?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Release 0.9.61.
[wine.git] / dlls / wined3d / glsl_shader.c
blobebf3c09cb4cf7a6f8865600e88287267bb09e37a
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[50];
43 char mask_str[6];
44 } glsl_dst_param_t;
45
46 typedef struct {
47 char reg_name[50];
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 IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
373 WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
374
375 GLhandleARB *constant_locations;
376 struct list *constant_list;
377 GLhandleARB programId;
378 struct glsl_shader_prog_link *prog = stateBlock->glsl_program;
379 unsigned int i;
380
381 if (!prog) {
382 /* No GLSL program set - nothing to do. */
383 return;
384 }
385 programId = prog->programId;
386
387 if (useVertexShader) {
388 IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
389
390 constant_locations = prog->vuniformF_locations;
391 constant_list = &stateBlock->set_vconstantsF;
392
393 /* Load DirectX 9 float constants/uniforms for vertex shader */
394 shader_glsl_load_constantsF(vshader, gl_info, GL_LIMITS(vshader_constantsF),
395 stateBlock->vertexShaderConstantF, constant_locations, constant_list);
396
397 /* Load DirectX 9 integer constants/uniforms for vertex shader */
398 shader_glsl_load_constantsI(vshader, gl_info, programId,
399 prog->vuniformI_locations, MAX_CONST_I,
400 stateBlock->vertexShaderConstantI,
401 stateBlock->changed.vertexShaderConstantsI);
402
403 /* Load DirectX 9 boolean constants/uniforms for vertex shader */
404 shader_glsl_load_constantsB(vshader, gl_info, programId, MAX_CONST_B,
405 stateBlock->vertexShaderConstantB,
406 stateBlock->changed.vertexShaderConstantsB);
407
408 /* Upload the position fixup params */
409 GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, &deviceImpl->posFixup[0]));
410 checkGLcall("glUniform4fvARB");
411 }
412
413 if (usePixelShader) {
414
415 IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
416
417 constant_locations = prog->puniformF_locations;
418 constant_list = &stateBlock->set_pconstantsF;
419
420 /* Load DirectX 9 float constants/uniforms for pixel shader */
421 shader_glsl_load_constantsF(pshader, gl_info, GL_LIMITS(pshader_constantsF),
422 stateBlock->pixelShaderConstantF, constant_locations, constant_list);
423
424 /* Load DirectX 9 integer constants/uniforms for pixel shader */
425 shader_glsl_load_constantsI(pshader, gl_info, programId,
426 prog->puniformI_locations, MAX_CONST_I,
427 stateBlock->pixelShaderConstantI,
428 stateBlock->changed.pixelShaderConstantsI);
429
430 /* Load DirectX 9 boolean constants/uniforms for pixel shader */
431 shader_glsl_load_constantsB(pshader, gl_info, programId, MAX_CONST_B,
432 stateBlock->pixelShaderConstantB,
433 stateBlock->changed.pixelShaderConstantsB);
434
435 /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
436 * It can't be 0 for a valid texbem instruction.
437 */
438 for(i = 0; i < ((IWineD3DPixelShaderImpl *) pshader)->numbumpenvmatconsts; i++) {
439 IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pshader;
440 int stage = ps->luminanceconst[i].texunit;
441
442 float *data = (float *) &stateBlock->textureState[(int) ps->bumpenvmatconst[i].texunit][WINED3DTSS_BUMPENVMAT00];
443 GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
444 checkGLcall("glUniformMatrix2fvARB");
445
446 /* texbeml needs the luminance scale and offset too. If texbeml is used, needsbumpmat
447 * is set too, so we can check that in the needsbumpmat check
448 */
449 if(ps->baseShader.reg_maps.luminanceparams[stage]) {
450 GLfloat *scale = (GLfloat *) &stateBlock->textureState[stage][WINED3DTSS_BUMPENVLSCALE];
451 GLfloat *offset = (GLfloat *) &stateBlock->textureState[stage][WINED3DTSS_BUMPENVLOFFSET];
452
453 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
454 checkGLcall("glUniform1fvARB");
455 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
456 checkGLcall("glUniform1fvARB");
457 }
458 }
459
460 if(((IWineD3DPixelShaderImpl *) pshader)->srgb_enabled &&
461 !((IWineD3DPixelShaderImpl *) pshader)->srgb_mode_hardcoded) {
462 float comparison[4];
463 float mul_low[4];
464
465 if(stateBlock->renderState[WINED3DRS_SRGBWRITEENABLE]) {
466 comparison[0] = srgb_cmp; comparison[1] = srgb_cmp;
467 comparison[2] = srgb_cmp; comparison[3] = srgb_cmp;
468
469 mul_low[0] = srgb_mul_low; mul_low[1] = srgb_mul_low;
470 mul_low[2] = srgb_mul_low; mul_low[3] = srgb_mul_low;
471 } else {
472 comparison[0] = 1.0 / 0.0; comparison[1] = 1.0 / 0.0;
473 comparison[2] = 1.0 / 0.0; comparison[3] = 1.0 / 0.0;
474
475 mul_low[0] = 1.0; mul_low[1] = 1.0;
476 mul_low[2] = 1.0; mul_low[3] = 1.0;
477 }
478
479 GL_EXTCALL(glUniform4fvARB(prog->srgb_comparison_location, 1, comparison));
480 GL_EXTCALL(glUniform4fvARB(prog->srgb_mul_low_location, 1, mul_low));
481 }
482 if(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
483 float correction_params[4];
484 if(deviceImpl->render_offscreen) {
485 correction_params[0] = 0.0;
486 correction_params[1] = 1.0;
487 } else {
488 /* position is window relative, not viewport relative */
489 correction_params[0] = ((IWineD3DSurfaceImpl *) deviceImpl->render_targets[0])->currentDesc.Height;
490 correction_params[1] = -1.0;
491 }
492 GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
493 }
494 }
495 }
496
497 /** Generate the variable & register declarations for the GLSL output target */
498 void shader_generate_glsl_declarations(
499 IWineD3DBaseShader *iface,
500 shader_reg_maps* reg_maps,
501 SHADER_BUFFER* buffer,
502 WineD3D_GL_Info* gl_info) {
503
504 IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
505 IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
506 int i;
507 unsigned int extra_constants_needed = 0;
508 local_constant* lconst;
509
510 /* There are some minor differences between pixel and vertex shaders */
511 char pshader = shader_is_pshader_version(This->baseShader.hex_version);
512 char prefix = pshader ? 'P' : 'V';
513
514 /* Prototype the subroutines */
515 for (i = 0; i < This->baseShader.limits.label; i++) {
516 if (reg_maps->labels[i])
517 shader_addline(buffer, "void subroutine%u();\n", i);
518 }
519
520 /* Declare the constants (aka uniforms) */
521 if (This->baseShader.limits.constant_float > 0) {
522 unsigned max_constantsF = min(This->baseShader.limits.constant_float,
523 (pshader ? GL_LIMITS(pshader_constantsF) : GL_LIMITS(vshader_constantsF)));
524 shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
525 }
526
527 if (This->baseShader.limits.constant_int > 0)
528 shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, This->baseShader.limits.constant_int);
529
530 if (This->baseShader.limits.constant_bool > 0)
531 shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, This->baseShader.limits.constant_bool);
532
533 if(!pshader) {
534 shader_addline(buffer, "uniform vec4 posFixup;\n");
535 /* Predeclaration; This function is added at link time based on the pixel shader.
536 * VS 3.0 shaders have an array OUT[] the shader writes to, earlier versions don't have
537 * that. We know the input to the reorder function at vertex shader compile time, so
538 * we can deal with that. The reorder function for a 1.x and 2.x vertex shader can just
539 * read gl_FrontColor. The output depends on the pixel shader. The reorder function for a
540 * 1.x and 2.x pshader or for fixed function will write gl_FrontColor, and for a 3.0 shader
541 * it will write to the varying array. Here we depend on the shader optimizer on sorting that
542 * out. The nvidia driver only does that if the parameter is inout instead of out, hence the
543 * inout.
544 */
545 if(This->baseShader.hex_version >= WINED3DVS_VERSION(3, 0)) {
546 shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
547 } else {
548 shader_addline(buffer, "void order_ps_input();\n");
549 }
550 } else {
551 IWineD3DPixelShaderImpl *ps_impl = (IWineD3DPixelShaderImpl *) This;
552
553 ps_impl->numbumpenvmatconsts = 0;
554 for(i = 0; i < (sizeof(reg_maps->bumpmat) / sizeof(reg_maps->bumpmat[0])); i++) {
555 if(!reg_maps->bumpmat[i]) {
556 continue;
557 }
558
559 ps_impl->bumpenvmatconst[(int) ps_impl->numbumpenvmatconsts].texunit = i;
560 shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
561
562 if(reg_maps->luminanceparams) {
563 ps_impl->luminanceconst[(int) ps_impl->numbumpenvmatconsts].texunit = i;
564 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
565 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
566 extra_constants_needed++;
567 } else {
568 ps_impl->luminanceconst[(int) ps_impl->numbumpenvmatconsts].texunit = -1;
569 }
570
571 extra_constants_needed++;
572 ps_impl->numbumpenvmatconsts++;
573 }
574
575 if(device->stateBlock->renderState[WINED3DRS_SRGBWRITEENABLE]) {
576 ps_impl->srgb_enabled = 1;
577 if(This->baseShader.limits.constant_float + extra_constants_needed + 1 < GL_LIMITS(pshader_constantsF)) {
578 shader_addline(buffer, "uniform vec4 srgb_mul_low;\n");
579 shader_addline(buffer, "uniform vec4 srgb_comparison;\n");
580 ps_impl->srgb_mode_hardcoded = 0;
581 extra_constants_needed++;
582 } else {
583 ps_impl->srgb_mode_hardcoded = 1;
584 shader_addline(buffer, "const vec4 srgb_mul_low = vec4(%f, %f, %f, %f);\n",
585 srgb_mul_low, srgb_mul_low, srgb_mul_low, srgb_mul_low);
586 shader_addline(buffer, "const vec4 srgb_comparison = vec4(%f, %f, %f, %f);\n",
587 srgb_cmp, srgb_cmp, srgb_cmp, srgb_cmp);
588 }
589 } else {
590 IWineD3DPixelShaderImpl *ps_impl = (IWineD3DPixelShaderImpl *) This;
591
592 /* Do not write any srgb fixup into the shader to save shader size and processing time.
593 * As a consequence, we can't toggle srgb write on without recompilation
594 */
595 ps_impl->srgb_enabled = 0;
596 ps_impl->srgb_mode_hardcoded = 1;
597 }
598 if(reg_maps->vpos || reg_maps->usesdsy) {
599 if(This->baseShader.limits.constant_float + extra_constants_needed + 1 < GL_LIMITS(pshader_constantsF)) {
600 shader_addline(buffer, "uniform vec4 ycorrection;\n");
601 ((IWineD3DPixelShaderImpl *) This)->vpos_uniform = 1;
602 extra_constants_needed++;
603 } else {
604 /* This happens because we do not have proper tracking of the constant registers that are
605 * actually used, only the max limit of the shader version
606 */
607 FIXME("Cannot find a free uniform for vpos correction params\n");
608 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
609 device->render_offscreen ? 0.0 : ((IWineD3DSurfaceImpl *) device->render_targets[0])->currentDesc.Height,
610 device->render_offscreen ? 1.0 : -1.0);
611 }
612 shader_addline(buffer, "vec4 vpos;\n");
613 }
614 }
615
616 /* Declare texture samplers */
617 for (i = 0; i < This->baseShader.limits.sampler; i++) {
618 if (reg_maps->samplers[i]) {
619
620 DWORD stype = reg_maps->samplers[i] & WINED3DSP_TEXTURETYPE_MASK;
621 switch (stype) {
622
623 case WINED3DSTT_1D:
624 shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
625 break;
626 case WINED3DSTT_2D:
627 if(device->stateBlock->textures[i] &&
628 IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[i]) == GL_TEXTURE_RECTANGLE_ARB) {
629 shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
630 } else {
631 shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
632 }
633 break;
634 case WINED3DSTT_CUBE:
635 shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
636 break;
637 case WINED3DSTT_VOLUME:
638 shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
639 break;
640 default:
641 shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
642 FIXME("Unrecognized sampler type: %#x\n", stype);
643 break;
644 }
645 }
646 }
647
648 /* Declare address variables */
649 for (i = 0; i < This->baseShader.limits.address; i++) {
650 if (reg_maps->address[i])
651 shader_addline(buffer, "ivec4 A%d;\n", i);
652 }
653
654 /* Declare texture coordinate temporaries and initialize them */
655 for (i = 0; i < This->baseShader.limits.texcoord; i++) {
656 if (reg_maps->texcoord[i])
657 shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
658 }
659
660 /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
661 * helper function shader that is linked in at link time
662 */
663 if(pshader && This->baseShader.hex_version >= WINED3DPS_VERSION(3, 0)) {
664 if(use_vs(device)) {
665 shader_addline(buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
666 } else {
667 /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
668 * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
669 * pixel shader that reads the fixed function color into the packed input registers.
