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
nine_ff_vs_key_hash(const void * key)127 static uint32_t nine_ff_vs_key_hash(const void *key)
128 {
129 const 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 }
nine_ff_vs_key_comp(const void * key1,const void * key2)136 static bool nine_ff_vs_key_comp(const void *key1, const 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)) == 0;
142 }
nine_ff_ps_key_hash(const void * key)143 static uint32_t nine_ff_ps_key_hash(const void *key)
144 {
145 const 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 }
nine_ff_ps_key_comp(const void * key1,const void * key2)152 static bool nine_ff_ps_key_comp(const void *key1, const 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)) == 0;
158 }
nine_ff_fvf_key_hash(const void * key)159 static uint32_t nine_ff_fvf_key_hash(const void *key)
160 {
161 return *(DWORD *)key;
162 }
nine_ff_fvf_key_comp(const void * key1,const void * key2)163 static bool nine_ff_fvf_key_comp(const void *key1, const 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
nine_ureg_tgsi_dump(struct ureg_program * ureg,boolean override)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
get_texcoord_sn(struct pipe_screen * screen)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
build_vs_add_input(struct vs_build_ctx * vs,uint16_t ndecl)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
ureg_normalize3(struct ureg_program * ureg,struct ureg_dst dst,struct ureg_src src)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 *
nine_ff_build_vs(struct NineDevice9 * device,struct vs_build_ctx * vs)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;
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 has_aNrm = need_aNrm && key->has_normal;
368
369 /* Declare and record used inputs (needed for linkage with vertex format):
370 * (texture coordinates handled later)
371 */
372 vs->aVtx = build_vs_add_input(vs,
373 key->position_t ? NINE_DECLUSAGE_POSITIONT : NINE_DECLUSAGE_POSITION);
374
375 vs->aNrm = ureg_imm1f(ureg, 0.0f);
376 if (has_aNrm)
377 vs->aNrm = build_vs_add_input(vs, NINE_DECLUSAGE_NORMAL);
378
379 vs->aCol[0] = ureg_imm1f(ureg, 1.0f);
380 vs->aCol[1] = ureg_imm1f(ureg, 0.0f);
381
382 if (key->lighting || key->darkness) {
383 const unsigned mask = key->mtl_diffuse | key->mtl_specular |
384 key->mtl_ambient | key->mtl_emissive;
385 if ((mask & 0x1) && !key->color0in_one)
386 vs->aCol[0] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 0));
387 if ((mask & 0x2) && !key->color1in_zero)
388 vs->aCol[1] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 1));
389
390 vs->mtlD = MATERIAL_CONST(1);
391 vs->mtlA = MATERIAL_CONST(2);
392 vs->mtlS = MATERIAL_CONST(3);
393 vs->mtlE = MATERIAL_CONST(5);
394 if (key->mtl_diffuse == 1) vs->mtlD = vs->aCol[0]; else
395 if (key->mtl_diffuse == 2) vs->mtlD = vs->aCol[1];
396 if (key->mtl_ambient == 1) vs->mtlA = vs->aCol[0]; else
397 if (key->mtl_ambient == 2) vs->mtlA = vs->aCol[1];
398 if (key->mtl_specular == 1) vs->mtlS = vs->aCol[0]; else
399 if (key->mtl_specular == 2) vs->mtlS = vs->aCol[1];
400 if (key->mtl_emissive == 1) vs->mtlE = vs->aCol[0]; else
401 if (key->mtl_emissive == 2) vs->mtlE = vs->aCol[1];
402 } else {
403 if (!key->color0in_one) vs->aCol[0] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 0));
404 if (!key->color1in_zero) vs->aCol[1] = build_vs_add_input(vs, NINE_DECLUSAGE_i(COLOR, 1));
405 }
406
407 if (key->vertexpointsize)
408 vs->aPsz = build_vs_add_input(vs, NINE_DECLUSAGE_PSIZE);
409
410 if (key->vertexblend_indexed || key->passthrough & (1 << NINE_DECLUSAGE_BLENDINDICES))
411 vs->aInd = build_vs_add_input(vs, NINE_DECLUSAGE_BLENDINDICES);
412 if (key->vertexblend || key->passthrough & (1 << NINE_DECLUSAGE_BLENDWEIGHT))
413 vs->aWgt = build_vs_add_input(vs, NINE_DECLUSAGE_BLENDWEIGHT);
414 if (key->vertextween) {
415 vs->aVtx1 = build_vs_add_input(vs, NINE_DECLUSAGE_i(POSITION,1));
416 vs->aNrm1 = build_vs_add_input(vs, NINE_DECLUSAGE_i(NORMAL,1));
417 }
418
419 /* Declare outputs:
420 */
421 oPos = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0); /* HPOS */
422 oCol[0] = ureg_saturate(ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0));
423 oCol[1] = ureg_saturate(ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 1));
424 if (key->fog || key->passthrough & (1 << NINE_DECLUSAGE_FOG)) {
425 oFog = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 16);
426 oFog = ureg_writemask(oFog, TGSI_WRITEMASK_X);
427 }
428
429 if (key->vertexpointsize || key->pointscale) {
430 oPsz = ureg_DECL_output_masked(ureg, TGSI_SEMANTIC_PSIZE, 0,
431 TGSI_WRITEMASK_X, 0, 1);
432 oPsz = ureg_writemask(oPsz, TGSI_WRITEMASK_X);
433 }
434
435 if (key->lighting || key->vertexblend)
436 AR = ureg_DECL_address(ureg);
437
438 /* === Vertex transformation / vertex blending:
439 */
440
441 if (key->position_t) {
442 if (device->driver_caps.window_space_position_support) {
443 ureg_MOV(ureg, oPos, vs->aVtx);
444 } else {
445 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
446 /* vs->aVtx contains the coordinates buffer wise.
447 * later in the pipeline, clipping, viewport and division
448 * by w (rhw = 1/w) are going to be applied, so do the reverse
449 * of these transformations (except clipping) to have the good
450 * position at the end.*/
451 ureg_MOV(ureg, tmp, vs->aVtx);
452 /* X from [X_min, X_min + width] to [-1, 1], same for Y. Z to [0, 1] */
453 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), ureg_negate(_CONST(101)));
454 ureg_MUL(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _CONST(100));
455 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XY), ureg_src(tmp), ureg_imm1f(ureg, -1.0f));
456 /* Y needs to be reversed */
457 ureg_MOV(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_Y), ureg_negate(ureg_src(tmp)));
458 /* inverse rhw */
459 ureg_RCP(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_W), _W(tmp));
460 /* multiply X, Y, Z by w */
461 ureg_MUL(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(tmp), _W(tmp));
462 ureg_MOV(ureg, oPos, ureg_src(tmp));
463 ureg_release_temporary(ureg, tmp);
464 }
465 } else if (key->vertexblend) {
466 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
467 struct ureg_dst tmp2 = ureg_DECL_temporary(ureg);
468 struct ureg_dst aVtx_dst = ureg_DECL_temporary(ureg);
469 struct ureg_dst aNrm_dst = ureg_DECL_temporary(ureg);
470 struct ureg_dst sum_blendweights = ureg_DECL_temporary(ureg);
471 struct ureg_src cWM[4];
472
473 for (i = 160; i <= 195; ++i)
474 ureg_DECL_constant(ureg, i);
475
476 /* translate world matrix index to constant file index */
477 if (key->vertexblend_indexed) {
478 ureg_MAD(ureg, tmp, vs->aInd, ureg_imm1f(ureg, 4.0f), ureg_imm1f(ureg, 160.0f));
479 ureg_ARL(ureg, AR, ureg_src(tmp));
480 }
481
482 ureg_MOV(ureg, aVtx_dst, ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 0.0f));
483 ureg_MOV(ureg, aNrm_dst, ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 0.0f));
484 ureg_MOV(ureg, sum_blendweights, ureg_imm4f(ureg, 1.0f, 1.0f, 1.0f, 1.0f));
485
486 for (i = 0; i < key->vertexblend; ++i) {
487 for (c = 0; c < 4; ++c) {
488 cWM[c] = ureg_src_dimension(ureg_src_register(TGSI_FILE_CONSTANT, (160 + i * 4) * !key->vertexblend_indexed + c), 0);
489 if (key->vertexblend_indexed)
490 cWM[c] = ureg_src_indirect(cWM[c], ureg_scalar(ureg_src(AR), i));
491 }
492
493 /* multiply by WORLD(index) */
494 ureg_MUL(ureg, tmp, _XXXX(vs->aVtx), cWM[0]);
495 ureg_MAD(ureg, tmp, _YYYY(vs->aVtx), cWM[1], ureg_src(tmp));
496 ureg_MAD(ureg, tmp, _ZZZZ(vs->aVtx), cWM[2], ureg_src(tmp));
497 ureg_MAD(ureg, tmp, _WWWW(vs->aVtx), cWM[3], ureg_src(tmp));
498
499 if (has_aNrm) {
500 /* Note: the spec says the transpose of the inverse of the
501 * WorldView matrices should be used, but all tests show
502 * otherwise.
503 * Only case unknown: D3DVBF_0WEIGHTS */
504 ureg_MUL(ureg, tmp2, _XXXX(vs->aNrm), cWM[0]);
505 ureg_MAD(ureg, tmp2, _YYYY(vs->aNrm), cWM[1], ureg_src(tmp2));
506 ureg_MAD(ureg, tmp2, _ZZZZ(vs->aNrm), cWM[2], ureg_src(tmp2));
507 }
508
509 if (i < (key->vertexblend - 1)) {
510 /* accumulate weighted position value */
511 ureg_MAD(ureg, aVtx_dst, ureg_src(tmp), ureg_scalar(vs->aWgt, i), ureg_src(aVtx_dst));
512 if (has_aNrm)
513 ureg_MAD(ureg, aNrm_dst, ureg_src(tmp2), ureg_scalar(vs->aWgt, i), ureg_src(aNrm_dst));
514 /* subtract weighted position value for last value */
515 ureg_ADD(ureg, sum_blendweights, ureg_src(sum_blendweights), ureg_negate(ureg_scalar(vs->aWgt, i)));
516 }
517 }
518
519 /* the last weighted position is always 1 - sum_of_previous_weights */
520 ureg_MAD(ureg, aVtx_dst, ureg_src(tmp), ureg_scalar(ureg_src(sum_blendweights), key->vertexblend - 1), ureg_src(aVtx_dst));
521 if (has_aNrm)
522 ureg_MAD(ureg, aNrm_dst, ureg_src(tmp2), ureg_scalar(ureg_src(sum_blendweights), key->vertexblend - 1), ureg_src(aNrm_dst));
523
524 /* multiply by VIEW_PROJ */
525 ureg_MUL(ureg, tmp, _X(aVtx_dst), _CONST(8));
526 ureg_MAD(ureg, tmp, _Y(aVtx_dst), _CONST(9), ureg_src(tmp));
527 ureg_MAD(ureg, tmp, _Z(aVtx_dst), _CONST(10), ureg_src(tmp));
528 ureg_MAD(ureg, oPos, _W(aVtx_dst), _CONST(11), ureg_src(tmp));
529
530 if (need_aVtx)
531 vs->aVtx = ureg_src(aVtx_dst);
532
533 ureg_release_temporary(ureg, tmp);
534 ureg_release_temporary(ureg, tmp2);
535 ureg_release_temporary(ureg, sum_blendweights);
536 if (!need_aVtx)
537 ureg_release_temporary(ureg, aVtx_dst);
538
539 if (has_aNrm) {
540 if (key->normalizenormals)
541 ureg_normalize3(ureg, aNrm_dst, ureg_src(aNrm_dst));
542 vs->aNrm = ureg_src(aNrm_dst);
543 } else
544 ureg_release_temporary(ureg, aNrm_dst);
545 } else {
546 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
547
548 if (key->vertextween) {
549 struct ureg_dst aVtx_dst = ureg_DECL_temporary(ureg);
550 ureg_LRP(ureg, aVtx_dst, _XXXX(_CONST(30)), vs->aVtx1, vs->aVtx);
551 vs->aVtx = ureg_src(aVtx_dst);
552 if (has_aNrm) {
553 struct ureg_dst aNrm_dst = ureg_DECL_temporary(ureg);
554 ureg_LRP(ureg, aNrm_dst, _XXXX(_CONST(30)), vs->aNrm1, vs->aNrm);
555 vs->aNrm = ureg_src(aNrm_dst);
556 }
557 }
558
559 /* position = vertex * WORLD_VIEW_PROJ */
560 ureg_MUL(ureg, tmp, _XXXX(vs->aVtx), _CONST(0));
561 ureg_MAD(ureg, tmp, _YYYY(vs->aVtx), _CONST(1), ureg_src(tmp));
562 ureg_MAD(ureg, tmp, _ZZZZ(vs->aVtx), _CONST(2), ureg_src(tmp));
563 ureg_MAD(ureg, oPos, _WWWW(vs->aVtx), _CONST(3), ureg_src(tmp));
564 ureg_release_temporary(ureg, tmp);
565
566 if (need_aVtx) {
567 struct ureg_dst aVtx_dst = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
568 ureg_MUL(ureg, aVtx_dst, _XXXX(vs->aVtx), _CONST(4));
569 ureg_MAD(ureg, aVtx_dst, _YYYY(vs->aVtx), _CONST(5), ureg_src(aVtx_dst));
570 ureg_MAD(ureg, aVtx_dst, _ZZZZ(vs->aVtx), _CONST(6), ureg_src(aVtx_dst));
571 ureg_MAD(ureg, aVtx_dst, _WWWW(vs->aVtx), _CONST(7), ureg_src(aVtx_dst));
572 vs->aVtx = ureg_src(aVtx_dst);
573 }
574 if (has_aNrm) {
575 struct ureg_dst aNrm_dst = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
576 ureg_MUL(ureg, aNrm_dst, _XXXX(vs->aNrm), _CONST(16));
577 ureg_MAD(ureg, aNrm_dst, _YYYY(vs->aNrm), _CONST(17), ureg_src(aNrm_dst));
578 ureg_MAD(ureg, aNrm_dst, _ZZZZ(vs->aNrm), _CONST(18), ureg_src(aNrm_dst));
579 if (key->normalizenormals)
580 ureg_normalize3(ureg, aNrm_dst, ureg_src(aNrm_dst));
581 vs->aNrm = ureg_src(aNrm_dst);
582 }
583 }
584
585 /* === Process point size:
586 */
587 if (key->vertexpointsize || key->pointscale) {
588 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
589 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
590 struct ureg_dst tmp_y = ureg_writemask(tmp, TGSI_WRITEMASK_Y);
591 struct ureg_dst tmp_z = ureg_writemask(tmp, TGSI_WRITEMASK_Z);
592 if (key->vertexpointsize) {
593 struct ureg_src cPsz1 = ureg_DECL_constant(ureg, 26);
594 ureg_MAX(ureg, tmp_z, _XXXX(vs->aPsz), _XXXX(cPsz1));
595 ureg_MIN(ureg, tmp_z, _Z(tmp), _YYYY(cPsz1));
596 } else {
597 struct ureg_src cPsz1 = ureg_DECL_constant(ureg, 26);
598 ureg_MOV(ureg, tmp_z, _ZZZZ(cPsz1));
599 }
600
601 if (key->pointscale) {
602 struct ureg_src cPsz1 = ureg_DECL_constant(ureg, 26);
603 struct ureg_src cPsz2 = ureg_DECL_constant(ureg, 27);
