?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
wined3d: Store the current GLSL program in the backend's private data.
[wine/hacks.git] / dlls / wined3d / glsl_shader.c
blob34b86a3b13f5a3f81072dff84bd1e3fda23049f2
1 /*
2 * GLSL pixel and vertex shader implementation
3 *
4 * Copyright 2006 Jason Green
5 * Copyright 2006-2007 Henri Verbeet
6 * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
7 *
8 * This library is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * This library is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with this library; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
21 */
22
23 /*
24 * D3D shader asm has swizzles on source parameters, and write masks for
25 * destination parameters. GLSL uses swizzles for both. The result of this is
26 * that for example "mov dst.xw, src.zyxw" becomes "dst.xw = src.zw" in GLSL.
27 * Ie, to generate a proper GLSL source swizzle, we need to take the D3D write
28 * mask for the destination parameter into account.
29 */
30
31 #include "config.h"
32 #include <stdio.h>
33 #include "wined3d_private.h"
34
35 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
36 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
37 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
38
39 #define GLINFO_LOCATION (*gl_info)
40
41 typedef struct {
42 char reg_name[150];
43 char mask_str[6];
44 } glsl_dst_param_t;
45
46 typedef struct {
47 char reg_name[150];
48 char param_str[100];
49 } glsl_src_param_t;
50
51 typedef struct {
52 const char *name;
53 DWORD coord_mask;
54 } glsl_sample_function_t;
55
56 /** Prints the GLSL info log which will contain error messages if they exist */
57 void print_glsl_info_log(WineD3D_GL_Info *gl_info, GLhandleARB obj) {
58
59 int infologLength = 0;
60 char *infoLog;
61 int i;
62 BOOL is_spam;
63
64 const char *spam[] = {
65 "Vertex shader was successfully compiled to run on hardware.\n", /* fglrx */
66 "Fragment shader was successfully compiled to run on hardware.\n", /* fglrx */
67 "Fragment shader(s) linked, vertex shader(s) linked. \n ", /* fglrx, with \n */
68 "Fragment shader(s) linked, vertex shader(s) linked.", /* fglrx, no \n */
69 "Vertex shader(s) linked, no fragment shader(s) defined. \n ", /* fglrx, with \n */
70 "Vertex shader(s) linked, no fragment shader(s) defined.", /* fglrx, no \n */
71 "Fragment shader was successfully compiled to run on hardware.\nWARNING: 0:1: extension 'GL_ARB_draw_buffers' is not supported",
72 "Fragment shader(s) linked, no vertex shader(s) defined.", /* fglrx, no \n */
73 "Fragment shader(s) linked, no vertex shader(s) defined. \n " /* fglrx, with \n */
74 };
75
76 GL_EXTCALL(glGetObjectParameterivARB(obj,
77 GL_OBJECT_INFO_LOG_LENGTH_ARB,
78 &infologLength));
79
80 /* A size of 1 is just a null-terminated string, so the log should be bigger than
81 * that if there are errors. */
82 if (infologLength > 1)
83 {
84 /* Fglrx doesn't terminate the string properly, but it tells us the proper length.
85 * So use HEAP_ZERO_MEMORY to avoid uninitialized bytes
86 */
87 infoLog = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, infologLength);
88 GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
89 is_spam = FALSE;
90
91 for(i = 0; i < sizeof(spam) / sizeof(spam[0]); i++) {
92 if(strcmp(infoLog, spam[i]) == 0) {
93 is_spam = TRUE;
94 break;
95 }
96 }
97 if(is_spam) {
98 TRACE("Spam received from GLSL shader #%u: %s\n", obj, debugstr_a(infoLog));
99 } else {
100 FIXME("Error received from GLSL shader #%u: %s\n", obj, debugstr_a(infoLog));
101 }
102 HeapFree(GetProcessHeap(), 0, infoLog);
103 }
104 }
105
106 /**
107 * Loads (pixel shader) samplers
108 */
109 static void shader_glsl_load_psamplers(
110 WineD3D_GL_Info *gl_info,
111 IWineD3DStateBlock* iface,
112 GLhandleARB programId) {
113
114 IWineD3DStateBlockImpl* stateBlock = (IWineD3DStateBlockImpl*) iface;
115 GLhandleARB name_loc;
116 int i;
117 char sampler_name[20];
118
119 for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
120 snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
121 name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
122 if (name_loc != -1) {
123 int mapped_unit = stateBlock->wineD3DDevice->texUnitMap[i];
124 if (mapped_unit != -1 && mapped_unit < GL_LIMITS(fragment_samplers)) {
125 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
126 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
127 checkGLcall("glUniform1iARB");
128 } else {
129 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
130 }
131 }
132 }
133 }
134
135 static void shader_glsl_load_vsamplers(WineD3D_GL_Info *gl_info, IWineD3DStateBlock* iface, GLhandleARB programId) {
136 IWineD3DStateBlockImpl* stateBlock = (IWineD3DStateBlockImpl*) iface;
137 GLhandleARB name_loc;
138 char sampler_name[20];
139 int i;
140
141 for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
142 snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
143 name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
144 if (name_loc != -1) {
145 int mapped_unit = stateBlock->wineD3DDevice->texUnitMap[MAX_FRAGMENT_SAMPLERS + i];
146 if (mapped_unit != -1 && mapped_unit < GL_LIMITS(combined_samplers)) {
147 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
148 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
149 checkGLcall("glUniform1iARB");
150 } else {
151 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
152 }
153 }
154 }
155 }
156
157 /**
158 * Loads floating point constants (aka uniforms) into the currently set GLSL program.
159 * When constant_list == NULL, it will load all the constants.
160 */
161 static void shader_glsl_load_constantsF(IWineD3DBaseShaderImpl* This, WineD3D_GL_Info *gl_info,
162 unsigned int max_constants, float* constants, GLhandleARB *constant_locations,
163 struct list *constant_list) {
164 constants_entry *constant;
165 local_constant* lconst;
166 GLhandleARB tmp_loc;
167 DWORD i, j, k;
168 DWORD *idx;
169
170 if (TRACE_ON(d3d_shader)) {
171 LIST_FOR_EACH_ENTRY(constant, constant_list, constants_entry, entry) {
172 idx = constant->idx;
173 j = constant->count;
174 while (j--) {
175 i = *idx++;
176 tmp_loc = constant_locations[i];
177 if (tmp_loc != -1) {
178 TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
179 constants[i * 4 + 0], constants[i * 4 + 1],
180 constants[i * 4 + 2], constants[i * 4 + 3]);
181 }
182 }
183 }
184 }
185
186 /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
187 if(WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) == 1 &&
188 shader_is_pshader_version(This->baseShader.hex_version)) {
189 float lcl_const[4];
190
191 LIST_FOR_EACH_ENTRY(constant, constant_list, constants_entry, entry) {
192 idx = constant->idx;
193 j = constant->count;
194 while (j--) {
195 i = *idx++;
196 tmp_loc = constant_locations[i];
197 if (tmp_loc != -1) {
198 /* We found this uniform name in the program - go ahead and send the data */
199 k = i * 4;
200 if(constants[k + 0] < -1.0) lcl_const[0] = -1.0;
201 else if(constants[k + 0] > 1.0) lcl_const[0] = 1.0;
202 else lcl_const[0] = constants[k + 0];
203 if(constants[k + 1] < -1.0) lcl_const[1] = -1.0;
204 else if(constants[k + 1] > 1.0) lcl_const[1] = 1.0;
205 else lcl_const[1] = constants[k + 1];
206 if(constants[k + 2] < -1.0) lcl_const[2] = -1.0;
207 else if(constants[k + 2] > 1.0) lcl_const[2] = 1.0;
208 else lcl_const[2] = constants[k + 2];
209 if(constants[k + 3] < -1.0) lcl_const[3] = -1.0;
210 else if(constants[k + 3] > 1.0) lcl_const[3] = 1.0;
211 else lcl_const[3] = constants[k + 3];
212
213 GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, lcl_const));
214 }
215 }
216 }
217 } else {
218 LIST_FOR_EACH_ENTRY(constant, constant_list, constants_entry, entry) {
219 idx = constant->idx;
220 j = constant->count;
221 while (j--) {
222 i = *idx++;
223 tmp_loc = constant_locations[i];
224 if (tmp_loc != -1) {
225 /* We found this uniform name in the program - go ahead and send the data */
226 GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, constants + (i * 4)));
227 }
228 }
229 }
230 }
231 checkGLcall("glUniform4fvARB()");
232
233 if(!This->baseShader.load_local_constsF) {
234 TRACE("No need to load local float constants for this shader\n");
235 return;
236 }
237
238 /* Load immediate constants */
239 if (TRACE_ON(d3d_shader)) {
240 LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
241 tmp_loc = constant_locations[lconst->idx];
242 if (tmp_loc != -1) {
243 GLfloat* values = (GLfloat*)lconst->value;
244 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
245 values[0], values[1], values[2], values[3]);
246 }
247 }
248 }
249 /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
250 LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
251 tmp_loc = constant_locations[lconst->idx];
252 if (tmp_loc != -1) {
253 /* We found this uniform name in the program - go ahead and send the data */
254 GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, (GLfloat*)lconst->value));
255 }
256 }
257 checkGLcall("glUniform4fvARB()");
258 }
259
260 /**
261 * Loads integer constants (aka uniforms) into the currently set GLSL program.
262 * When @constants_set == NULL, it will load all the constants.
263 */
264 static void shader_glsl_load_constantsI(
265 IWineD3DBaseShaderImpl* This,
266 WineD3D_GL_Info *gl_info,
267 GLhandleARB programId,
268 GLhandleARB locations[MAX_CONST_I],
269 unsigned max_constants,
270 int* constants,
271 BOOL* constants_set) {
272
273 int i;
274 struct list* ptr;
275
276 for (i=0; i<max_constants; ++i) {
277 if (NULL == constants_set || constants_set[i]) {
278
279 TRACE_(d3d_constants)("Loading constants %i: %i, %i, %i, %i\n",
280 i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
281
282 /* We found this uniform name in the program - go ahead and send the data */
283 GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
284 checkGLcall("glUniform4ivARB");
285 }
286 }
287
288 /* Load immediate constants */
289 ptr = list_head(&This->baseShader.constantsI);
290 while (ptr) {
291 local_constant* lconst = LIST_ENTRY(ptr, struct local_constant, entry);
292 unsigned int idx = lconst->idx;
293 GLint* values = (GLint*) lconst->value;
294
295 TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
296 values[0], values[1], values[2], values[3]);
297
298 /* We found this uniform name in the program - go ahead and send the data */
299 GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
300 checkGLcall("glUniform4ivARB");
301 ptr = list_next(&This->baseShader.constantsI, ptr);
302 }
303 }
304
305 /**
306 * Loads boolean constants (aka uniforms) into the currently set GLSL program.
307 * When @constants_set == NULL, it will load all the constants.
308 */
309 static void shader_glsl_load_constantsB(
310 IWineD3DBaseShaderImpl* This,
311 WineD3D_GL_Info *gl_info,
312 GLhandleARB programId,
313 unsigned max_constants,
314 BOOL* constants,
315 BOOL* constants_set) {
316
317 GLhandleARB tmp_loc;
318 int i;
319 char tmp_name[8];
320 char is_pshader = shader_is_pshader_version(This->baseShader.hex_version);
321 const char* prefix = is_pshader? "PB":"VB";
322 struct list* ptr;
323
324 for (i=0; i<max_constants; ++i) {
325 if (NULL == constants_set || constants_set[i]) {
326
327 TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
328
329 /* TODO: Benchmark and see if it would be beneficial to store the
330 * locations of the constants to avoid looking up each time */
331 snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
332 tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
333 if (tmp_loc != -1) {
334 /* We found this uniform name in the program - go ahead and send the data */
335 GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
336 checkGLcall("glUniform1ivARB");
337 }
338 }
339 }
340
341 /* Load immediate constants */
342 ptr = list_head(&This->baseShader.constantsB);
343 while (ptr) {
344 local_constant* lconst = LIST_ENTRY(ptr, struct local_constant, entry);
345 unsigned int idx = lconst->idx;
346 GLint* values = (GLint*) lconst->value;
347
348 TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
349
350 snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
351 tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
352 if (tmp_loc != -1) {
353 /* We found this uniform name in the program - go ahead and send the data */
354 GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
355 checkGLcall("glUniform1ivARB");
356 }
357 ptr = list_next(&This->baseShader.constantsB, ptr);
358 }
359 }
360
361
362
363 /**
364 * Loads the app-supplied constants into the currently set GLSL program.
365 */
366 void shader_glsl_load_constants(
367 IWineD3DDevice* device,
368 char usePixelShader,
369 char useVertexShader) {
370
371 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) device;
372 struct shader_glsl_priv *priv = (struct shader_glsl_priv *)deviceImpl->shader_priv;
373 IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
374 WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
375
376 GLhandleARB *constant_locations;
377 struct list *constant_list;
378 GLhandleARB programId;
379 struct glsl_shader_prog_link *prog = priv->glsl_program;
380 unsigned int i;
381
382 if (!prog) {
383 /* No GLSL program set - nothing to do. */
384 return;
385 }
386 programId = prog->programId;
387
388 if (useVertexShader) {
389 IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
390
391 constant_locations = prog->vuniformF_locations;
392 constant_list = &stateBlock->set_vconstantsF;
393
394 /* Load DirectX 9 float constants/uniforms for vertex shader */
395 shader_glsl_load_constantsF(vshader, gl_info, GL_LIMITS(vshader_constantsF),
396 stateBlock->vertexShaderConstantF, constant_locations, constant_list);
397
398 /* Load DirectX 9 integer constants/uniforms for vertex shader */
399 shader_glsl_load_constantsI(vshader, gl_info, programId,
400 prog->vuniformI_locations, MAX_CONST_I,
401 stateBlock->vertexShaderConstantI,
402 stateBlock->changed.vertexShaderConstantsI);
403
404 /* Load DirectX 9 boolean constants/uniforms for vertex shader */
405 shader_glsl_load_constantsB(vshader, gl_info, programId, MAX_CONST_B,
406 stateBlock->vertexShaderConstantB,
407 stateBlock->changed.vertexShaderConstantsB);
408
409 /* Upload the position fixup params */
410 GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, &deviceImpl->posFixup[0]));
411 checkGLcall("glUniform4fvARB");
412 }
413
414 if (usePixelShader) {
415
416 IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
417
418 constant_locations = prog->puniformF_locations;
419 constant_list = &stateBlock->set_pconstantsF;
420
421 /* Load DirectX 9 float constants/uniforms for pixel shader */
422 shader_glsl_load_constantsF(pshader, gl_info, GL_LIMITS(pshader_constantsF),
423 stateBlock->pixelShaderConstantF, constant_locations, constant_list);
424
425 /* Load DirectX 9 integer constants/uniforms for pixel shader */
426 shader_glsl_load_constantsI(pshader, gl_info, programId,
427 prog->puniformI_locations, MAX_CONST_I,
428 stateBlock->pixelShaderConstantI,
429 stateBlock->changed.pixelShaderConstantsI);
430
431 /* Load DirectX 9 boolean constants/uniforms for pixel shader */
432 shader_glsl_load_constantsB(pshader, gl_info, programId, MAX_CONST_B,
433 stateBlock->pixelShaderConstantB,
434 stateBlock->changed.pixelShaderConstantsB);
435
436 /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
437 * It can't be 0 for a valid texbem instruction.
438 */
439 for(i = 0; i < ((IWineD3DPixelShaderImpl *) pshader)->numbumpenvmatconsts; i++) {
440 IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pshader;
441 int stage = ps->luminanceconst[i].texunit;
442
443 float *data = (float *) &stateBlock->textureState[(int) ps->bumpenvmatconst[i].texunit][WINED3DTSS_BUMPENVMAT00];
444 GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
445 checkGLcall("glUniformMatrix2fvARB");
446
447 /* texbeml needs the luminance scale and offset too. If texbeml is used, needsbumpmat
448 * is set too, so we can check that in the needsbumpmat check
449 */
450 if(ps->baseShader.reg_maps.luminanceparams[stage]) {
451 GLfloat *scale = (GLfloat *) &stateBlock->textureState[stage][WINED3DTSS_BUMPENVLSCALE];
452 GLfloat *offset = (GLfloat *) &stateBlock->textureState[stage][WINED3DTSS_BUMPENVLOFFSET];
453
454 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
455 checkGLcall("glUniform1fvARB");
456 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
457 checkGLcall("glUniform1fvARB");
458 }
459 }
460
461 if(((IWineD3DPixelShaderImpl *) pshader)->srgb_enabled &&
462 !((IWineD3DPixelShaderImpl *) pshader)->srgb_mode_hardcoded) {
463 float comparison[4];
464 float mul_low[4];
465
466 if(stateBlock->renderState[WINED3DRS_SRGBWRITEENABLE]) {
467 comparison[0] = srgb_cmp; comparison[1] = srgb_cmp;
468 comparison[2] = srgb_cmp; comparison[3] = srgb_cmp;
469
470 mul_low[0] = srgb_mul_low; mul_low[1] = srgb_mul_low;
471 mul_low[2] = srgb_mul_low; mul_low[3] = srgb_mul_low;
472 } else {
473 comparison[0] = 1.0 / 0.0; comparison[1] = 1.0 / 0.0;
474 comparison[2] = 1.0 / 0.0; comparison[3] = 1.0 / 0.0;
475
476 mul_low[0] = 1.0; mul_low[1] = 1.0;
477 mul_low[2] = 1.0; mul_low[3] = 1.0;
478 }
479
480 GL_EXTCALL(glUniform4fvARB(prog->srgb_comparison_location, 1, comparison));
481 GL_EXTCALL(glUniform4fvARB(prog->srgb_mul_low_location, 1, mul_low));
482 }
483 if(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
484 float correction_params[4];
485 if(deviceImpl->render_offscreen) {
486 correction_params[0] = 0.0;
487 correction_params[1] = 1.0;
488 } else {
489 /* position is window relative, not viewport relative */
490 correction_params[0] = ((IWineD3DSurfaceImpl *) deviceImpl->render_targets[0])->currentDesc.Height;
491 correction_params[1] = -1.0;
492 }
493 GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
494 }
495 }
496 }
497
498 /** Generate the variable & register declarations for the GLSL output target */
499 void shader_generate_glsl_declarations(
500 IWineD3DBaseShader *iface,
501 shader_reg_maps* reg_maps,
502 SHADER_BUFFER* buffer,
503 WineD3D_GL_Info* gl_info) {
504
505 IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
506 IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
507 int i;
508 unsigned int extra_constants_needed = 0;
509 local_constant* lconst;
510
511 /* There are some minor differences between pixel and vertex shaders */
512 char pshader = shader_is_pshader_version(This->baseShader.hex_version);
513 char prefix = pshader ? 'P' : 'V';
514
515 /* Prototype the subroutines */
516 for (i = 0; i < This->baseShader.limits.label; i++) {
517 if (reg_maps->labels[i])
518 shader_addline(buffer, "void subroutine%u();\n", i);
519 }
520
521 /* Declare the constants (aka uniforms) */
522 if (This->baseShader.limits.constant_float > 0) {
523 unsigned max_constantsF = min(This->baseShader.limits.constant_float,
524 (pshader ? GL_LIMITS(pshader_constantsF) : GL_LIMITS(vshader_constantsF)));
525 shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
526 }
527
528 if (This->baseShader.limits.constant_int > 0)
529 shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, This->baseShader.limits.constant_int);
530
531 if (This->baseShader.limits.constant_bool > 0)
532 shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, This->baseShader.limits.constant_bool);
533
534 if(!pshader) {
535 shader_addline(buffer, "uniform vec4 posFixup;\n");
536 /* Predeclaration; This function is added at link time based on the pixel shader.
