projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
st/nine: Compact ff vs constants a bit
[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 192
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 normalizenormals : 1;
61 uint32_t pad1 : 5;
62 uint32_t tc_dim_input: 16; /* 8 * 2 bits */
63 uint32_t pad2 : 16;
64 uint32_t tc_dim_output: 24; /* 8 * 3 bits */
65 uint32_t pad3 : 8;
66 uint32_t tc_gen : 24; /* 8 * 3 bits */
67 uint32_t pad4 : 8;
68 uint32_t tc_idx : 24;
69 uint32_t pad5 : 8;
70 uint32_t passthrough;
71 };
72 uint64_t value64[3]; /* don't forget to resize VertexShader9.ff_key */
73 uint32_t value32[6];
74 };
75 };
76
77 /* Texture stage state:
78 *
79 * COLOROP D3DTOP 5 bit
80 * ALPHAOP D3DTOP 5 bit
81 * COLORARG0 D3DTA 3 bit
82 * COLORARG1 D3DTA 3 bit
83 * COLORARG2 D3DTA 3 bit
84 * ALPHAARG0 D3DTA 3 bit
85 * ALPHAARG1 D3DTA 3 bit
86 * ALPHAARG2 D3DTA 3 bit
87 * RESULTARG D3DTA 1 bit (CURRENT:0 or TEMP:1)
88 * TEXCOORDINDEX 0 - 7 3 bit
89 * ===========================
90 * 32 bit per stage
91 */
92 struct nine_ff_ps_key
93 {
94 union {
95 struct {
96 struct {
97 uint32_t colorop : 5;
98 uint32_t alphaop : 5;
99 uint32_t colorarg0 : 3;
100 uint32_t colorarg1 : 3;
101 uint32_t colorarg2 : 3;
102 uint32_t alphaarg0 : 3;
103 uint32_t alphaarg1 : 3;
104 uint32_t alphaarg2 : 3;
105 uint32_t resultarg : 1; /* CURRENT:0 or TEMP:1 */
106 uint32_t textarget : 2; /* 1D/2D/3D/CUBE */
107 uint32_t pad : 1;
108 /* that's 32 bit exactly */
109 } ts[8];
110 uint32_t projected : 16;
111 uint32_t fog : 1; /* for vFog coming from VS */
112 uint32_t fog_mode : 2;
113 uint32_t specular : 1;
114 uint32_t pad1 : 12; /* 9 32-bit words with this */
115 uint8_t colorarg_b4[3];
116 uint8_t colorarg_b5[3];
117 uint8_t alphaarg_b4[3]; /* 11 32-bit words plus a byte */
118 uint8_t pad2[3];
119 };
120 uint64_t value64[6]; /* don't forget to resize PixelShader9.ff_key */
121 uint32_t value32[12];
122 };
123 };
124
125 static unsigned nine_ff_vs_key_hash(void *key)
126 {
127 struct nine_ff_vs_key *vs = key;
128 unsigned i;
129 uint32_t hash = vs->value32[0];
130 for (i = 1; i < ARRAY_SIZE(vs->value32); ++i)
131 hash ^= vs->value32[i];
132 return hash;
133 }
134 static int nine_ff_vs_key_comp(void *key1, void *key2)
135 {
136 struct nine_ff_vs_key *a = (struct nine_ff_vs_key *)key1;
137 struct nine_ff_vs_key *b = (struct nine_ff_vs_key *)key2;
138
139 return memcmp(a->value64, b->value64, sizeof(a->value64));
140 }
141 static unsigned nine_ff_ps_key_hash(void *key)
142 {
143 struct nine_ff_ps_key *ps = key;
144 unsigned i;
145 uint32_t hash = ps->value32[0];
146 for (i = 1; i < ARRAY_SIZE(ps->value32); ++i)
147 hash ^= ps->value32[i];
148 return hash;
149 }
150 static int nine_ff_ps_key_comp(void *key1, void *key2)
151 {
152 struct nine_ff_ps_key *a = (struct nine_ff_ps_key *)key1;
153 struct nine_ff_ps_key *b = (struct nine_ff_ps_key *)key2;
154
155 return memcmp(a->value64, b->value64, sizeof(a->value64));
156 }
157 static unsigned nine_ff_fvf_key_hash(void *key)
158 {
159 return *(DWORD *)key;
160 }
161 static int nine_ff_fvf_key_comp(void *key1, void *key2)
162 {
163 return *(DWORD *)key1 != *(DWORD *)key2;
164 }
165
166 static void nine_ff_prune_vs(struct NineDevice9 *);
167 static void nine_ff_prune_ps(struct NineDevice9 *);
168
169 static void nine_ureg_tgsi_dump(struct ureg_program *ureg, boolean override)
170 {
171 if (debug_get_bool_option("NINE_FF_DUMP", FALSE) || override) {
172 unsigned count;
173 const struct tgsi_token *toks = ureg_get_tokens(ureg, &count);
174 tgsi_dump(toks, 0);
175 ureg_free_tokens(toks);
176 }
177 }
178
179 #define _X(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_X)
180 #define _Y(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_Y)
181 #define _Z(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_Z)
182 #define _W(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_W)
183
184 #define _XXXX(r) ureg_scalar(r, TGSI_SWIZZLE_X)
185 #define _YYYY(r) ureg_scalar(r, TGSI_SWIZZLE_Y)
186 #define _ZZZZ(r) ureg_scalar(r, TGSI_SWIZZLE_Z)
187 #define _WWWW(r) ureg_scalar(r, TGSI_SWIZZLE_W)
188
189 #define _XYZW(r) (r)
190
191 /* AL should contain base address of lights table. */
192 #define LIGHT_CONST(i) \
193 ureg_src_indirect(ureg_DECL_constant(ureg, i), _X(AL))
194
195 #define MATERIAL_CONST(i) \
196 ureg_DECL_constant(ureg, 19 + (i))
197
198 #define _CONST(n) ureg_DECL_constant(ureg, n)
199
200 /* VS FF constants layout:
201 *
202 * CONST[ 0.. 3] D3DTS_WORLD * D3DTS_VIEW * D3DTS_PROJECTION
203 * CONST[ 4.. 7] D3DTS_WORLD * D3DTS_VIEW
204 * CONST[ 8..11] D3DTS_PROJECTION
205 * CONST[12..15] D3DTS_VIEW
206 * CONST[16..18] Normal matrix
207 *
208 * CONST[19] MATERIAL.Emissive + Material.Ambient * RS.Ambient
209 * CONST[20] MATERIAL.Diffuse
210 * CONST[21] MATERIAL.Ambient
211 * CONST[22] MATERIAL.Specular
212 * CONST[23].x___ MATERIAL.Power
213 * CONST[24] MATERIAL.Emissive
214 * CONST[25] RS.Ambient
215 *
216 * CONST[26].x___ RS.PointSizeMin
217 * CONST[26]._y__ RS.PointSizeMax
218 * CONST[26].__z_ RS.PointSize
219 * CONST[26].___w RS.PointScaleA
220 * CONST[27].x___ RS.PointScaleB
221 * CONST[27]._y__ RS.PointScaleC
222 *
223 * CONST[28].x___ RS.FogEnd
224 * CONST[28]._y__ 1.0f / (RS.FogEnd - RS.FogStart)
225 * CONST[28].__z_ RS.FogDensity
226
227 * CONST[30].x___ TWEENFACTOR
228 *
229 * CONST[32].x___ LIGHT[0].Type
230 * CONST[32]._yzw LIGHT[0].Attenuation0,1,2
231 * CONST[33] LIGHT[0].Diffuse
232 * CONST[34] LIGHT[0].Specular
233 * CONST[35] LIGHT[0].Ambient
234 * CONST[36].xyz_ LIGHT[0].Position
235 * CONST[36].___w LIGHT[0].Range
236 * CONST[37].xyz_ LIGHT[0].Direction
237 * CONST[37].___w LIGHT[0].Falloff
238 * CONST[38].x___ cos(LIGHT[0].Theta / 2)
239 * CONST[38]._y__ cos(LIGHT[0].Phi / 2)
240 * CONST[38].__z_ 1.0f / (cos(LIGHT[0].Theta / 2) - cos(Light[0].Phi / 2))
241 * CONST[39].xyz_ LIGHT[0].HalfVector (for directional lights)
242 * CONST[39].___w 1 if this is the last active light, 0 if not
243 * CONST[40] LIGHT[1]
244 * CONST[48] LIGHT[2]
245 * CONST[56] LIGHT[3]
246 * CONST[64] LIGHT[4]
247 * CONST[72] LIGHT[5]
248 * CONST[80] LIGHT[6]
249 * CONST[88] LIGHT[7]
250 * NOTE: no lighting code is generated if there are no active lights
251 *
252 * CONST[100].x___ Viewport 2/width
253 * CONST[100]._y__ Viewport 2/height
254 * CONST[100].__z_ Viewport 1/(zmax - zmin)
255 * CONST[100].___w Viewport width
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[160] D3DTS_WORLDMATRIX[0] * D3DTS_VIEW
270 * CONST[164] D3DTS_WORLDMATRIX[1] * D3DTS_VIEW
271 * ...
272 * CONST[188] D3DTS_WORLDMATRIX[7] * D3DTS_VIEW
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 {
321 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
322 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
323
324 ureg_DP3(ureg, tmp_x, src, src);
325 ureg_RSQ(ureg, tmp_x, _X(tmp));
326 ureg_MUL(ureg, dst, src, _X(tmp));
327 ureg_release_temporary(ureg, tmp);
328 }
329
330 static void *
331 nine_ff_build_vs(struct NineDevice9 *device, struct vs_build_ctx *vs)
332 {
333 const struct nine_ff_vs_key *key = vs->key;
334 struct ureg_program *ureg = ureg_create(PIPE_SHADER_VERTEX);
335 struct ureg_dst oPos, oCol[2], oPsz, oFog;
336 struct ureg_dst AR;
337 unsigned i, c;
338 unsigned label[32], l = 0;
339 boolean need_aNrm = key->lighting || key->passthrough & (1 << NINE_DECLUSAGE_NORMAL);
340 boolean need_aVtx = key->lighting || key->fog_mode || key->pointscale;
341 const unsigned texcoord_sn = get_texcoord_sn(device->screen);
342
343 vs->ureg = ureg;
344
345 /* Check which inputs we should transform. */
346 for (i = 0; i < 8 * 3; i += 3) {
347 switch ((key->tc_gen >> i) & 0x3) {
348 case NINED3DTSS_TCI_CAMERASPACENORMAL:
349 need_aNrm = TRUE;
350 break;
351 case NINED3DTSS_TCI_CAMERASPACEPOSITION:
352 need_aVtx = TRUE;
353 break;
354 case NINED3DTSS_TCI_CAMERASPACEREFLECTIONVECTOR:
355 need_aVtx = need_aNrm = TRUE;
356 break;
357 default:
358 break;
359 }
360 }
361
362 /* Declare and record used inputs (needed for linkage with vertex format):
363 * (texture coordinates handled later)
364 */
365 vs->aVtx = build_vs_add_input(vs,
366 key->position_t ? NINE_DECLUSAGE_POSITIONT : NINE_DECLUSAGE_POSITION);
367
368 if (need_aNrm)
369 vs->aNrm = build_vs_add_input(vs, NINE_DECLUSAGE_NORMAL);
370
371 vs->aCol[0] = ureg_imm1f(ureg, 1.0f);
372 vs->aCol[1] = ureg_imm1f(ureg, 1.0f);
373
374 if (key->lighting || key->darkness) {
375 const unsigned mask = key->mtl_diffuse | key->mtl_specular |
376 key->mtl_ambient | key->mtl_emissive;
377 if ((mask & 0x1) && !key->color0in_one)
378 vs->aCol[0] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 0));
379 if ((mask & 0x2) && !key->color1in_one)
380 vs->aCol[1] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 1));
381
382 vs->mtlD = MATERIAL_CONST(1);
383 vs->mtlA = MATERIAL_CONST(2);
384 vs->mtlS = MATERIAL_CONST(3);
385 vs->mtlE = MATERIAL_CONST(5);
386 if (key->mtl_diffuse == 1) vs->mtlD = vs->aCol[0]; else
387 if (key->mtl_diffuse == 2) vs->mtlD = vs->aCol[1];
388 if (key->mtl_ambient == 1) vs->mtlA = vs->aCol[0]; else
389 if (key->mtl_ambient == 2) vs->mtlA = vs->aCol[1];
390 if (key->mtl_specular == 1) vs->mtlS = vs->aCol[0]; else
391 if (key->mtl_specular == 2) vs->mtlS = vs->aCol[1];
392 if (key->mtl_emissive == 1) vs->mtlE = vs->aCol[0]; else
393 if (key->mtl_emissive == 2) vs->mtlE = vs->aCol[1];
394 } else {
395 if (!key->color0in_one) vs->aCol[0] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 0));
396 if (!key->color1in_one) vs->aCol[1] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 1));
397 }
398
399 if (key->vertexpointsize)
400 vs->aPsz = build_vs_add_input(vs, NINE_DECLUSAGE_PSIZE);
401
402 if (key->vertexblend_indexed || key->passthrough & (1 << NINE_DECLUSAGE_BLENDINDICES))
403 vs->aInd = build_vs_add_input(vs, NINE_DECLUSAGE_BLENDINDICES);
404 if (key->vertexblend || key->passthrough & (1 << NINE_DECLUSAGE_BLENDWEIGHT))
405 vs->aWgt = build_vs_add_input(vs, NINE_DECLUSAGE_BLENDWEIGHT);
406 if (key->vertextween) {
407 vs->aVtx1 = build_vs_add_input(vs, NINE_DECLUSAGE_i(POSITION,1));
408 vs->aNrm1 = build_vs_add_input(vs, NINE_DECLUSAGE_i(NORMAL,1));
409 }
410
411 /* Declare outputs:
412 */
413 oPos = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0); /* HPOS */
414 oCol[0] = ureg_saturate(ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0));
415 oCol[1] = ureg_saturate(ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 1));
416 if (key->fog || key->passthrough & (1 << NINE_DECLUSAGE_FOG)) {
417 oFog = ureg_DECL_output(ureg, TGSI_SEMANTIC_FOG, 0);
418 oFog = ureg_writemask(oFog, TGSI_WRITEMASK_X);
419 }
420
421 if (key->vertexpointsize || key->pointscale) {
422 oPsz = ureg_DECL_output_masked(ureg, TGSI_SEMANTIC_PSIZE, 0,
423 TGSI_WRITEMASK_X, 0, 1);
424 oPsz = ureg_writemask(oPsz, TGSI_WRITEMASK_X);
425 }
426
427 if (key->lighting || key->vertexblend)
428 AR = ureg_DECL_address(ureg);
429
430 /* === Vertex transformation / vertex blending:
431 */
432
433 if (key->position_t) {
434 if (device->driver_caps.window_space_position_support) {
435 ureg_MOV(ureg, oPos, vs->aVtx);
436 } else {
437 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
438 /* vs->aVtx contains the coordinates buffer wise.
