projects / mesa.git / blob
? search:


2175bdbcc51d9cf645a9e9dafe21fe1e4e1cbaeb
[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 196
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_zero : 1;
58 uint32_t has_normal : 1;
59 uint32_t fog : 1;
60 uint32_t normalizenormals : 1;
61 uint32_t ucp : 1;
62 uint32_t pad1 : 4;
63 uint32_t tc_dim_input: 16; /* 8 * 2 bits */
64 uint32_t pad2 : 16;
65 uint32_t tc_dim_output: 24; /* 8 * 3 bits */
66 uint32_t pad3 : 8;
67 uint32_t tc_gen : 24; /* 8 * 3 bits */
68 uint32_t pad4 : 8;
69 uint32_t tc_idx : 24;
70 uint32_t pad5 : 8;
71 uint32_t passthrough;
72 };
73 uint64_t value64[3]; /* don't forget to resize VertexShader9.ff_key */
74 uint32_t value32[6];
75 };
76 };
77
78 /* Texture stage state:
79 *
80 * COLOROP D3DTOP 5 bit
81 * ALPHAOP D3DTOP 5 bit
82 * COLORARG0 D3DTA 3 bit
83 * COLORARG1 D3DTA 3 bit
84 * COLORARG2 D3DTA 3 bit
85 * ALPHAARG0 D3DTA 3 bit
86 * ALPHAARG1 D3DTA 3 bit
87 * ALPHAARG2 D3DTA 3 bit
88 * RESULTARG D3DTA 1 bit (CURRENT:0 or TEMP:1)
89 * TEXCOORDINDEX 0 - 7 3 bit
90 * ===========================
91 * 32 bit per stage
92 */
93 struct nine_ff_ps_key
94 {
95 union {
96 struct {
97 struct {
98 uint32_t colorop : 5;
99 uint32_t alphaop : 5;
100 uint32_t colorarg0 : 3;
101 uint32_t colorarg1 : 3;
102 uint32_t colorarg2 : 3;
103 uint32_t alphaarg0 : 3;
104 uint32_t alphaarg1 : 3;
105 uint32_t alphaarg2 : 3;
106 uint32_t resultarg : 1; /* CURRENT:0 or TEMP:1 */
107 uint32_t textarget : 2; /* 1D/2D/3D/CUBE */
108 uint32_t pad : 1;
109 /* that's 32 bit exactly */
110 } ts[8];
111 uint32_t projected : 16;
112 uint32_t fog : 1; /* for vFog coming from VS */
113 uint32_t fog_mode : 2;
114 uint32_t fog_source : 1; /* 0: Z, 1: W */
115 uint32_t specular : 1;
116 uint32_t pad1 : 11; /* 9 32-bit words with this */
117 uint8_t colorarg_b4[3];
118 uint8_t colorarg_b5[3];
119 uint8_t alphaarg_b4[3]; /* 11 32-bit words plus a byte */
120 uint8_t pad2[3];
121 };
122 uint64_t value64[6]; /* don't forget to resize PixelShader9.ff_key */
123 uint32_t value32[12];
124 };
125 };
126
127 static unsigned nine_ff_vs_key_hash(void *key)
128 {
129 struct nine_ff_vs_key *vs = key;
130 unsigned i;
131 uint32_t hash = vs->value32[0];
132 for (i = 1; i < ARRAY_SIZE(vs->value32); ++i)
133 hash ^= vs->value32[i];
134 return hash;
135 }
136 static int nine_ff_vs_key_comp(void *key1, void *key2)
137 {
138 struct nine_ff_vs_key *a = (struct nine_ff_vs_key *)key1;
139 struct nine_ff_vs_key *b = (struct nine_ff_vs_key *)key2;
140
141 return memcmp(a->value64, b->value64, sizeof(a->value64));
142 }
143 static unsigned nine_ff_ps_key_hash(void *key)
144 {
145 struct nine_ff_ps_key *ps = key;
146 unsigned i;
147 uint32_t hash = ps->value32[0];
148 for (i = 1; i < ARRAY_SIZE(ps->value32); ++i)
149 hash ^= ps->value32[i];
150 return hash;
151 }
152 static int nine_ff_ps_key_comp(void *key1, void *key2)
153 {
154 struct nine_ff_ps_key *a = (struct nine_ff_ps_key *)key1;
155 struct nine_ff_ps_key *b = (struct nine_ff_ps_key *)key2;
156
157 return memcmp(a->value64, b->value64, sizeof(a->value64));
158 }
159 static unsigned nine_ff_fvf_key_hash(void *key)
160 {
161 return *(DWORD *)key;
162 }
163 static int nine_ff_fvf_key_comp(void *key1, void *key2)
164 {
165 return *(DWORD *)key1 != *(DWORD *)key2;
166 }
167
168 static void nine_ff_prune_vs(struct NineDevice9 *);
169 static void nine_ff_prune_ps(struct NineDevice9 *);
170
171 static void nine_ureg_tgsi_dump(struct ureg_program *ureg, boolean override)
172 {
173 if (debug_get_bool_option("NINE_FF_DUMP", FALSE) || override) {
174 const struct tgsi_token *toks = ureg_get_tokens(ureg, NULL);
175 tgsi_dump(toks, 0);
176 ureg_free_tokens(toks);
177 }
178 }
179
180 #define _X(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_X)
181 #define _Y(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_Y)
182 #define _Z(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_Z)
183 #define _W(r) ureg_scalar(ureg_src(r), TGSI_SWIZZLE_W)
184
185 #define _XXXX(r) ureg_scalar(r, TGSI_SWIZZLE_X)
186 #define _YYYY(r) ureg_scalar(r, TGSI_SWIZZLE_Y)
187 #define _ZZZZ(r) ureg_scalar(r, TGSI_SWIZZLE_Z)
188 #define _WWWW(r) ureg_scalar(r, TGSI_SWIZZLE_W)
189
190 #define _XYZW(r) (r)
191
192 /* AL should contain base address of lights table. */
193 #define LIGHT_CONST(i) \
194 ureg_src_indirect(ureg_DECL_constant(ureg, i), _X(AL))
195
196 #define MATERIAL_CONST(i) \
197 ureg_DECL_constant(ureg, 19 + (i))
198
199 #define _CONST(n) ureg_DECL_constant(ureg, n)
200
201 /* VS FF constants layout:
202 *
203 * CONST[ 0.. 3] D3DTS_WORLD * D3DTS_VIEW * D3DTS_PROJECTION
204 * CONST[ 4.. 7] D3DTS_WORLD * D3DTS_VIEW
205 * CONST[ 8..11] D3DTS_PROJECTION
206 * CONST[12..15] D3DTS_VIEW^(-1)
207 * CONST[16..18] Normal matrix
208 *
209 * CONST[19].xyz MATERIAL.Emissive + Material.Ambient * RS.Ambient
210 * CONST[20] MATERIAL.Diffuse
211 * CONST[21] MATERIAL.Ambient
212 * CONST[22] MATERIAL.Specular
213 * CONST[23].x___ MATERIAL.Power
214 * CONST[24] MATERIAL.Emissive
215 * CONST[25] RS.Ambient
216 *
217 * CONST[26].x___ RS.PointSizeMin
218 * CONST[26]._y__ RS.PointSizeMax
219 * CONST[26].__z_ RS.PointSize
220 * CONST[26].___w RS.PointScaleA
221 * CONST[27].x___ RS.PointScaleB
222 * CONST[27]._y__ RS.PointScaleC
223 *
224 * CONST[28].x___ RS.FogEnd
225 * CONST[28]._y__ 1.0f / (RS.FogEnd - RS.FogStart)
226 * CONST[28].__z_ RS.FogDensity
227
228 * CONST[30].x___ TWEENFACTOR
229 *
230 * CONST[32].x___ LIGHT[0].Type
231 * CONST[32]._yzw LIGHT[0].Attenuation0,1,2
232 * CONST[33] LIGHT[0].Diffuse
233 * CONST[34] LIGHT[0].Specular
234 * CONST[35] LIGHT[0].Ambient
235 * CONST[36].xyz_ LIGHT[0].Position
236 * CONST[36].___w LIGHT[0].Range
237 * CONST[37].xyz_ LIGHT[0].Direction
238 * CONST[37].___w LIGHT[0].Falloff
239 * CONST[38].x___ cos(LIGHT[0].Theta / 2)
240 * CONST[38]._y__ cos(LIGHT[0].Phi / 2)
241 * CONST[38].__z_ 1.0f / (cos(LIGHT[0].Theta / 2) - cos(Light[0].Phi / 2))
242 * CONST[39].xyz_ LIGHT[0].HalfVector (for directional lights)
243 * CONST[39].___w 1 if this is the last active light, 0 if not
244 * CONST[40] LIGHT[1]
245 * CONST[48] LIGHT[2]
246 * CONST[56] LIGHT[3]
247 * CONST[64] LIGHT[4]
248 * CONST[72] LIGHT[5]
249 * CONST[80] LIGHT[6]
250 * CONST[88] LIGHT[7]
251 * NOTE: no lighting code is generated if there are no active lights
252 *
253 * CONST[100].x___ Viewport 2/width
254 * CONST[100]._y__ Viewport 2/height
255 * CONST[100].__z_ Viewport 1/(zmax - zmin)
256 * CONST[100].___w Viewport width
257 * CONST[101].x___ Viewport x0
258 * CONST[101]._y__ Viewport y0
259 * CONST[101].__z_ Viewport z0
260 *
261 * CONST[128..131] D3DTS_TEXTURE0
262 * CONST[132..135] D3DTS_TEXTURE1
263 * CONST[136..139] D3DTS_TEXTURE2
264 * CONST[140..143] D3DTS_TEXTURE3
265 * CONST[144..147] D3DTS_TEXTURE4
266 * CONST[148..151] D3DTS_TEXTURE5
267 * CONST[152..155] D3DTS_TEXTURE6
268 * CONST[156..159] D3DTS_TEXTURE7
269 *
270 * CONST[160] D3DTS_WORLDMATRIX[0] * D3DTS_VIEW
271 * CONST[164] D3DTS_WORLDMATRIX[1] * D3DTS_VIEW
272 * ...
273 * CONST[192] D3DTS_WORLDMATRIX[8] * D3DTS_VIEW
274 */
275 struct vs_build_ctx
276 {
277 struct ureg_program *ureg;
278 const struct nine_ff_vs_key *key;
279
280 uint16_t input[PIPE_MAX_ATTRIBS];
281 unsigned num_inputs;
282
283 struct ureg_src aVtx;
284 struct ureg_src aNrm;
285 struct ureg_src aCol[2];
286 struct ureg_src aTex[8];
287 struct ureg_src aPsz;
288 struct ureg_src aInd;
289 struct ureg_src aWgt;
290
291 struct ureg_src aVtx1; /* tweening */
292 struct ureg_src aNrm1;
293
294 struct ureg_src mtlA;
295 struct ureg_src mtlD;
296 struct ureg_src mtlS;
297 struct ureg_src mtlE;
298 };
299
300 static inline unsigned
301 get_texcoord_sn(struct pipe_screen *screen)
302 {
303 if (screen->get_param(screen, PIPE_CAP_TGSI_TEXCOORD))
304 return TGSI_SEMANTIC_TEXCOORD;
305 return TGSI_SEMANTIC_GENERIC;
306 }
307
308 static inline struct ureg_src
309 build_vs_add_input(struct vs_build_ctx *vs, uint16_t ndecl)
310 {
311 const unsigned i = vs->num_inputs++;
312 assert(i < PIPE_MAX_ATTRIBS);
313 vs->input[i] = ndecl;
314 return ureg_DECL_vs_input(vs->ureg, i);
315 }
316
317 /* NOTE: dst may alias src */
318 static inline void
319 ureg_normalize3(struct ureg_program *ureg,
320 struct ureg_dst dst, struct ureg_src src)
321 {
322 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
323 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
324
325 ureg_DP3(ureg, tmp_x, src, src);
326 ureg_RSQ(ureg, tmp_x, _X(tmp));
327 ureg_MUL(ureg, dst, src, _X(tmp));
328 ureg_release_temporary(ureg, tmp);
329 }
330
331 static void *
332 nine_ff_build_vs(struct NineDevice9 *device, struct vs_build_ctx *vs)
333 {
334 const struct nine_ff_vs_key *key = vs->key;
335 struct ureg_program *ureg = ureg_create(PIPE_SHADER_VERTEX);
336 struct ureg_dst oPos, oCol[2], oPsz, oFog;
337 struct ureg_dst AR;
338 unsigned i, c;
339 unsigned label[32], l = 0;
340 boolean need_aNrm = key->lighting || key->passthrough & (1 << NINE_DECLUSAGE_NORMAL);
341 boolean has_aNrm = need_aNrm && key->has_normal;
342 boolean need_aVtx = key->lighting || key->fog_mode || key->pointscale || key->ucp;
343 const unsigned texcoord_sn = get_texcoord_sn(device->screen);
344
345 vs->ureg = ureg;
346
347 /* Check which inputs we should transform. */
348 for (i = 0; i < 8 * 3; i += 3) {
349 switch ((key->tc_gen >> i) & 0x7) {
350 case NINED3DTSS_TCI_CAMERASPACENORMAL:
351 need_aNrm = TRUE;
352 break;
353 case NINED3DTSS_TCI_CAMERASPACEPOSITION:
354 need_aVtx = TRUE;
355 break;
356 case NINED3DTSS_TCI_CAMERASPACEREFLECTIONVECTOR:
357 need_aVtx = need_aNrm = TRUE;
358 break;
359 case NINED3DTSS_TCI_SPHEREMAP:
360 need_aVtx = need_aNrm = TRUE;
361 break;
362 default:
363 break;
364 }
365 }
366
367 /* Declare and record used inputs (needed for linkage with vertex format):
368 * (texture coordinates handled later)
369 */
370 vs->aVtx = build_vs_add_input(vs,
371 key->position_t ? NINE_DECLUSAGE_POSITIONT : NINE_DECLUSAGE_POSITION);
372
373 vs->aNrm = ureg_imm1f(ureg, 0.0f);
374 if (has_aNrm)
375 vs->aNrm = build_vs_add_input(vs, NINE_DECLUSAGE_NORMAL);
376
377 vs->aCol[0] = ureg_imm1f(ureg, 1.0f);
378 vs->aCol[1] = ureg_imm1f(ureg, 0.0f);
379
380 if (key->lighting || key->darkness) {
381 const unsigned mask = key->mtl_diffuse | key->mtl_specular |
382 key->mtl_ambient | key->mtl_emissive;
383 if ((mask & 0x1) && !key->color0in_one)
384 vs->aCol[0] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 0));
385 if ((mask & 0x2) && !key->color1in_zero)
386 vs->aCol[1] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 1));
387
388 vs->mtlD = MATERIAL_CONST(1);
389 vs->mtlA = MATERIAL_CONST(2);
390 vs->mtlS = MATERIAL_CONST(3);
391 vs->mtlE = MATERIAL_CONST(5);
392 if (key->mtl_diffuse == 1) vs->mtlD = vs->aCol[0]; else
393 if (key->mtl_diffuse == 2) vs->mtlD = vs->aCol[1];
394 if (key->mtl_ambient == 1) vs->mtlA = vs->aCol[0]; else
395 if (key->mtl_ambient == 2) vs->mtlA = vs->aCol[1];
396 if (key->mtl_specular == 1) vs->mtlS = vs->aCol[0]; else
397 if (key->mtl_specular == 2) vs->mtlS = vs->aCol[1];
398 if (key->mtl_emissive == 1) vs->mtlE = vs->aCol[0]; else
399 if (key->mtl_emissive == 2) vs->mtlE = vs->aCol[1];
400 } else {
401 if (!key->color0in_one) vs->aCol[0] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 0));
402 if (!key->color1in_zero) vs->aCol[1] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 1));
403 }
404
405 if (key->vertexpointsize)
406 vs->aPsz = build_vs_add_input(vs, NINE_DECLUSAGE_PSIZE);
407
408 if (key->vertexblend_indexed || key->passthrough & (1 << NINE_DECLUSAGE_BLENDINDICES))
409 vs->aInd = build_vs_add_input(vs, NINE_DECLUSAGE_BLENDINDICES);
410 if (key->vertexblend || key->passthrough & (1 << NINE_DECLUSAGE_BLENDWEIGHT))
411 vs->aWgt = build_vs_add_input(vs, NINE_DECLUSAGE_BLENDWEIGHT);
412 if (key->vertextween) {
413 vs->aVtx1 = build_vs_add_input(vs, NINE_DECLUSAGE_i(POSITION,1));
414 vs->aNrm1 = build_vs_add_input(vs, NINE_DECLUSAGE_i(NORMAL,1));
415 }
416
417 /* Declare outputs:
418 */
419 oPos = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0); /* HPOS */
420 oCol[0] = ureg_saturate(ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0));
421 oCol[1] = ureg_saturate(ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 1));
422 if (key->fog || key->passthrough & (1 << NINE_DECLUSAGE_FOG)) {
423 oFog = ureg_DECL_output(ureg, TGSI_SEMANTIC_FOG, 0);
424 oFog = ureg_writemask(oFog, TGSI_WRITEMASK_X);
425 }
426
427 if (key->vertexpointsize || key->pointscale) {
428 oPsz = ureg_DECL_output_masked(ureg, TGSI_SEMANTIC_PSIZE, 0,
429 TGSI_WRITEMASK_X, 0, 1);
430 oPsz = ureg_writemask(oPsz, TGSI_WRITEMASK_X);
431 }
432
433 if (key->lighting || key->vertexblend)
434 AR = ureg_DECL_address(ureg);
435
436 /* === Vertex transformation / vertex blending:
437 */
438
439 if (key->position_t) {
440 if (device->driver_caps.window_space_position_support) {
441 ureg_MOV(ureg, oPos, vs->aVtx);
442 } else {
443 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
444 /* vs->aVtx contains the coordinates buffer wise.