537 * VS 3.0 shaders have an array OUT[] the shader writes to, earlier versions don't have
538 * that. We know the input to the reorder function at vertex shader compile time, so
539 * we can deal with that. The reorder function for a 1.x and 2.x vertex shader can just
540 * read gl_FrontColor. The output depends on the pixel shader. The reorder function for a
541 * 1.x and 2.x pshader or for fixed function will write gl_FrontColor, and for a 3.0 shader
542 * it will write to the varying array. Here we depend on the shader optimizer on sorting that
543 * out. The nvidia driver only does that if the parameter is inout instead of out, hence the
544 * inout.
545 */
546 if(This->baseShader.hex_version >= WINED3DVS_VERSION(3, 0)) {
547 shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
548 } else {
549 shader_addline(buffer, "void order_ps_input();\n");
550 }
551 } else {
552 IWineD3DPixelShaderImpl *ps_impl = (IWineD3DPixelShaderImpl *) This;
553
554 ps_impl->numbumpenvmatconsts = 0;
555 for(i = 0; i < (sizeof(reg_maps->bumpmat) / sizeof(reg_maps->bumpmat[0])); i++) {
556 if(!reg_maps->bumpmat[i]) {
557 continue;
558 }
559
560 ps_impl->bumpenvmatconst[(int) ps_impl->numbumpenvmatconsts].texunit = i;
561 shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
562
563 if(reg_maps->luminanceparams) {
564 ps_impl->luminanceconst[(int) ps_impl->numbumpenvmatconsts].texunit = i;
565 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
566 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
567 extra_constants_needed++;
568 } else {
569 ps_impl->luminanceconst[(int) ps_impl->numbumpenvmatconsts].texunit = -1;
570 }
571
572 extra_constants_needed++;
573 ps_impl->numbumpenvmatconsts++;
574 }
575
576 if(device->stateBlock->renderState[WINED3DRS_SRGBWRITEENABLE]) {
577 ps_impl->srgb_enabled = 1;
578 if(This->baseShader.limits.constant_float + extra_constants_needed + 1 < GL_LIMITS(pshader_constantsF)) {
579 shader_addline(buffer, "uniform vec4 srgb_mul_low;\n");
580 shader_addline(buffer, "uniform vec4 srgb_comparison;\n");
581 ps_impl->srgb_mode_hardcoded = 0;
582 extra_constants_needed++;
583 } else {
584 ps_impl->srgb_mode_hardcoded = 1;
585 shader_addline(buffer, "const vec4 srgb_mul_low = vec4(%f, %f, %f, %f);\n",
586 srgb_mul_low, srgb_mul_low, srgb_mul_low, srgb_mul_low);
587 shader_addline(buffer, "const vec4 srgb_comparison = vec4(%f, %f, %f, %f);\n",
588 srgb_cmp, srgb_cmp, srgb_cmp, srgb_cmp);
589 }
590 } else {
591 IWineD3DPixelShaderImpl *ps_impl = (IWineD3DPixelShaderImpl *) This;
592
593 /* Do not write any srgb fixup into the shader to save shader size and processing time.
594 * As a consequence, we can't toggle srgb write on without recompilation
595 */
596 ps_impl->srgb_enabled = 0;
597 ps_impl->srgb_mode_hardcoded = 1;
598 }
599 if(reg_maps->vpos || reg_maps->usesdsy) {
600 if(This->baseShader.limits.constant_float + extra_constants_needed + 1 < GL_LIMITS(pshader_constantsF)) {
601 shader_addline(buffer, "uniform vec4 ycorrection;\n");
602 ((IWineD3DPixelShaderImpl *) This)->vpos_uniform = 1;
603 extra_constants_needed++;
604 } else {
605 /* This happens because we do not have proper tracking of the constant registers that are
606 * actually used, only the max limit of the shader version
607 */
608 FIXME("Cannot find a free uniform for vpos correction params\n");
609 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
610 device->render_offscreen ? 0.0 : ((IWineD3DSurfaceImpl *) device->render_targets[0])->currentDesc.Height,
611 device->render_offscreen ? 1.0 : -1.0);
612 }
613 shader_addline(buffer, "vec4 vpos;\n");
614 }
615 }
616
617 /* Declare texture samplers */
618 for (i = 0; i < This->baseShader.limits.sampler; i++) {
619 if (reg_maps->samplers[i]) {
620
621 DWORD stype = reg_maps->samplers[i] & WINED3DSP_TEXTURETYPE_MASK;
622 switch (stype) {
623
624 case WINED3DSTT_1D:
625 shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
626 break;
627 case WINED3DSTT_2D:
628 if(device->stateBlock->textures[i] &&
629 IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[i]) == GL_TEXTURE_RECTANGLE_ARB) {
630 shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
631 } else {
632 shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
633 }
634 break;
635 case WINED3DSTT_CUBE:
636 shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
637 break;
638 case WINED3DSTT_VOLUME:
639 shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
640 break;
641 default:
642 shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
643 FIXME("Unrecognized sampler type: %#x\n", stype);
644 break;
645 }
646 }
647 }
648
649 /* Declare address variables */
650 for (i = 0; i < This->baseShader.limits.address; i++) {
651 if (reg_maps->address[i])
652 shader_addline(buffer, "ivec4 A%d;\n", i);
653 }
654
655 /* Declare texture coordinate temporaries and initialize them */
656 for (i = 0; i < This->baseShader.limits.texcoord; i++) {
657 if (reg_maps->texcoord[i])
658 shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
659 }
660
661 /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
662 * helper function shader that is linked in at link time
663 */
664 if(pshader && This->baseShader.hex_version >= WINED3DPS_VERSION(3, 0)) {
665 if(use_vs(device)) {
666 shader_addline(buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
667 } else {
668 /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
669 * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
670 * pixel shader that reads the fixed function color into the packed input registers.
671 */
672 shader_addline(buffer, "vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
673 }
674 }
675
676 /* Declare output register temporaries */
677 if(This->baseShader.limits.packed_output) {
678 shader_addline(buffer, "vec4 OUT[%u];\n", This->baseShader.limits.packed_output);
679 }
680
681 /* Declare temporary variables */
682 for(i = 0; i < This->baseShader.limits.temporary; i++) {
683 if (reg_maps->temporary[i])
684 shader_addline(buffer, "vec4 R%u;\n", i);
685 }
686
687 /* Declare attributes */
688 for (i = 0; i < This->baseShader.limits.attributes; i++) {
689 if (reg_maps->attributes[i])
690 shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
691 }
692
693 /* Declare loop registers aLx */
694 for (i = 0; i < reg_maps->loop_depth; i++) {
695 shader_addline(buffer, "int aL%u;\n", i);
696 shader_addline(buffer, "int tmpInt%u;\n", i);
697 }
698
699 /* Temporary variables for matrix operations */
700 shader_addline(buffer, "vec4 tmp0;\n");
701 shader_addline(buffer, "vec4 tmp1;\n");
702
703 /* Local constants use a different name so they can be loaded once at shader link time
704 * They can't be hardcoded into the shader text via LC = {x, y, z, w}; because the
705 * float -> string conversion can cause precision loss.
706 */
707 if(!This->baseShader.load_local_constsF) {
708 LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
709 shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
710 }
711 }
712
713 /* Start the main program */
714 shader_addline(buffer, "void main() {\n");
715 if(pshader && reg_maps->vpos) {
716 /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
717 * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
718 * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
719 * precision troubles when we just substract 0.5.
720 *
721 * To deal with that just floor() the position. This will eliminate the fraction on all cards.
722 *
723 * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
724 *
725 * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
726 * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
727 * coordinates specify the pixel centers instead of the pixel corners. This code will behave
728 * correctly on drivers that returns integer values.
729 */
730 shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
731 }
732 }
733
734 /*****************************************************************************
735 * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
736 *
737 * For more information, see http://wiki.winehq.org/DirectX-Shaders
738 ****************************************************************************/
739
740 /* Prototypes */
741 static void shader_glsl_add_src_param(SHADER_OPCODE_ARG* arg, const DWORD param,
742 const DWORD addr_token, DWORD mask, glsl_src_param_t *src_param);
743
744 /** Used for opcode modifiers - They multiply the result by the specified amount */
745 static const char * const shift_glsl_tab[] = {
746 "", /* 0 (none) */
747 "2.0 * ", /* 1 (x2) */
748 "4.0 * ", /* 2 (x4) */
749 "8.0 * ", /* 3 (x8) */
750 "16.0 * ", /* 4 (x16) */
751 "32.0 * ", /* 5 (x32) */
752 "", /* 6 (x64) */
753 "", /* 7 (x128) */
754 "", /* 8 (d256) */
755 "", /* 9 (d128) */
756 "", /* 10 (d64) */
757 "", /* 11 (d32) */
758 "0.0625 * ", /* 12 (d16) */
759 "0.125 * ", /* 13 (d8) */
760 "0.25 * ", /* 14 (d4) */
761 "0.5 * " /* 15 (d2) */
762 };
763
764 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
765 static void shader_glsl_gen_modifier (
766 const DWORD instr,
767 const char *in_reg,
768 const char *in_regswizzle,
769 char *out_str) {
770
771 out_str[0] = 0;
772
773 if (instr == WINED3DSIO_TEXKILL)
774 return;
775
776 switch (instr & WINED3DSP_SRCMOD_MASK) {
777 case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
778 case WINED3DSPSM_DW:
779 case WINED3DSPSM_NONE:
780 sprintf(out_str, "%s%s", in_reg, in_regswizzle);
781 break;
782 case WINED3DSPSM_NEG:
783 sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
784 break;
785 case WINED3DSPSM_NOT:
786 sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
787 break;
788 case WINED3DSPSM_BIAS:
789 sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
790 break;
791 case WINED3DSPSM_BIASNEG:
792 sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
793 break;
794 case WINED3DSPSM_SIGN:
795 sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
796 break;
797 case WINED3DSPSM_SIGNNEG:
798 sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
799 break;
800 case WINED3DSPSM_COMP:
801 sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
802 break;
803 case WINED3DSPSM_X2:
804 sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
805 break;
806 case WINED3DSPSM_X2NEG:
807 sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
808 break;
809 case WINED3DSPSM_ABS:
810 sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
811 break;
812 case WINED3DSPSM_ABSNEG:
813 sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
814 break;
815 default:
816 FIXME("Unhandled modifier %u\n", (instr & WINED3DSP_SRCMOD_MASK));
817 sprintf(out_str, "%s%s", in_reg, in_regswizzle);
818 }
819 }
820
821 /** Writes the GLSL variable name that corresponds to the register that the
822 * DX opcode parameter is trying to access */
823 static void shader_glsl_get_register_name(
824 const DWORD param,
825 const DWORD addr_token,
826 char* regstr,
827 BOOL* is_color,
828 SHADER_OPCODE_ARG* arg) {
829
830 /* oPos, oFog and oPts in D3D */
831 static const char * const hwrastout_reg_names[] = { "gl_Position", "gl_FogFragCoord", "gl_PointSize" };
832
833 DWORD reg = param & WINED3DSP_REGNUM_MASK;
834 DWORD regtype = shader_get_regtype(param);
835 IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) arg->shader;
836 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
837 WineD3D_GL_Info* gl_info = &deviceImpl->adapter->gl_info;
838
839 char pshader = shader_is_pshader_version(This->baseShader.hex_version);
840 char tmpStr[150];
841
842 *is_color = FALSE;
843
844 switch (regtype) {
845 case WINED3DSPR_TEMP:
846 sprintf(tmpStr, "R%u", reg);
847 break;
848 case WINED3DSPR_INPUT:
849 if (pshader) {
850 /* Pixel shaders >= 3.0 */
851 if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) >= 3) {
852 DWORD in_count = GL_LIMITS(glsl_varyings) / 4;
853
854 if (param & WINED3DSHADER_ADDRMODE_RELATIVE) {
855 glsl_src_param_t rel_param;
856 shader_glsl_add_src_param(arg, addr_token, 0, WINED3DSP_WRITEMASK_0, &rel_param);
857
858 /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
859 * operation there
860 */
861 if(((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg]) {
862 if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count) {
863 sprintf(tmpStr, "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
864 rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg], in_count - 1,
865 rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg], in_count,
866 rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg]);
867 } else {
868 sprintf(tmpStr, "IN[%s + %u]", rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg]);
869 }
870 } else {
871 if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count) {
872 sprintf(tmpStr, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
873 rel_param.param_str, in_count - 1,
874 rel_param.param_str, in_count,
875 rel_param.param_str);
876 } else {
877 sprintf(tmpStr, "IN[%s]", rel_param.param_str);
878 }
879 }
880 } else {
881 DWORD idx = ((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg];
882 if (idx == in_count) {
883 sprintf(tmpStr, "gl_Color");
884 } else if (idx == in_count + 1) {
885 sprintf(tmpStr, "gl_SecondaryColor");
886 } else {
887 sprintf(tmpStr, "IN[%u]", idx);
888 }
889 }
890 } else {
891 if (reg==0)
892 strcpy(tmpStr, "gl_Color");
893 else
894 strcpy(tmpStr, "gl_SecondaryColor");
895 }
896 } else {
897 if (vshader_input_is_color((IWineD3DVertexShader*) This, reg))
898 *is_color = TRUE;
899 sprintf(tmpStr, "attrib%u", reg);
900 }
901 break;
902 case WINED3DSPR_CONST:
903 {
904 const char prefix = pshader? 'P':'V';
905
906 /* Relative addressing */
907 if (param & WINED3DSHADER_ADDRMODE_RELATIVE) {
908
909 /* Relative addressing on shaders 2.0+ have a relative address token,
910 * prior to that, it was hard-coded as "A0.x" because there's only 1 register */
911 if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) >= 2) {
912 glsl_src_param_t rel_param;
913 shader_glsl_add_src_param(arg, addr_token, 0, WINED3DSP_WRITEMASK_0, &rel_param);
914 if(reg) {
915 sprintf(tmpStr, "%cC[%s + %u]", prefix, rel_param.param_str, reg);
916 } else {
917 sprintf(tmpStr, "%cC[%s]", prefix, rel_param.param_str);
918 }
919 } else {
920 if(reg) {
921 sprintf(tmpStr, "%cC[A0.x + %u]", prefix, reg);
922 } else {
923 sprintf(tmpStr, "%cC[A0.x]", prefix);
924 }
925 }
926
927 } else {
928 if(shader_constant_is_local(This, reg)) {
929 sprintf(tmpStr, "%cLC%u", prefix, reg);
930 } else {
931 sprintf(tmpStr, "%cC[%u]", prefix, reg);
932 }
933 }
934
935 break;
936 }
937 case WINED3DSPR_CONSTINT:
938 if (pshader)
939 sprintf(tmpStr, "PI[%u]", reg);
940 else
941 sprintf(tmpStr, "VI[%u]", reg);
942 break;
943 case WINED3DSPR_CONSTBOOL:
944 if (pshader)
945 sprintf(tmpStr, "PB[%u]", reg);
946 else
947 sprintf(tmpStr, "VB[%u]", reg);
948 break;
949 case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
950 if (pshader) {
951 sprintf(tmpStr, "T%u", reg);
952 } else {
953 sprintf(tmpStr, "A%u", reg);
954 }
955 break;
956 case WINED3DSPR_LOOP:
957 sprintf(tmpStr, "aL%u", This->baseShader.cur_loop_regno - 1);
958 break;
959 case WINED3DSPR_SAMPLER:
960 if (pshader)
961 sprintf(tmpStr, "Psampler%u", reg);
962 else
963 sprintf(tmpStr, "Vsampler%u", reg);
964 break;
965 case WINED3DSPR_COLOROUT:
966 if (reg >= GL_LIMITS(buffers)) {
967 WARN("Write to render target %u, only %d supported\n", reg, 4);
968 }
969 if (GL_SUPPORT(ARB_DRAW_BUFFERS)) {
970 sprintf(tmpStr, "gl_FragData[%u]", reg);
971 } else { /* On older cards with GLSL support like the GeforceFX there's only one buffer. */
972 sprintf(tmpStr, "gl_FragColor");
973 }
974 break;
975 case WINED3DSPR_RASTOUT:
976 sprintf(tmpStr, "%s", hwrastout_reg_names[reg]);
977 break;
978 case WINED3DSPR_DEPTHOUT:
979 sprintf(tmpStr, "gl_FragDepth");
980 break;
981 case WINED3DSPR_ATTROUT:
982 if (reg == 0) {
983 sprintf(tmpStr, "gl_FrontColor");
984 } else {
985 sprintf(tmpStr, "gl_FrontSecondaryColor");
986 }
987 break;
988 case WINED3DSPR_TEXCRDOUT:
989 /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
990 if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) >= 3)
991 sprintf(tmpStr, "OUT[%u]", reg);
992 else
993 sprintf(tmpStr, "gl_TexCoord[%u]", reg);
994 break;
995 case WINED3DSPR_MISCTYPE:
996 if (reg == 0) {
997 /* vPos */
998 sprintf(tmpStr, "vpos");
999 } else if (reg == 1){
1000 /* Note that gl_FrontFacing is a bool, while vFace is
1001 * a float for which the sign determines front/back
1002 */
1003 sprintf(tmpStr, "(gl_FrontFacing ? 1.0 : -1.0)");
1004 } else {
1005 FIXME("Unhandled misctype register %d\n", reg);
1006 sprintf(tmpStr, "unrecognized_register");
1007 }
1008 break;
1009 default:
1010 FIXME("Unhandled register name Type(%d)\n", regtype);
1011 sprintf(tmpStr, "unrecognized_register");
1012 break;
1013 }
1014
1015 strcat(regstr, tmpStr);
1016 }
1017
1018 /* Get the GLSL write mask for the destination register */
1019 static DWORD shader_glsl_get_write_mask(const DWORD param, char *write_mask) {
1020 char *ptr = write_mask;
1021 DWORD mask = param & WINED3DSP_WRITEMASK_ALL;
1022
1023 if (shader_is_scalar(param)) {
1024 mask = WINED3DSP_WRITEMASK_0;
1025 } else {
1026 *ptr++ = '.';
1027 if (param & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1028 if (param & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1029 if (param & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1030 if (param & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1031 }
1032
1033 *ptr = '\0';
1034
1035 return mask;
1036 }
1037
1038 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1039 unsigned int size = 0;
1040
1041 if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1042 if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1043 if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1044 if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1045
1046 return size;
1047 }
1048
1049 static void shader_glsl_get_swizzle(const DWORD param, BOOL fixup, DWORD mask, char *swizzle_str) {
1050 /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1051 * but addressed as "rgba". To fix this we need to swap the register's x
1052 * and z components. */
1053 DWORD swizzle = (param & WINED3DSP_SWIZZLE_MASK) >> WINED3DSP_SWIZZLE_SHIFT;
1054 const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1055 char *ptr = swizzle_str;
1056
1057 if (!shader_is_scalar(param)) {
1058 *ptr++ = '.';
1059 /* swizzle bits fields: wwzzyyxx */
1060 if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = swizzle_chars[swizzle & 0x03];
1061 if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = swizzle_chars[(swizzle >> 2) & 0x03];
1062 if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = swizzle_chars[(swizzle >> 4) & 0x03];
1063 if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = swizzle_chars[(swizzle >> 6) & 0x03];
1064 }
1065
1066 *ptr = '\0';
1067 }
1068
1069 /* From a given parameter token, generate the corresponding GLSL string.