604
605 ureg_DP3(ureg, tmp_x, vs->aVtx, vs->aVtx);
606 ureg_RSQ(ureg, tmp_y, _X(tmp));
607 ureg_MUL(ureg, tmp_y, _Y(tmp), _X(tmp));
608 ureg_CMP(ureg, tmp_y, ureg_negate(_Y(tmp)), _Y(tmp), ureg_imm1f(ureg, 0.0f));
609 ureg_MAD(ureg, tmp_x, _Y(tmp), _YYYY(cPsz2), _XXXX(cPsz2));
610 ureg_MAD(ureg, tmp_x, _Y(tmp), _X(tmp), _WWWW(cPsz1));
611 ureg_RSQ(ureg, tmp_x, _X(tmp));
612 ureg_MUL(ureg, tmp_x, _X(tmp), _Z(tmp));
613 ureg_MUL(ureg, tmp_x, _X(tmp), _WWWW(_CONST(100)));
614 ureg_MAX(ureg, tmp_x, _X(tmp), _XXXX(cPsz1));
615 ureg_MIN(ureg, tmp_z, _X(tmp), _YYYY(cPsz1));
616 }
617
618 ureg_MOV(ureg, oPsz, _Z(tmp));
619 ureg_release_temporary(ureg, tmp);
620 }
621
622 for (i = 0; i < 8; ++i) {
623 struct ureg_dst tmp, tmp_x, tmp2;
624 struct ureg_dst oTex, input_coord, transformed, t, aVtx_normed;
625 unsigned c, writemask;
626 const unsigned tci = (key->tc_gen >> (i * 3)) & 0x7;
627 const unsigned idx = (key->tc_idx >> (i * 3)) & 0x7;
628 unsigned dim_input = 1 + ((key->tc_dim_input >> (i * 2)) & 0x3);
629 const unsigned dim_output = (key->tc_dim_output >> (i * 3)) & 0x7;
630
631 /* No texture output of index s */
632 if (tci == NINED3DTSS_TCI_DISABLE)
633 continue;
634 oTex = ureg_DECL_output(ureg, texcoord_sn, i);
635 tmp = ureg_DECL_temporary(ureg);
636 tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
637 input_coord = ureg_DECL_temporary(ureg);
638 transformed = ureg_DECL_temporary(ureg);
639
640 /* Get the coordinate */
641 switch (tci) {
642 case NINED3DTSS_TCI_PASSTHRU:
643 /* NINED3DTSS_TCI_PASSTHRU => Use texcoord coming from index idx *
644 * Else the idx is used only to determine wrapping mode. */
645 vs->aTex[idx] = build_vs_add_input(vs, NINE_DECLUSAGE_i(TEXCOORD,idx));
646 ureg_MOV(ureg, input_coord, vs->aTex[idx]);
647 break;
648 case NINED3DTSS_TCI_CAMERASPACENORMAL:
649 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XYZ), vs->aNrm);
650 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
651 dim_input = 4;
652 break;
653 case NINED3DTSS_TCI_CAMERASPACEPOSITION:
654 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XYZ), vs->aVtx);
655 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
656 dim_input = 4;
657 break;
658 case NINED3DTSS_TCI_CAMERASPACEREFLECTIONVECTOR:
659 tmp.WriteMask = TGSI_WRITEMASK_XYZ;
660 aVtx_normed = ureg_DECL_temporary(ureg);
661 ureg_normalize3(ureg, aVtx_normed, vs->aVtx);
662 ureg_DP3(ureg, tmp_x, ureg_src(aVtx_normed), vs->aNrm);
663 ureg_MUL(ureg, tmp, vs->aNrm, _X(tmp));
664 ureg_ADD(ureg, tmp, ureg_src(tmp), ureg_src(tmp));
665 ureg_ADD(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XYZ), ureg_src(aVtx_normed), ureg_negate(ureg_src(tmp)));
666 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_W), ureg_imm1f(ureg, 1.0f));
667 ureg_release_temporary(ureg, aVtx_normed);
668 dim_input = 4;
669 tmp.WriteMask = TGSI_WRITEMASK_XYZW;
670 break;
671 case NINED3DTSS_TCI_SPHEREMAP:
672 /* Implement the formula of GL_SPHERE_MAP */
673 tmp.WriteMask = TGSI_WRITEMASK_XYZ;
674 aVtx_normed = ureg_DECL_temporary(ureg);
675 tmp2 = ureg_DECL_temporary(ureg);
676 ureg_normalize3(ureg, aVtx_normed, vs->aVtx);
677 ureg_DP3(ureg, tmp_x, ureg_src(aVtx_normed), vs->aNrm);
678 ureg_MUL(ureg, tmp, vs->aNrm, _X(tmp));
679 ureg_ADD(ureg, tmp, ureg_src(tmp), ureg_src(tmp));
680 ureg_ADD(ureg, tmp, ureg_src(aVtx_normed), ureg_negate(ureg_src(tmp)));
681 /* now tmp = normed(Vtx) - 2 dot3(normed(Vtx), Nrm) Nrm */
682 ureg_MOV(ureg, ureg_writemask(tmp2, TGSI_WRITEMASK_XYZ), ureg_src(tmp));
683 ureg_MUL(ureg, tmp2, ureg_src(tmp2), ureg_src(tmp2));
684 ureg_DP3(ureg, ureg_writemask(tmp2, TGSI_WRITEMASK_X), ureg_src(tmp2), ureg_src(tmp2));
685 ureg_RSQ(ureg, ureg_writemask(tmp2, TGSI_WRITEMASK_X), ureg_src(tmp2));
686 ureg_MUL(ureg, ureg_writemask(tmp2, TGSI_WRITEMASK_X), ureg_src(tmp2), ureg_imm1f(ureg, 0.5f));
687 /* tmp2 = 0.5 / sqrt(tmp.x^2 + tmp.y^2 + (tmp.z+1)^2)
688 * TODO: z coordinates are a bit different gl vs d3d, should the formula be adapted ? */
689 ureg_MUL(ureg, tmp, ureg_src(tmp), _X(tmp2));
690 ureg_ADD(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_XY), ureg_src(tmp), ureg_imm1f(ureg, 0.5f));
691 ureg_MOV(ureg, ureg_writemask(input_coord, TGSI_WRITEMASK_ZW), ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 1.0f));
692 ureg_release_temporary(ureg, aVtx_normed);
693 ureg_release_temporary(ureg, tmp2);
694 dim_input = 4;
695 tmp.WriteMask = TGSI_WRITEMASK_XYZW;
696 break;
697 default:
698 assert(0);
699 break;
700 }
701
702 /* Apply the transformation */
703 /* dim_output == 0 => do not transform the components.
704 * XYZRHW also disables transformation */
705 if (!dim_output || key->position_t) {
706 ureg_release_temporary(ureg, transformed);
707 transformed = input_coord;
708 writemask = TGSI_WRITEMASK_XYZW;
709 } else {
710 for (c = 0; c < dim_output; c++) {
711 t = ureg_writemask(transformed, 1 << c);
712 switch (dim_input) {
713 /* dim_input = 1 2 3: -> we add trailing 1 to input*/
714 case 1: ureg_MAD(ureg, t, _X(input_coord), _XXXX(_CONST(128 + i * 4 + c)), _YYYY(_CONST(128 + i * 4 + c)));
715 break;
716 case 2: ureg_DP2(ureg, t, ureg_src(input_coord), _CONST(128 + i * 4 + c));
717 ureg_ADD(ureg, t, ureg_src(transformed), _ZZZZ(_CONST(128 + i * 4 + c)));
718 break;
719 case 3: ureg_DP3(ureg, t, ureg_src(input_coord), _CONST(128 + i * 4 + c));
720 ureg_ADD(ureg, t, ureg_src(transformed), _WWWW(_CONST(128 + i * 4 + c)));
721 break;
722 case 4: ureg_DP4(ureg, t, ureg_src(input_coord), _CONST(128 + i * 4 + c)); break;
723 default:
724 assert(0);
725 }
726 }
727 writemask = (1 << dim_output) - 1;
728 ureg_release_temporary(ureg, input_coord);
729 }
730
731 ureg_MOV(ureg, ureg_writemask(oTex, writemask), ureg_src(transformed));
732 ureg_release_temporary(ureg, transformed);
733 ureg_release_temporary(ureg, tmp);
734 }
735
736 /* === Lighting:
737 *
738 * DIRECTIONAL: Light at infinite distance, parallel rays, no attenuation.
739 * POINT: Finite distance to scene, divergent rays, isotropic, attenuation.
740 * SPOT: Finite distance, divergent rays, angular dependence, attenuation.
741 *
742 * vec3 normal = normalize(in.Normal * NormalMatrix);
743 * vec3 hitDir = light.direction;
744 * float atten = 1.0;
745 *
746 * if (light.type != DIRECTIONAL)
747 * {
748 * vec3 hitVec = light.position - eyeVertex;
749 * float d = length(hitVec);
750 * hitDir = hitVec / d;
751 * atten = 1 / ((light.atten2 * d + light.atten1) * d + light.atten0);
752 * }
753 *
754 * if (light.type == SPOTLIGHT)
755 * {
756 * float rho = dp3(-hitVec, light.direction);
757 * if (rho < cos(light.phi / 2))
758 * atten = 0;
759 * if (rho < cos(light.theta / 2))
760 * atten *= pow(some_func(rho), light.falloff);
761 * }
762 *
763 * float nDotHit = dp3_sat(normal, hitVec);
764 * float powFact = 0.0;
765 *
766 * if (nDotHit > 0.0)
767 * {
768 * vec3 midVec = normalize(hitDir + eye);
769 * float nDotMid = dp3_sat(normal, midVec);
770 * pFact = pow(nDotMid, material.power);
771 * }
772 *
773 * ambient += light.ambient * atten;
774 * diffuse += light.diffuse * atten * nDotHit;
775 * specular += light.specular * atten * powFact;
776 */
777 if (key->lighting) {
778 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
779 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
780 struct ureg_dst tmp_y = ureg_writemask(tmp, TGSI_WRITEMASK_Y);
781 struct ureg_dst tmp_z = ureg_writemask(tmp, TGSI_WRITEMASK_Z);
782 struct ureg_dst rAtt = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_W);
783 struct ureg_dst rHit = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
784 struct ureg_dst rMid = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
785
786 struct ureg_dst rCtr = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_W);
787
788 struct ureg_dst AL = ureg_writemask(AR, TGSI_WRITEMASK_X);
789
790 /* Light.*.Alpha is not used. */
791 struct ureg_dst rD = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
792 struct ureg_dst rA = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_XYZ);
793 struct ureg_dst rS = ureg_DECL_temporary(ureg);
794
795 struct ureg_src mtlP = _XXXX(MATERIAL_CONST(4));
796
797 struct ureg_src cLKind = _XXXX(LIGHT_CONST(0));
798 struct ureg_src cLAtt0 = _YYYY(LIGHT_CONST(0));
799 struct ureg_src cLAtt1 = _ZZZZ(LIGHT_CONST(0));
800 struct ureg_src cLAtt2 = _WWWW(LIGHT_CONST(0));
801 struct ureg_src cLColD = _XYZW(LIGHT_CONST(1));
802 struct ureg_src cLColS = _XYZW(LIGHT_CONST(2));
803 struct ureg_src cLColA = _XYZW(LIGHT_CONST(3));
804 struct ureg_src cLPos = _XYZW(LIGHT_CONST(4));
805 struct ureg_src cLRng = _WWWW(LIGHT_CONST(4));
806 struct ureg_src cLDir = _XYZW(LIGHT_CONST(5));
807 struct ureg_src cLFOff = _WWWW(LIGHT_CONST(5));
808 struct ureg_src cLTht = _XXXX(LIGHT_CONST(6));
809 struct ureg_src cLPhi = _YYYY(LIGHT_CONST(6));
810 struct ureg_src cLSDiv = _ZZZZ(LIGHT_CONST(6));
811 struct ureg_src cLLast = _WWWW(LIGHT_CONST(7));
812
813 const unsigned loop_label = l++;
814
815 /* Declare all light constants to allow indirect adressing */
816 for (i = 32; i < 96; i++)
817 ureg_DECL_constant(ureg, i);
818
819 ureg_MOV(ureg, rCtr, ureg_imm1f(ureg, 32.0f)); /* &lightconst(0) */
820 ureg_MOV(ureg, rD, ureg_imm1f(ureg, 0.0f));
821 ureg_MOV(ureg, rA, ureg_imm1f(ureg, 0.0f));
822 ureg_MOV(ureg, rS, ureg_imm1f(ureg, 0.0f));
823
824 /* loop management */
825 ureg_BGNLOOP(ureg, &label[loop_label]);
826 ureg_ARL(ureg, AL, _W(rCtr));
827
828 /* if (not DIRECTIONAL light): */
829 ureg_SNE(ureg, tmp_x, cLKind, ureg_imm1f(ureg, D3DLIGHT_DIRECTIONAL));
830 ureg_MOV(ureg, rHit, ureg_negate(cLDir));
831 ureg_MOV(ureg, rAtt, ureg_imm1f(ureg, 1.0f));
832 ureg_IF(ureg, _X(tmp), &label[l++]);
833 {
834 /* hitDir = light.position - eyeVtx
835 * d = length(hitDir)
836 */
837 ureg_ADD(ureg, rHit, cLPos, ureg_negate(vs->aVtx));
838 ureg_DP3(ureg, tmp_x, ureg_src(rHit), ureg_src(rHit));
839 ureg_RSQ(ureg, tmp_y, _X(tmp));
840 ureg_MUL(ureg, tmp_x, _X(tmp), _Y(tmp)); /* length */
841
842 /* att = 1.0 / (light.att0 + (light.att1 + light.att2 * d) * d) */
843 ureg_MAD(ureg, rAtt, _X(tmp), cLAtt2, cLAtt1);
844 ureg_MAD(ureg, rAtt, _X(tmp), _W(rAtt), cLAtt0);
845 ureg_RCP(ureg, rAtt, _W(rAtt));
846 /* cut-off if distance exceeds Light.Range */
847 ureg_SLT(ureg, tmp_x, _X(tmp), cLRng);
848 ureg_MUL(ureg, rAtt, _W(rAtt), _X(tmp));
849 }
850 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
851 ureg_ENDIF(ureg);
852
853 /* normalize hitDir */
854 ureg_normalize3(ureg, rHit, ureg_src(rHit));
855
856 /* if (SPOT light) */
857 ureg_SEQ(ureg, tmp_x, cLKind, ureg_imm1f(ureg, D3DLIGHT_SPOT));
858 ureg_IF(ureg, _X(tmp), &label[l++]);
859 {
860 /* rho = dp3(-hitDir, light.spotDir)
861 *
862 * if (rho > light.ctht2) NOTE: 0 <= phi <= pi, 0 <= theta <= phi
863 * spotAtt = 1
864 * else
865 * if (rho <= light.cphi2)
866 * spotAtt = 0
867 * else
868 * spotAtt = (rho - light.cphi2) / (light.ctht2 - light.cphi2) ^ light.falloff
869 */
870 ureg_DP3(ureg, tmp_y, ureg_negate(ureg_src(rHit)), cLDir); /* rho */
871 ureg_ADD(ureg, tmp_x, _Y(tmp), ureg_negate(cLPhi));
872 ureg_MUL(ureg, tmp_x, _X(tmp), cLSDiv);
873 ureg_POW(ureg, tmp_x, _X(tmp), cLFOff); /* spotAtten */
874 ureg_SGE(ureg, tmp_z, _Y(tmp), cLTht); /* if inside theta && phi */
875 ureg_SGE(ureg, tmp_y, _Y(tmp), cLPhi); /* if inside phi */
876 ureg_MAD(ureg, ureg_saturate(tmp_x), _X(tmp), _Y(tmp), _Z(tmp));
877 ureg_MUL(ureg, rAtt, _W(rAtt), _X(tmp));
878 }
879 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
880 ureg_ENDIF(ureg);
881
882 /* directional factors, let's not use LIT because of clarity */
883
884 if (has_aNrm) {
885 if (key->localviewer) {
886 ureg_normalize3(ureg, rMid, vs->aVtx);
887 ureg_ADD(ureg, rMid, ureg_src(rHit), ureg_negate(ureg_src(rMid)));
888 } else {
889 ureg_ADD(ureg, rMid, ureg_src(rHit), ureg_imm3f(ureg, 0.0f, 0.0f, -1.0f));
890 }
891 ureg_normalize3(ureg, rMid, ureg_src(rMid));
892 ureg_DP3(ureg, ureg_saturate(tmp_x), vs->aNrm, ureg_src(rHit));
893 ureg_DP3(ureg, ureg_saturate(tmp_y), vs->aNrm, ureg_src(rMid));
894 ureg_MUL(ureg, tmp_z, _X(tmp), _Y(tmp));
895 /* Tests show that specular is computed only if (dp3(normal,hitDir) > 0).