445 * later in the pipeline, clipping, viewport and division
446 * by w (rhw = 1/w) are going to be applied, so do the reverse
447 * of these transformations (except clipping) to have the good
448 * position at the end.*/
449 ureg_MOV(ureg, tmp, vs->aVtx);
450 /* X from [X_min, X_min + width] to [-1, 1], same for Y. Z to [0, 1] */
451 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), ureg_negate(_CONST(101)));
452 ureg_MUL(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _CONST(100));
453 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XY), ureg_src(tmp), ureg_imm1f(ureg, -1.0f));
454 /* Y needs to be reversed */
455 ureg_MOV(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_Y), ureg_negate(ureg_src(tmp)));
456 /* inverse rhw */
457 ureg_RCP(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_W), _W(tmp));
458 /* multiply X, Y, Z by w */
459 ureg_MUL(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _W(tmp));
460 ureg_MOV(ureg, oPos, ureg_src(tmp));
461 ureg_release_temporary(ureg, tmp);
462 }
463 } else if (key->vertexblend) {
464 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
465 struct ureg_dst tmp2 = ureg_DECL_temporary(ureg);
466 struct ureg_dst aVtx_dst = ureg_DECL_temporary(ureg);
467 struct ureg_dst aNrm_dst = ureg_DECL_temporary(ureg);
468 struct ureg_dst sum_blendweights = ureg_DECL_temporary(ureg);
469 struct ureg_src cWM[4];
470
471 for (i = 160; i <= 195; ++i)
472 ureg_DECL_constant(ureg, i);
473
474 /* translate world matrix index to constant file index */
475 if (key->vertexblend_indexed) {
476 ureg_MAD(ureg, tmp, vs->aInd, ureg_imm1f(ureg, 4.0f), ureg_imm1f(ureg, 160.0f));
477 ureg_ARL(ureg, AR, ureg_src(tmp));
478 }
479
480 ureg_MOV(ureg, aVtx_dst, ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 0.0f));
481 ureg_MOV(ureg, aNrm_dst, ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 0.0f));
482 ureg_MOV(ureg, sum_blendweights, ureg_imm4f(ureg, 1.0f, 1.0f, 1.0f, 1.0f));
483
484 for (i = 0; i < key->vertexblend; ++i) {
485 for (c = 0; c < 4; ++c) {
486 cWM[c] = ureg_src_register(TGSI_FILE_CONSTANT, (160 + i * 4) * !key->vertexblend_indexed + c);
487 if (key->vertexblend_indexed)
488 cWM[c] = ureg_src_indirect(cWM[c], ureg_scalar(ureg_src(AR), i));
489 }
490
491 /* multiply by WORLD(index) */
492 ureg_MUL(ureg, tmp, _XXXX(vs->aVtx), cWM[0]);
493 ureg_MAD(ureg, tmp, _YYYY(vs->aVtx), cWM[1], ureg_src(tmp));
494 ureg_MAD(ureg, tmp, _ZZZZ(vs->aVtx), cWM[2], ureg_src(tmp));
495 ureg_MAD(ureg, tmp, _WWWW(vs->aVtx), cWM[3], ureg_src(tmp));
496
497 if (has_aNrm) {
498 /* Note: the spec says the transpose of the inverse of the
499 * WorldView matrices should be used, but all tests show
500 * otherwise.
501 * Only case unknown: D3DVBF_0WEIGHTS */
502 ureg_MUL(ureg, tmp2, _XXXX(vs->aNrm), cWM[0]);
503 ureg_MAD(ureg, tmp2, _YYYY(vs->aNrm), cWM[1], ureg_src(tmp2));
504 ureg_MAD(ureg, tmp2, _ZZZZ(vs->aNrm), cWM[2], ureg_src(tmp2));
505 }
506
507 if (i < (key->vertexblend - 1)) {
508 /* accumulate weighted position value */
509 ureg_MAD(ureg, aVtx_dst, ureg_src(tmp), ureg_scalar(vs->aWgt, i), ureg_src(aVtx_dst));
510 if (has_aNrm)
511 ureg_MAD(ureg, aNrm_dst, ureg_src(tmp2), ureg_scalar(vs->aWgt, i), ureg_src(aNrm_dst));
512 /* subtract weighted position value for last value */
513 ureg_ADD(ureg, sum_blendweights, ureg_src(sum_blendweights), ureg_negate(ureg_scalar(vs->aWgt, i)));
514 }
515 }
516
517 /* the last weighted position is always 1 - sum_of_previous_weights */
518 ureg_MAD(ureg, aVtx_dst, ureg_src(tmp), ureg_scalar(ureg_src(sum_blendweights), key->vertexblend - 1), ureg_src(aVtx_dst));
519 if (has_aNrm)
520 ureg_MAD(ureg, aNrm_dst, ureg_src(tmp2), ureg_scalar(ureg_src(sum_blendweights), key->vertexblend - 1), ureg_src(aNrm_dst));
521
522 /* multiply by VIEW_PROJ */
523 ureg_MUL(ureg, tmp, _X(aVtx_dst), _CONST(8));
524 ureg_MAD(ureg, tmp, _Y(aVtx_dst), _CONST(9), ureg_src(tmp));
525 ureg_MAD(ureg, tmp, _Z(aVtx_dst), _CONST(10), ureg_src(tmp));
526 ureg_MAD(ureg, oPos, _W(aVtx_dst), _CONST(11), ureg_src(tmp));
527
528 if (need_aVtx)
529 vs->aVtx = ureg_src(aVtx_dst);
530
531 ureg_release_temporary(ureg, tmp);
532 ureg_release_temporary(ureg, tmp2);
533 ureg_release_temporary(ureg, sum_blendweights);
534 if (!need_aVtx)
535 ureg_release_temporary(ureg, aVtx_dst);
536
537 if (has_aNrm) {
538 if (key->normalizenormals)
539 ureg_normalize3(ureg, aNrm_dst, ureg_src(aNrm_dst));
540 vs->aNrm = ureg_src(aNrm_dst);
541 } else
542 ureg_release_temporary(ureg, aNrm_dst);
543 } else {
544 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
545
546 if (key->vertextween) {
547 struct ureg_dst aVtx_dst = ureg_DECL_temporary(ureg);
548 ureg_LRP(ureg, aVtx_dst, _XXXX(_CONST(30)), vs->aVtx1, vs->aVtx);
549 vs->aVtx = ureg_src(aVtx_dst);
550 if (has_aNrm) {
551 struct ureg_dst aNrm_dst = ureg_DECL_temporary(ureg);
552 ureg_LRP(ureg, aNrm_dst, _XXXX(_CONST(30)), vs->aNrm1, vs->aNrm);
553 vs->aNrm = ureg_src(aNrm_dst);
554 }
555 }
556
557 /* position = vertex * WORLD_VIEW_PROJ */
558 ureg_MUL(ureg, tmp, _XXXX(vs->aVtx), _CONST(0));
559 ureg_MAD(ureg, tmp, _YYYY(vs->aVtx), _CONST(1), ureg_src(tmp));
560 ureg_MAD(ureg, tmp, _ZZZZ(vs->aVtx), _CONST(2), ureg_src(tmp));
561 ureg_MAD(ureg, oPos, _WWWW(vs->aVtx), _CONST(3), ureg_src(tmp));
562 ureg_release_temporary(ureg, tmp);
563
564 if (need_aVtx) {
565 struct ureg_dst aVtx_dst = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
566 ureg_MUL(ureg, aVtx_dst, _XXXX(vs->aVtx), _CONST(4));
567 ureg_MAD(ureg, aVtx_dst, _YYYY(vs->aVtx), _CONST(5), ureg_src(aVtx_dst));
568 ureg_MAD(ureg, aVtx_dst, _ZZZZ(vs->aVtx), _CONST(6), ureg_src(aVtx_dst));
569 ureg_MAD(ureg, aVtx_dst, _WWWW(vs->aVtx), _CONST(7), ureg_src(aVtx_dst));
570 vs->aVtx = ureg_src(aVtx_dst);
571 }
572 if (has_aNrm) {
573 struct ureg_dst aNrm_dst = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
574 ureg_MUL(ureg, aNrm_dst, _XXXX(vs->aNrm), _CONST(16));
575 ureg_MAD(ureg, aNrm_dst, _YYYY(vs->aNrm), _CONST(17), ureg_src(aNrm_dst));
576 ureg_MAD(ureg, aNrm_dst, _ZZZZ(vs->aNrm), _CONST(18), ureg_src(aNrm_dst));
577 if (key->normalizenormals)
578 ureg_normalize3(ureg, aNrm_dst, ureg_src(aNrm_dst));
579 vs->aNrm = ureg_src(aNrm_dst);
580 }
581 }
582
583 /* === Process point size:
584 */
585 if (key->vertexpointsize || key->pointscale) {
586 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
587 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
588 struct ureg_dst tmp_y = ureg_writemask(tmp, TGSI_WRITEMASK_Y);
589 struct ureg_dst tmp_z = ureg_writemask(tmp, TGSI_WRITEMASK_Z);
590 if (key->vertexpointsize) {
591 struct ureg_src cPsz1 = ureg_DECL_constant(ureg, 26);
592 ureg_MAX(ureg, tmp_z, _XXXX(vs->aPsz), _XXXX(cPsz1));
593 ureg_MIN(ureg, tmp_z, _Z(tmp), _YYYY(cPsz1));
594 } else {
595 struct ureg_src cPsz1 = ureg_DECL_constant(ureg, 26);
596 ureg_MOV(ureg, tmp_z, _ZZZZ(cPsz1));
597 }
598
599 if (key->pointscale) {
600 struct ureg_src cPsz1 = ureg_DECL_constant(ureg, 26);
601 struct ureg_src cPsz2 = ureg_DECL_constant(ureg, 27);
602
603 ureg_DP3(ureg, tmp_x, vs->aVtx, vs->aVtx);
604 ureg_RSQ(ureg, tmp_y, _X(tmp));
605 ureg_MUL(ureg, tmp_y, _Y(tmp), _X(tmp));
606 ureg_CMP(ureg, tmp_y, ureg_negate(_Y(tmp)), _Y(tmp), ureg_imm1f(ureg, 0.0f));
607 ureg_MAD(ureg, tmp_x, _Y(tmp), _YYYY(cPsz2), _XXXX(cPsz2));
608 ureg_MAD(ureg, tmp_x, _Y(tmp), _X(tmp), _WWWW(cPsz1));
609 ureg_RSQ(ureg, tmp_x, _X(tmp));
610 ureg_MUL(ureg, tmp_x, _X(tmp), _Z(tmp));
611 ureg_MUL(ureg, tmp_x, _X(tmp), _WWWW(_CONST(100)));
612 ureg_MAX(ureg, tmp_x, _X(tmp), _XXXX(cPsz1));
613 ureg_MIN(ureg, tmp_z, _X(tmp), _YYYY(cPsz1));
614 }
615
616 ureg_MOV(ureg, oPsz, _Z(tmp));
617 ureg_release_temporary(ureg, tmp);
618 }
619
620 for (i = 0; i < 8; ++i) {
621 struct ureg_dst tmp, tmp_x, tmp2;
622 struct ureg_dst oTex, input_coord, transformed, t, aVtx_normed;
623 unsigned c, writemask;
624 const unsigned tci = (key->tc_gen >> (i * 3)) & 0x7;
625 const unsigned idx = (key->tc_idx >> (i * 3)) & 0x7;
626 unsigned dim_input = 1 + ((key->tc_dim_input >> (i * 2)) & 0x3);
627 const unsigned dim_output = (key->tc_dim_output >> (i * 3)) & 0x7;
628
629 /* No texture output of index s */
630 if (tci == NINED3DTSS_TCI_DISABLE)
631 continue;
632 oTex = ureg_DECL_output(ureg, texcoord_sn, i);
633 tmp = ureg_DECL_temporary(ureg);
634 tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
635 input_coord = ureg_DECL_temporary(ureg);
636 transformed = ureg_DECL_temporary(ureg);
637
638 /* Get the coordinate */
639 switch (tci) {
640 case NINED3DTSS_TCI_PASSTHRU:
641 /* NINED3DTSS_TCI_PASSTHRU => Use texcoord coming from index idx *
642 * Else the idx is used only to determine wrapping mode. */
643 vs->aTex[idx] = build_vs_add_input(vs, NINE_DECLUSAGE_i(TEXCOORD,idx));
644 ureg_MOV(ureg, input_coord, vs->aTex[idx]);
645 break;
646 case NINED3DTSS_TCI_CAMERASPACENORMAL:
647 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XYZ), vs->aNrm);
648 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
649 dim_input = 4;
650 break;
651 case NINED3DTSS_TCI_CAMERASPACEPOSITION:
652 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XYZ), vs->aVtx);
653 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
654 dim_input = 4;
655 break;
656 case NINED3DTSS_TCI_CAMERASPACEREFLECTIONVECTOR:
657 tmp.WriteMask = TGSI_WRITEMASK_XYZ;
658 aVtx_normed = ureg_DECL_temporary(ureg);
659 ureg_normalize3(ureg, aVtx_normed, vs->aVtx);
660 ureg_DP3(ureg, tmp_x, ureg_src(aVtx_normed), vs->aNrm);
661 ureg_MUL(ureg, tmp, vs->aNrm, _X(tmp));
662 ureg_ADD(ureg, tmp, ureg_src(tmp), ureg_src(tmp));
663 ureg_ADD(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XYZ), ureg_src(aVtx_normed), ureg_negate(ureg_src(tmp)));
664 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
665 ureg_release_temporary(ureg, aVtx_normed);
666 dim_input = 4;
667 tmp.WriteMask = TGSI_WRITEMASK_XYZW;
668 break;
669 case NINED3DTSS_TCI_SPHEREMAP:
670 /* Implement the formula of GL_SPHERE_MAP */
671 tmp.WriteMask = TGSI_WRITEMASK_XYZ;
672 aVtx_normed = ureg_DECL_temporary(ureg);
673 tmp2 = ureg_DECL_temporary(ureg);
674 ureg_normalize3(ureg, aVtx_normed, vs->aVtx);
675 ureg_DP3(ureg, tmp_x, ureg_src(aVtx_normed), vs->aNrm);
676 ureg_MUL(ureg, tmp, vs->aNrm, _X(tmp));
677 ureg_ADD(ureg, tmp, ureg_src(tmp), ureg_src(tmp));
678 ureg_ADD(ureg, tmp, ureg_src(aVtx_normed), ureg_negate(ureg_src(tmp)));
679 /* now tmp = normed(Vtx) - 2 dot3(normed(Vtx), Nrm) Nrm */
680 ureg_MOV(ureg, ureg_writemask(tmp2, TGSI_WRITEMASK_XYZ), ureg_src(tmp));
681 ureg_MUL(ureg, tmp2, ureg_src(tmp2), ureg_src(tmp2));
682 ureg_DP3(ureg, ureg_writemask(tmp2, TGSI_WRITEMASK_X), ureg_src(tmp2), ureg_src(tmp2));
683 ureg_RSQ(ureg, ureg_writemask(tmp2, TGSI_WRITEMASK_X), ureg_src(tmp2));
684 ureg_MUL(ureg, ureg_writemask(tmp2, TGSI_WRITEMASK_X), ureg_src(tmp2), ureg_imm1f(ureg, 0.5f));
685 /* tmp2 = 0.5 / sqrt(tmp.x^2 + tmp.y^2 + (tmp.z+1)^2)
686 * TODO: z coordinates are a bit different gl vs d3d, should the formula be adapted ? */
687 ureg_MUL(ureg, tmp, ureg_src(tmp), _X(tmp2));
688 ureg_ADD(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XY), ureg_src(tmp), ureg_imm1f(ureg, 0.5f));
689 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_ZW), ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 1.0f));
690 ureg_release_temporary(ureg, aVtx_normed);
691 ureg_release_temporary(ureg, tmp2);
692 dim_input = 4;
693 tmp.WriteMask = TGSI_WRITEMASK_XYZW;
694 break;
695 default:
696 assert(0);
697 break;
698 }
699
700 /* Apply the transformation */
701 /* dim_output == 0 => do not transform the components.