1070 * Also, return the actual register name and swizzle in case the
1071 * caller needs this information as well. */
1072 static void shader_glsl_add_src_param(SHADER_OPCODE_ARG* arg, const DWORD param,
1073 const DWORD addr_token, DWORD mask, glsl_src_param_t *src_param) {
1074 BOOL is_color = FALSE;
1075 char swizzle_str[6];
1076
1077 src_param->reg_name[0] = '\0';
1078 src_param->param_str[0] = '\0';
1079 swizzle_str[0] = '\0';
1080
1081 shader_glsl_get_register_name(param, addr_token, src_param->reg_name, &is_color, arg);
1082
1083 shader_glsl_get_swizzle(param, is_color, mask, swizzle_str);
1084 shader_glsl_gen_modifier(param, src_param->reg_name, swizzle_str, src_param->param_str);
1085 }
1086
1087 /* From a given parameter token, generate the corresponding GLSL string.
1088 * Also, return the actual register name and swizzle in case the
1089 * caller needs this information as well. */
1090 static DWORD shader_glsl_add_dst_param(SHADER_OPCODE_ARG* arg, const DWORD param,
1091 const DWORD addr_token, glsl_dst_param_t *dst_param) {
1092 BOOL is_color = FALSE;
1093
1094 dst_param->mask_str[0] = '\0';
1095 dst_param->reg_name[0] = '\0';
1096
1097 shader_glsl_get_register_name(param, addr_token, dst_param->reg_name, &is_color, arg);
1098 return shader_glsl_get_write_mask(param, dst_param->mask_str);
1099 }
1100
1101 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1102 static DWORD shader_glsl_append_dst_ext(SHADER_BUFFER *buffer, SHADER_OPCODE_ARG *arg, const DWORD param) {
1103 glsl_dst_param_t dst_param;
1104 DWORD mask;
1105 int shift;
1106
1107 mask = shader_glsl_add_dst_param(arg, param, arg->dst_addr, &dst_param);
1108
1109 if(mask) {
1110 shift = (param & WINED3DSP_DSTSHIFT_MASK) >> WINED3DSP_DSTSHIFT_SHIFT;
1111 shader_addline(buffer, "%s%s = %s(", dst_param.reg_name, dst_param.mask_str, shift_glsl_tab[shift]);
1112 }
1113
1114 return mask;
1115 }
1116
1117 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1118 static DWORD shader_glsl_append_dst(SHADER_BUFFER *buffer, SHADER_OPCODE_ARG *arg) {
1119 return shader_glsl_append_dst_ext(buffer, arg, arg->dst);
1120 }
1121
1122 /** Process GLSL instruction modifiers */
1123 void shader_glsl_add_instruction_modifiers(SHADER_OPCODE_ARG* arg) {
1124
1125 DWORD mask = arg->dst & WINED3DSP_DSTMOD_MASK;
1126
1127 if (arg->opcode->dst_token && mask != 0) {
1128 glsl_dst_param_t dst_param;
1129
1130 shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
1131
1132 if (mask & WINED3DSPDM_SATURATE) {
1133 /* _SAT means to clamp the value of the register to between 0 and 1 */
1134 shader_addline(arg->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1135 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1136 }
1137 if (mask & WINED3DSPDM_MSAMPCENTROID) {
1138 FIXME("_centroid modifier not handled\n");
1139 }
1140 if (mask & WINED3DSPDM_PARTIALPRECISION) {
1141 /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1142 }
1143 }
1144 }
1145
1146 static inline const char* shader_get_comp_op(
1147 const DWORD opcode) {
1148
1149 DWORD op = (opcode & INST_CONTROLS_MASK) >> INST_CONTROLS_SHIFT;
1150 switch (op) {
1151 case COMPARISON_GT: return ">";
1152 case COMPARISON_EQ: return "==";
1153 case COMPARISON_GE: return ">=";
1154 case COMPARISON_LT: return "<";
1155 case COMPARISON_NE: return "!=";
1156 case COMPARISON_LE: return "<=";
1157 default:
1158 FIXME("Unrecognized comparison value: %u\n", op);
1159 return "(\?\?)";
1160 }
1161 }
1162
1163 static void shader_glsl_get_sample_function(DWORD sampler_type, BOOL projected, BOOL texrect, glsl_sample_function_t *sample_function) {
1164 /* Note that there's no such thing as a projected cube texture. */
1165 switch(sampler_type) {
1166 case WINED3DSTT_1D:
1167 sample_function->name = projected ? "texture1DProj" : "texture1D";
1168 sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1169 break;
1170 case WINED3DSTT_2D:
1171 if(texrect) {
1172 sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1173 } else {
1174 sample_function->name = projected ? "texture2DProj" : "texture2D";
1175 }
1176 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1177 break;
1178 case WINED3DSTT_CUBE:
1179 sample_function->name = "textureCube";
1180 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1181 break;
1182 case WINED3DSTT_VOLUME:
1183 sample_function->name = projected ? "texture3DProj" : "texture3D";
1184 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1185 break;
1186 default:
1187 sample_function->name = "";
1188 FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1189 break;
1190 }
1191 }
1192
1193 static void shader_glsl_color_correction(SHADER_OPCODE_ARG* arg) {
1194 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1195 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) shader->baseShader.device;
1196 WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
1197 glsl_dst_param_t dst_param;
1198 glsl_dst_param_t dst_param2;
1199 WINED3DFORMAT fmt;
1200 WINED3DFORMAT conversion_group;
1201 IWineD3DBaseTextureImpl *texture;
1202 DWORD mask, mask_size;
1203 UINT i;
1204 BOOL recorded = FALSE;
1205 DWORD sampler_idx;
1206 DWORD hex_version = shader->baseShader.hex_version;
1207
1208 switch(arg->opcode->opcode) {
1209 case WINED3DSIO_TEX:
1210 if (hex_version < WINED3DPS_VERSION(2,0)) {
1211 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
1212 } else {
1213 sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
1214 }
1215 break;
1216
1217 case WINED3DSIO_TEXLDL:
1218 FIXME("Add color fixup for vertex texture WINED3DSIO_TEXLDL\n");
1219 return;
1220
1221 case WINED3DSIO_TEXDP3TEX:
1222 case WINED3DSIO_TEXM3x3TEX:
1223 case WINED3DSIO_TEXM3x3SPEC:
1224 case WINED3DSIO_TEXM3x3VSPEC:
1225 case WINED3DSIO_TEXBEM:
1226 case WINED3DSIO_TEXREG2AR:
1227 case WINED3DSIO_TEXREG2GB:
1228 case WINED3DSIO_TEXREG2RGB:
1229 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
1230 break;
1231
1232 default:
1233 /* Not a texture sampling instruction, nothing to do */
1234 return;
1235 };
1236
1237 texture = (IWineD3DBaseTextureImpl *) deviceImpl->stateBlock->textures[sampler_idx];
1238 if(texture) {
1239 fmt = texture->resource.format;
1240 conversion_group = texture->baseTexture.shader_conversion_group;
1241 } else {
1242 fmt = WINED3DFMT_UNKNOWN;
1243 conversion_group = WINED3DFMT_UNKNOWN;
1244 }
1245
1246 /* before doing anything, record the sampler with the format in the format conversion list,
1247 * but check if it's not there already
1248 */
1249 for(i = 0; i < shader->baseShader.num_sampled_samplers; i++) {
1250 if(shader->baseShader.sampled_samplers[i] == sampler_idx) {
1251 recorded = TRUE;
1252 break;
1253 }
1254 }
1255 if(!recorded) {
1256 shader->baseShader.sampled_samplers[shader->baseShader.num_sampled_samplers] = sampler_idx;
1257 shader->baseShader.num_sampled_samplers++;
1258 shader->baseShader.sampled_format[sampler_idx] = conversion_group;
1259 }
1260
1261 switch(fmt) {
1262 case WINED3DFMT_V8U8:
1263 case WINED3DFMT_V16U16:
1264 if(GL_SUPPORT(NV_TEXTURE_SHADER) ||
1265 (GL_SUPPORT(ATI_ENVMAP_BUMPMAP) && fmt == WINED3DFMT_V8U8)) {
1266 /* The 3rd channel returns 1.0 in d3d, but 0.0 in gl. Fix this while we're at it :-) */
1267 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_2, &dst_param);
1268 mask_size = shader_glsl_get_write_mask_size(mask);
1269 if(mask_size >= 3) {
1270 shader_addline(arg->buffer, "%s.%c = 1.0;\n", dst_param.reg_name, dst_param.mask_str[3]);
1271 }
1272 } else {
1273 /* Correct the sign, but leave the blue as it is - it was loaded correctly already */
1274 mask = shader_glsl_add_dst_param(arg, arg->dst,
1275 WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1,
1276 &dst_param);
1277 mask_size = shader_glsl_get_write_mask_size(mask);
1278 if(mask_size >= 2) {
1279 shader_addline(arg->buffer, "%s.%c%c = %s.%c%c * 2.0 - 1.0;\n",
1280 dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2],
1281 dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2]);
1282 } else if(mask_size == 1) {
1283 shader_addline(arg->buffer, "%s.%c = %s.%c * 2.0 - 1.0;\n", dst_param.reg_name, dst_param.mask_str[1],
1284 dst_param.reg_name, dst_param.mask_str[1]);
1285 }
1286 }
1287 break;
1288
1289 case WINED3DFMT_X8L8V8U8:
1290 if(!GL_SUPPORT(NV_TEXTURE_SHADER)) {
1291 /* Red and blue are the signed channels, fix them up; Blue(=L) is correct already,
1292 * and a(X) is always 1.0
1293 */
1294 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &dst_param);
1295 mask_size = shader_glsl_get_write_mask_size(mask);
1296 if(mask_size >= 2) {
1297 shader_addline(arg->buffer, "%s.%c%c = %s.%c%c * 2.0 - 1.0;\n",
1298 dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2],
1299 dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2]);
1300 } else if(mask_size == 1) {
1301 shader_addline(arg->buffer, "%s.%c = %s.%c * 2.0 - 1.0;\n",
1302 dst_param.reg_name, dst_param.mask_str[1],
1303 dst_param.reg_name, dst_param.mask_str[1]);
1304 }
1305 }
1306 break;
1307
1308 case WINED3DFMT_L6V5U5:
1309 if(!GL_SUPPORT(NV_TEXTURE_SHADER)) {
1310 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &dst_param);
1311 mask_size = shader_glsl_get_write_mask_size(mask);
1312 shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_2, &dst_param2);
1313 if(mask_size >= 3) {
1314 /* Swap y and z (U and L), and do a sign conversion on x and the new y(V and U) */
1315 shader_addline(arg->buffer, "tmp0.g = %s.%c;\n",
1316 dst_param.reg_name, dst_param.mask_str[2]);
1317 shader_addline(arg->buffer, "%s.%c%c = %s.%c%c * 2.0 - 1.0;\n",
1318 dst_param.reg_name, dst_param.mask_str[2], dst_param.mask_str[1],
1319 dst_param2.reg_name, dst_param.mask_str[1], dst_param.mask_str[3]);
1320 shader_addline(arg->buffer, "%s.%c = tmp0.g;\n", dst_param.reg_name,
1321 dst_param.mask_str[3]);
1322 } else if(mask_size == 2) {
1323 /* This is bad: We have VL, but we need VU */
1324 FIXME("2 components sampled from a converted L6V5U5 texture\n");
1325 } else {
1326 shader_addline(arg->buffer, "%s.%c = %s.%c * 2.0 - 1.0;\n",
1327 dst_param.reg_name, dst_param.mask_str[1],
1328 dst_param2.reg_name, dst_param.mask_str[1]);
1329 }
1330 }
1331 break;
1332
1333 case WINED3DFMT_Q8W8V8U8:
1334 if(!GL_SUPPORT(NV_TEXTURE_SHADER)) {
1335 /* Correct the sign in all channels. The writemask just applies as-is, no
1336 * need for checking the mask size
1337 */
1338 shader_glsl_add_dst_param(arg, arg->dst,
1339 WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 |
1340 WINED3DSP_WRITEMASK_2 | WINED3DSP_WRITEMASK_3,
1341 &dst_param);
1342 shader_addline(arg->buffer, "%s%s = %s%s * 2.0 - 1.0;\n", dst_param.reg_name, dst_param.mask_str,
1343 dst_param.reg_name, dst_param.mask_str);
1344 }
1345 break;
1346
1347 /* stupid compiler */
1348 default:
1349 break;
1350 }
1351 }
1352
1353 /*****************************************************************************
1354 *
1355 * Begin processing individual instruction opcodes
1356 *
1357 ****************************************************************************/
1358
1359 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
1360 void shader_glsl_arith(SHADER_OPCODE_ARG* arg) {
1361 CONST SHADER_OPCODE* curOpcode = arg->opcode;
1362 SHADER_BUFFER* buffer = arg->buffer;
1363 glsl_src_param_t src0_param;
1364 glsl_src_param_t src1_param;
1365 DWORD write_mask;
1366 char op;
1367
1368 /* Determine the GLSL operator to use based on the opcode */
1369 switch (curOpcode->opcode) {
1370 case WINED3DSIO_MUL: op = '*'; break;
1371 case WINED3DSIO_ADD: op = '+'; break;
1372 case WINED3DSIO_SUB: op = '-'; break;
1373 default:
1374 op = ' ';
1375 FIXME("Opcode %s not yet handled in GLSL\n", curOpcode->name);
1376 break;
1377 }
1378
1379 write_mask = shader_glsl_append_dst(buffer, arg);
1380 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1381 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1382 shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
1383 }
1384
1385 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
1386 void shader_glsl_mov(SHADER_OPCODE_ARG* arg) {
1387 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1388 SHADER_BUFFER* buffer = arg->buffer;
1389 glsl_src_param_t src0_param;
1390 DWORD write_mask;
1391
1392 write_mask = shader_glsl_append_dst(buffer, arg);
1393 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1394
1395 /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
1396 * shader versions WINED3DSIO_MOVA is used for this. */
1397 if ((WINED3DSHADER_VERSION_MAJOR(shader->baseShader.hex_version) == 1 &&
1398 !shader_is_pshader_version(shader->baseShader.hex_version) &&
1399 shader_get_regtype(arg->dst) == WINED3DSPR_ADDR)) {
1400 /* This is a simple floor() */
1401 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1402 if (mask_size > 1) {
1403 shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
1404 } else {
1405 shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
1406 }
1407 } else if(arg->opcode->opcode == WINED3DSIO_MOVA) {
1408 /* We need to *round* to the nearest int here. */
1409 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1410 if (mask_size > 1) {
1411 shader_addline(buffer, "ivec%d(floor(abs(%s) + vec%d(0.5)) * sign(%s)));\n", mask_size, src0_param.param_str, mask_size, src0_param.param_str);
1412 } else {
1413 shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n", src0_param.param_str, src0_param.param_str);
1414 }
1415 } else {
1416 shader_addline(buffer, "%s);\n", src0_param.param_str);
1417 }
1418 }
1419
1420 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
1421 void shader_glsl_dot(SHADER_OPCODE_ARG* arg) {
1422 CONST SHADER_OPCODE* curOpcode = arg->opcode;
1423 SHADER_BUFFER* buffer = arg->buffer;
1424 glsl_src_param_t src0_param;
1425 glsl_src_param_t src1_param;
1426 DWORD dst_write_mask, src_write_mask;
1427 unsigned int dst_size = 0;
1428
1429 dst_write_mask = shader_glsl_append_dst(buffer, arg);
1430 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1431
1432 /* dp3 works on vec3, dp4 on vec4 */
1433 if (curOpcode->opcode == WINED3DSIO_DP4) {
1434 src_write_mask = WINED3DSP_WRITEMASK_ALL;
1435 } else {
1436 src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1437 }
1438
1439 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_write_mask, &src0_param);
1440 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_write_mask, &src1_param);
1441
1442 if (dst_size > 1) {
1443 shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1444 } else {
1445 shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
1446 }
1447 }
1448
1449 /* Note that this instruction has some restrictions. The destination write mask
1450 * can't contain the w component, and the source swizzles have to be .xyzw */
1451 void shader_glsl_cross(SHADER_OPCODE_ARG *arg) {
1452 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1453 glsl_src_param_t src0_param;
1454 glsl_src_param_t src1_param;
1455 char dst_mask[6];
1456
1457 shader_glsl_get_write_mask(arg->dst, dst_mask);
1458 shader_glsl_append_dst(arg->buffer, arg);
1459 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
1460 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_mask, &src1_param);
1461 shader_addline(arg->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
1462 }
1463
1464 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
1465 * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
1466 * GLSL uses the value as-is. */
1467 void shader_glsl_pow(SHADER_OPCODE_ARG *arg) {
1468 SHADER_BUFFER *buffer = arg->buffer;
1469 glsl_src_param_t src0_param;
1470 glsl_src_param_t src1_param;
1471 DWORD dst_write_mask;
1472 unsigned int dst_size;
1473
1474 dst_write_mask = shader_glsl_append_dst(buffer, arg);
1475 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1476
1477 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1478 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
1479
1480 if (dst_size > 1) {
1481 shader_addline(buffer, "vec%d(pow(abs(%s), %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1482 } else {
