projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
tgsi/ureg: add support for GS input array declarations
[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
26 #define NINE_TGSI_LAZY_DEVS 1
27
28 #define DBG_CHANNEL DBG_FF
29
30 #define NINE_FF_NUM_VS_CONST 256
31 #define NINE_FF_NUM_PS_CONST 24
32
33 #define NINED3DTSS_TCI_DISABLE 0
34 #define NINED3DTSS_TCI_PASSTHRU 1
35 #define NINED3DTSS_TCI_CAMERASPACENORMAL 2
36 #define NINED3DTSS_TCI_CAMERASPACEPOSITION 3
37 #define NINED3DTSS_TCI_CAMERASPACEREFLECTIONVECTOR 4
38 #define NINED3DTSS_TCI_SPHEREMAP 5
39
40 struct fvec4
41 {
42 float x, y, z, w;
43 };
44
45 struct nine_ff_vs_key
46 {
47 union {
48 struct {
49 uint32_t position_t : 1;
50 uint32_t lighting : 1;
51 uint32_t darkness : 1; /* lighting enabled but no active lights */
52 uint32_t localviewer : 1;
53 uint32_t vertexpointsize : 1;
54 uint32_t pointscale : 1;
55 uint32_t vertexblend : 3;
56 uint32_t vertexblend_indexed : 1;
57 uint32_t vertextween : 1;
58 uint32_t mtl_diffuse : 2; /* 0 = material, 1 = color1, 2 = color2 */
59 uint32_t mtl_ambient : 2;
60 uint32_t mtl_specular : 2;
61 uint32_t mtl_emissive : 2;
62 uint32_t fog_mode : 2;
63 uint32_t fog_range : 1;
64 uint32_t color0in_one : 1;
65 uint32_t color1in_one : 1;
66 uint32_t pad1 : 8;
67 uint32_t tc_gen : 24; /* 8 * 3 bits */
68 uint32_t pad2 : 8;
69 uint32_t tc_idx : 24;
70 uint32_t pad3 : 8;
71 uint32_t tc_dim : 24; /* 8 * 3 bits */
72 uint32_t pad4 : 8;
73 };
74 uint64_t value64[2]; /* don't forget to resize VertexShader9.ff_key */
75 uint32_t value32[4];
76 };
77 };
78
79 /* Texture stage state:
80 *
81 * COLOROP D3DTOP 5 bit
82 * ALPHAOP D3DTOP 5 bit
83 * COLORARG0 D3DTA 3 bit
84 * COLORARG1 D3DTA 3 bit
85 * COLORARG2 D3DTA 3 bit
86 * ALPHAARG0 D3DTA 3 bit
87 * ALPHAARG1 D3DTA 3 bit
88 * ALPHAARG2 D3DTA 3 bit
89 * RESULTARG D3DTA 1 bit (CURRENT:0 or TEMP:1)
90 * TEXCOORDINDEX 0 - 7 3 bit
91 * ===========================
92 * 32 bit per stage
93 */
94 struct nine_ff_ps_key
95 {
96 union {
97 struct {
98 struct {
99 uint32_t colorop : 5;
100 uint32_t alphaop : 5;
101 uint32_t colorarg0 : 3;
102 uint32_t colorarg1 : 3;
103 uint32_t colorarg2 : 3;
104 uint32_t alphaarg0 : 3;
105 uint32_t alphaarg1 : 3;
106 uint32_t alphaarg2 : 3;
107 uint32_t resultarg : 1; /* CURRENT:0 or TEMP:1 */
108 uint32_t textarget : 2; /* 1D/2D/3D/CUBE */
109 uint32_t projected : 1;
110 /* that's 32 bit exactly */
111 } ts[8];
112 uint32_t fog : 1; /* for vFog with programmable VS */
113 uint32_t fog_mode : 2;
114 uint32_t specular : 1; /* 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 };
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 * CONST[29] RS.FogColor
226
227 * CONST[30].x___ TWEENFACTOR
228 *
229 * CONST[32].x___ LIGHT[0].Type
230 * CONST[32]._yzw LIGHT[0].Attenuation0,1,2
231 * CONST[33] LIGHT[0].Diffuse
232 * CONST[34] LIGHT[0].Specular
233 * CONST[35] LIGHT[0].Ambient
234 * CONST[36].xyz_ LIGHT[0].Position
235 * CONST[36].___w LIGHT[0].Range
236 * CONST[37].xyz_ LIGHT[0].Direction
237 * CONST[37].___w LIGHT[0].Falloff
238 * CONST[38].x___ cos(LIGHT[0].Theta / 2)
239 * CONST[38]._y__ cos(LIGHT[0].Phi / 2)
240 * CONST[38].__z_ 1.0f / (cos(LIGHT[0].Theta / 2) - cos(Light[0].Phi / 2))
241 * CONST[39].xyz_ LIGHT[0].HalfVector (for directional lights)
242 * CONST[39].___w 1 if this is the last active light, 0 if not
243 * CONST[40] LIGHT[1]
244 * CONST[48] LIGHT[2]
245 * CONST[56] LIGHT[3]
246 * CONST[64] LIGHT[4]
247 * CONST[72] LIGHT[5]
248 * CONST[80] LIGHT[6]
249 * CONST[88] LIGHT[7]
250 * NOTE: no lighting code is generated if there are no active lights
251 *
252 * CONST[100].x___ Viewport 2/width
253 * CONST[100]._y__ Viewport 2/height
254 * CONST[100].__z_ Viewport 1/(zmax - zmin)
255 * CONST[101].x___ Viewport x0
256 * CONST[101]._y__ Viewport y0
257 * CONST[101].__z_ Viewport z0
258 *
259 * CONST[128..131] D3DTS_TEXTURE0
260 * CONST[132..135] D3DTS_TEXTURE1
261 * CONST[136..139] D3DTS_TEXTURE2
262 * CONST[140..143] D3DTS_TEXTURE3
263 * CONST[144..147] D3DTS_TEXTURE4
264 * CONST[148..151] D3DTS_TEXTURE5
265 * CONST[152..155] D3DTS_TEXTURE6
266 * CONST[156..159] D3DTS_TEXTURE7
267 *
268 * CONST[224] D3DTS_WORLDMATRIX[0]
269 * CONST[228] D3DTS_WORLDMATRIX[1]
270 * ...
271 * CONST[252] D3DTS_WORLDMATRIX[7]
272 */
273 struct vs_build_ctx
274 {
275 struct ureg_program *ureg;
276 const struct nine_ff_vs_key *key;
277
278 uint16_t input[PIPE_MAX_ATTRIBS];
279 unsigned num_inputs;
280
281 struct ureg_src aVtx;
282 struct ureg_src aNrm;
283 struct ureg_src aCol[2];
284 struct ureg_src aTex[8];
285 struct ureg_src aPsz;
286 struct ureg_src aInd;
287 struct ureg_src aWgt;
288
289 struct ureg_src aVtx1; /* tweening */
290 struct ureg_src aNrm1;
291
292 struct ureg_src mtlA;
293 struct ureg_src mtlD;
294 struct ureg_src mtlS;
295 struct ureg_src mtlE;
296 };
297
298 static INLINE unsigned
299 get_texcoord_sn(struct pipe_screen *screen)
300 {
301 if (screen->get_param(screen, PIPE_CAP_TGSI_TEXCOORD))
302 return TGSI_SEMANTIC_TEXCOORD;
303 return TGSI_SEMANTIC_GENERIC;
304 }
305
306 static INLINE struct ureg_src
307 build_vs_add_input(struct vs_build_ctx *vs, uint16_t ndecl)
308 {
309 const unsigned i = vs->num_inputs++;
310 assert(i < PIPE_MAX_ATTRIBS);
311 vs->input[i] = ndecl;
312 return ureg_DECL_vs_input(vs->ureg, i);
313 }
314
315 /* NOTE: dst may alias src */
316 static INLINE void
317 ureg_normalize3(struct ureg_program *ureg,
318 struct ureg_dst dst, struct ureg_src src,
319 struct ureg_dst tmp)
320 {
321 #ifdef NINE_TGSI_LAZY_DEVS
322 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
323
324 ureg_DP3(ureg, tmp_x, src, src);
325 ureg_RSQ(ureg, tmp_x, _X(tmp));
326 ureg_MUL(ureg, dst, src, _X(tmp));
327 #else
328 ureg_NRM(ureg, dst, src);
329 #endif
330 }
331
332 static void *
333 nine_ff_build_vs(struct NineDevice9 *device, struct vs_build_ctx *vs)
334 {
335 const struct nine_ff_vs_key *key = vs->key;
336 struct ureg_program *ureg = ureg_create(TGSI_PROCESSOR_VERTEX);
337 struct ureg_dst oPos, oCol[2], oTex[8], oPsz, oFog;
338 struct ureg_dst rCol[2]; /* oCol if no fog, TEMP otherwise */
339 struct ureg_dst rVtx, rNrm;
340 struct ureg_dst r[8];
341 struct ureg_dst AR;
342 struct ureg_dst tmp, tmp_x, tmp_z;
343 unsigned i, c;
344 unsigned label[32], l = 0;
345 unsigned num_r = 8;
346 boolean need_rNrm = key->lighting || key->pointscale;
347 boolean need_rVtx = key->lighting || key->fog_mode;
348 const unsigned texcoord_sn = get_texcoord_sn(device->screen);
349
350 vs->ureg = ureg;
351
352 /* Check which inputs we should transform. */
353 for (i = 0; i < 8 * 3; i += 3) {
354 switch ((key->tc_gen >> i) & 0x3) {
355 case NINED3DTSS_TCI_CAMERASPACENORMAL:
356 need_rNrm = TRUE;
357 break;
358 case NINED3DTSS_TCI_CAMERASPACEPOSITION:
359 need_rVtx = TRUE;
360 break;
361 case NINED3DTSS_TCI_CAMERASPACEREFLECTIONVECTOR:
362 need_rVtx = need_rNrm = TRUE;
363 break;
364 default:
365 break;
366 }
367 }
368
369 /* Declare and record used inputs (needed for linkage with vertex format):
370 * (texture coordinates handled later)
371 */
372 vs->aVtx = build_vs_add_input(vs,
373 key->position_t ? NINE_DECLUSAGE_POSITIONT : NINE_DECLUSAGE_POSITION);
374
375 if (need_rNrm)
376 vs->aNrm = build_vs_add_input(vs, NINE_DECLUSAGE_NORMAL);
377
378 vs->aCol[0] = ureg_imm1f(ureg, 1.0f);
379 vs->aCol[1] = ureg_imm1f(ureg, 1.0f);
380
381 if (key->lighting || key->darkness) {
382 const unsigned mask = key->mtl_diffuse | key->mtl_specular |
383 key->mtl_ambient | key->mtl_emissive;
384 if ((mask & 0x1) && !key->color0in_one)
385 vs->aCol[0] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 0));
386 if ((mask & 0x2) && !key->color1in_one)
387 vs->aCol[1] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 1));
388
389 vs->mtlD = MATERIAL_CONST(1);
390 vs->mtlA = MATERIAL_CONST(2);
391 vs->mtlS = MATERIAL_CONST(3);
392 vs->mtlE = MATERIAL_CONST(5);
393 if (key->mtl_diffuse == 1) vs->mtlD = vs->aCol[0]; else
394 if (key->mtl_diffuse == 2) vs->mtlD = vs->aCol[1];
395 if (key->mtl_ambient == 1) vs->mtlA = vs->aCol[0]; else
396 if (key->mtl_ambient == 2) vs->mtlA = vs->aCol[1];
397 if (key->mtl_specular == 1) vs->mtlS = vs->aCol[0]; else
398 if (key->mtl_specular == 2) vs->mtlS = vs->aCol[1];
399 if (key->mtl_emissive == 1) vs->mtlE = vs->aCol[0]; else
400 if (key->mtl_emissive == 2) vs->mtlE = vs->aCol[1];
401 } else {
402 if (!key->color0in_one) vs->aCol[0] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 0));
403 if (!key->color1in_one) vs->aCol[1] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 1));
404 }
405
406 if (key->vertexpointsize)
407 vs->aPsz = build_vs_add_input(vs, NINE_DECLUSAGE_PSIZE);
408
409 if (key->vertexblend_indexed)
410 vs->aInd = build_vs_add_input(vs, NINE_DECLUSAGE_BLENDINDICES);
411 if (key->vertexblend)
412 vs->aWgt = build_vs_add_input(vs, NINE_DECLUSAGE_BLENDWEIGHT);
413 if (key->vertextween) {
414 vs->aVtx1 = build_vs_add_input(vs, NINE_DECLUSAGE_i(POSITION,1));
415 vs->aNrm1 = build_vs_add_input(vs, NINE_DECLUSAGE_i(NORMAL,1));
416 }
417
418 /* Declare outputs:
419 */
420 oPos = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0); /* HPOS */
421 oCol[0] = ureg_saturate(ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0));
422 oCol[1] = ureg_saturate(ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 1));
423
424 if (key->vertexpointsize || key->pointscale) {
425 oPsz = ureg_DECL_output_masked(ureg, TGSI_SEMANTIC_PSIZE, 0, TGSI_WRITEMASK_X);
426 oPsz = ureg_writemask(oPsz, TGSI_WRITEMASK_X);
427 }
428 if (key->fog_mode) {
429 /* We apply fog to the vertex colors, oFog is for programmable shaders only ?