670 */
671 shader_addline(buffer, "vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
672 }
673 }
674
675 /* Declare output register temporaries */
676 if(This->baseShader.limits.packed_output) {
677 shader_addline(buffer, "vec4 OUT[%u];\n", This->baseShader.limits.packed_output);
678 }
679
680 /* Declare temporary variables */
681 for(i = 0; i < This->baseShader.limits.temporary; i++) {
682 if (reg_maps->temporary[i])
683 shader_addline(buffer, "vec4 R%u;\n", i);
684 }
685
686 /* Declare attributes */
687 for (i = 0; i < This->baseShader.limits.attributes; i++) {
688 if (reg_maps->attributes[i])
689 shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
690 }
691
692 /* Declare loop registers aLx */
693 for (i = 0; i < reg_maps->loop_depth; i++) {
694 shader_addline(buffer, "int aL%u;\n", i);
695 shader_addline(buffer, "int tmpInt%u;\n", i);
696 }
697
698 /* Temporary variables for matrix operations */
699 shader_addline(buffer, "vec4 tmp0;\n");
700 shader_addline(buffer, "vec4 tmp1;\n");
701
702 /* Hardcodable local constants */
703 if(!This->baseShader.load_local_constsF) {
704 LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
705 float *value = (float *) lconst->value;
706 shader_addline(buffer, "const vec4 LC%u = vec4(%f, %f, %f, %f);\n", lconst->idx,
707 value[0], value[1], value[2], value[3]);
708 }
709 }
710
711 /* Start the main program */
712 shader_addline(buffer, "void main() {\n");
713 if(pshader && reg_maps->vpos) {
714 /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
715 * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
716 * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
717 * precision troubles when we just substract 0.5.
718 *
719 * To deal with that just floor() the position. This will eliminate the fraction on all cards.
720 *
721 * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
722 *
723 * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
724 * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
725 * coordinates specify the pixel centers instead of the pixel corners. This code will behave
726 * correctly on drivers that returns integer values.
727 */
728 shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
729 }
730 }
731
732 /*****************************************************************************
733 * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
734 *
735 * For more information, see http://wiki.winehq.org/DirectX-Shaders
736 ****************************************************************************/
737
738 /* Prototypes */
739 static void shader_glsl_add_src_param(SHADER_OPCODE_ARG* arg, const DWORD param,
740 const DWORD addr_token, DWORD mask, glsl_src_param_t *src_param);
741
742 /** Used for opcode modifiers - They multiply the result by the specified amount */
743 static const char * const shift_glsl_tab[] = {
744 "", /* 0 (none) */
745 "2.0 * ", /* 1 (x2) */
746 "4.0 * ", /* 2 (x4) */
747 "8.0 * ", /* 3 (x8) */
748 "16.0 * ", /* 4 (x16) */
749 "32.0 * ", /* 5 (x32) */
750 "", /* 6 (x64) */
751 "", /* 7 (x128) */
752 "", /* 8 (d256) */
753 "", /* 9 (d128) */
754 "", /* 10 (d64) */
755 "", /* 11 (d32) */
756 "0.0625 * ", /* 12 (d16) */
757 "0.125 * ", /* 13 (d8) */
758 "0.25 * ", /* 14 (d4) */
759 "0.5 * " /* 15 (d2) */
760 };
761
762 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
763 static void shader_glsl_gen_modifier (
764 const DWORD instr,
765 const char *in_reg,
766 const char *in_regswizzle,
767 char *out_str) {
768
769 out_str[0] = 0;
770
771 if (instr == WINED3DSIO_TEXKILL)
772 return;
773
774 switch (instr & WINED3DSP_SRCMOD_MASK) {
775 case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
776 case WINED3DSPSM_DW:
777 case WINED3DSPSM_NONE:
778 sprintf(out_str, "%s%s", in_reg, in_regswizzle);
779 break;
780 case WINED3DSPSM_NEG:
781 sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
782 break;
783 case WINED3DSPSM_NOT:
784 sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
785 break;
786 case WINED3DSPSM_BIAS:
787 sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
788 break;
789 case WINED3DSPSM_BIASNEG:
790 sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
791 break;
792 case WINED3DSPSM_SIGN:
793 sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
794 break;
795 case WINED3DSPSM_SIGNNEG:
796 sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
797 break;
798 case WINED3DSPSM_COMP:
799 sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
800 break;
801 case WINED3DSPSM_X2:
802 sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
803 break;
804 case WINED3DSPSM_X2NEG:
805 sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
806 break;
807 case WINED3DSPSM_ABS:
808 sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
809 break;
810 case WINED3DSPSM_ABSNEG:
811 sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
812 break;
813 default:
814 FIXME("Unhandled modifier %u\n", (instr & WINED3DSP_SRCMOD_MASK));
815 sprintf(out_str, "%s%s", in_reg, in_regswizzle);
816 }
817 }
818
819 /** Writes the GLSL variable name that corresponds to the register that the
820 * DX opcode parameter is trying to access */
821 static void shader_glsl_get_register_name(
822 const DWORD param,
823 const DWORD addr_token,
824 char* regstr,
825 BOOL* is_color,
826 SHADER_OPCODE_ARG* arg) {
827
828 /* oPos, oFog and oPts in D3D */
829 static const char * const hwrastout_reg_names[] = { "gl_Position", "gl_FogFragCoord", "gl_PointSize" };
830
831 DWORD reg = param & WINED3DSP_REGNUM_MASK;
832 DWORD regtype = shader_get_regtype(param);
833 IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) arg->shader;
834 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
835 WineD3D_GL_Info* gl_info = &deviceImpl->adapter->gl_info;
836
837 char pshader = shader_is_pshader_version(This->baseShader.hex_version);
838 char tmpStr[50];
839
840 *is_color = FALSE;
841
842 switch (regtype) {
843 case WINED3DSPR_TEMP:
844 sprintf(tmpStr, "R%u", reg);
845 break;
846 case WINED3DSPR_INPUT:
847 if (pshader) {
848 /* Pixel shaders >= 3.0 */
849 if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) >= 3) {
850 if (param & WINED3DSHADER_ADDRMODE_RELATIVE) {
851 glsl_src_param_t rel_param;
852 shader_glsl_add_src_param(arg, addr_token, 0, WINED3DSP_WRITEMASK_0, &rel_param);
853
854 /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
855 * operation there
856 */
857 if(((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg]) {
858 sprintf(tmpStr, "IN[%s + %u]", rel_param.param_str,
859 ((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg]);
860 } else {
861 sprintf(tmpStr, "IN[%s]", rel_param.param_str);
862 }
863 } else {
864 sprintf(tmpStr, "IN[%u]",
865 ((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg]);
866 }
867 } else {
868 if (reg==0)
869 strcpy(tmpStr, "gl_Color");
870 else
871 strcpy(tmpStr, "gl_SecondaryColor");
872 }
873 } else {
874 if (vshader_input_is_color((IWineD3DVertexShader*) This, reg))
875 *is_color = TRUE;
876 sprintf(tmpStr, "attrib%u", reg);
877 }
878 break;
879 case WINED3DSPR_CONST:
880 {
881 const char* prefix = pshader? "PC":"VC";
882
883 /* Relative addressing */
884 if (param & WINED3DSHADER_ADDRMODE_RELATIVE) {
885
886 /* Relative addressing on shaders 2.0+ have a relative address token,
887 * prior to that, it was hard-coded as "A0.x" because there's only 1 register */
888 if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) >= 2) {
889 glsl_src_param_t rel_param;
890 shader_glsl_add_src_param(arg, addr_token, 0, WINED3DSP_WRITEMASK_0, &rel_param);
891 if(reg) {
892 sprintf(tmpStr, "%s[%s + %u]", prefix, rel_param.param_str, reg);
893 } else {
894 sprintf(tmpStr, "%s[%s]", prefix, rel_param.param_str);
895 }
896 } else {
897 if(reg) {
898 sprintf(tmpStr, "%s[A0.x + %u]", prefix, reg);
899 } else {
900 sprintf(tmpStr, "%s[A0.x]", prefix);
901 }
902 }
903
904 } else {
905 if(shader_constant_is_local(This, reg)) {
906 sprintf(tmpStr, "LC%u", reg);
907 } else {
908 sprintf(tmpStr, "%s[%u]", prefix, reg);
909 }
910 }
911
912 break;
913 }
914 case WINED3DSPR_CONSTINT:
915 if (pshader)
916 sprintf(tmpStr, "PI[%u]", reg);
917 else
918 sprintf(tmpStr, "VI[%u]", reg);
919 break;
920 case WINED3DSPR_CONSTBOOL:
921 if (pshader)
922 sprintf(tmpStr, "PB[%u]", reg);
923 else
924 sprintf(tmpStr, "VB[%u]", reg);
925 break;
926 case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
927 if (pshader) {
928 sprintf(tmpStr, "T%u", reg);
929 } else {
930 sprintf(tmpStr, "A%u", reg);
931 }
932 break;
933 case WINED3DSPR_LOOP:
934 sprintf(tmpStr, "aL%u", This->baseShader.cur_loop_regno - 1);
935 break;
936 case WINED3DSPR_SAMPLER:
937 if (pshader)
938 sprintf(tmpStr, "Psampler%u", reg);
939 else
940 sprintf(tmpStr, "Vsampler%u", reg);
941 break;
942 case WINED3DSPR_COLOROUT:
943 if (reg >= GL_LIMITS(buffers)) {
944 WARN("Write to render target %u, only %d supported\n", reg, 4);
945 }
946 if (GL_SUPPORT(ARB_DRAW_BUFFERS)) {
947 sprintf(tmpStr, "gl_FragData[%u]", reg);
948 } else { /* On older cards with GLSL support like the GeforceFX there's only one buffer. */
949 sprintf(tmpStr, "gl_FragColor");
950 }
951 break;
952 case WINED3DSPR_RASTOUT:
953 sprintf(tmpStr, "%s", hwrastout_reg_names[reg]);
954 break;
955 case WINED3DSPR_DEPTHOUT:
956 sprintf(tmpStr, "gl_FragDepth");
957 break;
958 case WINED3DSPR_ATTROUT:
959 if (reg == 0) {
960 sprintf(tmpStr, "gl_FrontColor");
961 } else {
962 sprintf(tmpStr, "gl_FrontSecondaryColor");
963 }
964 break;
965 case WINED3DSPR_TEXCRDOUT:
966 /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
967 if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) >= 3)
968 sprintf(tmpStr, "OUT[%u]", reg);
969 else
970 sprintf(tmpStr, "gl_TexCoord[%u]", reg);
971 break;
972 case WINED3DSPR_MISCTYPE:
973 if (reg == 0) {
974 /* vPos */
975 sprintf(tmpStr, "vpos");
976 } else if (reg == 1){
977 /* Note that gl_FrontFacing is a bool, while vFace is
978 * a float for which the sign determines front/back
979 */
980 sprintf(tmpStr, "(gl_FrontFacing ? 1.0 : -1.0)");
981 } else {
982 FIXME("Unhandled misctype register %d\n", reg);
983 sprintf(tmpStr, "unrecognized_register");
984 }
985 break;
986 default:
987 FIXME("Unhandled register name Type(%d)\n", regtype);
988 sprintf(tmpStr, "unrecognized_register");
989 break;
990 }
991
992 strcat(regstr, tmpStr);
993 }
994
995 /* Get the GLSL write mask for the destination register */
996 static DWORD shader_glsl_get_write_mask(const DWORD param, char *write_mask) {
997 char *ptr = write_mask;
998 DWORD mask = param & WINED3DSP_WRITEMASK_ALL;
999
1000 if (shader_is_scalar(param)) {
1001 mask = WINED3DSP_WRITEMASK_0;
1002 } else {
1003 *ptr++ = '.';
1004 if (param & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1005 if (param & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1006 if (param & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1007 if (param & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1008 }
1009
1010 *ptr = '\0';
1011
1012 return mask;
1013 }
1014
1015 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1016 unsigned int size = 0;
1017
1018 if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1019 if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1020 if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1021 if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1022
1023 return size;
1024 }
1025
1026 static void shader_glsl_get_swizzle(const DWORD param, BOOL fixup, DWORD mask, char *swizzle_str) {
1027 /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1028 * but addressed as "rgba". To fix this we need to swap the register's x
1029 * and z components. */
1030 DWORD swizzle = (param & WINED3DSP_SWIZZLE_MASK) >> WINED3DSP_SWIZZLE_SHIFT;
1031 const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1032 char *ptr = swizzle_str;
1033
1034 if (!shader_is_scalar(param)) {
1035 *ptr++ = '.';
1036 /* swizzle bits fields: wwzzyyxx */
1037 if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = swizzle_chars[swizzle & 0x03];
1038 if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = swizzle_chars[(swizzle >> 2) & 0x03];
1039 if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = swizzle_chars[(swizzle >> 4) & 0x03];
1040 if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = swizzle_chars[(swizzle >> 6) & 0x03];
1041 }
1042
1043 *ptr = '\0';
1044 }
1045
1046 /* From a given parameter token, generate the corresponding GLSL string.
1047 * Also, return the actual register name and swizzle in case the
1048 * caller needs this information as well. */
1049 static void shader_glsl_add_src_param(SHADER_OPCODE_ARG* arg, const DWORD param,
1050 const DWORD addr_token, DWORD mask, glsl_src_param_t *src_param) {
1051 BOOL is_color = FALSE;
1052 char swizzle_str[6];
1053
1054 src_param->reg_name[0] = '\0';
1055 src_param->param_str[0] = '\0';
1056 swizzle_str[0] = '\0';
1057
1058 shader_glsl_get_register_name(param, addr_token, src_param->reg_name, &is_color, arg);
1059
1060 shader_glsl_get_swizzle(param, is_color, mask, swizzle_str);
1061 shader_glsl_gen_modifier(param, src_param->reg_name, swizzle_str, src_param->param_str);
1062 }
1063
1064 /* From a given parameter token, generate the corresponding GLSL string.