439 * later in the pipeline, clipping, viewport and division
440 * by w (rhw = 1/w) are going to be applied, so do the reverse
441 * of these transformations (except clipping) to have the good
442 * position at the end.*/
443 ureg_MOV(ureg, tmp, vs->aVtx);
444 /* X from [X_min, X_min + width] to [-1, 1], same for Y. Z to [0, 1] */
445 ureg_SUB(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _CONST(101));
446 ureg_MUL(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _CONST(100));
447 ureg_SUB(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XY), ureg_src(tmp), ureg_imm1f(ureg, 1.0f));
448 /* Y needs to be reversed */
449 ureg_MOV(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_Y), ureg_negate(ureg_src(tmp)));
450 /* inverse rhw */
451 ureg_RCP(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_W), _W(tmp));
452 /* multiply X, Y, Z by w */
453 ureg_MUL(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _W(tmp));
454 ureg_MOV(ureg, oPos, ureg_src(tmp));
455 ureg_release_temporary(ureg, tmp);
456 }
457 } else if (key->vertexblend) {
458 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
459 struct ureg_dst aVtx_dst = ureg_DECL_temporary(ureg);
460 struct ureg_dst sum_blendweights = ureg_DECL_temporary(ureg);
461 struct ureg_src cWM[4];
462
463 for (i = 160; i <= 191; ++i)
464 ureg_DECL_constant(ureg, i);
465
466 /* translate world matrix index to constant file index */
467 if (key->vertexblend_indexed) {
468 ureg_MAD(ureg, tmp, vs->aInd, ureg_imm1f(ureg, 4.0f), ureg_imm1f(ureg, 160.0f));
469 ureg_ARL(ureg, AR, ureg_src(tmp));
470 }
471
472 ureg_MOV(ureg, aVtx_dst, ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 0.0f));
473 ureg_MOV(ureg, sum_blendweights, ureg_imm4f(ureg, 1.0f, 1.0f, 1.0f, 1.0f));
474
475 for (i = 0; i < key->vertexblend; ++i) {
476 for (c = 0; c < 4; ++c) {
477 cWM[c] = ureg_src_register(TGSI_FILE_CONSTANT, (160 + i * 4) * !key->vertexblend_indexed + c);
478 if (key->vertexblend_indexed)
479 cWM[c] = ureg_src_indirect(cWM[c], ureg_scalar(ureg_src(AR), i));
480 }
481 /* multiply by WORLD(index) */
482 ureg_MUL(ureg, tmp, _XXXX(vs->aVtx), cWM[0]);
483 ureg_MAD(ureg, tmp, _YYYY(vs->aVtx), cWM[1], ureg_src(tmp));
484 ureg_MAD(ureg, tmp, _ZZZZ(vs->aVtx), cWM[2], ureg_src(tmp));
485 ureg_MAD(ureg, tmp, _WWWW(vs->aVtx), cWM[3], ureg_src(tmp));
486
487 if (i < (key->vertexblend - 1)) {
488 /* accumulate weighted position value */
489 ureg_MAD(ureg, aVtx_dst, ureg_src(tmp), ureg_scalar(vs->aWgt, i), ureg_src(aVtx_dst));
490 /* subtract weighted position value for last value */
491 ureg_SUB(ureg, sum_blendweights, ureg_src(sum_blendweights), ureg_scalar(vs->aWgt, i));
492 }
493 }
494
495 /* the last weighted position is always 1 - sum_of_previous_weights */
496 ureg_MAD(ureg, aVtx_dst, ureg_src(tmp), ureg_scalar(ureg_src(sum_blendweights), key->vertexblend - 1), ureg_src(aVtx_dst));
497
498 /* multiply by VIEW_PROJ */
499 ureg_MUL(ureg, tmp, _X(aVtx_dst), _CONST(8));
500 ureg_MAD(ureg, tmp, _Y(aVtx_dst), _CONST(9), ureg_src(tmp));
501 ureg_MAD(ureg, tmp, _Z(aVtx_dst), _CONST(10), ureg_src(tmp));
502 ureg_MAD(ureg, oPos, _W(aVtx_dst), _CONST(11), ureg_src(tmp));
503
504 if (need_aVtx)
505 vs->aVtx = ureg_src(aVtx_dst);
506
507 ureg_release_temporary(ureg, tmp);
508 ureg_release_temporary(ureg, sum_blendweights);
509 if (!need_aVtx)
510 ureg_release_temporary(ureg, aVtx_dst);
511
512 if (need_aNrm) {
513 struct ureg_dst aNrm_dst = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
514 ureg_MUL(ureg, aNrm_dst, _XXXX(vs->aNrm), _CONST(16));
515 ureg_MAD(ureg, aNrm_dst, _YYYY(vs->aNrm), _CONST(17), ureg_src(aNrm_dst));
516 ureg_MAD(ureg, aNrm_dst, _ZZZZ(vs->aNrm), _CONST(18), ureg_src(aNrm_dst));
517 if (key->normalizenormals)
518 ureg_normalize3(ureg, aNrm_dst, ureg_src(aNrm_dst));
519 vs->aNrm = ureg_src(aNrm_dst);
520 }
521 } else {
522 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
523
524 if (key->vertextween) {
525 struct ureg_dst aVtx_dst = ureg_DECL_temporary(ureg);
526 ureg_LRP(ureg, aVtx_dst, _XXXX(_CONST(30)), vs->aVtx, vs->aVtx1);
527 vs->aVtx = ureg_src(aVtx_dst);
528 if (need_aNrm) {
529 struct ureg_dst aNrm_dst = ureg_DECL_temporary(ureg);
530 ureg_LRP(ureg, aNrm_dst, _XXXX(_CONST(30)), vs->aNrm, vs->aNrm1);
531 vs->aNrm = ureg_src(aNrm_dst);
532 }
533 }
534
535 /* position = vertex * WORLD_VIEW_PROJ */
536 ureg_MUL(ureg, tmp, _XXXX(vs->aVtx), _CONST(0));
537 ureg_MAD(ureg, tmp, _YYYY(vs->aVtx), _CONST(1), ureg_src(tmp));
538 ureg_MAD(ureg, tmp, _ZZZZ(vs->aVtx), _CONST(2), ureg_src(tmp));
539 ureg_MAD(ureg, oPos, _WWWW(vs->aVtx), _CONST(3), ureg_src(tmp));
540 ureg_release_temporary(ureg, tmp);
541
542 if (need_aVtx) {
543 struct ureg_dst aVtx_dst = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
544 ureg_MUL(ureg, aVtx_dst, _XXXX(vs->aVtx), _CONST(4));
545 ureg_MAD(ureg, aVtx_dst, _YYYY(vs->aVtx), _CONST(5), ureg_src(aVtx_dst));
546 ureg_MAD(ureg, aVtx_dst, _ZZZZ(vs->aVtx), _CONST(6), ureg_src(aVtx_dst));
547 ureg_MAD(ureg, aVtx_dst, _WWWW(vs->aVtx), _CONST(7), ureg_src(aVtx_dst));
548 vs->aVtx = ureg_src(aVtx_dst);
549 }
550 if (need_aNrm) {
551 struct ureg_dst aNrm_dst = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
552 ureg_MUL(ureg, aNrm_dst, _XXXX(vs->aNrm), _CONST(16));
553 ureg_MAD(ureg, aNrm_dst, _YYYY(vs->aNrm), _CONST(17), ureg_src(aNrm_dst));
554 ureg_MAD(ureg, aNrm_dst, _ZZZZ(vs->aNrm), _CONST(18), ureg_src(aNrm_dst));
555 if (key->normalizenormals)
556 ureg_normalize3(ureg, aNrm_dst, ureg_src(aNrm_dst));
557 vs->aNrm = ureg_src(aNrm_dst);
558 }
559 }
560
561 /* === Process point size:
562 */
563 if (key->vertexpointsize || key->pointscale) {
564 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
565 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
566 struct ureg_dst tmp_y = ureg_writemask(tmp, TGSI_WRITEMASK_Y);
567 struct ureg_dst tmp_z = ureg_writemask(tmp, TGSI_WRITEMASK_Z);
568 if (key->vertexpointsize) {
569 struct ureg_src cPsz1 = ureg_DECL_constant(ureg, 26);
570 ureg_MAX(ureg, tmp_z, _XXXX(vs->aPsz), _XXXX(cPsz1));
571 ureg_MIN(ureg, tmp_z, _Z(tmp), _YYYY(cPsz1));
572 } else {
573 struct ureg_src cPsz1 = ureg_DECL_constant(ureg, 26);
574 ureg_MOV(ureg, tmp_z, _ZZZZ(cPsz1));
575 }
576
577 if (key->pointscale) {
578 struct ureg_src cPsz1 = ureg_DECL_constant(ureg, 26);
579 struct ureg_src cPsz2 = ureg_DECL_constant(ureg, 27);
580
581 ureg_DP3(ureg, tmp_x, vs->aVtx, vs->aVtx);
582 ureg_RSQ(ureg, tmp_y, _X(tmp));
583 ureg_MUL(ureg, tmp_y, _Y(tmp), _X(tmp));
584 ureg_CMP(ureg, tmp_y, ureg_negate(_Y(tmp)), _Y(tmp), ureg_imm1f(ureg, 0.0f));
585 ureg_MAD(ureg, tmp_x, _Y(tmp), _YYYY(cPsz2), _XXXX(cPsz2));
586 ureg_MAD(ureg, tmp_x, _Y(tmp), _X(tmp), _WWWW(cPsz1));
587 ureg_RSQ(ureg, tmp_x, _X(tmp));
588 ureg_MUL(ureg, tmp_x, _X(tmp), _Z(tmp));
589 ureg_MUL(ureg, tmp_x, _X(tmp), _WWWW(_CONST(100)));
590 ureg_MAX(ureg, tmp_x, _X(tmp), _XXXX(cPsz1));
591 ureg_MIN(ureg, tmp_z, _X(tmp), _YYYY(cPsz1));
592 }
593
594 ureg_MOV(ureg, oPsz, _Z(tmp));
595 ureg_release_temporary(ureg, tmp);
596 }
597
598 for (i = 0; i < 8; ++i) {
599 struct ureg_dst tmp, tmp_x;
600 struct ureg_dst oTex, input_coord, transformed, t;
601 unsigned c, writemask;
602 const unsigned tci = (key->tc_gen >> (i * 3)) & 0x7;
603 const unsigned idx = (key->tc_idx >> (i * 3)) & 0x7;
604 unsigned dim_input = 1 + ((key->tc_dim_input >> (i * 2)) & 0x3);
605 const unsigned dim_output = (key->tc_dim_output >> (i * 3)) & 0x7;
606
607 /* No texture output of index s */
608 if (tci == NINED3DTSS_TCI_DISABLE)
609 continue;
610 oTex = ureg_DECL_output(ureg, texcoord_sn, i);
611 tmp = ureg_DECL_temporary(ureg);
612 tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
613 input_coord = ureg_DECL_temporary(ureg);
614 transformed = ureg_DECL_temporary(ureg);
615
616 /* Get the coordinate */
617 switch (tci) {
618 case NINED3DTSS_TCI_PASSTHRU:
619 /* NINED3DTSS_TCI_PASSTHRU => Use texcoord coming from index idx *
620 * Else the idx is used only to determine wrapping mode. */
621 vs->aTex[idx] = build_vs_add_input(vs, NINE_DECLUSAGE_i(TEXCOORD,idx));
622 ureg_MOV(ureg, input_coord, vs->aTex[idx]);
623 break;
624 case NINED3DTSS_TCI_CAMERASPACENORMAL:
625 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XYZ), vs->aNrm);
626 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
627 dim_input = 4;
628 break;
629 case NINED3DTSS_TCI_CAMERASPACEPOSITION:
630 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XYZ), vs->aVtx);
631 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
632 dim_input = 4;
633 break;
634 case NINED3DTSS_TCI_CAMERASPACEREFLECTIONVECTOR:
635 tmp.WriteMask = TGSI_WRITEMASK_XYZ;
636 ureg_DP3(ureg, tmp_x, vs->aVtx, vs->aNrm);
637 ureg_MUL(ureg, tmp, vs->aNrm, _X(tmp));
638 ureg_ADD(ureg, tmp, ureg_src(tmp), ureg_src(tmp));
639 ureg_SUB(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XYZ), vs->aVtx, ureg_src(tmp));
640 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
641 dim_input = 4;
642 tmp.WriteMask = TGSI_WRITEMASK_XYZW;
643 break;
644 case NINED3DTSS_TCI_SPHEREMAP:
645 assert(!"TODO");
646 break;
647 default:
648 assert(0);
649 break;
650 }
651
652 /* Apply the transformation */
653 /* dim_output == 0 => do not transform the components.
654 * XYZRHW also disables transformation */
655 if (!dim_output || key->position_t) {
656 ureg_release_temporary(ureg, transformed);
657 transformed = input_coord;
658 writemask = TGSI_WRITEMASK_XYZW;
659 } else {
660 for (c = 0; c < dim_output; c++) {
661 t = ureg_writemask(transformed, 1 << c);
662 switch (dim_input) {
663 /* dim_input = 1 2 3: -> we add trailing 1 to input*/
664 case 1: ureg_MAD(ureg, t, _X(input_coord), _XXXX(_CONST(128 + i * 4 + c)), _YYYY(_CONST(128 + i * 4 + c)));
665 break;
666 case 2: ureg_DP2(ureg, t, ureg_src(input_coord), _CONST(128 + i * 4 + c));
667 ureg_ADD(ureg, t, ureg_src(transformed), _ZZZZ(_CONST(128 + i * 4 + c)));
668 break;
669 case 3: ureg_DP3(ureg, t, ureg_src(input_coord), _CONST(128 + i * 4 + c));
670 ureg_ADD(ureg, t, ureg_src(transformed), _WWWW(_CONST(128 + i * 4 + c)));
671 break;
672 case 4: ureg_DP4(ureg, t, ureg_src(input_coord), _CONST(128 + i * 4 + c)); break;
673 default:
674 assert(0);
675 }
676 }
677 writemask = (1 << dim_output) - 1;
678 ureg_release_temporary(ureg, input_coord);
679 }
680
681 ureg_MOV(ureg, ureg_writemask(oTex, writemask), ureg_src(transformed));
682 ureg_release_temporary(ureg, transformed);
683 ureg_release_temporary(ureg, tmp);
684 }
685
686 /* === Lighting:
687 *
688 * DIRECTIONAL: Light at infinite distance, parallel rays, no attenuation.
689 * POINT: Finite distance to scene, divergent rays, isotropic, attenuation.
690 * SPOT: Finite distance, divergent rays, angular dependence, attenuation.