1483 shader_addline(buffer, "pow(abs(%s), %s));\n", src0_param.param_str, src1_param.param_str);
1484 }
1485 }
1486
1487 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
1488 * Src0 is a scalar. Note that D3D uses the absolute of src0, while
1489 * GLSL uses the value as-is. */
1490 void shader_glsl_log(SHADER_OPCODE_ARG *arg) {
1491 SHADER_BUFFER *buffer = arg->buffer;
1492 glsl_src_param_t src0_param;
1493 DWORD dst_write_mask;
1494 unsigned int dst_size;
1495
1496 dst_write_mask = shader_glsl_append_dst(buffer, arg);
1497 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1498
1499 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1500
1501 if (dst_size > 1) {
1502 shader_addline(buffer, "vec%d(log2(abs(%s))));\n", dst_size, src0_param.param_str);
1503 } else {
1504 shader_addline(buffer, "log2(abs(%s)));\n", src0_param.param_str);
1505 }
1506 }
1507
1508 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
1509 void shader_glsl_map2gl(SHADER_OPCODE_ARG* arg) {
1510 CONST SHADER_OPCODE* curOpcode = arg->opcode;
1511 SHADER_BUFFER* buffer = arg->buffer;
1512 glsl_src_param_t src_param;
1513 const char *instruction;
1514 char arguments[256];
1515 DWORD write_mask;
1516 unsigned i;
1517
1518 /* Determine the GLSL function to use based on the opcode */
1519 /* TODO: Possibly make this a table for faster lookups */
1520 switch (curOpcode->opcode) {
1521 case WINED3DSIO_MIN: instruction = "min"; break;
1522 case WINED3DSIO_MAX: instruction = "max"; break;
1523 case WINED3DSIO_ABS: instruction = "abs"; break;
1524 case WINED3DSIO_FRC: instruction = "fract"; break;
1525 case WINED3DSIO_NRM: instruction = "normalize"; break;
1526 case WINED3DSIO_LOGP:
1527 case WINED3DSIO_LOG: instruction = "log2"; break;
1528 case WINED3DSIO_EXP: instruction = "exp2"; break;
1529 case WINED3DSIO_SGN: instruction = "sign"; break;
1530 case WINED3DSIO_DSX: instruction = "dFdx"; break;
1531 case WINED3DSIO_DSY: instruction = "ycorrection.y * dFdy"; break;
1532 default: instruction = "";
1533 FIXME("Opcode %s not yet handled in GLSL\n", curOpcode->name);
1534 break;
1535 }
1536
1537 write_mask = shader_glsl_append_dst(buffer, arg);
1538
1539 arguments[0] = '\0';
1540 if (curOpcode->num_params > 0) {
1541 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src_param);
1542 strcat(arguments, src_param.param_str);
1543 for (i = 2; i < curOpcode->num_params; ++i) {
1544 strcat(arguments, ", ");
1545 shader_glsl_add_src_param(arg, arg->src[i-1], arg->src_addr[i-1], write_mask, &src_param);
1546 strcat(arguments, src_param.param_str);
1547 }
1548 }
1549
1550 shader_addline(buffer, "%s(%s));\n", instruction, arguments);
1551 }
1552
1553 /** Process the WINED3DSIO_EXPP instruction in GLSL:
1554 * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
1555 * dst.x = 2^(floor(src))
1556 * dst.y = src - floor(src)
1557 * dst.z = 2^src (partial precision is allowed, but optional)
1558 * dst.w = 1.0;
1559 * For 2.0 shaders, just do this (honoring writemask and swizzle):
1560 * dst = 2^src; (partial precision is allowed, but optional)
1561 */
1562 void shader_glsl_expp(SHADER_OPCODE_ARG* arg) {
1563 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)arg->shader;
1564 glsl_src_param_t src_param;
1565
1566 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src_param);
1567
1568 if (shader->baseShader.hex_version < WINED3DPS_VERSION(2,0)) {
1569 char dst_mask[6];
1570
1571 shader_addline(arg->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
1572 shader_addline(arg->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
1573 shader_addline(arg->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
1574 shader_addline(arg->buffer, "tmp0.w = 1.0;\n");
1575
1576 shader_glsl_append_dst(arg->buffer, arg);
1577 shader_glsl_get_write_mask(arg->dst, dst_mask);
1578 shader_addline(arg->buffer, "tmp0%s);\n", dst_mask);
1579 } else {
1580 DWORD write_mask;
1581 unsigned int mask_size;
1582
1583 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1584 mask_size = shader_glsl_get_write_mask_size(write_mask);
1585
1586 if (mask_size > 1) {
1587 shader_addline(arg->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
1588 } else {
1589 shader_addline(arg->buffer, "exp2(%s));\n", src_param.param_str);
1590 }
1591 }
1592 }
1593
1594 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
1595 void shader_glsl_rcp(SHADER_OPCODE_ARG* arg) {
1596 glsl_src_param_t src_param;
1597 DWORD write_mask;
1598 unsigned int mask_size;
1599
1600 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1601 mask_size = shader_glsl_get_write_mask_size(write_mask);
1602 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src_param);
1603
1604 if (mask_size > 1) {
1605 shader_addline(arg->buffer, "vec%d(1.0 / %s));\n", mask_size, src_param.param_str);
1606 } else {
1607 shader_addline(arg->buffer, "1.0 / %s);\n", src_param.param_str);
1608 }
1609 }
1610
1611 void shader_glsl_rsq(SHADER_OPCODE_ARG* arg) {
1612 SHADER_BUFFER* buffer = arg->buffer;
1613 glsl_src_param_t src_param;
1614 DWORD write_mask;
1615 unsigned int mask_size;
1616
1617 write_mask = shader_glsl_append_dst(buffer, arg);
1618 mask_size = shader_glsl_get_write_mask_size(write_mask);
1619
1620 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src_param);
1621
1622 if (mask_size > 1) {
1623 shader_addline(buffer, "vec%d(inversesqrt(%s)));\n", mask_size, src_param.param_str);
1624 } else {
1625 shader_addline(buffer, "inversesqrt(%s));\n", src_param.param_str);
1626 }
1627 }
1628
1629 /** Process signed comparison opcodes in GLSL. */
1630 void shader_glsl_compare(SHADER_OPCODE_ARG* arg) {
1631 glsl_src_param_t src0_param;
1632 glsl_src_param_t src1_param;
1633 DWORD write_mask;
1634 unsigned int mask_size;
1635
1636 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1637 mask_size = shader_glsl_get_write_mask_size(write_mask);
1638 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1639 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1640
1641 if (mask_size > 1) {
1642 const char *compare;
1643
1644 switch(arg->opcode->opcode) {
1645 case WINED3DSIO_SLT: compare = "lessThan"; break;
1646 case WINED3DSIO_SGE: compare = "greaterThanEqual"; break;
1647 default: compare = "";
1648 FIXME("Can't handle opcode %s\n", arg->opcode->name);
1649 }
1650
1651 shader_addline(arg->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
1652 src0_param.param_str, src1_param.param_str);
1653 } else {
1654 switch(arg->opcode->opcode) {
1655 case WINED3DSIO_SLT:
1656 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
1657 * to return 0.0 but step returns 1.0 because step is not < x
1658 * An alternative is a bvec compare padded with an unused second component.
1659 * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
1660 * issue. Playing with not() is not possible either because not() does not accept
1661 * a scalar.
1662 */
1663 shader_addline(arg->buffer, "(%s < %s) ? 1.0 : 0.0);\n", src0_param.param_str, src1_param.param_str);
1664 break;
1665 case WINED3DSIO_SGE:
1666 /* Here we can use the step() function and safe a conditional */
1667 shader_addline(arg->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
1668 break;
1669 default:
1670 FIXME("Can't handle opcode %s\n", arg->opcode->name);
1671 }
1672
1673 }
1674 }
1675
1676 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
1677 void shader_glsl_cmp(SHADER_OPCODE_ARG* arg) {
1678 glsl_src_param_t src0_param;
1679 glsl_src_param_t src1_param;
1680 glsl_src_param_t src2_param;
1681 DWORD write_mask, cmp_channel = 0;
1682 unsigned int i, j;
1683 char mask_char[6];
1684 BOOL temp_destination = FALSE;
1685
1686 if(shader_is_scalar(arg->src[0])) {
1687 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1688
1689 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
1690 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1691 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1692
1693 shader_addline(arg->buffer, "%s >= 0.0 ? %s : %s);\n",
1694 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1695 } else {
1696 DWORD src0reg = arg->src[0] & WINED3DSP_REGNUM_MASK;
1697 DWORD src1reg = arg->src[1] & WINED3DSP_REGNUM_MASK;
1698 DWORD src2reg = arg->src[2] & WINED3DSP_REGNUM_MASK;
1699 DWORD src0regtype = shader_get_regtype(arg->src[0]);
1700 DWORD src1regtype = shader_get_regtype(arg->src[1]);
1701 DWORD src2regtype = shader_get_regtype(arg->src[2]);
1702 DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
1703 DWORD dstregtype = shader_get_regtype(arg->dst);
1704
1705 /* Cycle through all source0 channels */
1706 for (i=0; i<4; i++) {
1707 write_mask = 0;
1708 /* Find the destination channels which use the current source0 channel */
1709 for (j=0; j<4; j++) {
1710 if ( ((arg->src[0] >> (WINED3DSP_SWIZZLE_SHIFT + 2*j)) & 0x3) == i ) {
1711 write_mask |= WINED3DSP_WRITEMASK_0 << j;
1712 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
1713 }
1714 }
1715
1716 /* Splitting the cmp instruction up in multiple lines imposes a problem:
1717 * The first lines may overwrite source parameters of the following lines.
1718 * Deal with that by using a temporary destination register if needed
1719 */
1720 if((src0reg == dstreg && src0regtype == dstregtype) ||
1721 (src1reg == dstreg && src1regtype == dstregtype) ||
1722 (src2reg == dstreg && src2regtype == dstregtype)) {
1723
1724 write_mask = shader_glsl_get_write_mask(arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask), mask_char);
1725 if (!write_mask) continue;
1726 shader_addline(arg->buffer, "tmp0%s = (", mask_char);
1727 temp_destination = TRUE;
1728 } else {
1729 write_mask = shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask));
1730 if (!write_mask) continue;
1731 }
1732
1733 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], cmp_channel, &src0_param);
1734 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1735 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1736
1737 shader_addline(arg->buffer, "%s >= 0.0 ? %s : %s);\n",
1738 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1739 }
1740
1741 if(temp_destination) {
1742 shader_glsl_get_write_mask(arg->dst, mask_char);
1743 shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst);
1744 shader_addline(arg->buffer, "tmp0%s);\n", mask_char);
1745 }
1746 }
1747
1748 }
1749
1750 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
1751 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
1752 * the compare is done per component of src0. */
1753 void shader_glsl_cnd(SHADER_OPCODE_ARG* arg) {
1754 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1755 glsl_src_param_t src0_param;
1756 glsl_src_param_t src1_param;
1757 glsl_src_param_t src2_param;
1758 DWORD write_mask, cmp_channel = 0;
1759 unsigned int i, j;
1760
1761 if (shader->baseShader.hex_version < WINED3DPS_VERSION(1, 4)) {
1762 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1763 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1764 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1765 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1766
1767 /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
1768 if(arg->opcode_token & WINED3DSI_COISSUE) {
1769 shader_addline(arg->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
1770 } else {
1771 shader_addline(arg->buffer, "%s > 0.5 ? %s : %s);\n",
1772 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1773 }
1774 return;
1775 }
1776 /* Cycle through all source0 channels */
1777 for (i=0; i<4; i++) {
1778 write_mask = 0;
1779 /* Find the destination channels which use the current source0 channel */
1780 for (j=0; j<4; j++) {
1781 if ( ((arg->src[0] >> (WINED3DSP_SWIZZLE_SHIFT + 2*j)) & 0x3) == i ) {
1782 write_mask |= WINED3DSP_WRITEMASK_0 << j;
1783 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
1784 }
1785 }
1786 write_mask = shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask));
1787 if (!write_mask) continue;
1788
1789 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], cmp_channel, &src0_param);
1790 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1791 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1792
1793 shader_addline(arg->buffer, "%s > 0.5 ? %s : %s);\n",
1794 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1795 }
1796 }
1797
1798 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
1799 void shader_glsl_mad(SHADER_OPCODE_ARG* arg) {
1800 glsl_src_param_t src0_param;
1801 glsl_src_param_t src1_param;
1802 glsl_src_param_t src2_param;
1803 DWORD write_mask;
1804
1805 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1806 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1807 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1808 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1809 shader_addline(arg->buffer, "(%s * %s) + %s);\n",
1810 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1811 }
1812
1813 /** Handles transforming all WINED3DSIO_M?x? opcodes for
1814 Vertex shaders to GLSL codes */
1815 void shader_glsl_mnxn(SHADER_OPCODE_ARG* arg) {
1816 int i;
1817 int nComponents = 0;
1818 SHADER_OPCODE_ARG tmpArg;
1819
1820 memset(&tmpArg, 0, sizeof(SHADER_OPCODE_ARG));
1821
1822 /* Set constants for the temporary argument */
1823 tmpArg.shader = arg->shader;
1824 tmpArg.buffer = arg->buffer;
1825 tmpArg.src[0] = arg->src[0];
1826 tmpArg.src_addr[0] = arg->src_addr[0];
1827 tmpArg.src_addr[1] = arg->src_addr[1];
1828 tmpArg.reg_maps = arg->reg_maps;
1829
1830 switch(arg->opcode->opcode) {
1831 case WINED3DSIO_M4x4:
1832 nComponents = 4;
1833 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP4);
1834 break;
1835 case WINED3DSIO_M4x3:
1836 nComponents = 3;
1837 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP4);
1838 break;
1839 case WINED3DSIO_M3x4:
1840 nComponents = 4;
1841 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
1842 break;
1843 case WINED3DSIO_M3x3:
1844 nComponents = 3;
1845 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
1846 break;
1847 case WINED3DSIO_M3x2:
1848 nComponents = 2;
1849 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
1850 break;
1851 default:
1852 break;
1853 }
1854
1855 for (i = 0; i < nComponents; i++) {
1856 tmpArg.dst = ((arg->dst) & ~WINED3DSP_WRITEMASK_ALL)|(WINED3DSP_WRITEMASK_0<<i);
1857 tmpArg.src[1] = arg->src[1]+i;
1858 shader_glsl_dot(&tmpArg);
1859 }
1860 }
1861
1862 /**
1863 The LRP instruction performs a component-wise linear interpolation
1864 between the second and third operands using the first operand as the
1865 blend factor. Equation: (dst = src2 + src0 * (src1 - src2))
1866 This is equivalent to mix(src2, src1, src0);
1867 */
1868 void shader_glsl_lrp(SHADER_OPCODE_ARG* arg) {
1869 glsl_src_param_t src0_param;
1870 glsl_src_param_t src1_param;
1871 glsl_src_param_t src2_param;
1872 DWORD write_mask;
1873
1874 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1875
1876 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1877 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1878 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1879
1880 shader_addline(arg->buffer, "mix(%s, %s, %s));\n",
1881 src2_param.param_str, src1_param.param_str, src0_param.param_str);
1882 }
1883
1884 /** Process the WINED3DSIO_LIT instruction in GLSL:
1885 * dst.x = dst.w = 1.0
1886 * dst.y = (src0.x > 0) ? src0.x
1887 * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
1888 * where src.w is clamped at +- 128
1889 */
1890 void shader_glsl_lit(SHADER_OPCODE_ARG* arg) {
1891 glsl_src_param_t src0_param;
1892 glsl_src_param_t src1_param;
1893 glsl_src_param_t src3_param;
1894 char dst_mask[6];
1895
1896 shader_glsl_append_dst(arg->buffer, arg);
1897 shader_glsl_get_write_mask(arg->dst, dst_mask);
1898
1899 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1900 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_1, &src1_param);
1901 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src3_param);
1902
1903 /* The sdk specifies the instruction like this
1904 * dst.x = 1.0;
1905 * if(src.x > 0.0) dst.y = src.x
1906 * else dst.y = 0.0.