702 * XYZRHW also disables transformation */
703 if (!dim_output || key->position_t) {
704 ureg_release_temporary(ureg, transformed);
705 transformed = input_coord;
706 writemask = TGSI_WRITEMASK_XYZW;
707 } else {
708 for (c = 0; c < dim_output; c++) {
709 t = ureg_writemask(transformed, 1 << c);
710 switch (dim_input) {
711 /* dim_input = 1 2 3: -> we add trailing 1 to input*/
712 case 1: ureg_MAD(ureg, t, _X(input_coord), _XXXX(_CONST(128 + i * 4 + c)), _YYYY(_CONST(128 + i * 4 + c)));
713 break;
714 case 2: ureg_DP2(ureg, t, ureg_src(input_coord), _CONST(128 + i * 4 + c));
715 ureg_ADD(ureg, t, ureg_src(transformed), _ZZZZ(_CONST(128 + i * 4 + c)));
716 break;
717 case 3: ureg_DP3(ureg, t, ureg_src(input_coord), _CONST(128 + i * 4 + c));
718 ureg_ADD(ureg, t, ureg_src(transformed), _WWWW(_CONST(128 + i * 4 + c)));
719 break;
720 case 4: ureg_DP4(ureg, t, ureg_src(input_coord), _CONST(128 + i * 4 + c)); break;
721 default:
722 assert(0);
723 }
724 }
725 writemask = (1 << dim_output) - 1;
726 ureg_release_temporary(ureg, input_coord);
727 }
728
729 ureg_MOV(ureg, ureg_writemask(oTex, writemask), ureg_src(transformed));
730 ureg_release_temporary(ureg, transformed);
731 ureg_release_temporary(ureg, tmp);
732 }
733
734 /* === Lighting:
735 *
736 * DIRECTIONAL: Light at infinite distance, parallel rays, no attenuation.
737 * POINT: Finite distance to scene, divergent rays, isotropic, attenuation.
738 * SPOT: Finite distance, divergent rays, angular dependence, attenuation.
739 *
740 * vec3 normal = normalize(in.Normal * NormalMatrix);
741 * vec3 hitDir = light.direction;
742 * float atten = 1.0;
743 *
744 * if (light.type != DIRECTIONAL)
745 * {
746 * vec3 hitVec = light.position - eyeVertex;
747 * float d = length(hitVec);
748 * hitDir = hitVec / d;
749 * atten = 1 / ((light.atten2 * d + light.atten1) * d + light.atten0);
750 * }
751 *
752 * if (light.type == SPOTLIGHT)
753 * {
754 * float rho = dp3(-hitVec, light.direction);
755 * if (rho < cos(light.phi / 2))
756 * atten = 0;
757 * if (rho < cos(light.theta / 2))
758 * atten *= pow(some_func(rho), light.falloff);
759 * }
760 *
761 * float nDotHit = dp3_sat(normal, hitVec);
762 * float powFact = 0.0;
763 *
764 * if (nDotHit > 0.0)
765 * {
766 * vec3 midVec = normalize(hitDir + eye);
767 * float nDotMid = dp3_sat(normal, midVec);
768 * pFact = pow(nDotMid, material.power);
769 * }
770 *
771 * ambient += light.ambient * atten;
772 * diffuse += light.diffuse * atten * nDotHit;
773 * specular += light.specular * atten * powFact;
774 */
775 if (key->lighting) {
776 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
777 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
778 struct ureg_dst tmp_y = ureg_writemask(tmp, TGSI_WRITEMASK_Y);
779 struct ureg_dst tmp_z = ureg_writemask(tmp, TGSI_WRITEMASK_Z);
780 struct ureg_dst rAtt = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_W);
781 struct ureg_dst rHit = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
782 struct ureg_dst rMid = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
783
784 struct ureg_dst rCtr = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_W);
785
786 struct ureg_dst AL = ureg_writemask(AR, TGSI_WRITEMASK_X);
787
788 /* Light.*.Alpha is not used. */
789 struct ureg_dst rD = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
790 struct ureg_dst rA = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
791 struct ureg_dst rS = ureg_DECL_temporary(ureg);
792
793 struct ureg_src mtlP = _XXXX(MATERIAL_CONST(4));
794
795 struct ureg_src cLKind = _XXXX(LIGHT_CONST(0));
796 struct ureg_src cLAtt0 = _YYYY(LIGHT_CONST(0));
797 struct ureg_src cLAtt1 = _ZZZZ(LIGHT_CONST(0));
798 struct ureg_src cLAtt2 = _WWWW(LIGHT_CONST(0));
799 struct ureg_src cLColD = _XYZW(LIGHT_CONST(1));
800 struct ureg_src cLColS = _XYZW(LIGHT_CONST(2));
801 struct ureg_src cLColA = _XYZW(LIGHT_CONST(3));
802 struct ureg_src cLPos = _XYZW(LIGHT_CONST(4));
803 struct ureg_src cLRng = _WWWW(LIGHT_CONST(4));
804 struct ureg_src cLDir = _XYZW(LIGHT_CONST(5));
805 struct ureg_src cLFOff = _WWWW(LIGHT_CONST(5));
806 struct ureg_src cLTht = _XXXX(LIGHT_CONST(6));
807 struct ureg_src cLPhi = _YYYY(LIGHT_CONST(6));
808 struct ureg_src cLSDiv = _ZZZZ(LIGHT_CONST(6));
809 struct ureg_src cLLast = _WWWW(LIGHT_CONST(7));
810
811 const unsigned loop_label = l++;
812
813 ureg_MOV(ureg, rCtr, ureg_imm1f(ureg, 32.0f)); /* &lightconst(0) */
814 ureg_MOV(ureg, rD, ureg_imm1f(ureg, 0.0f));
815 ureg_MOV(ureg, rA, ureg_imm1f(ureg, 0.0f));
816 ureg_MOV(ureg, rS, ureg_imm1f(ureg, 0.0f));
817
818 /* loop management */
819 ureg_BGNLOOP(ureg, &label[loop_label]);
820 ureg_ARL(ureg, AL, _W(rCtr));
821
822 /* if (not DIRECTIONAL light): */
823 ureg_SNE(ureg, tmp_x, cLKind, ureg_imm1f(ureg, D3DLIGHT_DIRECTIONAL));
824 ureg_MOV(ureg, rHit, ureg_negate(cLDir));
825 ureg_MOV(ureg, rAtt, ureg_imm1f(ureg, 1.0f));
826 ureg_IF(ureg, _X(tmp), &label[l++]);
827 {
828 /* hitDir = light.position - eyeVtx
829 * d = length(hitDir)
830 */
831 ureg_ADD(ureg, rHit, cLPos, ureg_negate(vs->aVtx));
832 ureg_DP3(ureg, tmp_x, ureg_src(rHit), ureg_src(rHit));
833 ureg_RSQ(ureg, tmp_y, _X(tmp));
834 ureg_MUL(ureg, tmp_x, _X(tmp), _Y(tmp)); /* length */
835
836 /* att = 1.0 / (light.att0 + (light.att1 + light.att2 * d) * d) */
837 ureg_MAD(ureg, rAtt, _X(tmp), cLAtt2, cLAtt1);
838 ureg_MAD(ureg, rAtt, _X(tmp), _W(rAtt), cLAtt0);
839 ureg_RCP(ureg, rAtt, _W(rAtt));
840 /* cut-off if distance exceeds Light.Range */
841 ureg_SLT(ureg, tmp_x, _X(tmp), cLRng);
842 ureg_MUL(ureg, rAtt, _W(rAtt), _X(tmp));
843 }
844 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
845 ureg_ENDIF(ureg);
846
847 /* normalize hitDir */
848 ureg_normalize3(ureg, rHit, ureg_src(rHit));
849
850 /* if (SPOT light) */
851 ureg_SEQ(ureg, tmp_x, cLKind, ureg_imm1f(ureg, D3DLIGHT_SPOT));
852 ureg_IF(ureg, _X(tmp), &label[l++]);
853 {
854 /* rho = dp3(-hitDir, light.spotDir)
855 *
856 * if (rho > light.ctht2) NOTE: 0 <= phi <= pi, 0 <= theta <= phi
857 * spotAtt = 1
858 * else
859 * if (rho <= light.cphi2)
860 * spotAtt = 0
861 * else
862 * spotAtt = (rho - light.cphi2) / (light.ctht2 - light.cphi2) ^ light.falloff
863 */
864 ureg_DP3(ureg, tmp_y, ureg_negate(ureg_src(rHit)), cLDir); /* rho */
865 ureg_ADD(ureg, tmp_x, _Y(tmp), ureg_negate(cLPhi));
866 ureg_MUL(ureg, tmp_x, _X(tmp), cLSDiv);
867 ureg_POW(ureg, tmp_x, _X(tmp), cLFOff); /* spotAtten */
868 ureg_SGE(ureg, tmp_z, _Y(tmp), cLTht); /* if inside theta && phi */
869 ureg_SGE(ureg, tmp_y, _Y(tmp), cLPhi); /* if inside phi */
870 ureg_MAD(ureg, ureg_saturate(tmp_x), _X(tmp), _Y(tmp), _Z(tmp));
871 ureg_MUL(ureg, rAtt, _W(rAtt), _X(tmp));
872 }
873 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
874 ureg_ENDIF(ureg);
875
876 /* directional factors, let's not use LIT because of clarity */
877
878 if (has_aNrm) {
879 if (key->localviewer) {
880 ureg_normalize3(ureg, rMid, vs->aVtx);
881 ureg_ADD(ureg, rMid, ureg_src(rHit), ureg_negate(ureg_src(rMid)));
882 } else {
883 ureg_ADD(ureg, rMid, ureg_src(rHit), ureg_imm3f(ureg, 0.0f, 0.0f, -1.0f));
884 }
885 ureg_normalize3(ureg, rMid, ureg_src(rMid));
886 ureg_DP3(ureg, ureg_saturate(tmp_x), vs->aNrm, ureg_src(rHit));
887 ureg_DP3(ureg, ureg_saturate(tmp_y), vs->aNrm, ureg_src(rMid));
888 ureg_MUL(ureg, tmp_z, _X(tmp), _Y(tmp));
889 /* Tests show that specular is computed only if (dp3(normal,hitDir) > 0).
890 * For front facing, it is more restrictive than test (dp3(normal,mid) > 0).
891 * No tests were made for backfacing, so add the two conditions */
892 ureg_IF(ureg, _Z(tmp), &label[l++]);
893 {
894 ureg_DP3(ureg, ureg_saturate(tmp_y), vs->aNrm, ureg_src(rMid));
895 ureg_POW(ureg, tmp_y, _Y(tmp), mtlP);
896 ureg_MUL(ureg, tmp_y, _W(rAtt), _Y(tmp)); /* power factor * att */
897 ureg_MAD(ureg, rS, cLColS, _Y(tmp), ureg_src(rS)); /* accumulate specular */
898 }
899 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
900 ureg_ENDIF(ureg);
901
902 ureg_MUL(ureg, tmp_x, _W(rAtt), _X(tmp)); /* dp3(normal,hitDir) * att */
903 ureg_MAD(ureg, rD, cLColD, _X(tmp), ureg_src(rD)); /* accumulate diffuse */
904 }
905
906 ureg_MAD(ureg, rA, cLColA, _W(rAtt), ureg_src(rA)); /* accumulate ambient */
907
908 /* break if this was the last light */
909 ureg_IF(ureg, cLLast, &label[l++]);
910 ureg_BRK(ureg);
911 ureg_ENDIF(ureg);
912 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
913
914 ureg_ADD(ureg, rCtr, _W(rCtr), ureg_imm1f(ureg, 8.0f));
915 ureg_fixup_label(ureg, label[loop_label], ureg_get_instruction_number(ureg));
916 ureg_ENDLOOP(ureg, &label[loop_label]);
917
918 /* Apply to material:
919 *
920 * oCol[0] = (material.emissive + material.ambient * rs.ambient) +
921 * material.ambient * ambient +
922 * material.diffuse * diffuse +
923 * oCol[1] = material.specular * specular;
924 */
925 if (key->mtl_emissive == 0 && key->mtl_ambient == 0)
926 ureg_MAD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(rA), vs->mtlA, _CONST(19));
927 else {
928 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(rA), _CONST(25));
929 ureg_MAD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), vs->mtlA, ureg_src(tmp), vs->mtlE);
930 }
931
932 ureg_MAD(ureg, ureg_writemask(oCol[0], TGSI_WRITEMASK_XYZ), ureg_src(rD), vs->mtlD, ureg_src(tmp));
933 ureg_MOV(ureg, ureg_writemask(oCol[0], TGSI_WRITEMASK_W), vs->mtlD);
934 ureg_MUL(ureg, oCol[1], ureg_src(rS), vs->mtlS);
935 ureg_release_temporary(ureg, rAtt);
936 ureg_release_temporary(ureg, rHit);
937 ureg_release_temporary(ureg, rMid);
938 ureg_release_temporary(ureg, rCtr);
939 ureg_release_temporary(ureg, rD);
940 ureg_release_temporary(ureg, rA);
941 ureg_release_temporary(ureg, rS);
942 ureg_release_temporary(ureg, rAtt);
943 ureg_release_temporary(ureg, tmp);
944 } else
945 /* COLOR */
946 if (key->darkness) {
947 if (key->mtl_emissive == 0 && key->mtl_ambient == 0)
948 ureg_MOV(ureg, ureg_writemask(oCol[0], TGSI_WRITEMASK_XYZ), _CONST(19));
949 else
950 ureg_MAD(ureg, ureg_writemask(oCol[0], TGSI_WRITEMASK_XYZ), vs->mtlA, _CONST(25), vs->mtlE);
951 ureg_MOV(ureg, ureg_writemask(oCol[0], TGSI_WRITEMASK_W), vs->mtlD);
952 ureg_MOV(ureg, oCol[1], ureg_imm1f(ureg, 0.0f));
953 } else {
954 ureg_MOV(ureg, oCol[0], vs->aCol[0]);
955 ureg_MOV(ureg, oCol[1], vs->aCol[1]);
956 }
957
958 /* === Process fog.
959 *
960 * exp(x) = ex2(log2(e) * x)
961 */
962 if (key->fog_mode) {
963 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
964 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
965 struct ureg_dst tmp_z = ureg_writemask(tmp, TGSI_WRITEMASK_Z);
966 if (key->fog_range) {
967 ureg_DP3(ureg, tmp_x, vs->aVtx, vs->aVtx);
968 ureg_RSQ(ureg, tmp_z, _X(tmp));
969 ureg_MUL(ureg, tmp_z, _Z(tmp), _X(tmp));
970 } else {
971 ureg_MOV(ureg, tmp_z, ureg_abs(_ZZZZ(vs->aVtx)));
972 }
973
974 if (key->fog_mode == D3DFOG_EXP) {
975 ureg_MUL(ureg, tmp_x, _Z(tmp), _ZZZZ(_CONST(28)));
976 ureg_MUL(ureg, tmp_x, _X(tmp), ureg_imm1f(ureg, -1.442695f));
977 ureg_EX2(ureg, tmp_x, _X(tmp));
978 } else
979 if (key->fog_mode == D3DFOG_EXP2) {
980 ureg_MUL(ureg, tmp_x, _Z(tmp), _ZZZZ(_CONST(28)));
981 ureg_MUL(ureg, tmp_x, _X(tmp), _X(tmp));
982 ureg_MUL(ureg, tmp_x, _X(tmp), ureg_imm1f(ureg, -1.442695f));
983 ureg_EX2(ureg, tmp_x, _X(tmp));
984 } else
985 if (key->fog_mode == D3DFOG_LINEAR) {
986 ureg_ADD(ureg, tmp_x, _XXXX(_CONST(28)), ureg_negate(_Z(tmp)));
987 ureg_MUL(ureg, ureg_saturate(tmp_x), _X(tmp), _YYYY(_CONST(28)));
988 }
989 ureg_MOV(ureg, oFog, _X(tmp));
990 ureg_release_temporary(ureg, tmp);
991 } else if (key->fog && !(key->passthrough & (1 << NINE_DECLUSAGE_FOG))) {
992 ureg_MOV(ureg, oFog, ureg_scalar(vs->aCol[1], TGSI_SWIZZLE_W));
993 }
994
995 if (key->passthrough & (1 << NINE_DECLUSAGE_BLENDWEIGHT)) {
996 struct ureg_src input;
997 struct ureg_dst output;
998 input = vs->aWgt;
999 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 18);
1000 ureg_MOV(ureg, output, input);
1001 }
1002 if (key->passthrough & (1 << NINE_DECLUSAGE_BLENDINDICES)) {
1003 struct ureg_src input;
1004 struct ureg_dst output;
1005 input = vs->aInd;
1006 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 19);
1007 ureg_MOV(ureg, output, input);
1008 }
1009 if (key->passthrough & (1 << NINE_DECLUSAGE_NORMAL)) {
1010 struct ureg_src input;
1011 struct ureg_dst output;
1012 input = vs->aNrm;
1013 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 20);
1014 ureg_MOV(ureg, output, input);
1015 }
1016 if (key->passthrough & (1 << NINE_DECLUSAGE_TANGENT)) {
1017 struct ureg_src input;
1018 struct ureg_dst output;
1019 input = build_vs_add_input(vs, NINE_DECLUSAGE_TANGENT);
1020 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 21);
1021 ureg_MOV(ureg, output, input);
1022 }
1023 if (key->passthrough & (1 << NINE_DECLUSAGE_BINORMAL)) {
1024 struct ureg_src input;
1025 struct ureg_dst output;
1026 input = build_vs_add_input(vs, NINE_DECLUSAGE_BINORMAL);
1027 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 22);
1028 ureg_MOV(ureg, output, input);
1029 }
1030 if (key->passthrough & (1 << NINE_DECLUSAGE_FOG)) {
1031 struct ureg_src input;
1032 struct ureg_dst output;
1033 input = build_vs_add_input(vs, NINE_DECLUSAGE_FOG);
1034 input = ureg_scalar(input, TGSI_SWIZZLE_X);
1035 output = oFog;
1036 ureg_MOV(ureg, output, input);
1037 }
1038 if (key->passthrough & (1 << NINE_DECLUSAGE_DEPTH)) {
1039 (void) 0; /* TODO: replace z of position output ? */
1040 }
1041
1042 /* ucp for ff applies on world coordinates.