896 * For front facing, it is more restrictive than test (dp3(normal,mid) > 0).
897 * No tests were made for backfacing, so add the two conditions */
898 ureg_IF(ureg, _Z(tmp), &label[l++]);
899 {
900 ureg_DP3(ureg, ureg_saturate(tmp_y), vs->aNrm, ureg_src(rMid));
901 ureg_POW(ureg, tmp_y, _Y(tmp), mtlP);
902 ureg_MUL(ureg, tmp_y, _W(rAtt), _Y(tmp)); /* power factor * att */
903 ureg_MAD(ureg, rS, cLColS, _Y(tmp), ureg_src(rS)); /* accumulate specular */
904 }
905 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
906 ureg_ENDIF(ureg);
907
908 ureg_MUL(ureg, tmp_x, _W(rAtt), _X(tmp)); /* dp3(normal,hitDir) * att */
909 ureg_MAD(ureg, rD, cLColD, _X(tmp), ureg_src(rD)); /* accumulate diffuse */
910 }
911
912 ureg_MAD(ureg, rA, cLColA, _W(rAtt), ureg_src(rA)); /* accumulate ambient */
913
914 /* break if this was the last light */
915 ureg_IF(ureg, cLLast, &label[l++]);
916 ureg_BRK(ureg);
917 ureg_ENDIF(ureg);
918 ureg_fixup_label(ureg, label[l-1], ureg_get_instruction_number(ureg));
919
920 ureg_ADD(ureg, rCtr, _W(rCtr), ureg_imm1f(ureg, 8.0f));
921 ureg_fixup_label(ureg, label[loop_label], ureg_get_instruction_number(ureg));
922 ureg_ENDLOOP(ureg, &label[loop_label]);
923
924 /* Apply to material:
925 *
926 * oCol[0] = (material.emissive + material.ambient * rs.ambient) +
927 * material.ambient * ambient +
928 * material.diffuse * diffuse +
929 * oCol[1] = material.specular * specular;
930 */
931 if (key->mtl_emissive == 0 && key->mtl_ambient == 0)
932 ureg_MAD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(rA), vs->mtlA, _CONST(19));
933 else {
934 ureg_ADD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), ureg_src(rA), _CONST(25));
935 ureg_MAD(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ), vs->mtlA, ureg_src(tmp), vs->mtlE);
936 }
937
938 ureg_MAD(ureg, ureg_writemask(oCol[0], TGSI_WRITEMASK_XYZ), ureg_src(rD), vs->mtlD, ureg_src(tmp));
939 ureg_MOV(ureg, ureg_writemask(oCol[0], TGSI_WRITEMASK_W), vs->mtlD);
940 ureg_MUL(ureg, oCol[1], ureg_src(rS), vs->mtlS);
941 ureg_release_temporary(ureg, rAtt);
942 ureg_release_temporary(ureg, rHit);
943 ureg_release_temporary(ureg, rMid);
944 ureg_release_temporary(ureg, rCtr);
945 ureg_release_temporary(ureg, rD);
946 ureg_release_temporary(ureg, rA);
947 ureg_release_temporary(ureg, rS);
948 ureg_release_temporary(ureg, rAtt);
949 ureg_release_temporary(ureg, tmp);
950 } else
951 /* COLOR */
952 if (key->darkness) {
953 if (key->mtl_emissive == 0 && key->mtl_ambient == 0)
954 ureg_MOV(ureg, ureg_writemask(oCol[0], TGSI_WRITEMASK_XYZ), _CONST(19));
955 else
956 ureg_MAD(ureg, ureg_writemask(oCol[0], TGSI_WRITEMASK_XYZ), vs->mtlA, _CONST(25), vs->mtlE);
957 ureg_MOV(ureg, ureg_writemask(oCol[0], TGSI_WRITEMASK_W), vs->mtlD);
958 ureg_MOV(ureg, oCol[1], ureg_imm1f(ureg, 0.0f));
959 } else {
960 ureg_MOV(ureg, oCol[0], vs->aCol[0]);
961 ureg_MOV(ureg, oCol[1], vs->aCol[1]);
962 }
963
964 /* === Process fog.
965 *
966 * exp(x) = ex2(log2(e) * x)
967 */
968 if (key->fog_mode) {
969 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
970 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
971 struct ureg_dst tmp_z = ureg_writemask(tmp, TGSI_WRITEMASK_Z);
972 if (key->fog_range) {
973 ureg_DP3(ureg, tmp_x, vs->aVtx, vs->aVtx);
974 ureg_RSQ(ureg, tmp_z, _X(tmp));
975 ureg_MUL(ureg, tmp_z, _Z(tmp), _X(tmp));
976 } else {
977 ureg_MOV(ureg, tmp_z, ureg_abs(_ZZZZ(vs->aVtx)));
978 }
979
980 if (key->fog_mode == D3DFOG_EXP) {
981 ureg_MUL(ureg, tmp_x, _Z(tmp), _ZZZZ(_CONST(28)));
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_EXP2) {
986 ureg_MUL(ureg, tmp_x, _Z(tmp), _ZZZZ(_CONST(28)));
987 ureg_MUL(ureg, tmp_x, _X(tmp), _X(tmp));
988 ureg_MUL(ureg, tmp_x, _X(tmp), ureg_imm1f(ureg, -1.442695f));
989 ureg_EX2(ureg, tmp_x, _X(tmp));
990 } else
991 if (key->fog_mode == D3DFOG_LINEAR) {
992 ureg_ADD(ureg, tmp_x, _XXXX(_CONST(28)), ureg_negate(_Z(tmp)));
993 ureg_MUL(ureg, ureg_saturate(tmp_x), _X(tmp), _YYYY(_CONST(28)));
994 }
995 ureg_MOV(ureg, oFog, _X(tmp));
996 ureg_release_temporary(ureg, tmp);
997 } else if (key->fog && !(key->passthrough & (1 << NINE_DECLUSAGE_FOG))) {
998 ureg_MOV(ureg, oFog, ureg_scalar(vs->aCol[1], TGSI_SWIZZLE_W));
999 }
1000
1001 if (key->passthrough & (1 << NINE_DECLUSAGE_BLENDWEIGHT)) {
1002 struct ureg_src input;
1003 struct ureg_dst output;
1004 input = vs->aWgt;
1005 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 19);
1006 ureg_MOV(ureg, output, input);
1007 }
1008 if (key->passthrough & (1 << NINE_DECLUSAGE_BLENDINDICES)) {
1009 struct ureg_src input;
1010 struct ureg_dst output;
1011 input = vs->aInd;
1012 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 20);
1013 ureg_MOV(ureg, output, input);
1014 }
1015 if (key->passthrough & (1 << NINE_DECLUSAGE_NORMAL)) {
1016 struct ureg_src input;
1017 struct ureg_dst output;
1018 input = vs->aNrm;
1019 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 21);
1020 ureg_MOV(ureg, output, input);
1021 }
1022 if (key->passthrough & (1 << NINE_DECLUSAGE_TANGENT)) {
1023 struct ureg_src input;
1024 struct ureg_dst output;
1025 input = build_vs_add_input(vs, NINE_DECLUSAGE_TANGENT);
1026 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 22);
1027 ureg_MOV(ureg, output, input);
1028 }
1029 if (key->passthrough & (1 << NINE_DECLUSAGE_BINORMAL)) {
1030 struct ureg_src input;
1031 struct ureg_dst output;
1032 input = build_vs_add_input(vs, NINE_DECLUSAGE_BINORMAL);
1033 output = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 23);
1034 ureg_MOV(ureg, output, input);
1035 }
1036 if (key->passthrough & (1 << NINE_DECLUSAGE_FOG)) {
1037 struct ureg_src input;
1038 struct ureg_dst output;
1039 input = build_vs_add_input(vs, NINE_DECLUSAGE_FOG);
1040 input = ureg_scalar(input, TGSI_SWIZZLE_X);
1041 output = oFog;
1042 ureg_MOV(ureg, output, input);
1043 }
1044 if (key->passthrough & (1 << NINE_DECLUSAGE_DEPTH)) {
1045 (void) 0; /* TODO: replace z of position output ? */
1046 }
1047
1048 /* ucp for ff applies on world coordinates.
1049 * aVtx is in worldview coordinates. */
1050 if (key->ucp) {
1051 struct ureg_dst clipVect = ureg_DECL_output(ureg, TGSI_SEMANTIC_CLIPVERTEX, 0);
1052 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
1053 ureg_MUL(ureg, tmp, _XXXX(vs->aVtx), _CONST(12));
1054 ureg_MAD(ureg, tmp, _YYYY(vs->aVtx), _CONST(13), ureg_src(tmp));
1055 ureg_MAD(ureg, tmp, _ZZZZ(vs->aVtx), _CONST(14), ureg_src(tmp));
1056 ureg_ADD(ureg, clipVect, _CONST(15), ureg_src(tmp));
1057 ureg_release_temporary(ureg, tmp);
1058 }
1059
1060 if (key->position_t && device->driver_caps.window_space_position_support)
1061 ureg_property(ureg, TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION, TRUE);
1062
1063 ureg_END(ureg);
1064 nine_ureg_tgsi_dump(ureg, FALSE);
1065 return nine_create_shader_with_so_and_destroy(ureg, device->context.pipe, NULL);
1066 }
1067
1068 /* PS FF constants layout:
1069 *
1070 * CONST[ 0.. 7] stage[i].D3DTSS_CONSTANT
1071 * CONST[ 8..15].x___ stage[i].D3DTSS_BUMPENVMAT00
1072 * CONST[ 8..15]._y__ stage[i].D3DTSS_BUMPENVMAT01
1073 * CONST[ 8..15].__z_ stage[i].D3DTSS_BUMPENVMAT10
1074 * CONST[ 8..15].___w stage[i].D3DTSS_BUMPENVMAT11
1075 * CONST[16..19].x_z_ stage[i].D3DTSS_BUMPENVLSCALE
1076 * CONST[17..19]._y_w stage[i].D3DTSS_BUMPENVLOFFSET
1077 *
1078 * CONST[20] D3DRS_TEXTUREFACTOR
1079 * CONST[21] D3DRS_FOGCOLOR
1080 * CONST[22].x___ RS.FogEnd
1081 * CONST[22]._y__ 1.0f / (RS.FogEnd - RS.FogStart)
1082 * CONST[22].__z_ RS.FogDensity
1083 */
1084 struct ps_build_ctx
1085 {
1086 struct ureg_program *ureg;
1087
1088 struct ureg_src vC[2]; /* DIFFUSE, SPECULAR */
1089 struct ureg_src vT[8]; /* TEXCOORD[i] */
1090 struct ureg_dst rCur; /* D3DTA_CURRENT */
1091 struct ureg_dst rMod;
1092 struct ureg_src rCurSrc;
1093 struct ureg_dst rTmp; /* D3DTA_TEMP */
1094 struct ureg_src rTmpSrc;
1095 struct ureg_dst rTex;
1096 struct ureg_src rTexSrc;
1097 struct ureg_src cBEM[8];
1098 struct ureg_src s[8];
1099
1100 struct {
1101 unsigned index;
1102 unsigned index_pre_mod;
1103 } stage;
1104 };
1105
1106 static struct ureg_src
ps_get_ts_arg(struct ps_build_ctx * ps,unsigned ta)1107 ps_get_ts_arg(struct ps_build_ctx *ps, unsigned ta)
1108 {
1109 struct ureg_src reg;
1110
1111 switch (ta & D3DTA_SELECTMASK) {
1112 case D3DTA_CONSTANT:
1113 reg = ureg_DECL_constant(ps->ureg, ps->stage.index);
1114 break;
1115 case D3DTA_CURRENT:
1116 reg = (ps->stage.index == ps->stage.index_pre_mod) ? ureg_src(ps->rMod) : ps->rCurSrc;
1117 break;
1118 case D3DTA_DIFFUSE:
1119 reg = ureg_DECL_fs_input(ps->ureg, TGSI_SEMANTIC_COLOR, 0, TGSI_INTERPOLATE_COLOR);
1120 break;
1121 case D3DTA_SPECULAR:
1122 reg = ureg_DECL_fs_input(ps->ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_COLOR);
1123 break;
1124 case D3DTA_TEMP:
1125 reg = ps->rTmpSrc;
1126 break;
1127 case D3DTA_TEXTURE:
1128 reg = ps->rTexSrc;
1129 break;
1130 case D3DTA_TFACTOR:
1131 reg = ureg_DECL_constant(ps->ureg, 20);
1132 break;
1133 default:
1134 assert(0);
1135 reg = ureg_src_undef();
1136 break;
1137 }