430 */
431 oFog = ureg_DECL_output_masked(ureg, TGSI_SEMANTIC_FOG, 0, TGSI_WRITEMASK_X);
432 oFog = ureg_writemask(oFog, TGSI_WRITEMASK_X);
433 }
434
435 /* Declare TEMPs:
436 */
437 for (i = 0; i < num_r; ++i)
438 r[i] = ureg_DECL_local_temporary(ureg);
439 tmp = r[0];
440 tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
441 tmp_z = ureg_writemask(tmp, TGSI_WRITEMASK_Z);
442 if (key->lighting || key->vertexblend)
443 AR = ureg_DECL_address(ureg);
444
445 if (key->fog_mode) {
446 rCol[0] = r[2];
447 rCol[1] = r[3];
448 } else {
449 rCol[0] = oCol[0];
450 rCol[1] = oCol[1];
451 }
452
453 rVtx = ureg_writemask(r[1], TGSI_WRITEMASK_XYZ);
454 rNrm = ureg_writemask(r[2], TGSI_WRITEMASK_XYZ);
455
456 /* === Vertex transformation / vertex blending:
457 */
458 if (key->vertextween) {
459 assert(!key->vertexblend);
460 ureg_LRP(ureg, r[2], _XXXX(_CONST(30)), vs->aVtx, vs->aVtx1);
461 if (need_rNrm)
462 ureg_LRP(ureg, r[3], _XXXX(_CONST(30)), vs->aNrm, vs->aNrm1);
463 vs->aVtx = ureg_src(r[2]);
464 vs->aNrm = ureg_src(r[3]);
465 }
466
467 if (key->vertexblend) {
468 struct ureg_src cWM[4];
469
470 for (i = 224; i <= 255; ++i)
471 ureg_DECL_constant(ureg, i);
472
473 /* translate world matrix index to constant file index */
474 if (key->vertexblend_indexed) {
475 ureg_MAD(ureg, tmp, vs->aInd, ureg_imm1f(ureg, 4.0f), ureg_imm1f(ureg, 224.0f));
476 ureg_ARL(ureg, AR, ureg_src(tmp));
477 }
478 for (i = 0; i < key->vertexblend; ++i) {
479 for (c = 0; c < 4; ++c) {
480 cWM[c] = ureg_src_register(TGSI_FILE_CONSTANT, (224 + i * 4) * !key->vertexblend_indexed + c);
481 if (key->vertexblend_indexed)
482 cWM[c] = ureg_src_indirect(cWM[c], ureg_scalar(ureg_src(AR), i));
483 }
484 /* multiply by WORLD(index) */
485 ureg_MUL(ureg, r[0], _XXXX(vs->aVtx), cWM[0]);
486 ureg_MAD(ureg, r[0], _YYYY(vs->aVtx), cWM[1], ureg_src(r[0]));
487 ureg_MAD(ureg, r[0], _ZZZZ(vs->aVtx), cWM[2], ureg_src(r[0]));
488 ureg_MAD(ureg, r[0], _WWWW(vs->aVtx), cWM[3], ureg_src(r[0]));
489
490 /* accumulate weighted position value */
491 if (i)
492 ureg_MAD(ureg, r[2], ureg_src(r[0]), ureg_scalar(vs->aWgt, i), ureg_src(r[2]));
493 else
494 ureg_MUL(ureg, r[2], ureg_src(r[0]), ureg_scalar(vs->aWgt, 0));
495 }
496 /* multiply by VIEW_PROJ */
497 ureg_MUL(ureg, r[0], _X(r[2]), _CONST(8));
498 ureg_MAD(ureg, r[0], _Y(r[2]), _CONST(9), ureg_src(r[0]));
499 ureg_MAD(ureg, r[0], _Z(r[2]), _CONST(10), ureg_src(r[0]));
500 ureg_MAD(ureg, oPos, _W(r[2]), _CONST(11), ureg_src(r[0]));
501
502 if (need_rVtx)
503 vs->aVtx = ureg_src(r[2]);
504 } else
505 if (key->position_t && device->driver_caps.window_space_position_support) {
506 ureg_MOV(ureg, oPos, vs->aVtx);
507 } else if (key->position_t) {
508 /* vs->aVtx contains the coordinates buffer wise.
509 * later in the pipeline, clipping, viewport and division
510 * by w (rhw = 1/w) are going to be applied, so do the reverse
511 * of these transformations (except clipping) to have the good
512 * position at the end.*/
513 ureg_MOV(ureg, tmp, vs->aVtx);
514 /* X from [X_min, X_min + width] to [-1, 1], same for Y. Z to [0, 1] */
515 ureg_SUB(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _CONST(101));
516 ureg_MUL(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _CONST(100));
517 ureg_SUB(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XY), ureg_src(tmp), ureg_imm1f(ureg, 1.0f));
518 /* Y needs to be reversed */
519 ureg_MOV(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_Y), ureg_negate(ureg_src(tmp)));
520 /* inverse rhw */
521 ureg_RCP(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_W), _W(tmp));
522 /* multiply X, Y, Z by w */
523 ureg_MUL(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _W(tmp));
524 ureg_MOV(ureg, oPos, ureg_src(tmp));
525 } else {
526 /* position = vertex * WORLD_VIEW_PROJ */
527 ureg_MUL(ureg, r[0], _XXXX(vs->aVtx), _CONST(0));
528 ureg_MAD(ureg, r[0], _YYYY(vs->aVtx), _CONST(1), ureg_src(r[0]));
529 ureg_MAD(ureg, r[0], _ZZZZ(vs->aVtx), _CONST(2), ureg_src(r[0]));
530 ureg_MAD(ureg, oPos, _WWWW(vs->aVtx), _CONST(3), ureg_src(r[0]));
531 }
532
533 if (need_rVtx) {
534 ureg_MUL(ureg, rVtx, _XXXX(vs->aVtx), _CONST(4));
535 ureg_MAD(ureg, rVtx, _YYYY(vs->aVtx), _CONST(5), ureg_src(rVtx));
536 ureg_MAD(ureg, rVtx, _ZZZZ(vs->aVtx), _CONST(6), ureg_src(rVtx));
537 ureg_MAD(ureg, rVtx, _WWWW(vs->aVtx), _CONST(7), ureg_src(rVtx));
538 }
539 if (need_rNrm) {
540 ureg_MUL(ureg, rNrm, _XXXX(vs->aNrm), _CONST(16));
541 ureg_MAD(ureg, rNrm, _YYYY(vs->aNrm), _CONST(17), ureg_src(rNrm));
542 ureg_MAD(ureg, rNrm, _ZZZZ(vs->aNrm), _CONST(18), ureg_src(rNrm));
543 ureg_normalize3(ureg, rNrm, ureg_src(rNrm), tmp);
544 }
545 /* NOTE: don't use vs->aVtx, vs->aNrm after this line */
546
547 /* === Process point size:
548 */
549 if (key->vertexpointsize) {
550 struct ureg_src cPsz1 = ureg_DECL_constant(ureg, 26);
551 #ifdef NINE_TGSI_LAZY_DEVS
552 struct ureg_dst tmp_clamp = ureg_DECL_temporary(ureg);
553
554 ureg_MAX(ureg, tmp_clamp, vs->aPsz, _XXXX(cPsz1));
555 ureg_MIN(ureg, oPsz, ureg_src(tmp_clamp), _YYYY(cPsz1));
556 ureg_release_temporary(ureg, tmp_clamp);
557 #else
558 ureg_CLAMP(ureg, oPsz, vs->aPsz, _XXXX(cPsz1), _YYYY(cPsz1));
559 #endif
560 } else if (key->pointscale) {
561 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
562 struct ureg_dst tmp_y = ureg_writemask(tmp, TGSI_WRITEMASK_Y);
563 struct ureg_src cPsz1 = ureg_DECL_constant(ureg, 26);
564 struct ureg_src cPsz2 = ureg_DECL_constant(ureg, 27);
565
566 ureg_DP3(ureg, tmp_x, ureg_src(r[1]), ureg_src(r[1]));
567 ureg_SQRT(ureg, tmp_y, _X(tmp));
568 ureg_MAD(ureg, tmp_x, _Y(tmp), _YYYY(cPsz2), _XXXX(cPsz2));
569 ureg_MAD(ureg, tmp_x, _Y(tmp), _X(tmp), _WWWW(cPsz1));
570 ureg_RCP(ureg, tmp_x, ureg_src(tmp));
571 ureg_MUL(ureg, tmp_x, ureg_src(tmp), _ZZZZ(cPsz1));
572 #ifdef NINE_TGSI_LAZY_DEVS
573 struct ureg_dst tmp_clamp = ureg_DECL_temporary(ureg);
574
575 ureg_MAX(ureg, tmp_clamp, _X(tmp), _XXXX(cPsz1));
576 ureg_MIN(ureg, oPsz, ureg_src(tmp_clamp), _YYYY(cPsz1));
577 ureg_release_temporary(ureg, tmp_clamp);
578 #else
579 ureg_CLAMP(ureg, oPsz, _X(tmp), _XXXX(cPsz1), _YYYY(cPsz1));
580 #endif
581 }
582
583 /* Texture coordinate generation:
584 * XXX: D3DTTFF_PROJECTED, transform matrix
585 */
586 for (i = 0; i < 8; ++i) {
587 struct ureg_dst dst[5];
588 struct ureg_src src;
589 unsigned c;
590 const unsigned tci = (key->tc_gen >> (i * 3)) & 0x7;
591 const unsigned idx = (key->tc_idx >> (i * 3)) & 0x7;
592 const unsigned dim = (key->tc_dim >> (i * 3)) & 0x7;
593
594 if (tci == NINED3DTSS_TCI_DISABLE)
595 continue;
596 oTex[i] = ureg_DECL_output(ureg, texcoord_sn, i);
597
598 if (tci == NINED3DTSS_TCI_PASSTHRU)
599 vs->aTex[idx] = build_vs_add_input(vs, NINE_DECLUSAGE_i(TEXCOORD,idx));
600
601 if (!dim) {
602 dst[c = 4] = oTex[i];
603 } else {
604 dst[4] = r[5];
605 src = ureg_src(dst[4]);
606 for (c = 0; c < (dim - 1); ++c)
607 dst[c] = ureg_writemask(tmp, (1 << dim) - 1);
608 dst[c] = ureg_writemask(oTex[i], (1 << dim) - 1);
609 }
610
611 switch (tci) {
612 case NINED3DTSS_TCI_PASSTHRU:
613 ureg_MOV(ureg, dst[4], vs->aTex[idx]);
614 break;
615 case NINED3DTSS_TCI_CAMERASPACENORMAL:
616 assert(dim <= 3);
617 ureg_MOV(ureg, ureg_writemask(dst[4], TGSI_WRITEMASK_XYZ), ureg_src(rNrm));
618 ureg_MOV(ureg, ureg_writemask(dst[4], TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
619 break;
620 case NINED3DTSS_TCI_CAMERASPACEPOSITION:
621 ureg_MOV(ureg, ureg_writemask(dst[4], TGSI_WRITEMASK_XYZ), ureg_src(rVtx));
622 ureg_MOV(ureg, ureg_writemask(dst[4], TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
623 break;
624 case NINED3DTSS_TCI_CAMERASPACEREFLECTIONVECTOR:
625 tmp.WriteMask = TGSI_WRITEMASK_XYZ;
626 ureg_DP3(ureg, tmp_x, ureg_src(rVtx), ureg_src(rNrm));
627 ureg_MUL(ureg, tmp, ureg_src(rNrm), _X(tmp));
628 ureg_ADD(ureg, tmp, ureg_src(tmp), ureg_src(tmp));
629 ureg_SUB(ureg, ureg_writemask(dst[4], TGSI_WRITEMASK_XYZ), ureg_src(rVtx), ureg_src(tmp));
630 ureg_MOV(ureg, ureg_writemask(dst[4], TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
631 tmp.WriteMask = TGSI_WRITEMASK_XYZW;
632 break;
633 case NINED3DTSS_TCI_SPHEREMAP:
634 assert(!"TODO");
635 break;
636 default:
637 break;
638 }
639 if (!dim)
640 continue;
641 dst[c].WriteMask = ~dst[c].WriteMask;
642 if (dst[c].WriteMask)
643 ureg_MOV(ureg, dst[c], src); /* store untransformed components */
644 dst[c].WriteMask = ~dst[c].WriteMask;
645 if (dim > 0) ureg_MUL(ureg, dst[0], _XXXX(src), _CONST(128 + i * 4));
646 if (dim > 1) ureg_MAD(ureg, dst[1], _YYYY(src), _CONST(129 + i * 4), ureg_src(tmp));
647 if (dim > 2) ureg_MAD(ureg, dst[2], _ZZZZ(src), _CONST(130 + i * 4), ureg_src(tmp));
648 if (dim > 3) ureg_MAD(ureg, dst[3], _WWWW(src), _CONST(131 + i * 4), ureg_src(tmp));
649 }
650
651 /* === Lighting:
652 *
653 * DIRECTIONAL: Light at infinite distance, parallel rays, no attenuation.
654 * POINT: Finite distance to scene, divergent rays, isotropic, attenuation.
655 * SPOT: Finite distance, divergent rays, angular dependence, attenuation.