691 *
692 * vec3 normal = normalize(in.Normal * NormalMatrix);
693 * vec3 hitDir = light.direction;
694 * float atten = 1.0;
695 *
696 * if (light.type != DIRECTIONAL)
697 * {
698 * vec3 hitVec = light.position - eyeVertex;
699 * float d = length(hitVec);
700 * hitDir = hitVec / d;
701 * atten = 1 / ((light.atten2 * d + light.atten1) * d + light.atten0);
702 * }
703 *
704 * if (light.type == SPOTLIGHT)
705 * {
706 * float rho = dp3(-hitVec, light.direction);
707 * if (rho < cos(light.phi / 2))
708 * atten = 0;
709 * if (rho < cos(light.theta / 2))
710 * atten *= pow(some_func(rho), light.falloff);
711 * }
712 *
713 * float nDotHit = dp3_sat(normal, hitVec);
714 * float powFact = 0.0;
715 *
716 * if (nDotHit > 0.0)
717 * {
718 * vec3 midVec = normalize(hitDir + eye);
719 * float nDotMid = dp3_sat(normal, midVec);
720 * pFact = pow(nDotMid, material.power);
721 * }
722 *
723 * ambient += light.ambient * atten;
724 * diffuse += light.diffuse * atten * nDotHit;
725 * specular += light.specular * atten * powFact;
726 */
727 if (key->lighting) {
728 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
729 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
730 struct ureg_dst tmp_y = ureg_writemask(tmp, TGSI_WRITEMASK_Y);
731 struct ureg_dst tmp_z = ureg_writemask(tmp, TGSI_WRITEMASK_Z);
732 struct ureg_dst rAtt = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_W);
733 struct ureg_dst rHit = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
734 struct ureg_dst rMid = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
735
736 struct ureg_dst rCtr = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_W);
737
738 struct ureg_dst AL = ureg_writemask(AR, TGSI_WRITEMASK_X);
739
740 /* Light.*.Alpha is not used. */
741 struct ureg_dst rD = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
742 struct ureg_dst rA = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
743 struct ureg_dst rS = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
744
745 struct ureg_src mtlP = _XXXX(MATERIAL_CONST(4));
746
747 struct ureg_src cLKind = _XXXX(LIGHT_CONST(0));
748 struct ureg_src cLAtt0 = _YYYY(LIGHT_CONST(0));
749 struct ureg_src cLAtt1 = _ZZZZ(LIGHT_CONST(0));
750 struct ureg_src cLAtt2 = _WWWW(LIGHT_CONST(0));
751 struct ureg_src cLColD = _XYZW(LIGHT_CONST(1));
752 struct ureg_src cLColS = _XYZW(LIGHT_CONST(2));
753 struct ureg_src cLColA = _XYZW(LIGHT_CONST(3));
754 struct ureg_src cLPos = _XYZW(LIGHT_CONST(4));
755 struct ureg_src cLRng = _WWWW(LIGHT_CONST(4));
756 struct ureg_src cLDir = _XYZW(LIGHT_CONST(5));
757 struct ureg_src cLFOff = _WWWW(LIGHT_CONST(5));
758 struct ureg_src cLTht = _XXXX(LIGHT_CONST(6));
759 struct ureg_src cLPhi = _YYYY(LIGHT_CONST(6));
760 struct ureg_src cLSDiv = _ZZZZ(LIGHT_CONST(6));
761 struct ureg_src cLLast = _WWWW(LIGHT_CONST(7));
762
763 const unsigned loop_label = l++;
764
765 ureg_MOV(ureg, rCtr, ureg_imm1f(ureg, 32.0f)); /* &lightconst(0) */
766 ureg_MOV(ureg, rD, ureg_imm1f(ureg, 0.0f));
767 ureg_MOV(ureg, rA, ureg_imm1f(ureg, 0.0f));
768 ureg_MOV(ureg, rS, ureg_imm1f(ureg, 0.0f));
769 rD = ureg_saturate(rD);
770 rA = ureg_saturate(rA);
771 rS = ureg_saturate(rS);
772
773
774 /* loop management */
775 ureg_BGNLOOP(ureg, &label[loop_label]);
776 ureg_ARL(ureg, AL, _W(rCtr));
777
778 /* if (not DIRECTIONAL light): */
779 ureg_SNE(ureg, tmp_x, cLKind, ureg_imm1f(ureg, D3DLIGHT_DIRECTIONAL));
780 ureg_MOV(ureg, rHit, ureg_negate(cLDir));
781 ureg_MOV(ureg, rAtt, ureg_imm1f(ureg, 1.0f));
782 ureg_IF(ureg, _X(tmp), &label[l++]);
783 {
784 /* hitDir = light.position - eyeVtx
785 * d = length(hitDir)
786 */
787 ureg_SUB(ureg, rHit, cLPos, vs->aVtx);
788 ureg_DP3(ureg, tmp_x, ureg_src(rHit), ureg_src(rHit));
789 ureg_RSQ(ureg, tmp_y, _X(tmp));
790 ureg_MUL(ureg, tmp_x, _X(tmp), _Y(tmp)); /* length */
791
792 /* att = 1.0 / (light.att0 + (light.att1 + light.att2 * d) * d) */
793 ureg_MAD(ureg, rAtt, _X(tmp), cLAtt2, cLAtt1);
794 ureg_MAD(ureg, rAtt, _X(tmp), _W(rAtt), cLAtt0);
795 ureg_RCP(ureg, rAtt, _W(rAtt));
796 /* cut-off if distance exceeds Light.Range */
797 ureg_SLT(ureg, tmp_x, _X(tmp), cLRng);
798 ureg_MUL(ureg, rAtt, _W(rAtt), _X(tmp));
799 }
800 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
801 ureg_ENDIF(ureg);
802
803 /* normalize hitDir */
804 ureg_normalize3(ureg, rHit, ureg_src(rHit));
805
806 /* if (SPOT light) */
807 ureg_SEQ(ureg, tmp_x, cLKind, ureg_imm1f(ureg, D3DLIGHT_SPOT));
808 ureg_IF(ureg, _X(tmp), &label[l++]);
809 {
810 /* rho = dp3(-hitDir, light.spotDir)
811 *
812 * if (rho > light.ctht2) NOTE: 0 <= phi <= pi, 0 <= theta <= phi
813 * spotAtt = 1
814 * else
815 * if (rho <= light.cphi2)
816 * spotAtt = 0
817 * else
818 * spotAtt = (rho - light.cphi2) / (light.ctht2 - light.cphi2) ^ light.falloff
819 */
820 ureg_DP3(ureg, tmp_y, ureg_negate(ureg_src(rHit)), cLDir); /* rho */
821 ureg_SUB(ureg, tmp_x, _Y(tmp), cLPhi);
822 ureg_MUL(ureg, tmp_x, _X(tmp), cLSDiv);
823 ureg_POW(ureg, tmp_x, _X(tmp), cLFOff); /* spotAtten */
824 ureg_SGE(ureg, tmp_z, _Y(tmp), cLTht); /* if inside theta && phi */
825 ureg_SGE(ureg, tmp_y, _Y(tmp), cLPhi); /* if inside phi */
826 ureg_MAD(ureg, ureg_saturate(tmp_x), _X(tmp), _Y(tmp), _Z(tmp));
827 ureg_MUL(ureg, rAtt, _W(rAtt), _X(tmp));
828 }
829 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
830 ureg_ENDIF(ureg);
831
832 /* directional factors, let's not use LIT because of clarity */
833 ureg_DP3(ureg, ureg_saturate(tmp_x), vs->aNrm, ureg_src(rHit));
834 ureg_MOV(ureg, tmp_y, ureg_imm1f(ureg, 0.0f));
835 ureg_IF(ureg, _X(tmp), &label[l++]);
836 {
837 /* midVec = normalize(hitDir + eyeDir) */
838 if (key->localviewer) {
839 ureg_normalize3(ureg, rMid, vs->aVtx);
840 ureg_SUB(ureg, rMid, ureg_src(rHit), ureg_src(rMid));
841 } else {
842 ureg_SUB(ureg, rMid, ureg_src(rHit), ureg_imm3f(ureg, 0.0f, 0.0f, 1.0f));
843 }
844 ureg_normalize3(ureg, rMid, ureg_src(rMid));
845 ureg_DP3(ureg, ureg_saturate(tmp_y), vs->aNrm, ureg_src(rMid));
846 ureg_POW(ureg, tmp_y, _Y(tmp), mtlP);
847
848 ureg_MUL(ureg, tmp_x, _W(rAtt), _X(tmp)); /* dp3(normal,hitDir) * att */
849 ureg_MUL(ureg, tmp_y, _W(rAtt), _Y(tmp)); /* power factor * att */
850 ureg_MAD(ureg, rD, cLColD, _X(tmp), ureg_src(rD)); /* accumulate diffuse */
851 ureg_MAD(ureg, rS, cLColS, _Y(tmp), ureg_src(rS)); /* accumulate specular */
852 }
853 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
854 ureg_ENDIF(ureg);
855
856 ureg_MAD(ureg, rA, cLColA, _W(rAtt), ureg_src(rA)); /* accumulate ambient */
857
858 /* break if this was the last light */
859 ureg_IF(ureg, cLLast, &label[l++]);
860 ureg_BRK(ureg);
861 ureg_ENDIF(ureg);
862 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
863
864 ureg_ADD(ureg, rCtr, _W(rCtr), ureg_imm1f(ureg, 8.0f));
865 ureg_fixup_label(ureg, label[loop_label], ureg_get_instruction_number(ureg));
866 ureg_ENDLOOP(ureg, &label[loop_label]);
867
868 /* Set alpha factors of illumination to 1.0 for the multiplications. */
869 rD.WriteMask = TGSI_WRITEMASK_W; rD.Saturate = 0;
870 rS.WriteMask = TGSI_WRITEMASK_W; rS.Saturate = 0;
871 rA.WriteMask = TGSI_WRITEMASK_W; rA.Saturate = 0;
872 ureg_MOV(ureg, rD, ureg_imm1f(ureg, 1.0f));
873 ureg_MOV(ureg, rS, ureg_imm1f(ureg, 1.0f));
874
875 /* Apply to material:
876 *
877 * oCol[0] = (material.emissive + material.ambient * rs.ambient) +
878 * material.ambient * ambient +
879 * material.diffuse * diffuse +
880 * oCol[1] = material.specular * specular;
881 */
882 if (key->mtl_emissive == 0 && key->mtl_ambient == 0) {
883 ureg_MOV(ureg, rA, ureg_imm1f(ureg, 1.0f));
884 ureg_MAD(ureg, tmp, ureg_src(rA), vs->mtlA, _CONST(19));
885 } else {
886 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(rA), _CONST(25));
887 ureg_MAD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), vs->mtlA, ureg_src(tmp), vs->mtlE);
888 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_W ), vs->mtlA, vs->mtlE);
889 }
890
891 if (key->specular_enable) {
892 /* add oCol[1] to oCol[0] */
893 ureg_MAD(ureg, tmp, ureg_src(rD), vs->mtlD, ureg_src(tmp));
894 ureg_MAD(ureg, oCol[0], ureg_src(rS), vs->mtlS, ureg_src(tmp));
895 } else {
896 ureg_MAD(ureg, oCol[0], ureg_src(rD), vs->mtlD, ureg_src(tmp));
897 }
898 ureg_MUL(ureg, oCol[1], ureg_src(rS), vs->mtlS);
899 ureg_release_temporary(ureg, rAtt);
900 ureg_release_temporary(ureg, rHit);
901 ureg_release_temporary(ureg, rMid);
902 ureg_release_temporary(ureg, rCtr);
903 ureg_release_temporary(ureg, rD);
904 ureg_release_temporary(ureg, rA);
905 ureg_release_temporary(ureg, rS);
906 ureg_release_temporary(ureg, rAtt);
907 ureg_release_temporary(ureg, tmp);
908 } else
909 /* COLOR */
910 if (key->darkness) {
911 if (key->mtl_emissive == 0 && key->mtl_ambient == 0) {
912 ureg_MAD(ureg, oCol[0], vs->mtlD, ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 1.0f), _CONST(19));
913 } else {
914 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
915 ureg_MAD(ureg, ureg_writemask(oCol[0], TGSI_WRITEMASK_XYZ), vs->mtlA, _CONST(25), vs->mtlE);
916 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_W), vs->mtlA, vs->mtlE);
917 ureg_ADD(ureg, ureg_writemask(oCol[0], TGSI_WRITEMASK_W), vs->mtlD, _W(tmp));
918 ureg_release_temporary(ureg, tmp);
919 }
920 ureg_MUL(ureg, oCol[1], ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 1.0f), vs->mtlS);
921 } else {
922 ureg_MOV(ureg, oCol[0], vs->aCol[0]);
923 ureg_MOV(ureg, oCol[1], vs->aCol[1]);
924 }
925
926 /* === Process fog.
