invalid D3DVBF");
1517 break;
1518 }
1519 }
1520
1521 for (s = 0; s < 8; ++s) {
1522 unsigned gen = (state->ff.tex_stage[s][D3DTSS_TEXCOORDINDEX] >> 16) + 1;
1523 unsigned dim;
1524
1525 if (key.position_t && gen > NINED3DTSS_TCI_PASSTHRU)
1526 gen = NINED3DTSS_TCI_PASSTHRU;
1527
1528 if (!input_texture_coord[s] && gen == NINED3DTSS_TCI_PASSTHRU)
1529 gen = NINED3DTSS_TCI_DISABLE;
1530
1531 key.tc_gen |= gen << (s * 3);
1532 key.tc_idx |= (state->ff.tex_stage[s][D3DTSS_TEXCOORDINDEX] & 7) << (s * 3);
1533 key.tc_dim_input |= ((input_texture_coord[s]-1) & 0x3) << (s * 2);
1534
1535 dim = state->ff.tex_stage[s][D3DTSS_TEXTURETRANSFORMFLAGS] & 0x7;
1536 if (dim > 4)
1537 dim = input_texture_coord[s];
1538 if (dim == 1) /* NV behaviour */
1539 dim = 0;
1540 key.tc_dim_output |= dim << (s * 3);
1541 }
1542
1543 vs = util_hash_table_get(device->ff.ht_vs, &key);
1544 if (vs)
1545 return vs;
1546 NineVertexShader9_new(device, &vs, NULL, nine_ff_build_vs(device, &bld));
1547
1548 nine_ff_prune_vs(device);
1549 if (vs) {
1550 unsigned n;
1551
1552 memcpy(&vs->ff_key, &key, sizeof(vs->ff_key));
1553
1554 err = util_hash_table_set(device->ff.ht_vs, &vs->ff_key, vs);
1555 (void)err;
1556 assert(err == PIPE_OK);
1557 device->ff.num_vs++;
1558 NineUnknown_ConvertRefToBind(NineUnknown(vs));
1559
1560 vs->num_inputs = bld.num_inputs;
1561 for (n = 0; n < bld.num_inputs; ++n)
1562 vs->input_map[n].ndecl = bld.input[n];
1563
1564 vs->position_t = key.position_t;
1565 vs->point_size = key.vertexpointsize | key.pointscale;
1566 }
1567 return vs;
1568 }
1569
1570 static struct NinePixelShader9 *
1571 nine_ff_get_ps(struct NineDevice9 *device)
1572 {
1573 struct nine_state *state = &device->state;
1574 struct NinePixelShader9 *ps;
1575 enum pipe_error err;
1576 struct nine_ff_ps_key key;
1577 unsigned s;
1578 uint8_t sampler_mask = 0;
1579
1580 assert(sizeof(key) <= sizeof(key.value32));
1581
1582 memset(&key, 0, sizeof(key));
1583 for (s = 0; s < 8; ++s) {
1584 key.ts[s].colorop = state->ff.tex_stage[s][D3DTSS_COLOROP];
1585 key.ts[s].alphaop = state->ff.tex_stage[s][D3DTSS_ALPHAOP];
1586 /* MSDN says D3DTOP_DISABLE disables this and all subsequent stages. */
1587 /* ALPHAOP cannot be disabled if COLOROP is enabled. */
1588 if (key.ts[s].colorop == D3DTOP_DISABLE) {
1589 key.ts[s].alphaop = D3DTOP_DISABLE; /* DISABLE == 1, avoid degenerate keys */
1590 break;
1591 }
1592
1593 if (!state->texture[s] &&
1594 state->ff.tex_stage[s][D3DTSS_COLORARG1] == D3DTA_TEXTURE) {
1595 /* This should also disable the stage. */
1596 key.ts[s].colorop = key.ts[s].alphaop = D3DTOP_DISABLE;
1597 break;
1598 }
1599
1600 if (state->ff.tex_stage[s][D3DTSS_COLORARG1] == D3DTA_TEXTURE)
1601 sampler_mask |= (1 << s);
1602
1603 if (key.ts[s].colorop != D3DTOP_DISABLE) {
1604 uint8_t used_c = ps_d3dtop_args_mask(key.ts[s].colorop);
1605 if (used_c & 0x1) key.ts[s].colorarg0 = state->ff.tex_stage[s][D3DTSS_COLORARG0];
1606 if (used_c & 0x2) key.ts[s].colorarg1 = state->ff.tex_stage[s][D3DTSS_COLORARG1];
1607 if (used_c & 0x4) key.ts[s].colorarg2 = state->ff.tex_stage[s][D3DTSS_COLORARG2];
1608 if (used_c & 0x1) key.colorarg_b4[0] |= (state->ff.tex_stage[s][D3DTSS_COLORARG0] >> 4) << s;
1609 if (used_c & 0x1) key.colorarg_b5[0] |= (state->ff.tex_stage[s][D3DTSS_COLORARG0] >> 5) << s;
1610 if (used_c & 0x2) key.colorarg_b4[1] |= (state->ff.tex_stage[s][D3DTSS_COLORARG1] >> 4) << s;
1611 if (used_c & 0x2) key.colorarg_b5[1] |= (state->ff.tex_stage[s][D3DTSS_COLORARG1] >> 5) << s;
1612 if (used_c & 0x4) key.colorarg_b4[2] |= (state->ff.tex_stage[s][D3DTSS_COLORARG2] >> 4) << s;
1613 if (used_c & 0x4) key.colorarg_b5[2] |= (state->ff.tex_stage[s][D3DTSS_COLORARG2] >> 5) << s;
1614 }
1615 if (key.ts[s].alphaop != D3DTOP_DISABLE) {