1138 if (ta & D3DTA_COMPLEMENT) {
1139 struct ureg_dst dst = ureg_DECL_temporary(ps->ureg);
1140 ureg_ADD(ps->ureg, dst, ureg_imm1f(ps->ureg, 1.0f), ureg_negate(reg));
1141 reg = ureg_src(dst);
1142 }
1143 if (ta & D3DTA_ALPHAREPLICATE)
1144 reg = _WWWW(reg);
1145 return reg;
1146 }
1147
1148 static struct ureg_dst
ps_get_ts_dst(struct ps_build_ctx * ps,unsigned ta)1149 ps_get_ts_dst(struct ps_build_ctx *ps, unsigned ta)
1150 {
1151 assert(!(ta & (D3DTA_COMPLEMENT | D3DTA_ALPHAREPLICATE)));
1152
1153 switch (ta & D3DTA_SELECTMASK) {
1154 case D3DTA_CURRENT:
1155 return ps->rCur;
1156 case D3DTA_TEMP:
1157 return ps->rTmp;
1158 default:
1159 assert(0);
1160 return ureg_dst_undef();
1161 }
1162 }
1163
ps_d3dtop_args_mask(D3DTEXTUREOP top)1164 static uint8_t ps_d3dtop_args_mask(D3DTEXTUREOP top)
1165 {
1166 switch (top) {
1167 case D3DTOP_DISABLE:
1168 return 0x0;
1169 case D3DTOP_SELECTARG1:
1170 case D3DTOP_PREMODULATE:
1171 return 0x2;
1172 case D3DTOP_SELECTARG2:
1173 return 0x4;
1174 case D3DTOP_MULTIPLYADD:
1175 case D3DTOP_LERP:
1176 return 0x7;
1177 default:
1178 return 0x6;
1179 }
1180 }
1181
1182 static inline boolean
is_MOV_no_op(struct ureg_dst dst,struct ureg_src src)1183 is_MOV_no_op(struct ureg_dst dst, struct ureg_src src)
1184 {
1185 return !dst.WriteMask ||
1186 (dst.File == src.File &&
1187 dst.Index == src.Index &&
1188 !dst.Indirect &&
1189 !dst.Saturate &&
1190 !src.Indirect &&
1191 !src.Negate &&
1192 !src.Absolute &&
1193 (!(dst.WriteMask & TGSI_WRITEMASK_X) || (src.SwizzleX == TGSI_SWIZZLE_X)) &&
1194 (!(dst.WriteMask & TGSI_WRITEMASK_Y) || (src.SwizzleY == TGSI_SWIZZLE_Y)) &&
1195 (!(dst.WriteMask & TGSI_WRITEMASK_Z) || (src.SwizzleZ == TGSI_SWIZZLE_Z)) &&
1196 (!(dst.WriteMask & TGSI_WRITEMASK_W) || (src.SwizzleW == TGSI_SWIZZLE_W)));
1197
1198 }
1199
1200 static void
ps_do_ts_op(struct ps_build_ctx * ps,unsigned top,struct ureg_dst dst,struct ureg_src * arg)1201 ps_do_ts_op(struct ps_build_ctx *ps, unsigned top, struct ureg_dst dst, struct ureg_src *arg)
1202 {
1203 struct ureg_program *ureg = ps->ureg;
1204 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
1205 struct ureg_dst tmp2 = ureg_DECL_temporary(ureg);
1206 struct ureg_dst tmp_x = ureg_writemask(tmp, TGSI_WRITEMASK_X);
1207
1208 tmp.WriteMask = dst.WriteMask;
1209
1210 if (top != D3DTOP_SELECTARG1 && top != D3DTOP_SELECTARG2 &&
1211 top != D3DTOP_MODULATE && top != D3DTOP_PREMODULATE &&
1212 top != D3DTOP_BLENDDIFFUSEALPHA && top != D3DTOP_BLENDTEXTUREALPHA &&
1213 top != D3DTOP_BLENDFACTORALPHA && top != D3DTOP_BLENDCURRENTALPHA &&
1214 top != D3DTOP_BUMPENVMAP && top != D3DTOP_BUMPENVMAPLUMINANCE &&
1215 top != D3DTOP_LERP)
1216 dst = ureg_saturate(dst);
1217
1218 switch (top) {
1219 case D3DTOP_SELECTARG1:
1220 if (!is_MOV_no_op(dst, arg[1]))
1221 ureg_MOV(ureg, dst, arg[1]);
1222 break;
1223 case D3DTOP_SELECTARG2:
1224 if (!is_MOV_no_op(dst, arg[2]))
1225 ureg_MOV(ureg, dst, arg[2]);
1226 break;
1227 case D3DTOP_MODULATE:
1228 ureg_MUL(ureg, dst, arg[1], arg[2]);
1229 break;
1230 case D3DTOP_MODULATE2X:
1231 ureg_MUL(ureg, tmp, arg[1], arg[2]);
1232 ureg_ADD(ureg, dst, ureg_src(tmp), ureg_src(tmp));
1233 break;
1234 case D3DTOP_MODULATE4X:
1235 ureg_MUL(ureg, tmp, arg[1], arg[2]);
1236 ureg_MUL(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 4.0f));
1237 break;
1238 case D3DTOP_ADD:
1239 ureg_ADD(ureg, dst, arg[1], arg[2]);
1240 break;
1241 case D3DTOP_ADDSIGNED:
1242 ureg_ADD(ureg, tmp, arg[1], arg[2]);
1243 ureg_ADD(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, -0.5f));
1244 break;
1245 case D3DTOP_ADDSIGNED2X:
1246 ureg_ADD(ureg, tmp, arg[1], arg[2]);
1247 ureg_MAD(ureg, dst, ureg_src(tmp), ureg_imm1f(ureg, 2.0f), ureg_imm1f(ureg, -1.0f));
1248 break;
1249 case D3DTOP_SUBTRACT:
1250 ureg_ADD(ureg, dst, arg[1], ureg_negate(arg[2]));
1251 break;
1252 case D3DTOP_ADDSMOOTH:
1253 ureg_ADD(ureg, tmp, ureg_imm1f(ureg, 1.0f), ureg_negate(arg[1]));
1254 ureg_MAD(ureg, dst, ureg_src(tmp), arg[2], arg[1]);
1255 break;
1256 case D3DTOP_BLENDDIFFUSEALPHA:
1257 ureg_LRP(ureg, dst, _WWWW(ps->vC[0]), arg[1], arg[2]);
1258 break;
1259 case D3DTOP_BLENDTEXTUREALPHA:
1260 /* XXX: alpha taken from previous stage, texture or result ? */
1261 ureg_LRP(ureg, dst, _W(ps->rTex), arg[1], arg[2]);
1262 break;
1263 case D3DTOP_BLENDFACTORALPHA:
1264 ureg_LRP(ureg, dst, _WWWW(_CONST(20)), arg[1], arg[2]);
1265 break;
1266 case D3DTOP_BLENDTEXTUREALPHAPM:
1267 ureg_ADD(ureg, tmp_x, ureg_imm1f(ureg, 1.0f), ureg_negate(_W(ps->rTex)));
1268 ureg_MAD(ureg, dst, arg[2], _X(tmp), arg[1]);
1269 break;
1270 case D3DTOP_BLENDCURRENTALPHA:
1271 ureg_LRP(ureg, dst, _WWWW(ps->rCurSrc), arg[1], arg[2]);
1272 break;
1273 case D3DTOP_PREMODULATE:
1274 ureg_MOV(ureg, dst, arg[1]);
1275 ps->stage.index_pre_mod = ps->stage.index + 1;
1276 break;
1277 case D3DTOP_MODULATEALPHA_ADDCOLOR:
1278 ureg_MAD(ureg, dst, _WWWW(arg[1]), arg[2], arg[1]);
1279 break;
1280 case D3DTOP_MODULATECOLOR_ADDALPHA:
1281 ureg_MAD(ureg, dst, arg[1], arg[2], _WWWW(arg[1]));
1282 break;
1283 case D3DTOP_MODULATEINVALPHA_ADDCOLOR:
1284 ureg_ADD(ureg, tmp_x, ureg_imm1f(ureg, 1.0f), ureg_negate(_WWWW(arg[1])));
1285 ureg_MAD(ureg, dst, _X(tmp), arg[2], arg[1]);
1286 break;
1287 case D3DTOP_MODULATEINVCOLOR_ADDALPHA:
1288 ureg_ADD(ureg, tmp, ureg_imm1f(ureg, 1.0f), ureg_negate(arg[1]));
1289 ureg_MAD(ureg, dst, ureg_src(tmp), arg[2], _WWWW(arg[1]));
1290 break;
1291 case D3DTOP_BUMPENVMAP:
1292 break;
1293 case D3DTOP_BUMPENVMAPLUMINANCE:
1294 break;
1295 case D3DTOP_DOTPRODUCT3:
1296 ureg_ADD(ureg, tmp, arg[1], ureg_imm4f(ureg,-0.5,-0.5,-0.5,-0.5));
1297 ureg_ADD(ureg, tmp2, arg[2] , ureg_imm4f(ureg,-0.5,-0.5,-0.5,-0.5));
1298 ureg_DP3(ureg, tmp, ureg_src(tmp), ureg_src(tmp2));
1299 ureg_MUL(ureg, ureg_saturate(dst), ureg_src(tmp), ureg_imm4f(ureg,4.0,4.0,4.0,4.0));
1300 break;
1301 case D3DTOP_MULTIPLYADD:
1302 ureg_MAD(ureg, dst, arg[1], arg[2], arg[0]);
1303 break;
1304 case D3DTOP_LERP:
1305 ureg_LRP(ureg, dst, arg[0], arg[1], arg[2]);
1306 break;
1307 case D3DTOP_DISABLE:
1308 /* no-op ? */
1309 break;
1310 default:
1311 assert(!"invalid D3DTOP");
1312 break;
1313 }
1314 ureg_release_temporary(ureg, tmp);
1315 ureg_release_temporary(ureg, tmp2);
1316 }
1317
1318 static void *
nine_ff_build_ps(struct NineDevice9 * device,struct nine_ff_ps_key * key)1319 nine_ff_build_ps(struct NineDevice9 *device, struct nine_ff_ps_key *key)
1320 {
1321 struct ps_build_ctx ps;
1322 struct ureg_program *ureg = ureg_create(PIPE_SHADER_FRAGMENT);
1323 struct ureg_dst oCol;
1324 unsigned s;
1325 const unsigned texcoord_sn = get_texcoord_sn(device->screen);
1326
1327 memset(&ps, 0, sizeof(ps));
1328 ps.ureg = ureg;
1329 ps.stage.index_pre_mod = -1;
1330
1331 ps.vC[0] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 0, TGSI_INTERPOLATE_COLOR);
1332
1333 ps.rCur = ureg_DECL_temporary(ureg);
1334 ps.rTmp = ureg_DECL_temporary(ureg);
1335 ps.rTex = ureg_DECL_temporary(ureg);
1336 ps.rCurSrc = ureg_src(ps.rCur);
1337 ps.rTmpSrc = ureg_src(ps.rTmp);
1338 ps.rTexSrc = ureg_src(ps.rTex);
1339
1340 /* Initial values */
1341 ureg_MOV(ureg, ps.rCur, ps.vC[0]);
1342 ureg_MOV(ureg, ps.rTmp, ureg_imm1f(ureg, 0.0f));
1343 ureg_MOV(ureg, ps.rTex, ureg_imm1f(ureg, 0.0f));
1344
1345 for (s = 0; s < 8; ++s) {
1346 ps.s[s] = ureg_src_undef();
1347
1348 if (key->ts[s].colorop != D3DTOP_DISABLE) {
1349 if (key->ts[s].colorarg0 == D3DTA_SPECULAR ||
1350 key->ts[s].colorarg1 == D3DTA_SPECULAR ||
1351 key->ts[s].colorarg2 == D3DTA_SPECULAR)
1352 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_COLOR);
1353
1354 if (key->ts[s].colorarg0 == D3DTA_TEXTURE ||
1355 key->ts[s].colorarg1 == D3DTA_TEXTURE ||
1356 key->ts[s].colorarg2 == D3DTA_TEXTURE ||
1357 key->ts[s].colorop == D3DTOP_BLENDTEXTUREALPHA ||
1358 key->ts[s].colorop == D3DTOP_BLENDTEXTUREALPHAPM) {
1359 ps.s[s] = ureg_DECL_sampler(ureg, s);
1360 ps.vT[s] = ureg_DECL_fs_input(ureg, texcoord_sn, s, TGSI_INTERPOLATE_PERSPECTIVE);
1361 }
1362 if (s && (key->ts[s - 1].colorop == D3DTOP_PREMODULATE ||
1363 key->ts[s - 1].alphaop == D3DTOP_PREMODULATE))
1364 ps.s[s] = ureg_DECL_sampler(ureg, s);
1365 }
1366
1367 if (key->ts[s].alphaop != D3DTOP_DISABLE) {
1368 if (key->ts[s].alphaarg0 == D3DTA_SPECULAR ||
1369 key->ts[s].alphaarg1 == D3DTA_SPECULAR ||
1370 key->ts[s].alphaarg2 == D3DTA_SPECULAR)
1371 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_COLOR);
1372
1373 if (key->ts[s].alphaarg0 == D3DTA_TEXTURE ||
1374 key->ts[s].alphaarg1 == D3DTA_TEXTURE ||
1375 key->ts[s].alphaarg2 == D3DTA_TEXTURE ||
1376 key->ts[s].colorop == D3DTOP_BLENDTEXTUREALPHA ||
1377 key->ts[s].colorop == D3DTOP_BLENDTEXTUREALPHAPM) {
1378 ps.s[s] = ureg_DECL_sampler(ureg, s);
1379 ps.vT[s] = ureg_DECL_fs_input(ureg, texcoord_sn, s, TGSI_INTERPOLATE_PERSPECTIVE);
1380 }
1381 }
1382 }
1383 if (key->specular)
1384 ps.vC[1] = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_COLOR, 1, TGSI_INTERPOLATE_COLOR);
1385
1386 oCol = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
1387
1388 /* Run stages.