927 *
928 * exp(x) = ex2(log2(e) * x)
929 */
930 if (key->fog_mode) {
931 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
932 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
933 struct ureg_dst tmp_z = ureg_writemask(tmp, TGSI_WRITEMASK_Z);
934 if (key->fog_range) {
935 ureg_DP3(ureg, tmp_x, vs->aVtx, vs->aVtx);
936 ureg_RSQ(ureg, tmp_z, _X(tmp));
937 ureg_MUL(ureg, tmp_z, _Z(tmp), _X(tmp));
938 } else {
939 ureg_MOV(ureg, tmp_z, ureg_abs(_ZZZZ(vs->aVtx)));
940 }
941
942 if (key->fog_mode == D3DFOG_EXP) {
943 ureg_MUL(ureg, tmp_x, _Z(tmp), _ZZZZ(_CONST(28)));
944 ureg_MUL(ureg, tmp_x, _X(tmp), ureg_imm1f(ureg, -1.442695f));
945 ureg_EX2(ureg, tmp_x, _X(tmp));
946 } else
947 if (key->fog_mode == D3DFOG_EXP2) {
948 ureg_MUL(ureg, tmp_x, _Z(tmp), _ZZZZ(_CONST(28)));
949 ureg_MUL(ureg, tmp_x, _X(tmp), _X(tmp));
950 ureg_MUL(ureg, tmp_x, _X(tmp), ureg_imm1f(ureg, -1.442695f));
951 ureg_EX2(ureg, tmp_x, _X(tmp));
952 } else
953 if (key->fog_mode == D3DFOG_LINEAR) {
954 ureg_SUB(ureg, tmp_x, _XXXX(_CONST(28)), _Z(tmp));
955 ureg_MUL(ureg, ureg_saturate(tmp_x), _X(tmp), _YYYY(_CONST(28)));
956 }
957 ureg_MOV(ureg, oFog, _X(tmp));
958 ureg_release_temporary(ureg, tmp);
959 } else if (key->fog && !(key->passthrough & (1 << NINE_DECLUSAGE_FOG))) {
960 ureg_MOV(ureg, oFog, ureg_scalar(vs->aCol[1], TGSI_SWIZZLE_W));
961 }
962
963 if (key->passthrough & (1 << NINE_DECLUSAGE_BLENDWEIGHT)) {
964 struct ureg_src input;
965 struct ureg_dst output;
966 input = vs->aWgt;
967 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 18);
968 ureg_MOV(ureg, output, input);
969 }
970 if (key->passthrough & (1 << NINE_DECLUSAGE_BLENDINDICES)) {
971 struct ureg_src input;
972 struct ureg_dst output;
973 input = vs->aInd;
974 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 19);
975 ureg_MOV(ureg, output, input);
976 }
977 if (key->passthrough & (1 << NINE_DECLUSAGE_NORMAL)) {
978 struct ureg_src input;
979 struct ureg_dst output;
980 input = vs->aNrm;
981 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 20);
982 ureg_MOV(ureg, output, input);
983 }
984 if (key->passthrough & (1 << NINE_DECLUSAGE_TANGENT)) {
985 struct ureg_src input;
986 struct ureg_dst output;
987 input = build_vs_add_input(vs, NINE_DECLUSAGE_TANGENT);
988 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 21);
989 ureg_MOV(ureg, output, input);
990 }
991 if (key->passthrough & (1 << NINE_DECLUSAGE_BINORMAL)) {
992 struct ureg_src input;
993 struct ureg_dst output;
994 input = build_vs_add_input(vs, NINE_DECLUSAGE_BINORMAL);
995 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 22);
996 ureg_MOV(ureg, output, input);
997 }
998 if (key->passthrough & (1 << NINE_DECLUSAGE_FOG)) {
999 struct ureg_src input;
1000 struct ureg_dst output;
1001 input = build_vs_add_input(vs, NINE_DECLUSAGE_FOG);
1002 input = ureg_scalar(input, TGSI_SWIZZLE_X);
1003 output = oFog;
1004 ureg_MOV(ureg, output, input);
1005 }
1006 if (key->passthrough & (1 << NINE_DECLUSAGE_DEPTH)) {
1007 (void) 0; /* TODO: replace z of position output ? */
1008 }
1009
1010
1011 if (key->position_t && device->driver_caps.window_space_position_support)
1012 ureg_property(ureg, TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION, TRUE);
1013
1014 ureg_END(ureg);
1015 nine_ureg_tgsi_dump(ureg, FALSE);
1016 return ureg_create_shader_and_destroy(ureg, device->pipe);
1017 }
1018
1019 /* PS FF constants layout:
1020 *
1021 * CONST[ 0.. 7] stage[i].D3DTSS_CONSTANT
1022 * CONST[ 8..15].x___ stage[i].D3DTSS_BUMPENVMAT00
1023 * CONST[ 8..15]._y__ stage[i].D3DTSS_BUMPENVMAT01
1024 * CONST[ 8..15].__z_ stage[i].D3DTSS_BUMPENVMAT10
1025 * CONST[ 8..15].___w stage[i].D3DTSS_BUMPENVMAT11
1026 * CONST[16..19].x_z_ stage[i].D3DTSS_BUMPENVLSCALE
1027 * CONST[17..19]._y_w stage[i].D3DTSS_BUMPENVLOFFSET
1028 *
1029 * CONST[20] D3DRS_TEXTUREFACTOR
1030 * CONST[21] D3DRS_FOGCOLOR
1031 * CONST[22].x___ RS.FogEnd
1032 * CONST[22]._y__ 1.0f / (RS.FogEnd - RS.FogStart)
1033 * CONST[22].__z_ RS.FogDensity
1034 */
1035 struct ps_build_ctx
1036 {
1037 struct ureg_program *ureg;
1038
1039 struct ureg_src vC[2]; /* DIFFUSE, SPECULAR */
1040 struct ureg_src vT[8]; /* TEXCOORD[i] */
1041 struct ureg_dst r[6]; /* TEMPs */
1042 struct ureg_dst rCur; /* D3DTA_CURRENT */
1043 struct ureg_dst rMod;
1044 struct ureg_src rCurSrc;
1045 struct ureg_dst rTmp; /* D3DTA_TEMP */
1046 struct ureg_src rTmpSrc;
1047 struct ureg_dst rTex;
1048 struct ureg_src rTexSrc;
1049 struct ureg_src cBEM[8];
1050 struct ureg_src s[8];
1051
1052 struct {
1053 unsigned index;
1054 unsigned index_pre_mod;
1055 unsigned num_regs;
1056 } stage;
1057 };
1058
1059 static struct ureg_src
1060 ps_get_ts_arg(struct ps_build_ctx *ps, unsigned ta)
1061 {
1062 struct ureg_src reg;
1063
1064 switch (ta & D3DTA_SELECTMASK) {
1065 case D3DTA_CONSTANT:
1066 reg = ureg_DECL_constant(ps->ureg, ps->stage.index);
1067 break;
1068 case D3DTA_CURRENT:
1069 reg = (ps->stage.index == ps->stage.index_pre_mod) ? ureg_src(ps->rMod) : ps->rCurSrc;
1070 break;
1071 case D3DTA_DIFFUSE:
1072 reg = ureg_DECL_fs_input(ps->ureg, TGSI_SEMANTIC_COLOR, 0, TGSI_INTERPOLATE_COLOR);
1073 break;
1074 case D3DTA_SPECULAR:
1075 reg = ureg_DECL_fs_input(ps->ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_COLOR);
1076 break;
1077 case D3DTA_TEMP:
1078 reg = ps->rTmpSrc;
1079 break;
1080 case D3DTA_TEXTURE:
1081 reg = ps->rTexSrc;
1082 break;
1083 case D3DTA_TFACTOR:
1084 reg = ureg_DECL_constant(ps->ureg, 20);
1085 break;
1086 default:
1087 assert(0);
1088 reg = ureg_src_undef();
1089 break;
1090 }
1091 if (ta & D3DTA_COMPLEMENT) {
1092 struct ureg_dst dst = ps->r[ps->stage.num_regs++];
1093 ureg_SUB(ps->ureg, dst, ureg_imm1f(ps->ureg, 1.0f), reg);
1094 reg = ureg_src(dst);
1095 }
1096 if (ta & D3DTA_ALPHAREPLICATE)
1097 reg = _WWWW(reg);
1098 return reg;
1099 }
1100
1101 static struct ureg_dst
1102 ps_get_ts_dst(struct ps_build_ctx *ps, unsigned ta)
1103 {
1104 assert(!(ta & (D3DTA_COMPLEMENT | D3DTA_ALPHAREPLICATE)));
1105
1106 switch (ta & D3DTA_SELECTMASK) {
1107 case D3DTA_CURRENT:
1108 return ps->rCur;
1109 case D3DTA_TEMP:
1110 return ps->rTmp;
1111 default:
1112 assert(0);
1113 return ureg_dst_undef();
1114 }
1115 }
1116
1117 static uint8_t ps_d3dtop_args_mask(D3DTEXTUREOP top)
1118 {
1119 switch (top) {
1120 case D3DTOP_DISABLE:
1121 return 0x0;
1122 case D3DTOP_SELECTARG1:
1123 case D3DTOP_PREMODULATE:
1124 return 0x2;
1125 case D3DTOP_SELECTARG2:
1126 return 0x4;
1127 case D3DTOP_MULTIPLYADD:
1128 case D3DTOP_LERP:
1129 return 0x7;
1130 default:
1131 return 0x6;
1132 }
1133 }
1134
1135 static inline boolean
1136 is_MOV_no_op(struct ureg_dst dst, struct ureg_src src)
1137 {
1138 return !dst.WriteMask ||
1139 (dst.File == src.File &&
1140 dst.Index == src.Index &&
1141 !dst.Indirect &&
1142 !dst.Saturate &&
1143 !src.Indirect &&
1144 !src.Negate &&
1145 !src.Absolute &&
1146 (!(dst.WriteMask & TGSI_WRITEMASK_X) || (src.SwizzleX == TGSI_SWIZZLE_X)) &&
1147 (!(dst.WriteMask & TGSI_WRITEMASK_Y) || (src.SwizzleY == TGSI_SWIZZLE_Y)) &&
1148 (!(dst.WriteMask & TGSI_WRITEMASK_Z) || (src.SwizzleZ == TGSI_SWIZZLE_Z)) &&
1149 (!(dst.WriteMask & TGSI_WRITEMASK_W) || (src.SwizzleW == TGSI_SWIZZLE_W)));
1150
1151 }
1152
1153 static void
1154 ps_do_ts_op(struct ps_build_ctx *ps, unsigned top, struct ureg_dst dst, struct ureg_src *arg)
1155 {
1156 struct ureg_program *ureg = ps->ureg;
1157 struct ureg_dst tmp = ps->r[ps->stage.num_regs];
1158 struct ureg_dst tmp2 = ps->r[ps->stage.num_regs+1];
1159 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
1160
1161 tmp.WriteMask = dst.WriteMask;
1162
1163 if (top != D3DTOP_SELECTARG1 && top != D3DTOP_SELECTARG2 &&
1164 top != D3DTOP_MODULATE && top != D3DTOP_PREMODULATE &&
1165 top != D3DTOP_BLENDDIFFUSEALPHA && top != D3DTOP_BLENDTEXTUREALPHA &&
1166 top != D3DTOP_BLENDFACTORALPHA && top != D3DTOP_BLENDCURRENTALPHA &&
1167 top != D3DTOP_BUMPENVMAP && top != D3DTOP_BUMPENVMAPLUMINANCE &&
1168 top != D3DTOP_LERP)
1169 dst = ureg_saturate(dst);
1170
1171 switch (top) {
1172 case D3DTOP_SELECTARG1:
1173 if (!is_MOV_no_op(dst, arg[1]))
1174 ureg_MOV(ureg, dst, arg[1]);
1175 break;
1176 case D3DTOP_SELECTARG2:
1177 if (!is_MOV_no_op(dst, arg[2]))
1178 ureg_MOV(ureg, dst, arg[2]);
1179 break;
1180 case D3DTOP_MODULATE:
1181 ureg_MUL(ureg, dst, arg[1], arg[2]);
1182 break;
1183 case D3DTOP_MODULATE2X:
1184 ureg_MUL(ureg, tmp, arg[1], arg[2]);
1185 ureg_ADD(ureg, dst, ureg_src(tmp), ureg_src(tmp));
1186 break;
1187 case D3DTOP_MODULATE4X:
1188 ureg_MUL(ureg, tmp, arg[1], arg[2]);
1189 ureg_MUL(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 4.0f));
1190 break;
1191 case D3DTOP_ADD:
1192 ureg_ADD(ureg, dst, arg[1], arg[2]);
1193 break;
1194 case D3DTOP_ADDSIGNED:
1195 ureg_ADD(ureg, tmp, arg[1], arg[2]);
1196 ureg_SUB(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 0.5f));
1197 break;
1198 case D3DTOP_ADDSIGNED2X:
1199 ureg_ADD(ureg, tmp, arg[1], arg[2]);
1200 ureg_MAD(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 2.0f), ureg_imm1f(ureg, -1.0f));
1201 break;
1202 case D3DTOP_SUBTRACT:
1203 ureg_SUB(ureg, dst, arg[1], arg[2]);
1204 break;
1205 case D3DTOP_ADDSMOOTH:
1206 ureg_SUB(ureg, tmp, ureg_imm1f(ureg, 1.0f), arg[1]);
1207 ureg_MAD(ureg, dst, ureg_src(tmp), arg[2], arg[1]);
1208 break;
1209 case D3DTOP_BLENDDIFFUSEALPHA:
1210 ureg_LRP(ureg, dst, _WWWW(ps->vC[0]), arg[1], arg[2]);
1211 break;
1212 case D3DTOP_BLENDTEXTUREALPHA:
1213 /* XXX: alpha taken from previous stage, texture or result ? */
1214 ureg_LRP(ureg, dst, _W(ps->rTex), arg[1], arg[2]);
1215 break;
1216 case D3DTOP_BLENDFACTORALPHA:
1217 ureg_LRP(ureg, dst, _WWWW(_CONST(20)), arg[1], arg[2]);
1218 break;
1219 case D3DTOP_BLENDTEXTUREALPHAPM:
1220 ureg_SUB(ureg, tmp_x, ureg_imm1f(ureg, 1.0f), _W(ps->rTex));
1221 ureg_MAD(ureg, dst, arg[2], _X(tmp), arg[1]);
1222 break;
1223 case D3DTOP_BLENDCURRENTALPHA:
1224 ureg_LRP(ureg, dst, _WWWW(ps->rCurSrc), arg[1], arg[2]);
1225 break;
1226 case D3DTOP_PREMODULATE:
1227 ureg_MOV(ureg, dst, arg[1]);
1228 ps->stage.index_pre_mod = ps->stage.index + 1;
1229 break;
1230 case D3DTOP_MODULATEALPHA_ADDCOLOR:
1231 ureg_MAD(ureg, dst, _WWWW(arg[1]), arg[2], arg[1]);
1232 break;
1233 case D3DTOP_MODULATECOLOR_ADDALPHA:
1234 ureg_MAD(ureg, dst, arg[1], arg[2], _WWWW(arg[1]));
1235 break;
1236 case D3DTOP_MODULATEINVALPHA_ADDCOLOR:
1237 ureg_SUB(ureg, tmp_x, ureg_imm1f(ureg, 1.0f), _WWWW(arg[1]));
1238 ureg_MAD(ureg, dst, _X(tmp), arg[2], arg[1]);
1239 break;
1240 case D3DTOP_MODULATEINVCOLOR_ADDALPHA:
1241 ureg_SUB(ureg, tmp, ureg_imm1f(ureg, 1.0f), arg[1]);
1242 ureg_MAD(ureg, dst, ureg_src(tmp), arg[2], _WWWW(arg[1]));
1243 break;
1244 case D3DTOP_BUMPENVMAP:
1245 break;
1246 case D3DTOP_BUMPENVMAPLUMINANCE:
1247 break;
1248 case D3DTOP_DOTPRODUCT3:
1249 ureg_SUB(ureg, tmp, arg[1], ureg_imm4f(ureg,0.5,0.5,0.5,0.5));
1250 ureg_SUB(ureg, tmp2, arg[2] , ureg_imm4f(ureg,0.5,0.5,0.5,0.5));
1251 ureg_DP3(ureg, tmp, ureg_src(tmp), ureg_src(tmp2));
1252 ureg_MUL(ureg, ureg_saturate(dst), ureg_src(tmp), ureg_imm4f(ureg,4.0,4.0,4.0,4.0));
1253 break;
1254 case D3DTOP_MULTIPLYADD:
1255 ureg_MAD(ureg, dst, arg[1], arg[2], arg[0]);
1256 break;
1257 case D3DTOP_LERP:
1258 ureg_LRP(ureg, dst, arg[0], arg[1], arg[2]);
1259 break;
1260 case D3DTOP_DISABLE:
1261 /* no-op ? */
1262 break;
1263 default:
1264 assert(!"invalid D3DTOP");
1265 break;
1266 }
1267 }
1268
1269 static void *
1270 nine_ff_build_ps(struct NineDevice9 *device, struct nine_ff_ps_key *key)
1271 {
1272 struct ps_build_ctx ps;
1273 struct ureg_program *ureg = ureg_create(PIPE_SHADER_FRAGMENT);
1274 struct ureg_dst oCol;
1275 unsigned i, s;
1276 const unsigned texcoord_sn = get_texcoord_sn(device->screen);
1277
1278 memset(&ps, 0, sizeof(ps));
1279 ps.ureg = ureg;
1280 ps.stage.index_pre_mod = -1;
1281
1282 ps.vC[0] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 0, TGSI_INTERPOLATE_COLOR);
1283
1284 /* Declare all TEMPs we might need, serious drivers have a register allocator. */
1285 for (i = 0; i < ARRAY_SIZE(ps.r); ++i)
1286 ps.r[i] = ureg_DECL_temporary(ureg);
1287 ps.rCur = ps.r[0];
1288 ps.rTmp = ps.r[1];
1289 ps.rTex = ps.r[2];
1290 ps.rCurSrc = ureg_src(ps.rCur);
1291 ps.rTmpSrc = ureg_src(ps.rTmp);
1292 ps.rTexSrc = ureg_src(ps.rTex);
1293
1294 for (s = 0; s < 8; ++s) {
1295 ps.s[s] = ureg_src_undef();
1296
1297 if (key->ts[s].colorop != D3DTOP_DISABLE) {
1298 if (key->ts[s].colorarg0 == D3DTA_SPECULAR ||
1299 key->ts[s].colorarg1 == D3DTA_SPECULAR ||
1300 key->ts[s].colorarg2 == D3DTA_SPECULAR)
1301 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_COLOR);
1302
1303 if (key->ts[s].colorarg0 == D3DTA_TEXTURE ||
1304 key->ts[s].colorarg1 == D3DTA_TEXTURE ||
1305 key->ts[s].colorarg2 == D3DTA_TEXTURE) {
1306 ps.s[s] = ureg_DECL_sampler(ureg, s);
1307 ps.vT[s] = ureg_DECL_fs_input(ureg, texcoord_sn, s, TGSI_INTERPOLATE_PERSPECTIVE);
1308 }
1309 if (s && (key->ts[s - 1].colorop == D3DTOP_PREMODULATE ||
1310 key->ts[s - 1].alphaop == D3DTOP_PREMODULATE))
1311 ps.s[s] = ureg_DECL_sampler(ureg, s);
1312 }
1313
1314 if (key->ts[s].alphaop != D3DTOP_DISABLE) {
1315 if (key->ts[s].alphaarg0 == D3DTA_SPECULAR ||
1316 key->ts[s].alphaarg1 == D3DTA_SPECULAR ||
1317 key->ts[s].alphaarg2 == D3DTA_SPECULAR)
1318 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_COLOR);
1319
1320 if (key->ts[s].alphaarg0 == D3DTA_TEXTURE ||
1321 key->ts[s].alphaarg1 == D3DTA_TEXTURE ||
1322 key->ts[s].alphaarg2 == D3DTA_TEXTURE) {
1323 ps.s[s] = ureg_DECL_sampler(ureg, s);
1324 ps.vT[s] = ureg_DECL_fs_input(ureg, texcoord_sn, s, TGSI_INTERPOLATE_PERSPECTIVE);
1325 }
1326 }
1327 }
1328 if (key->specular)
1329 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_COLOR);
1330
1331 oCol = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
1332
1333 if (key->ts[0].colorop == D3DTOP_DISABLE &&
1334 key->ts[0].alphaop == D3DTOP_DISABLE)
1335 ureg_MOV(ureg, ps.rCur, ps.vC[0]);
1336 /* Or is it undefined then ? */
1337
1338 /* Run stages.