1616 uint8_t used_a = ps_d3dtop_args_mask(key.ts[s].alphaop);
1617 if (used_a & 0x1) key.ts[s].alphaarg0 = state->ff.tex_stage[s][D3DTSS_ALPHAARG0];
1618 if (used_a & 0x2) key.ts[s].alphaarg1 = state->ff.tex_stage[s][D3DTSS_ALPHAARG1];
1619 if (used_a & 0x4) key.ts[s].alphaarg2 = state->ff.tex_stage[s][D3DTSS_ALPHAARG2];
1620 if (used_a & 0x1) key.alphaarg_b4[0] |= (state->ff.tex_stage[s][D3DTSS_ALPHAARG0] >> 4) << s;
1621 if (used_a & 0x2) key.alphaarg_b4[1] |= (state->ff.tex_stage[s][D3DTSS_ALPHAARG1] >> 4) << s;
1622 if (used_a & 0x4) key.alphaarg_b4[2] |= (state->ff.tex_stage[s][D3DTSS_ALPHAARG2] >> 4) << s;
1623 }
1624 key.ts[s].resultarg = state->ff.tex_stage[s][D3DTSS_RESULTARG] == D3DTA_TEMP;
1625
1626 if (state->texture[s]) {
1627 switch (state->texture[s]->base.type) {
1628 case D3DRTYPE_TEXTURE: key.ts[s].textarget = 1; break;
1629 case D3DRTYPE_VOLUMETEXTURE: key.ts[s].textarget = 2; break;
1630 case D3DRTYPE_CUBETEXTURE: key.ts[s].textarget = 3; break;
1631 default:
1632 assert(!"unexpected texture type");
1633 break;
1634 }
1635 } else {
1636 key.ts[s].textarget = 1;
1637 }
1638 }
1639
1640 key.projected = nine_ff_get_projected_key(state);
1641
1642 for (; s < 8; ++s)
1643 key.ts[s].colorop = key.ts[s].alphaop = D3DTOP_DISABLE;
1644 if (state->rs[D3DRS_FOGENABLE])
1645 key.fog_mode = state->rs[D3DRS_FOGTABLEMODE];
1646 key.fog = !!state->rs[D3DRS_FOGENABLE];
1647
1648 ps = util_hash_table_get(device->ff.ht_ps, &key);
1649 if (ps)
1650 return ps;
1651 NinePixelShader9_new(device, &ps, NULL, nine_ff_build_ps(device, &key));
1652
1653 nine_ff_prune_ps(device);
1654 if (ps) {
1655 memcpy(&ps->ff_key, &key, sizeof(ps->ff_key));
1656
1657 err = util_hash_table_set(device->ff.ht_ps, &ps->ff_key, ps);
1658 (void)err;
1659 assert(err == PIPE_OK);
1660 device->ff.num_ps++;
1661 NineUnknown_ConvertRefToBind(NineUnknown(ps));
1662
1663 ps->rt_mask = 0x1;
1664 ps->sampler_mask = sampler_mask;
1665 }
1666 return ps;
1667 }
1668
1669 #define GET_D3DTS(n) nine_state_access_transform(state, D3DTS_##n, FALSE)
1670 #define IS_D3DTS_DIRTY(s,n) ((s)->ff.changed.transform[(D3DTS_##n) / 32] & (1 << ((D3DTS_##n) % 32)))
1671 static void
1672 nine_ff_load_vs_transforms(struct NineDevice9 *device)
1673 {
1674 struct nine_state *state = &device->state;
1675 D3DMATRIX T;
1676 D3DMATRIX *M = (D3DMATRIX *)device->ff.vs_const;
1677 unsigned i;
1678
1679 /* TODO: make this nicer, and only upload the ones we need */
1680 /* TODO: use ff.vs_const as storage of W, V, P matrices */
1681
1682 if (IS_D3DTS_DIRTY(state, WORLD) ||
1683 IS_D3DTS_DIRTY(state, VIEW) ||
1684 IS_D3DTS_DIRTY(state, PROJECTION)) {
1685 /* WVP, WV matrices */
1686 nine_d3d_matrix_matrix_mul(&M[1], GET_D3DTS(WORLD), GET_D3DTS(VIEW));
1687 nine_d3d_matrix_matrix_mul(&M[0], &M[1], GET_D3DTS(PROJECTION));
1688
1689 /* normal matrix == transpose(inverse(WV)) */
1690 nine_d3d_matrix_inverse_3x3(&T, &M[1]);
1691 nine_d3d_matrix_transpose(&M[4], &T);
1692
1693 /* VP matrix */
1694 nine_d3d_matrix_matrix_mul(&M[2], GET_D3DTS(VIEW), GET_D3DTS(PROJECTION));
1695
1696 /* V and W matrix */
1697 M[3] = *GET_D3DTS(VIEW);
1698 M[56] = *GET_D3DTS(WORLD);
1699 }
1700
1701 if (state->rs[D3DRS_VERTEXBLEND] != D3DVBF_DISABLE) {
1702 /* load other world matrices */
1703 for (i = 1; i <= 7; ++i)
1704 M[56 + i] = *GET_D3DTS(WORLDMATRIX(i));
1705 }
1706
1707 device->ff.vs_const[30 * 4] = asfloat(state->rs[D3DRS_TWEENFACTOR]);
1708 }
1709
1710 static void
1711 nine_ff_load_lights(struct NineDevice9 *device)
1712 {
1713 struct nine_state *state = &device->state;
1714 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1715 unsigned l;
1716
1717 if (state->changed.group & NINE_STATE_FF_MATERIAL) {
1718 const D3DMATERIAL9 *mtl = &state->ff.material;
1719
1720 memcpy(&dst[20], &mtl->Diffuse, 4 * sizeof(float));