1065 * Also, return the actual register name and swizzle in case the
1066 * caller needs this information as well. */
1067 static DWORD shader_glsl_add_dst_param(SHADER_OPCODE_ARG* arg, const DWORD param,
1068 const DWORD addr_token, glsl_dst_param_t *dst_param) {
1069 BOOL is_color = FALSE;
1070
1071 dst_param->mask_str[0] = '\0';
1072 dst_param->reg_name[0] = '\0';
1073
1074 shader_glsl_get_register_name(param, addr_token, dst_param->reg_name, &is_color, arg);
1075 return shader_glsl_get_write_mask(param, dst_param->mask_str);
1076 }
1077
1078 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1079 static DWORD shader_glsl_append_dst_ext(SHADER_BUFFER *buffer, SHADER_OPCODE_ARG *arg, const DWORD param) {
1080 glsl_dst_param_t dst_param;
1081 DWORD mask;
1082 int shift;
1083
1084 mask = shader_glsl_add_dst_param(arg, param, arg->dst_addr, &dst_param);
1085
1086 if(mask) {
1087 shift = (param & WINED3DSP_DSTSHIFT_MASK) >> WINED3DSP_DSTSHIFT_SHIFT;
1088 shader_addline(buffer, "%s%s = %s(", dst_param.reg_name, dst_param.mask_str, shift_glsl_tab[shift]);
1089 }
1090
1091 return mask;
1092 }
1093
1094 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1095 static DWORD shader_glsl_append_dst(SHADER_BUFFER *buffer, SHADER_OPCODE_ARG *arg) {
1096 return shader_glsl_append_dst_ext(buffer, arg, arg->dst);
1097 }
1098
1099 /** Process GLSL instruction modifiers */
1100 void shader_glsl_add_instruction_modifiers(SHADER_OPCODE_ARG* arg) {
1101
1102 DWORD mask = arg->dst & WINED3DSP_DSTMOD_MASK;
1103
1104 if (arg->opcode->dst_token && mask != 0) {
1105 glsl_dst_param_t dst_param;
1106
1107 shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
1108
1109 if (mask & WINED3DSPDM_SATURATE) {
1110 /* _SAT means to clamp the value of the register to between 0 and 1 */
1111 shader_addline(arg->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1112 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1113 }
1114 if (mask & WINED3DSPDM_MSAMPCENTROID) {
1115 FIXME("_centroid modifier not handled\n");
1116 }
1117 if (mask & WINED3DSPDM_PARTIALPRECISION) {
1118 /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1119 }
1120 }
1121 }
1122
1123 static inline const char* shader_get_comp_op(
1124 const DWORD opcode) {
1125
1126 DWORD op = (opcode & INST_CONTROLS_MASK) >> INST_CONTROLS_SHIFT;
1127 switch (op) {
1128 case COMPARISON_GT: return ">";
1129 case COMPARISON_EQ: return "==";
1130 case COMPARISON_GE: return ">=";
1131 case COMPARISON_LT: return "<";
1132 case COMPARISON_NE: return "!=";
1133 case COMPARISON_LE: return "<=";
1134 default:
1135 FIXME("Unrecognized comparison value: %u\n", op);
1136 return "(\?\?)";
1137 }
1138 }
1139
1140 static void shader_glsl_get_sample_function(DWORD sampler_type, BOOL projected, BOOL texrect, glsl_sample_function_t *sample_function) {
1141 /* Note that there's no such thing as a projected cube texture. */
1142 switch(sampler_type) {
1143 case WINED3DSTT_1D:
1144 sample_function->name = projected ? "texture1DProj" : "texture1D";
1145 sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1146 break;
1147 case WINED3DSTT_2D:
1148 if(texrect) {
1149 sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1150 } else {
1151 sample_function->name = projected ? "texture2DProj" : "texture2D";
1152 }
1153 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1154 break;
1155 case WINED3DSTT_CUBE:
1156 sample_function->name = "textureCube";
1157 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1158 break;
1159 case WINED3DSTT_VOLUME:
1160 sample_function->name = projected ? "texture3DProj" : "texture3D";
1161 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1162 break;
1163 default:
1164 sample_function->name = "";
1165 FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1166 break;
1167 }
1168 }
1169
1170 static void shader_glsl_color_correction(SHADER_OPCODE_ARG* arg) {
1171 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1172 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) shader->baseShader.device;
1173 WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
1174 glsl_dst_param_t dst_param;
1175 glsl_dst_param_t dst_param2;
1176 WINED3DFORMAT fmt;
1177 WINED3DFORMAT conversion_group;
1178 IWineD3DBaseTextureImpl *texture;
1179 DWORD mask, mask_size;
1180 UINT i;
1181 BOOL recorded = FALSE;
1182 DWORD sampler_idx;
1183 DWORD hex_version = shader->baseShader.hex_version;
1184
1185 switch(arg->opcode->opcode) {
1186 case WINED3DSIO_TEX:
1187 if (hex_version < WINED3DPS_VERSION(2,0)) {
1188 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
1189 } else {
1190 sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
1191 }
1192 break;
1193
1194 case WINED3DSIO_TEXLDL:
1195 FIXME("Add color fixup for vertex texture WINED3DSIO_TEXLDL\n");
1196 return;
1197
1198 case WINED3DSIO_TEXDP3TEX:
1199 case WINED3DSIO_TEXM3x3TEX:
1200 case WINED3DSIO_TEXM3x3SPEC:
1201 case WINED3DSIO_TEXM3x3VSPEC:
1202 case WINED3DSIO_TEXBEM:
1203 case WINED3DSIO_TEXREG2AR:
1204 case WINED3DSIO_TEXREG2GB:
1205 case WINED3DSIO_TEXREG2RGB:
1206 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
1207 break;
1208
1209 default:
1210 /* Not a texture sampling instruction, nothing to do */
1211 return;
1212 };
1213
1214 texture = (IWineD3DBaseTextureImpl *) deviceImpl->stateBlock->textures[sampler_idx];
1215 if(texture) {
1216 fmt = texture->resource.format;
1217 conversion_group = texture->baseTexture.shader_conversion_group;
1218 } else {
1219 fmt = WINED3DFMT_UNKNOWN;
1220 conversion_group = WINED3DFMT_UNKNOWN;
1221 }
1222
1223 /* before doing anything, record the sampler with the format in the format conversion list,
1224 * but check if it's not there already
1225 */
1226 for(i = 0; i < shader->baseShader.num_sampled_samplers; i++) {
1227 if(shader->baseShader.sampled_samplers[i] == sampler_idx) {
1228 recorded = TRUE;
1229 break;
1230 }
1231 }
1232 if(!recorded) {
1233 shader->baseShader.sampled_samplers[shader->baseShader.num_sampled_samplers] = sampler_idx;
1234 shader->baseShader.num_sampled_samplers++;
1235 shader->baseShader.sampled_format[sampler_idx] = conversion_group;
1236 }
1237
1238 switch(fmt) {
1239 case WINED3DFMT_V8U8:
1240 case WINED3DFMT_V16U16:
1241 if(GL_SUPPORT(NV_TEXTURE_SHADER) ||
1242 (GL_SUPPORT(ATI_ENVMAP_BUMPMAP) && fmt == WINED3DFMT_V8U8)) {
1243 /* The 3rd channel returns 1.0 in d3d, but 0.0 in gl. Fix this while we're at it :-) */
1244 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_2, &dst_param);
1245 mask_size = shader_glsl_get_write_mask_size(mask);
1246 if(mask_size >= 3) {
1247 shader_addline(arg->buffer, "%s.%c = 1.0;\n", dst_param.reg_name, dst_param.mask_str[3]);
1248 }
1249 } else {
1250 /* Correct the sign, but leave the blue as it is - it was loaded correctly already */
1251 mask = shader_glsl_add_dst_param(arg, arg->dst,
1252 WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1,
1253 &dst_param);
1254 mask_size = shader_glsl_get_write_mask_size(mask);
1255 if(mask_size >= 2) {
1256 shader_addline(arg->buffer, "%s.%c%c = %s.%c%c * 2.0 - 1.0;\n",
1257 dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2],
1258 dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2]);
1259 } else if(mask_size == 1) {
1260 shader_addline(arg->buffer, "%s.%c = %s.%c * 2.0 - 1.0;\n", dst_param.reg_name, dst_param.mask_str[1],
1261 dst_param.reg_name, dst_param.mask_str[1]);
1262 }
1263 }
1264 break;
1265
1266 case WINED3DFMT_X8L8V8U8:
1267 if(!GL_SUPPORT(NV_TEXTURE_SHADER)) {
1268 /* Red and blue are the signed channels, fix them up; Blue(=L) is correct already,
1269 * and a(X) is always 1.0
1270 */
1271 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &dst_param);
1272 mask_size = shader_glsl_get_write_mask_size(mask);
1273 if(mask_size >= 2) {
1274 shader_addline(arg->buffer, "%s.%c%c = %s.%c%c * 2.0 - 1.0;\n",
1275 dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2],
1276 dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2]);
1277 } else if(mask_size == 1) {
1278 shader_addline(arg->buffer, "%s.%c = %s.%c * 2.0 - 1.0;\n",
1279 dst_param.reg_name, dst_param.mask_str[1],
1280 dst_param.reg_name, dst_param.mask_str[1]);
1281 }
1282 }
1283 break;
1284
1285 case WINED3DFMT_L6V5U5:
1286 if(!GL_SUPPORT(NV_TEXTURE_SHADER)) {
1287 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &dst_param);
1288 mask_size = shader_glsl_get_write_mask_size(mask);
1289 shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_2, &dst_param2);
1290 if(mask_size >= 3) {
1291 /* Swap y and z (U and L), and do a sign conversion on x and the new y(V and U) */
1292 shader_addline(arg->buffer, "tmp0.g = %s.%c;\n",
1293 dst_param.reg_name, dst_param.mask_str[2]);
1294 shader_addline(arg->buffer, "%s.%c%c = %s.%c%c * 2.0 - 1.0;\n",
1295 dst_param.reg_name, dst_param.mask_str[2], dst_param.mask_str[1],
1296 dst_param2.reg_name, dst_param.mask_str[1], dst_param.mask_str[3]);
1297 shader_addline(arg->buffer, "%s.%c = tmp0.g;\n", dst_param.reg_name,
1298 dst_param.mask_str[3]);
1299 } else if(mask_size == 2) {
1300 /* This is bad: We have VL, but we need VU */
1301 FIXME("2 components sampled from a converted L6V5U5 texture\n");
1302 } else {
1303 shader_addline(arg->buffer, "%s.%c = %s.%c * 2.0 - 1.0;\n",
1304 dst_param.reg_name, dst_param.mask_str[1],
1305 dst_param2.reg_name, dst_param.mask_str[1]);
1306 }
1307 }
1308 break;
1309
1310 case WINED3DFMT_Q8W8V8U8:
1311 if(!GL_SUPPORT(NV_TEXTURE_SHADER)) {
1312 /* Correct the sign in all channels. The writemask just applies as-is, no
1313 * need for checking the mask size
1314 */
1315 shader_glsl_add_dst_param(arg, arg->dst,
1316 WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 |
1317 WINED3DSP_WRITEMASK_2 | WINED3DSP_WRITEMASK_3,
1318 &dst_param);
1319 shader_addline(arg->buffer, "%s%s = %s%s * 2.0 - 1.0;\n", dst_param.reg_name, dst_param.mask_str,
1320 dst_param.reg_name, dst_param.mask_str);
1321 }
1322 break;
1323
1324 /* stupid compiler */
1325 default:
1326 break;
1327 }
1328 }
1329
1330 /*****************************************************************************
1331 *
1332 * Begin processing individual instruction opcodes
1333 *
1334 ****************************************************************************/
1335
1336 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
1337 void shader_glsl_arith(SHADER_OPCODE_ARG* arg) {
1338 CONST SHADER_OPCODE* curOpcode = arg->opcode;
1339 SHADER_BUFFER* buffer = arg->buffer;
1340 glsl_src_param_t src0_param;
1341 glsl_src_param_t src1_param;
1342 DWORD write_mask;
1343 char op;
1344
1345 /* Determine the GLSL operator to use based on the opcode */
1346 switch (curOpcode->opcode) {
1347 case WINED3DSIO_MUL: op = '*'; break;
1348 case WINED3DSIO_ADD: op = '+'; break;
1349 case WINED3DSIO_SUB: op = '-'; break;
1350 default:
1351 op = ' ';
1352 FIXME("Opcode %s not yet handled in GLSL\n", curOpcode->name);
1353 break;
1354 }
1355
1356 write_mask = shader_glsl_append_dst(buffer, arg);
1357 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1358 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1359 shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
1360 }
1361
1362 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
1363 void shader_glsl_mov(SHADER_OPCODE_ARG* arg) {
1364 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1365 SHADER_BUFFER* buffer = arg->buffer;
1366 glsl_src_param_t src0_param;
1367 DWORD write_mask;
1368
1369 write_mask = shader_glsl_append_dst(buffer, arg);
1370 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1371
1372 /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
1373 * shader versions WINED3DSIO_MOVA is used for this. */
1374 if ((WINED3DSHADER_VERSION_MAJOR(shader->baseShader.hex_version) == 1 &&
1375 !shader_is_pshader_version(shader->baseShader.hex_version) &&
1376 shader_get_regtype(arg->dst) == WINED3DSPR_ADDR)) {
1377 /* This is a simple floor() */
1378 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1379 if (mask_size > 1) {
1380 shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
1381 } else {
1382 shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