1339 */
1340 for (s = 0; s < 8; ++s) {
1341 unsigned colorarg[3];
1342 unsigned alphaarg[3];
1343 const uint8_t used_c = ps_d3dtop_args_mask(key->ts[s].colorop);
1344 const uint8_t used_a = ps_d3dtop_args_mask(key->ts[s].alphaop);
1345 struct ureg_dst dst;
1346 struct ureg_src arg[3];
1347
1348 if (key->ts[s].colorop == D3DTOP_DISABLE &&
1349 key->ts[s].alphaop == D3DTOP_DISABLE)
1350 continue;
1351 ps.stage.index = s;
1352 ps.stage.num_regs = 3;
1353
1354 DBG("STAGE[%u]: colorop=%s alphaop=%s\n", s,
1355 nine_D3DTOP_to_str(key->ts[s].colorop),
1356 nine_D3DTOP_to_str(key->ts[s].alphaop));
1357
1358 if (!ureg_src_is_undef(ps.s[s])) {
1359 unsigned target;
1360 struct ureg_src texture_coord = ps.vT[s];
1361 struct ureg_dst delta;
1362 switch (key->ts[s].textarget) {
1363 case 0: target = TGSI_TEXTURE_1D; break;
1364 case 1: target = TGSI_TEXTURE_2D; break;
1365 case 2: target = TGSI_TEXTURE_3D; break;
1366 case 3: target = TGSI_TEXTURE_CUBE; break;
1367 /* this is a 2 bit bitfield, do I really need a default case ? */
1368 }
1369
1370 /* Modify coordinates */
1371 if (s >= 1 &&
1372 (key->ts[s-1].colorop == D3DTOP_BUMPENVMAP ||
1373 key->ts[s-1].colorop == D3DTOP_BUMPENVMAPLUMINANCE)) {
1374 delta = ureg_DECL_temporary(ureg);
1375 /* Du' = D3DTSS_BUMPENVMAT00(stage s-1)*t(s-1)R + D3DTSS_BUMPENVMAT10(stage s-1)*t(s-1)G */
1376 ureg_MUL(ureg, ureg_writemask(delta, TGSI_WRITEMASK_X), _X(ps.rTex), _XXXX(_CONST(8 + s - 1)));
1377 ureg_MAD(ureg, ureg_writemask(delta, TGSI_WRITEMASK_X), _Y(ps.rTex), _ZZZZ(_CONST(8 + s - 1)), ureg_src(delta));
1378 /* Dv' = D3DTSS_BUMPENVMAT01(stage s-1)*t(s-1)R + D3DTSS_BUMPENVMAT11(stage s-1)*t(s-1)G */
1379 ureg_MUL(ureg, ureg_writemask(delta, TGSI_WRITEMASK_Y), _X(ps.rTex), _YYYY(_CONST(8 + s - 1)));
1380 ureg_MAD(ureg, ureg_writemask(delta, TGSI_WRITEMASK_Y), _Y(ps.rTex), _WWWW(_CONST(8 + s - 1)), ureg_src(delta));
1381 texture_coord = ureg_src(ureg_DECL_temporary(ureg));
1382 ureg_MOV(ureg, ureg_writemask(ureg_dst(texture_coord), ureg_dst(ps.vT[s]).WriteMask), ps.vT[s]);
1383 ureg_ADD(ureg, ureg_writemask(ureg_dst(texture_coord), TGSI_WRITEMASK_XY), texture_coord, ureg_src(delta));
1384 /* Prepare luminance multiplier
1385 * t(s)RGBA = t(s)RGBA * clamp[(t(s-1)B * D3DTSS_BUMPENVLSCALE(stage s-1)) + D3DTSS_BUMPENVLOFFSET(stage s-1)] */
1386 if (key->ts[s-1].colorop == D3DTOP_BUMPENVMAPLUMINANCE) {
1387 struct ureg_src bumpenvlscale = ((s-1) & 1) ? _ZZZZ(_CONST(16 + (s-1) / 2)) : _XXXX(_CONST(16 + (s-1) / 2));
1388 struct ureg_src bumpenvloffset = ((s-1) & 1) ? _WWWW(_CONST(16 + (s-1) / 2)) : _YYYY(_CONST(16 + (s-1) / 2));
1389
1390 ureg_MAD(ureg, ureg_saturate(ureg_writemask(delta, TGSI_WRITEMASK_X)), _Z(ps.rTex), bumpenvlscale, bumpenvloffset);
1391 }
1392 }
1393 if (key->projected & (3 << (s *2))) {
1394 unsigned dim = 1 + ((key->projected >> (2 * s)) & 3);
1395 if (dim == 4)
1396 ureg_TXP(ureg, ps.rTex, target, texture_coord, ps.s[s]);
1397 else {
1398 ureg_RCP(ureg, ureg_writemask(ps.rTmp, TGSI_WRITEMASK_X), ureg_scalar(texture_coord, dim-1));
1399 ureg_MUL(ureg, ps.rTmp, _XXXX(ps.rTmpSrc), texture_coord);
1400 ureg_TEX(ureg, ps.rTex, target, ps.rTmpSrc, ps.s[s]);
1401 }
1402 } else {
1403 ureg_TEX(ureg, ps.rTex, target, texture_coord, ps.s[s]);
1404 }
1405 if (s >= 1 && key->ts[s-1].colorop == D3DTOP_BUMPENVMAPLUMINANCE)
1406 ureg_MUL(ureg, ps.rTex, ureg_src(ps.rTex), _X(delta));
1407 }
1408
1409 if (((s == 0 && key->ts[0].colorop != D3DTOP_BUMPENVMAP &&
1410 key->ts[0].colorop != D3DTOP_BUMPENVMAPLUMINANCE) ||
1411 (s == 1 &&
1412 (key->ts[0].colorop == D3DTOP_BUMPENVMAP ||
1413 key->ts[0].colorop == D3DTOP_BUMPENVMAPLUMINANCE)))&&
1414 (key->ts[s].resultarg != 0 /* not current */ ||
1415 key->ts[s].colorop == D3DTOP_DISABLE ||
1416 key->ts[s].alphaop == D3DTOP_DISABLE ||
1417 key->ts[s].colorop == D3DTOP_BLENDCURRENTALPHA ||
1418 key->ts[s].alphaop == D3DTOP_BLENDCURRENTALPHA ||
1419 key->ts[s].colorarg0 == D3DTA_CURRENT ||
1420 key->ts[s].colorarg1 == D3DTA_CURRENT ||
1421 key->ts[s].colorarg2 == D3DTA_CURRENT ||
1422 key->ts[s].alphaarg0 == D3DTA_CURRENT ||
1423 key->ts[s].alphaarg1 == D3DTA_CURRENT ||
1424 key->ts[s].alphaarg2 == D3DTA_CURRENT)) {
1425 /* Initialize D3DTA_CURRENT.
1426 * (Yes we can do this before the loop but not until
1427 * NVE4 has an instruction scheduling pass.)
1428 */
1429 ureg_MOV(ureg, ps.rCur, ps.vC[0]);
1430 }
1431
1432 if (key->ts[s].colorop == D3DTOP_BUMPENVMAP ||
1433 key->ts[s].colorop == D3DTOP_BUMPENVMAPLUMINANCE)
1434 continue;
1435
1436 dst = ps_get_ts_dst(&ps, key->ts[s].resultarg ? D3DTA_TEMP : D3DTA_CURRENT);
1437
1438 if (ps.stage.index_pre_mod == ps.stage.index) {
1439 ps.rMod = ps.r[ps.stage.num_regs++];
1440 ureg_MUL(ureg, ps.rMod, ps.rCurSrc, ps.rTexSrc);
1441 }
1442
1443 colorarg[0] = (key->ts[s].colorarg0 | ((key->colorarg_b4[0] >> s) << 4) | ((key->colorarg_b5[0] >> s) << 5)) & 0x3f;
1444 colorarg[1] = (key->ts[s].colorarg1 | ((key->colorarg_b4[1] >> s) << 4) | ((key->colorarg_b5[1] >> s) << 5)) & 0x3f;
1445 colorarg[2] = (key->ts[s].colorarg2 | ((key->colorarg_b4[2] >> s) << 4) | ((key->colorarg_b5[2] >> s) << 5)) & 0x3f;
1446 alphaarg[0] = (key->ts[s].alphaarg0 | ((key->alphaarg_b4[0] >> s) << 4)) & 0x1f;
1447 alphaarg[1] = (key->ts[s].alphaarg1 | ((key->alphaarg_b4[1] >> s) << 4)) & 0x1f;
1448 alphaarg[2] = (key->ts[s].alphaarg2 | ((key->alphaarg_b4[2] >> s) << 4)) & 0x1f;
1449
1450 if (key->ts[s].colorop != key->ts[s].alphaop ||
1451 colorarg[0] != alphaarg[0] ||
1452 colorarg[1] != alphaarg[1] ||
1453 colorarg[2] != alphaarg[2])
1454 dst.WriteMask = TGSI_WRITEMASK_XYZ;
1455
1456 /* Special DOTPRODUCT behaviour (see wine tests) */
1457 if (key->ts[s].colorop == D3DTOP_DOTPRODUCT3)
1458 dst.WriteMask = TGSI_WRITEMASK_XYZW;
1459
1460 if (used_c & 0x1) arg[0] = ps_get_ts_arg(&ps, colorarg[0]);
1461 if (used_c & 0x2) arg[1] = ps_get_ts_arg(&ps, colorarg[1]);
1462 if (used_c & 0x4) arg[2] = ps_get_ts_arg(&ps, colorarg[2]);
1463 ps_do_ts_op(&ps, key->ts[s].colorop, dst, arg);
1464
1465 if (dst.WriteMask != TGSI_WRITEMASK_XYZW) {
1466 dst.WriteMask = TGSI_WRITEMASK_W;
1467
1468 if (used_a & 0x1) arg[0] = ps_get_ts_arg(&ps, alphaarg[0]);
1469 if (used_a & 0x2) arg[1] = ps_get_ts_arg(&ps, alphaarg[1]);
1470 if (used_a & 0x4) arg[2] = ps_get_ts_arg(&ps, alphaarg[2]);
1471 ps_do_ts_op(&ps, key->ts[s].alphaop, dst, arg);
1472 }
1473 }
1474
1475 if (key->specular)
1476 ureg_ADD(ureg, ps.rCur, ps.rCurSrc, ps.vC[1]);
1477
1478 /* Fog.
1479 */
1480 if (key->fog_mode) {
1481 struct ureg_src vPos;
1482 if (device->screen->get_param(device->screen,
1483 PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL)) {
1484 vPos = ureg_DECL_system_value(ureg, TGSI_SEMANTIC_POSITION, 0);
1485 } else {
1486 vPos = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_POSITION, 0,
1487 TGSI_INTERPOLATE_LINEAR);
1488 }
1489
1490 struct ureg_dst rFog = ureg_writemask(ps.rTmp, TGSI_WRITEMASK_X);
1491 if (key->fog_mode == D3DFOG_EXP) {
1492 ureg_MUL(ureg, rFog, _ZZZZ(vPos), _ZZZZ(_CONST(22)));
1493 ureg_MUL(ureg, rFog, _X(rFog), ureg_imm1f(ureg, -1.442695f));
1494 ureg_EX2(ureg, rFog, _X(rFog));
1495 } else
1496 if (key->fog_mode == D3DFOG_EXP2) {
1497 ureg_MUL(ureg, rFog, _ZZZZ(vPos), _ZZZZ(_CONST(22)));
1498 ureg_MUL(ureg, rFog, _X(rFog), _X(rFog));
1499 ureg_MUL(ureg, rFog, _X(rFog), ureg_imm1f(ureg, -1.442695f));
1500 ureg_EX2(ureg, rFog, _X(rFog));
1501 } else
1502 if (key->fog_mode == D3DFOG_LINEAR) {
1503 ureg_SUB(ureg, rFog, _XXXX(_CONST(22)), _ZZZZ(vPos));
1504 ureg_MUL(ureg, ureg_saturate(rFog), _X(rFog), _YYYY(_CONST(22)));
1505 }
1506 ureg_LRP(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_XYZ), _X(rFog), ps.rCurSrc, _CONST(21));
1507 ureg_MOV(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_W), ps.rCurSrc);
1508 } else
1509 if (key->fog) {
1510 struct ureg_src vFog = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_FOG, 0, TGSI_INTERPOLATE_PERSPECTIVE);
1511 ureg_LRP(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_XYZ), _XXXX(vFog), ps.rCurSrc, _CONST(21));
1512 ureg_MOV(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_W), ps.rCurSrc);
1513 } else {
1514 ureg_MOV(ureg, oCol, ps.rCurSrc);
1515 }
1516
1517 ureg_END(ureg);
1518 nine_ureg_tgsi_dump(ureg, FALSE);
1519 return ureg_create_shader_and_destroy(ureg, device->pipe);
1520 }
1521
1522 static struct NineVertexShader9 *
1523 nine_ff_get_vs(struct NineDevice9 *device)
1524 {
1525 const struct nine_state *state = &device->state;
1526 struct NineVertexShader9 *vs;
1527 enum pipe_error err;
1528 struct vs_build_ctx bld;
1529 struct nine_ff_vs_key key;
1530 unsigned s, i;
1531 char input_texture_coord[8];
1532
1533 assert(sizeof(key) <= sizeof(key.value32));
1534
1535 memset(&key, 0, sizeof(key));
1536 memset(&bld, 0, sizeof(bld));
1537 memset(&input_texture_coord, 0, sizeof(input_texture_coord));
1538
1539 bld.key = &key;
1540
1541 /* FIXME: this shouldn't be NULL, but it is on init */
1542 if (state->vdecl) {
1543 key.color0in_one = 1;
1544 key.color1in_one = 1;
1545 for (i = 0; i < state->vdecl->nelems; i++) {
1546 uint16_t usage = state->vdecl->usage_map[i];
1547 if (usage == NINE_DECLUSAGE_POSITIONT)
1548 key.position_t = 1;
1549 else if (usage == NINE_DECLUSAGE_i(COLOR, 0))
1550 key.color0in_one = 0;
1551 else if (usage == NINE_DECLUSAGE_i(COLOR, 1))
1552 key.color1in_one = 0;
1553 else if (usage == NINE_DECLUSAGE_PSIZE)
1554 key.vertexpointsize = 1;
1555 else if (usage % NINE_DECLUSAGE_COUNT == NINE_DECLUSAGE_TEXCOORD) {
1556 s = usage / NINE_DECLUSAGE_COUNT;
1557 if (s < 8)
1558 input_texture_coord[s] = nine_decltype_get_dim(state->vdecl->decls[i].Type);
1559 else
1560 DBG("FF given texture coordinate >= 8. Ignoring\n");
1561 } else if (usage < NINE_DECLUSAGE_NONE)
1562 key.passthrough |= 1 << usage;
1563 }
1564 }
1565 /* ff vs + ps 3.0: some elements are passed to the ps (wine test).