1721 memcpy(&dst[21], &mtl->Ambient, 4 * sizeof(float));
1722 memcpy(&dst[22], &mtl->Specular, 4 * sizeof(float));
1723 dst[23].x = mtl->Power;
1724 memcpy(&dst[24], &mtl->Emissive, 4 * sizeof(float));
1725 d3dcolor_to_rgba(&dst[25].x, state->rs[D3DRS_AMBIENT]);
1726 dst[19].x = dst[25].x * mtl->Ambient.r + mtl->Emissive.r;
1727 dst[19].y = dst[25].y * mtl->Ambient.g + mtl->Emissive.g;
1728 dst[19].z = dst[25].z * mtl->Ambient.b + mtl->Emissive.b;
1729 dst[19].w = mtl->Ambient.a + mtl->Emissive.a;
1730 }
1731
1732 if (!(state->changed.group & NINE_STATE_FF_LIGHTING))
1733 return;
1734
1735 for (l = 0; l < state->ff.num_lights_active; ++l) {
1736 const D3DLIGHT9 *light = &state->ff.light[state->ff.active_light[l]];
1737
1738 dst[32 + l * 8].x = light->Type;
1739 dst[32 + l * 8].y = light->Attenuation0;
1740 dst[32 + l * 8].z = light->Attenuation1;
1741 dst[32 + l * 8].w = light->Attenuation2;
1742 memcpy(&dst[33 + l * 8].x, &light->Diffuse, sizeof(light->Diffuse));
1743 memcpy(&dst[34 + l * 8].x, &light->Specular, sizeof(light->Specular));
1744 memcpy(&dst[35 + l * 8].x, &light->Ambient, sizeof(light->Ambient));
1745 nine_d3d_vector4_matrix_mul((D3DVECTOR *)&dst[36 + l * 8].x, &light->Position, GET_D3DTS(VIEW));
1746 nine_d3d_vector3_matrix_mul((D3DVECTOR *)&dst[37 + l * 8].x, &light->Direction, GET_D3DTS(VIEW));
1747 dst[36 + l * 8].w = light->Type == D3DLIGHT_DIRECTIONAL ? 1e9f : light->Range;
1748 dst[37 + l * 8].w = light->Falloff;
1749 dst[38 + l * 8].x = cosf(light->Theta * 0.5f);
1750 dst[38 + l * 8].y = cosf(light->Phi * 0.5f);
1751 dst[38 + l * 8].z = 1.0f / (dst[38 + l * 8].x - dst[38 + l * 8].y);
1752 dst[39 + l * 8].w = (l + 1) == state->ff.num_lights_active;
1753 }
1754 }
1755
1756 static void
1757 nine_ff_load_point_and_fog_params(struct NineDevice9 *device)
1758 {
1759 const struct nine_state *state = &device->state;
1760 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1761
1762 if (!(state->changed.group & NINE_STATE_FF_OTHER))
1763 return;
1764 dst[26].x = asfloat(state->rs[D3DRS_POINTSIZE_MIN]);
1765 dst[26].y = asfloat(state->rs[D3DRS_POINTSIZE_MAX]);
1766 dst[26].z = asfloat(state->rs[D3DRS_POINTSIZE]);
1767 dst[26].w = asfloat(state->rs[D3DRS_POINTSCALE_A]);
1768 dst[27].x = asfloat(state->rs[D3DRS_POINTSCALE_B]);
1769 dst[27].y = asfloat(state->rs[D3DRS_POINTSCALE_C]);
1770 dst[28].x = asfloat(state->rs[D3DRS_FOGEND]);
1771 dst[28].y = 1.0f / (asfloat(state->rs[D3DRS_FOGEND]) - asfloat(state->rs[D3DRS_FOGSTART]));
1772 if (isinf(dst[28].y))
1773 dst[28].y = 0.0f;
1774 dst[28].z = asfloat(state->rs[D3DRS_FOGDENSITY]);
1775 }
1776
1777 static void
1778 nine_ff_load_tex_matrices(struct NineDevice9 *device)
1779 {
1780 struct nine_state *state = &device->state;
1781 D3DMATRIX *M = (D3DMATRIX *)device->ff.vs_const;
1782 unsigned s;
1783
1784 if (!(state->ff.changed.transform[0] & 0xff0000))
1785 return;
1786 for (s = 0; s < 8; ++s) {
1787 if (IS_D3DTS_DIRTY(state, TEXTURE0 + s))
1788 nine_d3d_matrix_transpose(&M[32 + s], nine_state_access_transform(state, D3DTS_TEXTURE0 + s, FALSE));
1789 }
1790 }
1791
1792 static void
1793 nine_ff_load_ps_params(struct NineDevice9 *device)
1794 {
1795 const struct nine_state *state = &device->state;
1796 struct fvec4 *dst = (struct fvec4 *)device->ff.ps_const;
1797 unsigned s;
1798
1799 if (!(state->changed.group & (NINE_STATE_FF_PSSTAGES | NINE_STATE_FF_OTHER)))
1800 return;
1801
1802 for (s = 0; s < 8; ++s)
1803 d3dcolor_to_rgba(&dst[s].x, state->ff.tex_stage[s][D3DTSS_CONSTANT]);
1804
1805 for (s = 0; s < 8; ++s) {
1806 dst[8 + s].x = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT00]);
1807 dst[8 + s].y = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT01]);
1808 dst[8 + s].z = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT10]);