1907 * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
1908 * else dst.z = 0.0;
1909 * dst.w = 1.0;
1910 *
1911 * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
1912 * dst.x = 1.0 ... No further explanation needed
1913 * dst.y = max(src.y, 0.0); ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
1914 * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0; ... 0 ^ power is 0, and otherwise we use y anyway
1915 * dst.w = 1.0. ... Nothing fancy.
1916 *
1917 * So we still have one conditional in there. So do this:
1918 * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
1919 *
1920 * step(0.0, x) will return 1 if src.x > 0.0, and 0 otherwise. So if y is 0 we get pow(0.0 * 1.0, power),
1921 * which sets dst.z to 0. If y > 0, but x = 0.0, we get pow(y * 0.0, power), which results in 0 too.
1922 * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to
1923 */
1924 shader_addline(arg->buffer, "vec4(1.0, max(%s, 0.0), pow(max(0.0, %s) * step(0.0, %s), clamp(%s, -128.0, 128.0)), 1.0)%s);\n",
1925 src0_param.param_str, src1_param.param_str, src0_param.param_str, src3_param.param_str, dst_mask);
1926 }
1927
1928 /** Process the WINED3DSIO_DST instruction in GLSL:
1929 * dst.x = 1.0
1930 * dst.y = src0.x * src0.y
1931 * dst.z = src0.z
1932 * dst.w = src1.w
1933 */
1934 void shader_glsl_dst(SHADER_OPCODE_ARG* arg) {
1935 glsl_src_param_t src0y_param;
1936 glsl_src_param_t src0z_param;
1937 glsl_src_param_t src1y_param;
1938 glsl_src_param_t src1w_param;
1939 char dst_mask[6];
1940
1941 shader_glsl_append_dst(arg->buffer, arg);
1942 shader_glsl_get_write_mask(arg->dst, dst_mask);
1943
1944 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_1, &src0y_param);
1945 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &src0z_param);
1946 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_1, &src1y_param);
1947 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_3, &src1w_param);
1948
1949 shader_addline(arg->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
1950 src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
1951 }
1952
1953 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
1954 * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
1955 * can handle it. But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
1956 *
1957 * dst.x = cos(src0.?)
1958 * dst.y = sin(src0.?)
1959 * dst.z = dst.z
1960 * dst.w = dst.w
1961 */
1962 void shader_glsl_sincos(SHADER_OPCODE_ARG* arg) {
1963 glsl_src_param_t src0_param;
1964 DWORD write_mask;
1965
1966 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1967 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1968
1969 switch (write_mask) {
1970 case WINED3DSP_WRITEMASK_0:
1971 shader_addline(arg->buffer, "cos(%s));\n", src0_param.param_str);
1972 break;
1973
1974 case WINED3DSP_WRITEMASK_1:
1975 shader_addline(arg->buffer, "sin(%s));\n", src0_param.param_str);
1976 break;
1977
1978 case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
1979 shader_addline(arg->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
1980 break;
1981
1982 default:
1983 ERR("Write mask should be .x, .y or .xy\n");
1984 break;
1985 }
1986 }
1987
1988 /** Process the WINED3DSIO_LOOP instruction in GLSL:
1989 * Start a for() loop where src1.y is the initial value of aL,
1990 * increment aL by src1.z for a total of src1.x iterations.
1991 * Need to use a temporary variable for this operation.
1992 */
1993 /* FIXME: I don't think nested loops will work correctly this way. */
1994 void shader_glsl_loop(SHADER_OPCODE_ARG* arg) {
1995 glsl_src_param_t src1_param;
1996 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1997 DWORD regtype = shader_get_regtype(arg->src[1]);
1998 DWORD reg = arg->src[1] & WINED3DSP_REGNUM_MASK;
1999 const DWORD *control_values = NULL;
2000 local_constant *constant;
2001
2002 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2003
2004 /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2005 * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2006 * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2007 * addressing.
2008 */
2009 if(regtype == WINED3DSPR_CONSTINT) {
2010 LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
2011 if(constant->idx == reg) {
2012 control_values = constant->value;
2013 break;
2014 }
2015 }
2016 }
2017
2018 if(control_values) {
2019 if(control_values[2] > 0) {
2020 shader_addline(arg->buffer, "for (aL%u = %d; aL%u < (%d * %d + %d); aL%u += %d) {\n",
2021 shader->baseShader.cur_loop_depth, control_values[1],
2022 shader->baseShader.cur_loop_depth, control_values[0], control_values[2], control_values[1],
2023 shader->baseShader.cur_loop_depth, control_values[2]);
2024 } else if(control_values[2] == 0) {
2025 shader_addline(arg->buffer, "for (aL%u = %d, tmpInt%u = 0; tmpInt%u < %d; tmpInt%u++) {\n",
2026 shader->baseShader.cur_loop_depth, control_values[1], shader->baseShader.cur_loop_depth,
2027 shader->baseShader.cur_loop_depth, control_values[0],
2028 shader->baseShader.cur_loop_depth);
2029 } else {
2030 shader_addline(arg->buffer, "for (aL%u = %d; aL%u > (%d * %d + %d); aL%u += %d) {\n",
2031 shader->baseShader.cur_loop_depth, control_values[1],
2032 shader->baseShader.cur_loop_depth, control_values[0], control_values[2], control_values[1],
2033 shader->baseShader.cur_loop_depth, control_values[2]);
2034 }
2035 } else {
2036 shader_addline(arg->buffer, "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2037 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno,
2038 src1_param.reg_name, shader->baseShader.cur_loop_depth, src1_param.reg_name,
2039 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno, src1_param.reg_name);
2040 }
2041
2042 shader->baseShader.cur_loop_depth++;
2043 shader->baseShader.cur_loop_regno++;
2044 }
2045
2046 void shader_glsl_end(SHADER_OPCODE_ARG* arg) {
2047 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
2048
2049 shader_addline(arg->buffer, "}\n");
2050
2051 if(arg->opcode->opcode == WINED3DSIO_ENDLOOP) {
2052 shader->baseShader.cur_loop_depth--;
2053 shader->baseShader.cur_loop_regno--;
2054 }
2055 if(arg->opcode->opcode == WINED3DSIO_ENDREP) {
2056 shader->baseShader.cur_loop_depth--;
2057 }
2058 }
2059
2060 void shader_glsl_rep(SHADER_OPCODE_ARG* arg) {
2061 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
2062 glsl_src_param_t src0_param;
2063
2064 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2065 shader_addline(arg->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2066 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2067 src0_param.param_str, shader->baseShader.cur_loop_depth);
2068 shader->baseShader.cur_loop_depth++;
2069 }
2070
2071 void shader_glsl_if(SHADER_OPCODE_ARG* arg) {
2072 glsl_src_param_t src0_param;
2073
2074 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2075 shader_addline(arg->buffer, "if (%s) {\n", src0_param.param_str);
2076 }
2077
2078 void shader_glsl_ifc(SHADER_OPCODE_ARG* arg) {
2079 glsl_src_param_t src0_param;
2080 glsl_src_param_t src1_param;
2081
2082 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2083 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2084
2085 shader_addline(arg->buffer, "if (%s %s %s) {\n",
2086 src0_param.param_str, shader_get_comp_op(arg->opcode_token), src1_param.param_str);
2087 }
2088
2089 void shader_glsl_else(SHADER_OPCODE_ARG* arg) {
2090 shader_addline(arg->buffer, "} else {\n");
2091 }
2092
2093 void shader_glsl_break(SHADER_OPCODE_ARG* arg) {
2094 shader_addline(arg->buffer, "break;\n");
2095 }
2096
2097 /* FIXME: According to MSDN the compare is done per component. */
2098 void shader_glsl_breakc(SHADER_OPCODE_ARG* arg) {
2099 glsl_src_param_t src0_param;
2100 glsl_src_param_t src1_param;
2101
2102 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2103 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2104
2105 shader_addline(arg->buffer, "if (%s %s %s) break;\n",
2106 src0_param.param_str, shader_get_comp_op(arg->opcode_token), src1_param.param_str);
2107 }
2108
2109 void shader_glsl_label(SHADER_OPCODE_ARG* arg) {
2110
2111 DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2112 shader_addline(arg->buffer, "}\n");
2113 shader_addline(arg->buffer, "void subroutine%u () {\n", snum);
2114 }
2115
2116 void shader_glsl_call(SHADER_OPCODE_ARG* arg) {
2117 DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2118 shader_addline(arg->buffer, "subroutine%u();\n", snum);
2119 }
2120
2121 void shader_glsl_callnz(SHADER_OPCODE_ARG* arg) {
2122 glsl_src_param_t src1_param;
2123
2124 DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2125 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2126 shader_addline(arg->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, snum);
2127 }
2128
2129 /*********************************************
2130 * Pixel Shader Specific Code begins here
2131 ********************************************/
2132 void pshader_glsl_tex(SHADER_OPCODE_ARG* arg) {
2133 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2134 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2135 DWORD hex_version = This->baseShader.hex_version;
2136 char dst_swizzle[6];
2137 glsl_sample_function_t sample_function;
2138 DWORD sampler_type;
2139 DWORD sampler_idx;
2140 BOOL projected, texrect = FALSE;
2141 DWORD mask = 0;
2142
2143 /* All versions have a destination register */
2144 shader_glsl_append_dst(arg->buffer, arg);
2145
2146 /* 1.0-1.4: Use destination register as sampler source.
2147 * 2.0+: Use provided sampler source. */
2148 if (hex_version < WINED3DPS_VERSION(1,4)) {
2149 DWORD flags;
2150
2151 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2152 flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2153
2154 if (flags & WINED3DTTFF_PROJECTED) {
2155 projected = TRUE;
2156 switch (flags & ~WINED3DTTFF_PROJECTED) {
2157 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2158 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
2159 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
2160 case WINED3DTTFF_COUNT4:
2161 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
2162 }
2163 } else {
2164 projected = FALSE;
2165 }
2166 } else if (hex_version < WINED3DPS_VERSION(2,0)) {
2167 DWORD src_mod = arg->src[0] & WINED3DSP_SRCMOD_MASK;
2168 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2169
2170 if (src_mod == WINED3DSPSM_DZ) {
2171 projected = TRUE;
2172 mask = WINED3DSP_WRITEMASK_2;
2173 } else if (src_mod == WINED3DSPSM_DW) {
2174 projected = TRUE;
2175 mask = WINED3DSP_WRITEMASK_3;
2176 } else {
2177 projected = FALSE;
2178 }
2179 } else {
2180 sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
2181 if(arg->opcode_token & WINED3DSI_TEXLD_PROJECT) {
2182 /* ps 2.0 texldp instruction always divides by the fourth component. */
2183 projected = TRUE;
2184 mask = WINED3DSP_WRITEMASK_3;
2185 } else {
2186 projected = FALSE;
2187 }
2188 }
2189
2190 if(deviceImpl->stateBlock->textures[sampler_idx] &&
2191 IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2192 texrect = TRUE;
2193 }
2194
2195 sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2196 shader_glsl_get_sample_function(sampler_type, projected, texrect, &sample_function);
2197 mask |= sample_function.coord_mask;
2198
2199 if (hex_version < WINED3DPS_VERSION(2,0)) {
2200 shader_glsl_get_write_mask(arg->dst, dst_swizzle);
2201 } else {
2202 shader_glsl_get_swizzle(arg->src[1], FALSE, arg->dst, dst_swizzle);
2203 }
2204
2205 /* 1.0-1.3: Use destination register as coordinate source.
2206 1.4+: Use provided coordinate source register. */
2207 if (hex_version < WINED3DPS_VERSION(1,4)) {
2208 char coord_mask[6];
2209 shader_glsl_get_write_mask(mask, coord_mask);
2210 shader_addline(arg->buffer, "%s(Psampler%u, T%u%s)%s);\n",
2211 sample_function.name, sampler_idx, sampler_idx, coord_mask, dst_swizzle);
2212 } else {
2213 glsl_src_param_t coord_param;
2214 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], mask, &coord_param);
2215 if(arg->opcode_token & WINED3DSI_TEXLD_BIAS) {
2216 glsl_src_param_t bias;
2217 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &bias);
2218
2219 shader_addline(arg->buffer, "%s(Psampler%u, %s, %s)%s);\n",
2220 sample_function.name, sampler_idx, coord_param.param_str,
2221 bias.param_str, dst_swizzle);
2222 } else {
2223 shader_addline(arg->buffer, "%s(Psampler%u, %s)%s);\n",
2224 sample_function.name, sampler_idx, coord_param.param_str, dst_swizzle);
2225 }
2226 }
2227 }
2228
2229 void shader_glsl_texldl(SHADER_OPCODE_ARG* arg) {
2230 IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*)arg->shader;
2231 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2232 glsl_sample_function_t sample_function;
2233 glsl_src_param_t coord_param, lod_param;
2234 char dst_swizzle[6];
2235 DWORD sampler_type;
2236 DWORD sampler_idx;
2237 BOOL texrect = FALSE;
2238
2239 shader_glsl_append_dst(arg->buffer, arg);
2240 shader_glsl_get_swizzle(arg->src[1], FALSE, arg->dst, dst_swizzle);
2241
2242 sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
2243 sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2244 if(deviceImpl->stateBlock->textures[sampler_idx] &&
2245 IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2246 texrect = TRUE;
2247 }
2248 shader_glsl_get_sample_function(sampler_type, FALSE, texrect, &sample_function); shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], sample_function.coord_mask, &coord_param);
2249
2250 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &lod_param);
2251
2252 if (shader_is_pshader_version(This->baseShader.hex_version)) {
2253 /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
2254 * However, they seem to work just fine in fragment shaders as well. */
2255 WARN("Using %sLod in fragment shader.\n", sample_function.name);
2256 shader_addline(arg->buffer, "%sLod(Psampler%u, %s, %s)%s);\n",
2257 sample_function.name, sampler_idx, coord_param.param_str, lod_param.param_str, dst_swizzle);
2258 } else {
2259 shader_addline(arg->buffer, "%sLod(Vsampler%u, %s, %s)%s);\n",
2260 sample_function.name, sampler_idx, coord_param.param_str, lod_param.param_str, dst_swizzle);
2261 }
2262 }
2263
2264 void pshader_glsl_texcoord(SHADER_OPCODE_ARG* arg) {
2265
2266 /* FIXME: Make this work for more than just 2D textures */
2267
2268 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2269 SHADER_BUFFER* buffer = arg->buffer;
2270 DWORD hex_version = This->baseShader.hex_version;
2271 DWORD write_mask;
2272 char dst_mask[6];
2273
2274 write_mask = shader_glsl_append_dst(arg->buffer, arg);
2275 shader_glsl_get_write_mask(write_mask, dst_mask);
2276
2277 if (hex_version != WINED3DPS_VERSION(1,4)) {
2278 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2279 shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n", reg, dst_mask);
2280 } else {
2281 DWORD reg = arg->src[0] & WINED3DSP_REGNUM_MASK;
2282 DWORD src_mod = arg->src[0] & WINED3DSP_SRCMOD_MASK;
2283 char dst_swizzle[6];
2284
2285 shader_glsl_get_swizzle(arg->src[0], FALSE, write_mask, dst_swizzle);
2286
2287 if (src_mod == WINED3DSPSM_DZ) {
2288 glsl_src_param_t div_param;
2289 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2290 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &div_param);
2291
2292 if (mask_size > 1) {
2293 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2294 } else {
2295 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2296 }
2297 } else if (src_mod == WINED3DSPSM_DW) {
2298 glsl_src_param_t div_param;
2299 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2300 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &div_param);
2301
2302 if (mask_size > 1) {
2303 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2304 } else {
2305 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2306 }
2307 } else {
2308 shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
2309 }
2310 }
2311 }
2312
2313 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
2314 * Take a 3-component dot product of the TexCoord[dstreg] and src,
2315 * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2316 void pshader_glsl_texdp3tex(SHADER_OPCODE_ARG* arg) {
2317 glsl_src_param_t src0_param;
2318 char dst_mask[6];
2319 glsl_sample_function_t sample_function;
2320 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2321 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2322 DWORD sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2323
2324 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2325
2326 shader_glsl_append_dst(arg->buffer, arg);
2327 shader_glsl_get_write_mask(arg->dst, dst_mask);
2328
2329 /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
2330 * scalar, and projected sampling would require 4.