1566 * We do restrict to indices 0 */
1567 key.passthrough &= ~((1 << NINE_DECLUSAGE_POSITION) | (1 << NINE_DECLUSAGE_PSIZE) |
1568 (1 << NINE_DECLUSAGE_TEXCOORD) | (1 << NINE_DECLUSAGE_POSITIONT) |
1569 (1 << NINE_DECLUSAGE_TESSFACTOR) | (1 << NINE_DECLUSAGE_SAMPLE));
1570 key.pointscale = !!state->rs[D3DRS_POINTSCALEENABLE];
1571
1572 key.lighting = !!state->rs[D3DRS_LIGHTING] && state->ff.num_lights_active;
1573 key.darkness = !!state->rs[D3DRS_LIGHTING] && !state->ff.num_lights_active;
1574 if (key.position_t) {
1575 key.darkness = 0; /* |= key.lighting; */ /* XXX ? */
1576 key.lighting = 0;
1577 }
1578 if ((key.lighting | key.darkness) && state->rs[D3DRS_COLORVERTEX]) {
1579 key.mtl_diffuse = state->rs[D3DRS_DIFFUSEMATERIALSOURCE];
1580 key.mtl_ambient = state->rs[D3DRS_AMBIENTMATERIALSOURCE];
1581 key.mtl_specular = state->rs[D3DRS_SPECULARMATERIALSOURCE];
1582 key.mtl_emissive = state->rs[D3DRS_EMISSIVEMATERIALSOURCE];
1583 }
1584 key.fog = !!state->rs[D3DRS_FOGENABLE];
1585 key.fog_mode = (!key.position_t && state->rs[D3DRS_FOGENABLE]) ? state->rs[D3DRS_FOGVERTEXMODE] : 0;
1586 if (key.fog_mode)
1587 key.fog_range = state->rs[D3DRS_RANGEFOGENABLE];
1588
1589 key.localviewer = !!state->rs[D3DRS_LOCALVIEWER];
1590 key.specular_enable = !!state->rs[D3DRS_SPECULARENABLE];
1591 key.normalizenormals = !!state->rs[D3DRS_NORMALIZENORMALS];
1592
1593 if (state->rs[D3DRS_VERTEXBLEND] != D3DVBF_DISABLE) {
1594 key.vertexblend_indexed = !!state->rs[D3DRS_INDEXEDVERTEXBLENDENABLE];
1595
1596 switch (state->rs[D3DRS_VERTEXBLEND]) {
1597 case D3DVBF_0WEIGHTS: key.vertexblend = key.vertexblend_indexed; break;
1598 case D3DVBF_1WEIGHTS: key.vertexblend = 2; break;
1599 case D3DVBF_2WEIGHTS: key.vertexblend = 3; break;
1600 case D3DVBF_3WEIGHTS: key.vertexblend = 4; break;
1601 case D3DVBF_TWEENING: key.vertextween = 1; break;
1602 default:
1603 assert(!"invalid D3DVBF");
1604 break;
1605 }
1606 }
1607
1608 for (s = 0; s < 8; ++s) {
1609 unsigned gen = (state->ff.tex_stage[s][D3DTSS_TEXCOORDINDEX] >> 16) + 1;
1610 unsigned dim;
1611
1612 if (key.position_t && gen > NINED3DTSS_TCI_PASSTHRU)
1613 gen = NINED3DTSS_TCI_PASSTHRU;
1614
1615 if (!input_texture_coord[s] && gen == NINED3DTSS_TCI_PASSTHRU)
1616 gen = NINED3DTSS_TCI_DISABLE;
1617
1618 key.tc_gen |= gen << (s * 3);
1619 key.tc_idx |= (state->ff.tex_stage[s][D3DTSS_TEXCOORDINDEX] & 7) << (s * 3);
1620 key.tc_dim_input |= ((input_texture_coord[s]-1) & 0x3) << (s * 2);
1621
1622 dim = state->ff.tex_stage[s][D3DTSS_TEXTURETRANSFORMFLAGS] & 0x7;
1623 if (dim > 4)
1624 dim = input_texture_coord[s];
1625 if (dim == 1) /* NV behaviour */
1626 dim = 0;
1627 key.tc_dim_output |= dim << (s * 3);
1628 }
1629
1630 vs = util_hash_table_get(device->ff.ht_vs, &key);
1631 if (vs)
1632 return vs;
1633 NineVertexShader9_new(device, &vs, NULL, nine_ff_build_vs(device, &bld));
1634
1635 nine_ff_prune_vs(device);
1636 if (vs) {
1637 unsigned n;
1638
1639 memcpy(&vs->ff_key, &key, sizeof(vs->ff_key));
1640
1641 err = util_hash_table_set(device->ff.ht_vs, &vs->ff_key, vs);
1642 (void)err;
1643 assert(err == PIPE_OK);
1644 device->ff.num_vs++;
1645 NineUnknown_ConvertRefToBind(NineUnknown(vs));
1646
1647 vs->num_inputs = bld.num_inputs;
1648 for (n = 0; n < bld.num_inputs; ++n)
1649 vs->input_map[n].ndecl = bld.input[n];
1650
1651 vs->position_t = key.position_t;
1652 vs->point_size = key.vertexpointsize | key.pointscale;
1653 }
1654 return vs;
1655 }
1656
1657 static struct NinePixelShader9 *
1658 nine_ff_get_ps(struct NineDevice9 *device)
1659 {
1660 struct nine_state *state = &device->state;
1661 struct NinePixelShader9 *ps;
1662 enum pipe_error err;
1663 struct nine_ff_ps_key key;
1664 unsigned s;
1665 uint8_t sampler_mask = 0;
1666
1667 assert(sizeof(key) <= sizeof(key.value32));
1668
1669 memset(&key, 0, sizeof(key));
1670 for (s = 0; s < 8; ++s) {
1671 key.ts[s].colorop = state->ff.tex_stage[s][D3DTSS_COLOROP];
1672 key.ts[s].alphaop = state->ff.tex_stage[s][D3DTSS_ALPHAOP];
1673 /* MSDN says D3DTOP_DISABLE disables this and all subsequent stages. */
1674 /* ALPHAOP cannot be disabled if COLOROP is enabled. */
1675 if (key.ts[s].colorop == D3DTOP_DISABLE) {
1676 key.ts[s].alphaop = D3DTOP_DISABLE; /* DISABLE == 1, avoid degenerate keys */
1677 break;
1678 }
1679
1680 if (!state->texture[s] &&
1681 state->ff.tex_stage[s][D3DTSS_COLORARG1] == D3DTA_TEXTURE) {
1682 /* This should also disable the stage. */
1683 key.ts[s].colorop = key.ts[s].alphaop = D3DTOP_DISABLE;
1684 break;
1685 }
1686
1687 if (state->ff.tex_stage[s][D3DTSS_COLORARG1] == D3DTA_TEXTURE)
1688 sampler_mask |= (1 << s);
1689
1690 if (key.ts[s].colorop != D3DTOP_DISABLE) {
1691 uint8_t used_c = ps_d3dtop_args_mask(key.ts[s].colorop);
1692 if (used_c & 0x1) key.ts[s].colorarg0 = state->ff.tex_stage[s][D3DTSS_COLORARG0];
1693 if (used_c & 0x2) key.ts[s].colorarg1 = state->ff.tex_stage[s][D3DTSS_COLORARG1];
1694 if (used_c & 0x4) key.ts[s].colorarg2 = state->ff.tex_stage[s][D3DTSS_COLORARG2];
1695 if (used_c & 0x1) key.colorarg_b4[0] |= (state->ff.tex_stage[s][D3DTSS_COLORARG0] >> 4) << s;
1696 if (used_c & 0x1) key.colorarg_b5[0] |= (state->ff.tex_stage[s][D3DTSS_COLORARG0] >> 5) << s;
1697 if (used_c & 0x2) key.colorarg_b4[1] |= (state->ff.tex_stage[s][D3DTSS_COLORARG1] >> 4) << s;
1698 if (used_c & 0x2) key.colorarg_b5[1] |= (state->ff.tex_stage[s][D3DTSS_COLORARG1] >> 5) << s;
1699 if (used_c & 0x4) key.colorarg_b4[2] |= (state->ff.tex_stage[s][D3DTSS_COLORARG2] >> 4) << s;
1700 if (used_c & 0x4) key.colorarg_b5[2] |= (state->ff.tex_stage[s][D3DTSS_COLORARG2] >> 5) << s;
1701 }
1702 if (key.ts[s].alphaop != D3DTOP_DISABLE) {
1703 uint8_t used_a = ps_d3dtop_args_mask(key.ts[s].alphaop);
1704 if (used_a & 0x1) key.ts[s].alphaarg0 = state->ff.tex_stage[s][D3DTSS_ALPHAARG0];
1705 if (used_a & 0x2) key.ts[s].alphaarg1 = state->ff.tex_stage[s][D3DTSS_ALPHAARG1];
1706 if (used_a & 0x4) key.ts[s].alphaarg2 = state->ff.tex_stage[s][D3DTSS_ALPHAARG2];
1707 if (used_a & 0x1) key.alphaarg_b4[0] |= (state->ff.tex_stage[s][D3DTSS_ALPHAARG0] >> 4) << s;
1708 if (used_a & 0x2) key.alphaarg_b4[1] |= (state->ff.tex_stage[s][D3DTSS_ALPHAARG1] >> 4) << s;
1709 if (used_a & 0x4) key.alphaarg_b4[2] |= (state->ff.tex_stage[s][D3DTSS_ALPHAARG2] >> 4) << s;
1710 }
1711 key.ts[s].resultarg = state->ff.tex_stage[s][D3DTSS_RESULTARG] == D3DTA_TEMP;
1712
1713 if (state->texture[s]) {
1714 switch (state->texture[s]->base.type) {
1715 case D3DRTYPE_TEXTURE: key.ts[s].textarget = 1; break;
1716 case D3DRTYPE_VOLUMETEXTURE: key.ts[s].textarget = 2; break;
1717 case D3DRTYPE_CUBETEXTURE: key.ts[s].textarget = 3; break;
1718 default:
1719 assert(!"unexpected texture type");
1720 break;
1721 }
1722 } else {
1723 key.ts[s].textarget = 1;
1724 }
1725 }
1726
1727 key.projected = nine_ff_get_projected_key(state);
1728
1729 for (; s < 8; ++s)
1730 key.ts[s].colorop = key.ts[s].alphaop = D3DTOP_DISABLE;
1731 if (state->rs[D3DRS_FOGENABLE])
1732 key.fog_mode = state->rs[D3DRS_FOGTABLEMODE];
1733 key.fog = !!state->rs[D3DRS_FOGENABLE];
1734
1735 ps = util_hash_table_get(device->ff.ht_ps, &key);
1736 if (ps)
1737 return ps;
1738 NinePixelShader9_new(device, &ps, NULL, nine_ff_build_ps(device, &key));
1739
1740 nine_ff_prune_ps(device);
1741 if (ps) {
1742 memcpy(&ps->ff_key, &key, sizeof(ps->ff_key));
1743
1744 err = util_hash_table_set(device->ff.ht_ps, &ps->ff_key, ps);
1745 (void)err;
1746 assert(err == PIPE_OK);
1747 device->ff.num_ps++;
1748 NineUnknown_ConvertRefToBind(NineUnknown(ps));
1749
1750 ps->rt_mask = 0x1;
1751 ps->sampler_mask = sampler_mask;
1752 }
1753 return ps;
1754 }
1755
1756 #define GET_D3DTS(n) nine_state_access_transform(state, D3DTS_##n, FALSE)
1757 #define IS_D3DTS_DIRTY(s,n) ((s)->ff.changed.transform[(D3DTS_##n) / 32] & (1 << ((D3DTS_##n) % 32)))
1758 static void
1759 nine_ff_load_vs_transforms(struct NineDevice9 *device)
1760 {
1761 struct nine_state *state = &device->state;
1762 D3DMATRIX T;
1763 D3DMATRIX *M = (D3DMATRIX *)device->ff.vs_const;
1764 unsigned i;
1765
1766 /* TODO: make this nicer, and only upload the ones we need */
1767 /* TODO: use ff.vs_const as storage of W, V, P matrices */
1768
1769 if (IS_D3DTS_DIRTY(state, WORLD) ||
1770 IS_D3DTS_DIRTY(state, VIEW) ||
1771 IS_D3DTS_DIRTY(state, PROJECTION)) {
1772 /* WVP, WV matrices */
1773 nine_d3d_matrix_matrix_mul(&M[1], GET_D3DTS(WORLD), GET_D3DTS(VIEW));
1774 nine_d3d_matrix_matrix_mul(&M[0], &M[1], GET_D3DTS(PROJECTION));
1775
1776 /* normal matrix == transpose(inverse(WV)) */
1777 nine_d3d_matrix_inverse_3x3(&T, &M[1]);
1778 nine_d3d_matrix_transpose(&M[4], &T);
1779
1780 /* P matrix */
1781 M[2] = *GET_D3DTS(PROJECTION);
1782
1783 /* V and W matrix */
1784 M[3] = *GET_D3DTS(VIEW);
1785 M[40] = M[1];
1786 }
1787
1788 if (state->rs[D3DRS_VERTEXBLEND] != D3DVBF_DISABLE) {
1789 /* load other world matrices */
1790 for (i = 1; i <= 7; ++i) {
1791 nine_d3d_matrix_matrix_mul(&M[40 + i], GET_D3DTS(WORLDMATRIX(i)), GET_D3DTS(VIEW));
1792 }
1793 }
1794
1795 device->ff.vs_const[30 * 4] = asfloat(state->rs[D3DRS_TWEENFACTOR]);
1796 }
1797
1798 static void
1799 nine_ff_load_lights(struct NineDevice9 *device)
1800 {
1801 struct nine_state *state = &device->state;
1802 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1803 unsigned l;
1804
1805 if (state->changed.group & NINE_STATE_FF_MATERIAL) {
1806 const D3DMATERIAL9 *mtl = &state->ff.material;
1807
1808 memcpy(&dst[20], &mtl->Diffuse, 4 * sizeof(float));
1809 memcpy(&dst[21], &mtl->Ambient, 4 * sizeof(float));
1810 memcpy(&dst[22], &mtl->Specular, 4 * sizeof(float));
1811 dst[23].x = mtl->Power;
1812 memcpy(&dst[24], &mtl->Emissive, 4 * sizeof(float));
1813 d3dcolor_to_rgba(&dst[25].x, state->rs[D3DRS_AMBIENT]);
1814 dst[19].x = dst[25].x * mtl->Ambient.r + mtl->Emissive.r;
1815 dst[19].y = dst[25].y * mtl->Ambient.g + mtl->Emissive.g;
1816 dst[19].z = dst[25].z * mtl->Ambient.b + mtl->Emissive.b;