1043 * aVtx is in worldview coordinates. */
1044 if (key->ucp) {
1045 struct ureg_dst clipVect = ureg_DECL_output(ureg, TGSI_SEMANTIC_CLIPVERTEX, 0);
1046 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
1047 ureg_MUL(ureg, tmp, _XXXX(vs->aVtx), _CONST(12));
1048 ureg_MAD(ureg, tmp, _YYYY(vs->aVtx), _CONST(13), ureg_src(tmp));
1049 ureg_MAD(ureg, tmp, _ZZZZ(vs->aVtx), _CONST(14), ureg_src(tmp));
1050 ureg_ADD(ureg, clipVect, _CONST(15), ureg_src(tmp));
1051 ureg_release_temporary(ureg, tmp);
1052 }
1053
1054 if (key->position_t && device->driver_caps.window_space_position_support)
1055 ureg_property(ureg, TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION, TRUE);
1056
1057 ureg_END(ureg);
1058 nine_ureg_tgsi_dump(ureg, FALSE);
1059 return ureg_create_shader_and_destroy(ureg, device->context.pipe);
1060 }
1061
1062 /* PS FF constants layout:
1063 *
1064 * CONST[ 0.. 7] stage[i].D3DTSS_CONSTANT
1065 * CONST[ 8..15].x___ stage[i].D3DTSS_BUMPENVMAT00
1066 * CONST[ 8..15]._y__ stage[i].D3DTSS_BUMPENVMAT01
1067 * CONST[ 8..15].__z_ stage[i].D3DTSS_BUMPENVMAT10
1068 * CONST[ 8..15].___w stage[i].D3DTSS_BUMPENVMAT11
1069 * CONST[16..19].x_z_ stage[i].D3DTSS_BUMPENVLSCALE
1070 * CONST[17..19]._y_w stage[i].D3DTSS_BUMPENVLOFFSET
1071 *
1072 * CONST[20] D3DRS_TEXTUREFACTOR
1073 * CONST[21] D3DRS_FOGCOLOR
1074 * CONST[22].x___ RS.FogEnd
1075 * CONST[22]._y__ 1.0f / (RS.FogEnd - RS.FogStart)
1076 * CONST[22].__z_ RS.FogDensity
1077 */
1078 struct ps_build_ctx
1079 {
1080 struct ureg_program *ureg;
1081
1082 struct ureg_src vC[2]; /* DIFFUSE, SPECULAR */
1083 struct ureg_src vT[8]; /* TEXCOORD[i] */
1084 struct ureg_dst rCur; /* D3DTA_CURRENT */
1085 struct ureg_dst rMod;
1086 struct ureg_src rCurSrc;
1087 struct ureg_dst rTmp; /* D3DTA_TEMP */
1088 struct ureg_src rTmpSrc;
1089 struct ureg_dst rTex;
1090 struct ureg_src rTexSrc;
1091 struct ureg_src cBEM[8];
1092 struct ureg_src s[8];
1093
1094 struct {
1095 unsigned index;
1096 unsigned index_pre_mod;
1097 } stage;
1098 };
1099
1100 static struct ureg_src
1101 ps_get_ts_arg(struct ps_build_ctx *ps, unsigned ta)
1102 {
1103 struct ureg_src reg;
1104
1105 switch (ta & D3DTA_SELECTMASK) {
1106 case D3DTA_CONSTANT:
1107 reg = ureg_DECL_constant(ps->ureg, ps->stage.index);
1108 break;
1109 case D3DTA_CURRENT:
1110 reg = (ps->stage.index == ps->stage.index_pre_mod) ? ureg_src(ps->rMod) : ps->rCurSrc;
1111 break;
1112 case D3DTA_DIFFUSE:
1113 reg = ureg_DECL_fs_input(ps->ureg, TGSI_SEMANTIC_COLOR, 0, TGSI_INTERPOLATE_COLOR);
1114 break;
1115 case D3DTA_SPECULAR:
1116 reg = ureg_DECL_fs_input(ps->ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_COLOR);
1117 break;
1118 case D3DTA_TEMP:
1119 reg = ps->rTmpSrc;
1120 break;
1121 case D3DTA_TEXTURE:
1122 reg = ps->rTexSrc;
1123 break;
1124 case D3DTA_TFACTOR:
1125 reg = ureg_DECL_constant(ps->ureg, 20);
1126 break;
1127 default:
1128 assert(0);
1129 reg = ureg_src_undef();
1130 break;
1131 }
1132 if (ta & D3DTA_COMPLEMENT) {
1133 struct ureg_dst dst = ureg_DECL_temporary(ps->ureg);
1134 ureg_ADD(ps->ureg, dst, ureg_imm1f(ps->ureg, 1.0f), ureg_negate(reg));
1135 reg = ureg_src(dst);
1136 }
1137 if (ta & D3DTA_ALPHAREPLICATE)
1138 reg = _WWWW(reg);
1139 return reg;
1140 }
1141
1142 static struct ureg_dst
1143 ps_get_ts_dst(struct ps_build_ctx *ps, unsigned ta)
1144 {
1145 assert(!(ta & (D3DTA_COMPLEMENT | D3DTA_ALPHAREPLICATE)));
1146
1147 switch (ta & D3DTA_SELECTMASK) {
1148 case D3DTA_CURRENT:
1149 return ps->rCur;
1150 case D3DTA_TEMP:
1151 return ps->rTmp;
1152 default:
1153 assert(0);
1154 return ureg_dst_undef();
1155 }
1156 }
1157
1158 static uint8_t ps_d3dtop_args_mask(D3DTEXTUREOP top)
1159 {
1160 switch (top) {
1161 case D3DTOP_DISABLE:
1162 return 0x0;
1163 case D3DTOP_SELECTARG1:
1164 case D3DTOP_PREMODULATE:
1165 return 0x2;
1166 case D3DTOP_SELECTARG2:
1167 return 0x4;
1168 case D3DTOP_MULTIPLYADD:
1169 case D3DTOP_LERP:
1170 return 0x7;
1171 default:
1172 return 0x6;
1173 }
1174 }
1175
1176 static inline boolean
1177 is_MOV_no_op(struct ureg_dst dst, struct ureg_src src)
1178 {
1179 return !dst.WriteMask ||
1180 (dst.File == src.File &&
1181 dst.Index == src.Index &&
1182 !dst.Indirect &&
1183 !dst.Saturate &&
1184 !src.Indirect &&
1185 !src.Negate &&
1186 !src.Absolute &&
1187 (!(dst.WriteMask & TGSI_WRITEMASK_X) || (src.SwizzleX == TGSI_SWIZZLE_X)) &&
1188 (!(dst.WriteMask & TGSI_WRITEMASK_Y) || (src.SwizzleY == TGSI_SWIZZLE_Y)) &&
1189 (!(dst.WriteMask & TGSI_WRITEMASK_Z) || (src.SwizzleZ == TGSI_SWIZZLE_Z)) &&
1190 (!(dst.WriteMask & TGSI_WRITEMASK_W) || (src.SwizzleW == TGSI_SWIZZLE_W)));
1191
1192 }
1193
1194 static void
1195 ps_do_ts_op(struct ps_build_ctx *ps, unsigned top, struct ureg_dst dst, struct ureg_src *arg)
1196 {
1197 struct ureg_program *ureg = ps->ureg;
1198 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
1199 struct ureg_dst tmp2 = ureg_DECL_temporary(ureg);
1200 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
1201
1202 tmp.WriteMask = dst.WriteMask;
1203
1204 if (top != D3DTOP_SELECTARG1 && top != D3DTOP_SELECTARG2 &&
1205 top != D3DTOP_MODULATE && top != D3DTOP_PREMODULATE &&
1206 top != D3DTOP_BLENDDIFFUSEALPHA && top != D3DTOP_BLENDTEXTUREALPHA &&
1207 top != D3DTOP_BLENDFACTORALPHA && top != D3DTOP_BLENDCURRENTALPHA &&
1208 top != D3DTOP_BUMPENVMAP && top != D3DTOP_BUMPENVMAPLUMINANCE &&
1209 top != D3DTOP_LERP)
1210 dst = ureg_saturate(dst);
1211
1212 switch (top) {
1213 case D3DTOP_SELECTARG1:
1214 if (!is_MOV_no_op(dst, arg[1]))
1215 ureg_MOV(ureg, dst, arg[1]);
1216 break;
1217 case D3DTOP_SELECTARG2:
1218 if (!is_MOV_no_op(dst, arg[2]))
1219 ureg_MOV(ureg, dst, arg[2]);
1220 break;
1221 case D3DTOP_MODULATE:
1222 ureg_MUL(ureg, dst, arg[1], arg[2]);
1223 break;
1224 case D3DTOP_MODULATE2X:
1225 ureg_MUL(ureg, tmp, arg[1], arg[2]);
1226 ureg_ADD(ureg, dst, ureg_src(tmp), ureg_src(tmp));
1227 break;
1228 case D3DTOP_MODULATE4X:
1229 ureg_MUL(ureg, tmp, arg[1], arg[2]);
1230 ureg_MUL(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 4.0f));
1231 break;
1232 case D3DTOP_ADD:
1233 ureg_ADD(ureg, dst, arg[1], arg[2]);
1234 break;
1235 case D3DTOP_ADDSIGNED:
1236 ureg_ADD(ureg, tmp, arg[1], arg[2]);
1237 ureg_ADD(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, -0.5f));
1238 break;
1239 case D3DTOP_ADDSIGNED2X:
1240 ureg_ADD(ureg, tmp, arg[1], arg[2]);
1241 ureg_MAD(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 2.0f), ureg_imm1f(ureg, -1.0f));
1242 break;
1243 case D3DTOP_SUBTRACT:
1244 ureg_ADD(ureg, dst, arg[1], ureg_negate(arg[2]));
1245 break;
1246 case D3DTOP_ADDSMOOTH:
1247 ureg_ADD(ureg, tmp, ureg_imm1f(ureg, 1.0f), ureg_negate(arg[1]));
1248 ureg_MAD(ureg, dst, ureg_src(tmp), arg[2], arg[1]);
1249 break;
1250 case D3DTOP_BLENDDIFFUSEALPHA:
1251 ureg_LRP(ureg, dst, _WWWW(ps->vC[0]), arg[1], arg[2]);
1252 break;
1253 case D3DTOP_BLENDTEXTUREALPHA:
1254 /* XXX: alpha taken from previous stage, texture or result ? */
1255 ureg_LRP(ureg, dst, _W(ps->rTex), arg[1], arg[2]);
1256 break;
1257 case D3DTOP_BLENDFACTORALPHA:
1258 ureg_LRP(ureg, dst, _WWWW(_CONST(20)), arg[1], arg[2]);
1259 break;
1260 case D3DTOP_BLENDTEXTUREALPHAPM:
1261 ureg_ADD(ureg, tmp_x, ureg_imm1f(ureg, 1.0f), ureg_negate(_W(ps->rTex)));
1262 ureg_MAD(ureg, dst, arg[2], _X(tmp), arg[1]);
1263 break;
1264 case D3DTOP_BLENDCURRENTALPHA:
1265 ureg_LRP(ureg, dst, _WWWW(ps->rCurSrc), arg[1], arg[2]);
1266 break;
1267 case D3DTOP_PREMODULATE:
1268 ureg_MOV(ureg, dst, arg[1]);
1269 ps->stage.index_pre_mod = ps->stage.index + 1;
1270 break;
1271 case D3DTOP_MODULATEALPHA_ADDCOLOR:
1272 ureg_MAD(ureg, dst, _WWWW(arg[1]), arg[2], arg[1]);
1273 break;
1274 case D3DTOP_MODULATECOLOR_ADDALPHA:
1275 ureg_MAD(ureg, dst, arg[1], arg[2], _WWWW(arg[1]));
1276 break;
1277 case D3DTOP_MODULATEINVALPHA_ADDCOLOR:
1278 ureg_ADD(ureg, tmp_x, ureg_imm1f(ureg, 1.0f), ureg_negate(_WWWW(arg[1])));
1279 ureg_MAD(ureg, dst, _X(tmp), arg[2], arg[1]);
1280 break;
1281 case D3DTOP_MODULATEINVCOLOR_ADDALPHA:
1282 ureg_ADD(ureg, tmp, ureg_imm1f(ureg, 1.0f), ureg_negate(arg[1]));
1283 ureg_MAD(ureg, dst, ureg_src(tmp), arg[2], _WWWW(arg[1]));
1284 break;
1285 case D3DTOP_BUMPENVMAP:
1286 break;
1287 case D3DTOP_BUMPENVMAPLUMINANCE:
1288 break;
1289 case D3DTOP_DOTPRODUCT3:
1290 ureg_ADD(ureg, tmp, arg[1], ureg_imm4f(ureg,-0.5,-0.5,-0.5,-0.5));
1291 ureg_ADD(ureg, tmp2, arg[2] , ureg_imm4f(ureg,-0.5,-0.5,-0.5,-0.5));
1292 ureg_DP3(ureg, tmp, ureg_src(tmp), ureg_src(tmp2));
1293 ureg_MUL(ureg, ureg_saturate(dst), ureg_src(tmp), ureg_imm4f(ureg,4.0,4.0,4.0,4.0));
1294 break;
1295 case D3DTOP_MULTIPLYADD:
1296 ureg_MAD(ureg, dst, arg[1], arg[2], arg[0]);
1297 break;
1298 case D3DTOP_LERP:
1299 ureg_LRP(ureg, dst, arg[0], arg[1], arg[2]);
1300 break;
1301 case D3DTOP_DISABLE:
1302 /* no-op ? */
1303 break;
1304 default:
1305 assert(!"invalid D3DTOP");
1306 break;
1307 }
1308 ureg_release_temporary(ureg, tmp);
1309 ureg_release_temporary(ureg, tmp2);
1310 }
1311
1312 static void *
1313 nine_ff_build_ps(struct NineDevice9 *device, struct nine_ff_ps_key *key)
1314 {
1315 struct ps_build_ctx ps;
1316 struct ureg_program *ureg = ureg_create(PIPE_SHADER_FRAGMENT);
1317 struct ureg_dst oCol;
1318 unsigned s;
1319 const unsigned texcoord_sn = get_texcoord_sn(device->screen);
1320
1321 memset(&ps, 0, sizeof(ps));
1322 ps.ureg = ureg;
1323 ps.stage.index_pre_mod = -1;
1324
1325 ps.vC[0] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 0, TGSI_INTERPOLATE_COLOR);
1326
1327 ps.rCur = ureg_DECL_temporary(ureg);
1328 ps.rTmp = ureg_DECL_temporary(ureg);
1329 ps.rTex = ureg_DECL_temporary(ureg);
1330 ps.rCurSrc = ureg_src(ps.rCur);
1331 ps.rTmpSrc = ureg_src(ps.rTmp);
1332 ps.rTexSrc = ureg_src(ps.rTex);
1333
1334 /* Initial values */
1335 ureg_MOV(ureg, ps.rCur, ps.vC[0]);
1336 ureg_MOV(ureg, ps.rTmp, ureg_imm1f(ureg, 0.0f));
1337 ureg_MOV(ureg, ps.rTex, ureg_imm1f(ureg, 0.0f));
1338
1339 for (s = 0; s < 8; ++s) {
1340 ps.s[s] = ureg_src_undef();
1341
1342 if (key->ts[s].colorop != D3DTOP_DISABLE) {
1343 if (key->ts[s].colorarg0 == D3DTA_SPECULAR ||
1344 key->ts[s].colorarg1 == D3DTA_SPECULAR ||
1345 key->ts[s].colorarg2 == D3DTA_SPECULAR)
1346 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_COLOR);
1347
1348 if (key->ts[s].colorarg0 == D3DTA_TEXTURE ||
1349 key->ts[s].colorarg1 == D3DTA_TEXTURE ||
1350 key->ts[s].colorarg2 == D3DTA_TEXTURE) {
1351 ps.s[s] = ureg_DECL_sampler(ureg, s);
1352 ps.vT[s] = ureg_DECL_fs_input(ureg, texcoord_sn, s, TGSI_INTERPOLATE_PERSPECTIVE);
1353 }
1354 if (s && (key->ts[s - 1].colorop == D3DTOP_PREMODULATE ||
1355 key->ts[s - 1].alphaop == D3DTOP_PREMODULATE))
1356 ps.s[s] = ureg_DECL_sampler(ureg, s);
1357 }
1358
1359 if (key->ts[s].alphaop != D3DTOP_DISABLE) {
1360 if (key->ts[s].alphaarg0 == D3DTA_SPECULAR ||
1361 key->ts[s].alphaarg1 == D3DTA_SPECULAR ||
1362 key->ts[s].alphaarg2 == D3DTA_SPECULAR)
1363 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_COLOR);
1364
1365 if (key->ts[s].alphaarg0 == D3DTA_TEXTURE ||
1366 key->ts[s].alphaarg1 == D3DTA_TEXTURE ||
1367 key->ts[s].alphaarg2 == D3DTA_TEXTURE) {
1368 ps.s[s] = ureg_DECL_sampler(ureg, s);
1369 ps.vT[s] = ureg_DECL_fs_input(ureg, texcoord_sn, s, TGSI_INTERPOLATE_PERSPECTIVE);
1370 }
1371 }
1372 }
1373 if (key->specular)
1374 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_COLOR);
1375
1376 oCol = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
1377
1378 /* Run stages.