2331 *
2332 * It is a dependent read - not valid with conditional NP2 textures
2333 */
2334 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2335
2336 switch(count_bits(sample_function.coord_mask)) {
2337 case 1:
2338 shader_addline(arg->buffer, "%s(Psampler%u, dot(gl_TexCoord[%u].xyz, %s))%s);\n",
2339 sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2340 break;
2341
2342 case 2:
2343 shader_addline(arg->buffer, "%s(Psampler%u, vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0))%s);\n",
2344 sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2345 break;
2346
2347 case 3:
2348 shader_addline(arg->buffer, "%s(Psampler%u, vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0))%s);\n",
2349 sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2350 break;
2351 default:
2352 FIXME("Unexpected mask bitcount %d\n", count_bits(sample_function.coord_mask));
2353 }
2354 }
2355
2356 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
2357 * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2358 void pshader_glsl_texdp3(SHADER_OPCODE_ARG* arg) {
2359 glsl_src_param_t src0_param;
2360 DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
2361 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2362 DWORD dst_mask;
2363 unsigned int mask_size;
2364
2365 dst_mask = shader_glsl_append_dst(arg->buffer, arg);
2366 mask_size = shader_glsl_get_write_mask_size(dst_mask);
2367 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2368
2369 if (mask_size > 1) {
2370 shader_addline(arg->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
2371 } else {
2372 shader_addline(arg->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
2373 }
2374 }
2375
2376 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
2377 * Calculate the depth as dst.x / dst.y */
2378 void pshader_glsl_texdepth(SHADER_OPCODE_ARG* arg) {
2379 glsl_dst_param_t dst_param;
2380
2381 shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
2382
2383 /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
2384 * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
2385 * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
2386 * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
2387 * >= 1.0 or < 0.0
2388 */
2389 shader_addline(arg->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n", dst_param.reg_name, dst_param.reg_name);
2390 }
2391
2392 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
2393 * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2394 * Calculate tmp0.y = TexCoord[dstreg] . src.xyz; (tmp0.x has already been calculated)
2395 * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2396 */
2397 void pshader_glsl_texm3x2depth(SHADER_OPCODE_ARG* arg) {
2398 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2399 DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
2400 glsl_src_param_t src0_param;
2401
2402 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2403
2404 shader_addline(arg->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
2405 shader_addline(arg->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
2406 }
2407
2408 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
2409 * Calculate the 1st of a 2-row matrix multiplication. */
2410 void pshader_glsl_texm3x2pad(SHADER_OPCODE_ARG* arg) {
2411 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2412 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2413 SHADER_BUFFER* buffer = arg->buffer;
2414 glsl_src_param_t src0_param;
2415
2416 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2417 shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2418 }
2419
2420 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
2421 * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
2422 void pshader_glsl_texm3x3pad(SHADER_OPCODE_ARG* arg) {
2423
2424 IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2425 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2426 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2427 SHADER_BUFFER* buffer = arg->buffer;
2428 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2429 glsl_src_param_t src0_param;
2430
2431 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2432 shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + current_state->current_row, reg, src0_param.param_str);
2433 current_state->texcoord_w[current_state->current_row++] = reg;
2434 }
2435
2436 void pshader_glsl_texm3x2tex(SHADER_OPCODE_ARG* arg) {
2437 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2438 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2439 SHADER_BUFFER* buffer = arg->buffer;
2440 glsl_src_param_t src0_param;
2441 char dst_mask[6];
2442
2443 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2444 shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2445
2446 shader_glsl_append_dst(buffer, arg);
2447 shader_glsl_get_write_mask(arg->dst, dst_mask);
2448
2449 /* Sample the texture using the calculated coordinates */
2450 shader_addline(buffer, "texture2D(Psampler%u, tmp0.xy)%s);\n", reg, dst_mask);
2451 }
2452
2453 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
2454 * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
2455 void pshader_glsl_texm3x3tex(SHADER_OPCODE_ARG* arg) {
2456 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2457 glsl_src_param_t src0_param;
2458 char dst_mask[6];
2459 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2460 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2461 SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2462 DWORD sampler_type = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2463 glsl_sample_function_t sample_function;
2464
2465 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2466 shader_addline(arg->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2467
2468 shader_glsl_append_dst(arg->buffer, arg);
2469 shader_glsl_get_write_mask(arg->dst, dst_mask);
2470 /* Dependent read, not valid with conditional NP2 */
2471 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2472
2473 /* Sample the texture using the calculated coordinates */
2474 shader_addline(arg->buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2475
2476 current_state->current_row = 0;
2477 }
2478
2479 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
2480 * Perform the 3rd row of a 3x3 matrix multiply */
2481 void pshader_glsl_texm3x3(SHADER_OPCODE_ARG* arg) {
2482 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2483 glsl_src_param_t src0_param;
2484 char dst_mask[6];
2485 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2486 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2487 SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2488
2489 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2490
2491 shader_glsl_append_dst(arg->buffer, arg);
2492 shader_glsl_get_write_mask(arg->dst, dst_mask);
2493 shader_addline(arg->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
2494
2495 current_state->current_row = 0;
2496 }
2497
2498 /** Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
2499 * Peform the final texture lookup based on the previous 2 3x3 matrix multiplies */
2500 void pshader_glsl_texm3x3spec(SHADER_OPCODE_ARG* arg) {
2501
2502 IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2503 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2504 glsl_src_param_t src0_param;
2505 glsl_src_param_t src1_param;
2506 char dst_mask[6];
2507 SHADER_BUFFER* buffer = arg->buffer;
2508 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2509 DWORD stype = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2510 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2511 glsl_sample_function_t sample_function;
2512
2513 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2514 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_mask, &src1_param);
2515
2516 /* Perform the last matrix multiply operation */
2517 shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2518 /* Reflection calculation */
2519 shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
2520
2521 shader_glsl_append_dst(buffer, arg);
2522 shader_glsl_get_write_mask(arg->dst, dst_mask);
2523 /* Dependent read, not valid with conditional NP2 */
2524 shader_glsl_get_sample_function(stype, FALSE, FALSE, &sample_function);
2525
2526 /* Sample the texture */
2527 shader_addline(buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2528
2529 current_state->current_row = 0;
2530 }
2531
2532 /** Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
2533 * Peform the final texture lookup based on the previous 2 3x3 matrix multiplies */
2534 void pshader_glsl_texm3x3vspec(SHADER_OPCODE_ARG* arg) {
2535
2536 IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2537 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2538 SHADER_BUFFER* buffer = arg->buffer;
2539 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2540 glsl_src_param_t src0_param;
2541 char dst_mask[6];
2542 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2543 DWORD sampler_type = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2544 glsl_sample_function_t sample_function;
2545
2546 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2547
2548 /* Perform the last matrix multiply operation */
2549 shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
2550
2551 /* Construct the eye-ray vector from w coordinates */
2552 shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
2553 current_state->texcoord_w[0], current_state->texcoord_w[1], reg);
2554 shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
2555
2556 shader_glsl_append_dst(buffer, arg);
2557 shader_glsl_get_write_mask(arg->dst, dst_mask);
2558 /* Dependent read, not valid with conditional NP2 */
2559 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2560
2561 /* Sample the texture using the calculated coordinates */
2562 shader_addline(buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2563
2564 current_state->current_row = 0;
2565 }
2566
2567 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
2568 * Apply a fake bump map transform.
2569 * texbem is pshader <= 1.3 only, this saves a few version checks
2570 */
2571 void pshader_glsl_texbem(SHADER_OPCODE_ARG* arg) {
2572 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2573 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2574 char dst_swizzle[6];
2575 glsl_sample_function_t sample_function;
2576 glsl_src_param_t coord_param;
2577 DWORD sampler_type;
2578 DWORD sampler_idx;
2579 DWORD mask;
2580 DWORD flags;
2581 char coord_mask[6];
2582
2583 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2584 flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2585
2586 sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2587 /* Dependent read, not valid with conditional NP2 */
2588 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2589 mask = sample_function.coord_mask;
2590
2591 shader_glsl_get_write_mask(arg->dst, dst_swizzle);
2592
2593 shader_glsl_get_write_mask(mask, coord_mask);
2594
2595 /* with projective textures, texbem only divides the static texture coord, not the displacement,
2596 * so we can't let the GL handle this.
2597 */
2598 if (flags & WINED3DTTFF_PROJECTED) {
2599 DWORD div_mask=0;
2600 char coord_div_mask[3];
2601 switch (flags & ~WINED3DTTFF_PROJECTED) {
2602 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2603 case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
2604 case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
2605 case WINED3DTTFF_COUNT4:
2606 case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
2607 }
2608 shader_glsl_get_write_mask(div_mask, coord_div_mask);
2609 shader_addline(arg->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
2610 }
2611
2612 shader_glsl_append_dst(arg->buffer, arg);
2613 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &coord_param);
2614 if(arg->opcode->opcode == WINED3DSIO_TEXBEML) {
2615 glsl_src_param_t luminance_param;
2616 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &luminance_param);
2617 shader_addline(arg->buffer, "(%s(Psampler%u, T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s )*(%s * luminancescale%d + luminanceoffset%d))%s);\n",
2618 sample_function.name, sampler_idx, sampler_idx, coord_mask, sampler_idx, coord_param.param_str, coord_mask,
2619 luminance_param.param_str, sampler_idx, sampler_idx, dst_swizzle);
2620 } else {
2621 shader_addline(arg->buffer, "%s(Psampler%u, T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s )%s);\n",
2622 sample_function.name, sampler_idx, sampler_idx, coord_mask, sampler_idx, coord_param.param_str, coord_mask, dst_swizzle);
2623 }
2624 }
2625
2626 void pshader_glsl_bem(SHADER_OPCODE_ARG* arg) {
2627 glsl_src_param_t src0_param, src1_param;
2628 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2629
2630 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &src0_param);
2631 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &src1_param);
2632
2633 shader_glsl_append_dst(arg->buffer, arg);
2634 shader_addline(arg->buffer, "%s + bumpenvmat%d * %s);\n",
2635 src0_param.param_str, sampler_idx, src1_param.param_str);
2636 }
2637
2638 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
2639 * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
2640 void pshader_glsl_texreg2ar(SHADER_OPCODE_ARG* arg) {
2641
2642 glsl_src_param_t src0_param;
2643 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2644 char dst_mask[6];
2645
2646 shader_glsl_append_dst(arg->buffer, arg);
2647 shader_glsl_get_write_mask(arg->dst, dst_mask);
2648 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2649
2650 shader_addline(arg->buffer, "texture2D(Psampler%u, %s.wx)%s);\n", sampler_idx, src0_param.reg_name, dst_mask);
2651 }
2652
2653 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
2654 * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
2655 void pshader_glsl_texreg2gb(SHADER_OPCODE_ARG* arg) {
2656 glsl_src_param_t src0_param;
2657 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2658 char dst_mask[6];
2659
2660 shader_glsl_append_dst(arg->buffer, arg);
2661 shader_glsl_get_write_mask(arg->dst, dst_mask);
2662 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2663
2664 shader_addline(arg->buffer, "texture2D(Psampler%u, %s.yz)%s);\n", sampler_idx, src0_param.reg_name, dst_mask);
2665 }
2666
2667 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
2668 * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
2669 void pshader_glsl_texreg2rgb(SHADER_OPCODE_ARG* arg) {
2670 glsl_src_param_t src0_param;
2671 char dst_mask[6];
2672 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2673 DWORD sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2674 glsl_sample_function_t sample_function;
2675
2676 shader_glsl_append_dst(arg->buffer, arg);
2677 shader_glsl_get_write_mask(arg->dst, dst_mask);
2678 /* Dependent read, not valid with conditional NP2 */
2679 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2680 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], sample_function.coord_mask, &src0_param);
2681
2682 shader_addline(arg->buffer, "%s(Psampler%u, %s)%s);\n", sample_function.name, sampler_idx, src0_param.param_str, dst_mask);
2683 }
2684
2685 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
2686 * If any of the first 3 components are < 0, discard this pixel */
2687 void pshader_glsl_texkill(SHADER_OPCODE_ARG* arg) {
2688 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2689 DWORD hex_version = This->baseShader.hex_version;
2690 glsl_dst_param_t dst_param;
2691
2692 /* The argument is a destination parameter, and no writemasks are allowed */
2693 shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
2694 if((hex_version >= WINED3DPS_VERSION(2,0))) {
2695 /* 2.0 shaders compare all 4 components in texkill */
2696 shader_addline(arg->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
2697 } else {
2698 /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
2699 * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
2700 * 4 components are defined, only the first 3 are used
2701 */
2702 shader_addline(arg->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
2703 }
2704 }
2705
2706 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
2707 * dst = dot2(src0, src1) + src2 */
2708 void pshader_glsl_dp2add(SHADER_OPCODE_ARG* arg) {
2709 glsl_src_param_t src0_param;
2710 glsl_src_param_t src1_param;
2711 glsl_src_param_t src2_param;
2712 DWORD write_mask;
2713 unsigned int mask_size;
2714
2715 write_mask = shader_glsl_append_dst(arg->buffer, arg);
2716 mask_size = shader_glsl_get_write_mask_size(write_mask);
2717
2718 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
2719 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
2720 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], WINED3DSP_WRITEMASK_0, &src2_param);
2721
2722 if (mask_size > 1) {
2723 shader_addline(arg->buffer, "vec%d(dot(%s, %s) + %s));\n", mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
2724 } else {
2725 shader_addline(arg->buffer, "dot(%s, %s) + %s);\n", src0_param.param_str, src1_param.param_str, src2_param.param_str);
2726 }
2727 }
2728
2729 void pshader_glsl_input_pack(
2730 SHADER_BUFFER* buffer,
2731 semantic* semantics_in,
2732 IWineD3DPixelShader *iface) {
2733
2734 unsigned int i;
2735 IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *) iface;
2736
2737 for (i = 0; i < MAX_REG_INPUT; i++) {
2738
2739 DWORD usage_token = semantics_in[i].usage;
2740 DWORD register_token = semantics_in[i].reg;
2741 DWORD usage, usage_idx;
2742 char reg_mask[6];
2743
2744 /* Uninitialized */
2745 if (!usage_token) continue;
2746 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2747 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2748 shader_glsl_get_write_mask(register_token, reg_mask);
2749
2750 switch(usage) {
2751
2752 case WINED3DDECLUSAGE_TEXCOORD:
2753 if(usage_idx < 8 && This->vertexprocessing == pretransformed) {
2754 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
2755 This->input_reg_map[i], reg_mask, usage_idx, reg_mask);
2756 } else {
2757 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2758 This->input_reg_map[i], reg_mask, reg_mask);
2759 }
2760 break;
2761
2762 case WINED3DDECLUSAGE_COLOR:
2763 if (usage_idx == 0)
2764 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
2765 This->input_reg_map[i], reg_mask, reg_mask);
2766 else if (usage_idx == 1)
2767 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
2768 This->input_reg_map[i], reg_mask, reg_mask);
2769 else
2770 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2771 This->input_reg_map[i], reg_mask, reg_mask);
2772 break;
2773
2774 default:
2775 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2776 This->input_reg_map[i], reg_mask, reg_mask);
2777 }
2778 }
2779 }
2780
2781 /*********************************************
2782 * Vertex Shader Specific Code begins here
2783 ********************************************/
2784
2785 static void add_glsl_program_entry(IWineD3DDeviceImpl *device, struct glsl_shader_prog_link *entry) {
2786 glsl_program_key_t *key;
2787
2788 key = HeapAlloc(GetProcessHeap(), 0, sizeof(glsl_program_key_t));
2789 key->vshader = entry->vshader;
2790 key->pshader = entry->pshader;
2791
2792 hash_table_put(device->glsl_program_lookup, key, entry);
2793 }
2794
2795 static struct glsl_shader_prog_link *get_glsl_program_entry(IWineD3DDeviceImpl *device,
2796 GLhandleARB vshader, GLhandleARB pshader) {
2797 glsl_program_key_t key;
2798
2799 key.vshader = vshader;
2800 key.pshader = pshader;
2801
2802 return (struct glsl_shader_prog_link *)hash_table_get(device->glsl_program_lookup, &key);
2803 }
2804
2805 void delete_glsl_program_entry(IWineD3DDevice *iface, struct glsl_shader_prog_link *entry) {
2806 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
2807 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
2808 glsl_program_key_t *key;
2809
2810 key = HeapAlloc(GetProcessHeap(), 0, sizeof(glsl_program_key_t));
2811 key->vshader = entry->vshader;
2812 key->pshader = entry->pshader;
2813 hash_table_remove(This->glsl_program_lookup, key);
2814
2815 GL_EXTCALL(glDeleteObjectARB(entry->programId));
2816 if (entry->vshader) list_remove(&entry->vshader_entry);
2817 if (entry->pshader) list_remove(&entry->pshader_entry);
2818 HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
2819 HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
2820 HeapFree(GetProcessHeap(), 0, entry);
2821 }
2822
2823 static void handle_ps3_input(SHADER_BUFFER *buffer, semantic *semantics_in, semantic *semantics_out, WineD3D_GL_Info *gl_info, DWORD *map) {
2824 unsigned int i, j;
2825 DWORD usage_token, usage_token_out;
2826 DWORD register_token, register_token_out;
2827 DWORD usage, usage_idx, usage_out, usage_idx_out;
2828 DWORD *set;
2829 DWORD in_idx;
2830 DWORD in_count = GL_LIMITS(glsl_varyings) / 4;
2831 char reg_mask[6], reg_mask_out[6];
2832 char destination[50];
2833
2834 set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
2835
2836 if (!semantics_out) {
2837 /* Save gl_FrontColor & gl_FrontSecondaryColor before overwriting them. */
2838 shader_addline(buffer, "vec4 front_color = gl_FrontColor;\n");
2839 shader_addline(buffer, "vec4 front_secondary_color = gl_FrontSecondaryColor;\n");
2840 }
2841
2842 for(i = 0; i < MAX_REG_INPUT; i++) {
2843 usage_token = semantics_in[i].usage;
2844 if (!usage_token) continue;
2845
2846 in_idx = map[i];
2847 if (in_idx >= (in_count + 2)) {
2848 FIXME("More input varyings declared than supported, expect issues\n");
2849 continue;
2850 } else if(map[i] == -1) {
2851 /* Declared, but not read register */
2852 continue;
2853 }
2854
2855 if (in_idx == in_count) {
2856 sprintf(destination, "gl_FrontColor");
2857 } else if (in_idx == in_count + 1) {
2858 sprintf(destination, "gl_FrontSecondaryColor");
2859 } else {
2860 sprintf(destination, "IN[%u]", in_idx);
2861 }
2862
2863 register_token = semantics_in[i].reg;
2864
2865 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2866 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2867 set[map[i]] = shader_glsl_get_write_mask(register_token, reg_mask);
2868
2869 if(!semantics_out) {
2870 switch(usage) {
2871 case WINED3DDECLUSAGE_COLOR:
2872 if (usage_idx == 0)
2873 shader_addline(buffer, "%s%s = front_color%s;\n",
2874 destination, reg_mask, reg_mask);
2875 else if (usage_idx == 1)
2876 shader_addline(buffer, "%s%s = front_secondary_color%s;\n",
2877 destination, reg_mask, reg_mask);
2878 else
2879 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2880 destination, reg_mask, reg_mask);
2881 break;
2882
2883 case WINED3DDECLUSAGE_TEXCOORD:
2884 if (usage_idx < 8) {
2885 shader_addline(buffer, "%s%s = gl_TexCoord[%u]%s;\n",
2886 destination, reg_mask, usage_idx, reg_mask);
2887 } else {
2888 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2889 destination, reg_mask, reg_mask);
2890 }
2891 break;
2892
2893 case WINED3DDECLUSAGE_FOG:
2894 shader_addline(buffer, "%s%s = vec4(gl_FogFragCoord, 0.0, 0.0, 0.0)%s;\n",
2895 destination, reg_mask, reg_mask);
2896 break;
2897
2898 default:
2899 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2900 destination, reg_mask, reg_mask);
2901 }
2902 } else {
2903 BOOL found = FALSE;
2904 for(j = 0; j < MAX_REG_OUTPUT; j++) {
2905 usage_token_out = semantics_out[j].usage;
2906 if (!usage_token_out) continue;
2907 register_token_out = semantics_out[j].reg;
2908
2909 usage_out = (usage_token_out & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2910 usage_idx_out = (usage_token_out & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2911 shader_glsl_get_write_mask(register_token_out, reg_mask_out);
2912
2913 if(usage == usage_out &&
2914 usage_idx == usage_idx_out) {
2915 shader_addline(buffer, "%s%s = OUT[%u]%s;\n",
2916 destination, reg_mask, j, reg_mask);
2917 found = TRUE;
2918 }
2919 }
2920 if(!found) {
2921 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2922 destination, reg_mask, reg_mask);
2923 }
2924 }
2925 }
2926
2927 /* This is solely to make the compiler / linker happy and avoid warning about undefined
2928 * varyings. It shouldn't result in any real code executed on the GPU, since all read
2929 * input varyings are assigned above, if the optimizer works properly.