1817 dst[19].w = mtl->Ambient.a + mtl->Emissive.a;
1818 }
1819
1820 if (!(state->changed.group & NINE_STATE_FF_LIGHTING))
1821 return;
1822
1823 for (l = 0; l < state->ff.num_lights_active; ++l) {
1824 const D3DLIGHT9 *light = &state->ff.light[state->ff.active_light[l]];
1825
1826 dst[32 + l * 8].x = light->Type;
1827 dst[32 + l * 8].y = light->Attenuation0;
1828 dst[32 + l * 8].z = light->Attenuation1;
1829 dst[32 + l * 8].w = light->Attenuation2;
1830 memcpy(&dst[33 + l * 8].x, &light->Diffuse, sizeof(light->Diffuse));
1831 memcpy(&dst[34 + l * 8].x, &light->Specular, sizeof(light->Specular));
1832 memcpy(&dst[35 + l * 8].x, &light->Ambient, sizeof(light->Ambient));
1833 nine_d3d_vector4_matrix_mul((D3DVECTOR *)&dst[36 + l * 8].x, &light->Position, GET_D3DTS(VIEW));
1834 nine_d3d_vector3_matrix_mul((D3DVECTOR *)&dst[37 + l * 8].x, &light->Direction, GET_D3DTS(VIEW));
1835 dst[36 + l * 8].w = light->Type == D3DLIGHT_DIRECTIONAL ? 1e9f : light->Range;
1836 dst[37 + l * 8].w = light->Falloff;
1837 dst[38 + l * 8].x = cosf(light->Theta * 0.5f);
1838 dst[38 + l * 8].y = cosf(light->Phi * 0.5f);
1839 dst[38 + l * 8].z = 1.0f / (dst[38 + l * 8].x - dst[38 + l * 8].y);
1840 dst[39 + l * 8].w = (l + 1) == state->ff.num_lights_active;
1841 }
1842 }
1843
1844 static void
1845 nine_ff_load_point_and_fog_params(struct NineDevice9 *device)
1846 {
1847 const struct nine_state *state = &device->state;
1848 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1849
1850 if (!(state->changed.group & NINE_STATE_FF_OTHER))
1851 return;
1852 dst[26].x = asfloat(state->rs[D3DRS_POINTSIZE_MIN]);
1853 dst[26].y = asfloat(state->rs[D3DRS_POINTSIZE_MAX]);
1854 dst[26].z = asfloat(state->rs[D3DRS_POINTSIZE]);
1855 dst[26].w = asfloat(state->rs[D3DRS_POINTSCALE_A]);
1856 dst[27].x = asfloat(state->rs[D3DRS_POINTSCALE_B]);
1857 dst[27].y = asfloat(state->rs[D3DRS_POINTSCALE_C]);
1858 dst[28].x = asfloat(state->rs[D3DRS_FOGEND]);
1859 dst[28].y = 1.0f / (asfloat(state->rs[D3DRS_FOGEND]) - asfloat(state->rs[D3DRS_FOGSTART]));
1860 if (isinf(dst[28].y))
1861 dst[28].y = 0.0f;
1862 dst[28].z = asfloat(state->rs[D3DRS_FOGDENSITY]);
1863 }
1864
1865 static void
1866 nine_ff_load_tex_matrices(struct NineDevice9 *device)
1867 {
1868 struct nine_state *state = &device->state;
1869 D3DMATRIX *M = (D3DMATRIX *)device->ff.vs_const;
1870 unsigned s;
1871
1872 if (!(state->ff.changed.transform[0] & 0xff0000))
1873 return;
1874 for (s = 0; s < 8; ++s) {
1875 if (IS_D3DTS_DIRTY(state, TEXTURE0 + s))
1876 nine_d3d_matrix_transpose(&M[32 + s], nine_state_access_transform(state, D3DTS_TEXTURE0 + s, FALSE));
1877 }
1878 }
1879
1880 static void
1881 nine_ff_load_ps_params(struct NineDevice9 *device)
1882 {
1883 const struct nine_state *state = &device->state;
1884 struct fvec4 *dst = (struct fvec4 *)device->ff.ps_const;
1885 unsigned s;
1886
1887 if (!(state->changed.group & (NINE_STATE_FF_PSSTAGES | NINE_STATE_FF_OTHER)))
1888 return;
1889
1890 for (s = 0; s < 8; ++s)
1891 d3dcolor_to_rgba(&dst[s].x, state->ff.tex_stage[s][D3DTSS_CONSTANT]);
1892
1893 for (s = 0; s < 8; ++s) {
1894 dst[8 + s].x = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT00]);
1895 dst[8 + s].y = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT01]);
1896 dst[8 + s].z = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT10]);
1897 dst[8 + s].w = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT11]);
1898 if (s & 1) {
1899 dst[16 + s / 2].z = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLSCALE]);
1900 dst[16 + s / 2].w = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLOFFSET]);
1901 } else {
1902 dst[16 + s / 2].x = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLSCALE]);
1903 dst[16 + s / 2].y = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLOFFSET]);
1904 }
1905 }
1906
1907 d3dcolor_to_rgba(&dst[20].x, state->rs[D3DRS_TEXTUREFACTOR]);
1908 d3dcolor_to_rgba(&dst[21].x, state->rs[D3DRS_FOGCOLOR]);
1909 dst[22].x = asfloat(state->rs[D3DRS_FOGEND]);
1910 dst[22].y = 1.0f / (asfloat(state->rs[D3DRS_FOGEND]) - asfloat(state->rs[D3DRS_FOGSTART]));
1911 dst[22].z = asfloat(state->rs[D3DRS_FOGDENSITY]);
1912 }
1913
1914 static void
1915 nine_ff_load_viewport_info(struct NineDevice9 *device)
1916 {
1917 D3DVIEWPORT9 *viewport = &device->state.viewport;
1918 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1919 float diffZ = viewport->MaxZ - viewport->MinZ;
1920
1921 /* Note: the other functions avoids to fill the const again if nothing changed.
1922 * But we don't have much to fill, and adding code to allow that may be complex
1923 * so just fill it always */
1924 dst[100].x = 2.0f / (float)(viewport->Width);
1925 dst[100].y = 2.0f / (float)(viewport->Height);
1926 dst[100].z = (diffZ == 0.0f) ? 0.0f : (1.0f / diffZ);
1927 dst[100].w = (float)(viewport->Width);
1928 dst[101].x = (float)(viewport->X);
1929 dst[101].y = (float)(viewport->Y);
1930 dst[101].z = (float)(viewport->MinZ);
1931 }
1932
1933 void
1934 nine_ff_update(struct NineDevice9 *device)
1935 {
1936 struct nine_state *state = &device->state;
1937 struct pipe_constant_buffer cb;
1938
1939 DBG("vs=%p ps=%p\n", device->state.vs, device->state.ps);
1940
1941 /* NOTE: the only reference belongs to the hash table */
1942 if (!state->programmable_vs) {
1943 device->ff.vs = nine_ff_get_vs(device);
1944 device->state.changed.group |= NINE_STATE_VS;
1945 }
1946 if (!device->state.ps) {
1947 device->ff.ps = nine_ff_get_ps(device);
1948 device->state.changed.group |= NINE_STATE_PS;
1949 }
1950
1951 if (!state->programmable_vs) {
1952 nine_ff_load_vs_transforms(device);
1953 nine_ff_load_tex_matrices(device);
1954 nine_ff_load_lights(device);
1955 nine_ff_load_point_and_fog_params(device);
1956 nine_ff_load_viewport_info(device);
1957
1958 memset(state->ff.changed.transform, 0, sizeof(state->ff.changed.transform));
1959
1960 cb.buffer_offset = 0;
1961 cb.buffer = NULL;
1962 cb.user_buffer = device->ff.vs_const;
1963 cb.buffer_size = NINE_FF_NUM_VS_CONST * 4 * sizeof(float);
1964
1965 if (!device->driver_caps.user_cbufs) {
1966 u_upload_data(device->constbuf_uploader,
1967 0,
1968 cb.buffer_size,
1969 device->constbuf_alignment,
1970 cb.user_buffer,
1971 &cb.buffer_offset,
1972 &cb.buffer);
1973 u_upload_unmap(device->constbuf_uploader);
1974 cb.user_buffer = NULL;
1975 }
1976 state->pipe.cb_vs_ff = cb;
1977 state->commit |= NINE_STATE_COMMIT_CONST_VS;
1978 }
1979
1980 if (!device->state.ps) {
1981 nine_ff_load_ps_params(device);
1982
1983 cb.buffer_offset = 0;
1984 cb.buffer = NULL;
1985 cb.user_buffer = device->ff.ps_const;
1986 cb.buffer_size = NINE_FF_NUM_PS_CONST * 4 * sizeof(float);
1987
1988 if (!device->driver_caps.user_cbufs) {
1989 u_upload_data(device->constbuf_uploader,
1990 0,
1991 cb.buffer_size,
1992 device->constbuf_alignment,
1993 cb.user_buffer,
1994 &cb.buffer_offset,
1995 &cb.buffer);
1996 u_upload_unmap(device->constbuf_uploader);
1997 cb.user_buffer = NULL;
1998 }
1999 state->pipe.cb_ps_ff = cb;
2000 state->commit |= NINE_STATE_COMMIT_CONST_PS;
2001 }
2002
2003 device->state.changed.group &= ~NINE_STATE_FF;
2004 }
2005
2006
2007 boolean
2008 nine_ff_init(struct NineDevice9 *device)
2009 {
2010 device->ff.ht_vs = util_hash_table_create(nine_ff_vs_key_hash,
2011 nine_ff_vs_key_comp);
2012 device->ff.ht_ps = util_hash_table_create(nine_ff_ps_key_hash,
2013 nine_ff_ps_key_comp);
2014
2015 device->ff.ht_fvf = util_hash_table_create(nine_ff_fvf_key_hash,
2016 nine_ff_fvf_key_comp);
2017
2018 device->ff.vs_const = CALLOC(NINE_FF_NUM_VS_CONST, 4 * sizeof(float));
2019 device->ff.ps_const = CALLOC(NINE_FF_NUM_PS_CONST, 4 * sizeof(float));
2020
2021 return device->ff.ht_vs && device->ff.ht_ps &&
2022 device->ff.ht_fvf &&
2023 device->ff.vs_const && device->ff.ps_const;
2024 }
2025
2026 static enum pipe_error nine_ff_ht_delete_cb(void *key, void *value, void *data)
2027 {
2028 NineUnknown_Unbind(NineUnknown(value));
2029 return PIPE_OK;
2030 }
2031
2032 void
2033 nine_ff_fini(struct NineDevice9 *device)
2034 {
2035 if (device->ff.ht_vs) {
2036 util_hash_table_foreach(device->ff.ht_vs, nine_ff_ht_delete_cb, NULL);
2037 util_hash_table_destroy(device->ff.ht_vs);
2038 }
2039 if (device->ff.ht_ps) {
2040 util_hash_table_foreach(device->ff.ht_ps, nine_ff_ht_delete_cb, NULL);
2041 util_hash_table_destroy(device->ff.ht_ps);
2042 }
2043 if (device->ff.ht_fvf) {
2044 util_hash_table_foreach(device->ff.ht_fvf, nine_ff_ht_delete_cb, NULL);
2045 util_hash_table_destroy(device->ff.ht_fvf);
2046 }
2047 device->ff.vs = NULL; /* destroyed by unbinding from hash table */
2048 device->ff.ps = NULL;
2049
2050 FREE(device->ff.vs_const);
2051 FREE(device->ff.ps_const);
2052 }
2053
2054 static void
2055 nine_ff_prune_vs(struct NineDevice9 *device)
2056 {
2057 if (device->ff.num_vs > 100) {
2058 /* could destroy the bound one here, so unbind */
2059 device->pipe->bind_vs_state(device->pipe, NULL);
2060 util_hash_table_foreach(device->ff.ht_vs, nine_ff_ht_delete_cb, NULL);
2061 util_hash_table_clear(device->ff.ht_vs);
2062 device->ff.num_vs = 0;
2063 device->state.changed.group |= NINE_STATE_VS;
2064 }
2065 }
2066 static void
2067 nine_ff_prune_ps(struct NineDevice9 *device)
2068 {
2069 if (device->ff.num_ps > 100) {
2070 /* could destroy the bound one here, so unbind */
2071 device->pipe->bind_fs_state(device->pipe, NULL);
2072 util_hash_table_foreach(device->ff.ht_ps, nine_ff_ht_delete_cb, NULL);
2073 util_hash_table_clear(device->ff.ht_ps);
2074 device->ff.num_ps = 0;
2075 device->state.changed.group |= NINE_STATE_PS;
2076 }
2077 }
2078
2079 /* ========================================================================== */
2080
2081 /* Matrix multiplication:
2082 *
2083 * in memory: 0 1 2 3 (row major)
2084 * 4 5 6 7
2085 * 8 9 a b
2086 * c d e f
2087 *
2088 * cA cB cC cD
2089 * r0 = (r0 * cA) (r0 * cB) . .
2090 * r1 = (r1 * cA) (r1 * cB)
2091 * r2 = (r2 * cA) .
2092 * r3 = (r3 * cA) .