1389 */
1390 for (s = 0; s < 8; ++s) {
1391 unsigned colorarg[3];
1392 unsigned alphaarg[3];
1393 const uint8_t used_c = ps_d3dtop_args_mask(key->ts[s].colorop);
1394 const uint8_t used_a = ps_d3dtop_args_mask(key->ts[s].alphaop);
1395 struct ureg_dst dst;
1396 struct ureg_src arg[3];
1397
1398 if (key->ts[s].colorop == D3DTOP_DISABLE) {
1399 assert (key->ts[s].alphaop == D3DTOP_DISABLE);
1400 continue;
1401 }
1402 ps.stage.index = s;
1403
1404 DBG("STAGE[%u]: colorop=%s alphaop=%s\n", s,
1405 nine_D3DTOP_to_str(key->ts[s].colorop),
1406 nine_D3DTOP_to_str(key->ts[s].alphaop));
1407
1408 if (!ureg_src_is_undef(ps.s[s])) {
1409 unsigned target;
1410 struct ureg_src texture_coord = ps.vT[s];
1411 struct ureg_dst delta;
1412 switch (key->ts[s].textarget) {
1413 case 0: target = TGSI_TEXTURE_1D; break;
1414 case 1: target = TGSI_TEXTURE_2D; break;
1415 case 2: target = TGSI_TEXTURE_3D; break;
1416 case 3: target = TGSI_TEXTURE_CUBE; break;
1417 /* this is a 2 bit bitfield, do I really need a default case ? */
1418 }
1419
1420 /* Modify coordinates */
1421 if (s >= 1 &&
1422 (key->ts[s-1].colorop == D3DTOP_BUMPENVMAP ||
1423 key->ts[s-1].colorop == D3DTOP_BUMPENVMAPLUMINANCE)) {
1424 delta = ureg_DECL_temporary(ureg);
1425 /* Du' = D3DTSS_BUMPENVMAT00(stage s-1)*t(s-1)R + D3DTSS_BUMPENVMAT10(stage s-1)*t(s-1)G */
1426 ureg_MUL(ureg, ureg_writemask(delta, TGSI_WRITEMASK_X), _X(ps.rTex), _XXXX(_CONST(8 + s - 1)));
1427 ureg_MAD(ureg, ureg_writemask(delta, TGSI_WRITEMASK_X), _Y(ps.rTex), _ZZZZ(_CONST(8 + s - 1)), ureg_src(delta));
1428 /* Dv' = D3DTSS_BUMPENVMAT01(stage s-1)*t(s-1)R + D3DTSS_BUMPENVMAT11(stage s-1)*t(s-1)G */
1429 ureg_MUL(ureg, ureg_writemask(delta, TGSI_WRITEMASK_Y), _X(ps.rTex), _YYYY(_CONST(8 + s - 1)));
1430 ureg_MAD(ureg, ureg_writemask(delta, TGSI_WRITEMASK_Y), _Y(ps.rTex), _WWWW(_CONST(8 + s - 1)), ureg_src(delta));
1431 texture_coord = ureg_src(ureg_DECL_temporary(ureg));
1432 ureg_MOV(ureg, ureg_writemask(ureg_dst(texture_coord), ureg_dst(ps.vT[s]).WriteMask), ps.vT[s]);
1433 ureg_ADD(ureg, ureg_writemask(ureg_dst(texture_coord), TGSI_WRITEMASK_XY), texture_coord, ureg_src(delta));
1434 /* Prepare luminance multiplier
1435 * t(s)RGBA = t(s)RGBA * clamp[(t(s-1)B * D3DTSS_BUMPENVLSCALE(stage s-1)) + D3DTSS_BUMPENVLOFFSET(stage s-1)] */
1436 if (key->ts[s-1].colorop == D3DTOP_BUMPENVMAPLUMINANCE) {
1437 struct ureg_src bumpenvlscale = ((s-1) & 1) ? _ZZZZ(_CONST(16 + (s-1) / 2)) : _XXXX(_CONST(16 + (s-1) / 2));
1438 struct ureg_src bumpenvloffset = ((s-1) & 1) ? _WWWW(_CONST(16 + (s-1) / 2)) : _YYYY(_CONST(16 + (s-1) / 2));
1439
1440 ureg_MAD(ureg, ureg_saturate(ureg_writemask(delta, TGSI_WRITEMASK_X)), _Z(ps.rTex), bumpenvlscale, bumpenvloffset);
1441 }
1442 }
1443 if (key->projected & (3 << (s *2))) {
1444 unsigned dim = 1 + ((key->projected >> (2 * s)) & 3);
1445 if (dim == 4)
1446 ureg_TXP(ureg, ps.rTex, target, texture_coord, ps.s[s]);
1447 else {
1448 struct ureg_dst tmp = ureg_DECL_temporary(ureg);
1449 ureg_RCP(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_X), ureg_scalar(texture_coord, dim-1));
1450 ureg_MUL(ureg, ps.rTmp, _X(tmp), texture_coord);
1451 ureg_TEX(ureg, ps.rTex, target, ps.rTmpSrc, ps.s[s]);
1452 ureg_release_temporary(ureg, tmp);
1453 }
1454 } else {
1455 ureg_TEX(ureg, ps.rTex, target, texture_coord, ps.s[s]);
1456 }
1457 if (s >= 1 && key->ts[s-1].colorop == D3DTOP_BUMPENVMAPLUMINANCE)
1458 ureg_MUL(ureg, ps.rTex, ureg_src(ps.rTex), _X(delta));
1459 }
1460
1461 if (key->ts[s].colorop == D3DTOP_BUMPENVMAP ||
1462 key->ts[s].colorop == D3DTOP_BUMPENVMAPLUMINANCE)
1463 continue;
1464
1465 dst = ps_get_ts_dst(&ps, key->ts[s].resultarg ? D3DTA_TEMP : D3DTA_CURRENT);
1466
1467 if (ps.stage.index_pre_mod == ps.stage.index) {
1468 ps.rMod = ureg_DECL_temporary(ureg);
1469 ureg_MUL(ureg, ps.rMod, ps.rCurSrc, ps.rTexSrc);
1470 }
1471
1472 colorarg[0] = (key->ts[s].colorarg0 | (((key->colorarg_b4[0] >> s) & 0x1) << 4) | ((key->colorarg_b5[0] >> s) << 5)) & 0x3f;
1473 colorarg[1] = (key->ts[s].colorarg1 | (((key->colorarg_b4[1] >> s) & 0x1) << 4) | ((key->colorarg_b5[1] >> s) << 5)) & 0x3f;
1474 colorarg[2] = (key->ts[s].colorarg2 | (((key->colorarg_b4[2] >> s) & 0x1) << 4) | ((key->colorarg_b5[2] >> s) << 5)) & 0x3f;
1475 alphaarg[0] = (key->ts[s].alphaarg0 | ((key->alphaarg_b4[0] >> s) << 4)) & 0x1f;
1476 alphaarg[1] = (key->ts[s].alphaarg1 | ((key->alphaarg_b4[1] >> s) << 4)) & 0x1f;
1477 alphaarg[2] = (key->ts[s].alphaarg2 | ((key->alphaarg_b4[2] >> s) << 4)) & 0x1f;
1478
1479 if (key->ts[s].colorop != key->ts[s].alphaop ||
1480 colorarg[0] != alphaarg[0] ||
1481 colorarg[1] != alphaarg[1] ||
1482 colorarg[2] != alphaarg[2])
1483 dst.WriteMask = TGSI_WRITEMASK_XYZ;
1484
1485 /* Special DOTPRODUCT behaviour (see wine tests) */
1486 if (key->ts[s].colorop == D3DTOP_DOTPRODUCT3)
1487 dst.WriteMask = TGSI_WRITEMASK_XYZW;
1488
1489 if (used_c & 0x1) arg[0] = ps_get_ts_arg(&ps, colorarg[0]);
1490 if (used_c & 0x2) arg[1] = ps_get_ts_arg(&ps, colorarg[1]);
1491 if (used_c & 0x4) arg[2] = ps_get_ts_arg(&ps, colorarg[2]);
1492 ps_do_ts_op(&ps, key->ts[s].colorop, dst, arg);
1493
1494 if (dst.WriteMask != TGSI_WRITEMASK_XYZW) {
1495 dst.WriteMask = TGSI_WRITEMASK_W;
1496
1497 if (used_a & 0x1) arg[0] = ps_get_ts_arg(&ps, alphaarg[0]);
1498 if (used_a & 0x2) arg[1] = ps_get_ts_arg(&ps, alphaarg[1]);
1499 if (used_a & 0x4) arg[2] = ps_get_ts_arg(&ps, alphaarg[2]);
1500 ps_do_ts_op(&ps, key->ts[s].alphaop, dst, arg);
1501 }
1502 }
1503
1504 if (key->specular)
1505 ureg_ADD(ureg, ureg_writemask(ps.rCur, TGSI_WRITEMASK_XYZ), ps.rCurSrc, ps.vC[1]);
1506
1507 /* Fog.
1508 */
1509 if (key->fog_mode) {
1510 struct ureg_dst rFog = ureg_writemask(ps.rTmp, TGSI_WRITEMASK_X);
1511 struct ureg_src vPos;
1512 if (device->screen->get_param(device->screen,
1513 PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL)) {
1514 vPos = ureg_DECL_system_value(ureg, TGSI_SEMANTIC_POSITION, 0);
1515 } else {
1516 vPos = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_POSITION, 0,
1517 TGSI_INTERPOLATE_LINEAR);
1518 }
1519
1520 /* Source is either W or Z.
1521 * When we use vs ff,
1522 * Z is when an orthogonal projection matrix is detected,
1523 * W (WFOG) else.
1524 * Z is used for programmable vs.
1525 * Note: Tests indicate that the projection matrix coefficients do
1526 * actually affect pixel fog (and not vertex fog) when vs ff is used,
1527 * which justifies taking the position's w instead of taking the z coordinate
1528 * before the projection in the vs shader.
1529 */
1530 if (!key->fog_source)
1531 ureg_MOV(ureg, rFog, _ZZZZ(vPos));
1532 else
1533 /* Position's w is 1/w */
1534 ureg_RCP(ureg, rFog, _WWWW(vPos));
1535
1536 if (key->fog_mode == D3DFOG_EXP) {
1537 ureg_MUL(ureg, rFog, _X(rFog), _ZZZZ(_CONST(22)));
1538 ureg_MUL(ureg, rFog, _X(rFog), ureg_imm1f(ureg, -1.442695f));
1539 ureg_EX2(ureg, rFog, _X(rFog));
1540 } else
1541 if (key->fog_mode == D3DFOG_EXP2) {
1542 ureg_MUL(ureg, rFog, _X(rFog), _ZZZZ(_CONST(22)));
1543 ureg_MUL(ureg, rFog, _X(rFog), _X(rFog));
1544 ureg_MUL(ureg, rFog, _X(rFog), ureg_imm1f(ureg, -1.442695f));
1545 ureg_EX2(ureg, rFog, _X(rFog));
1546 } else
1547 if (key->fog_mode == D3DFOG_LINEAR) {
1548 ureg_ADD(ureg, rFog, _XXXX(_CONST(22)), ureg_negate(_X(rFog)));
1549 ureg_MUL(ureg, ureg_saturate(rFog), _X(rFog), _YYYY(_CONST(22)));
1550 }
1551 ureg_LRP(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_XYZ), _X(rFog), ps.rCurSrc, _CONST(21));
1552 ureg_MOV(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_W), ps.rCurSrc);
1553 } else
1554 if (key->fog) {
1555 struct ureg_src vFog = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_GENERIC, 16, TGSI_INTERPOLATE_PERSPECTIVE);
1556 ureg_LRP(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_XYZ), _XXXX(vFog), ps.rCurSrc, _CONST(21));
1557 ureg_MOV(ureg, ureg_writemask(oCol, TGSI_WRITEMASK_W), ps.rCurSrc);
1558 } else {
1559 ureg_MOV(ureg, oCol, ps.rCurSrc);
1560 }
1561
1562 ureg_END(ureg);
1563 nine_ureg_tgsi_dump(ureg, FALSE);
1564 return nine_create_shader_with_so_and_destroy(ureg, device->context.pipe, NULL);
1565 }
1566
1567 static struct NineVertexShader9 *
nine_ff_get_vs(struct NineDevice9 * device)1568 nine_ff_get_vs(struct NineDevice9 *device)
1569 {
1570 const struct nine_context *context = &device->context;
1571 struct NineVertexShader9 *vs;
1572 struct vs_build_ctx bld;
1573 struct nine_ff_vs_key key;
1574 unsigned s, i;
1575 boolean has_indexes = false;
1576 boolean has_weights = false;
1577 char input_texture_coord[8];
1578
1579 assert(sizeof(key) <= sizeof(key.value32));
1580
1581 memset(&key, 0, sizeof(key));
1582 memset(&bld, 0, sizeof(bld));
1583 memset(&input_texture_coord, 0, sizeof(input_texture_coord));
1584
1585 bld.key = &key;
1586
1587 /* FIXME: this shouldn't be NULL, but it is on init */
1588 if (context->vdecl) {
1589 key.color0in_one = 1;
1590 key.color1in_zero = 1;
1591 for (i = 0; i < context->vdecl->nelems; i++) {
1592 uint16_t usage = context->vdecl->usage_map[i];
1593 if (usage == NINE_DECLUSAGE_POSITIONT)
1594 key.position_t = 1;
1595 else if (usage == NINE_DECLUSAGE_i(COLOR, 0))
1596 key.color0in_one = 0;
1597 else if (usage == NINE_DECLUSAGE_i(COLOR, 1))
1598 key.color1in_zero = 0;
1599 else if (usage == NINE_DECLUSAGE_i(BLENDINDICES, 0)) {
1600 has_indexes = true;
1601 key.passthrough |= 1 << usage;
1602 } else if (usage == NINE_DECLUSAGE_i(BLENDWEIGHT, 0)) {
1603 has_weights = true;
1604 key.passthrough |= 1 << usage;
1605 } else if (usage == NINE_DECLUSAGE_i(NORMAL, 0)) {
1606 key.has_normal = 1;
1607 key.passthrough |= 1 << usage;
1608 } else if (usage == NINE_DECLUSAGE_PSIZE)
1609 key.vertexpointsize = 1;
1610 else if (usage % NINE_DECLUSAGE_COUNT == NINE_DECLUSAGE_TEXCOORD) {
1611 s = usage / NINE_DECLUSAGE_COUNT;
1612 if (s < 8)
1613 input_texture_coord[s] = nine_decltype_get_dim(context->vdecl->decls[i].Type);
1614 else
1615 DBG("FF given texture coordinate >= 8. Ignoring\n");
1616 } else if (usage < NINE_DECLUSAGE_NONE)
1617 key.passthrough |= 1 << usage;
1618 }
1619 }
1620 /* ff vs + ps 3.0: some elements are passed to the ps (wine test).
1621 * We do restrict to indices 0 */
1622 key.passthrough &= ~((1 << NINE_DECLUSAGE_POSITION) | (1 << NINE_DECLUSAGE_PSIZE) |
1623 (1 << NINE_DECLUSAGE_TEXCOORD) | (1 << NINE_DECLUSAGE_POSITIONT) |
1624 (1 << NINE_DECLUSAGE_TESSFACTOR) | (1 << NINE_DECLUSAGE_SAMPLE));
1625 if (!key.position_t)
1626 key.passthrough = 0;
1627 key.pointscale = !!context->rs[D3DRS_POINTSCALEENABLE];
1628
1629 key.lighting = !!context->rs[D3DRS_LIGHTING] && context->ff.num_lights_active;
1630 key.darkness = !!context->rs[D3DRS_LIGHTING] && !context->ff.num_lights_active;
1631 if (key.position_t) {
1632 key.darkness = 0; /* |= key.lighting; */ /* XXX ? */
1633 key.lighting = 0;
1634 }
1635 if ((key.lighting | key.darkness) && context->rs[D3DRS_COLORVERTEX]) {
1636 uint32_t mask = (key.color0in_one ? 0 : 1) | (key.color1in_zero ? 0 : 2);
1637 key.mtl_diffuse = context->rs[D3DRS_DIFFUSEMATERIALSOURCE] & mask;
1638 key.mtl_ambient = context->rs[D3DRS_AMBIENTMATERIALSOURCE] & mask;
1639 key.mtl_specular = context->rs[D3DRS_SPECULARMATERIALSOURCE] & mask;
1640 key.mtl_emissive = context->rs[D3DRS_EMISSIVEMATERIALSOURCE] & mask;
1641 }
1642 key.fog = !!context->rs[D3DRS_FOGENABLE];
1643 key.fog_mode = (!key.position_t && context->rs[D3DRS_FOGENABLE]) ? context->rs[D3DRS_FOGVERTEXMODE] : 0;
1644 if (key.fog_mode)
1645 key.fog_range = context->rs[D3DRS_RANGEFOGENABLE];
1646
1647 key.localviewer = !!context->rs[D3DRS_LOCALVIEWER];
1648 key.normalizenormals = !!context->rs[D3DRS_NORMALIZENORMALS];
1649 key.ucp = !!context->rs[D3DRS_CLIPPLANEENABLE];
1650
1651 if (context->rs[D3DRS_VERTEXBLEND] != D3DVBF_DISABLE) {
1652 key.vertexblend_indexed = !!context->rs[D3DRS_INDEXEDVERTEXBLENDENABLE] && has_indexes;
1653
1654 switch (context->rs[D3DRS_VERTEXBLEND]) {
1655 case D3DVBF_0WEIGHTS: key.vertexblend = key.vertexblend_indexed; break;
1656 case D3DVBF_1WEIGHTS: key.vertexblend = 2; break;
1657 case D3DVBF_2WEIGHTS: key.vertexblend = 3; break;
1658 case D3DVBF_3WEIGHTS: key.vertexblend = 4; break;
1659 case D3DVBF_TWEENING: key.vertextween = 1; break;
1660 default:
1661 assert(!"invalid D3DVBF");
1662 break;
1663 }
1664 if (!has_weights && context->rs[D3DRS_VERTEXBLEND] != D3DVBF_0WEIGHTS)
1665 key.vertexblend = 0; /* TODO: if key.vertexblend_indexed, perhaps it should use 1.0 as weight, or revert to D3DVBF_0WEIGHTS */
1666 }
1667
1668 for (s = 0; s < 8; ++s) {
1669 unsigned gen = (context->ff.tex_stage[s][D3DTSS_TEXCOORDINDEX] >> 16) + 1;
1670 unsigned idx = context->ff.tex_stage[s][D3DTSS_TEXCOORDINDEX] & 7;
1671 unsigned dim;
1672
1673 if (key.position_t && gen > NINED3DTSS_TCI_PASSTHRU)
1674 gen = NINED3DTSS_TCI_PASSTHRU;
1675
1676 if (!input_texture_coord[idx] && gen == NINED3DTSS_TCI_PASSTHRU)
1677 gen = NINED3DTSS_TCI_DISABLE;
1678
1679 key.tc_gen |= gen << (s * 3);
1680 key.tc_idx |= idx << (s * 3);
1681 key.tc_dim_input |= ((input_texture_coord[idx]-1) & 0x3) << (s * 2);
1682
1683 dim = context->ff.tex_stage[s][D3DTSS_TEXTURETRANSFORMFLAGS] & 0x7;
1684 if (dim > 4)
1685 dim = input_texture_coord[idx];
1686 if (dim == 1) /* NV behaviour */
1687 dim = 0;
1688 key.tc_dim_output |= dim << (s * 3);
1689 }
1690
1691 DBG("VS ff key hash: %x\n", nine_ff_vs_key_hash(&key));
1692 vs = util_hash_table_get(device->ff.ht_vs, &key);
1693 if (vs)
1694 return vs;
1695 NineVertexShader9_new(device, &vs, NULL, nine_ff_build_vs(device, &bld));
1696
1697 nine_ff_prune_vs(device);
1698 if (vs) {
1699 unsigned n;
1700
1701 memcpy(&vs->ff_key, &key, sizeof(vs->ff_key));
1702
1703 _mesa_hash_table_insert(device->ff.ht_vs, &vs->ff_key, vs);
1704 device->ff.num_vs++;
1705
1706 vs->num_inputs = bld.num_inputs;
1707 for (n = 0; n < bld.num_inputs; ++n)
1708 vs->input_map[n].ndecl = bld.input[n];
1709
1710 vs->position_t = key.position_t;
1711 vs->point_size = key.vertexpointsize | key.pointscale;
1712 }
1713 return vs;
1714 }
1715
1716 #define GET_D3DTS(n) nine_state_access_transform(&context->ff, D3DTS_##n, FALSE)
1717 #define IS_D3DTS_DIRTY(s,n) ((s)->ff.changed.transform[(D3DTS_##n) / 32] & (1 << ((D3DTS_##n) % 32)))
1718
1719 static struct NinePixelShader9 *
nine_ff_get_ps(struct NineDevice9 * device)1720 nine_ff_get_ps(struct NineDevice9 *device)
1721 {
1722 struct nine_context *context = &device->context;
1723 D3DMATRIX *projection_matrix = GET_D3DTS(PROJECTION);
1724 struct NinePixelShader9 *ps;
1725 struct nine_ff_ps_key key;
1726 unsigned s;
1727 uint8_t sampler_mask = 0;
1728
1729 assert(sizeof(key) <= sizeof(key.value32));
1730
1731 memset(&key, 0, sizeof(key));
1732 for (s = 0; s < 8; ++s) {
1733 key.ts[s].colorop = context->ff.tex_stage[s][D3DTSS_COLOROP];
1734 key.ts[s].alphaop = context->ff.tex_stage[s][D3DTSS_ALPHAOP];
1735 const uint8_t used_c = ps_d3dtop_args_mask(key.ts[s].colorop);
1736 const uint8_t used_a = ps_d3dtop_args_mask(key.ts[s].alphaop);
1737 /* MSDN says D3DTOP_DISABLE disables this and all subsequent stages.