1379 */
1380 for (s = 0; s < 8; ++s) {
1381 unsigned colorarg[3];
1382 unsigned alphaarg[3];
1383 const uint8_t used_c = ps_d3dtop_args_mask(key->ts[s].colorop);
1384 const uint8_t used_a = ps_d3dtop_args_mask(key->ts[s].alphaop);
1385 struct ureg_dst dst;
1386 struct ureg_src arg[3];
1387
1388 if (key->ts[s].colorop == D3DTOP_DISABLE) {
1389 assert (key->ts[s].alphaop == D3DTOP_DISABLE);
1390 continue;
1391 }
1392 ps.stage.index = s;
1393
1394 DBG("STAGE[%u]: colorop=%s alphaop=%s\n", s,
1395 nine_D3DTOP_to_str(key->ts[s].colorop),
1396 nine_D3DTOP_to_str(key->ts[s].alphaop));
1397
1398 if (!ureg_src_is_undef(ps.s[s])) {
1399 unsigned target;
1400 struct ureg_src texture_coord = ps.vT[s];
1401 struct ureg_dst delta;
1402 switch (key->ts[s].textarget) {
1403 case 0: target = TGSI_TEXTURE_1D; break;
1404 case 1: target = TGSI_TEXTURE_2D; break;
1405 case 2: target = TGSI_TEXTURE_3D; break;
1406 case 3: target = TGSI_TEXTURE_CUBE; break;
1407 /* this is a 2 bit bitfield, do I really need a default case ? */
1408 }
1409
1410 /* Modify coordinates */
1411 if (s >= 1 &&
1412 (key->ts[s-1].colorop == D3DTOP_BUMPENVMAP ||
1413 key->ts[s-1].colorop == D3DTOP_BUMPENVMAPLUMINANCE)) {
1414 delta = ureg_DECL_temporary(ureg);
1415 /* Du' = D3DTSS_BUMPENVMAT00(stage s-1)*t(s-1)R + D3DTSS_BUMPENVMAT10(stage s-1)*t(s-1)G */
1416 ureg_MUL(ureg, ureg_writemask(delta, TGSI_WRITEMASK_X), _X(ps.rTex), _XXXX(_CONST(8 + s - 1)));
1417 ureg_MAD(ureg, ureg_writemask(delta, TGSI_WRITEMASK_X), _Y(ps.rTex), _ZZZZ(_CONST(8 + s - 1)), ureg_src(delta));
1418 /* Dv' = D3DTSS_BUMPENVMAT01(stage s-1)*t(s-1)R + D3DTSS_BUMPENVMAT11(stage s-1)*t(s-1)G */
1419 ureg_MUL(ureg, ureg_writemask(delta, TGSI_WRITEMASK_Y), _X(ps.rTex), _YYYY(_CONST(8 + s - 1)));
1420 ureg_MAD(ureg, ureg_writemask(delta, TGSI_WRITEMASK_Y), _Y(ps.rTex), _WWWW(_CONST(8 + s - 1)), ureg_src(delta));
1421 texture_coord = ureg_src(ureg_DECL_temporary(ureg));
1422 ureg_MOV(ureg, ureg_writemask(ureg_dst(texture_coord), ureg_dst(ps.vT[s]).WriteMask), ps.vT[s]);
1423 ureg_ADD(ureg, ureg_writemask(ureg_dst(texture_coord), TGSI_WRITEMASK_XY), texture_coord, ureg_src(delta));
1424 /* Prepare luminance multiplier
1425 * t(s)RGBA = t(s)RGBA * clamp[(t(s-1)B * D3DTSS_BUMPENVLSCALE(stage s-1)) + D3DTSS_BUMPENVLOFFSET(stage s-1)] */
1426 if (key->ts[s-1].colorop == D3DTOP_BUMPENVMAPLUMINANCE) {
1427 struct ureg_src bumpenvlscale = ((s-1) & 1) ? _ZZZZ(_CONST(16 + (s-1) / 2)) : _XXXX(_CONST(16 + (s-1) / 2));
1428 struct ureg_src bumpenvloffset = ((s-1) & 1) ? _WWWW(_CONST(16 + (s-1) / 2)) : _YYYY(_CONST(16 + (s-1) / 2));
1429
1430 ureg_MAD(ureg, ureg_saturate(ureg_writemask(delta, TGSI_WRITEMASK_X)), _Z(ps.rTex), bumpenvlscale, bumpenvloffset);
1431 }
1432 }
1433 if (key->projected & (3 << (s *2))) {
1434 unsigned dim = 1 + ((key->projected >> (2 * s)) & 3);
1435 if (dim == 4)
1436 ureg_TXP(ureg, ps.rTex, target, texture_coord, ps.s[s]);
1437 else {
1438 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
1439 ureg_RCP(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_X), ureg_scalar(texture_coord, dim-1));
1440 ureg_MUL(ureg, ps.rTmp, _X(tmp), texture_coord);
1441 ureg_TEX(ureg, ps.rTex, target, ps.rTmpSrc, ps.s[s]);
1442 ureg_release_temporary(ureg, tmp);
1443 }
1444 } else {
1445 ureg_TEX(ureg, ps.rTex, target, texture_coord, ps.s[s]);
1446 }
1447 if (s >= 1 && key->ts[s-1].colorop == D3DTOP_BUMPENVMAPLUMINANCE)
1448 ureg_MUL(ureg, ps.rTex, ureg_src(ps.rTex), _X(delta));
1449 }
1450
1451 if (key->ts[s].colorop == D3DTOP_BUMPENVMAP ||
1452 key->ts[s].colorop == D3DTOP_BUMPENVMAPLUMINANCE)
1453 continue;
1454
1455 dst = ps_get_ts_dst(&ps, key->ts[s].resultarg ? D3DTA_TEMP : D3DTA_CURRENT);
1456
1457 if (ps.stage.index_pre_mod == ps.stage.index) {
1458 ps.rMod = ureg_DECL_temporary(ureg);
1459 ureg_MUL(ureg, ps.rMod, ps.rCurSrc, ps.rTexSrc);
1460 }
1461
1462 colorarg[0] = (key->ts[s].colorarg0 | ((key->colorarg_b4[0] >> s) << 4) | ((key->colorarg_b5[0] >> s) << 5)) & 0x3f;
1463 colorarg[1] = (key->ts[s].colorarg1 | ((key->colorarg_b4[1] >> s) << 4) | ((key->colorarg_b5[1] >> s) << 5)) & 0x3f;
1464 colorarg[2] = (key->ts[s].colorarg2 | ((key->colorarg_b4[2] >> s) << 4) | ((key->colorarg_b5[2] >> s) << 5)) & 0x3f;
1465 alphaarg[0] = (key->ts[s].alphaarg0 | ((key->alphaarg_b4[0] >> s) << 4)) & 0x1f;
1466 alphaarg[1] = (key->ts[s].alphaarg1 | ((key->alphaarg_b4[1] >> s) << 4)) & 0x1f;
1467 alphaarg[2] = (key->ts[s].alphaarg2 | ((key->alphaarg_b4[2] >> s) << 4)) & 0x1f;
1468
1469 if (key->ts[s].colorop != key->ts[s].alphaop ||
1470 colorarg[0] != alphaarg[0] ||
1471 colorarg[1] != alphaarg[1] ||
1472 colorarg[2] != alphaarg[2])
1473 dst.WriteMask = TGSI_WRITEMASK_XYZ;
1474
1475 /* Special DOTPRODUCT behaviour (see wine tests) */
1476 if (key->ts[s].colorop == D3DTOP_DOTPRODUCT3)
1477 dst.WriteMask = TGSI_WRITEMASK_XYZW;
1478
1479 if (used_c & 0x1) arg[0] = ps_get_ts_arg(&ps, colorarg[0]);
1480 if (used_c & 0x2) arg[1] = ps_get_ts_arg(&ps, colorarg[1]);
1481 if (used_c & 0x4) arg[2] = ps_get_ts_arg(&ps, colorarg[2]);
1482 ps_do_ts_op(&ps, key->ts[s].colorop, dst, arg);
1483
1484 if (dst.WriteMask != TGSI_WRITEMASK_XYZW) {
1485 dst.WriteMask = TGSI_WRITEMASK_W;
1486
1487 if (used_a & 0x1) arg[0] = ps_get_ts_arg(&ps, alphaarg[0]);
1488 if (used_a & 0x2) arg[1] = ps_get_ts_arg(&ps, alphaarg[1]);
1489 if (used_a & 0x4) arg[2] = ps_get_ts_arg(&ps, alphaarg[2]);
1490 ps_do_ts_op(&ps, key->ts[s].alphaop, dst, arg);
1491 }
1492 }
1493
1494 if (key->specular)
1495 ureg_ADD(ureg, ureg_writemask(ps.rCur, TGSI_WRITEMASK_XYZ), ps.rCurSrc, ps.vC[1]);
1496
1497 /* Fog.
1498 */
1499 if (key->fog_mode) {
1500 struct ureg_dst rFog = ureg_writemask(ps.rTmp, TGSI_WRITEMASK_X);
1501 struct ureg_src vPos;
1502 if (device->screen->get_param(device->screen,
1503 PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL)) {
1504 vPos = ureg_DECL_system_value(ureg, TGSI_SEMANTIC_POSITION, 0);
1505 } else {
1506 vPos = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_POSITION, 0,
1507 TGSI_INTERPOLATE_LINEAR);
1508 }
1509
1510 /* Source is either W or Z.
1511 * When we use vs ff,
1512 * Z is when an orthogonal projection matrix is detected,
1513 * W (WFOG) else.
1514 * Z is used for programmable vs.
1515 * Note: Tests indicate that the projection matrix coefficients do
1516 * actually affect pixel fog (and not vertex fog) when vs ff is used,
1517 * which justifies taking the position's w instead of taking the z coordinate
1518 * before the projection in the vs shader.
1519 */
1520 if (!key->fog_source)
1521 ureg_MOV(ureg, rFog, _ZZZZ(vPos));
1522 else
1523 /* Position's w is 1/w */
1524 ureg_RCP(ureg, rFog, _WWWW(vPos));
1525
1526 if (key->fog_mode == D3DFOG_EXP) {
1527 ureg_MUL(ureg, rFog, _X(rFog), _ZZZZ(_CONST(22)));
1528 ureg_MUL(ureg, rFog, _X(rFog), ureg_imm1f(ureg, -1.442695f));
1529 ureg_EX2(ureg, rFog, _X(rFog));
1530 } else
1531 if (key->fog_mode == D3DFOG_EXP2) {
1532 ureg_MUL(ureg, rFog, _X(rFog), _ZZZZ(_CONST(22)));
1533 ureg_MUL(ureg, rFog, _X(rFog), _X(rFog));
1534 ureg_MUL(ureg, rFog, _X(rFog), ureg_imm1f(ureg, -1.442695f));
1535 ureg_EX2(ureg, rFog, _X(rFog));
1536 } else
1537 if (key->fog_mode == D3DFOG_LINEAR) {
1538 ureg_ADD(ureg, rFog, _XXXX(_CONST(22)), ureg_negate(_X(rFog)));
1539 ureg_MUL(ureg, ureg_saturate(rFog), _X(rFog), _YYYY(_CONST(22)));
1540 }
1541 ureg_LRP(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_XYZ), _X(rFog), ps.rCurSrc, _CONST(21));
1542 ureg_MOV(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_W), ps.rCurSrc);
1543 } else
1544 if (key->fog) {
1545 struct ureg_src vFog = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_FOG, 0, TGSI_INTERPOLATE_PERSPECTIVE);
1546 ureg_LRP(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_XYZ), _XXXX(vFog), ps.rCurSrc, _CONST(21));
1547 ureg_MOV(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_W), ps.rCurSrc);
1548 } else {
1549 ureg_MOV(ureg, oCol, ps.rCurSrc);
1550 }
1551
1552 ureg_END(ureg);
1553 nine_ureg_tgsi_dump(ureg, FALSE);
1554 return ureg_create_shader_and_destroy(ureg, device->context.pipe);
1555 }
1556
1557 static struct NineVertexShader9 *
1558 nine_ff_get_vs(struct NineDevice9 *device)
1559 {
1560 const struct nine_context *context = &device->context;
1561 struct NineVertexShader9 *vs;
1562 enum pipe_error err;
1563 struct vs_build_ctx bld;
1564 struct nine_ff_vs_key key;
1565 unsigned s, i;
1566 boolean has_indexes = false;
1567 boolean has_weights = false;
1568 char input_texture_coord[8];
1569
1570 assert(sizeof(key) <= sizeof(key.value32));
1571
1572 memset(&key, 0, sizeof(key));
1573 memset(&bld, 0, sizeof(bld));
1574 memset(&input_texture_coord, 0, sizeof(input_texture_coord));
1575
1576 bld.key = &key;
1577
1578 /* FIXME: this shouldn't be NULL, but it is on init */
1579 if (context->vdecl) {
1580 key.color0in_one = 1;
1581 key.color1in_zero = 1;
1582 for (i = 0; i < context->vdecl->nelems; i++) {
1583 uint16_t usage = context->vdecl->usage_map[i];
1584 if (usage == NINE_DECLUSAGE_POSITIONT)
1585 key.position_t = 1;
1586 else if (usage == NINE_DECLUSAGE_i(COLOR, 0))
1587 key.color0in_one = 0;
1588 else if (usage == NINE_DECLUSAGE_i(COLOR, 1))
1589 key.color1in_zero = 0;
1590 else if (usage == NINE_DECLUSAGE_i(BLENDINDICES, 0)) {
1591 has_indexes = true;
1592 key.passthrough |= 1 << usage;
1593 } else if (usage == NINE_DECLUSAGE_i(BLENDWEIGHT, 0)) {
1594 has_weights = true;
1595 key.passthrough |= 1 << usage;
1596 } else if (usage == NINE_DECLUSAGE_i(NORMAL, 0)) {
1597 key.has_normal = 1;
1598 key.passthrough |= 1 << usage;
1599 } else if (usage == NINE_DECLUSAGE_PSIZE)
1600 key.vertexpointsize = 1;
1601 else if (usage % NINE_DECLUSAGE_COUNT == NINE_DECLUSAGE_TEXCOORD) {
1602 s = usage / NINE_DECLUSAGE_COUNT;
1603 if (s < 8)
1604 input_texture_coord[s] = nine_decltype_get_dim(context->vdecl->decls[i].Type);
1605 else
1606 DBG("FF given texture coordinate >= 8. Ignoring\n");
1607 } else if (usage < NINE_DECLUSAGE_NONE)
1608 key.passthrough |= 1 << usage;
1609 }
1610 }
1611 /* ff vs + ps 3.0: some elements are passed to the ps (wine test).