2093 *
2094 * r: (11) (12) (13) (14)
2095 * (21) (22) (23) (24)
2096 * (31) (32) (33) (34)
2097 * (41) (42) (43) (44)
2098 * l: (11 12 13 14)
2099 * (21 22 23 24)
2100 * (31 32 33 34)
2101 * (41 42 43 44)
2102 *
2103 * v: (x y z 1 )
2104 *
2105 * t.xyzw = MUL(v.xxxx, r[0]);
2106 * t.xyzw = MAD(v.yyyy, r[1], t.xyzw);
2107 * t.xyzw = MAD(v.zzzz, r[2], t.xyzw);
2108 * v.xyzw = MAD(v.wwww, r[3], t.xyzw);
2109 *
2110 * v.x = DP4(v, c[0]);
2111 * v.y = DP4(v, c[1]);
2112 * v.z = DP4(v, c[2]);
2113 * v.w = DP4(v, c[3]) = 1
2114 */
2115
2116 /*
2117 static void
2118 nine_D3DMATRIX_print(const D3DMATRIX *M)
2119 {
2120 DBG("\n(%f %f %f %f)\n"
2121 "(%f %f %f %f)\n"
2122 "(%f %f %f %f)\n"
2123 "(%f %f %f %f)\n",
2124 M->m[0][0], M->m[0][1], M->m[0][2], M->m[0][3],
2125 M->m[1][0], M->m[1][1], M->m[1][2], M->m[1][3],
2126 M->m[2][0], M->m[2][1], M->m[2][2], M->m[2][3],
2127 M->m[3][0], M->m[3][1], M->m[3][2], M->m[3][3]);
2128 }
2129 */
2130
2131 static inline float
2132 nine_DP4_row_col(const D3DMATRIX *A, int r, const D3DMATRIX *B, int c)
2133 {
2134 return A->m[r][0] * B->m[0][c] +
2135 A->m[r][1] * B->m[1][c] +
2136 A->m[r][2] * B->m[2][c] +
2137 A->m[r][3] * B->m[3][c];
2138 }
2139
2140 static inline float
2141 nine_DP4_vec_col(const D3DVECTOR *v, const D3DMATRIX *M, int c)
2142 {
2143 return v->x * M->m[0][c] +
2144 v->y * M->m[1][c] +
2145 v->z * M->m[2][c] +
2146 1.0f * M->m[3][c];
2147 }
2148
2149 static inline float
2150 nine_DP3_vec_col(const D3DVECTOR *v, const D3DMATRIX *M, int c)
2151 {
2152 return v->x * M->m[0][c] +
2153 v->y * M->m[1][c] +
2154 v->z * M->m[2][c];
2155 }
2156
2157 void
2158 nine_d3d_matrix_matrix_mul(D3DMATRIX *D, const D3DMATRIX *L, const D3DMATRIX *R)
2159 {
2160 D->_11 = nine_DP4_row_col(L, 0, R, 0);
2161 D->_12 = nine_DP4_row_col(L, 0, R, 1);
2162 D->_13 = nine_DP4_row_col(L, 0, R, 2);
2163 D->_14 = nine_DP4_row_col(L, 0, R, 3);
2164
2165 D->_21 = nine_DP4_row_col(L, 1, R, 0);
2166 D->_22 = nine_DP4_row_col(L, 1, R, 1);
2167 D->_23 = nine_DP4_row_col(L, 1, R, 2);
2168 D->_24 = nine_DP4_row_col(L, 1, R, 3);
2169
2170 D->_31 = nine_DP4_row_col(L, 2, R, 0);
2171 D->_32 = nine_DP4_row_col(L, 2, R, 1);
2172 D->_33 = nine_DP4_row_col(L, 2, R, 2);
2173 D->_34 = nine_DP4_row_col(L, 2, R, 3);
2174
2175 D->_41 = nine_DP4_row_col(L, 3, R, 0);
2176 D->_42 = nine_DP4_row_col(L, 3, R, 1);
2177 D->_43 = nine_DP4_row_col(L, 3, R, 2);
2178 D->_44 = nine_DP4_row_col(L, 3, R, 3);
2179 }
2180
2181 void
2182 nine_d3d_vector4_matrix_mul(D3DVECTOR *d, const D3DVECTOR *v, const D3DMATRIX *M)
2183 {
2184 d->x = nine_DP4_vec_col(v, M, 0);
2185 d->y = nine_DP4_vec_col(v, M, 1);
2186 d->z = nine_DP4_vec_col(v, M, 2);
2187 }
2188
2189 void
2190 nine_d3d_vector3_matrix_mul(D3DVECTOR *d, const D3DVECTOR *v, const D3DMATRIX *M)
2191 {
2192 d->x = nine_DP3_vec_col(v, M, 0);
2193 d->y = nine_DP3_vec_col(v, M, 1);
2194 d->z = nine_DP3_vec_col(v, M, 2);
2195 }
2196
2197 void
2198 nine_d3d_matrix_transpose(D3DMATRIX *D, const D3DMATRIX *M)
2199 {
2200 unsigned i, j;
2201 for (i = 0; i < 4; ++i)
2202 for (j = 0; j < 4; ++j)
2203 D->m[i][j] = M->m[j][i];
2204 }
2205
2206 #define _M_ADD_PROD_1i_2j_3k_4l(i,j,k,l) do { \
2207 float t = M->_1##i * M->_2##j * M->_3##k * M->_4##l; \
2208 if (t > 0.0f) pos += t; else neg += t; } while(0)
2209
2210 #define _M_SUB_PROD_1i_2j_3k_4l(i,j,k,l) do { \
2211 float t = M->_1##i * M->_2##j * M->_3##k * M->_4##l; \
2212 if (t > 0.0f) neg -= t; else pos -= t; } while(0)
2213 float
2214 nine_d3d_matrix_det(const D3DMATRIX *M)
2215 {
2216 float pos = 0.0f;
2217 float neg = 0.0f;
2218
2219 _M_ADD_PROD_1i_2j_3k_4l(1, 2, 3, 4);
2220 _M_ADD_PROD_1i_2j_3k_4l(1, 3, 4, 2);
2221 _M_ADD_PROD_1i_2j_3k_4l(1, 4, 2, 3);
2222
2223 _M_ADD_PROD_1i_2j_3k_4l(2, 1, 4, 3);
2224 _M_ADD_PROD_1i_2j_3k_4l(2, 3, 1, 4);
2225 _M_ADD_PROD_1i_2j_3k_4l(2, 4, 3, 1);
2226
2227 _M_ADD_PROD_1i_2j_3k_4l(3, 1, 2, 4);
2228 _M_ADD_PROD_1i_2j_3k_4l(3, 2, 4, 1);
2229 _M_ADD_PROD_1i_2j_3k_4l(3, 4, 1, 2);
2230
2231 _M_ADD_PROD_1i_2j_3k_4l(4, 1, 3, 2);
2232 _M_ADD_PROD_1i_2j_3k_4l(4, 2, 1, 3);
2233 _M_ADD_PROD_1i_2j_3k_4l(4, 3, 2, 1);
2234
2235 _M_SUB_PROD_1i_2j_3k_4l(1, 2, 4, 3);
2236 _M_SUB_PROD_1i_2j_3k_4l(1, 3, 2, 4);
2237 _M_SUB_PROD_1i_2j_3k_4l(1, 4, 3, 2);
2238
2239 _M_SUB_PROD_1i_2j_3k_4l(2, 1, 3, 4);
2240 _M_SUB_PROD_1i_2j_3k_4l(2, 3, 4, 1);
2241 _M_SUB_PROD_1i_2j_3k_4l(2, 4, 1, 3);
2242
2243 _M_SUB_PROD_1i_2j_3k_4l(3, 1, 4, 2);
2244 _M_SUB_PROD_1i_2j_3k_4l(3, 2, 1, 4);
2245 _M_SUB_PROD_1i_2j_3k_4l(3, 4, 2, 1);
2246
2247 _M_SUB_PROD_1i_2j_3k_4l(4, 1, 2, 3);
2248 _M_SUB_PROD_1i_2j_3k_4l(4, 2, 3, 1);
2249 _M_SUB_PROD_1i_2j_3k_4l(4, 3, 1, 2);
2250
2251 return pos + neg;
2252 }
2253
2254 /* XXX: Probably better to just use src/mesa/math/m_matrix.c because
2255 * I have no idea where this code came from.
2256 */
2257 void
2258 nine_d3d_matrix_inverse(D3DMATRIX *D, const D3DMATRIX *M)
2259 {
2260 int i, k;
2261 float det;
2262
2263 D->m[0][0] =
2264 M->m[1][1] * M->m[2][2] * M->m[3][3] -
2265 M->m[1][1] * M->m[3][2] * M->m[2][3] -
2266 M->m[1][2] * M->m[2][1] * M->m[3][3] +
2267 M->m[1][2] * M->m[3][1] * M->m[2][3] +
2268 M->m[1][3] * M->m[2][1] * M->m[3][2] -
2269 M->m[1][3] * M->m[3][1] * M->m[2][2];
2270
2271 D->m[0][1] =
2272 -M->m[0][1] * M->m[2][2] * M->m[3][3] +
2273 M->m[0][1] * M->m[3][2] * M->m[2][3] +
2274 M->m[0][2] * M->m[2][1] * M->m[3][3] -
2275 M->m[0][2] * M->m[3][1] * M->m[2][3] -
2276 M->m[0][3] * M->m[2][1] * M->m[3][2] +
2277 M->m[0][3] * M->m[3][1] * M->m[2][2];
2278
2279 D->m[0][2] =
2280 M->m[0][1] * M->m[1][2] * M->m[3][3] -
2281 M->m[0][1] * M->m[3][2] * M->m[1][3] -
2282 M->m[0][2] * M->m[1][1] * M->m[3][3] +
2283 M->m[0][2] * M->m[3][1] * M->m[1][3] +
2284 M->m[0][3] * M->m[1][1] * M->m[3][2] -
2285 M->m[0][3] * M->m[3][1] * M->m[1][2];
2286
2287 D->m[0][3] =
2288 -M->m[0][1] * M->m[1][2] * M->m[2][3] +
2289 M->m[0][1] * M->m[2][2] * M->m[1][3] +
2290 M->m[0][2] * M->m[1][1] * M->m[2][3] -
2291 M->m[0][2] * M->m[2][1] * M->m[1][3] -
2292 M->m[0][3] * M->m[1][1] * M->m[2][2] +
2293 M->m[0][3] * M->m[2][1] * M->m[1][2];
2294
2295 D->m[1][0] =
2296 -M->m[1][0] * M->m[2][2] * M->m[3][3] +
2297 M->m[1][0] * M->m[3][2] * M->m[2][3] +
2298 M->m[1][2] * M->m[2][0] * M->m[3][3] -
2299 M->m[1][2] * M->m[3][0] * M->m[2][3] -
2300 M->m[1][3] * M->m[2][0] * M->m[3][2] +
2301 M->m[1][3] * M->m[3][0] * M->m[2][2];
2302
2303 D->m[1][1] =
2304 M->m[0][0] * M->m[2][2] * M->m[3][3] -
2305 M->m[0][0] * M->m[3][2] * M->m[2][3] -
2306 M->m[0][2] * M->m[2][0] * M->m[3][3] +
2307 M->m[0][2] * M->m[3][0] * M->m[2][3] +
2308 M->m[0][3] * M->m[2][0] * M->m[3][2] -
2309 M->m[0][3] * M->m[3][0] * M->m[2][2];
2310
2311 D->m[1][2] =
2312 -M->m[0][0] * M->m[1][2] * M->m[3][3] +
2313 M->m[0][0] * M->m[3][2] * M->m[1][3] +
2314 M->m[0][2] * M->m[1][0] * M->m[3][3] -
2315 M->m[0][2] * M->m[3][0] * M->m[1][3] -
2316 M->m[0][3] * M->m[1][0] * M->m[3][2] +
2317 M->m[0][3] * M->m[3][0] * M->m[1][2];
2318
2319 D->m[1][3] =
2320 M->m[0][0] * M->m[1][2] * M->m[2][3] -
2321 M->m[0][0] * M->m[2][2] * M->m[1][3] -
2322 M->m[0][2] * M->m[1][0] * M->m[2][3] +
2323 M->m[0][2] * M->m[2][0] * M->m[1][3] +
2324 M->m[0][3] * M->m[1][0] * M->m[2][2] -
2325 M->m[0][3] * M->m[2][0] * M->m[1][2];
2326
2327 D->m[2][0] =
2328 M->m[1][0] * M->m[2][1] * M->m[3][3] -
2329 M->m[1][0] * M->m[3][1] * M->m[2][3] -
2330 M->m[1][1] * M->m[2][0] * M->m[3][3] +
2331 M->m[1][1] * M->m[3][0] * M->m[2][3] +
2332 M->m[1][3] * M->m[2][0] * M->m[3][1] -
2333 M->m[1][3] * M->m[3][0] * M->m[2][1];
2334
2335 D->m[2][1] =
2336 -M->m[0][0] * M->m[2][1] * M->m[3][3] +
2337 M->m[0][0] * M->m[3][1] * M->m[2][3] +
2338 M->m[0][1] * M->m[2][0] * M->m[3][3] -
2339 M->m[0][1] * M->m[3][0] * M->m[2][3] -
2340 M->m[0][3] * M->m[2][0] * M->m[3][1] +
2341 M->m[0][3] * M->m[3][0] * M->m[2][1];
2342
2343 D->m[2][2] =
2344 M->m[0][0] * M->m[1][1] * M->m[3][3] -
2345 M->m[0][0] * M->m[3][1] * M->m[1][3] -
2346 M->m[0][1] * M->m[1][0] * M->m[3][3] +
2347 M->m[0][1] * M->m[3][0] * M->m[1][3] +
2348 M->m[0][3] * M->m[1][0] * M->m[3][1] -
2349 M->m[0][3] * M->m[3][0] * M->m[1][1];
2350
2351 D->m[2][3] =
2352 -M->m[0][0] * M->m[1][1] * M->m[2][3] +
2353 M->m[0][0] * M->m[2][1] * M->m[1][3] +
2354 M->m[0][1] * M->m[1][0] * M->m[2][3] -
2355 M->m[0][1] * M->m[2][0] * M->m[1][3] -
2356 M->m[0][3] * M->m[1][0] * M->m[2][1] +
2357 M->m[0][3] * M->m[2][0] * M->m[1][1];
2358
2359 D->m[3][0] =
2360 -M->m[1][0] * M->m[2][1] * M->m[3][2] +
2361 M->m[1][0] * M->m[3][1] * M->m[2][2] +
2362 M->m[1][1] * M->m[2][0] * M->m[3][2] -
2363 M->m[1][1] * M->m[3][0] * M->m[2][2] -
2364 M->m[1][2] * M->m[2][0] * M->m[3][1] +
2365 M->m[1][2] * M->m[3][0] * M->m[2][1];
2366
2367 D->m[3][1] =
2368 M->m[0][0] * M->m[2][1] * M->m[3][2] -
2369 M->m[0][0] * M->m[3][1] * M->m[2][2] -
2370 M->m[0][1] * M->m[2][0] * M->m[3][2] +
2371 M->m[0][1] * M->m[3][0] * M->m[2][2] +
2372 M->m[0][2] * M->m[2][0] * M->m[3][1] -
2373 M->m[0][2] * M->m[3][0] * M->m[2][1];
2374
2375 D->m[3][2] =
2376 -M->m[0][0] * M->m[1][1] * M->m[3][2] +
2377 M->m[0][0] * M->m[3][1] * M->m[1][2] +
2378 M->m[0][1] * M->m[1][0] * M->m[3][2] -
2379 M->m[0][1] * M->m[3][0] * M->m[1][2] -
2380 M->m[0][2] * M->m[1][0] * M->m[3][1] +
2381 M->m[0][2] * M->m[3][0] * M->m[1][1];
2382
2383 D->m[3][3] =
2384 M->m[0][0] * M->m[1][1] * M->m[2][2] -
2385 M->m[0][0] * M->m[2][1] * M->m[1][2] -
2386 M->m[0][1] * M->m[1][0] * M->m[2][2] +
2387 M->m[0][1] * M->m[2][0] * M->m[1][2] +
2388 M->m[0][2] * M->m[1][0] * M->m[2][1] -
2389 M->m[0][2] * M->m[2][0] * M->m[1][1];
2390
2391 det =
2392 M->m[0][0] * D->m[0][0] +
2393 M->m[1][0] * D->m[0][1] +
2394 M->m[2][0] * D->m[0][2] +
2395 M->m[3][0] * D->m[0][3];
2396
2397 det = 1.0 / det;
2398
2399 for (i = 0; i < 4; i++)
2400 for (k = 0; k < 4; k++)
2401 D->m[i][k] *= det;
2402
2403 #ifdef DEBUG
2404 {
2405 D3DMATRIX I;
2406
2407 nine_d3d_matrix_matrix_mul(&I, D, M);
2408
2409 for (i = 0; i < 4; ++i)
2410 for (k = 0; k < 4; ++k)
2411 if (fabsf(I.m[i][k] - (float)(i == k)) > 1e-3)
2412 DBG("Matrix inversion check FAILED !\n");
2413 }
2414 #endif
2415 }
2416
2417 /* TODO: don't use 4x4 inverse, unless this gets all nicely inlined ? */
2418 void
2419 nine_d3d_matrix_inverse_3x3(D3DMATRIX *D, const D3DMATRIX *M)
2420 {
2421 D3DMATRIX T;
2422 unsigned i, j;
2423
2424 for (i = 0; i < 3; ++i)
2425 for (j = 0; j < 3; ++j)
2426 T.m[i][j] = M->m[i][j];
2427 for (i = 0; i < 3; ++i) {
2428 T.m[i][3] = 0.0f;
2429 T.m[3][i] = 0.0f;
2430 }
2431 T.m[3][3] = 1.0f;
2432
2433 nine_d3d_matrix_inverse(D, &T);
2434 }