1738 * ALPHAOP cannot be enabled if COLOROP is disabled.
1739 * Verified on Windows. */
1740 if (key.ts[s].colorop == D3DTOP_DISABLE) {
1741 key.ts[s].alphaop = D3DTOP_DISABLE; /* DISABLE == 1, avoid degenerate keys */
1742 break;
1743 }
1744
1745 if (!context->texture[s].enabled &&
1746 ((context->ff.tex_stage[s][D3DTSS_COLORARG0] == D3DTA_TEXTURE &&
1747 used_c & 0x1) ||
1748 (context->ff.tex_stage[s][D3DTSS_COLORARG1] == D3DTA_TEXTURE &&
1749 used_c & 0x2) ||
1750 (context->ff.tex_stage[s][D3DTSS_COLORARG2] == D3DTA_TEXTURE &&
1751 used_c & 0x4))) {
1752 /* Tested on Windows: Invalid texture read disables the stage
1753 * and the subsequent ones, but only for colorop. For alpha,
1754 * it's as if the texture had alpha of 1.0, which is what
1755 * has our dummy texture in that case. Invalid color also
1756 * disabled the following alpha stages. */
1757 key.ts[s].colorop = key.ts[s].alphaop = D3DTOP_DISABLE;
1758 break;
1759 }
1760
1761 if (context->ff.tex_stage[s][D3DTSS_COLORARG0] == D3DTA_TEXTURE ||
1762 context->ff.tex_stage[s][D3DTSS_COLORARG1] == D3DTA_TEXTURE ||
1763 context->ff.tex_stage[s][D3DTSS_COLORARG2] == D3DTA_TEXTURE ||
1764 context->ff.tex_stage[s][D3DTSS_ALPHAARG0] == D3DTA_TEXTURE ||
1765 context->ff.tex_stage[s][D3DTSS_ALPHAARG1] == D3DTA_TEXTURE ||
1766 context->ff.tex_stage[s][D3DTSS_ALPHAARG2] == D3DTA_TEXTURE)
1767 sampler_mask |= (1 << s);
1768
1769 if (key.ts[s].colorop != D3DTOP_DISABLE) {
1770 if (used_c & 0x1) key.ts[s].colorarg0 = context->ff.tex_stage[s][D3DTSS_COLORARG0] & 0x7;
1771 if (used_c & 0x2) key.ts[s].colorarg1 = context->ff.tex_stage[s][D3DTSS_COLORARG1] & 0x7;
1772 if (used_c & 0x4) key.ts[s].colorarg2 = context->ff.tex_stage[s][D3DTSS_COLORARG2] & 0x7;
1773 if (used_c & 0x1) key.colorarg_b4[0] |= ((context->ff.tex_stage[s][D3DTSS_COLORARG0] >> 4) & 0x1) << s;
1774 if (used_c & 0x1) key.colorarg_b5[0] |= ((context->ff.tex_stage[s][D3DTSS_COLORARG0] >> 5) & 0x1) << s;
1775 if (used_c & 0x2) key.colorarg_b4[1] |= ((context->ff.tex_stage[s][D3DTSS_COLORARG1] >> 4) & 0x1) << s;
1776 if (used_c & 0x2) key.colorarg_b5[1] |= ((context->ff.tex_stage[s][D3DTSS_COLORARG1] >> 5) & 0x1) << s;
1777 if (used_c & 0x4) key.colorarg_b4[2] |= ((context->ff.tex_stage[s][D3DTSS_COLORARG2] >> 4) & 0x1) << s;
1778 if (used_c & 0x4) key.colorarg_b5[2] |= ((context->ff.tex_stage[s][D3DTSS_COLORARG2] >> 5) & 0x1) << s;
1779 }
1780 if (key.ts[s].alphaop != D3DTOP_DISABLE) {
1781 if (used_a & 0x1) key.ts[s].alphaarg0 = context->ff.tex_stage[s][D3DTSS_ALPHAARG0] & 0x7;
1782 if (used_a & 0x2) key.ts[s].alphaarg1 = context->ff.tex_stage[s][D3DTSS_ALPHAARG1] & 0x7;
1783 if (used_a & 0x4) key.ts[s].alphaarg2 = context->ff.tex_stage[s][D3DTSS_ALPHAARG2] & 0x7;
1784 if (used_a & 0x1) key.alphaarg_b4[0] |= ((context->ff.tex_stage[s][D3DTSS_ALPHAARG0] >> 4) & 0x1) << s;
1785 if (used_a & 0x2) key.alphaarg_b4[1] |= ((context->ff.tex_stage[s][D3DTSS_ALPHAARG1] >> 4) & 0x1) << s;
1786 if (used_a & 0x4) key.alphaarg_b4[2] |= ((context->ff.tex_stage[s][D3DTSS_ALPHAARG2] >> 4) & 0x1) << s;
1787 }
1788 key.ts[s].resultarg = context->ff.tex_stage[s][D3DTSS_RESULTARG] == D3DTA_TEMP;
1789
1790 if (context->texture[s].enabled) {
1791 switch (context->texture[s].type) {
1792 case D3DRTYPE_TEXTURE: key.ts[s].textarget = 1; break;
1793 case D3DRTYPE_VOLUMETEXTURE: key.ts[s].textarget = 2; break;
1794 case D3DRTYPE_CUBETEXTURE: key.ts[s].textarget = 3; break;
1795 default:
1796 assert(!"unexpected texture type");
1797 break;
1798 }
1799 } else {
1800 key.ts[s].textarget = 1;
1801 }
1802 }
1803
1804 /* Note: If colorop is D3DTOP_DISABLE for the first stage
1805 * (which implies alphaop is too), nothing particular happens,
1806 * that is, current is equal to diffuse (which is the case anyway,
1807 * because it is how it is initialized).
1808 * Special case seems if alphaop is D3DTOP_DISABLE and not colorop,
1809 * because then if the resultarg is TEMP, then diffuse alpha is written
1810 * to it. */
1811 if (key.ts[0].colorop != D3DTOP_DISABLE &&
1812 key.ts[0].alphaop == D3DTOP_DISABLE &&
1813 key.ts[0].resultarg != 0) {
1814 key.ts[0].alphaop = D3DTOP_SELECTARG1;
1815 key.ts[0].alphaarg1 = D3DTA_DIFFUSE;
1816 }
1817 /* When no alpha stage writes to current, diffuse alpha is taken.
1818 * Since we initialize current to diffuse, we have the behaviour. */
1819
1820 /* Last stage always writes to Current */
1821 if (s >= 1)
1822 key.ts[s-1].resultarg = 0;
1823
1824 key.projected = nine_ff_get_projected_key_ff(context);
1825 key.specular = !!context->rs[D3DRS_SPECULARENABLE];
1826
1827 for (; s < 8; ++s)
1828 key.ts[s].colorop = key.ts[s].alphaop = D3DTOP_DISABLE;
1829 if (context->rs[D3DRS_FOGENABLE])
1830 key.fog_mode = context->rs[D3DRS_FOGTABLEMODE];
1831 key.fog = !!context->rs[D3DRS_FOGENABLE];
1832 /* Pixel fog (with WFOG advertised): source is either Z or W.
1833 * W is the source if vs ff is used, and the
1834 * projection matrix is not orthogonal.
1835 * Tests on Win 10 seem to indicate _34
1836 * and _33 are checked against 0, 1. */
1837 if (key.fog_mode && key.fog)
1838 key.fog_source = !context->programmable_vs &&
1839 !(projection_matrix->_34 == 0.0f &&
1840 projection_matrix->_44 == 1.0f);
1841
1842 DBG("PS ff key hash: %x\n", nine_ff_ps_key_hash(&key));
1843 ps = util_hash_table_get(device->ff.ht_ps, &key);
1844 if (ps)
1845 return ps;
1846 NinePixelShader9_new(device, &ps, NULL, nine_ff_build_ps(device, &key));
1847
1848 nine_ff_prune_ps(device);
1849 if (ps) {
1850 memcpy(&ps->ff_key, &key, sizeof(ps->ff_key));
1851
1852 _mesa_hash_table_insert(device->ff.ht_ps, &ps->ff_key, ps);
1853 device->ff.num_ps++;
1854
1855 ps->rt_mask = 0x1;
1856 ps->sampler_mask = sampler_mask;
1857 }
1858 return ps;
1859 }
1860
1861 static void
nine_ff_load_vs_transforms(struct NineDevice9 * device)1862 nine_ff_load_vs_transforms(struct NineDevice9 *device)
1863 {
1864 struct nine_context *context = &device->context;
1865 D3DMATRIX T;
1866 D3DMATRIX *M = (D3DMATRIX *)device->ff.vs_const;
1867 unsigned i;
1868
1869 /* TODO: make this nicer, and only upload the ones we need */
1870 /* TODO: use ff.vs_const as storage of W, V, P matrices */
1871
1872 if (IS_D3DTS_DIRTY(context, WORLD) ||
1873 IS_D3DTS_DIRTY(context, VIEW) ||
1874 IS_D3DTS_DIRTY(context, PROJECTION)) {
1875 /* WVP, WV matrices */
1876 nine_d3d_matrix_matrix_mul(&M[1], GET_D3DTS(WORLD), GET_D3DTS(VIEW));
1877 nine_d3d_matrix_matrix_mul(&M[0], &M[1], GET_D3DTS(PROJECTION));
1878
1879 /* normal matrix == transpose(inverse(WV)) */
1880 nine_d3d_matrix_inverse(&T, &M[1]);
1881 nine_d3d_matrix_transpose(&M[4], &T);
1882
1883 /* P matrix */
1884 M[2] = *GET_D3DTS(PROJECTION);
1885
1886 /* V and W matrix */
1887 nine_d3d_matrix_inverse(&M[3], GET_D3DTS(VIEW));
1888 M[40] = M[1];
1889 }
1890
1891 if (context->rs[D3DRS_VERTEXBLEND] != D3DVBF_DISABLE) {
1892 /* load other world matrices */
1893 for (i = 1; i <= 8; ++i) {
1894 nine_d3d_matrix_matrix_mul(&M[40 + i], GET_D3DTS(WORLDMATRIX(i)), GET_D3DTS(VIEW));
1895 }
1896 }
1897
1898 device->ff.vs_const[30 * 4] = asfloat(context->rs[D3DRS_TWEENFACTOR]);
1899 }
1900
1901 static void
nine_ff_load_lights(struct NineDevice9 * device)1902 nine_ff_load_lights(struct NineDevice9 *device)
1903 {
1904 struct nine_context *context = &device->context;
1905 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1906 unsigned l;
1907
1908 if (context->changed.group & NINE_STATE_FF_MATERIAL) {
1909 const D3DMATERIAL9 *mtl = &context->ff.material;
1910
1911 memcpy(&dst[20], &mtl->Diffuse, 4 * sizeof(float));
1912 memcpy(&dst[21], &mtl->Ambient, 4 * sizeof(float));
1913 memcpy(&dst[22], &mtl->Specular, 4 * sizeof(float));
1914 dst[23].x = mtl->Power;
1915 memcpy(&dst[24], &mtl->Emissive, 4 * sizeof(float));
1916 d3dcolor_to_rgba(&dst[25].x, context->rs[D3DRS_AMBIENT]);
1917 dst[19].x = dst[25].x * mtl->Ambient.r + mtl->Emissive.r;
1918 dst[19].y = dst[25].y * mtl->Ambient.g + mtl->Emissive.g;
1919 dst[19].z = dst[25].z * mtl->Ambient.b + mtl->Emissive.b;
1920 }
1921
1922 if (!(context->changed.group & NINE_STATE_FF_LIGHTING))
1923 return;
1924
1925 for (l = 0; l < context->ff.num_lights_active; ++l) {
1926 const D3DLIGHT9 *light = &context->ff.light[context->ff.active_light[l]];
1927
1928 dst[32 + l * 8].x = light->Type;
1929 dst[32 + l * 8].y = light->Attenuation0;
1930 dst[32 + l * 8].z = light->Attenuation1;
1931 dst[32 + l * 8].w = light->Attenuation2;
1932 memcpy(&dst[33 + l * 8].x, &light->Diffuse, sizeof(light->Diffuse));
1933 memcpy(&dst[34 + l * 8].x, &light->Specular, sizeof(light->Specular));
1934 memcpy(&dst[35 + l * 8].x, &light->Ambient, sizeof(light->Ambient));
1935 nine_d3d_vector4_matrix_mul((D3DVECTOR *)&dst[36 + l * 8].x, &light->Position, GET_D3DTS(VIEW));
1936 nine_d3d_vector3_matrix_mul((D3DVECTOR *)&dst[37 + l * 8].x, &light->Direction, GET_D3DTS(VIEW));
1937 dst[36 + l * 8].w = light->Type == D3DLIGHT_DIRECTIONAL ? 1e9f : light->Range;
1938 dst[37 + l * 8].w = light->Falloff;
1939 dst[38 + l * 8].x = cosf(light->Theta * 0.5f);