656 *
657 * vec3 normal = normalize(in.Normal * NormalMatrix);
658 * vec3 hitDir = light.direction;
659 * float atten = 1.0;
660 *
661 * if (light.type != DIRECTIONAL)
662 * {
663 * vec3 hitVec = light.position - eyeVertex;
664 * float d = length(hitVec);
665 * hitDir = hitVec / d;
666 * atten = 1 / ((light.atten2 * d + light.atten1) * d + light.atten0);
667 * }
668 *
669 * if (light.type == SPOTLIGHT)
670 * {
671 * float rho = dp3(-hitVec, light.direction);
672 * if (rho < cos(light.phi / 2))
673 * atten = 0;
674 * if (rho < cos(light.theta / 2))
675 * atten *= pow(some_func(rho), light.falloff);
676 * }
677 *
678 * float nDotHit = dp3_sat(normal, hitVec);
679 * float powFact = 0.0;
680 *
681 * if (nDotHit > 0.0)
682 * {
683 * vec3 midVec = normalize(hitDir + eye);
684 * float nDotMid = dp3_sat(normal, midVec);
685 * pFact = pow(nDotMid, material.power);
686 * }
687 *
688 * ambient += light.ambient * atten;
689 * diffuse += light.diffuse * atten * nDotHit;
690 * specular += light.specular * atten * powFact;
691 */
692 if (key->lighting) {
693 struct ureg_dst tmp_y = ureg_writemask(tmp, TGSI_WRITEMASK_Y);
694
695 struct ureg_dst rAtt = ureg_writemask(r[1], TGSI_WRITEMASK_W);
696 struct ureg_dst rHit = ureg_writemask(r[3], TGSI_WRITEMASK_XYZ);
697 struct ureg_dst rMid = ureg_writemask(r[4], TGSI_WRITEMASK_XYZ);
698
699 struct ureg_dst rCtr = ureg_writemask(r[2], TGSI_WRITEMASK_W);
700
701 struct ureg_dst AL = ureg_writemask(AR, TGSI_WRITEMASK_X);
702
703 /* Light.*.Alpha is not used. */
704 struct ureg_dst rD = ureg_writemask(r[5], TGSI_WRITEMASK_XYZ);
705 struct ureg_dst rA = ureg_writemask(r[6], TGSI_WRITEMASK_XYZ);
706 struct ureg_dst rS = ureg_writemask(r[7], TGSI_WRITEMASK_XYZ);
707
708 struct ureg_src mtlP = _XXXX(MATERIAL_CONST(4));
709
710 struct ureg_src cLKind = _XXXX(LIGHT_CONST(0));
711 struct ureg_src cLAtt0 = _YYYY(LIGHT_CONST(0));
712 struct ureg_src cLAtt1 = _ZZZZ(LIGHT_CONST(0));
713 struct ureg_src cLAtt2 = _WWWW(LIGHT_CONST(0));
714 struct ureg_src cLColD = _XYZW(LIGHT_CONST(1));
715 struct ureg_src cLColS = _XYZW(LIGHT_CONST(2));
716 struct ureg_src cLColA = _XYZW(LIGHT_CONST(3));
717 struct ureg_src cLPos = _XYZW(LIGHT_CONST(4));
718 struct ureg_src cLRng = _WWWW(LIGHT_CONST(4));
719 struct ureg_src cLDir = _XYZW(LIGHT_CONST(5));
720 struct ureg_src cLFOff = _WWWW(LIGHT_CONST(5));
721 struct ureg_src cLTht = _XXXX(LIGHT_CONST(6));
722 struct ureg_src cLPhi = _YYYY(LIGHT_CONST(6));
723 struct ureg_src cLSDiv = _ZZZZ(LIGHT_CONST(6));
724 struct ureg_src cLLast = _WWWW(LIGHT_CONST(7));
725
726 const unsigned loop_label = l++;
727
728 ureg_MOV(ureg, rCtr, ureg_imm1f(ureg, 32.0f)); /* &lightconst(0) */
729 ureg_MOV(ureg, rD, ureg_imm1f(ureg, 0.0f));
730 ureg_MOV(ureg, rA, ureg_imm1f(ureg, 0.0f));
731 ureg_MOV(ureg, rS, ureg_imm1f(ureg, 0.0f));
732 rD = ureg_saturate(rD);
733 rA = ureg_saturate(rA);
734 rS = ureg_saturate(rS);
735
736
737 /* loop management */
738 ureg_BGNLOOP(ureg, &label[loop_label]);
739 ureg_ARL(ureg, AL, _W(rCtr));
740
741 /* if (not DIRECTIONAL light): */
742 ureg_SNE(ureg, tmp_x, cLKind, ureg_imm1f(ureg, D3DLIGHT_DIRECTIONAL));
743 ureg_MOV(ureg, rHit, ureg_negate(cLDir));
744 ureg_MOV(ureg, rAtt, ureg_imm1f(ureg, 1.0f));
745 ureg_IF(ureg, _X(tmp), &label[l++]);
746 {
747 /* hitDir = light.position - eyeVtx
748 * d = length(hitDir)
749 * hitDir /= d
750 */
751 ureg_SUB(ureg, rHit, cLPos, ureg_src(rVtx));
752 ureg_DP3(ureg, tmp_x, ureg_src(rHit), ureg_src(rHit));
753 ureg_RSQ(ureg, tmp_y, _X(tmp));
754 ureg_MUL(ureg, rHit, ureg_src(rHit), _Y(tmp)); /* normalize */
755 ureg_MUL(ureg, tmp_x, _X(tmp), _Y(tmp)); /* length */
756
757 /* att = 1.0 / (light.att0 + (light.att1 + light.att2 * d) * d) */
758 ureg_MAD(ureg, rAtt, _X(tmp), cLAtt2, cLAtt1);
759 ureg_MAD(ureg, rAtt, _X(tmp), _W(rAtt), cLAtt0);
760 ureg_RCP(ureg, rAtt, _W(rAtt));
761 /* cut-off if distance exceeds Light.Range */
762 ureg_SLT(ureg, tmp_x, _X(tmp), cLRng);
763 ureg_MUL(ureg, rAtt, _W(rAtt), _X(tmp));
764 }
765 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
766 ureg_ENDIF(ureg);
767
768 /* if (SPOT light) */
769 ureg_SEQ(ureg, tmp_x, cLKind, ureg_imm1f(ureg, D3DLIGHT_SPOT));
770 ureg_IF(ureg, _X(tmp), &label[l++]);
771 {
772 /* rho = dp3(-hitDir, light.spotDir)
773 *
774 * if (rho > light.ctht2) NOTE: 0 <= phi <= pi, 0 <= theta <= phi
775 * spotAtt = 1
776 * else
777 * if (rho <= light.cphi2)
778 * spotAtt = 0
779 * else
780 * spotAtt = (rho - light.cphi2) / (light.ctht2 - light.cphi2) ^ light.falloff
781 */
782 ureg_DP3(ureg, tmp_y, ureg_negate(ureg_src(rHit)), cLDir); /* rho */
783 ureg_SUB(ureg, tmp_x, _Y(tmp), cLPhi);
784 ureg_MUL(ureg, tmp_x, _X(tmp), cLSDiv);
785 ureg_POW(ureg, tmp_x, _X(tmp), cLFOff); /* spotAtten */
786 ureg_SGE(ureg, tmp_z, _Y(tmp), cLTht); /* if inside theta && phi */
787 ureg_SGE(ureg, tmp_y, _Y(tmp), cLPhi); /* if inside phi */
788 ureg_MAD(ureg, ureg_saturate(tmp_x), _X(tmp), _Y(tmp), _Z(tmp));
789 ureg_MUL(ureg, rAtt, _W(rAtt), _X(tmp));
790 }
791 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
792 ureg_ENDIF(ureg);
793
794 /* directional factors, let's not use LIT because of clarity */
795 ureg_DP3(ureg, ureg_saturate(tmp_x), ureg_src(rNrm), ureg_src(rHit));
796 ureg_MOV(ureg, tmp_y, ureg_imm1f(ureg, 0.0f));
797 ureg_IF(ureg, _X(tmp), &label[l++]);
798 {
799 /* midVec = normalize(hitDir + eyeDir) */
800 if (key->localviewer) {
801 ureg_normalize3(ureg, rMid, ureg_src(rVtx), tmp);
802 ureg_ADD(ureg, rMid, ureg_src(rHit), ureg_negate(ureg_src(rMid)));
803 } else {
804 ureg_ADD(ureg, rMid, ureg_src(rHit), ureg_imm3f(ureg, 0.0f, 0.0f, 1.0f));
805 }
806 ureg_normalize3(ureg, rMid, ureg_src(rMid), tmp);
807 ureg_DP3(ureg, ureg_saturate(tmp_y), ureg_src(rNrm), ureg_src(rMid));
808 ureg_POW(ureg, tmp_y, _Y(tmp), mtlP);
809
810 ureg_MUL(ureg, tmp_x, _W(rAtt), _X(tmp)); /* dp3(normal,hitDir) * att */
811 ureg_MUL(ureg, tmp_y, _W(rAtt), _Y(tmp)); /* power factor * att */
812 ureg_MAD(ureg, rD, cLColD, _X(tmp), ureg_src(rD)); /* accumulate diffuse */
813 ureg_MAD(ureg, rS, cLColS, _Y(tmp), ureg_src(rS)); /* accumulate specular */
814 }
815 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
816 ureg_ENDIF(ureg);
817
818 ureg_MAD(ureg, rA, cLColA, _W(rAtt), ureg_src(rA)); /* accumulate ambient */
819
820 /* break if this was the last light */
821 ureg_IF(ureg, cLLast, &label[l++]);
822 ureg_BRK(ureg);
823 ureg_ENDIF(ureg);
824 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
825
826 ureg_ADD(ureg, rCtr, _W(rCtr), ureg_imm1f(ureg, 8.0f));
827 ureg_fixup_label(ureg, label[loop_label], ureg_get_instruction_number(ureg));
828 ureg_ENDLOOP(ureg, &label[loop_label]);
829
830 /* Set alpha factors of illumination to 1.0 for the multiplications. */
831 rD.WriteMask = TGSI_WRITEMASK_W; rD.Saturate = 0;
832 rS.WriteMask = TGSI_WRITEMASK_W; rS.Saturate = 0;
833 rA.WriteMask = TGSI_WRITEMASK_W; rA.Saturate = 0;
834 ureg_MOV(ureg, rD, ureg_imm1f(ureg, 1.0f));
835 ureg_MOV(ureg, rS, ureg_imm1f(ureg, 1.0f));
836
837 /* Apply to material:
838 *
839 * oCol[0] = (material.emissive + material.ambient * rs.ambient) +
840 * material.ambient * ambient +
841 * material.diffuse * diffuse +
842 * oCol[1] = material.specular * specular;
843 */
844 if (key->mtl_emissive == 0 && key->mtl_ambient == 0) {
845 ureg_MOV(ureg, rA, ureg_imm1f(ureg, 1.0f));
846 ureg_MAD(ureg, tmp, ureg_src(rA), vs->mtlA, _CONST(19));
847 } else {
848 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(rA), _CONST(25));
849 ureg_MAD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), vs->mtlA, ureg_src(tmp), vs->mtlE);
850 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_W ), vs->mtlA, vs->mtlE);
851 }
852 ureg_MAD(ureg, rCol[0], ureg_src(rD), vs->mtlD, ureg_src(tmp));
853 ureg_MUL(ureg, rCol[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, rCol[0], vs->mtlD, ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 1.0f), _CONST(19));
859 } else {
860 ureg_MAD(ureg, ureg_writemask(rCol[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(rCol[0], TGSI_WRITEMASK_W), vs->mtlD, _W(tmp));
863 }
864 ureg_MUL(ureg, rCol[1], ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 1.0f), vs->mtlS);
865 } else {
866 ureg_MOV(ureg, rCol[0], vs->aCol[0]);
867 ureg_MOV(ureg, rCol[1], vs->aCol[1]);
868 }
869
870 /* === Process fog.