1809 dst[8 + s].w = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT11]);
1810 if (s & 1) {
1811 dst[8 + s / 2].z = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLSCALE]);
1812 dst[8 + s / 2].w = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLOFFSET]);
1813 } else {
1814 dst[8 + s / 2].x = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLSCALE]);
1815 dst[8 + s / 2].y = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLOFFSET]);
1816 }
1817 }
1818
1819 d3dcolor_to_rgba(&dst[20].x, state->rs[D3DRS_TEXTUREFACTOR]);
1820 d3dcolor_to_rgba(&dst[21].x, state->rs[D3DRS_FOGCOLOR]);
1821 dst[22].x = asfloat(state->rs[D3DRS_FOGEND]);
1822 dst[22].y = 1.0f / (asfloat(state->rs[D3DRS_FOGEND]) - asfloat(state->rs[D3DRS_FOGSTART]));
1823 dst[22].z = asfloat(state->rs[D3DRS_FOGDENSITY]);
1824 }
1825
1826 static void
1827 nine_ff_load_viewport_info(struct NineDevice9 *device)
1828 {
1829 D3DVIEWPORT9 *viewport = &device->state.viewport;
1830 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1831 float diffZ = viewport->MaxZ - viewport->MinZ;
1832
1833 /* Note: the other functions avoids to fill the const again if nothing changed.
1834 * But we don't have much to fill, and adding code to allow that may be complex
1835 * so just fill it always */
1836 dst[100].x = 2.0f / (float)(viewport->Width);
1837 dst[100].y = 2.0f / (float)(viewport->Height);
1838 dst[100].z = (diffZ == 0.0f) ? 0.0f : (1.0f / diffZ);
1839 dst[101].x = (float)(viewport->X);
1840 dst[101].y = (float)(viewport->Y);
1841 dst[101].z = (float)(viewport->MinZ);
1842 }
1843
1844 void
1845 nine_ff_update(struct NineDevice9 *device)
1846 {
1847 struct nine_state *state = &device->state;
1848 struct pipe_constant_buffer cb;
1849
1850 DBG("vs=%p ps=%p\n", device->state.vs, device->state.ps);
1851
1852 /* NOTE: the only reference belongs to the hash table */
1853 if (!state->programmable_vs) {
1854 device->ff.vs = nine_ff_get_vs(device);
1855 device->state.changed.group |= NINE_STATE_VS;
1856 }
1857 if (!device->state.ps) {
1858 device->ff.ps = nine_ff_get_ps(device);
1859 device->state.changed.group |= NINE_STATE_PS;
1860 }
1861
1862 if (!state->programmable_vs) {
1863 nine_ff_load_vs_transforms(device);
1864 nine_ff_load_tex_matrices(device);
1865 nine_ff_load_lights(device);
1866 nine_ff_load_point_and_fog_params(device);
1867 nine_ff_load_viewport_info(device);
1868
1869 memset(state->ff.changed.transform, 0, sizeof(state->ff.changed.transform));
1870
1871 cb.buffer_offset = 0;
1872 cb.buffer = NULL;
1873 cb.user_buffer = device->ff.vs_const;
1874 cb.buffer_size = NINE_FF_NUM_VS_CONST * 4 * sizeof(float);
1875
1876 if (!device->driver_caps.user_cbufs) {
1877 u_upload_data(device->constbuf_uploader,
1878 0,
1879 cb.buffer_size,
1880 device->constbuf_alignment,
1881 cb.user_buffer,
1882 &cb.buffer_offset,
1883 &cb.buffer);
1884 u_upload_unmap(device->constbuf_uploader);
1885 cb.user_buffer = NULL;
1886 }
1887 state->pipe.cb_vs_ff = cb;
1888 state->commit |= NINE_STATE_COMMIT_CONST_VS;
1889 }
1890
1891 if (!device->state.ps) {
1892 nine_ff_load_ps_params(device);
1893
1894 cb.buffer_offset = 0;
1895 cb.buffer = NULL;
1896 cb.user_buffer = device->ff.ps_const;
1897 cb.buffer_size = NINE_FF_NUM_PS_CONST * 4 * sizeof(float);
1898
1899 if (!device->driver_caps.user_cbufs) {
1900 u_upload_data(device->constbuf_uploader,
1901 0,
1902 cb.buffer_size,
1903 device->constbuf_alignment,
1904 cb.user_buffer,
1905 &cb.buffer_offset,
1906 &cb.buffer);
1907 u_upload_unmap(device->constbuf_uploader);
1908 cb.user_buffer = NULL;
1909 }
1910 state->pipe.cb_ps_ff = cb;
1911 state->commit |= NINE_STATE_COMMIT_CONST_PS;
1912 }
1913
1914 device->state.changed.group &= ~NINE_STATE_FF;
1915 }
1916
1917
1918 boolean