1383 }
1384 } else if(arg->opcode->opcode == WINED3DSIO_MOVA) {
1385 /* We need to *round* to the nearest int here. */
1386 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1387 if (mask_size > 1) {
1388 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);
1389 } else {
1390 shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n", src0_param.param_str, src0_param.param_str);
1391 }
1392 } else {
1393 shader_addline(buffer, "%s);\n", src0_param.param_str);
1394 }
1395 }
1396
1397 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
1398 void shader_glsl_dot(SHADER_OPCODE_ARG* arg) {
1399 CONST SHADER_OPCODE* curOpcode = arg->opcode;
1400 SHADER_BUFFER* buffer = arg->buffer;
1401 glsl_src_param_t src0_param;
1402 glsl_src_param_t src1_param;
1403 DWORD dst_write_mask, src_write_mask;
1404 unsigned int dst_size = 0;
1405
1406 dst_write_mask = shader_glsl_append_dst(buffer, arg);
1407 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1408
1409 /* dp3 works on vec3, dp4 on vec4 */
1410 if (curOpcode->opcode == WINED3DSIO_DP4) {
1411 src_write_mask = WINED3DSP_WRITEMASK_ALL;
1412 } else {
1413 src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1414 }
1415
1416 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_write_mask, &src0_param);
1417 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_write_mask, &src1_param);
1418
1419 if (dst_size > 1) {
1420 shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1421 } else {
1422 shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
1423 }
1424 }
1425
1426 /* Note that this instruction has some restrictions. The destination write mask
1427 * can't contain the w component, and the source swizzles have to be .xyzw */
1428 void shader_glsl_cross(SHADER_OPCODE_ARG *arg) {
1429 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1430 glsl_src_param_t src0_param;
1431 glsl_src_param_t src1_param;
1432 char dst_mask[6];
1433
1434 shader_glsl_get_write_mask(arg->dst, dst_mask);
1435 shader_glsl_append_dst(arg->buffer, arg);
1436 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
1437 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_mask, &src1_param);
1438 shader_addline(arg->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
1439 }
1440
1441 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
1442 * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
1443 * GLSL uses the value as-is. */
1444 void shader_glsl_pow(SHADER_OPCODE_ARG *arg) {
1445 SHADER_BUFFER *buffer = arg->buffer;
1446 glsl_src_param_t src0_param;
1447 glsl_src_param_t src1_param;
1448 DWORD dst_write_mask;
1449 unsigned int dst_size;
1450
1451 dst_write_mask = shader_glsl_append_dst(buffer, arg);
1452 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1453
1454 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1455 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
1456
1457 if (dst_size > 1) {
1458 shader_addline(buffer, "vec%d(pow(abs(%s), %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1459 } else {
1460 shader_addline(buffer, "pow(abs(%s), %s));\n", src0_param.param_str, src1_param.param_str);
1461 }
1462 }
1463
1464 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
1465 * Src0 is a scalar. Note that D3D uses the absolute of src0, while
1466 * GLSL uses the value as-is. */
1467 void shader_glsl_log(SHADER_OPCODE_ARG *arg) {
1468 SHADER_BUFFER *buffer = arg->buffer;
1469 glsl_src_param_t src0_param;
1470 DWORD dst_write_mask;
1471 unsigned int dst_size;
1472
1473 dst_write_mask = shader_glsl_append_dst(buffer, arg);
1474 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1475
1476 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1477
1478 if (dst_size > 1) {
1479 shader_addline(buffer, "vec%d(log2(abs(%s))));\n", dst_size, src0_param.param_str);
1480 } else {
1481 shader_addline(buffer, "log2(abs(%s)));\n", src0_param.param_str);
1482 }
1483 }
1484
1485 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
1486 void shader_glsl_map2gl(SHADER_OPCODE_ARG* arg) {
1487 CONST SHADER_OPCODE* curOpcode = arg->opcode;
1488 SHADER_BUFFER* buffer = arg->buffer;
1489 glsl_src_param_t src_param;
1490 const char *instruction;
1491 char arguments[256];
1492 DWORD write_mask;
1493 unsigned i;
1494
1495 /* Determine the GLSL function to use based on the opcode */
1496 /* TODO: Possibly make this a table for faster lookups */
1497 switch (curOpcode->opcode) {
1498 case WINED3DSIO_MIN: instruction = "min"; break;
1499 case WINED3DSIO_MAX: instruction = "max"; break;
1500 case WINED3DSIO_ABS: instruction = "abs"; break;
1501 case WINED3DSIO_FRC: instruction = "fract"; break;
1502 case WINED3DSIO_NRM: instruction = "normalize"; break;
1503 case WINED3DSIO_LOGP:
1504 case WINED3DSIO_LOG: instruction = "log2"; break;
1505 case WINED3DSIO_EXP: instruction = "exp2"; break;
1506 case WINED3DSIO_SGN: instruction = "sign"; break;
1507 case WINED3DSIO_DSX: instruction = "dFdx"; break;
1508 case WINED3DSIO_DSY: instruction = "ycorrection.y * dFdy"; break;
1509 default: instruction = "";
1510 FIXME("Opcode %s not yet handled in GLSL\n", curOpcode->name);
1511 break;
1512 }
1513
1514 write_mask = shader_glsl_append_dst(buffer, arg);
1515
1516 arguments[0] = '\0';
1517 if (curOpcode->num_params > 0) {
1518 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src_param);
1519 strcat(arguments, src_param.param_str);
1520 for (i = 2; i < curOpcode->num_params; ++i) {
1521 strcat(arguments, ", ");
1522 shader_glsl_add_src_param(arg, arg->src[i-1], arg->src_addr[i-1], write_mask, &src_param);
1523 strcat(arguments, src_param.param_str);
1524 }
1525 }
1526
1527 shader_addline(buffer, "%s(%s));\n", instruction, arguments);
1528 }
1529
1530 /** Process the WINED3DSIO_EXPP instruction in GLSL:
1531 * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
1532 * dst.x = 2^(floor(src))
1533 * dst.y = src - floor(src)
1534 * dst.z = 2^src (partial precision is allowed, but optional)
1535 * dst.w = 1.0;
1536 * For 2.0 shaders, just do this (honoring writemask and swizzle):
1537 * dst = 2^src; (partial precision is allowed, but optional)
1538 */
1539 void shader_glsl_expp(SHADER_OPCODE_ARG* arg) {
1540 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)arg->shader;
1541 glsl_src_param_t src_param;
1542
1543 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src_param);
1544
1545 if (shader->baseShader.hex_version < WINED3DPS_VERSION(2,0)) {
1546 char dst_mask[6];
1547
1548 shader_addline(arg->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
1549 shader_addline(arg->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
1550 shader_addline(arg->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
1551 shader_addline(arg->buffer, "tmp0.w = 1.0;\n");
1552
1553 shader_glsl_append_dst(arg->buffer, arg);
1554 shader_glsl_get_write_mask(arg->dst, dst_mask);
1555 shader_addline(arg->buffer, "tmp0%s);\n", dst_mask);
1556 } else {
1557 DWORD write_mask;
1558 unsigned int mask_size;
1559
1560 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1561 mask_size = shader_glsl_get_write_mask_size(write_mask);
1562
1563 if (mask_size > 1) {
1564 shader_addline(arg->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
1565 } else {
1566 shader_addline(arg->buffer, "exp2(%s));\n", src_param.param_str);
1567 }
1568 }
1569 }
1570
1571 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
1572 void shader_glsl_rcp(SHADER_OPCODE_ARG* arg) {
1573 glsl_src_param_t src_param;
1574 DWORD write_mask;
1575 unsigned int mask_size;
1576
1577 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1578 mask_size = shader_glsl_get_write_mask_size(write_mask);
1579 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src_param);
1580
1581 if (mask_size > 1) {
1582 shader_addline(arg->buffer, "vec%d(1.0 / %s));\n", mask_size, src_param.param_str);
1583 } else {
1584 shader_addline(arg->buffer, "1.0 / %s);\n", src_param.param_str);
1585 }
1586 }
1587
1588 void shader_glsl_rsq(SHADER_OPCODE_ARG* arg) {
1589 SHADER_BUFFER* buffer = arg->buffer;
1590 glsl_src_param_t src_param;
1591 DWORD write_mask;
1592 unsigned int mask_size;
1593
1594 write_mask = shader_glsl_append_dst(buffer, arg);
1595 mask_size = shader_glsl_get_write_mask_size(write_mask);
1596
1597 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src_param);
1598
1599 if (mask_size > 1) {
1600 shader_addline(buffer, "vec%d(inversesqrt(%s)));\n", mask_size, src_param.param_str);
1601 } else {
1602 shader_addline(buffer, "inversesqrt(%s));\n", src_param.param_str);
1603 }
1604 }
1605
1606 /** Process signed comparison opcodes in GLSL. */
1607 void shader_glsl_compare(SHADER_OPCODE_ARG* arg) {
1608 glsl_src_param_t src0_param;
1609 glsl_src_param_t src1_param;
1610 DWORD write_mask;
1611 unsigned int mask_size;
1612
1613 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1614 mask_size = shader_glsl_get_write_mask_size(write_mask);
1615 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1616 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1617
1618 if (mask_size > 1) {
1619 const char *compare;
1620
1621 switch(arg->opcode->opcode) {
1622 case WINED3DSIO_SLT: compare = "lessThan"; break;
1623 case WINED3DSIO_SGE: compare = "greaterThanEqual"; break;
1624 default: compare = "";
1625 FIXME("Can't handle opcode %s\n", arg->opcode->name);
1626 }
1627
1628 shader_addline(arg->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
1629 src0_param.param_str, src1_param.param_str);
1630 } else {
1631 switch(arg->opcode->opcode) {
1632 case WINED3DSIO_SLT:
1633 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
1634 * to return 0.0 but step returns 1.0 because step is not < x
1635 * An alternative is a bvec compare padded with an unused second component.
1636 * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
1637 * issue. Playing with not() is not possible either because not() does not accept
1638 * a scalar.
1639 */
1640 shader_addline(arg->buffer, "(%s < %s) ? 1.0 : 0.0);\n", src0_param.param_str, src1_param.param_str);
1641 break;
1642 case WINED3DSIO_SGE:
1643 /* Here we can use the step() function and safe a conditional */
1644 shader_addline(arg->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
1645 break;
1646 default:
1647 FIXME("Can't handle opcode %s\n", arg->opcode->name);
1648 }
1649
1650 }
1651 }
1652
1653 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
1654 void shader_glsl_cmp(SHADER_OPCODE_ARG* arg) {
1655 glsl_src_param_t src0_param;
1656 glsl_src_param_t src1_param;
1657 glsl_src_param_t src2_param;
1658 DWORD write_mask, cmp_channel = 0;
1659 unsigned int i, j;
1660 char mask_char[6];
1661 BOOL temp_destination = FALSE;
1662
1663 if(shader_is_scalar(arg->src[0])) {
1664 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1665
1666 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
1667 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1668 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1669
1670 shader_addline(arg->buffer, "%s >= 0.0 ? %s : %s);\n",
1671 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1672 } else {
1673 DWORD src0reg = arg->src[0] & WINED3DSP_REGNUM_MASK;
1674 DWORD src1reg = arg->src[1] & WINED3DSP_REGNUM_MASK;
1675 DWORD src2reg = arg->src[2] & WINED3DSP_REGNUM_MASK;
1676 DWORD src0regtype = shader_get_regtype(arg->src[0]);
1677 DWORD src1regtype = shader_get_regtype(arg->src[1]);
1678 DWORD src2regtype = shader_get_regtype(arg->src[2]);
1679 DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
1680 DWORD dstregtype = shader_get_regtype(arg->dst);
1681
1682 /* Cycle through all source0 channels */
1683 for (i=0; i<4; i++) {
1684 write_mask = 0;
1685 /* Find the destination channels which use the current source0 channel */
1686 for (j=0; j<4; j++) {
1687 if ( ((arg->src[0] >> (WINED3DSP_SWIZZLE_SHIFT + 2*j)) & 0x3) == i ) {
1688 write_mask |= WINED3DSP_WRITEMASK_0 << j;
1689 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
1690 }
1691 }
1692
1693 /* Splitting the cmp instruction up in multiple lines imposes a problem:
1694 * The first lines may overwrite source parameters of the following lines.