871 *
872 * exp(x) = ex2(log2(e) * x)
873 */
874 if (key->fog_mode) {
875 /* Fog doesn't affect alpha, TODO: combine with light code output */
876 ureg_MOV(ureg, ureg_writemask(oCol[0], TGSI_WRITEMASK_W), _W(rCol[0]));
877 ureg_MOV(ureg, ureg_writemask(oCol[1], TGSI_WRITEMASK_W), _W(rCol[1]));
878
879 if (key->position_t) {
880 ureg_MOV(ureg, ureg_saturate(tmp_x), ureg_scalar(vs->aCol[1], TGSI_SWIZZLE_W));
881 } else
882 if (key->fog_range) {
883 ureg_DP3(ureg, tmp_x, ureg_src(rVtx), ureg_src(rVtx));
884 ureg_RSQ(ureg, tmp_z, _X(tmp));
885 ureg_MUL(ureg, tmp_z, _Z(tmp), _X(tmp));
886 } else {
887 ureg_MOV(ureg, tmp_z, ureg_abs(_Z(rVtx)));
888 }
889
890 if (key->fog_mode == D3DFOG_EXP) {
891 ureg_MUL(ureg, tmp_x, _Z(tmp), _ZZZZ(_CONST(28)));
892 ureg_MUL(ureg, tmp_x, _X(tmp), ureg_imm1f(ureg, -1.442695f));
893 ureg_EX2(ureg, tmp_x, _X(tmp));
894 } else
895 if (key->fog_mode == D3DFOG_EXP2) {
896 ureg_MUL(ureg, tmp_x, _Z(tmp), _ZZZZ(_CONST(28)));
897 ureg_MUL(ureg, tmp_x, _X(tmp), _X(tmp));
898 ureg_MUL(ureg, tmp_x, _X(tmp), ureg_imm1f(ureg, -1.442695f));
899 ureg_EX2(ureg, tmp_x, _X(tmp));
900 } else
901 if (key->fog_mode == D3DFOG_LINEAR && !key->position_t) {
902 ureg_SUB(ureg, tmp_x, _XXXX(_CONST(28)), _Z(tmp));
903 ureg_MUL(ureg, ureg_saturate(tmp_x), _X(tmp), _YYYY(_CONST(28)));
904 }
905 ureg_MOV(ureg, oFog, _X(tmp));
906 ureg_LRP(ureg, ureg_writemask(oCol[0], TGSI_WRITEMASK_XYZ), _X(tmp), ureg_src(rCol[0]), _CONST(29));
907 ureg_LRP(ureg, ureg_writemask(oCol[1], TGSI_WRITEMASK_XYZ), _X(tmp), ureg_src(rCol[1]), _CONST(29));
908 }
909
910 if (key->position_t && device->driver_caps.window_space_position_support)
911 ureg_property(ureg, TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION, TRUE);
912
913 ureg_END(ureg);
914 nine_ureg_tgsi_dump(ureg, FALSE);
915 return ureg_create_shader_and_destroy(ureg, device->pipe);
916 }
917
918 /* PS FF constants layout:
919 *
920 * CONST[ 0.. 7] stage[i].D3DTSS_CONSTANT
921 * CONST[ 8..15].x___ stage[i].D3DTSS_BUMPENVMAT00
922 * CONST[ 8..15]._y__ stage[i].D3DTSS_BUMPENVMAT01
923 * CONST[ 8..15].__z_ stage[i].D3DTSS_BUMPENVMAT10
924 * CONST[ 8..15].___w stage[i].D3DTSS_BUMPENVMAT11
925 * CONST[16..19].x_z_ stage[i].D3DTSS_BUMPENVLSCALE
926 * CONST[17..19]._y_w stage[i].D3DTSS_BUMPENVLOFFSET
927 *
928 * CONST[20] D3DRS_TEXTUREFACTOR
929 * CONST[21] D3DRS_FOGCOLOR
930 * CONST[22].x___ RS.FogEnd
931 * CONST[22]._y__ 1.0f / (RS.FogEnd - RS.FogStart)
932 * CONST[22].__z_ RS.FogDensity
933 */
934 struct ps_build_ctx
935 {
936 struct ureg_program *ureg;
937
938 struct ureg_src vC[2]; /* DIFFUSE, SPECULAR */
939 struct ureg_src vT[8]; /* TEXCOORD[i] */
940 struct ureg_dst r[6]; /* TEMPs */
941 struct ureg_dst rCur; /* D3DTA_CURRENT */
942 struct ureg_dst rMod;
943 struct ureg_src rCurSrc;
944 struct ureg_dst rTmp; /* D3DTA_TEMP */
945 struct ureg_src rTmpSrc;
946 struct ureg_dst rTex;
947 struct ureg_src rTexSrc;
948 struct ureg_src cBEM[8];
949 struct ureg_src s[8];
950
951 struct {
952 unsigned index;
953 unsigned index_pre_mod;
954 unsigned num_regs;
955 } stage;
956 };
957
958 static struct ureg_src
959 ps_get_ts_arg(struct ps_build_ctx *ps, unsigned ta)
960 {
961 struct ureg_src reg;
962
963 switch (ta & D3DTA_SELECTMASK) {
964 case D3DTA_CONSTANT:
965 reg = ureg_DECL_constant(ps->ureg, ps->stage.index);
966 break;
967 case D3DTA_CURRENT:
968 reg = (ps->stage.index == ps->stage.index_pre_mod) ? ureg_src(ps->rMod) : ps->rCurSrc;
969 break;
970 case D3DTA_DIFFUSE:
971 reg = ureg_DECL_fs_input(ps->ureg, TGSI_SEMANTIC_COLOR, 0, TGSI_INTERPOLATE_PERSPECTIVE);
972 break;
973 case D3DTA_SPECULAR:
974 reg = ureg_DECL_fs_input(ps->ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_PERSPECTIVE);
975 break;
976 case D3DTA_TEMP:
977 reg = ps->rTmpSrc;
978 break;
979 case D3DTA_TEXTURE:
980 reg = ps->rTexSrc;
981 break;
982 case D3DTA_TFACTOR:
983 reg = ureg_DECL_constant(ps->ureg, 20);
984 break;
985 default:
986 assert(0);
987 reg = ureg_src_undef();
988 break;
989 }
990 if (ta & D3DTA_COMPLEMENT) {
991 struct ureg_dst dst = ps->r[ps->stage.num_regs++];
992 ureg_SUB(ps->ureg, dst, ureg_imm1f(ps->ureg, 1.0f), reg);
993 reg = ureg_src(dst);
994 }
995 if (ta & D3DTA_ALPHAREPLICATE)
996 reg = _WWWW(reg);
997 return reg;
998 }
999
1000 static struct ureg_dst
1001 ps_get_ts_dst(struct ps_build_ctx *ps, unsigned ta)
1002 {
1003 assert(!(ta & (D3DTA_COMPLEMENT | D3DTA_ALPHAREPLICATE)));
1004
1005 switch (ta & D3DTA_SELECTMASK) {
1006 case D3DTA_CURRENT:
1007 return ps->rCur;
1008 case D3DTA_TEMP:
1009 return ps->rTmp;
1010 default:
1011 assert(0);
1012 return ureg_dst_undef();
1013 }
1014 }
1015
1016 static uint8_t ps_d3dtop_args_mask(D3DTEXTUREOP top)
1017 {
1018 switch (top) {
1019 case D3DTOP_DISABLE:
1020 return 0x0;
1021 case D3DTOP_SELECTARG1:
1022 case D3DTOP_PREMODULATE:
1023 return 0x2;
1024 case D3DTOP_SELECTARG2:
1025 return 0x4;
1026 case D3DTOP_MULTIPLYADD:
1027 case D3DTOP_LERP:
1028 return 0x7;
1029 default:
1030 return 0x6;
1031 }
1032 }
1033
1034 static INLINE boolean
1035 is_MOV_no_op(struct ureg_dst dst, struct ureg_src src)
1036 {
1037 return !dst.WriteMask ||
1038 (dst.File == src.File &&
1039 dst.Index == src.Index &&
1040 !dst.Indirect &&
1041 !dst.Saturate &&
1042 !src.Indirect &&
1043 !src.Negate &&
1044 !src.Absolute &&
1045 (!(dst.WriteMask & TGSI_WRITEMASK_X) || (src.SwizzleX == TGSI_SWIZZLE_X)) &&
1046 (!(dst.WriteMask & TGSI_WRITEMASK_Y) || (src.SwizzleY == TGSI_SWIZZLE_Y)) &&
1047 (!(dst.WriteMask & TGSI_WRITEMASK_Z) || (src.SwizzleZ == TGSI_SWIZZLE_Z)) &&
1048 (!(dst.WriteMask & TGSI_WRITEMASK_W) || (src.SwizzleW == TGSI_SWIZZLE_W)));
1049
1050 }
1051
1052 static void
1053 ps_do_ts_op(struct ps_build_ctx *ps, unsigned top, struct ureg_dst dst, struct ureg_src *arg)
1054 {
1055 struct ureg_program *ureg = ps->ureg;
1056 struct ureg_dst tmp = ps->r[ps->stage.num_regs];
1057 struct ureg_dst tmp2 = ps->r[ps->stage.num_regs+1];
1058 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
1059
1060 tmp.WriteMask = dst.WriteMask;
1061
1062 if (top != D3DTOP_SELECTARG1 && top != D3DTOP_SELECTARG2 &&
1063 top != D3DTOP_MODULATE && top != D3DTOP_PREMODULATE &&
1064 top != D3DTOP_BLENDDIFFUSEALPHA && top != D3DTOP_BLENDTEXTUREALPHA &&
1065 top != D3DTOP_BLENDFACTORALPHA && top != D3DTOP_BLENDCURRENTALPHA &&
1066 top != D3DTOP_BUMPENVMAP && top != D3DTOP_BUMPENVMAPLUMINANCE &&
1067 top != D3DTOP_LERP)
1068 dst = ureg_saturate(dst);
1069
1070 switch (top) {
1071 case D3DTOP_SELECTARG1:
1072 if (!is_MOV_no_op(dst, arg[1]))
1073 ureg_MOV(ureg, dst, arg[1]);
1074 break;
1075 case D3DTOP_SELECTARG2:
1076 if (!is_MOV_no_op(dst, arg[2]))
1077 ureg_MOV(ureg, dst, arg[2]);
1078 break;
1079 case D3DTOP_MODULATE:
1080 ureg_MUL(ureg, dst, arg[1], arg[2]);
1081 break;
1082 case D3DTOP_MODULATE2X:
1083 ureg_MUL(ureg, tmp, arg[1], arg[2]);
1084 ureg_ADD(ureg, dst, ureg_src(tmp), ureg_src(tmp));
1085 break;
1086 case D3DTOP_MODULATE4X:
1087 ureg_MUL(ureg, tmp, arg[1], arg[2]);
1088 ureg_MUL(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 4.0f));
1089 break;
1090 case D3DTOP_ADD:
1091 ureg_ADD(ureg, dst, arg[1], arg[2]);
1092 break;
1093 case D3DTOP_ADDSIGNED:
1094 ureg_ADD(ureg, tmp, arg[1], arg[2]);
1095 ureg_SUB(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 0.5f));
1096 break;
1097 case D3DTOP_ADDSIGNED2X:
1098 ureg_ADD(ureg, tmp, arg[1], arg[2]);
1099 ureg_MAD(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 2.0f), ureg_imm1f(ureg, -1.0f));
1100 break;
1101 case D3DTOP_SUBTRACT:
1102 ureg_SUB(ureg, dst, arg[1], arg[2]);
1103 break;
1104 case D3DTOP_ADDSMOOTH:
1105 ureg_SUB(ureg, tmp, ureg_imm1f(ureg, 1.0f), arg[1]);
1106 ureg_MAD(ureg, dst, ureg_src(tmp), arg[2], arg[1]);
1107 break;
1108 case D3DTOP_BLENDDIFFUSEALPHA:
1109 ureg_LRP(ureg, dst, _WWWW(ps->vC[0]), arg[1], arg[2]);
1110 break;
1111 case D3DTOP_BLENDTEXTUREALPHA:
1112 /* XXX: alpha taken from previous stage, texture or result ? */
1113 ureg_LRP(ureg, dst, _W(ps->rTex), arg[1], arg[2]);
1114 break;
1115 case D3DTOP_BLENDFACTORALPHA:
1116 ureg_LRP(ureg, dst, _WWWW(_CONST(20)), arg[1], arg[2]);
1117 break;
1118 case D3DTOP_BLENDTEXTUREALPHAPM:
1119 ureg_SUB(ureg, tmp_x, ureg_imm1f(ureg, 1.0f), _W(ps->rTex));
1120 ureg_MAD(ureg, dst, arg[2], _X(tmp), arg[1]);
1121 break;
1122 case D3DTOP_BLENDCURRENTALPHA:
1123 ureg_LRP(ureg, dst, _WWWW(ps->rCurSrc), arg[1], arg[2]);
1124 break;
1125 case D3DTOP_PREMODULATE:
1126 ureg_MOV(ureg, dst, arg[1]);
1127 ps->stage.index_pre_mod = ps->stage.index + 1;
1128 break;
1129 case D3DTOP_MODULATEALPHA_ADDCOLOR:
1130 ureg_MAD(ureg, dst, _WWWW(arg[1]), arg[2], arg[1]);
1131 break;
1132 case D3DTOP_MODULATECOLOR_ADDALPHA:
1133 ureg_MAD(ureg, dst, arg[1], arg[2], _WWWW(arg[1]));
1134 break;
1135 case D3DTOP_MODULATEINVALPHA_ADDCOLOR:
1136 ureg_SUB(ureg, tmp_x, ureg_imm1f(ureg, 1.0f), _WWWW(arg[1]));
1137 ureg_MAD(ureg, dst, _X(tmp), arg[2], arg[1]);
1138 break;
1139 case D3DTOP_MODULATEINVCOLOR_ADDALPHA:
1140 ureg_SUB(ureg, tmp, ureg_imm1f(ureg, 1.0f), arg[1]);
1141 ureg_MAD(ureg, dst, ureg_src(tmp), arg[2], _WWWW(arg[1]));
1142 break;
1143 case D3DTOP_BUMPENVMAP:
1144 break;
1145 case D3DTOP_BUMPENVMAPLUMINANCE:
1146 break;
1147 case D3DTOP_DOTPRODUCT3:
1148 ureg_SUB(ureg, tmp, arg[1], ureg_imm4f(ureg,0.5,0.5,0.5,0.5));
1149 ureg_SUB(ureg, tmp2, arg[2] , ureg_imm4f(ureg,0.5,0.5,0.5,0.5));
1150 ureg_DP3(ureg, tmp, ureg_src(tmp), ureg_src(tmp2));
1151 ureg_MUL(ureg, ureg_saturate(dst), ureg_src(tmp), ureg_imm4f(ureg,4.0,4.0,4.0,4.0));
1152 break;
1153 case D3DTOP_MULTIPLYADD:
1154 ureg_MAD(ureg, dst, arg[1], arg[2], arg[0]);
1155 break;
1156 case D3DTOP_LERP:
1157 ureg_LRP(ureg, dst, arg[0], arg[1], arg[2]);
1158 break;
1159 case D3DTOP_DISABLE:
1160 /* no-op ? */
1161 break;
1162 default:
1163 assert(!"invalid D3DTOP");
1164 break;
1165 }
1166 }
1167
1168 static void *
1169 nine_ff_build_ps(struct NineDevice9 *device, struct nine_ff_ps_key *key)
1170 {
1171 struct ps_build_ctx ps;
1172 struct ureg_program *ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT);
1173 struct ureg_dst oCol;
1174 unsigned i, s;
1175 const unsigned texcoord_sn = get_texcoord_sn(device->screen);
1176
1177 memset(&ps, 0, sizeof(ps));
1178 ps.ureg = ureg;
1179 ps.stage.index_pre_mod = -1;
1180
1181 ps.vC[0] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 0, TGSI_INTERPOLATE_PERSPECTIVE);
1182
1183 /* Declare all TEMPs we might need, serious drivers have a register allocator. */
1184 for (i = 0; i < Elements(ps.r); ++i)
1185 ps.r[i] = ureg_DECL_local_temporary(ureg);
1186 ps.rCur = ps.r[0];
1187 ps.rTmp = ps.r[1];
1188 ps.rTex = ps.r[2];
1189 ps.rCurSrc = ureg_src(ps.rCur);
1190 ps.rTmpSrc = ureg_src(ps.rTmp);
1191 ps.rTexSrc = ureg_src(ps.rTex);
1192
1193 for (s = 0; s < 8; ++s) {
1194 ps.s[s] = ureg_src_undef();
1195
1196 if (key->ts[s].colorop != D3DTOP_DISABLE) {
1197 if (key->ts[s].colorarg0 == D3DTA_SPECULAR ||
1198 key->ts[s].colorarg1 == D3DTA_SPECULAR ||
1199 key->ts[s].colorarg2 == D3DTA_SPECULAR)
1200 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_PERSPECTIVE);
1201
1202 if (key->ts[s].colorarg0 == D3DTA_TEXTURE ||
1203 key->ts[s].colorarg1 == D3DTA_TEXTURE ||
1204 key->ts[s].colorarg2 == D3DTA_TEXTURE) {
1205 ps.s[s] = ureg_DECL_sampler(ureg, s);
1206 ps.vT[s] = ureg_DECL_fs_input(ureg, texcoord_sn, s, TGSI_INTERPOLATE_PERSPECTIVE);
1207 }
1208 if (s && (key->ts[s - 1].colorop == D3DTOP_PREMODULATE ||
1209 key->ts[s - 1].alphaop == D3DTOP_PREMODULATE))
1210 ps.s[s] = ureg_DECL_sampler(ureg, s);
1211 }
1212
1213 if (key->ts[s].alphaop != D3DTOP_DISABLE) {
1214 if (key->ts[s].alphaarg0 == D3DTA_SPECULAR ||
1215 key->ts[s].alphaarg1 == D3DTA_SPECULAR ||
1216 key->ts[s].alphaarg2 == D3DTA_SPECULAR)
1217 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_PERSPECTIVE);
1218
1219 if (key->ts[s].alphaarg0 == D3DTA_TEXTURE ||
1220 key->ts[s].alphaarg1 == D3DTA_TEXTURE ||
1221 key->ts[s].alphaarg2 == D3DTA_TEXTURE) {
1222 ps.s[s] = ureg_DECL_sampler(ureg, s);
1223 ps.vT[s] = ureg_DECL_fs_input(ureg, texcoord_sn, s, TGSI_INTERPOLATE_PERSPECTIVE);
1224 }
1225 }
1226 }
1227 if (key->specular)
1228 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_PERSPECTIVE);
1229
1230 oCol = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
1231
1232 if (key->ts[0].colorop == D3DTOP_DISABLE &&
1233 key->ts[0].alphaop == D3DTOP_DISABLE)
1234 ureg_MOV(ureg, ps.rCur, ps.vC[0]);
1235 /* Or is it undefined then ? */
1236
1237 /* Run stages.