2930 */
2931 for(i = 0; i < in_count + 2; i++) {
2932 if(set[i] != WINED3DSP_WRITEMASK_ALL) {
2933 unsigned int size = 0;
2934 memset(reg_mask, 0, sizeof(reg_mask));
2935 if(!(set[i] & WINED3DSP_WRITEMASK_0)) {
2936 reg_mask[size] = 'x';
2937 size++;
2938 }
2939 if(!(set[i] & WINED3DSP_WRITEMASK_1)) {
2940 reg_mask[size] = 'y';
2941 size++;
2942 }
2943 if(!(set[i] & WINED3DSP_WRITEMASK_2)) {
2944 reg_mask[size] = 'z';
2945 size++;
2946 }
2947 if(!(set[i] & WINED3DSP_WRITEMASK_3)) {
2948 reg_mask[size] = 'w';
2949 size++;
2950 }
2951
2952 if (i == in_count) {
2953 sprintf(destination, "gl_FrontColor");
2954 } else if (i == in_count + 1) {
2955 sprintf(destination, "gl_FrontSecondaryColor");
2956 } else {
2957 sprintf(destination, "IN[%u]", i);
2958 }
2959
2960 if (size == 1) {
2961 shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
2962 } else {
2963 shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
2964 }
2965 }
2966 }
2967
2968 HeapFree(GetProcessHeap(), 0, set);
2969 }
2970
2971 static GLhandleARB generate_param_reorder_function(IWineD3DVertexShader *vertexshader,
2972 IWineD3DPixelShader *pixelshader,
2973 WineD3D_GL_Info *gl_info) {
2974 GLhandleARB ret = 0;
2975 IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) vertexshader;
2976 IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pixelshader;
2977 DWORD vs_major = vs ? WINED3DSHADER_VERSION_MAJOR(vs->baseShader.hex_version) : 0;
2978 DWORD ps_major = ps ? WINED3DSHADER_VERSION_MAJOR(ps->baseShader.hex_version) : 0;
2979 unsigned int i;
2980 SHADER_BUFFER buffer;
2981 DWORD usage_token;
2982 DWORD register_token;
2983 DWORD usage, usage_idx, writemask;
2984 char reg_mask[6];
2985 semantic *semantics_out, *semantics_in;
2986
2987 buffer.buffer = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, SHADER_PGMSIZE);
2988 buffer.bsize = 0;
2989 buffer.lineNo = 0;
2990 buffer.newline = TRUE;
2991
2992 shader_addline(&buffer, "#version 120\n");
2993
2994 if(vs_major < 3 && ps_major < 3) {
2995 /* That one is easy: The vertex shader writes to the builtin varyings, the pixel shader reads from them.
2996 * Take care about the texcoord .w fixup though if we're using the fixed function fragment pipeline
2997 */
2998 if((GLINFO_LOCATION).set_texcoord_w && ps_major == 0 && vs_major > 0) {
2999 shader_addline(&buffer, "void order_ps_input() {\n");
3000 for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
3001 if(vs->baseShader.reg_maps.texcoord_mask[i] != 0 &&
3002 vs->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
3003 shader_addline(&buffer, "gl_TexCoord[%u].w = 1.0;\n", i);
3004 }
3005 }
3006 shader_addline(&buffer, "}\n");
3007 } else {
3008 shader_addline(&buffer, "void order_ps_input() { /* do nothing */ }\n");
3009 }
3010 } else if(ps_major < 3 && vs_major >= 3) {
3011 /* The vertex shader writes to its own varyings, the pixel shader needs them in the builtin ones */
3012 semantics_out = vs->semantics_out;
3013
3014 shader_addline(&buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3015 for(i = 0; i < MAX_REG_OUTPUT; i++) {
3016 usage_token = semantics_out[i].usage;
3017 if (!usage_token) continue;
3018 register_token = semantics_out[i].reg;
3019
3020 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
3021 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
3022 writemask = shader_glsl_get_write_mask(register_token, reg_mask);
3023
3024 switch(usage) {
3025 case WINED3DDECLUSAGE_COLOR:
3026 if (usage_idx == 0)
3027 shader_addline(&buffer, "gl_FrontColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3028 else if (usage_idx == 1)
3029 shader_addline(&buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3030 break;
3031
3032 case WINED3DDECLUSAGE_POSITION:
3033 shader_addline(&buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3034 break;
3035
3036 case WINED3DDECLUSAGE_TEXCOORD:
3037 if (usage_idx < 8) {
3038 if(!(GLINFO_LOCATION).set_texcoord_w || ps_major > 0) writemask |= WINED3DSP_WRITEMASK_3;
3039
3040 shader_addline(&buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
3041 usage_idx, reg_mask, i, reg_mask);
3042 if(!(writemask & WINED3DSP_WRITEMASK_3)) {
3043 shader_addline(&buffer, "gl_TexCoord[%u].w = 1.0;\n", usage_idx);
3044 }
3045 }
3046 break;
3047
3048 case WINED3DDECLUSAGE_PSIZE:
3049 shader_addline(&buffer, "gl_PointSize = OUT[%u].x;\n", i);
3050 break;
3051
3052 case WINED3DDECLUSAGE_FOG:
3053 shader_addline(&buffer, "gl_FogFragCoord = OUT[%u].%c;\n", i, reg_mask[1]);
3054 break;
3055
3056 default:
3057 break;
3058 }
3059 }
3060 shader_addline(&buffer, "}\n");
3061
3062 } else if(ps_major >= 3 && vs_major >= 3) {
3063 semantics_out = vs->semantics_out;
3064 semantics_in = ps->semantics_in;
3065
3066 /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3067 shader_addline(&buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
3068 shader_addline(&buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3069
3070 /* First, sort out position and point size. Those are not passed to the pixel shader */
3071 for(i = 0; i < MAX_REG_OUTPUT; i++) {
3072 usage_token = semantics_out[i].usage;
3073 if (!usage_token) continue;
3074 register_token = semantics_out[i].reg;
3075
3076 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
3077 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
3078 shader_glsl_get_write_mask(register_token, reg_mask);
3079
3080 switch(usage) {
3081 case WINED3DDECLUSAGE_POSITION:
3082 shader_addline(&buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3083 break;
3084
3085 case WINED3DDECLUSAGE_PSIZE:
3086 shader_addline(&buffer, "gl_PointSize = OUT[%u].x;\n", i);
3087 break;
3088
3089 default:
3090 break;
3091 }
3092 }
3093
3094 /* Then, fix the pixel shader input */
3095 handle_ps3_input(&buffer, semantics_in, semantics_out, gl_info, ps->input_reg_map);
3096
3097 shader_addline(&buffer, "}\n");
3098 } else if(ps_major >= 3 && vs_major < 3) {
3099 semantics_in = ps->semantics_in;
3100
3101 shader_addline(&buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
3102 shader_addline(&buffer, "void order_ps_input() {\n");
3103 /* The vertex shader wrote to the builtin varyings. There is no need to figure out position and
3104 * point size, but we depend on the optimizers kindness to find out that the pixel shader doesn't
3105 * read gl_TexCoord and gl_ColorX, otherwise we'll run out of varyings
3106 */
3107 handle_ps3_input(&buffer, semantics_in, NULL, gl_info, ps->input_reg_map);
3108 shader_addline(&buffer, "}\n");
3109 } else {
3110 ERR("Unexpected vertex and pixel shader version condition: vs: %d, ps: %d\n", vs_major, ps_major);
3111 }
3112
3113 ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3114 checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3115 GL_EXTCALL(glShaderSourceARB(ret, 1, (const char**)&buffer.buffer, NULL));
3116 checkGLcall("glShaderSourceARB(ret, 1, (const char**)&buffer.buffer, NULL)");
3117 GL_EXTCALL(glCompileShaderARB(ret));
3118 checkGLcall("glCompileShaderARB(ret)");
3119
3120 HeapFree(GetProcessHeap(), 0, buffer.buffer);
3121 return ret;
3122 }
3123
3124 static void hardcode_local_constants(IWineD3DBaseShaderImpl *shader, WineD3D_GL_Info *gl_info, GLhandleARB programId, char prefix) {
3125 local_constant* lconst;
3126 GLuint tmp_loc;
3127 float *value;
3128 char glsl_name[8];
3129
3130 LIST_FOR_EACH_ENTRY(lconst, &shader->baseShader.constantsF, local_constant, entry) {
3131 value = (float *) lconst->value;
3132 snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
3133 tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3134 GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
3135 }
3136 checkGLcall("Hardcoding local constants\n");
3137 }
3138
3139 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
3140 * It sets the programId on the current StateBlock (because it should be called
3141 * inside of the DrawPrimitive() part of the render loop).
3142 *
3143 * If a program for the given combination does not exist, create one, and store
3144 * the program in the hash table. If it creates a program, it will link the
3145 * given objects, too.
3146 */
3147 static void set_glsl_shader_program(IWineD3DDevice *iface, BOOL use_ps, BOOL use_vs) {
3148 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3149 struct shader_glsl_priv *priv = (struct shader_glsl_priv *)This->shader_priv;
3150 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3151 IWineD3DPixelShader *pshader = This->stateBlock->pixelShader;
3152 IWineD3DVertexShader *vshader = This->stateBlock->vertexShader;
3153 struct glsl_shader_prog_link *entry = NULL;
3154 GLhandleARB programId = 0;
3155 GLhandleARB reorder_shader_id = 0;
3156 int i;
3157 char glsl_name[8];
3158
3159 GLhandleARB vshader_id = use_vs ? ((IWineD3DBaseShaderImpl*)vshader)->baseShader.prgId : 0;
3160 GLhandleARB pshader_id = use_ps ? ((IWineD3DBaseShaderImpl*)pshader)->baseShader.prgId : 0;
3161 entry = get_glsl_program_entry(This, vshader_id, pshader_id);
3162 if (entry) {
3163 priv->glsl_program = entry;
3164 return;
3165 }
3166
3167 /* If we get to this point, then no matching program exists, so we create one */
3168 programId = GL_EXTCALL(glCreateProgramObjectARB());
3169 TRACE("Created new GLSL shader program %u\n", programId);
3170
3171 /* Create the entry */
3172 entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
3173 entry->programId = programId;
3174 entry->vshader = vshader_id;
3175 entry->pshader = pshader_id;
3176 /* Add the hash table entry */
3177 add_glsl_program_entry(This, entry);
3178
3179 /* Set the current program */
3180 priv->glsl_program = entry;
3181
3182 /* Attach GLSL vshader */
3183 if (vshader_id) {
3184 int max_attribs = 16; /* TODO: Will this always be the case? It is at the moment... */
3185 char tmp_name[10];
3186
3187 reorder_shader_id = generate_param_reorder_function(vshader, pshader, gl_info);
3188 TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
3189 GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
3190 checkGLcall("glAttachObjectARB");
3191 /* Flag the reorder function for deletion, then it will be freed automatically when the program
3192 * is destroyed
3193 */
3194 GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
3195
3196 TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
3197 GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
3198 checkGLcall("glAttachObjectARB");
3199
3200 /* Bind vertex attributes to a corresponding index number to match
3201 * the same index numbers as ARB_vertex_programs (makes loading
3202 * vertex attributes simpler). With this method, we can use the
3203 * exact same code to load the attributes later for both ARB and
3204 * GLSL shaders.
3205 *
3206 * We have to do this here because we need to know the Program ID
3207 * in order to make the bindings work, and it has to be done prior
3208 * to linking the GLSL program. */
3209 for (i = 0; i < max_attribs; ++i) {
3210 if (((IWineD3DBaseShaderImpl*)vshader)->baseShader.reg_maps.attributes[i]) {
3211 snprintf(tmp_name, sizeof(tmp_name), "attrib%i", i);
3212 GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
3213 }
3214 }
3215 checkGLcall("glBindAttribLocationARB");
3216
3217 list_add_head(&((IWineD3DBaseShaderImpl *)vshader)->baseShader.linked_programs, &entry->vshader_entry);
3218 }
3219
3220 /* Attach GLSL pshader */
3221 if (pshader_id) {
3222 TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
3223 GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
3224 checkGLcall("glAttachObjectARB");
3225
3226 list_add_head(&((IWineD3DBaseShaderImpl *)pshader)->baseShader.linked_programs, &entry->pshader_entry);
3227 }
3228
3229 /* Link the program */
3230 TRACE("Linking GLSL shader program %u\n", programId);
3231 GL_EXTCALL(glLinkProgramARB(programId));
3232 print_glsl_info_log(&GLINFO_LOCATION, programId);
3233
3234 entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0, sizeof(GLhandleARB) * GL_LIMITS(vshader_constantsF));
3235 for (i = 0; i < GL_LIMITS(vshader_constantsF); ++i) {
3236 snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
3237 entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3238 }
3239 for (i = 0; i < MAX_CONST_I; ++i) {
3240 snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
3241 entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3242 }
3243 entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0, sizeof(GLhandleARB) * GL_LIMITS(pshader_constantsF));
3244 for (i = 0; i < GL_LIMITS(pshader_constantsF); ++i) {
3245 snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
3246 entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3247 }
3248 for (i = 0; i < MAX_CONST_I; ++i) {
3249 snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
3250 entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3251 }
3252
3253 if(pshader) {
3254 for(i = 0; i < ((IWineD3DPixelShaderImpl*)pshader)->numbumpenvmatconsts; i++) {
3255 char name[32];
3256 sprintf(name, "bumpenvmat%d", ((IWineD3DPixelShaderImpl*)pshader)->bumpenvmatconst[i].texunit);
3257 entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3258 sprintf(name, "luminancescale%d", ((IWineD3DPixelShaderImpl*)pshader)->luminanceconst[i].texunit);
3259 entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3260 sprintf(name, "luminanceoffset%d", ((IWineD3DPixelShaderImpl*)pshader)->luminanceconst[i].texunit);
3261 entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3262 }
3263 }
3264
3265
3266 entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
3267 entry->srgb_comparison_location = GL_EXTCALL(glGetUniformLocationARB(programId, "srgb_comparison"));
3268 entry->srgb_mul_low_location = GL_EXTCALL(glGetUniformLocationARB(programId, "srgb_mul_low"));
3269 entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
3270 checkGLcall("Find glsl program uniform locations");
3271
3272 if (pshader && WINED3DSHADER_VERSION_MAJOR(((IWineD3DPixelShaderImpl *)pshader)->baseShader.hex_version) >= 3
3273 && ((IWineD3DPixelShaderImpl *)pshader)->declared_in_count > GL_LIMITS(glsl_varyings) / 4) {
3274 TRACE("Shader %d needs vertex color clamping disabled\n", programId);
3275 entry->vertex_color_clamp = GL_FALSE;
3276 } else {
3277 entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
3278 }
3279
3280 /* Set the shader to allow uniform loading on it */
3281 GL_EXTCALL(glUseProgramObjectARB(programId));
3282 checkGLcall("glUseProgramObjectARB(programId)");
3283
3284 /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
3285 * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
3286 * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
3287 * vertex shader with fixed function pixel processing is used we make sure that the card
3288 * supports enough samplers to allow the max number of vertex samplers with all possible
3289 * fixed function fragment processing setups. So once the program is linked these samplers
3290 * won't change.