1695 * Deal with that by using a temporary destination register if needed
1696 */
1697 if((src0reg == dstreg && src0regtype == dstregtype) ||
1698 (src1reg == dstreg && src1regtype == dstregtype) ||
1699 (src2reg == dstreg && src2regtype == dstregtype)) {
1700
1701 write_mask = shader_glsl_get_write_mask(arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask), mask_char);
1702 if (!write_mask) continue;
1703 shader_addline(arg->buffer, "tmp0%s = (", mask_char);
1704 temp_destination = TRUE;
1705 } else {
1706 write_mask = shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask));
1707 if (!write_mask) continue;
1708 }
1709
1710 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], cmp_channel, &src0_param);
1711 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1712 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1713
1714 shader_addline(arg->buffer, "%s >= 0.0 ? %s : %s);\n",
1715 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1716 }
1717
1718 if(temp_destination) {
1719 shader_glsl_get_write_mask(arg->dst, mask_char);
1720 shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst);
1721 shader_addline(arg->buffer, "tmp0%s);\n", mask_char);
1722 }
1723 }
1724
1725 }
1726
1727 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
1728 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
1729 * the compare is done per component of src0. */
1730 void shader_glsl_cnd(SHADER_OPCODE_ARG* arg) {
1731 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1732 glsl_src_param_t src0_param;
1733 glsl_src_param_t src1_param;
1734 glsl_src_param_t src2_param;
1735 DWORD write_mask, cmp_channel = 0;
1736 unsigned int i, j;
1737
1738 if (shader->baseShader.hex_version < WINED3DPS_VERSION(1, 4)) {
1739 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1740 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1741 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1742 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1743
1744 /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
1745 if(arg->opcode_token & WINED3DSI_COISSUE) {
1746 shader_addline(arg->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
1747 } else {
1748 shader_addline(arg->buffer, "%s > 0.5 ? %s : %s);\n",
1749 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1750 }
1751 return;
1752 }
1753 /* Cycle through all source0 channels */
1754 for (i=0; i<4; i++) {
1755 write_mask = 0;
1756 /* Find the destination channels which use the current source0 channel */
1757 for (j=0; j<4; j++) {
1758 if ( ((arg->src[0] >> (WINED3DSP_SWIZZLE_SHIFT + 2*j)) & 0x3) == i ) {
1759 write_mask |= WINED3DSP_WRITEMASK_0 << j;
1760 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
1761 }
1762 }
1763 write_mask = shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask));
1764 if (!write_mask) continue;
1765
1766 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], cmp_channel, &src0_param);
1767 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1768 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1769
1770 shader_addline(arg->buffer, "%s > 0.5 ? %s : %s);\n",
1771 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1772 }
1773 }
1774
1775 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
1776 void shader_glsl_mad(SHADER_OPCODE_ARG* arg) {
1777 glsl_src_param_t src0_param;
1778 glsl_src_param_t src1_param;
1779 glsl_src_param_t src2_param;
1780 DWORD write_mask;
1781
1782 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1783 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1784 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1785 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1786 shader_addline(arg->buffer, "(%s * %s) + %s);\n",
1787 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1788 }
1789
1790 /** Handles transforming all WINED3DSIO_M?x? opcodes for
1791 Vertex shaders to GLSL codes */
1792 void shader_glsl_mnxn(SHADER_OPCODE_ARG* arg) {
1793 int i;
1794 int nComponents = 0;
1795 SHADER_OPCODE_ARG tmpArg;
1796
1797 memset(&tmpArg, 0, sizeof(SHADER_OPCODE_ARG));
1798
1799 /* Set constants for the temporary argument */
1800 tmpArg.shader = arg->shader;
1801 tmpArg.buffer = arg->buffer;
1802 tmpArg.src[0] = arg->src[0];
1803 tmpArg.src_addr[0] = arg->src_addr[0];
1804 tmpArg.src_addr[1] = arg->src_addr[1];
1805 tmpArg.reg_maps = arg->reg_maps;
1806
1807 switch(arg->opcode->opcode) {
1808 case WINED3DSIO_M4x4:
1809 nComponents = 4;
1810 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP4);
1811 break;
1812 case WINED3DSIO_M4x3:
1813 nComponents = 3;
1814 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP4);
1815 break;
1816 case WINED3DSIO_M3x4:
1817 nComponents = 4;
1818 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
1819 break;
1820 case WINED3DSIO_M3x3:
1821 nComponents = 3;
1822 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
1823 break;
1824 case WINED3DSIO_M3x2:
1825 nComponents = 2;
1826 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
1827 break;
1828 default:
1829 break;
1830 }
1831
1832 for (i = 0; i < nComponents; i++) {
1833 tmpArg.dst = ((arg->dst) & ~WINED3DSP_WRITEMASK_ALL)|(WINED3DSP_WRITEMASK_0<<i);
1834 tmpArg.src[1] = arg->src[1]+i;
1835 shader_glsl_dot(&tmpArg);
1836 }
1837 }
1838
1839 /**
1840 The LRP instruction performs a component-wise linear interpolation
1841 between the second and third operands using the first operand as the
1842 blend factor. Equation: (dst = src2 + src0 * (src1 - src2))
1843 This is equivalent to mix(src2, src1, src0);
1844 */
1845 void shader_glsl_lrp(SHADER_OPCODE_ARG* arg) {
1846 glsl_src_param_t src0_param;
1847 glsl_src_param_t src1_param;
1848 glsl_src_param_t src2_param;
1849 DWORD write_mask;
1850
1851 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1852
1853 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1854 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1855 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1856
1857 shader_addline(arg->buffer, "mix(%s, %s, %s));\n",
1858 src2_param.param_str, src1_param.param_str, src0_param.param_str);
1859 }
1860
1861 /** Process the WINED3DSIO_LIT instruction in GLSL:
1862 * dst.x = dst.w = 1.0
1863 * dst.y = (src0.x > 0) ? src0.x
1864 * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
1865 * where src.w is clamped at +- 128
1866 */
1867 void shader_glsl_lit(SHADER_OPCODE_ARG* arg) {
1868 glsl_src_param_t src0_param;
1869 glsl_src_param_t src1_param;
1870 glsl_src_param_t src3_param;
1871 char dst_mask[6];
1872
1873 shader_glsl_append_dst(arg->buffer, arg);
1874 shader_glsl_get_write_mask(arg->dst, dst_mask);
1875
1876 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1877 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_1, &src1_param);
1878 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src3_param);
1879
1880 /* The sdk specifies the instruction like this
1881 * dst.x = 1.0;
1882 * if(src.x > 0.0) dst.y = src.x
1883 * else dst.y = 0.0.
1884 * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
1885 * else dst.z = 0.0;
1886 * dst.w = 1.0;
1887 *
1888 * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
1889 * dst.x = 1.0 ... No further explanation needed
1890 * dst.y = max(src.y, 0.0); ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
1891 * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0; ... 0 ^ power is 0, and otherwise we use y anyway
1892 * dst.w = 1.0. ... Nothing fancy.
1893 *
1894 * So we still have one conditional in there. So do this:
1895 * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
1896 *
1897 * 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),
1898 * 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.
1899 * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to
1900 */
1901 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",
1902 src0_param.param_str, src1_param.param_str, src0_param.param_str, src3_param.param_str, dst_mask);
1903 }
1904
1905 /** Process the WINED3DSIO_DST instruction in GLSL:
1906 * dst.x = 1.0
1907 * dst.y = src0.x * src0.y
1908 * dst.z = src0.z
1909 * dst.w = src1.w
1910 */
1911 void shader_glsl_dst(SHADER_OPCODE_ARG* arg) {
1912 glsl_src_param_t src0y_param;
1913 glsl_src_param_t src0z_param;
1914 glsl_src_param_t src1y_param;
1915 glsl_src_param_t src1w_param;
1916 char dst_mask[6];
1917
1918 shader_glsl_append_dst(arg->buffer, arg);
1919 shader_glsl_get_write_mask(arg->dst, dst_mask);
1920
1921 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_1, &src0y_param);
1922 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &src0z_param);
1923 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_1, &src1y_param);
1924 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_3, &src1w_param);
1925
1926 shader_addline(arg->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
1927 src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
1928 }
1929
1930 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
1931 * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
1932 * can handle it. But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
1933 *
1934 * dst.x = cos(src0.?)
1935 * dst.y = sin(src0.?)
1936 * dst.z = dst.z
1937 * dst.w = dst.w
1938 */
1939 void shader_glsl_sincos(SHADER_OPCODE_ARG* arg) {
1940 glsl_src_param_t src0_param;
1941 DWORD write_mask;
1942
1943 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1944 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1945
1946 switch (write_mask) {
1947 case WINED3DSP_WRITEMASK_0:
1948 shader_addline(arg->buffer, "cos(%s));\n", src0_param.param_str);
1949 break;
1950
1951 case WINED3DSP_WRITEMASK_1:
1952 shader_addline(arg->buffer, "sin(%s));\n", src0_param.param_str);
1953 break;
1954
1955 case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
1956 shader_addline(arg->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
1957 break;
1958
1959 default:
1960 ERR("Write mask should be .x, .y or .xy\n");
1961 break;
1962 }
1963 }
1964
1965 /** Process the WINED3DSIO_LOOP instruction in GLSL:
1966 * Start a for() loop where src1.y is the initial value of aL,
1967 * increment aL by src1.z for a total of src1.x iterations.
1968 * Need to use a temporary variable for this operation.
1969 */
1970 /* FIXME: I don't think nested loops will work correctly this way. */
1971 void shader_glsl_loop(SHADER_OPCODE_ARG* arg) {
1972 glsl_src_param_t src1_param;
1973 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1974 DWORD regtype = shader_get_regtype(arg->src[1]);
1975 DWORD reg = arg->src[1] & WINED3DSP_REGNUM_MASK;
1976 const DWORD *control_values = NULL;
1977 local_constant *constant;
1978
1979 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
1980
1981 /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
1982 * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
1983 * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
1984 * addressing.
1985 */
1986 if(regtype == WINED3DSPR_CONSTINT) {
1987 LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
1988 if(constant->idx == reg) {
1989 control_values = constant->value;
1990 break;
1991 }
1992 }
1993 }
1994
1995 if(control_values) {
1996 if(control_values[2] > 0) {
1997 shader_addline(arg->buffer, "for (aL%u = %d; aL%u < (%d * %d + %d); aL%u += %d) {\n",
1998 shader->baseShader.cur_loop_depth, control_values[1],
1999 shader->baseShader.cur_loop_depth, control_values[0], control_values[2], control_values[1],
2000 shader->baseShader.cur_loop_depth, control_values[2]);
2001 } else if(control_values[2] == 0) {
2002 shader_addline(arg->buffer, "for (aL%u = %d, tmpInt%u = 0; tmpInt%u < %d; tmpInt%u++) {\n",
2003 shader->baseShader.cur_loop_depth, control_values[1], shader->baseShader.cur_loop_depth,
2004 shader->baseShader.cur_loop_depth, control_values[0],
2005 shader->baseShader.cur_loop_depth);
2006 } else {
2007 shader_addline(arg->buffer, "for (aL%u = %d; aL%u > (%d * %d + %d); aL%u += %d) {\n",
2008 shader->baseShader.cur_loop_depth, control_values[1],
2009 shader->baseShader.cur_loop_depth, control_values[0], control_values[2], control_values[1],
2010 shader->baseShader.cur_loop_depth, control_values[2]);
2011 }
2012 } else {
2013 shader_addline(arg->buffer, "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2014 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno,
2015 src1_param.reg_name, shader->baseShader.cur_loop_depth, src1_param.reg_name,
2016 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno, src1_param.reg_name);
2017 }
2018
2019 shader->baseShader.cur_loop_depth++;
2020 shader->baseShader.cur_loop_regno++;
2021 }
2022
2023 void shader_glsl_end(SHADER_OPCODE_ARG* arg) {
2024 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
2025
2026 shader_addline(arg->buffer, "}\n");
2027
2028 if(arg->opcode->opcode == WINED3DSIO_ENDLOOP) {
2029 shader->baseShader.cur_loop_depth--;
2030 shader->baseShader.cur_loop_regno--;
2031 }
2032 if(arg->opcode->opcode == WINED3DSIO_ENDREP) {
2033 shader->baseShader.cur_loop_depth--;
2034 }
2035 }
2036
2037 void shader_glsl_rep(SHADER_OPCODE_ARG* arg) {
2038 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
2039 glsl_src_param_t src0_param;
2040
2041 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2042 shader_addline(arg->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2043 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2044 src0_param.param_str, shader->baseShader.cur_loop_depth);
2045 shader->baseShader.cur_loop_depth++;
2046 }
2047
2048 void shader_glsl_if(SHADER_OPCODE_ARG* arg) {
2049 glsl_src_param_t src0_param;
2050
2051 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2052 shader_addline(arg->buffer, "if (%s) {\n", src0_param.param_str);
2053 }
2054
2055 void shader_glsl_ifc(SHADER_OPCODE_ARG* arg) {
2056 glsl_src_param_t src0_param;
2057 glsl_src_param_t src1_param;
2058
2059 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2060 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2061
2062 shader_addline(arg->buffer, "if (%s %s %s) {\n",
2063 src0_param.param_str, shader_get_comp_op(arg->opcode_token), src1_param.param_str);
2064 }
2065
2066 void shader_glsl_else(SHADER_OPCODE_ARG* arg) {
2067 shader_addline(arg->buffer, "} else {\n");
2068 }
2069
2070 void shader_glsl_break(SHADER_OPCODE_ARG* arg) {
2071 shader_addline(arg->buffer, "break;\n");
2072 }
2073
2074 /* FIXME: According to MSDN the compare is done per component. */
2075 void shader_glsl_breakc(SHADER_OPCODE_ARG* arg) {
2076 glsl_src_param_t src0_param;
2077 glsl_src_param_t src1_param;
2078
2079 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2080 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2081
2082 shader_addline(arg->buffer, "if (%s %s %s) break;\n",
2083 src0_param.param_str, shader_get_comp_op(arg->opcode_token), src1_param.param_str);
2084 }
2085
2086 void shader_glsl_label(SHADER_OPCODE_ARG* arg) {
2087
2088 DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2089 shader_addline(arg->buffer, "}\n");
2090 shader_addline(arg->buffer, "void subroutine%u () {\n", snum);
2091 }
2092
2093 void shader_glsl_call(SHADER_OPCODE_ARG* arg) {
2094 DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2095 shader_addline(arg->buffer, "subroutine%u();\n", snum);
2096 }
2097
2098 void shader_glsl_callnz(SHADER_OPCODE_ARG* arg) {
2099 glsl_src_param_t src1_param;
2100
2101 DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2102 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2103 shader_addline(arg->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, snum);
2104 }
2105
2106 /*********************************************
2107 * Pixel Shader Specific Code begins here
2108 ********************************************/
2109 void pshader_glsl_tex(SHADER_OPCODE_ARG* arg) {
2110 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2111 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2112 DWORD hex_version = This->baseShader.hex_version;
2113 char dst_swizzle[6];
2114 glsl_sample_function_t sample_function;
2115 DWORD sampler_type;
2116 DWORD sampler_idx;
2117 BOOL projected, texrect = FALSE;
2118 DWORD mask = 0;
2119
2120 /* All versions have a destination register */
2121 shader_glsl_append_dst(arg->buffer, arg);
2122
2123 /* 1.0-1.4: Use destination register as sampler source.