1238 */
1239 for (s = 0; s < 8; ++s) {
1240 unsigned colorarg[3];
1241 unsigned alphaarg[3];
1242 const uint8_t used_c = ps_d3dtop_args_mask(key->ts[s].colorop);
1243 const uint8_t used_a = ps_d3dtop_args_mask(key->ts[s].alphaop);
1244 struct ureg_dst dst;
1245 struct ureg_src arg[3];
1246
1247 if (key->ts[s].colorop == D3DTOP_DISABLE &&
1248 key->ts[s].alphaop == D3DTOP_DISABLE)
1249 continue;
1250 ps.stage.index = s;
1251 ps.stage.num_regs = 3;
1252
1253 DBG("STAGE[%u]: colorop=%s alphaop=%s\n", s,
1254 nine_D3DTOP_to_str(key->ts[s].colorop),
1255 nine_D3DTOP_to_str(key->ts[s].alphaop));
1256
1257 if (!ureg_src_is_undef(ps.s[s])) {
1258 unsigned target;
1259 switch (key->ts[s].textarget) {
1260 case 0: target = TGSI_TEXTURE_1D; break;
1261 case 1: target = TGSI_TEXTURE_2D; break;
1262 case 2: target = TGSI_TEXTURE_3D; break;
1263 case 3: target = TGSI_TEXTURE_CUBE; break;
1264 /* this is a 2 bit bitfield, do I really need a default case ? */
1265 }
1266
1267 /* sample the texture */
1268 if (key->ts[s].colorop == D3DTOP_BUMPENVMAP ||
1269 key->ts[s].colorop == D3DTOP_BUMPENVMAPLUMINANCE) {
1270 }
1271 if (key->ts[s].projected)
1272 ureg_TXP(ureg, ps.rTex, target, ps.vT[s], ps.s[s]);
1273 else
1274 ureg_TEX(ureg, ps.rTex, target, ps.vT[s], ps.s[s]);
1275 }
1276
1277 if (s == 0 &&
1278 (key->ts[0].resultarg != 0 /* not current */ ||
1279 key->ts[0].colorop == D3DTOP_DISABLE ||
1280 key->ts[0].alphaop == D3DTOP_DISABLE ||
1281 key->ts[0].colorop == D3DTOP_BLENDCURRENTALPHA ||
1282 key->ts[0].alphaop == D3DTOP_BLENDCURRENTALPHA ||
1283 key->ts[0].colorarg0 == D3DTA_CURRENT ||
1284 key->ts[0].colorarg1 == D3DTA_CURRENT ||
1285 key->ts[0].colorarg2 == D3DTA_CURRENT ||
1286 key->ts[0].alphaarg0 == D3DTA_CURRENT ||
1287 key->ts[0].alphaarg1 == D3DTA_CURRENT ||
1288 key->ts[0].alphaarg2 == D3DTA_CURRENT)
1289 ) {
1290 /* Initialize D3DTA_CURRENT.
1291 * (Yes we can do this before the loop but not until
1292 * NVE4 has an instruction scheduling pass.)
1293 */
1294 ureg_MOV(ureg, ps.rCur, ps.vC[0]);
1295 }
1296
1297 dst = ps_get_ts_dst(&ps, key->ts[s].resultarg ? D3DTA_TEMP : D3DTA_CURRENT);
1298
1299 if (ps.stage.index_pre_mod == ps.stage.index) {
1300 ps.rMod = ps.r[ps.stage.num_regs++];
1301 ureg_MUL(ureg, ps.rMod, ps.rCurSrc, ps.rTexSrc);
1302 }
1303
1304 colorarg[0] = (key->ts[s].colorarg0 | ((key->colorarg_b4[0] >> s) << 4) | ((key->colorarg_b5[0] >> s) << 5)) & 0x3f;
1305 colorarg[1] = (key->ts[s].colorarg1 | ((key->colorarg_b4[1] >> s) << 4) | ((key->colorarg_b5[1] >> s) << 5)) & 0x3f;
1306 colorarg[2] = (key->ts[s].colorarg2 | ((key->colorarg_b4[2] >> s) << 4) | ((key->colorarg_b5[2] >> s) << 5)) & 0x3f;
1307 alphaarg[0] = (key->ts[s].alphaarg0 | ((key->alphaarg_b4[0] >> s) << 4)) & 0x1f;
1308 alphaarg[1] = (key->ts[s].alphaarg1 | ((key->alphaarg_b4[1] >> s) << 4)) & 0x1f;
1309 alphaarg[2] = (key->ts[s].alphaarg2 | ((key->alphaarg_b4[2] >> s) << 4)) & 0x1f;
1310
1311 if (key->ts[s].colorop != key->ts[s].alphaop ||
1312 colorarg[0] != alphaarg[0] ||
1313 colorarg[1] != alphaarg[1] ||
1314 colorarg[2] != alphaarg[2])
1315 dst.WriteMask = TGSI_WRITEMASK_XYZ;
1316
1317 if (used_c & 0x1) arg[0] = ps_get_ts_arg(&ps, colorarg[0]);
1318 if (used_c & 0x2) arg[1] = ps_get_ts_arg(&ps, colorarg[1]);
1319 if (used_c & 0x4) arg[2] = ps_get_ts_arg(&ps, colorarg[2]);
1320 ps_do_ts_op(&ps, key->ts[s].colorop, dst, arg);
1321
1322 if (dst.WriteMask != TGSI_WRITEMASK_XYZW) {
1323 dst.WriteMask = TGSI_WRITEMASK_W;
1324
1325 if (used_a & 0x1) arg[0] = ps_get_ts_arg(&ps, alphaarg[0]);
1326 if (used_a & 0x2) arg[1] = ps_get_ts_arg(&ps, alphaarg[1]);
1327 if (used_a & 0x4) arg[2] = ps_get_ts_arg(&ps, alphaarg[2]);
1328 ps_do_ts_op(&ps, key->ts[s].alphaop, dst, arg);
1329 }
1330 }
1331
1332 if (key->specular)
1333 ureg_ADD(ureg, ps.rCur, ps.rCurSrc, ps.vC[1]);
1334
1335 /* Fog.
1336 */
1337 if (key->fog_mode) {
1338 struct ureg_src vPos = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_POSITION, 0, TGSI_INTERPOLATE_LINEAR);
1339 struct ureg_dst rFog = ureg_writemask(ps.rTmp, TGSI_WRITEMASK_X);
1340 if (key->fog_mode == D3DFOG_EXP) {
1341 ureg_MUL(ureg, rFog, _ZZZZ(vPos), _ZZZZ(_CONST(22)));
1342 ureg_MUL(ureg, rFog, _X(rFog), ureg_imm1f(ureg, -1.442695f));
1343 ureg_EX2(ureg, rFog, _X(rFog));
1344 } else
1345 if (key->fog_mode == D3DFOG_EXP2) {
1346 ureg_MUL(ureg, rFog, _ZZZZ(vPos), _ZZZZ(_CONST(22)));
1347 ureg_MUL(ureg, rFog, _X(rFog), _X(rFog));
1348 ureg_MUL(ureg, rFog, _X(rFog), ureg_imm1f(ureg, -1.442695f));
1349 ureg_EX2(ureg, rFog, _X(rFog));
1350 } else
1351 if (key->fog_mode == D3DFOG_LINEAR) {
1352 ureg_SUB(ureg, rFog, _XXXX(_CONST(22)), _ZZZZ(vPos));
1353 ureg_MUL(ureg, ureg_saturate(rFog), _X(rFog), _YYYY(_CONST(22)));
1354 }
1355 ureg_LRP(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_XYZ), _X(rFog), ps.rCurSrc, _CONST(21));
1356 ureg_MOV(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_W), ps.rCurSrc);
1357 } else
1358 if (key->fog) {
1359 struct ureg_src vFog = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_FOG, 0, TGSI_INTERPOLATE_PERSPECTIVE);
1360 ureg_LRP(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_XYZ), _XXXX(vFog), ps.rCurSrc, _CONST(21));
1361 ureg_MOV(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_W), ps.rCurSrc);
1362 } else {
1363 ureg_MOV(ureg, oCol, ps.rCurSrc);
1364 }
1365
1366 ureg_END(ureg);
1367 nine_ureg_tgsi_dump(ureg, FALSE);
1368 return ureg_create_shader_and_destroy(ureg, device->pipe);
1369 }
1370
1371 static struct NineVertexShader9 *
1372 nine_ff_get_vs(struct NineDevice9 *device)
1373 {
1374 const struct nine_state *state = &device->state;
1375 struct NineVertexShader9 *vs;
1376 enum pipe_error err;
1377 struct vs_build_ctx bld;
1378 struct nine_ff_vs_key key;
1379 unsigned s, i;
1380 char input_texture_coord[8];
1381
1382 assert(sizeof(key) <= sizeof(key.value32));
1383
1384 memset(&key, 0, sizeof(key));
1385 memset(&bld, 0, sizeof(bld));
1386 memset(&input_texture_coord, 0, sizeof(input_texture_coord));
1387
1388 bld.key = &key;
1389
1390 /* FIXME: this shouldn't be NULL, but it is on init */
1391 if (state->vdecl) {
1392 key.color0in_one = 1;
1393 key.color1in_one = 1;
1394 for (i = 0; i < state->vdecl->nelems; i++) {
1395 uint16_t usage = state->vdecl->usage_map[i];
1396 if (usage == NINE_DECLUSAGE_POSITIONT)
1397 key.position_t = 1;
1398 else if (usage == NINE_DECLUSAGE_i(COLOR, 0))
1399 key.color0in_one = 0;
1400 else if (usage == NINE_DECLUSAGE_i(COLOR, 1))
1401 key.color1in_one = 0;
1402 else if (usage == NINE_DECLUSAGE_PSIZE)
1403 key.vertexpointsize = 1;
1404 else if (usage % NINE_DECLUSAGE_COUNT == NINE_DECLUSAGE_TEXCOORD) {
1405 s = usage / NINE_DECLUSAGE_COUNT;
1406 if (s < 8)
1407 input_texture_coord[s] = 1;
1408 else
1409 DBG("FF given texture coordinate >= 8. Ignoring\n");
1410 }
1411 }
1412 }
1413 if (!key.vertexpointsize)
1414 key.pointscale = !!state->rs[D3DRS_POINTSCALEENABLE];
1415
1416 key.lighting = !!state->rs[D3DRS_LIGHTING] && state->ff.num_lights_active;
1417 key.darkness = !!state->rs[D3DRS_LIGHTING] && !state->ff.num_lights_active;
1418 if (key.position_t) {
1419 key.darkness = 0; /* |= key.lighting; */ /* XXX ? */
1420 key.lighting = 0;
1421 }
1422 if ((key.lighting | key.darkness) && state->rs[D3DRS_COLORVERTEX]) {
1423 key.mtl_diffuse = state->rs[D3DRS_DIFFUSEMATERIALSOURCE];
1424 key.mtl_ambient = state->rs[D3DRS_AMBIENTMATERIALSOURCE];
1425 key.mtl_specular = state->rs[D3DRS_SPECULARMATERIALSOURCE];
1426 key.mtl_emissive = state->rs[D3DRS_EMISSIVEMATERIALSOURCE];
1427 }
1428 key.fog_mode = state->rs[D3DRS_FOGENABLE] ? state->rs[D3DRS_FOGVERTEXMODE] : 0;
1429 if (key.fog_mode)
1430 key.fog_range = !key.position_t && state->rs[D3DRS_RANGEFOGENABLE];
1431
1432 if (state->rs[D3DRS_VERTEXBLEND] != D3DVBF_DISABLE) {
1433 key.vertexblend_indexed = !!state->rs[D3DRS_INDEXEDVERTEXBLENDENABLE];
1434
1435 switch (state->rs[D3DRS_VERTEXBLEND]) {
1436 case D3DVBF_0WEIGHTS: key.vertexblend = key.vertexblend_indexed; break;
1437 case D3DVBF_1WEIGHTS: key.vertexblend = 2; break;
1438 case D3DVBF_2WEIGHTS: key.vertexblend = 3; break;
1439 case D3DVBF_3WEIGHTS: key.vertexblend = 4; break;
1440 case D3DVBF_TWEENING: key.vertextween = 1; break;
1441 default:
1442 assert(!"invalid D3DVBF");
1443 break;
1444 }
1445 }
1446
1447 for (s = 0; s < 8; ++s) {
1448 unsigned gen = (state->ff.tex_stage[s][D3DTSS_TEXCOORDINDEX] >> 16) + 1;
1449 unsigned dim = MIN2(state->ff.tex_stage[s][D3DTSS_TEXTURETRANSFORMFLAGS] & 0x7, 4);
1450
1451 if (key.position_t && gen > NINED3DTSS_TCI_PASSTHRU)
1452 gen = NINED3DTSS_TCI_PASSTHRU;
1453
1454 if (!input_texture_coord[s] && gen == NINED3DTSS_TCI_PASSTHRU)
1455 gen = NINED3DTSS_TCI_DISABLE;
1456
1457 key.tc_gen |= gen << (s * 3);
1458 key.tc_idx |= (state->ff.tex_stage[s][D3DTSS_TEXCOORDINDEX] & 7) << (s * 3);
1459 key.tc_dim |= dim << (s * 3);
1460 }
1461
1462 vs = util_hash_table_get(device->ff.ht_vs, &key);
1463 if (vs)
1464 return vs;
1465 NineVertexShader9_new(device, &vs, NULL, nine_ff_build_vs(device, &bld));
1466
1467 nine_ff_prune_vs(device);
1468 if (vs) {
1469 unsigned n;
1470
1471 memcpy(&vs->ff_key, &key, sizeof(vs->ff_key));
1472
1473 err = util_hash_table_set(device->ff.ht_vs, &vs->ff_key, vs);
1474 assert(err == PIPE_OK);
1475 device->ff.num_vs++;
1476 NineUnknown_ConvertRefToBind(NineUnknown(vs));
1477
1478 vs->num_inputs = bld.num_inputs;
1479 for (n = 0; n < bld.num_inputs; ++n)
1480 vs->input_map[n].ndecl = bld.input[n];
1481
1482 vs->position_t = key.position_t;
1483 vs->point_size = key.vertexpointsize | key.pointscale;
1484 }
1485 return vs;
1486 }
1487
1488 static struct NinePixelShader9 *
1489 nine_ff_get_ps(struct NineDevice9 *device)
1490 {
1491 struct nine_state *state = &device->state;
1492 struct NinePixelShader9 *ps;
1493 enum pipe_error err;
1494 struct nine_ff_ps_key key;
1495 unsigned s;
1496 uint8_t sampler_mask = 0;
1497
1498 assert(sizeof(key) <= sizeof(key.value32));
1499
1500 memset(&key, 0, sizeof(key));
1501 for (s = 0; s < 8; ++s) {
1502 key.ts[s].colorop = state->ff.tex_stage[s][D3DTSS_COLOROP];
1503 key.ts[s].alphaop = state->ff.tex_stage[s][D3DTSS_ALPHAOP];
1504 /* MSDN says D3DTOP_DISABLE disables this and all subsequent stages. */
1505 /* ALPHAOP cannot be disabled if COLOROP is enabled. */