1919 nine_ff_init(struct NineDevice9 *device)
1920 {
1921 device->ff.ht_vs = util_hash_table_create(nine_ff_vs_key_hash,
1922 nine_ff_vs_key_comp);
1923 device->ff.ht_ps = util_hash_table_create(nine_ff_ps_key_hash,
1924 nine_ff_ps_key_comp);
1925
1926 device->ff.ht_fvf = util_hash_table_create(nine_ff_fvf_key_hash,
1927 nine_ff_fvf_key_comp);
1928
1929 device->ff.vs_const = CALLOC(NINE_FF_NUM_VS_CONST, 4 * sizeof(float));
1930 device->ff.ps_const = CALLOC(NINE_FF_NUM_PS_CONST, 4 * sizeof(float));
1931
1932 return device->ff.ht_vs && device->ff.ht_ps &&
1933 device->ff.ht_fvf &&
1934 device->ff.vs_const && device->ff.ps_const;
1935 }
1936
1937 static enum pipe_error nine_ff_ht_delete_cb(void *key, void *value, void *data)
1938 {
1939 NineUnknown_Unbind(NineUnknown(value));
1940 return PIPE_OK;
1941 }
1942
1943 void
1944 nine_ff_fini(struct NineDevice9 *device)
1945 {
1946 if (device->ff.ht_vs) {
1947 util_hash_table_foreach(device->ff.ht_vs, nine_ff_ht_delete_cb, NULL);
1948 util_hash_table_destroy(device->ff.ht_vs);
1949 }
1950 if (device->ff.ht_ps) {
1951 util_hash_table_foreach(device->ff.ht_ps, nine_ff_ht_delete_cb, NULL);
1952 util_hash_table_destroy(device->ff.ht_ps);
1953 }
1954 if (device->ff.ht_fvf) {
1955 util_hash_table_foreach(device->ff.ht_fvf, nine_ff_ht_delete_cb, NULL);
1956 util_hash_table_destroy(device->ff.ht_fvf);
1957 }
1958 device->ff.vs = NULL; /* destroyed by unbinding from hash table */
1959 device->ff.ps = NULL;
1960
1961 FREE(device->ff.vs_const);
1962 FREE(device->ff.ps_const);
1963 }
1964
1965 static void
1966 nine_ff_prune_vs(struct NineDevice9 *device)
1967 {
1968 if (device->ff.num_vs > 100) {
1969 /* could destroy the bound one here, so unbind */
1970 device->pipe->bind_vs_state(device->pipe, NULL);
1971 util_hash_table_foreach(device->ff.ht_vs, nine_ff_ht_delete_cb, NULL);
1972 util_hash_table_clear(device->ff.ht_vs);
1973 device->ff.num_vs = 0;
1974 device->state.changed.group |= NINE_STATE_VS;
1975 }
1976 }
1977 static void
1978 nine_ff_prune_ps(struct NineDevice9 *device)
1979 {
1980 if (device->ff.num_ps > 100) {
1981 /* could destroy the bound one here, so unbind */
1982 device->pipe->bind_fs_state(device->pipe, NULL);
1983 util_hash_table_foreach(device->ff.ht_ps, nine_ff_ht_delete_cb, NULL);
1984 util_hash_table_clear(device->ff.ht_ps);
1985 device->ff.num_ps = 0;
1986 device->state.changed.group |= NINE_STATE_PS;
1987 }
1988 }
1989
1990 /* ========================================================================== */
1991
1992 /* Matrix multiplication:
1993 *
1994 * in memory: 0 1 2 3 (row major)
1995 * 4 5 6 7
1996 * 8 9 a b
1997 * c d e f
1998 *
1999 * cA cB cC cD
2000 * r0 = (r0 * cA) (r0 * cB) . .
2001 * r1 = (r1 * cA) (r1 * cB)
2002 * r2 = (r2 * cA) .
2003 * r3 = (r3 * cA) .
2004 *
2005 * r: (11) (12) (13) (14)
2006 * (21) (22) (23) (24)
2007 * (31) (32) (33) (34)
2008 * (41) (42) (43) (44)
2009 * l: (11 12 13 14)
2010 * (21 22 23 24)
2011 * (31 32 33 34)
2012 * (41 42 43 44)
2013 *
2014 * v: (x y z 1 )
2015 *
2016 * t.xyzw = MUL(v.xxxx, r[0]);
2017 * t.xyzw = MAD(v.yyyy, r[1], t.xyzw);
2018 * t.xyzw = MAD(v.zzzz, r[2], t.xyzw);
2019 * v.xyzw = MAD(v.wwww, r[3], t.xyzw);
2020 *
2021 * v.x = DP4(v, c[0]);
2022 * v.y = DP4(v, c[1]);
2023 * v.z = DP4(v, c[2]);
2024 * v.w = DP4(v, c[3]) = 1
2025 */
2026
2027 /*
2028 static void
2029 nine_D3DMATRIX_print(const D3DMATRIX *M)
2030 {
2031 DBG("\n(%f %f %f %f)\n"
2032 "(%f %f %f %f)\n"
2033 "(%f %f %f %f)\n"
2034 "(%f %f %f %f)\n",
2035 M->m[0][0], M->m[0][1], M->m[0][2], M->m[0][3],
2036 M->m[1][0], M->m[1][1], M->m[1][2], M->m[1][3],
2037 M->m[2][0], M->m[2][1], M->m[2][2], M->m[2][3],
2038 M->m[3][0], M->m[3][1], M->m[3][2], M->m[3][3]);
2039 }
2040 */
2041
2042 static inline float