2124 * 2.0+: Use provided sampler source. */
2125 if (hex_version < WINED3DPS_VERSION(1,4)) {
2126 DWORD flags;
2127
2128 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2129 flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2130
2131 if (flags & WINED3DTTFF_PROJECTED) {
2132 projected = TRUE;
2133 switch (flags & ~WINED3DTTFF_PROJECTED) {
2134 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2135 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
2136 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
2137 case WINED3DTTFF_COUNT4:
2138 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
2139 }
2140 } else {
2141 projected = FALSE;
2142 }
2143 } else if (hex_version < WINED3DPS_VERSION(2,0)) {
2144 DWORD src_mod = arg->src[0] & WINED3DSP_SRCMOD_MASK;
2145 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2146
2147 if (src_mod == WINED3DSPSM_DZ) {
2148 projected = TRUE;
2149 mask = WINED3DSP_WRITEMASK_2;
2150 } else if (src_mod == WINED3DSPSM_DW) {
2151 projected = TRUE;
2152 mask = WINED3DSP_WRITEMASK_3;
2153 } else {
2154 projected = FALSE;
2155 }
2156 } else {
2157 sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
2158 if(arg->opcode_token & WINED3DSI_TEXLD_PROJECT) {
2159 /* ps 2.0 texldp instruction always divides by the fourth component. */
2160 projected = TRUE;
2161 mask = WINED3DSP_WRITEMASK_3;
2162 } else {
2163 projected = FALSE;
2164 }
2165 }
2166
2167 if(deviceImpl->stateBlock->textures[sampler_idx] &&
2168 IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2169 texrect = TRUE;
2170 }
2171
2172 sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2173 shader_glsl_get_sample_function(sampler_type, projected, texrect, &sample_function);
2174 mask |= sample_function.coord_mask;
2175
2176 if (hex_version < WINED3DPS_VERSION(2,0)) {
2177 shader_glsl_get_write_mask(arg->dst, dst_swizzle);
2178 } else {
2179 shader_glsl_get_swizzle(arg->src[1], FALSE, arg->dst, dst_swizzle);
2180 }
2181
2182 /* 1.0-1.3: Use destination register as coordinate source.
2183 1.4+: Use provided coordinate source register. */
2184 if (hex_version < WINED3DPS_VERSION(1,4)) {
2185 char coord_mask[6];
2186 shader_glsl_get_write_mask(mask, coord_mask);
2187 shader_addline(arg->buffer, "%s(Psampler%u, T%u%s)%s);\n",
2188 sample_function.name, sampler_idx, sampler_idx, coord_mask, dst_swizzle);
2189 } else {
2190 glsl_src_param_t coord_param;
2191 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], mask, &coord_param);
2192 if(arg->opcode_token & WINED3DSI_TEXLD_BIAS) {
2193 glsl_src_param_t bias;
2194 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &bias);
2195
2196 shader_addline(arg->buffer, "%s(Psampler%u, %s, %s)%s);\n",
2197 sample_function.name, sampler_idx, coord_param.param_str,
2198 bias.param_str, dst_swizzle);
2199 } else {
2200 shader_addline(arg->buffer, "%s(Psampler%u, %s)%s);\n",
2201 sample_function.name, sampler_idx, coord_param.param_str, dst_swizzle);
2202 }
2203 }
2204 }
2205
2206 void shader_glsl_texldl(SHADER_OPCODE_ARG* arg) {
2207 IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*)arg->shader;
2208 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2209 glsl_sample_function_t sample_function;
2210 glsl_src_param_t coord_param, lod_param;
2211 char dst_swizzle[6];
2212 DWORD sampler_type;
2213 DWORD sampler_idx;
2214 BOOL texrect = FALSE;
2215
2216 shader_glsl_append_dst(arg->buffer, arg);
2217 shader_glsl_get_swizzle(arg->src[1], FALSE, arg->dst, dst_swizzle);
2218
2219 sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
2220 sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2221 if(deviceImpl->stateBlock->textures[sampler_idx] &&
2222 IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2223 texrect = TRUE;
2224 }
2225 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);
2226
2227 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &lod_param);
2228
2229 if (shader_is_pshader_version(This->baseShader.hex_version)) {
2230 /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
2231 * However, they seem to work just fine in fragment shaders as well. */
2232 WARN("Using %sLod in fragment shader.\n", sample_function.name);
2233 shader_addline(arg->buffer, "%sLod(Psampler%u, %s, %s)%s);\n",
2234 sample_function.name, sampler_idx, coord_param.param_str, lod_param.param_str, dst_swizzle);
2235 } else {
2236 shader_addline(arg->buffer, "%sLod(Vsampler%u, %s, %s)%s);\n",
2237 sample_function.name, sampler_idx, coord_param.param_str, lod_param.param_str, dst_swizzle);
2238 }
2239 }
2240
2241 void pshader_glsl_texcoord(SHADER_OPCODE_ARG* arg) {
2242
2243 /* FIXME: Make this work for more than just 2D textures */
2244
2245 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2246 SHADER_BUFFER* buffer = arg->buffer;
2247 DWORD hex_version = This->baseShader.hex_version;
2248 DWORD write_mask;
2249 char dst_mask[6];
2250
2251 write_mask = shader_glsl_append_dst(arg->buffer, arg);
2252 shader_glsl_get_write_mask(write_mask, dst_mask);
2253
2254 if (hex_version != WINED3DPS_VERSION(1,4)) {
2255 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2256 shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n", reg, dst_mask);
2257 } else {
2258 DWORD reg = arg->src[0] & WINED3DSP_REGNUM_MASK;
2259 DWORD src_mod = arg->src[0] & WINED3DSP_SRCMOD_MASK;
2260 char dst_swizzle[6];
2261
2262 shader_glsl_get_swizzle(arg->src[0], FALSE, write_mask, dst_swizzle);
2263
2264 if (src_mod == WINED3DSPSM_DZ) {
2265 glsl_src_param_t div_param;
2266 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2267 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &div_param);
2268
2269 if (mask_size > 1) {
2270 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2271 } else {
2272 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2273 }
2274 } else if (src_mod == WINED3DSPSM_DW) {
2275 glsl_src_param_t div_param;
2276 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2277 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &div_param);
2278
2279 if (mask_size > 1) {
2280 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2281 } else {
2282 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2283 }
2284 } else {
2285 shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
2286 }
2287 }
2288 }
2289
2290 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
2291 * Take a 3-component dot product of the TexCoord[dstreg] and src,
2292 * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2293 void pshader_glsl_texdp3tex(SHADER_OPCODE_ARG* arg) {
2294 glsl_src_param_t src0_param;
2295 char dst_mask[6];
2296 glsl_sample_function_t sample_function;
2297 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2298 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2299 DWORD sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2300
2301 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2302
2303 shader_glsl_append_dst(arg->buffer, arg);
2304 shader_glsl_get_write_mask(arg->dst, dst_mask);
2305
2306 /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
2307 * scalar, and projected sampling would require 4.
2308 *
2309 * It is a dependent read - not valid with conditional NP2 textures
2310 */
2311 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2312
2313 switch(count_bits(sample_function.coord_mask)) {
2314 case 1:
2315 shader_addline(arg->buffer, "%s(Psampler%u, dot(gl_TexCoord[%u].xyz, %s))%s);\n",
2316 sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2317 break;
2318
2319 case 2:
2320 shader_addline(arg->buffer, "%s(Psampler%u, vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0))%s);\n",
2321 sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2322 break;
2323
2324 case 3:
2325 shader_addline(arg->buffer, "%s(Psampler%u, vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0))%s);\n",
2326 sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2327 break;
2328 default:
2329 FIXME("Unexpected mask bitcount %d\n", count_bits(sample_function.coord_mask));
2330 }
2331 }
2332
2333 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
2334 * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2335 void pshader_glsl_texdp3(SHADER_OPCODE_ARG* arg) {
2336 glsl_src_param_t src0_param;
2337 DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
2338 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2339 DWORD dst_mask;
2340 unsigned int mask_size;
2341
2342 dst_mask = shader_glsl_append_dst(arg->buffer, arg);
2343 mask_size = shader_glsl_get_write_mask_size(dst_mask);
2344 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2345
2346 if (mask_size > 1) {
2347 shader_addline(arg->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
2348 } else {
2349 shader_addline(arg->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
2350 }
2351 }
2352
2353 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
2354 * Calculate the depth as dst.x / dst.y */
2355 void pshader_glsl_texdepth(SHADER_OPCODE_ARG* arg) {
2356 glsl_dst_param_t dst_param;
2357
2358 shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
2359
2360 /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
2361 * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
2362 * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
2363 * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
2364 * >= 1.0 or < 0.0
2365 */
2366 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);
2367 }
2368
2369 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
2370 * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2371 * Calculate tmp0.y = TexCoord[dstreg] . src.xyz; (tmp0.x has already been calculated)
2372 * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2373 */
2374 void pshader_glsl_texm3x2depth(SHADER_OPCODE_ARG* arg) {
2375 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2376 DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
2377 glsl_src_param_t src0_param;
2378
2379 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2380
2381 shader_addline(arg->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
2382 shader_addline(arg->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
2383 }
2384
2385 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
2386 * Calculate the 1st of a 2-row matrix multiplication. */
2387 void pshader_glsl_texm3x2pad(SHADER_OPCODE_ARG* arg) {
2388 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2389 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2390 SHADER_BUFFER* buffer = arg->buffer;
2391 glsl_src_param_t src0_param;
2392
2393 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2394 shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2395 }
2396
2397 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
2398 * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
2399 void pshader_glsl_texm3x3pad(SHADER_OPCODE_ARG* arg) {
2400
2401 IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2402 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2403 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2404 SHADER_BUFFER* buffer = arg->buffer;
2405 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2406 glsl_src_param_t src0_param;
2407
2408 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2409 shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + current_state->current_row, reg, src0_param.param_str);
2410 current_state->texcoord_w[current_state->current_row++] = reg;
2411 }
2412
2413 void pshader_glsl_texm3x2tex(SHADER_OPCODE_ARG* arg) {
2414 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2415 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2416 SHADER_BUFFER* buffer = arg->buffer;
2417 glsl_src_param_t src0_param;
2418 char dst_mask[6];
2419
2420 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2421 shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2422
2423 shader_glsl_append_dst(buffer, arg);
2424 shader_glsl_get_write_mask(arg->dst, dst_mask);
2425
2426 /* Sample the texture using the calculated coordinates */
2427 shader_addline(buffer, "texture2D(Psampler%u, tmp0.xy)%s);\n", reg, dst_mask);
2428 }
2429
2430 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
2431 * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
2432 void pshader_glsl_texm3x3tex(SHADER_OPCODE_ARG* arg) {
2433 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2434 glsl_src_param_t src0_param;
2435 char dst_mask[6];
2436 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2437 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2438 SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2439 DWORD sampler_type = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2440 glsl_sample_function_t sample_function;
2441
2442 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2443 shader_addline(arg->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2444
2445 shader_glsl_append_dst(arg->buffer, arg);
2446 shader_glsl_get_write_mask(arg->dst, dst_mask);
2447 /* Dependent read, not valid with conditional NP2 */
2448 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2449
2450 /* Sample the texture using the calculated coordinates */
2451 shader_addline(arg->buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2452
2453 current_state->current_row = 0;
2454 }
2455
2456 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
2457 * Perform the 3rd row of a 3x3 matrix multiply */
2458 void pshader_glsl_texm3x3(SHADER_OPCODE_ARG* arg) {
2459 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2460 glsl_src_param_t src0_param;
2461 char dst_mask[6];
2462 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2463 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2464 SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2465
2466 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2467
2468 shader_glsl_append_dst(arg->buffer, arg);
2469 shader_glsl_get_write_mask(arg->dst, dst_mask);
2470 shader_addline(arg->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
2471
2472 current_state->current_row = 0;
2473 }
2474
2475 /** Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
2476 * Peform the final texture lookup based on the previous 2 3x3 matrix multiplies */
2477 void pshader_glsl_texm3x3spec(SHADER_OPCODE_ARG* arg) {
2478
2479 IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2480 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2481 glsl_src_param_t src0_param;
2482 glsl_src_param_t src1_param;
2483 char dst_mask[6];
2484 SHADER_BUFFER* buffer = arg->buffer;
2485 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2486 DWORD stype = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2487 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2488 glsl_sample_function_t sample_function;
2489
2490 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2491 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_mask, &src1_param);
2492
2493 /* Perform the last matrix multiply operation */
2494 shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2495 /* Reflection calculation */
2496 shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
2497
2498 shader_glsl_append_dst(buffer, arg);
2499 shader_glsl_get_write_mask(arg->dst, dst_mask);
2500 /* Dependent read, not valid with conditional NP2 */
2501 shader_glsl_get_sample_function(stype, FALSE, FALSE, &sample_function);
2502
2503 /* Sample the texture */
2504 shader_addline(buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2505
2506 current_state->current_row = 0;
2507 }
2508
2509 /** Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
2510 * Peform the final texture lookup based on the previous 2 3x3 matrix multiplies */
2511 void pshader_glsl_texm3x3vspec(SHADER_OPCODE_ARG* arg) {
2512
2513 IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2514 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2515 SHADER_BUFFER* buffer = arg->buffer;
2516 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2517 glsl_src_param_t src0_param;
2518 char dst_mask[6];
2519 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2520 DWORD sampler_type = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2521 glsl_sample_function_t sample_function;
2522
2523 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2524
2525 /* Perform the last matrix multiply operation */
2526 shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
2527
2528 /* Construct the eye-ray vector from w coordinates */
2529 shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
2530 current_state->texcoord_w[0], current_state->texcoord_w[1], reg);
2531 shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
2532
2533 shader_glsl_append_dst(buffer, arg);
2534 shader_glsl_get_write_mask(arg->dst, dst_mask);
2535 /* Dependent read, not valid with conditional NP2 */
2536 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2537
2538 /* Sample the texture using the calculated coordinates */
2539 shader_addline(buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2540
2541 current_state->current_row = 0;
2542 }
2543
2544 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
2545 * Apply a fake bump map transform.