1612 * We do restrict to indices 0 */
1613 key.passthrough &= ~((1 << NINE_DECLUSAGE_POSITION) | (1 << NINE_DECLUSAGE_PSIZE) |
1614 (1 << NINE_DECLUSAGE_TEXCOORD) | (1 << NINE_DECLUSAGE_POSITIONT) |
1615 (1 << NINE_DECLUSAGE_TESSFACTOR) | (1 << NINE_DECLUSAGE_SAMPLE));
1616 if (!key.position_t)
1617 key.passthrough = 0;
1618 key.pointscale = !!context->rs[D3DRS_POINTSCALEENABLE];
1619
1620 key.lighting = !!context->rs[D3DRS_LIGHTING] && context->ff.num_lights_active;
1621 key.darkness = !!context->rs[D3DRS_LIGHTING] && !context->ff.num_lights_active;
1622 if (key.position_t) {
1623 key.darkness = 0; /* |= key.lighting; */ /* XXX ? */
1624 key.lighting = 0;
1625 }
1626 if ((key.lighting | key.darkness) && context->rs[D3DRS_COLORVERTEX]) {
1627 uint32_t mask = (key.color0in_one ? 0 : 1) | (key.color1in_zero ? 0 : 2);
1628 key.mtl_diffuse = context->rs[D3DRS_DIFFUSEMATERIALSOURCE] & mask;
1629 key.mtl_ambient = context->rs[D3DRS_AMBIENTMATERIALSOURCE] & mask;
1630 key.mtl_specular = context->rs[D3DRS_SPECULARMATERIALSOURCE] & mask;
1631 key.mtl_emissive = context->rs[D3DRS_EMISSIVEMATERIALSOURCE] & mask;
1632 }
1633 key.fog = !!context->rs[D3DRS_FOGENABLE];
1634 key.fog_mode = (!key.position_t && context->rs[D3DRS_FOGENABLE]) ? context->rs[D3DRS_FOGVERTEXMODE] : 0;
1635 if (key.fog_mode)
1636 key.fog_range = context->rs[D3DRS_RANGEFOGENABLE];
1637
1638 key.localviewer = !!context->rs[D3DRS_LOCALVIEWER];
1639 key.normalizenormals = !!context->rs[D3DRS_NORMALIZENORMALS];
1640 key.ucp = !!context->rs[D3DRS_CLIPPLANEENABLE];
1641
1642 if (context->rs[D3DRS_VERTEXBLEND] != D3DVBF_DISABLE) {
1643 key.vertexblend_indexed = !!context->rs[D3DRS_INDEXEDVERTEXBLENDENABLE] && has_indexes;
1644
1645 switch (context->rs[D3DRS_VERTEXBLEND]) {
1646 case D3DVBF_0WEIGHTS: key.vertexblend = key.vertexblend_indexed; break;
1647 case D3DVBF_1WEIGHTS: key.vertexblend = 2; break;
1648 case D3DVBF_2WEIGHTS: key.vertexblend = 3; break;
1649 case D3DVBF_3WEIGHTS: key.vertexblend = 4; break;
1650 case D3DVBF_TWEENING: key.vertextween = 1; break;
1651 default:
1652 assert(!"invalid D3DVBF");
1653 break;
1654 }
1655 if (!has_weights && context->rs[D3DRS_VERTEXBLEND] != D3DVBF_0WEIGHTS)
1656 key.vertexblend = 0; /* TODO: if key.vertexblend_indexed, perhaps it should use 1.0 as weight, or revert to D3DVBF_0WEIGHTS */
1657 }
1658
1659 for (s = 0; s < 8; ++s) {
1660 unsigned gen = (context->ff.tex_stage[s][D3DTSS_TEXCOORDINDEX] >> 16) + 1;
1661 unsigned idx = context->ff.tex_stage[s][D3DTSS_TEXCOORDINDEX] & 7;
1662 unsigned dim;
1663
1664 if (key.position_t && gen > NINED3DTSS_TCI_PASSTHRU)
1665 gen = NINED3DTSS_TCI_PASSTHRU;
1666
1667 if (!input_texture_coord[idx] && gen == NINED3DTSS_TCI_PASSTHRU)
1668 gen = NINED3DTSS_TCI_DISABLE;
1669
1670 key.tc_gen |= gen << (s * 3);
1671 key.tc_idx |= idx << (s * 3);
1672 key.tc_dim_input |= ((input_texture_coord[idx]-1) & 0x3) << (s * 2);
1673
1674 dim = context->ff.tex_stage[s][D3DTSS_TEXTURETRANSFORMFLAGS] & 0x7;
1675 if (dim > 4)
1676 dim = input_texture_coord[idx];
1677 if (dim == 1) /* NV behaviour */
1678 dim = 0;
1679 key.tc_dim_output |= dim << (s * 3);
1680 }
1681
1682 vs = util_hash_table_get(device->ff.ht_vs, &key);
1683 if (vs)
1684 return vs;
1685 NineVertexShader9_new(device, &vs, NULL, nine_ff_build_vs(device, &bld));
1686
1687 nine_ff_prune_vs(device);
1688 if (vs) {
1689 unsigned n;
1690
1691 memcpy(&vs->ff_key, &key, sizeof(vs->ff_key));
1692
1693 err = util_hash_table_set(device->ff.ht_vs, &vs->ff_key, vs);
1694 (void)err;
1695 assert(err == PIPE_OK);
1696 device->ff.num_vs++;
1697 NineUnknown_ConvertRefToBind(NineUnknown(vs));
1698
1699 vs->num_inputs = bld.num_inputs;
1700 for (n = 0; n < bld.num_inputs; ++n)
1701 vs->input_map[n].ndecl = bld.input[n];
1702
1703 vs->position_t = key.position_t;
1704 vs->point_size = key.vertexpointsize | key.pointscale;
1705 }
1706 return vs;
1707 }
1708
1709 #define GET_D3DTS(n) nine_state_access_transform(&context->ff, D3DTS_##n, FALSE)
1710 #define IS_D3DTS_DIRTY(s,n) ((s)->ff.changed.transform[(D3DTS_##n) / 32] & (1 << ((D3DTS_##n) % 32)))
1711
1712 static struct NinePixelShader9 *
1713 nine_ff_get_ps(struct NineDevice9 *device)
1714 {
1715 struct nine_context *context = &device->context;
1716 D3DMATRIX *projection_matrix = GET_D3DTS(PROJECTION);
1717 struct NinePixelShader9 *ps;
1718 enum pipe_error err;
1719 struct nine_ff_ps_key key;
1720 unsigned s;
1721 uint8_t sampler_mask = 0;
1722
1723 assert(sizeof(key) <= sizeof(key.value32));
1724
1725 memset(&key, 0, sizeof(key));
1726 for (s = 0; s < 8; ++s) {
1727 key.ts[s].colorop = context->ff.tex_stage[s][D3DTSS_COLOROP];
1728 key.ts[s].alphaop = context->ff.tex_stage[s][D3DTSS_ALPHAOP];
1729 const uint8_t used_c = ps_d3dtop_args_mask(key.ts[s].colorop);
1730 const uint8_t used_a = ps_d3dtop_args_mask(key.ts[s].alphaop);
1731 /* MSDN says D3DTOP_DISABLE disables this and all subsequent stages.
1732 * ALPHAOP cannot be enabled if COLOROP is disabled.
1733 * Verified on Windows. */
1734 if (key.ts[s].colorop == D3DTOP_DISABLE) {
1735 key.ts[s].alphaop = D3DTOP_DISABLE; /* DISABLE == 1, avoid degenerate keys */
1736 break;
1737 }
1738
1739 if (!context->texture[s].enabled &&
1740 ((context->ff.tex_stage[s][D3DTSS_COLORARG0] == D3DTA_TEXTURE &&
1741 used_c & 0x1) ||
1742 (context->ff.tex_stage[s][D3DTSS_COLORARG1] == D3DTA_TEXTURE &&
1743 used_c & 0x2) ||
1744 (context->ff.tex_stage[s][D3DTSS_COLORARG2] == D3DTA_TEXTURE &&
1745 used_c & 0x4))) {
1746 /* Tested on Windows: Invalid texture read disables the stage
1747 * and the subsequent ones, but only for colorop. For alpha,
1748 * it's as if the texture had alpha of 1.0, which is what
1749 * has our dummy texture in that case. Invalid color also
1750 * disabled the following alpha stages. */
1751 key.ts[s].colorop = key.ts[s].alphaop = D3DTOP_DISABLE;
1752 break;
1753 }
1754
1755 if (context->ff.tex_stage[s][D3DTSS_COLORARG0] == D3DTA_TEXTURE ||
1756 context->ff.tex_stage[s][D3DTSS_COLORARG1] == D3DTA_TEXTURE ||
1757 context->ff.tex_stage[s][D3DTSS_COLORARG2] == D3DTA_TEXTURE ||
1758 context->ff.tex_stage[s][D3DTSS_ALPHAARG0] == D3DTA_TEXTURE ||
1759 context->ff.tex_stage[s][D3DTSS_ALPHAARG1] == D3DTA_TEXTURE ||
1760 context->ff.tex_stage[s][D3DTSS_ALPHAARG2] == D3DTA_TEXTURE)
1761 sampler_mask |= (1 << s);
1762
1763 if (key.ts[s].colorop != D3DTOP_DISABLE) {
1764 if (used_c & 0x1) key.ts[s].colorarg0 = context->ff.tex_stage[s][D3DTSS_COLORARG0];
1765 if (used_c & 0x2) key.ts[s].colorarg1 = context->ff.tex_stage[s][D3DTSS_COLORARG1];
1766 if (used_c & 0x4) key.ts[s].colorarg2 = context->ff.tex_stage[s][D3DTSS_COLORARG2];
1767 if (used_c & 0x1) key.colorarg_b4[0] |= (context->ff.tex_stage[s][D3DTSS_COLORARG0] >> 4) << s;
1768 if (used_c & 0x1) key.colorarg_b5[0] |= (context->ff.tex_stage[s][D3DTSS_COLORARG0] >> 5) << s;
1769 if (used_c & 0x2) key.colorarg_b4[1] |= (context->ff.tex_stage[s][D3DTSS_COLORARG1] >> 4) << s;
1770 if (used_c & 0x2) key.colorarg_b5[1] |= (context->ff.tex_stage[s][D3DTSS_COLORARG1] >> 5) << s;
1771 if (used_c & 0x4) key.colorarg_b4[2] |= (context->ff.tex_stage[s][D3DTSS_COLORARG2] >> 4) << s;
1772 if (used_c & 0x4) key.colorarg_b5[2] |= (context->ff.tex_stage[s][D3DTSS_COLORARG2] >> 5) << s;
1773 }
1774 if (key.ts[s].alphaop != D3DTOP_DISABLE) {
1775 if (used_a & 0x1) key.ts[s].alphaarg0 = context->ff.tex_stage[s][D3DTSS_ALPHAARG0];
1776 if (used_a & 0x2) key.ts[s].alphaarg1 = context->ff.tex_stage[s][D3DTSS_ALPHAARG1];
1777 if (used_a & 0x4) key.ts[s].alphaarg2 = context->ff.tex_stage[s][D3DTSS_ALPHAARG2];
1778 if (used_a & 0x1) key.alphaarg_b4[0] |= (context->ff.tex_stage[s][D3DTSS_ALPHAARG0] >> 4) << s;
1779 if (used_a & 0x2) key.alphaarg_b4[1] |= (context->ff.tex_stage[s][D3DTSS_ALPHAARG1] >> 4) << s;
1780 if (used_a & 0x4) key.alphaarg_b4[2] |= (context->ff.tex_stage[s][D3DTSS_ALPHAARG2] >> 4) << s;
1781 }
1782 key.ts[s].resultarg = context->ff.tex_stage[s][D3DTSS_RESULTARG] == D3DTA_TEMP;
1783
1784 if (context->texture[s].enabled) {
1785 switch (context->texture[s].type) {
1786 case D3DRTYPE_TEXTURE: key.ts[s].textarget = 1; break;
1787 case D3DRTYPE_VOLUMETEXTURE: key.ts[s].textarget = 2; break;
1788 case D3DRTYPE_CUBETEXTURE: key.ts[s].textarget = 3; break;
1789 default:
1790 assert(!"unexpected texture type");
1791 break;
1792 }
1793 } else {
1794 key.ts[s].textarget = 1;
1795 }
1796 }
1797
1798 /* Note: If colorop is D3DTOP_DISABLE for the first stage
1799 * (which implies alphaop is too), nothing particular happens,
1800 * that is, current is equal to diffuse (which is the case anyway,
1801 * because it is how it is initialized).
1802 * Special case seems if alphaop is D3DTOP_DISABLE and not colorop,
1803 * because then if the resultarg is TEMP, then diffuse alpha is written
1804 * to it. */
1805 if (key.ts[0].colorop != D3DTOP_DISABLE &&
1806 key.ts[0].alphaop == D3DTOP_DISABLE &&
1807 key.ts[0].resultarg != 0) {
1808 key.ts[0].alphaop = D3DTOP_SELECTARG1;
1809 key.ts[0].alphaarg1 = D3DTA_DIFFUSE;
1810 }
1811 /* When no alpha stage writes to current, diffuse alpha is taken.
1812 * Since we initialize current to diffuse, we have the behaviour. */
1813
1814 /* Last stage always writes to Current */
1815 if (s >= 1)
1816 key.ts[s-1].resultarg = 0;
1817
1818 key.projected = nine_ff_get_projected_key(context);
1819 key.specular = !!context->rs[D3DRS_SPECULARENABLE];
1820
1821 for (; s < 8; ++s)
1822 key.ts[s].colorop = key.ts[s].alphaop = D3DTOP_DISABLE;
1823 if (context->rs[D3DRS_FOGENABLE])
1824 key.fog_mode = context->rs[D3DRS_FOGTABLEMODE];
1825 key.fog = !!context->rs[D3DRS_FOGENABLE];
1826 /* Pixel fog (with WFOG advertised): source is either Z or W.
1827 * W is the source if vs ff is used, and the
1828 * projection matrix is not orthogonal.