1506 if (key.ts[s].colorop == D3DTOP_DISABLE) {
1507 key.ts[s].alphaop = D3DTOP_DISABLE; /* DISABLE == 1, avoid degenerate keys */
1508 break;
1509 }
1510
1511 if (!state->texture[s] &&
1512 state->ff.tex_stage[s][D3DTSS_COLORARG1] == D3DTA_TEXTURE) {
1513 /* This should also disable the stage. */
1514 key.ts[s].colorop = key.ts[s].alphaop = D3DTOP_DISABLE;
1515 break;
1516 }
1517
1518 if (state->ff.tex_stage[s][D3DTSS_COLORARG1] == D3DTA_TEXTURE)
1519 sampler_mask |= (1 << s);
1520
1521 if (key.ts[s].colorop != D3DTOP_DISABLE) {
1522 uint8_t used_c = ps_d3dtop_args_mask(key.ts[s].colorop);
1523 if (used_c & 0x1) key.ts[s].colorarg0 = state->ff.tex_stage[s][D3DTSS_COLORARG0];
1524 if (used_c & 0x2) key.ts[s].colorarg1 = state->ff.tex_stage[s][D3DTSS_COLORARG1];
1525 if (used_c & 0x4) key.ts[s].colorarg2 = state->ff.tex_stage[s][D3DTSS_COLORARG2];
1526 if (used_c & 0x1) key.colorarg_b4[0] |= (state->ff.tex_stage[s][D3DTSS_COLORARG0] >> 4) << s;
1527 if (used_c & 0x1) key.colorarg_b5[0] |= (state->ff.tex_stage[s][D3DTSS_COLORARG0] >> 5) << s;
1528 if (used_c & 0x2) key.colorarg_b4[1] |= (state->ff.tex_stage[s][D3DTSS_COLORARG1] >> 4) << s;
1529 if (used_c & 0x2) key.colorarg_b5[1] |= (state->ff.tex_stage[s][D3DTSS_COLORARG1] >> 5) << s;
1530 if (used_c & 0x4) key.colorarg_b4[2] |= (state->ff.tex_stage[s][D3DTSS_COLORARG2] >> 4) << s;
1531 if (used_c & 0x4) key.colorarg_b5[2] |= (state->ff.tex_stage[s][D3DTSS_COLORARG2] >> 5) << s;
1532 }
1533 if (key.ts[s].alphaop != D3DTOP_DISABLE) {
1534 uint8_t used_a = ps_d3dtop_args_mask(key.ts[s].alphaop);
1535 if (used_a & 0x1) key.ts[s].alphaarg0 = state->ff.tex_stage[s][D3DTSS_ALPHAARG0];
1536 if (used_a & 0x2) key.ts[s].alphaarg1 = state->ff.tex_stage[s][D3DTSS_ALPHAARG1];
1537 if (used_a & 0x4) key.ts[s].alphaarg2 = state->ff.tex_stage[s][D3DTSS_ALPHAARG2];
1538 if (used_a & 0x1) key.alphaarg_b4[0] |= (state->ff.tex_stage[s][D3DTSS_ALPHAARG0] >> 4) << s;
1539 if (used_a & 0x2) key.alphaarg_b4[1] |= (state->ff.tex_stage[s][D3DTSS_ALPHAARG1] >> 4) << s;
1540 if (used_a & 0x4) key.alphaarg_b4[2] |= (state->ff.tex_stage[s][D3DTSS_ALPHAARG2] >> 4) << s;
1541 }
1542 key.ts[s].resultarg = state->ff.tex_stage[s][D3DTSS_RESULTARG] == D3DTA_TEMP;
1543
1544 key.ts[s].projected = !!(state->ff.tex_stage[s][D3DTSS_TEXTURETRANSFORMFLAGS] & D3DTTFF_PROJECTED);
1545
1546 if (state->texture[s]) {
1547 switch (state->texture[s]->base.type) {
1548 case D3DRTYPE_TEXTURE: key.ts[s].textarget = 1; break;
1549 case D3DRTYPE_VOLUMETEXTURE: key.ts[s].textarget = 2; break;
1550 case D3DRTYPE_CUBETEXTURE: key.ts[s].textarget = 3; break;
1551 default:
1552 assert(!"unexpected texture type");
1553 break;
1554 }
1555 } else {
1556 key.ts[s].textarget = 1;
1557 }
1558 }
1559 for (; s < 8; ++s)
1560 key.ts[s].colorop = key.ts[s].alphaop = D3DTOP_DISABLE;
1561 if (state->rs[D3DRS_FOGENABLE])
1562 key.fog_mode = state->rs[D3DRS_FOGTABLEMODE];
1563
1564 ps = util_hash_table_get(device->ff.ht_ps, &key);
1565 if (ps)
1566 return ps;
1567 NinePixelShader9_new(device, &ps, NULL, nine_ff_build_ps(device, &key));
1568
1569 nine_ff_prune_ps(device);
1570 if (ps) {
1571 memcpy(&ps->ff_key, &key, sizeof(ps->ff_key));
1572
1573 err = util_hash_table_set(device->ff.ht_ps, &ps->ff_key, ps);
1574 assert(err == PIPE_OK);
1575 device->ff.num_ps++;
1576 NineUnknown_ConvertRefToBind(NineUnknown(ps));
1577
1578 ps->rt_mask = 0x1;
1579 ps->sampler_mask = sampler_mask;
1580 }
1581 return ps;
1582 }
1583
1584 #define GET_D3DTS(n) nine_state_access_transform(state, D3DTS_##n, FALSE)
1585 #define IS_D3DTS_DIRTY(s,n) ((s)->ff.changed.transform[(D3DTS_##n) / 32] & (1 << ((D3DTS_##n) % 32)))
1586 static void
1587 nine_ff_load_vs_transforms(struct NineDevice9 *device)
1588 {
1589 struct nine_state *state = &device->state;
1590 D3DMATRIX T;
1591 D3DMATRIX *M = (D3DMATRIX *)device->ff.vs_const;
1592 unsigned i;
1593
1594 /* TODO: make this nicer, and only upload the ones we need */
1595 /* TODO: use ff.vs_const as storage of W, V, P matrices */
1596
1597 if (IS_D3DTS_DIRTY(state, WORLD) ||
1598 IS_D3DTS_DIRTY(state, VIEW) ||
1599 IS_D3DTS_DIRTY(state, PROJECTION)) {
1600 /* WVP, WV matrices */
1601 nine_d3d_matrix_matrix_mul(&M[1], GET_D3DTS(WORLD), GET_D3DTS(VIEW));
1602 nine_d3d_matrix_matrix_mul(&M[0], &M[1], GET_D3DTS(PROJECTION));
1603
1604 /* normal matrix == transpose(inverse(WV)) */
1605 nine_d3d_matrix_inverse_3x3(&T, &M[1]);
1606 nine_d3d_matrix_transpose(&M[4], &T);
1607
1608 /* VP matrix */
1609 nine_d3d_matrix_matrix_mul(&M[2], GET_D3DTS(VIEW), GET_D3DTS(PROJECTION));
1610
1611 /* V and W matrix */
1612 M[3] = *GET_D3DTS(VIEW);
1613 M[56] = *GET_D3DTS(WORLD);
1614 }
1615
1616 if (state->rs[D3DRS_VERTEXBLEND] != D3DVBF_DISABLE) {
1617 /* load other world matrices */
1618 for (i = 1; i <= 7; ++i)
1619 M[56 + i] = *GET_D3DTS(WORLDMATRIX(i));
1620 }
1621
1622 device->ff.vs_const[30 * 4] = asfloat(state->rs[D3DRS_TWEENFACTOR]);
1623 }
1624
1625 static void
1626 nine_ff_load_lights(struct NineDevice9 *device)
1627 {
1628 struct nine_state *state = &device->state;
1629 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1630 unsigned l;
1631
1632 if (state->changed.group & NINE_STATE_FF_MATERIAL) {
1633 const D3DMATERIAL9 *mtl = &state->ff.material;
1634
1635 memcpy(&dst[20], &mtl->Diffuse, 4 * sizeof(float));
1636 memcpy(&dst[21], &mtl->Ambient, 4 * sizeof(float));
1637 memcpy(&dst[22], &mtl->Specular, 4 * sizeof(float));
1638 dst[23].x = mtl->Power;
1639 memcpy(&dst[24], &mtl->Emissive, 4 * sizeof(float));
1640 d3dcolor_to_rgba(&dst[25].x, state->rs[D3DRS_AMBIENT]);
1641 dst[19].x = dst[25].x * mtl->Ambient.r + mtl->Emissive.r;
1642 dst[19].y = dst[25].y * mtl->Ambient.g + mtl->Emissive.g;
1643 dst[19].z = dst[25].z * mtl->Ambient.b + mtl->Emissive.b;
1644 dst[19].w = mtl->Ambient.a + mtl->Emissive.a;
1645 }
1646
1647 if (!(state->changed.group & NINE_STATE_FF_LIGHTING))
1648 return;
1649
1650 for (l = 0; l < state->ff.num_lights_active; ++l) {
1651 const D3DLIGHT9 *light = &state->ff.light[state->ff.active_light[l]];
1652
1653 dst[32 + l * 8].x = light->Type;
1654 dst[32 + l * 8].y = light->Attenuation0;
1655 dst[32 + l * 8].z = light->Attenuation1;
1656 dst[32 + l * 8].w = light->Attenuation2;
1657 memcpy(&dst[33 + l * 8].x, &light->Diffuse, sizeof(light->Diffuse));
1658 memcpy(&dst[34 + l * 8].x, &light->Specular, sizeof(light->Specular));
1659 memcpy(&dst[35 + l * 8].x, &light->Ambient, sizeof(light->Ambient));
1660 nine_d3d_vector4_matrix_mul((D3DVECTOR *)&dst[36 + l * 8].x, &light->Position, GET_D3DTS(VIEW));
1661 nine_d3d_vector3_matrix_mul((D3DVECTOR *)&dst[37 + l * 8].x, &light->Direction, GET_D3DTS(VIEW));
1662 dst[36 + l * 8].w = light->Type == D3DLIGHT_DIRECTIONAL ? 1e9f : light->Range;
1663 dst[37 + l * 8].w = light->Falloff;
1664 dst[38 + l * 8].x = cosf(light->Theta * 0.5f);
1665 dst[38 + l * 8].y = cosf(light->Phi * 0.5f);
1666 dst[38 + l * 8].z = 1.0f / (dst[38 + l * 8].x - dst[38 + l * 8].y);
1667 dst[39 + l * 8].w = (l + 1) == state->ff.num_lights_active;
1668 }
1669 }
1670
1671 static void
1672 nine_ff_load_point_and_fog_params(struct NineDevice9 *device)
1673 {
1674 const struct nine_state *state = &device->state;
1675 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1676
1677 if (!(state->changed.group & NINE_STATE_FF_OTHER))
1678 return;
1679 dst[26].x = asfloat(state->rs[D3DRS_POINTSIZE_MIN]);
1680 dst[26].y = asfloat(state->rs[D3DRS_POINTSIZE_MAX]);
1681 dst[26].z = asfloat(state->rs[D3DRS_POINTSIZE]);
1682 dst[26].w = asfloat(state->rs[D3DRS_POINTSCALE_A]);
1683 dst[27].x = asfloat(state->rs[D3DRS_POINTSCALE_B]);
1684 dst[27].y = asfloat(state->rs[D3DRS_POINTSCALE_C]);
1685 dst[28].x = asfloat(state->rs[D3DRS_FOGEND]);
1686 dst[28].y = 1.0f / (asfloat(state->rs[D3DRS_FOGEND]) - asfloat(state->rs[D3DRS_FOGSTART]));
1687 if (isinf(dst[28].y))
1688 dst[28].y = 0.0f;
1689 dst[28].z = asfloat(state->rs[D3DRS_FOGDENSITY]);
1690 d3dcolor_to_rgba(&dst[29].x, state->rs[D3DRS_FOGCOLOR]);
1691 }
1692
1693 static void
1694 nine_ff_load_tex_matrices(struct NineDevice9 *device)
1695 {
1696 struct nine_state *state = &device->state;
1697 D3DMATRIX *M = (D3DMATRIX *)device->ff.vs_const;
1698 unsigned s;
1699
1700 if (!(state->ff.changed.transform[0] & 0xff0000))
1701 return;
1702 for (s = 0; s < 8; ++s) {
1703 if (IS_D3DTS_DIRTY(state, TEXTURE0 + s))
1704 M[32 + s] = *nine_state_access_transform(state, D3DTS_TEXTURE0 + s, FALSE);
1705 }
1706 }
1707
1708 static void
1709 nine_ff_load_ps_params(struct NineDevice9 *device)
1710 {
1711 const struct nine_state *state = &device->state;
1712 struct fvec4 *dst = (struct fvec4 *)device->ff.ps_const;
1713 unsigned s;
1714
1715 if (!(state->changed.group & (NINE_STATE_FF_PSSTAGES | NINE_STATE_FF_OTHER)))
1716 return;
1717
1718 for (s = 0; s < 8; ++s)
1719 d3dcolor_to_rgba(&dst[s].x, state->ff.tex_stage[s][D3DTSS_CONSTANT]);
1720
1721 for (s = 0; s < 8; ++s) {
1722 dst[8 + s].x = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT00]);
1723 dst[8 + s].y = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT01]);
1724 dst[8 + s].z = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT10]);
1725 dst[8 + s].w = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVMAT11]);
1726 if (s & 1) {
1727 dst[8 + s / 2].z = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLSCALE]);
1728 dst[8 + s / 2].w = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLOFFSET]);
1729 } else {
1730 dst[8 + s / 2].x = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLSCALE]);
1731 dst[8 + s / 2].y = asfloat(state->ff.tex_stage[s][D3DTSS_BUMPENVLOFFSET]);
1732 }
1733 }
1734
1735 d3dcolor_to_rgba(&dst[20].x, state->rs[D3DRS_TEXTUREFACTOR]);
1736 d3dcolor_to_rgba(&dst[21].x, state->rs[D3DRS_FOGCOLOR]);
1737 dst[22].x = asfloat(state->rs[D3DRS_FOGEND]);
1738 dst[22].y = 1.0f / (asfloat(state->rs[D3DRS_FOGEND]) - asfloat(state->rs[D3DRS_FOGSTART]));
1739 dst[22].z = asfloat(state->rs[D3DRS_FOGDENSITY]);
1740 }
1741
1742 static void
1743 nine_ff_load_viewport_info(struct NineDevice9 *device)
1744 {
1745 D3DVIEWPORT9 *viewport = &device->state.viewport;
1746 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1747 float diffZ = viewport->MaxZ - viewport->MinZ;
1748
1749 /* Note: the other functions avoids to fill the const again if nothing changed.