2546 * texbem is pshader <= 1.3 only, this saves a few version checks
2547 */
2548 void pshader_glsl_texbem(SHADER_OPCODE_ARG* arg) {
2549 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2550 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2551 char dst_swizzle[6];
2552 glsl_sample_function_t sample_function;
2553 glsl_src_param_t coord_param;
2554 DWORD sampler_type;
2555 DWORD sampler_idx;
2556 DWORD mask;
2557 DWORD flags;
2558 char coord_mask[6];
2559
2560 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2561 flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2562
2563 sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2564 /* Dependent read, not valid with conditional NP2 */
2565 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2566 mask = sample_function.coord_mask;
2567
2568 shader_glsl_get_write_mask(arg->dst, dst_swizzle);
2569
2570 shader_glsl_get_write_mask(mask, coord_mask);
2571
2572 /* with projective textures, texbem only divides the static texture coord, not the displacement,
2573 * so we can't let the GL handle this.
2574 */
2575 if (flags & WINED3DTTFF_PROJECTED) {
2576 DWORD div_mask=0;
2577 char coord_div_mask[3];
2578 switch (flags & ~WINED3DTTFF_PROJECTED) {
2579 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2580 case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
2581 case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
2582 case WINED3DTTFF_COUNT4:
2583 case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
2584 }
2585 shader_glsl_get_write_mask(div_mask, coord_div_mask);
2586 shader_addline(arg->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
2587 }
2588
2589 shader_glsl_append_dst(arg->buffer, arg);
2590 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &coord_param);
2591 if(arg->opcode->opcode == WINED3DSIO_TEXBEML) {
2592 glsl_src_param_t luminance_param;
2593 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &luminance_param);
2594 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",
2595 sample_function.name, sampler_idx, sampler_idx, coord_mask, sampler_idx, coord_param.param_str, coord_mask,
2596 luminance_param.param_str, sampler_idx, sampler_idx, dst_swizzle);
2597 } else {
2598 shader_addline(arg->buffer, "%s(Psampler%u, T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s )%s);\n",
2599 sample_function.name, sampler_idx, sampler_idx, coord_mask, sampler_idx, coord_param.param_str, coord_mask, dst_swizzle);
2600 }
2601 }
2602
2603 void pshader_glsl_bem(SHADER_OPCODE_ARG* arg) {
2604 glsl_src_param_t src0_param, src1_param;
2605 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2606
2607 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &src0_param);
2608 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &src1_param);
2609
2610 shader_glsl_append_dst(arg->buffer, arg);
2611 shader_addline(arg->buffer, "%s + bumpenvmat%d * %s);\n",
2612 src0_param.param_str, sampler_idx, src1_param.param_str);
2613 }
2614
2615 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
2616 * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
2617 void pshader_glsl_texreg2ar(SHADER_OPCODE_ARG* arg) {
2618
2619 glsl_src_param_t src0_param;
2620 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2621 char dst_mask[6];
2622
2623 shader_glsl_append_dst(arg->buffer, arg);
2624 shader_glsl_get_write_mask(arg->dst, dst_mask);
2625 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2626
2627 shader_addline(arg->buffer, "texture2D(Psampler%u, %s.wx)%s);\n", sampler_idx, src0_param.reg_name, dst_mask);
2628 }
2629
2630 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
2631 * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
2632 void pshader_glsl_texreg2gb(SHADER_OPCODE_ARG* arg) {
2633 glsl_src_param_t src0_param;
2634 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2635 char dst_mask[6];
2636
2637 shader_glsl_append_dst(arg->buffer, arg);
2638 shader_glsl_get_write_mask(arg->dst, dst_mask);
2639 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2640
2641 shader_addline(arg->buffer, "texture2D(Psampler%u, %s.yz)%s);\n", sampler_idx, src0_param.reg_name, dst_mask);
2642 }
2643
2644 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
2645 * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
2646 void pshader_glsl_texreg2rgb(SHADER_OPCODE_ARG* arg) {
2647 glsl_src_param_t src0_param;
2648 char dst_mask[6];
2649 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2650 DWORD sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2651 glsl_sample_function_t sample_function;
2652
2653 shader_glsl_append_dst(arg->buffer, arg);
2654 shader_glsl_get_write_mask(arg->dst, dst_mask);
2655 /* Dependent read, not valid with conditional NP2 */
2656 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2657 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], sample_function.coord_mask, &src0_param);
2658
2659 shader_addline(arg->buffer, "%s(Psampler%u, %s)%s);\n", sample_function.name, sampler_idx, src0_param.param_str, dst_mask);
2660 }
2661
2662 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
2663 * If any of the first 3 components are < 0, discard this pixel */
2664 void pshader_glsl_texkill(SHADER_OPCODE_ARG* arg) {
2665 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2666 DWORD hex_version = This->baseShader.hex_version;
2667 glsl_dst_param_t dst_param;
2668
2669 /* The argument is a destination parameter, and no writemasks are allowed */
2670 shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
2671 if((hex_version >= WINED3DPS_VERSION(2,0))) {
2672 /* 2.0 shaders compare all 4 components in texkill */
2673 shader_addline(arg->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
2674 } else {
2675 /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
2676 * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
2677 * 4 components are defined, only the first 3 are used
2678 */
2679 shader_addline(arg->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
2680 }
2681 }
2682
2683 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
2684 * dst = dot2(src0, src1) + src2 */
2685 void pshader_glsl_dp2add(SHADER_OPCODE_ARG* arg) {
2686 glsl_src_param_t src0_param;
2687 glsl_src_param_t src1_param;
2688 glsl_src_param_t src2_param;
2689 DWORD write_mask;
2690 unsigned int mask_size;
2691
2692 write_mask = shader_glsl_append_dst(arg->buffer, arg);
2693 mask_size = shader_glsl_get_write_mask_size(write_mask);
2694
2695 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
2696 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
2697 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], WINED3DSP_WRITEMASK_0, &src2_param);
2698
2699 if (mask_size > 1) {
2700 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);
2701 } else {
2702 shader_addline(arg->buffer, "dot(%s, %s) + %s);\n", src0_param.param_str, src1_param.param_str, src2_param.param_str);
2703 }
2704 }
2705
2706 void pshader_glsl_input_pack(
2707 SHADER_BUFFER* buffer,
2708 semantic* semantics_in,
2709 IWineD3DPixelShader *iface) {
2710
2711 unsigned int i;
2712 IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *) iface;
2713
2714 for (i = 0; i < MAX_REG_INPUT; i++) {
2715
2716 DWORD usage_token = semantics_in[i].usage;
2717 DWORD register_token = semantics_in[i].reg;
2718 DWORD usage, usage_idx;
2719 char reg_mask[6];
2720
2721 /* Uninitialized */
2722 if (!usage_token) continue;
2723 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2724 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2725 shader_glsl_get_write_mask(register_token, reg_mask);
2726
2727 switch(usage) {
2728
2729 case WINED3DDECLUSAGE_TEXCOORD:
2730 if(usage_idx < 8 && This->vertexprocessing == pretransformed) {
2731 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
2732 This->input_reg_map[i], reg_mask, usage_idx, reg_mask);
2733 } else {
2734 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2735 This->input_reg_map[i], reg_mask, reg_mask);
2736 }
2737 break;
2738
2739 case WINED3DDECLUSAGE_COLOR:
2740 if (usage_idx == 0)
2741 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
2742 This->input_reg_map[i], reg_mask, reg_mask);
2743 else if (usage_idx == 1)
2744 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
2745 This->input_reg_map[i], reg_mask, reg_mask);
2746 else
2747 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2748 This->input_reg_map[i], reg_mask, reg_mask);
2749 break;
2750
2751 default:
2752 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2753 This->input_reg_map[i], reg_mask, reg_mask);
2754 }
2755 }
2756 }
2757
2758 /*********************************************
2759 * Vertex Shader Specific Code begins here
2760 ********************************************/
2761
2762 static void add_glsl_program_entry(IWineD3DDeviceImpl *device, struct glsl_shader_prog_link *entry) {
2763 glsl_program_key_t *key;
2764
2765 key = HeapAlloc(GetProcessHeap(), 0, sizeof(glsl_program_key_t));
2766 key->vshader = entry->vshader;
2767 key->pshader = entry->pshader;
2768
2769 hash_table_put(device->glsl_program_lookup, key, entry);
2770 }
2771
2772 static struct glsl_shader_prog_link *get_glsl_program_entry(IWineD3DDeviceImpl *device,
2773 GLhandleARB vshader, GLhandleARB pshader) {
2774 glsl_program_key_t key;
2775
2776 key.vshader = vshader;
2777 key.pshader = pshader;
2778
2779 return (struct glsl_shader_prog_link *)hash_table_get(device->glsl_program_lookup, &key);
2780 }
2781
2782 void delete_glsl_program_entry(IWineD3DDevice *iface, struct glsl_shader_prog_link *entry) {
2783 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
2784 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
2785 glsl_program_key_t *key;
2786
2787 key = HeapAlloc(GetProcessHeap(), 0, sizeof(glsl_program_key_t));
2788 key->vshader = entry->vshader;
2789 key->pshader = entry->pshader;
2790 hash_table_remove(This->glsl_program_lookup, key);
2791
2792 GL_EXTCALL(glDeleteObjectARB(entry->programId));
2793 if (entry->vshader) list_remove(&entry->vshader_entry);
2794 if (entry->pshader) list_remove(&entry->pshader_entry);
2795 HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
2796 HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
2797 HeapFree(GetProcessHeap(), 0, entry);
2798 }
2799
2800 static void handle_ps3_input(SHADER_BUFFER *buffer, semantic *semantics_in, semantic *semantics_out, WineD3D_GL_Info *gl_info, DWORD *map) {
2801 unsigned int i, j;
2802 DWORD usage_token, usage_token_out;
2803 DWORD register_token, register_token_out;
2804 DWORD usage, usage_idx, usage_out, usage_idx_out;
2805 DWORD *set;
2806 char reg_mask[6], reg_mask_out[6];
2807
2808 set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (GL_LIMITS(glsl_varyings) / 4));
2809
2810 for(i = 0; i < MAX_REG_INPUT; i++) {
2811 usage_token = semantics_in[i].usage;
2812 if (!usage_token) continue;
2813 if(map[i] >= (GL_LIMITS(glsl_varyings) / 4)) {
2814 FIXME("More input varyings declared than supported, expect issues\n");
2815 continue;
2816 } else if(map[i] == -1) {
2817 /* Declared, but not read register */
2818 continue;
2819 }
2820 register_token = semantics_in[i].reg;
2821
2822 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2823 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2824 set[map[i]] = shader_glsl_get_write_mask(register_token, reg_mask);
2825
2826 if(!semantics_out) {
2827 switch(usage) {
2828 case WINED3DDECLUSAGE_COLOR:
2829 if (usage_idx == 0)
2830 shader_addline(buffer, "IN[%u]%s = gl_FrontColor%s;\n",
2831 map[i], reg_mask, reg_mask);
2832 else if (usage_idx == 1)
2833 shader_addline(buffer, "IN[%u]%s = gl_FrontSecondaryColor%s;\n",
2834 map[i], reg_mask, reg_mask);
2835 else
2836 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2837 map[i], reg_mask, reg_mask);
2838 break;
2839
2840 case WINED3DDECLUSAGE_TEXCOORD:
2841 if (usage_idx < 8) {
2842 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
2843 map[i], reg_mask, usage_idx, reg_mask);
2844 } else {
2845 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2846 map[i], reg_mask, reg_mask);
2847 }
2848 break;
2849
2850 case WINED3DDECLUSAGE_FOG:
2851 shader_addline(buffer, "IN[%u]%s = vec4(gl_FogFragCoord, 0.0, 0.0, 0.0)%s;\n",
2852 map[i], reg_mask, reg_mask);
2853 break;
2854
2855 default:
2856 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2857 map[i], reg_mask, reg_mask);
2858 }
2859 } else {
2860 BOOL found = FALSE;
2861 for(j = 0; j < MAX_REG_OUTPUT; j++) {
2862 usage_token_out = semantics_out[j].usage;
2863 if (!usage_token_out) continue;
2864 register_token_out = semantics_out[j].reg;
2865
2866 usage_out = (usage_token_out & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2867 usage_idx_out = (usage_token_out & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2868 shader_glsl_get_write_mask(register_token_out, reg_mask_out);
2869
2870 if(usage == usage_out &&
2871 usage_idx == usage_idx_out) {
2872 shader_addline(buffer, "IN[%u]%s = OUT[%u]%s;\n",
2873 map[i], reg_mask, j, reg_mask);
2874 found = TRUE;
2875 }
2876 }
2877 if(!found) {
2878 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2879 map[i], reg_mask, reg_mask);
2880 }
2881 }
2882 }
2883
2884 /* This is solely to make the compiler / linker happy and avoid warning about undefined
2885 * varyings. It shouldn't result in any real code executed on the GPU, since all read
2886 * input varyings are assigned above, if the optimizer works properly.