3291 */
3292 if(vshader_id) {
3293 /* Load vertex shader samplers */
3294 shader_glsl_load_vsamplers(gl_info, (IWineD3DStateBlock*)This->stateBlock, programId);
3295 }
3296 if(pshader_id) {
3297 /* Load pixel shader samplers */
3298 shader_glsl_load_psamplers(gl_info, (IWineD3DStateBlock*)This->stateBlock, programId);
3299 }
3300
3301 /* If the local constants do not have to be loaded with the environment constants,
3302 * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
3303 * later
3304 */
3305 if(pshader && !((IWineD3DPixelShaderImpl*)pshader)->baseShader.load_local_constsF) {
3306 hardcode_local_constants((IWineD3DBaseShaderImpl *) pshader, gl_info, programId, 'P');
3307 }
3308 if(vshader && !((IWineD3DVertexShaderImpl*)vshader)->baseShader.load_local_constsF) {
3309 hardcode_local_constants((IWineD3DBaseShaderImpl *) vshader, gl_info, programId, 'V');
3310 }
3311 }
3312
3313 static GLhandleARB create_glsl_blt_shader(WineD3D_GL_Info *gl_info) {
3314 GLhandleARB program_id;
3315 GLhandleARB vshader_id, pshader_id;
3316 const char *blt_vshader[] = {
3317 "#version 120\n"
3318 "void main(void)\n"
3319 "{\n"
3320 " gl_Position = gl_Vertex;\n"
3321 " gl_FrontColor = vec4(1.0);\n"
3322 " gl_TexCoord[0].x = (gl_Vertex.x * 0.5) + 0.5;\n"
3323 " gl_TexCoord[0].y = (-gl_Vertex.y * 0.5) + 0.5;\n"
3324 "}\n"
3325 };
3326
3327 const char *blt_pshader[] = {
3328 "#version 120\n"
3329 "uniform sampler2D sampler;\n"
3330 "void main(void)\n"
3331 "{\n"
3332 " gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
3333 "}\n"
3334 };
3335
3336 vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3337 GL_EXTCALL(glShaderSourceARB(vshader_id, 1, blt_vshader, NULL));
3338 GL_EXTCALL(glCompileShaderARB(vshader_id));
3339
3340 pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3341 GL_EXTCALL(glShaderSourceARB(pshader_id, 1, blt_pshader, NULL));
3342 GL_EXTCALL(glCompileShaderARB(pshader_id));
3343
3344 program_id = GL_EXTCALL(glCreateProgramObjectARB());
3345 GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
3346 GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
3347 GL_EXTCALL(glLinkProgramARB(program_id));
3348
3349 print_glsl_info_log(&GLINFO_LOCATION, program_id);
3350
3351 /* Once linked we can mark the shaders for deletion. They will be deleted once the program
3352 * is destroyed
3353 */
3354 GL_EXTCALL(glDeleteObjectARB(vshader_id));
3355 GL_EXTCALL(glDeleteObjectARB(pshader_id));
3356 return program_id;
3357 }
3358
3359 static void shader_glsl_select(IWineD3DDevice *iface, BOOL usePS, BOOL useVS) {
3360 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3361 struct shader_glsl_priv *priv = (struct shader_glsl_priv *)This->shader_priv;
3362 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3363 GLhandleARB program_id = 0;
3364 GLenum old_vertex_color_clamp, current_vertex_color_clamp;
3365
3366 old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
3367
3368 if (useVS || usePS) set_glsl_shader_program(iface, usePS, useVS);
3369 else priv->glsl_program = NULL;
3370
3371 current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
3372
3373 if (old_vertex_color_clamp != current_vertex_color_clamp) {
3374 if (GL_SUPPORT(ARB_COLOR_BUFFER_FLOAT)) {
3375 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
3376 checkGLcall("glClampColorARB");
3377 } else {
3378 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
3379 }
3380 }
3381
3382 program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
3383 if (program_id) TRACE("Using GLSL program %u\n", program_id);
3384 GL_EXTCALL(glUseProgramObjectARB(program_id));
3385 checkGLcall("glUseProgramObjectARB");
3386 }
3387
3388 static void shader_glsl_select_depth_blt(IWineD3DDevice *iface) {
3389 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3390 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3391 struct shader_glsl_priv *priv = (struct shader_glsl_priv *) This->shader_priv;
3392 static GLhandleARB loc = -1;
3393
3394 if (!priv->depth_blt_glsl_program_id) {
3395 priv->depth_blt_glsl_program_id = create_glsl_blt_shader(gl_info);
3396 loc = GL_EXTCALL(glGetUniformLocationARB(priv->depth_blt_glsl_program_id, "sampler"));
3397 }
3398
3399 GL_EXTCALL(glUseProgramObjectARB(priv->depth_blt_glsl_program_id));
3400 GL_EXTCALL(glUniform1iARB(loc, 0));
3401 }
3402
3403 static void shader_glsl_destroy_depth_blt(IWineD3DDevice *iface) {
3404 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3405 struct shader_glsl_priv *priv = (struct shader_glsl_priv *) This->shader_priv;
3406 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3407
3408 if(priv->depth_blt_glsl_program_id) {
3409 GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_glsl_program_id));
3410 priv->depth_blt_glsl_program_id = 0;
3411 }
3412 }
3413
3414 static void shader_glsl_cleanup(IWineD3DDevice *iface) {
3415 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3416 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3417 GL_EXTCALL(glUseProgramObjectARB(0));
3418 }
3419
3420 static void shader_glsl_destroy(IWineD3DBaseShader *iface) {
3421 struct list *linked_programs;
3422 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) iface;
3423 WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *) This->baseShader.device)->adapter->gl_info;
3424
3425 /* Note: Do not use QueryInterface here to find out which shader type this is because this code
3426 * can be called from IWineD3DBaseShader::Release
3427 */
3428 char pshader = shader_is_pshader_version(This->baseShader.hex_version);
3429
3430 if(This->baseShader.prgId == 0) return;
3431 linked_programs = &This->baseShader.linked_programs;
3432
3433 TRACE("Deleting linked programs\n");
3434 if (linked_programs->next) {
3435 struct glsl_shader_prog_link *entry, *entry2;
3436
3437 if(pshader) {
3438 LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
3439 delete_glsl_program_entry(This->baseShader.device, entry);
3440 }
3441 } else {
3442 LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
3443 delete_glsl_program_entry(This->baseShader.device, entry);
3444 }
3445 }
3446 }
3447
3448 TRACE("Deleting shader object %u\n", This->baseShader.prgId);
3449 GL_EXTCALL(glDeleteObjectARB(This->baseShader.prgId));
3450 checkGLcall("glDeleteObjectARB");
3451 This->baseShader.prgId = 0;
3452 This->baseShader.is_compiled = FALSE;
3453 }
3454
3455 static unsigned int glsl_program_key_hash(void *key) {
3456 glsl_program_key_t *k = (glsl_program_key_t *)key;
3457
3458 unsigned int hash = k->vshader | k->pshader << 16;
3459 hash += ~(hash << 15);
3460 hash ^= (hash >> 10);
3461 hash += (hash << 3);
3462 hash ^= (hash >> 6);
3463 hash += ~(hash << 11);
3464 hash ^= (hash >> 16);
3465
3466 return hash;
3467 }
3468
3469 static BOOL glsl_program_key_compare(void *keya, void *keyb) {
3470 glsl_program_key_t *ka = (glsl_program_key_t *)keya;
3471 glsl_program_key_t *kb = (glsl_program_key_t *)keyb;
3472
3473 return ka->vshader == kb->vshader && ka->pshader == kb->pshader;
3474 }
3475
3476 static HRESULT shader_glsl_alloc(IWineD3DDevice *iface) {
3477 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3478 This->shader_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
3479 This->glsl_program_lookup = hash_table_create(glsl_program_key_hash, glsl_program_key_compare);
3480 return WINED3D_OK;
3481 }
3482
3483 static void shader_glsl_free(IWineD3DDevice *iface) {
3484 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3485 HeapFree(GetProcessHeap(), 0, This->shader_priv);
3486 This->shader_priv = NULL;
3487 }
3488
3489 static BOOL shader_glsl_dirty_const(IWineD3DDevice *iface) {
3490 /* TODO: GL_EXT_bindable_uniform can be used to share constants across shaders */
3491 return FALSE;
3492 }
3493
3494 static void shader_glsl_generate_pshader(IWineD3DPixelShader *iface, SHADER_BUFFER *buffer) {
3495 IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
3496 shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3497 CONST DWORD *function = This->baseShader.function;
3498 const char *fragcolor;
3499 WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3500
3501 /* Create the hw GLSL shader object and assign it as the baseShader.prgId */
3502 GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3503
3504 shader_addline(buffer, "#version 120\n");
3505
3506 if (GL_SUPPORT(ARB_DRAW_BUFFERS)) {
3507 shader_addline(buffer, "#extension GL_ARB_draw_buffers : enable\n");
3508 }
3509 if (GL_SUPPORT(ARB_TEXTURE_RECTANGLE)) {
3510 /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
3511 * drivers write a warning if we don't do so
3512 */
3513 shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
3514 }
3515
3516 /* Base Declarations */
3517 shader_generate_glsl_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION);
3518
3519 /* Pack 3.0 inputs */
3520 if (This->baseShader.hex_version >= WINED3DPS_VERSION(3,0)) {
3521
3522 if(((IWineD3DDeviceImpl *) This->baseShader.device)->strided_streams.u.s.position_transformed) {
3523 This->vertexprocessing = pretransformed;
3524 pshader_glsl_input_pack(buffer, This->semantics_in, iface);
3525 } else if(!use_vs((IWineD3DDeviceImpl *) This->baseShader.device)) {
3526 This->vertexprocessing = fixedfunction;
3527 pshader_glsl_input_pack(buffer, This->semantics_in, iface);
3528 } else {
3529 This->vertexprocessing = vertexshader;
3530 }
3531 }
3532
3533 /* Base Shader Body */
3534 shader_generate_main( (IWineD3DBaseShader*) This, buffer, reg_maps, function);
3535
3536 /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
3537 if (This->baseShader.hex_version < WINED3DPS_VERSION(2,0)) {
3538 /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
3539 if(GL_SUPPORT(ARB_DRAW_BUFFERS))
3540 shader_addline(buffer, "gl_FragData[0] = R0;\n");
3541 else
3542 shader_addline(buffer, "gl_FragColor = R0;\n");
3543 }
3544
3545 if(GL_SUPPORT(ARB_DRAW_BUFFERS)) {
3546 fragcolor = "gl_FragData[0]";
3547 } else {
3548 fragcolor = "gl_FragColor";
3549 }
3550 if(This->srgb_enabled) {
3551 shader_addline(buffer, "tmp0.xyz = pow(%s.xyz, vec3(%f, %f, %f)) * vec3(%f, %f, %f) - vec3(%f, %f, %f);\n",
3552 fragcolor, srgb_pow, srgb_pow, srgb_pow, srgb_mul_high, srgb_mul_high, srgb_mul_high,
3553 srgb_sub_high, srgb_sub_high, srgb_sub_high);
3554 shader_addline(buffer, "tmp1.xyz = %s.xyz * srgb_mul_low.xyz;\n", fragcolor);
3555 shader_addline(buffer, "%s.x = %s.x < srgb_comparison.x ? tmp1.x : tmp0.x;\n", fragcolor, fragcolor);
3556 shader_addline(buffer, "%s.y = %s.y < srgb_comparison.y ? tmp1.y : tmp0.y;\n", fragcolor, fragcolor);
3557 shader_addline(buffer, "%s.z = %s.z < srgb_comparison.z ? tmp1.z : tmp0.z;\n", fragcolor, fragcolor);
3558 shader_addline(buffer, "%s = clamp(%s, 0.0, 1.0);\n", fragcolor, fragcolor);
3559 }
3560 /* Pixel shader < 3.0 do not replace the fog stage.
3561 * This implements linear fog computation and blending.
3562 * TODO: non linear fog
3563 * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
3564 * -1/(e-s) and e/(e-s) respectively.
3565 */
3566 if(This->baseShader.hex_version < WINED3DPS_VERSION(3,0)) {
3567 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * gl_Fog.start + gl_Fog.end, 0.0, 1.0);\n");
3568 shader_addline(buffer, "%s.xyz = mix(gl_Fog.color.xyz, %s.xyz, Fog);\n", fragcolor, fragcolor);
3569 }
3570
3571 shader_addline(buffer, "}\n");
3572
3573 TRACE("Compiling shader object %u\n", shader_obj);
3574 GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
3575 GL_EXTCALL(glCompileShaderARB(shader_obj));
3576 print_glsl_info_log(&GLINFO_LOCATION, shader_obj);
3577
3578 /* Store the shader object */
3579 This->baseShader.prgId = shader_obj;
3580 }
3581
3582 static void shader_glsl_generate_vshader(IWineD3DVertexShader *iface, SHADER_BUFFER *buffer) {
3583 IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
3584 shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3585 CONST DWORD *function = This->baseShader.function;
3586 WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3587
3588 /* Create the hw GLSL shader program and assign it as the baseShader.prgId */
3589 GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3590
3591 shader_addline(buffer, "#version 120\n");
3592
3593 /* Base Declarations */
3594 shader_generate_glsl_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION);
3595
3596 /* Base Shader Body */
3597 shader_generate_main( (IWineD3DBaseShader*) This, buffer, reg_maps, function);
3598
3599 /* Unpack 3.0 outputs */
3600 if (This->baseShader.hex_version >= WINED3DVS_VERSION(3,0)) {
3601 shader_addline(buffer, "order_ps_input(OUT);\n");
3602 } else {
3603 shader_addline(buffer, "order_ps_input();\n");
3604 }
3605
3606 /* If this shader doesn't use fog copy the z coord to the fog coord so that we can use table fog */
3607 if (!reg_maps->fog)
3608 shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
3609
3610 /* Write the final position.
3611 *
3612 * OpenGL coordinates specify the center of the pixel while d3d coords specify
3613 * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3614 * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3615 * contains 1.0 to allow a mad.
3616 */
3617 shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
3618 shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
3619
3620 /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3621 *
3622 * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
3623 * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
3624 * which is the same as z = z / 2 - w.
3625 */
3626 shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
3627
3628 shader_addline(buffer, "}\n");
3629
3630 TRACE("Compiling shader object %u\n", shader_obj);
3631 GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
3632 GL_EXTCALL(glCompileShaderARB(shader_obj));
3633 print_glsl_info_log(&GLINFO_LOCATION, shader_obj);
3634
3635 /* Store the shader object */
3636 This->baseShader.prgId = shader_obj;
3637 }
3638
3639 static void shader_glsl_get_caps(WINED3DDEVTYPE devtype, WineD3D_GL_Info *gl_info, struct shader_caps *pCaps) {
3640 /* We don't have a GLSL fixed function pipeline yet, so let the none backend set its caps,
3641 * then overwrite the shader specific ones
3642 */
3643 none_shader_backend.shader_get_caps(devtype, gl_info, pCaps);
3644
3645 /* Nvidia Geforce6/7 or Ati R4xx/R5xx cards with GLSL support, support VS 3.0 but older Nvidia/Ati
3646 * models with GLSL support only support 2.0. In case of nvidia we can detect VS 2.0 support using
3647 * vs_nv_version which is based on NV_vertex_program.
3648 * For Ati cards there's no way using glsl (it abstracts the lowlevel info away) and also not
3649 * using ARB_vertex_program. It is safe to assume that when a card supports pixel shader 2.0 it
3650 * supports vertex shader 2.0 too and the way around. We can detect ps2.0 using the maximum number
3651 * of native instructions, so use that here. For more info see the pixel shader versioning code below.
3652 */
3653 if((GLINFO_LOCATION.vs_nv_version == VS_VERSION_20) || (GLINFO_LOCATION.ps_arb_max_instructions <= 512))
3654 pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
3655 else
3656 pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
3657 TRACE_(d3d_caps)("Hardware vertex shader version %d.%d enabled (GLSL)\n", (pCaps->VertexShaderVersion >> 8) & 0xff, pCaps->VertexShaderVersion & 0xff);
3658 pCaps->MaxVertexShaderConst = GL_LIMITS(vshader_constantsF);
3659
3660 /* Older DX9-class videocards (GeforceFX / Radeon >9500/X*00) only support pixel shader 2.0/2.0a/2.0b.
3661 * In OpenGL the extensions related to GLSL abstract lowlevel GL info away which is needed
3662 * to distinguish between 2.0 and 3.0 (and 2.0a/2.0b). In case of Nvidia we use their fragment
3663 * program extensions. On other hardware including ATI GL_ARB_fragment_program offers the info
3664 * in max native instructions. Intel and others also offer the info in this extension but they
3665 * don't support GLSL (at least on Windows).
3666 *
3667 * PS2.0 requires at least 96 instructions, 2.0a/2.0b go up to 512. Assume that if the number
3668 * of instructions is 512 or less we have to do with ps2.0 hardware.
3669 * NOTE: ps3.0 hardware requires 512 or more instructions but ati and nvidia offer 'enough' (1024 vs 4096) on their most basic ps3.0 hardware.
3670 */
3671 if((GLINFO_LOCATION.ps_nv_version == PS_VERSION_20) || (GLINFO_LOCATION.ps_arb_max_instructions <= 512))
3672 pCaps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
3673 else
3674 pCaps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
3675
3676 /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
3677 * Direct3D minimum requirement.
3678 *
3679 * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
3680 * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
3681 *
3682 * The problem is that the refrast clamps temporary results in the shader to
3683 * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
3684 * then applications may miss the clamping behavior. On the other hand, if it is smaller,
3685 * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
3686 * offer a way to query this.
3687 */
3688 pCaps->PixelShader1xMaxValue = 8.0;
3689 TRACE_(d3d_caps)("Hardware pixel shader version %d.%d enabled (GLSL)\n", (pCaps->PixelShaderVersion >> 8) & 0xff, pCaps->PixelShaderVersion & 0xff);
3690 }
3691
3692 static void shader_glsl_load_init(void) {}
3693
3694 static void shader_glsl_fragment_enable(IWineD3DDevice *iface, BOOL enable) {
3695 none_shader_backend.shader_fragment_enable(iface, enable);
3696 }
3697
3698 const shader_backend_t glsl_shader_backend = {
3699 shader_glsl_select,
3700 shader_glsl_select_depth_blt,
3701 shader_glsl_destroy_depth_blt,
3702 shader_glsl_load_constants,
3703 shader_glsl_cleanup,
3704 shader_glsl_color_correction,
3705 shader_glsl_destroy,
3706 shader_glsl_alloc,
3707 shader_glsl_free,
3708 shader_glsl_dirty_const,
3709 shader_glsl_generate_pshader,
3710 shader_glsl_generate_vshader,
3711 shader_glsl_get_caps,
3712 shader_glsl_load_init,
3713 shader_glsl_fragment_enable,
3714 FFPStateTable
3715 };
useful hacks and other patches not (yet) suitable for mainline