2043 nine_DP4_row_col(const D3DMATRIX *A, int r, const D3DMATRIX *B, int c)
2044 {
2045 return A->m[r][0] * B->m[0][c] +
2046 A->m[r][1] * B->m[1][c] +
2047 A->m[r][2] * B->m[2][c] +
2048 A->m[r][3] * B->m[3][c];
2049 }
2050
2051 static inline float
2052 nine_DP4_vec_col(const D3DVECTOR *v, const D3DMATRIX *M, int c)
2053 {
2054 return v->x * M->m[0][c] +
2055 v->y * M->m[1][c] +
2056 v->z * M->m[2][c] +
2057 1.0f * M->m[3][c];
2058 }
2059
2060 static inline float
2061 nine_DP3_vec_col(const D3DVECTOR *v, const D3DMATRIX *M, int c)
2062 {
2063 return v->x * M->m[0][c] +
2064 v->y * M->m[1][c] +
2065 v->z * M->m[2][c];
2066 }
2067
2068 void
2069 nine_d3d_matrix_matrix_mul(D3DMATRIX *D, const D3DMATRIX *L, const D3DMATRIX *R)
2070 {
2071 D->_11 = nine_DP4_row_col(L, 0, R, 0);
2072 D->_12 = nine_DP4_row_col(L, 0, R, 1);
2073 D->_13 = nine_DP4_row_col(L, 0, R, 2);
2074 D->_14 = nine_DP4_row_col(L, 0, R, 3);
2075
2076 D->_21 = nine_DP4_row_col(L, 1, R, 0);
2077 D->_22 = nine_DP4_row_col(L, 1, R, 1);
2078 D->_23 = nine_DP4_row_col(L, 1, R, 2);
2079 D->_24 = nine_DP4_row_col(L, 1, R, 3);
2080
2081 D->_31 = nine_DP4_row_col(L, 2, R, 0);
2082 D->_32 = nine_DP4_row_col(L, 2, R, 1);
2083 D->_33 = nine_DP4_row_col(L, 2, R, 2);
2084 D->_34 = nine_DP4_row_col(L, 2, R, 3);
2085
2086 D->_41 = nine_DP4_row_col(L, 3, R, 0);
2087 D->_42 = nine_DP4_row_col(L, 3, R, 1);
2088 D->_43 = nine_DP4_row_col(L, 3, R, 2);
2089 D->_44 = nine_DP4_row_col(L, 3, R, 3);
2090 }
2091
2092 void
2093 nine_d3d_vector4_matrix_mul(D3DVECTOR *d, const D3DVECTOR *v, const D3DMATRIX *M)
2094 {
2095 d->x = nine_DP4_vec_col(v, M, 0);
2096 d->y = nine_DP4_vec_col(v, M, 1);
2097 d->z = nine_DP4_vec_col(v, M, 2);
2098 }
2099
2100 void
2101 nine_d3d_vector3_matrix_mul(D3DVECTOR *d, const D3DVECTOR *v, const D3DMATRIX *M)
2102 {
2103 d->x = nine_DP3_vec_col(v, M, 0);
2104 d->y = nine_DP3_vec_col(v, M, 1);
2105 d->z = nine_DP3_vec_col(v, M, 2);
2106 }
2107
2108 void
2109 nine_d3d_matrix_transpose(D3DMATRIX *D, const D3DMATRIX *M)
2110 {
2111 unsigned i, j;
2112 for (i = 0; i < 4; ++i)
2113 for (j = 0; j < 4; ++j)
2114 D->m[i][j] = M->m[j][i];
2115 }
2116
2117 #define _M_ADD_PROD_1i_2j_3k_4l(i,j,k,l) do { \
2118 float t = M->_1##i * M->_2##j * M->_3##k * M->_4##l; \
2119 if (t > 0.0f) pos += t; else neg += t; } while(0)
2120
2121 #define _M_SUB_PROD_1i_2j_3k_4l(i,j,k,l) do { \
2122 float t = M->_1##i * M->_2##j * M->_3##k * M->_4##l; \
2123 if (t > 0.0f) neg -= t; else pos -= t; } while(0)
2124 float
2125 nine_d3d_matrix_det(const D3DMATRIX *M)
2126 {
2127 float pos = 0.0f;
2128 float neg = 0.0f;
2129
2130 _M_ADD_PROD_1i_2j_3k_4l(1, 2, 3, 4);
2131 _M_ADD_PROD_1i_2j_3k_4l(1, 3, 4, 2);
2132 _M_ADD_PROD_1i_2j_3k_4l(1, 4, 2, 3);
2133
2134 _M_ADD_PROD_1i_2j_3k_4l(2, 1, 4, 3);
2135 _M_ADD_PROD_1i_2j_3k_4l(2, 3, 1, 4);
2136 _M_ADD_PROD_1i_2j_3k_4l(2, 4, 3, 1);
2137
2138 _M_ADD_PROD_1i_2j_3k_4l(3, 1, 2, 4);
2139 _M_ADD_PROD_1i_2j_3k_4l(3, 2, 4, 1);
2140 _M_ADD_PROD_1i_2j_3k_4l(3, 4, 1, 2);
2141
2142 _M_ADD_PROD_1i_2j_3k_4l(4, 1, 3, 2);
2143 _M_ADD_PROD_1i_2j_3k_4l(4, 2, 1, 3);
2144 _M_ADD_PROD_1i_2j_3k_4l(4, 3, 2, 1);
2145
2146 _M_SUB_PROD_1i_2j_3k_4l(1, 2, 4, 3);
2147 _M_SUB_PROD_1i_2j_3k_4l(1, 3, 2, 4);
2148 _M_SUB_PROD_1i_2j_3k_4l(1, 4, 3, 2);
2149
2150 _M_SUB_PROD_1i_2j_3k_4l(2, 1, 3, 4);
2151 _M_SUB_PROD_1i_2j_3k_4l(2, 3, 4, 1);
2152 _M_SUB_PROD_1i_2j_3k_4l(2, 4, 1, 3);
2153
2154 _M_SUB_PROD_1i_2j_3k_4l(3, 1, 4, 2);
2155 _M_SUB_PROD_1i_2j_3k_4l(3, 2, 1, 4);
2156 _M_SUB_PROD_1i_2j_3k_4l(3, 4, 2, 1);
2157
2158 _M_SUB_PROD_1i_2j_3k_4l(4, 1, 2, 3);
2159 _M_SUB_PROD_1i_2j_3k_4l(4, 2, 3, 1);
2160 _M_SUB_PROD_1i_2j_3k_4l(4, 3, 1, 2);
2161
2162 return pos + neg;
2163 }
2164
2165 /* XXX: Probably better to just use src/mesa/math/m_matrix.c because
2166 * I have no idea where this code came from.