1940 dst[38 + l * 8].y = cosf(light->Phi * 0.5f);
1941 dst[38 + l * 8].z = 1.0f / (dst[38 + l * 8].x - dst[38 + l * 8].y);
1942 dst[39 + l * 8].w = (float)((l + 1) == context->ff.num_lights_active);
1943 }
1944 }
1945
1946 static void
nine_ff_load_point_and_fog_params(struct NineDevice9 * device)1947 nine_ff_load_point_and_fog_params(struct NineDevice9 *device)
1948 {
1949 struct nine_context *context = &device->context;
1950 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
1951
1952 if (!(context->changed.group & NINE_STATE_FF_VS_OTHER))
1953 return;
1954 dst[26].x = asfloat(context->rs[D3DRS_POINTSIZE_MIN]);
1955 dst[26].y = asfloat(context->rs[D3DRS_POINTSIZE_MAX]);
1956 dst[26].z = asfloat(context->rs[D3DRS_POINTSIZE]);
1957 dst[26].w = asfloat(context->rs[D3DRS_POINTSCALE_A]);
1958 dst[27].x = asfloat(context->rs[D3DRS_POINTSCALE_B]);
1959 dst[27].y = asfloat(context->rs[D3DRS_POINTSCALE_C]);
1960 dst[28].x = asfloat(context->rs[D3DRS_FOGEND]);
1961 dst[28].y = 1.0f / (asfloat(context->rs[D3DRS_FOGEND]) - asfloat(context->rs[D3DRS_FOGSTART]));
1962 if (isinf(dst[28].y))
1963 dst[28].y = 0.0f;
1964 dst[28].z = asfloat(context->rs[D3DRS_FOGDENSITY]);
1965 }
1966
1967 static void
nine_ff_load_tex_matrices(struct NineDevice9 * device)1968 nine_ff_load_tex_matrices(struct NineDevice9 *device)
1969 {
1970 struct nine_context *context = &device->context;
1971 D3DMATRIX *M = (D3DMATRIX *)device->ff.vs_const;
1972 unsigned s;
1973
1974 if (!(context->ff.changed.transform[0] & 0xff0000))
1975 return;
1976 for (s = 0; s < 8; ++s) {
1977 if (IS_D3DTS_DIRTY(context, TEXTURE0 + s))
1978 nine_d3d_matrix_transpose(&M[32 + s], nine_state_access_transform(&context->ff, D3DTS_TEXTURE0 + s, FALSE));
1979 }
1980 }
1981
1982 static void
nine_ff_load_ps_params(struct NineDevice9 * device)1983 nine_ff_load_ps_params(struct NineDevice9 *device)
1984 {
1985 struct nine_context *context = &device->context;
1986 struct fvec4 *dst = (struct fvec4 *)device->ff.ps_const;
1987 unsigned s;
1988
1989 if (!(context->changed.group & NINE_STATE_FF_PS_CONSTS))
1990 return;
1991
1992 for (s = 0; s < 8; ++s)
1993 d3dcolor_to_rgba(&dst[s].x, context->ff.tex_stage[s][D3DTSS_CONSTANT]);
1994
1995 for (s = 0; s < 8; ++s) {
1996 dst[8 + s].x = asfloat(context->ff.tex_stage[s][D3DTSS_BUMPENVMAT00]);
1997 dst[8 + s].y = asfloat(context->ff.tex_stage[s][D3DTSS_BUMPENVMAT01]);
1998 dst[8 + s].z = asfloat(context->ff.tex_stage[s][D3DTSS_BUMPENVMAT10]);
1999 dst[8 + s].w = asfloat(context->ff.tex_stage[s][D3DTSS_BUMPENVMAT11]);
2000 if (s & 1) {
2001 dst[16 + s / 2].z = asfloat(context->ff.tex_stage[s][D3DTSS_BUMPENVLSCALE]);
2002 dst[16 + s / 2].w = asfloat(context->ff.tex_stage[s][D3DTSS_BUMPENVLOFFSET]);
2003 } else {
2004 dst[16 + s / 2].x = asfloat(context->ff.tex_stage[s][D3DTSS_BUMPENVLSCALE]);
2005 dst[16 + s / 2].y = asfloat(context->ff.tex_stage[s][D3DTSS_BUMPENVLOFFSET]);
2006 }
2007 }
2008
2009 d3dcolor_to_rgba(&dst[20].x, context->rs[D3DRS_TEXTUREFACTOR]);
2010 d3dcolor_to_rgba(&dst[21].x, context->rs[D3DRS_FOGCOLOR]);
2011 dst[22].x = asfloat(context->rs[D3DRS_FOGEND]);
2012 dst[22].y = 1.0f / (asfloat(context->rs[D3DRS_FOGEND]) - asfloat(context->rs[D3DRS_FOGSTART]));
2013 dst[22].z = asfloat(context->rs[D3DRS_FOGDENSITY]);
2014 }
2015
2016 static void
nine_ff_load_viewport_info(struct NineDevice9 * device)2017 nine_ff_load_viewport_info(struct NineDevice9 *device)
2018 {
2019 D3DVIEWPORT9 *viewport = &device->context.viewport;
2020 struct fvec4 *dst = (struct fvec4 *)device->ff.vs_const;
2021 float diffZ = viewport->MaxZ - viewport->MinZ;
2022
2023 /* Note: the other functions avoids to fill the const again if nothing changed.
2024 * But we don't have much to fill, and adding code to allow that may be complex
2025 * so just fill it always */
2026 dst[100].x = 2.0f / (float)(viewport->Width);
2027 dst[100].y = 2.0f / (float)(viewport->Height);
2028 dst[100].z = (diffZ == 0.0f) ? 0.0f : (1.0f / diffZ);
2029 dst[100].w = (float)(viewport->Width);
2030 dst[101].x = (float)(viewport->X);
2031 dst[101].y = (float)(viewport->Y);
2032 dst[101].z = (float)(viewport->MinZ);
2033 }
2034
2035 void
nine_ff_update(struct NineDevice9 * device)2036 nine_ff_update(struct NineDevice9 *device)
2037 {
2038 struct nine_context *context = &device->context;
2039 struct pipe_constant_buffer cb;
2040
2041 DBG("vs=%p ps=%p\n", context->vs, context->ps);
2042
2043 /* NOTE: the only reference belongs to the hash table */
2044 if (!context->programmable_vs) {
2045 device->ff.vs = nine_ff_get_vs(device);
2046 context->changed.group |= NINE_STATE_VS;
2047 }
2048 if (!context->ps) {
2049 device->ff.ps = nine_ff_get_ps(device);
2050 context->changed.group |= NINE_STATE_PS;
2051 }
2052
2053 if (!context->programmable_vs) {
2054 nine_ff_load_vs_transforms(device);
2055 nine_ff_load_tex_matrices(device);
2056 nine_ff_load_lights(device);
2057 nine_ff_load_point_and_fog_params(device);
2058 nine_ff_load_viewport_info(device);
2059
2060 memset(context->ff.changed.transform, 0, sizeof(context->ff.changed.transform));
2061
2062 cb.buffer_offset = 0;
2063 cb.buffer = NULL;
2064 cb.user_buffer = device->ff.vs_const;
2065 cb.buffer_size = NINE_FF_NUM_VS_CONST * 4 * sizeof(float);
2066
2067 context->pipe_data.cb_vs_ff = cb;
2068 context->commit |= NINE_STATE_COMMIT_CONST_VS;
2069
2070 context->changed.group &= ~NINE_STATE_FF_VS;
2071 }
2072
2073 if (!context->ps) {
2074 nine_ff_load_ps_params(device);
2075
2076 cb.buffer_offset = 0;
2077 cb.buffer = NULL;
2078 cb.user_buffer = device->ff.ps_const;
2079 cb.buffer_size = NINE_FF_NUM_PS_CONST * 4 * sizeof(float);
2080
2081 context->pipe_data.cb_ps_ff = cb;
2082 context->commit |= NINE_STATE_COMMIT_CONST_PS;
2083
2084 context->changed.group &= ~NINE_STATE_FF_PS;
2085 }
2086 }
2087
2088
2089 boolean
nine_ff_init(struct NineDevice9 * device)2090 nine_ff_init(struct NineDevice9 *device)
2091 {
2092 device->ff.ht_vs = _mesa_hash_table_create(NULL, nine_ff_vs_key_hash,
2093 nine_ff_vs_key_comp);
2094 device->ff.ht_ps = _mesa_hash_table_create(NULL, nine_ff_ps_key_hash,
2095 nine_ff_ps_key_comp);
2096
2097 device->ff.ht_fvf = _mesa_hash_table_create(NULL, nine_ff_fvf_key_hash,
2098 nine_ff_fvf_key_comp);
2099
2100 device->ff.vs_const = CALLOC(NINE_FF_NUM_VS_CONST, 4 * sizeof(float));
2101 device->ff.ps_const = CALLOC(NINE_FF_NUM_PS_CONST, 4 * sizeof(float));
2102
2103 return device->ff.ht_vs && device->ff.ht_ps &&
2104 device->ff.ht_fvf &&
2105 device->ff.vs_const && device->ff.ps_const;
2106 }
2107
nine_ff_ht_delete_cb(void * key,void * value,void * data)2108 static enum pipe_error nine_ff_ht_delete_cb(void *key, void *value, void *data)
2109 {
2110 NineUnknown_Unbind(NineUnknown(value));
2111 return PIPE_OK;
2112 }
2113
2114 void
nine_ff_fini(struct NineDevice9 * device)2115 nine_ff_fini(struct NineDevice9 *device)
2116 {
2117 if (device->ff.ht_vs) {
2118 util_hash_table_foreach(device->ff.ht_vs, nine_ff_ht_delete_cb, NULL);
2119 _mesa_hash_table_destroy(device->ff.ht_vs, NULL);
2120 }
2121 if (device->ff.ht_ps) {
2122 util_hash_table_foreach(device->ff.ht_ps, nine_ff_ht_delete_cb, NULL);
2123 _mesa_hash_table_destroy(device->ff.ht_ps, NULL);
2124 }
2125 if (device->ff.ht_fvf) {
2126 util_hash_table_foreach(device->ff.ht_fvf, nine_ff_ht_delete_cb, NULL);
2127 _mesa_hash_table_destroy(device->ff.ht_fvf, NULL);
2128 }
2129 device->ff.vs = NULL; /* destroyed by unbinding from hash table */
2130 device->ff.ps = NULL;
2131
2132 FREE(device->ff.vs_const);
2133 FREE(device->ff.ps_const);
2134 }
2135
2136 static void
nine_ff_prune_vs(struct NineDevice9 * device)2137 nine_ff_prune_vs(struct NineDevice9 *device)
2138 {
2139 struct nine_context *context = &device->context;
2140
2141 if (device->ff.num_vs > 1024) {
2142 /* could destroy the bound one here, so unbind */
2143 context->pipe->bind_vs_state(context->pipe, NULL);
2144 util_hash_table_foreach(device->ff.ht_vs, nine_ff_ht_delete_cb, NULL);
2145 _mesa_hash_table_clear(device->ff.ht_vs, NULL);
2146 device->ff.num_vs = 0;
2147 context->changed.group |= NINE_STATE_VS;
2148 }
2149 }
2150 static void
nine_ff_prune_ps(struct NineDevice9 * device)2151 nine_ff_prune_ps(struct NineDevice9 *device)
2152 {
2153 struct nine_context *context = &device->context;
2154
2155 if (device->ff.num_ps > 1024) {
2156 /* could destroy the bound one here, so unbind */
2157 context->pipe->bind_fs_state(context->pipe, NULL);
2158 util_hash_table_foreach(device->ff.ht_ps, nine_ff_ht_delete_cb, NULL);
2159 _mesa_hash_table_clear(device->ff.ht_ps, NULL);
2160 device->ff.num_ps = 0;
2161 context->changed.group |= NINE_STATE_PS;
2162 }
2163 }
2164
2165 /* ========================================================================== */
2166
2167 /* Matrix multiplication:
2168 *
2169 * in memory: 0 1 2 3 (row major)
2170 * 4 5 6 7
2171 * 8 9 a b
2172 * c d e f
2173 *
2174 * cA cB cC cD
2175 * r0 = (r0 * cA) (r0 * cB) . .
2176 * r1 = (r1 * cA) (r1 * cB)
2177 * r2 = (r2 * cA) .
2178 * r3 = (r3 * cA) .