1750 * But we don't have much to fill, and adding code to allow that may be complex
1751 * so just fill it always */
1752 dst[100].x = 2.0f / (float)(viewport->Width);
1753 dst[100].y = 2.0f / (float)(viewport->Height);
1754 dst[100].z = (diffZ == 0.0f) ? 0.0f : (1.0f / diffZ);
1755 dst[101].x = (float)(viewport->X);
1756 dst[101].y = (float)(viewport->Y);
1757 dst[101].z = (float)(viewport->MinZ);
1758 }
1759
1760 void
1761 nine_ff_update(struct NineDevice9 *device)
1762 {
1763 struct pipe_context *pipe = device->pipe;
1764 struct nine_state *state = &device->state;
1765
1766 DBG("vs=%p ps=%p\n", device->state.vs, device->state.ps);
1767
1768 /* NOTE: the only reference belongs to the hash table */
1769 if (!device->state.vs)
1770 device->ff.vs = nine_ff_get_vs(device);
1771 if (!device->state.ps)
1772 device->ff.ps = nine_ff_get_ps(device);
1773
1774 if (!device->state.vs) {
1775 if (device->state.ff.clobber.vs_const) {
1776 device->state.ff.clobber.vs_const = FALSE;
1777 device->state.changed.group |=
1778 NINE_STATE_FF_VSTRANSF |
1779 NINE_STATE_FF_MATERIAL |
1780 NINE_STATE_FF_LIGHTING |
1781 NINE_STATE_FF_OTHER;
1782 device->state.ff.changed.transform[0] |= 0xff000c;
1783 device->state.ff.changed.transform[8] |= 0xff;
1784 }
1785 nine_ff_load_vs_transforms(device);
1786 nine_ff_load_tex_matrices(device);
1787 nine_ff_load_lights(device);
1788 nine_ff_load_point_and_fog_params(device);
1789 nine_ff_load_viewport_info(device);
1790
1791 memset(state->ff.changed.transform, 0, sizeof(state->ff.changed.transform));
1792
1793 device->state.changed.group |= NINE_STATE_VS;
1794 device->state.changed.group |= NINE_STATE_VS_CONST;
1795
1796 if (device->prefer_user_constbuf) {
1797 struct pipe_context *pipe = device->pipe;
1798 struct pipe_constant_buffer cb;
1799 cb.buffer_offset = 0;
1800 cb.buffer = NULL;
1801 cb.user_buffer = device->ff.vs_const;
1802 cb.buffer_size = NINE_FF_NUM_VS_CONST * 4 * sizeof(float);
1803 pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 0, &cb);
1804 } else {
1805 struct pipe_box box;
1806 u_box_1d(0, NINE_FF_NUM_VS_CONST * 4 * sizeof(float), &box);
1807 pipe->transfer_inline_write(pipe, device->constbuf_vs, 0,
1808 0, &box,
1809 device->ff.vs_const, 0, 0);
1810 nine_ranges_insert(&device->state.changed.vs_const_f, 0, NINE_FF_NUM_VS_CONST,
1811 &device->range_pool);
1812 }
1813 }
1814
1815 if (!device->state.ps) {
1816 if (device->state.ff.clobber.ps_const) {
1817 device->state.ff.clobber.ps_const = FALSE;
1818 device->state.changed.group |=
1819 NINE_STATE_FF_PSSTAGES |
1820 NINE_STATE_FF_OTHER;
1821 }
1822 nine_ff_load_ps_params(device);
1823
1824 device->state.changed.group |= NINE_STATE_PS;
1825 device->state.changed.group |= NINE_STATE_PS_CONST;
1826
1827 if (device->prefer_user_constbuf) {
1828 struct pipe_context *pipe = device->pipe;
1829 struct pipe_constant_buffer cb;
1830 cb.buffer_offset = 0;
1831 cb.buffer = NULL;
1832 cb.user_buffer = device->ff.ps_const;
1833 cb.buffer_size = NINE_FF_NUM_PS_CONST * 4 * sizeof(float);
1834 pipe->set_constant_buffer(pipe, PIPE_SHADER_FRAGMENT, 0, &cb);
1835 } else {
1836 struct pipe_box box;
1837 u_box_1d(0, NINE_FF_NUM_PS_CONST * 4 * sizeof(float), &box);
1838 pipe->transfer_inline_write(pipe, device->constbuf_ps, 0,
1839 0, &box,
1840 device->ff.ps_const, 0, 0);
1841 nine_ranges_insert(&device->state.changed.ps_const_f, 0, NINE_FF_NUM_PS_CONST,
1842 &device->range_pool);
1843 }
1844 }
1845
1846 device->state.changed.group &= ~NINE_STATE_FF;
1847 }
1848
1849
1850 boolean
1851 nine_ff_init(struct NineDevice9 *device)
1852 {
1853 device->ff.ht_vs = util_hash_table_create(nine_ff_vs_key_hash,
1854 nine_ff_vs_key_comp);
1855 device->ff.ht_ps = util_hash_table_create(nine_ff_ps_key_hash,
1856 nine_ff_ps_key_comp);
1857
1858 device->ff.ht_fvf = util_hash_table_create(nine_ff_fvf_key_hash,
1859 nine_ff_fvf_key_comp);
1860
1861 device->ff.vs_const = CALLOC(NINE_FF_NUM_VS_CONST, 4 * sizeof(float));
1862 device->ff.ps_const = CALLOC(NINE_FF_NUM_PS_CONST, 4 * sizeof(float));
1863
1864 return device->ff.ht_vs && device->ff.ht_ps &&
1865 device->ff.ht_fvf &&
1866 device->ff.vs_const && device->ff.ps_const;
1867 }
1868
1869 static enum pipe_error nine_ff_ht_delete_cb(void *key, void *value, void *data)
1870 {
1871 NineUnknown_Unbind(NineUnknown(value));
1872 return PIPE_OK;
1873 }
1874
1875 void
1876 nine_ff_fini(struct NineDevice9 *device)
1877 {
1878 if (device->ff.ht_vs) {
1879 util_hash_table_foreach(device->ff.ht_vs, nine_ff_ht_delete_cb, NULL);
1880 util_hash_table_destroy(device->ff.ht_vs);
1881 }
1882 if (device->ff.ht_ps) {
1883 util_hash_table_foreach(device->ff.ht_ps, nine_ff_ht_delete_cb, NULL);
1884 util_hash_table_destroy(device->ff.ht_ps);
1885 }
1886 if (device->ff.ht_fvf) {
1887 util_hash_table_foreach(device->ff.ht_fvf, nine_ff_ht_delete_cb, NULL);
1888 util_hash_table_destroy(device->ff.ht_fvf);
1889 }
1890 device->ff.vs = NULL; /* destroyed by unbinding from hash table */
1891 device->ff.ps = NULL;
1892
1893 FREE(device->ff.vs_const);
1894 FREE(device->ff.ps_const);
1895 }
1896
1897 static void
1898 nine_ff_prune_vs(struct NineDevice9 *device)
1899 {
1900 if (device->ff.num_vs > 100) {
1901 /* could destroy the bound one here, so unbind */
1902 device->pipe->bind_vs_state(device->pipe, NULL);
1903 util_hash_table_foreach(device->ff.ht_vs, nine_ff_ht_delete_cb, NULL);
1904 util_hash_table_clear(device->ff.ht_vs);
1905 device->ff.num_vs = 0;
1906 device->state.changed.group |= NINE_STATE_VS;
1907 }
1908 }
1909 static void
1910 nine_ff_prune_ps(struct NineDevice9 *device)
1911 {
1912 if (device->ff.num_ps > 100) {
1913 /* could destroy the bound one here, so unbind */
1914 device->pipe->bind_fs_state(device->pipe, NULL);
1915 util_hash_table_foreach(device->ff.ht_ps, nine_ff_ht_delete_cb, NULL);
1916 util_hash_table_clear(device->ff.ht_ps);
1917 device->ff.num_ps = 0;
1918 device->state.changed.group |= NINE_STATE_PS;
1919 }
1920 }
1921
1922 /* ========================================================================== */
1923
1924 /* Matrix multiplication:
1925 *
1926 * in memory: 0 1 2 3 (row major)
1927 * 4 5 6 7
1928 * 8 9 a b
1929 * c d e f
1930 *
1931 * cA cB cC cD
1932 * r0 = (r0 * cA) (r0 * cB) . .
1933 * r1 = (r1 * cA) (r1 * cB)
1934 * r2 = (r2 * cA) .
1935 * r3 = (r3 * cA) .