2167 */
2168 void
2169 nine_d3d_matrix_inverse(D3DMATRIX *D, const D3DMATRIX *M)
2170 {
2171 int i, k;
2172 float det;
2173
2174 D->m[0][0] =
2175 M->m[1][1] * M->m[2][2] * M->m[3][3] -
2176 M->m[1][1] * M->m[3][2] * M->m[2][3] -
2177 M->m[1][2] * M->m[2][1] * M->m[3][3] +
2178 M->m[1][2] * M->m[3][1] * M->m[2][3] +
2179 M->m[1][3] * M->m[2][1] * M->m[3][2] -
2180 M->m[1][3] * M->m[3][1] * M->m[2][2];
2181
2182 D->m[0][1] =
2183 -M->m[0][1] * M->m[2][2] * M->m[3][3] +
2184 M->m[0][1] * M->m[3][2] * M->m[2][3] +
2185 M->m[0][2] * M->m[2][1] * M->m[3][3] -
2186 M->m[0][2] * M->m[3][1] * M->m[2][3] -
2187 M->m[0][3] * M->m[2][1] * M->m[3][2] +
2188 M->m[0][3] * M->m[3][1] * M->m[2][2];
2189
2190 D->m[0][2] =
2191 M->m[0][1] * M->m[1][2] * M->m[3][3] -
2192 M->m[0][1] * M->m[3][2] * M->m[1][3] -
2193 M->m[0][2] * M->m[1][1] * M->m[3][3] +
2194 M->m[0][2] * M->m[3][1] * M->m[1][3] +
2195 M->m[0][3] * M->m[1][1] * M->m[3][2] -
2196 M->m[0][3] * M->m[3][1] * M->m[1][2];
2197
2198 D->m[0][3] =
2199 -M->m[0][1] * M->m[1][2] * M->m[2][3] +
2200 M->m[0][1] * M->m[2][2] * M->m[1][3] +
2201 M->m[0][2] * M->m[1][1] * M->m[2][3] -
2202 M->m[0][2] * M->m[2][1] * M->m[1][3] -
2203 M->m[0][3] * M->m[1][1] * M->m[2][2] +
2204 M->m[0][3] * M->m[2][1] * M->m[1][2];
2205
2206 D->m[1][0] =
2207 -M->m[1][0] * M->m[2][2] * M->m[3][3] +
2208 M->m[1][0] * M->m[3][2] * M->m[2][3] +
2209 M->m[1][2] * M->m[2][0] * M->m[3][3] -
2210 M->m[1][2] * M->m[3][0] * M->m[2][3] -
2211 M->m[1][3] * M->m[2][0] * M->m[3][2] +
2212 M->m[1][3] * M->m[3][0] * M->m[2][2];
2213
2214 D->m[1][1] =
2215 M->m[0][0] * M->m[2][2] * M->m[3][3] -
2216 M->m[0][0] * M->m[3][2] * M->m[2][3] -
2217 M->m[0][2] * M->m[2][0] * M->m[3][3] +
2218 M->m[0][2] * M->m[3][0] * M->m[2][3] +
2219 M->m[0][3] * M->m[2][0] * M->m[3][2] -
2220 M->m[0][3] * M->m[3][0] * M->m[2][2];
2221
2222 D->m[1][2] =
2223 -M->m[0][0] * M->m[1][2] * M->m[3][3] +
2224 M->m[0][0] * M->m[3][2] * M->m[1][3] +
2225 M->m[0][2] * M->m[1][0] * M->m[3][3] -
2226 M->m[0][2] * M->m[3][0] * M->m[1][3] -
2227 M->m[0][3] * M->m[1][0] * M->m[3][2] +
2228 M->m[0][3] * M->m[3][0] * M->m[1][2];
2229
2230 D->m[1][3] =
2231 M->m[0][0] * M->m[1][2] * M->m[2][3] -
2232 M->m[0][0] * M->m[2][2] * M->m[1][3] -
2233 M->m[0][2] * M->m[1][0] * M->m[2][3] +
2234 M->m[0][2] * M->m[2][0] * M->m[1][3] +
2235 M->m[0][3] * M->m[1][0] * M->m[2][2] -
2236 M->m[0][3] * M->m[2][0] * M->m[1][2];
2237
2238 D->m[2][0] =
2239 M->m[1][0] * M->m[2][1] * M->m[3][3] -
2240 M->m[1][0] * M->m[3][1] * M->m[2][3] -
2241 M->m[1][1] * M->m[2][0] * M->m[3][3] +
2242 M->m[1][1] * M->m[3][0] * M->m[2][3] +
2243 M->m[1][3] * M->m[2][0] * M->m[3][1] -
2244 M->m[1][3] * M->m[3][0] * M->m[2][1];
2245
2246 D->m[2][1] =
2247 -M->m[0][0] * M->m[2][1] * M->m[3][3] +
2248 M->m[0][0] * M->m[3][1] * M->m[2][3] +
2249 M->m[0][1] * M->m[2][0] * M->m[3][3] -