2179 *
2180 * r: (11) (12) (13) (14)
2181 * (21) (22) (23) (24)
2182 * (31) (32) (33) (34)
2183 * (41) (42) (43) (44)
2184 * l: (11 12 13 14)
2185 * (21 22 23 24)
2186 * (31 32 33 34)
2187 * (41 42 43 44)
2188 *
2189 * v: (x y z 1 )
2190 *
2191 * t.xyzw = MUL(v.xxxx, r[0]);
2192 * t.xyzw = MAD(v.yyyy, r[1], t.xyzw);
2193 * t.xyzw = MAD(v.zzzz, r[2], t.xyzw);
2194 * v.xyzw = MAD(v.wwww, r[3], t.xyzw);
2195 *
2196 * v.x = DP4(v, c[0]);
2197 * v.y = DP4(v, c[1]);
2198 * v.z = DP4(v, c[2]);
2199 * v.w = DP4(v, c[3]) = 1
2200 */
2201
2202 /*
2203 static void
2204 nine_D3DMATRIX_print(const D3DMATRIX *M)
2205 {
2206 DBG("\n(%f %f %f %f)\n"
2207 "(%f %f %f %f)\n"
2208 "(%f %f %f %f)\n"
2209 "(%f %f %f %f)\n",
2210 M->m[0][0], M->m[0][1], M->m[0][2], M->m[0][3],
2211 M->m[1][0], M->m[1][1], M->m[1][2], M->m[1][3],
2212 M->m[2][0], M->m[2][1], M->m[2][2], M->m[2][3],
2213 M->m[3][0], M->m[3][1], M->m[3][2], M->m[3][3]);
2214 }
2215 */
2216
2217 static inline float
nine_DP4_row_col(const D3DMATRIX * A,int r,const D3DMATRIX * B,int c)2218 nine_DP4_row_col(const D3DMATRIX *A, int r, const D3DMATRIX *B, int c)
2219 {
2220 return A->m[r][0] * B->m[0][c] +
2221 A->m[r][1] * B->m[1][c] +
2222 A->m[r][2] * B->m[2][c] +
2223 A->m[r][3] * B->m[3][c];
2224 }
2225
2226 static inline float
nine_DP4_vec_col(const D3DVECTOR * v,const D3DMATRIX * M,int c)2227 nine_DP4_vec_col(const D3DVECTOR *v, const D3DMATRIX *M, int c)
2228 {
2229 return v->x * M->m[0][c] +
2230 v->y * M->m[1][c] +
2231 v->z * M->m[2][c] +
2232 1.0f * M->m[3][c];
2233 }
2234
2235 static inline float
nine_DP3_vec_col(const D3DVECTOR * v,const D3DMATRIX * M,int c)2236 nine_DP3_vec_col(const D3DVECTOR *v, const D3DMATRIX *M, int c)
2237 {
2238 return v->x * M->m[0][c] +
2239 v->y * M->m[1][c] +
2240 v->z * M->m[2][c];
2241 }
2242
2243 void
nine_d3d_matrix_matrix_mul(D3DMATRIX * D,const D3DMATRIX * L,const D3DMATRIX * R)2244 nine_d3d_matrix_matrix_mul(D3DMATRIX *D, const D3DMATRIX *L, const D3DMATRIX *R)
2245 {
2246 D->_11 = nine_DP4_row_col(L, 0, R, 0);
2247 D->_12 = nine_DP4_row_col(L, 0, R, 1);
2248 D->_13 = nine_DP4_row_col(L, 0, R, 2);
2249 D->_14 = nine_DP4_row_col(L, 0, R, 3);
2250
2251 D->_21 = nine_DP4_row_col(L, 1, R, 0);
2252 D->_22 = nine_DP4_row_col(L, 1, R, 1);
2253 D->_23 = nine_DP4_row_col(L, 1, R, 2);
2254 D->_24 = nine_DP4_row_col(L, 1, R, 3);
2255
2256 D->_31 = nine_DP4_row_col(L, 2, R, 0);
2257 D->_32 = nine_DP4_row_col(L, 2, R, 1);
2258 D->_33 = nine_DP4_row_col(L, 2, R, 2);
2259 D->_34 = nine_DP4_row_col(L, 2, R, 3);
2260
2261 D->_41 = nine_DP4_row_col(L, 3, R, 0);
2262 D->_42 = nine_DP4_row_col(L, 3, R, 1);
2263 D->_43 = nine_DP4_row_col(L, 3, R, 2);
2264 D->_44 = nine_DP4_row_col(L, 3, R, 3);
2265 }
2266
2267 void
nine_d3d_vector4_matrix_mul(D3DVECTOR * d,const D3DVECTOR * v,const D3DMATRIX * M)2268 nine_d3d_vector4_matrix_mul(D3DVECTOR *d, const D3DVECTOR *v, const D3DMATRIX *M)
2269 {
2270 d->x = nine_DP4_vec_col(v, M, 0);
2271 d->y = nine_DP4_vec_col(v, M, 1);
2272 d->z = nine_DP4_vec_col(v, M, 2);
2273 }
2274
2275 void
nine_d3d_vector3_matrix_mul(D3DVECTOR * d,const D3DVECTOR * v,const D3DMATRIX * M)2276 nine_d3d_vector3_matrix_mul(D3DVECTOR *d, const D3DVECTOR *v, const D3DMATRIX *M)
2277 {
2278 d->x = nine_DP3_vec_col(v, M, 0);
2279 d->y = nine_DP3_vec_col(v, M, 1);
2280 d->z = nine_DP3_vec_col(v, M, 2);
2281 }
2282
2283 void
nine_d3d_matrix_transpose(D3DMATRIX * D,const D3DMATRIX * M)2284 nine_d3d_matrix_transpose(D3DMATRIX *D, const D3DMATRIX *M)
2285 {
2286 unsigned i, j;
2287 for (i = 0; i < 4; ++i)
2288 for (j = 0; j < 4; ++j)
2289 D->m[i][j] = M->m[j][i];
2290 }
2291
2292 #define _M_ADD_PROD_1i_2j_3k_4l(i,j,k,l) do { \
2293 float t = M->_1##i * M->_2##j * M->_3##k * M->_4##l; \
2294 if (t > 0.0f) pos += t; else neg += t; } while(0)
2295
2296 #define _M_SUB_PROD_1i_2j_3k_4l(i,j,k,l) do { \
2297 float t = M->_1##i * M->_2##j * M->_3##k * M->_4##l; \
2298 if (t > 0.0f) neg -= t; else pos -= t; } while(0)
2299 float
nine_d3d_matrix_det(const D3DMATRIX * M)2300 nine_d3d_matrix_det(const D3DMATRIX *M)
2301 {
2302 float pos = 0.0f;
2303 float neg = 0.0f;
2304
2305 _M_ADD_PROD_1i_2j_3k_4l(1, 2, 3, 4);
2306 _M_ADD_PROD_1i_2j_3k_4l(1, 3, 4, 2);
2307 _M_ADD_PROD_1i_2j_3k_4l(1, 4, 2, 3);
2308
2309 _M_ADD_PROD_1i_2j_3k_4l(2, 1, 4, 3);
2310 _M_ADD_PROD_1i_2j_3k_4l(2, 3, 1, 4);
2311 _M_ADD_PROD_1i_2j_3k_4l(2, 4, 3, 1);
2312
2313 _M_ADD_PROD_1i_2j_3k_4l(3, 1, 2, 4);
2314 _M_ADD_PROD_1i_2j_3k_4l(3, 2, 4, 1);
2315 _M_ADD_PROD_1i_2j_3k_4l(3, 4, 1, 2);
2316
2317 _M_ADD_PROD_1i_2j_3k_4l(4, 1, 3, 2);
2318 _M_ADD_PROD_1i_2j_3k_4l(4, 2, 1, 3);
2319 _M_ADD_PROD_1i_2j_3k_4l(4, 3, 2, 1);
2320
2321 _M_SUB_PROD_1i_2j_3k_4l(1, 2, 4, 3);
2322 _M_SUB_PROD_1i_2j_3k_4l(1, 3, 2, 4);
2323 _M_SUB_PROD_1i_2j_3k_4l(1, 4, 3, 2);
2324
2325 _M_SUB_PROD_1i_2j_3k_4l(2, 1, 3, 4);
2326 _M_SUB_PROD_1i_2j_3k_4l(2, 3, 4, 1);
2327 _M_SUB_PROD_1i_2j_3k_4l(2, 4, 1, 3);
2328
2329 _M_SUB_PROD_1i_2j_3k_4l(3, 1, 4, 2);
2330 _M_SUB_PROD_1i_2j_3k_4l(3, 2, 1, 4);
2331 _M_SUB_PROD_1i_2j_3k_4l(3, 4, 2, 1);
2332
2333 _M_SUB_PROD_1i_2j_3k_4l(4, 1, 2, 3);
2334 _M_SUB_PROD_1i_2j_3k_4l(4, 2, 3, 1);
2335 _M_SUB_PROD_1i_2j_3k_4l(4, 3, 1, 2);
2336
2337 return pos + neg;
2338 }
2339
2340 /* XXX: Probably better to just use src/mesa/math/m_matrix.c because
2341 * I have no idea where this code came from.
2342 */
2343 void
nine_d3d_matrix_inverse(D3DMATRIX * D,const D3DMATRIX * M)2344 nine_d3d_matrix_inverse(D3DMATRIX *D, const D3DMATRIX *M)
2345 {
2346 int i, k;
2347 float det;
2348
2349 D->m[0][0] =
2350 M->m[1][1] * M->m[2][2] * M->m[3][3] -
2351 M->m[1][1] * M->m[3][2] * M->m[2][3] -
2352 M->m[1][2] * M->m[2][1] * M->m[3][3] +
2353 M->m[1][2] * M->m[3][1] * M->m[2][3] +
2354 M->m[1][3] * M->m[2][1] * M->m[3][2] -
2355 M->m[1][3] * M->m[3][1] * M->m[2][2];
2356
2357 D->m[0][1] =
2358 -M->m[0][1] * M->m[2][2] * M->m[3][3] +
2359 M->m[0][1] * M->m[3][2] * M->m[2][3] +
2360 M->m[0][2] * M->m[2][1] * M->m[3][3] -
2361 M->m[0][2] * M->m[3][1] * M->m[2][3] -
2362 M->m[0][3] * M->m[2][1] * M->m[3][2] +
2363 M->m[0][3] * M->m[3][1] * M->m[2][2];
2364
2365 D->m[0][2] =
2366 M->m[0][1] * M->m[1][2] * M->m[3][3] -
2367 M->m[0][1] * M->m[3][2] * M->m[1][3] -
2368 M->m[0][2] * M->m[1][1] * M->m[3][3] +
2369 M->m[0][2] * M->m[3][1] * M->m[1][3] +
2370 M->m[0][3] * M->m[1][1] * M->m[3][2] -
2371 M->m[0][3] * M->m[3][1] * M->m[1][2];
2372
2373 D->m[0][3] =
2374 -M->m[0][1] * M->m[1][2] * M->m[2][3] +
2375 M->m[0][1] * M->m[2][2] * M->m[1][3] +
2376 M->m[0][2] * M->m[1][1] * M->m[2][3] -
2377 M->m[0][2] * M->m[2][1] * M->m[1][3] -
2378 M->m[0][3] * M->m[1][1] * M->m[2][2] +
2379 M->m[0][3] * M->m[2][1] * M->m[1][2];
2380
2381 D->m[1][0] =
2382 -M->m[1][0] * M->m[2][2] * M->m[3][3] +
2383 M->m[1][0] * M->m[3][2] * M->m[2][3] +
2384 M->m[1][2] * M->m[2][0] * M->m[3][3] -
2385 M->m[1][2] * M->m[3][0] * M->m[2][3] -
2386 M->m[1][3] * M->m[2][0] * M->m[3][2] +
2387 M->m[1][3] * M->m[3][0] * M->m[2][2];
2388
2389 D->m[1][1] =
2390 M->m[0][0] * M->m[2][2] * M->m[3][3] -
2391 M->m[0][0] * M->m[3][2] * M->m[2][3] -
2392 M->m[0][2] * M->m[2][0] * M->m[3][3] +
2393 M->m[0][2] * M->m[3][0] * M->m[2][3] +
2394 M->m[0][3] * M->m[2][0] * M->m[3][2] -
2395 M->m[0][3] * M->m[3][0] * M->m[2][2];
2396
2397 D->m[1][2] =
2398 -M->m[0][0] * M->m[1][2] * M->m[3][3] +
2399 M->m[0][0] * M->m[3][2] * M->m[1][3] +
2400 M->m[0][2] * M->m[1][0] * M->m[3][3] -
2401 M->m[0][2] * M->m[3][0] * M->m[1][3] -
2402 M->m[0][3] * M->m[1][0] * M->m[3][2] +
2403 M->m[0][3] * M->m[3][0] * M->m[1][2];
2404
2405 D->m[1][3] =
2406 M->m[0][0] * M->m[1][2] * M->m[2][3] -
2407 M->m[0][0] * M->m[2][2] * M->m[1][3] -
2408 M->m[0][2] * M->m[1][0] * M->m[2][3] +
2409 M->m[0][2] * M->m[2][0] * M->m[1][3] +
2410 M->m[0][3] * M->m[1][0] * M->m[2][2] -
2411 M->m[0][3] * M->m[2][0] * M->m[1][2];
2412
2413 D->m[2][0] =
2414 M->m[1][0] * M->m[2][1] * M->m[3][3] -
2415 M->m[1][0] * M->m[3][1] * M->m[2][3] -
2416 M->m[1][1] * M->m[2][0] * M->m[3][3] +
2417 M->m[1][1] * M->m[3][0] * M->m[2][3] +
2418 M->m[1][3] * M->m[2][0] * M->m[3][1] -
2419 M->m[1][3] * M->m[3][0] * M->m[2][1];
2420
2421 D->m[2][1] =
2422 -M->m[0][0] * M->m[2][1] * M->m[3][3] +
2423 M->m[0][0] * M->m[3][1] * M->m[2][3] +
2424 M->m[0][1] * M->m[2][0] * M->m[3][3] -
2425 M->m[0][1] * M->m[3][0] * M->m[2][3] -
2426 M->m[0][3] * M->m[2][0] * M->m[3][1] +
2427 M->m[0][3] * M->m[3][0] * M->m[2][1];
2428
2429 D->m[2][2] =
2430 M->m[0][0] * M->m[1][1] * M->m[3][3] -
2431 M->m[0][0] * M->m[3][1] * M->m[1][3] -
2432 M->m[0][1] * M->m[1][0] * M->m[3][3] +
2433 M->m[0][1] * M->m[3][0] * M->m[1][3] +
2434 M->m[0][3] * M->m[1][0] * M->m[3][1] -
2435 M->m[0][3] * M->m[3][0] * M->m[1][1];
2436
2437 D->m[2][3] =
2438 -M->m[0][0] * M->m[1][1] * M->m[2][3] +
2439 M->m[0][0] * M->m[2][1] * M->m[1][3] +
2440 M->m[0][1] * M->m[1][0] * M->m[2][3] -
2441 M->m[0][1] * M->m[2][0] * M->m[1][3] -
2442 M->m[0][3] * M->m[1][0] * M->m[2][1] +
2443 M->m[0][3] * M->m[2][0] * M->m[1][1];
2444
2445 D->m[3][0] =
2446 -M->m[1][0] * M->m[2][1] * M->m[3][2] +
2447 M->m[1][0] * M->m[3][1] * M->m[2][2] +
2448 M->m[1][1] * M->m[2][0] * M->m[3][2] -
2449 M->m[1][1] * M->m[3][0] * M->m[2][2] -
2450 M->m[1][2] * M->m[2][0] * M->m[3][1] +
2451 M->m[1][2] * M->m[3][0] * M->m[2][1];
2452
2453 D->m[3][1] =
2454 M->m[0][0] * M->m[2][1] * M->m[3][2] -
2455 M->m[0][0] * M->m[3][1] * M->m[2][2] -
2456 M->m[0][1] * M->m[2][0] * M->m[3][2] +
2457 M->m[0][1] * M->m[3][0] * M->m[2][2] +
2458 M->m[0][2] * M->m[2][0] * M->m[3][1] -
2459 M->m[0][2] * M->m[3][0] * M->m[2][1];
2460
2461 D->m[3][2] =
2462 -M->m[0][0] * M->m[1][1] * M->m[3][2] +
2463 M->m[0][0] * M->m[3][1] * M->m[1][2] +
2464 M->m[0][1] * M->m[1][0] * M->m[3][2] -
2465 M->m[0][1] * M->m[3][0] * M->m[1][2] -
2466 M->m[0][2] * M->m[1][0] * M->m[3][1] +
2467 M->m[0][2] * M->m[3][0] * M->m[1][1];
2468
2469 D->m[3][3] =
2470 M->m[0][0] * M->m[1][1] * M->m[2][2] -
2471 M->m[0][0] * M->m[2][1] * M->m[1][2] -
2472 M->m[0][1] * M->m[1][0] * M->m[2][2] +
2473 M->m[0][1] * M->m[2][0] * M->m[1][2] +
2474 M->m[0][2] * M->m[1][0] * M->m[2][1] -
2475 M->m[0][2] * M->m[2][0] * M->m[1][1];
2476
2477 det =
2478 M->m[0][0] * D->m[0][0] +
2479 M->m[1][0] * D->m[0][1] +
2480 M->m[2][0] * D->m[0][2] +
2481 M->m[3][0] * D->m[0][3];
2482
2483 if (fabsf(det) < 1e-30) {/* non inversible */
2484 *D = *M; /* wine tests */
2485 return;
2486 }
2487
2488 det = 1.0 / det;
2489
2490 for (i = 0; i < 4; i++)
2491 for (k = 0; k < 4; k++)
2492 D->m[i][k] *= det;
2493
2494 #if defined(DEBUG) || !defined(NDEBUG)
2495 {
2496 D3DMATRIX I;
2497
2498 nine_d3d_matrix_matrix_mul(&I, D, M);
2499
2500 for (i = 0; i < 4; ++i)
2501 for (k = 0; k < 4; ++k)
2502 if (fabsf(I.m[i][k] - (float)(i == k)) > 1e-3)
2503 DBG("Matrix inversion check FAILED !\n");
2504 }
2505 #endif
2506 }
2507