1936 *
1937 * r: (11) (12) (13) (14)
1938 * (21) (22) (23) (24)
1939 * (31) (32) (33) (34)
1940 * (41) (42) (43) (44)
1941 * l: (11 12 13 14)
1942 * (21 22 23 24)
1943 * (31 32 33 34)
1944 * (41 42 43 44)
1945 *
1946 * v: (x y z 1 )
1947 *
1948 * t.xyzw = MUL(v.xxxx, r[0]);
1949 * t.xyzw = MAD(v.yyyy, r[1], t.xyzw);
1950 * t.xyzw = MAD(v.zzzz, r[2], t.xyzw);
1951 * v.xyzw = MAD(v.wwww, r[3], t.xyzw);
1952 *
1953 * v.x = DP4(v, c[0]);
1954 * v.y = DP4(v, c[1]);
1955 * v.z = DP4(v, c[2]);
1956 * v.w = DP4(v, c[3]) = 1
1957 */
1958
1959 /*
1960 static void
1961 nine_D3DMATRIX_print(const D3DMATRIX *M)
1962 {
1963 DBG("\n(%f %f %f %f)\n"
1964 "(%f %f %f %f)\n"
1965 "(%f %f %f %f)\n"
1966 "(%f %f %f %f)\n",
1967 M->m[0][0], M->m[0][1], M->m[0][2], M->m[0][3],
1968 M->m[1][0], M->m[1][1], M->m[1][2], M->m[1][3],
1969 M->m[2][0], M->m[2][1], M->m[2][2], M->m[2][3],
1970 M->m[3][0], M->m[3][1], M->m[3][2], M->m[3][3]);
1971 }
1972 */
1973
1974 static INLINE float
1975 nine_DP4_row_col(const D3DMATRIX *A, int r, const D3DMATRIX *B, int c)
1976 {
1977 return A->m[r][0] * B->m[0][c] +
1978 A->m[r][1] * B->m[1][c] +
1979 A->m[r][2] * B->m[2][c] +
1980 A->m[r][3] * B->m[3][c];
1981 }
1982
1983 static INLINE float
1984 nine_DP4_vec_col(const D3DVECTOR *v, const D3DMATRIX *M, int c)
1985 {
1986 return v->x * M->m[0][c] +
1987 v->y * M->m[1][c] +
1988 v->z * M->m[2][c] +
1989 1.0f * M->m[3][c];
1990 }
1991
1992 static INLINE float
1993 nine_DP3_vec_col(const D3DVECTOR *v, const D3DMATRIX *M, int c)
1994 {
1995 return v->x * M->m[0][c] +
1996 v->y * M->m[1][c] +
1997 v->z * M->m[2][c];
1998 }
1999
2000 void
2001 nine_d3d_matrix_matrix_mul(D3DMATRIX *D, const D3DMATRIX *L, const D3DMATRIX *R)
2002 {
2003 D->_11 = nine_DP4_row_col(L, 0, R, 0);
2004 D->_12 = nine_DP4_row_col(L, 0, R, 1);
2005 D->_13 = nine_DP4_row_col(L, 0, R, 2);
2006 D->_14 = nine_DP4_row_col(L, 0, R, 3);
2007
2008 D->_21 = nine_DP4_row_col(L, 1, R, 0);
2009 D->_22 = nine_DP4_row_col(L, 1, R, 1);
2010 D->_23 = nine_DP4_row_col(L, 1, R, 2);
2011 D->_24 = nine_DP4_row_col(L, 1, R, 3);
2012
2013 D->_31 = nine_DP4_row_col(L, 2, R, 0);
2014 D->_32 = nine_DP4_row_col(L, 2, R, 1);
2015 D->_33 = nine_DP4_row_col(L, 2, R, 2);
2016 D->_34 = nine_DP4_row_col(L, 2, R, 3);
2017
2018 D->_41 = nine_DP4_row_col(L, 3, R, 0);
2019 D->_42 = nine_DP4_row_col(L, 3, R, 1);
2020 D->_43 = nine_DP4_row_col(L, 3, R, 2);
2021 D->_44 = nine_DP4_row_col(L, 3, R, 3);
2022 }
2023
2024 void
2025 nine_d3d_vector4_matrix_mul(D3DVECTOR *d, const D3DVECTOR *v, const D3DMATRIX *M)
2026 {
2027 d->x = nine_DP4_vec_col(v, M, 0);
2028 d->y = nine_DP4_vec_col(v, M, 1);
2029 d->z = nine_DP4_vec_col(v, M, 2);
2030 }
2031
2032 void
2033 nine_d3d_vector3_matrix_mul(D3DVECTOR *d, const D3DVECTOR *v, const D3DMATRIX *M)
2034 {
2035 d->x = nine_DP3_vec_col(v, M, 0);
2036 d->y = nine_DP3_vec_col(v, M, 1);
2037 d->z = nine_DP3_vec_col(v, M, 2);
2038 }
2039
2040 void
2041 nine_d3d_matrix_transpose(D3DMATRIX *D, const D3DMATRIX *M)
2042 {
2043 unsigned i, j;
2044 for (i = 0; i < 4; ++i)
2045 for (j = 0; j < 4; ++j)
2046 D->m[i][j] = M->m[j][i];
2047 }
2048
2049 #define _M_ADD_PROD_1i_2j_3k_4l(i,j,k,l) do { \
2050 float t = M->_1##i * M->_2##j * M->_3##k * M->_4##l; \
2051 if (t > 0.0f) pos += t; else neg += t; } while(0)
2052
2053 #define _M_SUB_PROD_1i_2j_3k_4l(i,j,k,l) do { \
2054 float t = M->_1##i * M->_2##j * M->_3##k * M->_4##l; \
2055 if (t > 0.0f) neg -= t; else pos -= t; } while(0)
2056 float
2057 nine_d3d_matrix_det(const D3DMATRIX *M)
2058 {
2059 float pos = 0.0f;
2060 float neg = 0.0f;
2061
2062 _M_ADD_PROD_1i_2j_3k_4l(1, 2, 3, 4);
2063 _M_ADD_PROD_1i_2j_3k_4l(1, 3, 4, 2);
2064 _M_ADD_PROD_1i_2j_3k_4l(1, 4, 2, 3);
2065
2066 _M_ADD_PROD_1i_2j_3k_4l(2, 1, 4, 3);
2067 _M_ADD_PROD_1i_2j_3k_4l(2, 3, 1, 4);
2068 _M_ADD_PROD_1i_2j_3k_4l(2, 4, 3, 1);
2069
2070 _M_ADD_PROD_1i_2j_3k_4l(3, 1, 2, 4);
2071 _M_ADD_PROD_1i_2j_3k_4l(3, 2, 4, 1);
2072 _M_ADD_PROD_1i_2j_3k_4l(3, 4, 1, 2);
2073
2074 _M_ADD_PROD_1i_2j_3k_4l(4, 1, 3, 2);
2075 _M_ADD_PROD_1i_2j_3k_4l(4, 2, 1, 3);
2076 _M_ADD_PROD_1i_2j_3k_4l(4, 3, 2, 1);
2077
2078 _M_SUB_PROD_1i_2j_3k_4l(1, 2, 4, 3);
2079 _M_SUB_PROD_1i_2j_3k_4l(1, 3, 2, 4);
2080 _M_SUB_PROD_1i_2j_3k_4l(1, 4, 3, 2);
2081
2082 _M_SUB_PROD_1i_2j_3k_4l(2, 1, 3, 4);
2083 _M_SUB_PROD_1i_2j_3k_4l(2, 3, 4, 1);
2084 _M_SUB_PROD_1i_2j_3k_4l(2, 4, 1, 3);
2085
2086 _M_SUB_PROD_1i_2j_3k_4l(3, 1, 4, 2);
2087 _M_SUB_PROD_1i_2j_3k_4l(3, 2, 1, 4);
2088 _M_SUB_PROD_1i_2j_3k_4l(3, 4, 2, 1);
2089
2090 _M_SUB_PROD_1i_2j_3k_4l(4, 1, 2, 3);
2091 _M_SUB_PROD_1i_2j_3k_4l(4, 2, 3, 1);
2092 _M_SUB_PROD_1i_2j_3k_4l(4, 3, 1, 2);
2093
2094 return pos + neg;
2095 }
2096
2097 /* XXX: Probably better to just use src/mesa/math/m_matrix.c because
2098 * I have no idea where this code came from.
2099 */
2100 void
2101 nine_d3d_matrix_inverse(D3DMATRIX *D, const D3DMATRIX *M)
2102 {
2103 int i, k;
2104 float det;
2105
2106 D->m[0][0] =
2107 M->m[1][1] * M->m[2][2] * M->m[3][3] -
2108 M->m[1][1] * M->m[3][2] * M->m[2][3] -
2109 M->m[1][2] * M->m[2][1] * M->m[3][3] +
2110 M->m[1][2] * M->m[3][1] * M->m[2][3] +
2111 M->m[1][3] * M->m[2][1] * M->m[3][2] -
2112 M->m[1][3] * M->m[3][1] * M->m[2][2];
2113
2114 D->m[0][1] =
2115 -M->m[0][1] * M->m[2][2] * M->m[3][3] +
2116 M->m[0][1] * M->m[3][2] * M->m[2][3] +
2117 M->m[0][2] * M->m[2][1] * M->m[3][3] -
2118 M->m[0][2] * M->m[3][1] * M->m[2][3] -
2119 M->m[0][3] * M->m[2][1] * M->m[3][2] +
2120 M->m[0][3] * M->m[3][1] * M->m[2][2];
2121
2122 D->m[0][2] =
2123 M->m[0][1] * M->m[1][2] * M->m[3][3] -
2124 M->m[0][1] * M->m[3][2] * M->m[1][3] -
2125 M->m[0][2] * M->m[1][1] * M->m[3][3] +
2126 M->m[0][2] * M->m[3][1] * M->m[1][3] +
2127 M->m[0][3] * M->m[1][1] * M->m[3][2] -
2128 M->m[0][3] * M->m[3][1] * M->m[1][2];
2129
2130 D->m[0][3] =
2131 -M->m[0][1] * M->m[1][2] * M->m[2][3] +
2132 M->m[0][1] * M->m[2][2] * M->m[1][3] +
2133 M->m[0][2] * M->m[1][1] * M->m[2][3] -
2134 M->m[0][2] * M->m[2][1] * M->m[1][3] -
2135 M->m[0][3] * M->m[1][1] * M->m[2][2] +
2136 M->m[0][3] * M->m[2][1] * M->m[1][2];
2137
2138 D->m[1][0] =
2139 -M->m[1][0] * M->m[2][2] * M->m[3][3] +
2140 M->m[1][0] * M->m[3][2] * M->m[2][3] +
2141 M->m[1][2] * M->m[2][0] * M->m[3][3] -
2142 M->m[1][2] * M->m[3][0] * M->m[2][3] -
2143 M->m[1][3] * M->m[2][0] * M->m[3][2] +
2144 M->m[1][3] * M->m[3][0] * M->m[2][2];
2145
2146 D->m[1][1] =
2147 M->m[0][0] * M->m[2][2] * M->m[3][3] -
2148 M->m[0][0] * M->m[3][2] * M->m[2][3] -
2149 M->m[0][2] * M->m[2][0] * M->m[3][3] +
2150 M->m[0][2] * M->m[3][0] * M->m[2][3] +
2151 M->m[0][3] * M->m[2][0] * M->m[3][2] -
2152 M->m[0][3] * M->m[3][0] * M->m[2][2];
2153
2154 D->m[1][2] =
2155 -M->m[0][0] * M->m[1][2] * M->m[3][3] +
2156 M->m[0][0] * M->m[3][2] * M->m[1][3] +
2157 M->m[0][2] * M->m[1][0] * M->m[3][3] -
2158 M->m[0][2] * M->m[3][0] * M->m[1][3] -
2159 M->m[0][3] * M->m[1][0] * M->m[3][2] +
2160 M->m[0][3] * M->m[3][0] * M->m[1][2];
2161
2162 D->m[1][3] =
2163 M->m[0][0] * M->m[1][2] * M->m[2][3] -
2164 M->m[0][0] * M->m[2][2] * M->m[1][3] -
2165 M->m[0][2] * M->m[1][0] * M->m[2][3] +
2166 M->m[0][2] * M->m[2][0] * M->m[1][3] +
2167 M->m[0][3] * M->m[1][0] * M->m[2][2] -
2168 M->m[0][3] * M->m[2][0] * M->m[1][2];
2169
2170 D->m[2][0] =
2171 M->m[1][0] * M->m[2][1] * M->m[3][3] -
2172 M->m[1][0] * M->m[3][1] * M->m[2][3] -
2173 M->m[1][1] * M->m[2][0] * M->m[3][3] +
2174 M->m[1][1] * M->m[3][0] * M->m[2][3] +
2175 M->m[1][3] * M->m[2][0] * M->m[3][1] -
2176 M->m[1][3] * M->m[3][0] * M->m[2][1];
2177
2178 D->m[2][1] =
2179 -M->m[0][0] * M->m[2][1] * M->m[3][3] +
2180 M->m[0][0] * M->m[3][1] * M->m[2][3] +
2181 M->m[0][1] * M->m[2][0] * M->m[3][3] -
2182 M->m[0][1] * M->m[3][0] * M->m[2][3] -
2183 M->m[0][3] * M->m[2][0] * M->m[3][1] +
2184 M->m[0][3] * M->m[3][0] * M->m[2][1];
2185
2186 D->m[2][2] =
2187 M->m[0][0] * M->m[1][1] * M->m[3][3] -
2188 M->m[0][0] * M->m[3][1] * M->m[1][3] -
2189 M->m[0][1] * M->m[1][0] * M->m[3][3] +
2190 M->m[0][1] * M->m[3][0] * M->m[1][3] +
2191 M->m[0][3] * M->m[1][0] * M->m[3][1] -
2192 M->m[0][3] * M->m[3][0] * M->m[1][1];
2193
2194 D->m[2][3] =
2195 -M->m[0][0] * M->m[1][1] * M->m[2][3] +
2196 M->m[0][0] * M->m[2][1] * M->m[1][3] +
2197 M->m[0][1] * M->m[1][0] * M->m[2][3] -
2198 M->m[0][1] * M->m[2][0] * M->m[1][3] -
2199 M->m[0][3] * M->m[1][0] * M->m[2][1] +
2200 M->m[0][3] * M->m[2][0] * M->m[1][1];
2201
2202 D->m[3][0] =
2203 -M->m[1][0] * M->m[2][1] * M->m[3][2] +
2204 M->m[1][0] * M->m[3][1] * M->m[2][2] +
2205 M->m[1][1] * M->m[2][0] * M->m[3][2] -
2206 M->m[1][1] * M->m[3][0] * M->m[2][2] -
2207 M->m[1][2] * M->m[2][0] * M->m[3][1] +
2208 M->m[1][2] * M->m[3][0] * M->m[2][1];
2209
2210 D->m[3][1] =
2211 M->m[0][0] * M->m[2][1] * M->m[3][2] -
2212 M->m[0][0] * M->m[3][1] * M->m[2][2] -
2213 M->m[0][1] * M->m[2][0] * M->m[3][2] +
2214 M->m[0][1] * M->m[3][0] * M->m[2][2] +
2215 M->m[0][2] * M->m[2][0] * M->m[3][1] -
2216 M->m[0][2] * M->m[3][0] * M->m[2][1];
2217
2218 D->m[3][2] =
2219 -M->m[0][0] * M->m[1][1] * M->m[3][2] +
2220 M->m[0][0] * M->m[3][1] * M->m[1][2] +
2221 M->m[0][1] * M->m[1][0] * M->m[3][2] -
2222 M->m[0][1] * M->m[3][0] * M->m[1][2] -
2223 M->m[0][2] * M->m[1][0] * M->m[3][1] +
2224 M->m[0][2] * M->m[3][0] * M->m[1][1];
2225
2226 D->m[3][3] =
2227 M->m[0][0] * M->m[1][1] * M->m[2][2] -
2228 M->m[0][0] * M->m[2][1] * M->m[1][2] -
2229 M->m[0][1] * M->m[1][0] * M->m[2][2] +
2230 M->m[0][1] * M->m[2][0] * M->m[1][2] +
2231 M->m[0][2] * M->m[1][0] * M->m[2][1] -
2232 M->m[0][2] * M->m[2][0] * M->m[1][1];
2233
2234 det =
2235 M->m[0][0] * D->m[0][0] +
2236 M->m[1][0] * D->m[0][1] +
2237 M->m[2][0] * D->m[0][2] +
2238 M->m[3][0] * D->m[0][3];
2239
2240 det = 1.0 / det;
2241
2242 for (i = 0; i < 4; i++)
2243 for (k = 0; k < 4; k++)
2244 D->m[i][k] *= det;
2245
2246 #ifdef DEBUG
2247 {
2248 D3DMATRIX I;
2249
2250 nine_d3d_matrix_matrix_mul(&I, D, M);
2251
2252 for (i = 0; i < 4; ++i)
2253 for (k = 0; k < 4; ++k)
2254 if (fabsf(I.m[i][k] - (float)(i == k)) > 1e-3)
2255 DBG("Matrix inversion check FAILED !\n");
2256 }
2257 #endif
2258 }
2259
2260 /* TODO: don't use 4x4 inverse, unless this gets all nicely inlined ? */
2261 void
2262 nine_d3d_matrix_inverse_3x3(D3DMATRIX *D, const D3DMATRIX *M)
2263 {
2264 D3DMATRIX T;
2265 unsigned i, j;
2266
2267 for (i = 0; i < 3; ++i)
2268 for (j = 0; j < 3; ++j)
2269 T.m[i][j] = M->m[i][j];
2270 for (i = 0; i < 3; ++i) {
2271 T.m[i][3] = 0.0f;
2272 T.m[3][i] = 0.0f;
2273 }
2274 T.m[3][3] = 1.0f;
2275
2276 nine_d3d_matrix_inverse(D, &T);
2277 }