1829 * Tests on Win 10 seem to indicate _34
1830 * and _33 are checked against 0, 1. */
1831 if (key.fog_mode && key.fog)
1832 key.fog_source = !context->programmable_vs &&
1833 !(projection_matrix->_34 == 0.0f &&
1834 projection_matrix->_44 == 1.0f);
1835
1836 ps = util_hash_table_get(device->ff.ht_ps, &key);
1837 if (ps)
1838 return ps;
1839 NinePixelShader9_new(device, &ps, NULL, nine_ff_build_ps(device, &key));
1840
1841 nine_ff_prune_ps(device);
1842 if (ps) {
1843 memcpy(&ps->ff_key, &key, sizeof(ps->ff_key));
1844
1845 err = util_hash_table_set(device->ff.ht_ps, &ps->ff_key, ps);
1846 (void)err;
1847 assert(err == PIPE_OK);
1848 device->ff.num_ps++;
1849 NineUnknown_ConvertRefToBind(NineUnknown(ps));
1850
1851 ps->rt_mask = 0x1;
1852 ps->sampler_mask = sampler_mask;
1853 }
1854 return ps;
1855 }
1856
1857 static void
1858 nine_ff_load_vs_transforms(struct NineDevice9 *device)
1859 {
1860 struct nine_context *context = &device->context;
1861 D3DMATRIX T;
1862 D3DMATRIX *M = (D3DMATRIX *)device->ff.vs_const;
1863 unsigned i;
1864
1865 /* TODO: make this nicer, and only upload the ones we need */
1866 /* TODO: use ff.vs_const as storage of W, V, P matrices */
1867
1868 if (IS_D3DTS_DIRTY(context, WORLD) ||
1869 IS_D3DTS_DIRTY(context, VIEW) ||
1870 IS_D3DTS_DIRTY(context, PROJECTION)) {
1871 /* WVP, WV matrices */
1872 nine_d3d_matrix_matrix_mul(&M[1], GET_D3DTS(WORLD), GET_D3DTS(VIEW));
1873 nine_d3d_matrix_matrix_mul(&M[0], &M[1], GET_D3DTS(PROJECTION));
1874
1875 /* normal matrix == transpose(inverse(WV)) */
1876 nine_d3d_matrix_inverse(&T, &M[1]);
1877 nine_d3d_matrix_transpose(&M[4], &T);
1878
1879 /* P matrix */
1880 M[2] = *GET_D3DTS(PROJECTION);
1881
1882 /* V and W matrix */
1883 nine_d3d_matrix_inverse(&M[3], GET_D3DTS(VIEW));
1884 M[40] = M[1];
1885 }
1886
1887 if (context->rs[D3DRS_VERTEXBLEND] != D3DVBF_DISABLE) {
1888 /* load other world matrices */
1889 for (i = 1; i <= 8; ++i) {
1890 nine_d3d_matrix_matrix_mul(&M[40 + i], GET_D3DTS(WORLDMATRIX(i)), GET_D3DTS(VIEW));
1891 }
1892 }
1893
1894 device->ff.vs_const[30 * 4] = asfloat(context->rs[D3DRS_TWEENFACTOR]);
1895 }
1896
1897 static void
1898 nine_ff_load_lights(struct NineDevice9 *device)
1899 {
1900 struct nine_context *context = &device->context;
1901 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1902 unsigned l;
1903
1904 if (context->changed.group & NINE_STATE_FF_MATERIAL) {
1905 const D3DMATERIAL9 *mtl = &context->ff.material;
1906
1907 memcpy(&dst[20], &mtl->Diffuse, 4 * sizeof(float));
1908 memcpy(&dst[21], &mtl->Ambient, 4 * sizeof(float));
1909 memcpy(&dst[22], &mtl->Specular, 4 * sizeof(float));
1910 dst[23].x = mtl->Power;
1911 memcpy(&dst[24], &mtl->Emissive, 4 * sizeof(float));
1912 d3dcolor_to_rgba(&dst[25].x, context->rs[D3DRS_AMBIENT]);
1913 dst[19].x = dst[25].x * mtl->Ambient.r + mtl->Emissive.r;
1914 dst[19].y = dst[25].y * mtl->Ambient.g + mtl->Emissive.g;
1915 dst[19].z = dst[25].z * mtl->Ambient.b + mtl->Emissive.b;
1916 }
1917
1918 if (!(context->changed.group & NINE_STATE_FF_LIGHTING))
1919 return;
1920
1921 for (l = 0; l < context->ff.num_lights_active; ++l) {
1922 const D3DLIGHT9 *light = &context->ff.light[context->ff.active_light[l]];
1923
1924 dst[32 + l * 8].x = light->Type;
1925 dst[32 + l * 8].y = light->Attenuation0;
1926 dst[32 + l * 8].z = light->Attenuation1;
1927 dst[32 + l * 8].w = light->Attenuation2;
1928 memcpy(&dst[33 + l * 8].x, &light->Diffuse, sizeof(light->Diffuse));
1929 memcpy(&dst[34 + l * 8].x, &light->Specular, sizeof(light->Specular));
1930 memcpy(&dst[35 + l * 8].x, &light->Ambient, sizeof(light->Ambient));
1931 nine_d3d_vector4_matrix_mul((D3DVECTOR *)&dst[36 + l * 8].x, &light->Position, GET_D3DTS(VIEW));
1932 nine_d3d_vector3_matrix_mul((D3DVECTOR *)&dst[37 + l * 8].x, &light->Direction, GET_D3DTS(VIEW));
1933 dst[36 + l * 8].w = light->Type == D3DLIGHT_DIRECTIONAL ? 1e9f : light->Range;
1934 dst[37 + l * 8].w = light->Falloff;
1935 dst[38 + l * 8].x = cosf(light->Theta * 0.5f);
1936 dst[38 + l * 8].y = cosf(light->Phi * 0.5f);
1937 dst[38 + l * 8].z = 1.0f / (dst[38 + l * 8].x - dst[38 + l * 8].y);
1938 dst[39 + l * 8].w = (l + 1) == context->ff.num_lights_active;
1939 }
1940 }
1941
1942 static void
1943 nine_ff_load_point_and_fog_params(struct NineDevice9 *device)
1944 {
1945 struct nine_context *context = &device->context;
1946 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1947
1948 if (!(context->changed.group & NINE_STATE_FF_OTHER))
1949 return;
1950 dst[26].x = asfloat(context->rs[D3DRS_POINTSIZE_MIN]);
1951 dst[26].y = asfloat(context->rs[D3DRS_POINTSIZE_MAX]);
1952 dst[26].z = asfloat(context->rs[D3DRS_POINTSIZE]);
1953 dst[26].w = asfloat(context->rs[D3DRS_POINTSCALE_A]);
1954 dst[27].x = asfloat(context->rs[D3DRS_POINTSCALE_B]);
1955 dst[27].y = asfloat(context->rs[D3DRS_POINTSCALE_C]);
1956 dst[28].x = asfloat(context->rs[D3DRS_FOGEND]);
1957 dst[28].y = 1.0f / (asfloat(context->rs[D3DRS_FOGEND]) - asfloat(context->rs[D3DRS_FOGSTART]));
1958 if (isinf(dst[28].y))
1959 dst[28].y = 0.0f;
1960 dst[28].z = asfloat(context->rs[D3DRS_FOGDENSITY]);
1961 }
1962
1963 static void
1964 nine_ff_load_tex_matrices(struct NineDevice9 *device)
1965 {
1966 struct nine_context *context = &device->context;
1967 D3DMATRIX *M = (D3DMATRIX *)device->ff.vs_const;
1968 unsigned s;
1969
1970 if (!(context->ff.changed.transform[0] & 0xff0000))
1971 return;
1972 for (s = 0; s < 8; ++s) {
1973 if (IS_D3DTS_DIRTY(context, TEXTURE0 + s))
1974 nine_d3d_matrix_transpose(&M[32 + s], nine_state_access_transform(&context->ff, D3DTS_TEXTURE0 + s, FALSE));
1975 }
1976 }
1977
1978 static void
1979 nine_ff_load_ps_params(struct NineDevice9 *device)
1980 {
1981 struct nine_context *context = &device->context;
1982 struct fvec4 *dst = (struct fvec4 *)device->ff.ps_const;
1983 unsigned s;
1984
1985 if (!(context->changed.group & (NINE_STATE_FF_PSSTAGES | NINE_STATE_FF_OTHER)))
1986 return;
1987
1988 for (s = 0; s < 8; ++s)
1989 d3dcolor_to_rgba(&dst[s].x, context->ff.tex_stage[s][D3DTSS_CONSTANT]);
1990
1991 for (s = 0; s < 8; ++s) {
1992 dst[8 + s].x = asfloat(context->ff.tex_stage[s][D3DTSS_BUMPENVMAT00]);
1993 dst[8 + s].y = asfloat(context->ff.tex_stage[s][D3DTSS_BUMPENVMAT01]);
1994 dst[8 + s].z = asfloat(context->ff.tex_stage[s][D3DTSS_BUMPENVMAT10]);
1995 dst[8 + s].w = asfloat(context->ff.tex_stage[s][D3DTSS_BUMPENVMAT11]);
1996 if (s & 1) {
1997 dst[16 + s / 2].z = asfloat(context->ff.tex_stage[s][D3DTSS_BUMPENVLSCALE]);
1998 dst[16 + s / 2].w = asfloat(context->ff.tex_stage[s][D3DTSS_BUMPENVLOFFSET]);
1999 } else {
2000 dst[16 + s / 2].x = asfloat(context->ff.tex_stage[s][D3DTSS_BUMPENVLSCALE]);
2001 dst[16 + s / 2].y = asfloat(context->ff.tex_stage[s][D3DTSS_BUMPENVLOFFSET]);
2002 }
2003 }
2004
2005 d3dcolor_to_rgba(&dst[20].x, context->rs[D3DRS_TEXTUREFACTOR]);
2006 d3dcolor_to_rgba(&dst[21].x, context->rs[D3DRS_FOGCOLOR]);
2007 dst[22].x = asfloat(context->rs[D3DRS_FOGEND]);
2008 dst[22].y = 1.0f / (asfloat(context->rs[D3DRS_FOGEND]) - asfloat(context->rs[D3DRS_FOGSTART]));
2009 dst[22].z = asfloat(context->rs[D3DRS_FOGDENSITY]);
2010 }
2011
2012 static void
2013 nine_ff_load_viewport_info(struct NineDevice9 *device)
2014 {
2015 D3DVIEWPORT9 *viewport = &device->context.viewport;
2016 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
2017 float diffZ = viewport->MaxZ - viewport->MinZ;
2018
2019 /* Note: the other functions avoids to fill the const again if nothing changed.
2020 * But we don't have much to fill, and adding code to allow that may be complex
2021 * so just fill it always */
2022 dst[100].x = 2.0f / (float)(viewport->Width);
2023 dst[100].y = 2.0f / (float)(viewport->Height);
2024 dst[100].z = (diffZ == 0.0f) ? 0.0f : (1.0f / diffZ);
2025 dst[100].w = (float)(viewport->Width);
2026 dst[101].x = (float)(viewport->X);
2027 dst[101].y = (float)(viewport->Y);
2028 dst[101].z = (float)(viewport->MinZ);
2029 }
2030
2031 void
2032 nine_ff_update(struct NineDevice9 *device)
2033 {
2034 struct nine_context *context = &device->context;
2035 struct pipe_constant_buffer cb;
2036
2037 DBG("vs=%p ps=%p\n", context->vs, context->ps);
2038
2039 /* NOTE: the only reference belongs to the hash table */
2040 if (!context->programmable_vs) {
2041 device->ff.vs = nine_ff_get_vs(device);
2042 context->changed.group |= NINE_STATE_VS;
2043 }
2044 if (!context->ps) {
2045 device->ff.ps = nine_ff_get_ps(device);
2046 context->changed.group |= NINE_STATE_PS;
2047 }
2048
2049 if (!context->programmable_vs) {
2050 nine_ff_load_vs_transforms(device);
2051 nine_ff_load_tex_matrices(device);
2052 nine_ff_load_lights(device);
2053 nine_ff_load_point_and_fog_params(device);
2054 nine_ff_load_viewport_info(device);
2055
2056 memset(context->ff.changed.transform, 0, sizeof(context->ff.changed.transform));
2057
2058 cb.buffer_offset = 0;
2059 cb.buffer = NULL;
2060 cb.user_buffer = device->ff.vs_const;
2061 cb.buffer_size = NINE_FF_NUM_VS_CONST * 4 * sizeof(float);
2062
2063 if (!device->driver_caps.user_cbufs) {
2064 context->pipe_data.cb_vs_ff.buffer_size = cb.buffer_size;
2065 u_upload_data(device->context.pipe->const_uploader,
2066 0,
2067 cb.buffer_size,
2068 device->constbuf_alignment,
2069 cb.user_buffer,
2070 &context->pipe_data.cb_vs_ff.buffer_offset,
2071 &context->pipe_data.cb_vs_ff.buffer);
2072 u_upload_unmap(device->context.pipe->const_uploader);
2073 context->pipe_data.cb_vs_ff.user_buffer = NULL;
2074 } else
2075 context->pipe_data.cb_vs_ff = cb;
2076 context->commit |= NINE_STATE_COMMIT_CONST_VS;
2077 }
2078
2079 if (!context->ps) {
2080 nine_ff_load_ps_params(device);
2081
2082 cb.buffer_offset = 0;
2083 cb.buffer = NULL;
2084 cb.user_buffer = device->ff.ps_const;
2085 cb.buffer_size = NINE_FF_NUM_PS_CONST * 4 * sizeof(float);
2086
2087 if (!device->driver_caps.user_cbufs) {
2088 context->pipe_data.cb_ps_ff.buffer_size = cb.buffer_size;
2089 u_upload_data(device->context.pipe->const_uploader,
2090 0,
2091 cb.buffer_size,
2092 device->constbuf_alignment,
2093 cb.user_buffer,
2094 &context->pipe_data.cb_ps_ff.buffer_offset,
2095 &context->pipe_data.cb_ps_ff.buffer);
2096 u_upload_unmap(device->context.pipe->const_uploader);
2097 context->pipe_data.cb_ps_ff.user_buffer = NULL;
2098 } else
2099 context->pipe_data.cb_ps_ff = cb;
2100 context->commit |= NINE_STATE_COMMIT_CONST_PS;
2101 }
2102
2103 context->changed.group &= ~NINE_STATE_FF;
2104 }
2105
2106
2107 boolean
2108 nine_ff_init(struct NineDevice9 *device)
2109 {
2110 device->ff.ht_vs = util_hash_table_create(nine_ff_vs_key_hash,
2111 nine_ff_vs_key_comp);
2112 device->ff.ht_ps = util_hash_table_create(nine_ff_ps_key_hash,
2113 nine_ff_ps_key_comp);
2114
2115 device->ff.ht_fvf = util_hash_table_create(nine_ff_fvf_key_hash,
2116 nine_ff_fvf_key_comp);
2117
2118 device->ff.vs_const = CALLOC(NINE_FF_NUM_VS_CONST, 4 * sizeof(float));
2119 device->ff.ps_const = CALLOC(NINE_FF_NUM_PS_CONST, 4 * sizeof(float));
2120
2121 return device->ff.ht_vs && device->ff.ht_ps &&
2122 device->ff.ht_fvf &&
2123 device->ff.vs_const && device->ff.ps_const;
2124 }
2125
2126 static enum pipe_error nine_ff_ht_delete_cb(void *key, void *value, void *data)
2127 {
2128 NineUnknown_Unbind(NineUnknown(value));
2129 return PIPE_OK;
2130 }
2131
2132 void
2133 nine_ff_fini(struct NineDevice9 *device)
2134 {
2135 if (device->ff.ht_vs) {
2136 util_hash_table_foreach(device->ff.ht_vs, nine_ff_ht_delete_cb, NULL);
2137 util_hash_table_destroy(device->ff.ht_vs);
2138 }
2139 if (device->ff.ht_ps) {
2140 util_hash_table_foreach(device->ff.ht_ps, nine_ff_ht_delete_cb, NULL);
2141 util_hash_table_destroy(device->ff.ht_ps);
2142 }
2143 if (device->ff.ht_fvf) {
2144 util_hash_table_foreach(device->ff.ht_fvf, nine_ff_ht_delete_cb, NULL);
2145 util_hash_table_destroy(device->ff.ht_fvf);
2146 }
2147 device->ff.vs = NULL; /* destroyed by unbinding from hash table */
2148 device->ff.ps = NULL;
2149
2150 FREE(device->ff.vs_const);
2151 FREE(device->ff.ps_const);
2152 }
2153
2154 static void
2155 nine_ff_prune_vs(struct NineDevice9 *device)
2156 {
2157 struct nine_context *context = &device->context;
2158
2159 if (device->ff.num_vs > 100) {
2160 /* could destroy the bound one here, so unbind */
2161 context->pipe->bind_vs_state(context->pipe, NULL);
2162 util_hash_table_foreach(device->ff.ht_vs, nine_ff_ht_delete_cb, NULL);
2163 util_hash_table_clear(device->ff.ht_vs);
2164 device->ff.num_vs = 0;
2165 context->changed.group |= NINE_STATE_VS;
2166 }
2167 }
2168 static void
2169 nine_ff_prune_ps(struct NineDevice9 *device)
2170 {
2171 struct nine_context *context = &device->context;
2172
2173 if (device->ff.num_ps > 100) {
2174 /* could destroy the bound one here, so unbind */
2175 context->pipe->bind_fs_state(context->pipe, NULL);
2176 util_hash_table_foreach(device->ff.ht_ps, nine_ff_ht_delete_cb, NULL);
2177 util_hash_table_clear(device->ff.ht_ps);
2178 device->ff.num_ps = 0;
2179 context->changed.group |= NINE_STATE_PS;
2180 }
2181 }
2182
2183 /* ========================================================================== */
2184
2185 /* Matrix multiplication:
2186 *
2187 * in memory: 0 1 2 3 (row major)
2188 * 4 5 6 7
2189 * 8 9 a b
2190 * c d e f
2191 *
2192 * cA cB cC cD
2193 * r0 = (r0 * cA) (r0 * cB) . .
2194 * r1 = (r1 * cA) (r1 * cB)
2195 * r2 = (r2 * cA) .
2196 * r3 = (r3 * cA) .