2250 M->m[0][1] * M->m[3][0] * M->m[2][3] -
2251 M->m[0][3] * M->m[2][0] * M->m[3][1] +
2252 M->m[0][3] * M->m[3][0] * M->m[2][1];
2253
2254 D->m[2][2] =
2255 M->m[0][0] * M->m[1][1] * M->m[3][3] -
2256 M->m[0][0] * M->m[3][1] * M->m[1][3] -
2257 M->m[0][1] * M->m[1][0] * M->m[3][3] +
2258 M->m[0][1] * M->m[3][0] * M->m[1][3] +
2259 M->m[0][3] * M->m[1][0] * M->m[3][1] -
2260 M->m[0][3] * M->m[3][0] * M->m[1][1];
2261
2262 D->m[2][3] =
2263 -M->m[0][0] * M->m[1][1] * M->m[2][3] +
2264 M->m[0][0] * M->m[2][1] * M->m[1][3] +
2265 M->m[0][1] * M->m[1][0] * M->m[2][3] -
2266 M->m[0][1] * M->m[2][0] * M->m[1][3] -
2267 M->m[0][3] * M->m[1][0] * M->m[2][1] +
2268 M->m[0][3] * M->m[2][0] * M->m[1][1];
2269
2270 D->m[3][0] =
2271 -M->m[1][0] * M->m[2][1] * M->m[3][2] +
2272 M->m[1][0] * M->m[3][1] * M->m[2][2] +
2273 M->m[1][1] * M->m[2][0] * M->m[3][2] -
2274 M->m[1][1] * M->m[3][0] * M->m[2][2] -
2275 M->m[1][2] * M->m[2][0] * M->m[3][1] +
2276 M->m[1][2] * M->m[3][0] * M->m[2][1];
2277
2278 D->m[3][1] =
2279 M->m[0][0] * M->m[2][1] * M->m[3][2] -
2280 M->m[0][0] * M->m[3][1] * M->m[2][2] -
2281 M->m[0][1] * M->m[2][0] * M->m[3][2] +
2282 M->m[0][1] * M->m[3][0] * M->m[2][2] +
2283 M->m[0][2] * M->m[2][0] * M->m[3][1] -
2284 M->m[0][2] * M->m[3][0] * M->m[2][1];
2285
2286 D->m[3][2] =
2287 -M->m[0][0] * M->m[1][1] * M->m[3][2] +
2288 M->m[0][0] * M->m[3][1] * M->m[1][2] +
2289 M->m[0][1] * M->m[1][0] * M->m[3][2] -
2290 M->m[0][1] * M->m[3][0] * M->m[1][2] -
2291 M->m[0][2] * M->m[1][0] * M->m[3][1] +
2292 M->m[0][2] * M->m[3][0] * M->m[1][1];
2293
2294 D->m[3][3] =
2295 M->m[0][0] * M->m[1][1] * M->m[2][2] -
2296 M->m[0][0] * M->m[2][1] * M->m[1][2] -
2297 M->m[0][1] * M->m[1][0] * M->m[2][2] +
2298 M->m[0][1] * M->m[2][0] * M->m[1][2] +
2299 M->m[0][2] * M->m[1][0] * M->m[2][1] -
2300 M->m[0][2] * M->m[2][0] * M->m[1][1];
2301
2302 det =
2303 M->m[0][0] * D->m[0][0] +
2304 M->m[1][0] * D->m[0][1] +
2305 M->m[2][0] * D->m[0][2] +
2306 M->m[3][0] * D->m[0][3];
2307
2308 det = 1.0 / det;
2309
2310 for (i = 0; i < 4; i++)
2311 for (k = 0; k < 4; k++)
2312 D->m[i][k] *= det;
2313
2314 #ifdef DEBUG
2315 {
2316 D3DMATRIX I;
2317
2318 nine_d3d_matrix_matrix_mul(&I, D, M);
2319
2320 for (i = 0; i < 4; ++i)
2321 for (k = 0; k < 4; ++k)
2322 if (fabsf(I.m[i][k] - (float)(i == k)) > 1e-3)
2323 DBG("Matrix inversion check FAILED !\n");
2324 }
2325 #endif
2326 }
2327
2328 /* TODO: don't use 4x4 inverse, unless this gets all nicely inlined ? */
2329 void
2330 nine_d3d_matrix_inverse_3x3(D3DMATRIX *D, const D3DMATRIX *M)
2331 {
2332 D3DMATRIX T;
2333 unsigned i, j;
2334
2335 for (i = 0; i < 3; ++i)
2336 for (j = 0; j < 3; ++j)
2337 T.m[i][j] = M->m[i][j];
2338 for (i = 0; i < 3; ++i) {
2339 T.m[i][3] = 0.0f;
2340 T.m[3][i] = 0.0f;
2341 }
2342 T.m[3][3] = 1.0f;
2343
2344 nine_d3d_matrix_inverse(D, &T);
2345 }