2197 *
2198 * r: (11) (12) (13) (14)
2199 * (21) (22) (23) (24)
2200 * (31) (32) (33) (34)
2201 * (41) (42) (43) (44)
2202 * l: (11 12 13 14)
2203 * (21 22 23 24)
2204 * (31 32 33 34)
2205 * (41 42 43 44)
2206 *
2207 * v: (x y z 1 )
2208 *
2209 * t.xyzw = MUL(v.xxxx, r[0]);
2210 * t.xyzw = MAD(v.yyyy, r[1], t.xyzw);
2211 * t.xyzw = MAD(v.zzzz, r[2], t.xyzw);
2212 * v.xyzw = MAD(v.wwww, r[3], t.xyzw);
2213 *
2214 * v.x = DP4(v, c[0]);
2215 * v.y = DP4(v, c[1]);
2216 * v.z = DP4(v, c[2]);
2217 * v.w = DP4(v, c[3]) = 1
2218 */
2219
2220 /*
2221 static void
2222 nine_D3DMATRIX_print(const D3DMATRIX *M)
2223 {
2224 DBG("\n(%f %f %f %f)\n"
2225 "(%f %f %f %f)\n"
2226 "(%f %f %f %f)\n"
2227 "(%f %f %f %f)\n",
2228 M->m[0][0], M->m[0][1], M->m[0][2], M->m[0][3],
2229 M->m[1][0], M->m[1][1], M->m[1][2], M->m[1][3],
2230 M->m[2][0], M->m[2][1], M->m[2][2], M->m[2][3],
2231 M->m[3][0], M->m[3][1], M->m[3][2], M->m[3][3]);
2232 }
2233 */
2234
2235 static inline float
2236 nine_DP4_row_col(const D3DMATRIX *A, int r, const D3DMATRIX *B, int c)
2237 {
2238 return A->m[r][0] * B->m[0][c] +
2239 A->m[r][1] * B->m[1][c] +
2240 A->m[r][2] * B->m[2][c] +
2241 A->m[r][3] * B->m[3][c];
2242 }
2243
2244 static inline float
2245 nine_DP4_vec_col(const D3DVECTOR *v, const D3DMATRIX *M, int c)
2246 {
2247 return v->x * M->m[0][c] +
2248 v->y * M->m[1][c] +
2249 v->z * M->m[2][c] +
2250 1.0f * M->m[3][c];
2251 }
2252
2253 static inline float
2254 nine_DP3_vec_col(const D3DVECTOR *v, const D3DMATRIX *M, int c)
2255 {
2256 return v->x * M->m[0][c] +
2257 v->y * M->m[1][c] +
2258 v->z * M->m[2][c];
2259 }
2260
2261 void
2262 nine_d3d_matrix_matrix_mul(D3DMATRIX *D, const D3DMATRIX *L, const D3DMATRIX *R)
2263 {
2264 D->_11 = nine_DP4_row_col(L, 0, R, 0);
2265 D->_12 = nine_DP4_row_col(L, 0, R, 1);
2266 D->_13 = nine_DP4_row_col(L, 0, R, 2);
2267 D->_14 = nine_DP4_row_col(L, 0, R, 3);
2268
2269 D->_21 = nine_DP4_row_col(L, 1, R, 0);
2270 D->_22 = nine_DP4_row_col(L, 1, R, 1);
2271 D->_23 = nine_DP4_row_col(L, 1, R, 2);
2272 D->_24 = nine_DP4_row_col(L, 1, R, 3);
2273
2274 D->_31 = nine_DP4_row_col(L, 2, R, 0);
2275 D->_32 = nine_DP4_row_col(L, 2, R, 1);
2276 D->_33 = nine_DP4_row_col(L, 2, R, 2);
2277 D->_34 = nine_DP4_row_col(L, 2, R, 3);
2278
2279 D->_41 = nine_DP4_row_col(L, 3, R, 0);
2280 D->_42 = nine_DP4_row_col(L, 3, R, 1);
2281 D->_43 = nine_DP4_row_col(L, 3, R, 2);
2282 D->_44 = nine_DP4_row_col(L, 3, R, 3);
2283 }
2284
2285 void
2286 nine_d3d_vector4_matrix_mul(D3DVECTOR *d, const D3DVECTOR *v, const D3DMATRIX *M)
2287 {
2288 d->x = nine_DP4_vec_col(v, M, 0);
2289 d->y = nine_DP4_vec_col(v, M, 1);
2290 d->z = nine_DP4_vec_col(v, M, 2);
2291 }
2292
2293 void
2294 nine_d3d_vector3_matrix_mul(D3DVECTOR *d, const D3DVECTOR *v, const D3DMATRIX *M)
2295 {
2296 d->x = nine_DP3_vec_col(v, M, 0);
2297 d->y = nine_DP3_vec_col(v, M, 1);
2298 d->z = nine_DP3_vec_col(v, M, 2);
2299 }
2300
2301 void
2302 nine_d3d_matrix_transpose(D3DMATRIX *D, const D3DMATRIX *M)
2303 {
2304 unsigned i, j;
2305 for (i = 0; i < 4; ++i)
2306 for (j = 0; j < 4; ++j)
2307 D->m[i][j] = M->m[j][i];
2308 }
2309
2310 #define _M_ADD_PROD_1i_2j_3k_4l(i,j,k,l) do { \
2311 float t = M->_1##i * M->_2##j * M->_3##k * M->_4##l; \
2312 if (t > 0.0f) pos += t; else neg += t; } while(0)
2313
2314 #define _M_SUB_PROD_1i_2j_3k_4l(i,j,k,l) do { \
2315 float t = M->_1##i * M->_2##j * M->_3##k * M->_4##l; \
2316 if (t > 0.0f) neg -= t; else pos -= t; } while(0)
2317 float
2318 nine_d3d_matrix_det(const D3DMATRIX *M)
2319 {
2320 float pos = 0.0f;
2321 float neg = 0.0f;
2322
2323 _M_ADD_PROD_1i_2j_3k_4l(1, 2, 3, 4);
2324 _M_ADD_PROD_1i_2j_3k_4l(1, 3, 4, 2);
2325 _M_ADD_PROD_1i_2j_3k_4l(1, 4, 2, 3);
2326
2327 _M_ADD_PROD_1i_2j_3k_4l(2, 1, 4, 3);
2328 _M_ADD_PROD_1i_2j_3k_4l(2, 3, 1, 4);
2329 _M_ADD_PROD_1i_2j_3k_4l(2, 4, 3, 1);
2330
2331 _M_ADD_PROD_1i_2j_3k_4l(3, 1, 2, 4);
2332 _M_ADD_PROD_1i_2j_3k_4l(3, 2, 4, 1);
2333 _M_ADD_PROD_1i_2j_3k_4l(3, 4, 1, 2);
2334
2335 _M_ADD_PROD_1i_2j_3k_4l(4, 1, 3, 2);
2336 _M_ADD_PROD_1i_2j_3k_4l(4, 2, 1, 3);
2337 _M_ADD_PROD_1i_2j_3k_4l(4, 3, 2, 1);
2338
2339 _M_SUB_PROD_1i_2j_3k_4l(1, 2, 4, 3);
2340 _M_SUB_PROD_1i_2j_3k_4l(1, 3, 2, 4);
2341 _M_SUB_PROD_1i_2j_3k_4l(1, 4, 3, 2);
2342
2343 _M_SUB_PROD_1i_2j_3k_4l(2, 1, 3, 4);
2344 _M_SUB_PROD_1i_2j_3k_4l(2, 3, 4, 1);
2345 _M_SUB_PROD_1i_2j_3k_4l(2, 4, 1, 3);
2346
2347 _M_SUB_PROD_1i_2j_3k_4l(3, 1, 4, 2);
2348 _M_SUB_PROD_1i_2j_3k_4l(3, 2, 1, 4);
2349 _M_SUB_PROD_1i_2j_3k_4l(3, 4, 2, 1);
2350
2351 _M_SUB_PROD_1i_2j_3k_4l(4, 1, 2, 3);
2352 _M_SUB_PROD_1i_2j_3k_4l(4, 2, 3, 1);
2353 _M_SUB_PROD_1i_2j_3k_4l(4, 3, 1, 2);
2354
2355 return pos + neg;
2356 }
2357
2358 /* XXX: Probably better to just use src/mesa/math/m_matrix.c because
2359 * I have no idea where this code came from.
2360 */
2361 void
2362 nine_d3d_matrix_inverse(D3DMATRIX *D, const D3DMATRIX *M)
2363 {
2364 int i, k;
2365 float det;
2366
2367 D->m[0][0] =
2368 M->m[1][1] * M->m[2][2] * M->m[3][3] -
2369 M->m[1][1] * M->m[3][2] * M->m[2][3] -
2370 M->m[1][2] * M->m[2][1] * M->m[3][3] +
2371 M->m[1][2] * M->m[3][1] * M->m[2][3] +
2372 M->m[1][3] * M->m[2][1] * M->m[3][2] -
2373 M->m[1][3] * M->m[3][1] * M->m[2][2];
2374
2375 D->m[0][1] =
2376 -M->m[0][1] * M->m[2][2] * M->m[3][3] +
2377 M->m[0][1] * M->m[3][2] * M->m[2][3] +
2378 M->m[0][2] * M->m[2][1] * M->m[3][3] -
2379 M->m[0][2] * M->m[3][1] * M->m[2][3] -
2380 M->m[0][3] * M->m[2][1] * M->m[3][2] +
2381 M->m[0][3] * M->m[3][1] * M->m[2][2];
2382
2383 D->m[0][2] =
2384 M->m[0][1] * M->m[1][2] * M->m[3][3] -
2385 M->m[0][1] * M->m[3][2] * M->m[1][3] -
2386 M->m[0][2] * M->m[1][1] * M->m[3][3] +
2387 M->m[0][2] * M->m[3][1] * M->m[1][3] +
2388 M->m[0][3] * M->m[1][1] * M->m[3][2] -
2389 M->m[0][3] * M->m[3][1] * M->m[1][2];
2390
2391 D->m[0][3] =
2392 -M->m[0][1] * M->m[1][2] * M->m[2][3] +
2393 M->m[0][1] * M->m[2][2] * M->m[1][3] +
2394 M->m[0][2] * M->m[1][1] * M->m[2][3] -
2395 M->m[0][2] * M->m[2][1] * M->m[1][3] -
2396 M->m[0][3] * M->m[1][1] * M->m[2][2] +
2397 M->m[0][3] * M->m[2][1] * M->m[1][2];
2398
2399 D->m[1][0] =
2400 -M->m[1][0] * M->m[2][2] * M->m[3][3] +
2401 M->m[1][0] * M->m[3][2] * M->m[2][3] +
2402 M->m[1][2] * M->m[2][0] * M->m[3][3] -
2403 M->m[1][2] * M->m[3][0] * M->m[2][3] -
2404 M->m[1][3] * M->m[2][0] * M->m[3][2] +
2405 M->m[1][3] * M->m[3][0] * M->m[2][2];
2406
2407 D->m[1][1] =
2408 M->m[0][0] * M->m[2][2] * M->m[3][3] -
2409 M->m[0][0] * M->m[3][2] * M->m[2][3] -
2410 M->m[0][2] * M->m[2][0] * M->m[3][3] +
2411 M->m[0][2] * M->m[3][0] * M->m[2][3] +
2412 M->m[0][3] * M->m[2][0] * M->m[3][2] -
2413 M->m[0][3] * M->m[3][0] * M->m[2][2];
2414
2415 D->m[1][2] =
2416 -M->m[0][0] * M->m[1][2] * M->m[3][3] +
2417 M->m[0][0] * M->m[3][2] * M->m[1][3] +
2418 M->m[0][2] * M->m[1][0] * M->m[3][3] -
2419 M->m[0][2] * M->m[3][0] * M->m[1][3] -
2420 M->m[0][3] * M->m[1][0] * M->m[3][2] +
2421 M->m[0][3] * M->m[3][0] * M->m[1][2];
2422
2423 D->m[1][3] =
2424 M->m[0][0] * M->m[1][2] * M->m[2][3] -
2425 M->m[0][0] * M->m[2][2] * M->m[1][3] -
2426 M->m[0][2] * M->m[1][0] * M->m[2][3] +
2427 M->m[0][2] * M->m[2][0] * M->m[1][3] +
2428 M->m[0][3] * M->m[1][0] * M->m[2][2] -
2429 M->m[0][3] * M->m[2][0] * M->m[1][2];
2430
2431 D->m[2][0] =
2432 M->m[1][0] * M->m[2][1] * M->m[3][3] -
2433 M->m[1][0] * M->m[3][1] * M->m[2][3] -
2434 M->m[1][1] * M->m[2][0] * M->m[3][3] +
2435 M->m[1][1] * M->m[3][0] * M->m[2][3] +
2436 M->m[1][3] * M->m[2][0] * M->m[3][1] -
2437 M->m[1][3] * M->m[3][0] * M->m[2][1];
2438
2439 D->m[2][1] =
2440 -M->m[0][0] * M->m[2][1] * M->m[3][3] +
2441 M->m[0][0] * M->m[3][1] * M->m[2][3] +
2442 M->m[0][1] * M->m[2][0] * M->m[3][3] -
2443 M->m[0][1] * M->m[3][0] * M->m[2][3] -
2444 M->m[0][3] * M->m[2][0] * M->m[3][1] +
2445 M->m[0][3] * M->m[3][0] * M->m[2][1];
2446
2447 D->m[2][2] =
2448 M->m[0][0] * M->m[1][1] * M->m[3][3] -
2449 M->m[0][0] * M->m[3][1] * M->m[1][3] -
2450 M->m[0][1] * M->m[1][0] * M->m[3][3] +
2451 M->m[0][1] * M->m[3][0] * M->m[1][3] +
2452 M->m[0][3] * M->m[1][0] * M->m[3][1] -
2453 M->m[0][3] * M->m[3][0] * M->m[1][1];
2454
2455 D->m[2][3] =
2456 -M->m[0][0] * M->m[1][1] * M->m[2][3] +
2457 M->m[0][0] * M->m[2][1] * M->m[1][3] +
2458 M->m[0][1] * M->m[1][0] * M->m[2][3] -
2459 M->m[0][1] * M->m[2][0] * M->m[1][3] -
2460 M->m[0][3] * M->m[1][0] * M->m[2][1] +
2461 M->m[0][3] * M->m[2][0] * M->m[1][1];
2462
2463 D->m[3][0] =
2464 -M->m[1][0] * M->m[2][1] * M->m[3][2] +
2465 M->m[1][0] * M->m[3][1] * M->m[2][2] +
2466 M->m[1][1] * M->m[2][0] * M->m[3][2] -
2467 M->m[1][1] * M->m[3][0] * M->m[2][2] -
2468 M->m[1][2] * M->m[2][0] * M->m[3][1] +
2469 M->m[1][2] * M->m[3][0] * M->m[2][1];
2470
2471 D->m[3][1] =
2472 M->m[0][0] * M->m[2][1] * M->m[3][2] -
2473 M->m[0][0] * M->m[3][1] * M->m[2][2] -
2474 M->m[0][1] * M->m[2][0] * M->m[3][2] +
2475 M->m[0][1] * M->m[3][0] * M->m[2][2] +
2476 M->m[0][2] * M->m[2][0] * M->m[3][1] -
2477 M->m[0][2] * M->m[3][0] * M->m[2][1];
2478
2479 D->m[3][2] =
2480 -M->m[0][0] * M->m[1][1] * M->m[3][2] +
2481 M->m[0][0] * M->m[3][1] * M->m[1][2] +
2482 M->m[0][1] * M->m[1][0] * M->m[3][2] -
2483 M->m[0][1] * M->m[3][0] * M->m[1][2] -
2484 M->m[0][2] * M->m[1][0] * M->m[3][1] +
2485 M->m[0][2] * M->m[3][0] * M->m[1][1];
2486
2487 D->m[3][3] =
2488 M->m[0][0] * M->m[1][1] * M->m[2][2] -
2489 M->m[0][0] * M->m[2][1] * M->m[1][2] -
2490 M->m[0][1] * M->m[1][0] * M->m[2][2] +
2491 M->m[0][1] * M->m[2][0] * M->m[1][2] +
2492 M->m[0][2] * M->m[1][0] * M->m[2][1] -
2493 M->m[0][2] * M->m[2][0] * M->m[1][1];
2494
2495 det =
2496 M->m[0][0] * D->m[0][0] +
2497 M->m[1][0] * D->m[0][1] +
2498 M->m[2][0] * D->m[0][2] +
2499 M->m[3][0] * D->m[0][3];
2500
2501 if (det < 1e-30) {/* non inversible */
2502 *D = *M; /* wine tests */
2503 return;
2504 }
2505
2506 det = 1.0 / det;
2507
2508 for (i = 0; i < 4; i++)
2509 for (k = 0; k < 4; k++)
2510 D->m[i][k] *= det;
2511
2512 #ifdef DEBUG
2513 {
2514 D3DMATRIX I;
2515
2516 nine_d3d_matrix_matrix_mul(&I, D, M);
2517
2518 for (i = 0; i < 4; ++i)
2519 for (k = 0; k < 4; ++k)
2520 if (fabsf(I.m[i][k] - (float)(i == k)) > 1e-3)
2521 DBG("Matrix inversion check FAILED !\n");
2522 }
2523 #endif
2524 }