2887 */
2888 for(i = 0; i < GL_LIMITS(glsl_varyings) / 4; i++) {
2889 if(set[i] != WINED3DSP_WRITEMASK_ALL) {
2890 unsigned int size = 0;
2891 memset(reg_mask, 0, sizeof(reg_mask));
2892 if(!(set[i] & WINED3DSP_WRITEMASK_0)) {
2893 reg_mask[size] = 'x';
2894 size++;
2895 }
2896 if(!(set[i] & WINED3DSP_WRITEMASK_1)) {
2897 reg_mask[size] = 'y';
2898 size++;
2899 }
2900 if(!(set[i] & WINED3DSP_WRITEMASK_2)) {
2901 reg_mask[size] = 'z';
2902 size++;
2903 }
2904 if(!(set[i] & WINED3DSP_WRITEMASK_3)) {
2905 reg_mask[size] = 'w';
2906 size++;
2907 }
2908 switch(size) {
2909 case 1:
2910 shader_addline(buffer, "IN[%u].%s = 0.0;\n", i, reg_mask);
2911 break;
2912 case 2:
2913 shader_addline(buffer, "IN[%u].%s = vec2(0.0, 0.0);\n", i, reg_mask);
2914 break;
2915 case 3:
2916 shader_addline(buffer, "IN[%u].%s = vec3(0.0, 0.0, 0.0);\n", i, reg_mask);
2917 break;
2918 case 4:
2919 shader_addline(buffer, "IN[%u].%s = vec4(0.0, 0.0, 0.0, 0.0);\n", i, reg_mask);
2920 break;
2921 }
2922 }
2923 }
2924
2925 HeapFree(GetProcessHeap(), 0, set);
2926 }
2927
2928 static GLhandleARB generate_param_reorder_function(IWineD3DVertexShader *vertexshader,
2929 IWineD3DPixelShader *pixelshader,
2930 WineD3D_GL_Info *gl_info) {
2931 GLhandleARB ret = 0;
2932 IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) vertexshader;
2933 IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pixelshader;
2934 DWORD vs_major = vs ? WINED3DSHADER_VERSION_MAJOR(vs->baseShader.hex_version) : 0;
2935 DWORD ps_major = ps ? WINED3DSHADER_VERSION_MAJOR(ps->baseShader.hex_version) : 0;
2936 unsigned int i;
2937 SHADER_BUFFER buffer;
2938 DWORD usage_token;
2939 DWORD register_token;
2940 DWORD usage, usage_idx, writemask;
2941 char reg_mask[6];
2942 semantic *semantics_out, *semantics_in;
2943
2944 buffer.buffer = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, SHADER_PGMSIZE);
2945 buffer.bsize = 0;
2946 buffer.lineNo = 0;
2947 buffer.newline = TRUE;
2948
2949 if(vs_major < 3 && ps_major < 3) {
2950 /* That one is easy: The vertex shader writes to the builtin varyings, the pixel shader reads from them.
2951 * Take care about the texcoord .w fixup though if we're using the fixed function fragment pipeline
2952 */
2953 if((GLINFO_LOCATION).set_texcoord_w && ps_major == 0 && vs_major > 0) {
2954 shader_addline(&buffer, "void order_ps_input() {\n");
2955 for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
2956 if(vs->baseShader.reg_maps.texcoord_mask[i] != 0 &&
2957 vs->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
2958 shader_addline(&buffer, "gl_TexCoord[%u].w = 1.0;\n", i);
2959 }
2960 }
2961 shader_addline(&buffer, "}\n");
2962 } else {
2963 shader_addline(&buffer, "void order_ps_input() { /* do nothing */ }\n");
2964 }
2965 } else if(ps_major < 3 && vs_major >= 3) {
2966 /* The vertex shader writes to its own varyings, the pixel shader needs them in the builtin ones */
2967 semantics_out = vs->semantics_out;
2968
2969 shader_addline(&buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
2970 for(i = 0; i < MAX_REG_OUTPUT; i++) {
2971 usage_token = semantics_out[i].usage;
2972 if (!usage_token) continue;
2973 register_token = semantics_out[i].reg;
2974
2975 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2976 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2977 writemask = shader_glsl_get_write_mask(register_token, reg_mask);
2978
2979 switch(usage) {
2980 case WINED3DDECLUSAGE_COLOR:
2981 if (usage_idx == 0)
2982 shader_addline(&buffer, "gl_FrontColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
2983 else if (usage_idx == 1)
2984 shader_addline(&buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
2985 break;
2986
2987 case WINED3DDECLUSAGE_POSITION:
2988 shader_addline(&buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
2989 break;
2990
2991 case WINED3DDECLUSAGE_TEXCOORD:
2992 if (usage_idx < 8) {
2993 if(!(GLINFO_LOCATION).set_texcoord_w || ps_major > 0) writemask |= WINED3DSP_WRITEMASK_3;
2994
2995 shader_addline(&buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
2996 usage_idx, reg_mask, i, reg_mask);
2997 if(!(writemask & WINED3DSP_WRITEMASK_3)) {
2998 shader_addline(&buffer, "gl_TexCoord[%u].w = 1.0;\n", usage_idx);
2999 }
3000 }
3001 break;
3002
3003 case WINED3DDECLUSAGE_PSIZE:
3004 shader_addline(&buffer, "gl_PointSize = OUT[%u].x;\n", i);
3005 break;
3006
3007 case WINED3DDECLUSAGE_FOG:
3008 shader_addline(&buffer, "gl_FogFragCoord = OUT[%u].%c;\n", i, reg_mask[1]);
3009 break;
3010
3011 default:
3012 break;
3013 }
3014 }
3015 shader_addline(&buffer, "}\n");
3016
3017 } else if(ps_major >= 3 && vs_major >= 3) {
3018 semantics_out = vs->semantics_out;
3019 semantics_in = ps->semantics_in;
3020
3021 /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3022 shader_addline(&buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
3023 shader_addline(&buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3024
3025 /* First, sort out position and point size. Those are not passed to the pixel shader */
3026 for(i = 0; i < MAX_REG_OUTPUT; i++) {
3027 usage_token = semantics_out[i].usage;
3028 if (!usage_token) continue;
3029 register_token = semantics_out[i].reg;
3030
3031 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
3032 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
3033 shader_glsl_get_write_mask(register_token, reg_mask);
3034
3035 switch(usage) {
3036 case WINED3DDECLUSAGE_POSITION:
3037 shader_addline(&buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3038 break;
3039
3040 case WINED3DDECLUSAGE_PSIZE:
3041 shader_addline(&buffer, "gl_PointSize = OUT[%u].x;\n", i);
3042 break;
3043
3044 default:
3045 break;
3046 }
3047 }
3048
3049 /* Then, fix the pixel shader input */
3050 handle_ps3_input(&buffer, semantics_in, semantics_out, gl_info, ps->input_reg_map);
3051
3052 shader_addline(&buffer, "}\n");
3053 } else if(ps_major >= 3 && vs_major < 3) {
3054 semantics_in = ps->semantics_in;
3055
3056 shader_addline(&buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
3057 shader_addline(&buffer, "void order_ps_input() {\n");
3058 /* The vertex shader wrote to the builtin varyings. There is no need to figure out position and
3059 * point size, but we depend on the optimizers kindness to find out that the pixel shader doesn't
3060 * read gl_TexCoord and gl_ColorX, otherwise we'll run out of varyings
3061 */
3062 handle_ps3_input(&buffer, semantics_in, NULL, gl_info, ps->input_reg_map);
3063 shader_addline(&buffer, "}\n");
3064 } else {
3065 ERR("Unexpected vertex and pixel shader version condition: vs: %d, ps: %d\n", vs_major, ps_major);
3066 }
3067
3068 ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3069 checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3070 GL_EXTCALL(glShaderSourceARB(ret, 1, (const char**)&buffer.buffer, NULL));
3071 checkGLcall("glShaderSourceARB(ret, 1, (const char**)&buffer.buffer, NULL)");
3072 GL_EXTCALL(glCompileShaderARB(ret));
3073 checkGLcall("glCompileShaderARB(ret)");
3074
3075 HeapFree(GetProcessHeap(), 0, buffer.buffer);
3076 return ret;
3077 }
3078
3079 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
3080 * It sets the programId on the current StateBlock (because it should be called
3081 * inside of the DrawPrimitive() part of the render loop).
3082 *
3083 * If a program for the given combination does not exist, create one, and store
3084 * the program in the hash table. If it creates a program, it will link the
3085 * given objects, too.
3086 */
3087 static void set_glsl_shader_program(IWineD3DDevice *iface, BOOL use_ps, BOOL use_vs) {
3088 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3089 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3090 IWineD3DPixelShader *pshader = This->stateBlock->pixelShader;
3091 IWineD3DVertexShader *vshader = This->stateBlock->vertexShader;
3092 struct glsl_shader_prog_link *entry = NULL;
3093 GLhandleARB programId = 0;
3094 GLhandleARB reorder_shader_id = 0;
3095 int i;
3096 char glsl_name[8];
3097
3098 GLhandleARB vshader_id = use_vs ? ((IWineD3DBaseShaderImpl*)vshader)->baseShader.prgId : 0;
3099 GLhandleARB pshader_id = use_ps ? ((IWineD3DBaseShaderImpl*)pshader)->baseShader.prgId : 0;
3100 entry = get_glsl_program_entry(This, vshader_id, pshader_id);
3101 if (entry) {
3102 This->stateBlock->glsl_program = entry;
3103 return;
3104 }
3105
3106 /* If we get to this point, then no matching program exists, so we create one */
3107 programId = GL_EXTCALL(glCreateProgramObjectARB());
3108 TRACE("Created new GLSL shader program %u\n", programId);
3109
3110 /* Create the entry */
3111 entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
3112 entry->programId = programId;
3113 entry->vshader = vshader_id;
3114 entry->pshader = pshader_id;
3115 /* Add the hash table entry */
3116 add_glsl_program_entry(This, entry);
3117
3118 /* Set the current program */
3119 This->stateBlock->glsl_program = entry;
3120
3121 /* Attach GLSL vshader */
3122 if (vshader_id) {
3123 int max_attribs = 16; /* TODO: Will this always be the case? It is at the moment... */
3124 char tmp_name[10];
3125
3126 reorder_shader_id = generate_param_reorder_function(vshader, pshader, gl_info);
3127 TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
3128 GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
3129 checkGLcall("glAttachObjectARB");
3130 /* Flag the reorder function for deletion, then it will be freed automatically when the program
3131 * is destroyed
3132 */
3133 GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
3134
3135 TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
3136 GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
3137 checkGLcall("glAttachObjectARB");
3138
3139 /* Bind vertex attributes to a corresponding index number to match
3140 * the same index numbers as ARB_vertex_programs (makes loading
3141 * vertex attributes simpler). With this method, we can use the
3142 * exact same code to load the attributes later for both ARB and
3143 * GLSL shaders.
3144 *
3145 * We have to do this here because we need to know the Program ID
3146 * in order to make the bindings work, and it has to be done prior
3147 * to linking the GLSL program. */
3148 for (i = 0; i < max_attribs; ++i) {
3149 if (((IWineD3DBaseShaderImpl*)vshader)->baseShader.reg_maps.attributes[i]) {
3150 snprintf(tmp_name, sizeof(tmp_name), "attrib%i", i);
3151 GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
3152 }
3153 }
3154 checkGLcall("glBindAttribLocationARB");
3155
3156 list_add_head(&((IWineD3DBaseShaderImpl *)vshader)->baseShader.linked_programs, &entry->vshader_entry);
3157 }
3158
3159 /* Attach GLSL pshader */
3160 if (pshader_id) {
3161 TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
3162 GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
3163 checkGLcall("glAttachObjectARB");
3164
3165 list_add_head(&((IWineD3DBaseShaderImpl *)pshader)->baseShader.linked_programs, &entry->pshader_entry);
3166 }
3167
3168 /* Link the program */
3169 TRACE("Linking GLSL shader program %u\n", programId);
3170 GL_EXTCALL(glLinkProgramARB(programId));
3171 print_glsl_info_log(&GLINFO_LOCATION, programId);
3172
3173 entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0, sizeof(GLhandleARB) * GL_LIMITS(vshader_constantsF));
3174 for (i = 0; i < GL_LIMITS(vshader_constantsF); ++i) {
3175 snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
3176 entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3177 }
3178 for (i = 0; i < MAX_CONST_I; ++i) {
3179 snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
3180 entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3181 }
3182 entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0, sizeof(GLhandleARB) * GL_LIMITS(pshader_constantsF));
3183 for (i = 0; i < GL_LIMITS(pshader_constantsF); ++i) {
3184 snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
3185 entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3186 }
3187 for (i = 0; i < MAX_CONST_I; ++i) {
3188 snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
3189 entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3190 }
3191
3192 if(pshader) {
3193 for(i = 0; i < ((IWineD3DPixelShaderImpl*)pshader)->numbumpenvmatconsts; i++) {
3194 char name[32];
3195 sprintf(name, "bumpenvmat%d", ((IWineD3DPixelShaderImpl*)pshader)->bumpenvmatconst[i].texunit);
3196 entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3197 sprintf(name, "luminancescale%d", ((IWineD3DPixelShaderImpl*)pshader)->luminanceconst[i].texunit);
3198 entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3199 sprintf(name, "luminanceoffset%d", ((IWineD3DPixelShaderImpl*)pshader)->luminanceconst[i].texunit);
3200 entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3201 }
3202 }
3203
3204
3205 entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
3206 entry->srgb_comparison_location = GL_EXTCALL(glGetUniformLocationARB(programId, "srgb_comparison"));
3207 entry->srgb_mul_low_location = GL_EXTCALL(glGetUniformLocationARB(programId, "srgb_mul_low"));
3208 entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
3209 checkGLcall("Find glsl program uniform locations");
3210
3211 /* Set the shader to allow uniform loading on it */
3212 GL_EXTCALL(glUseProgramObjectARB(programId));
3213 checkGLcall("glUseProgramObjectARB(programId)");
3214
3215 /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
3216 * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
3217 * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
3218 * vertex shader with fixed function pixel processing is used we make sure that the card
3219 * supports enough samplers to allow the max number of vertex samplers with all possible
3220 * fixed function fragment processing setups. So once the program is linked these samplers
3221 * won't change.
3222 */
3223 if(vshader_id) {
3224 /* Load vertex shader samplers */
3225 shader_glsl_load_vsamplers(gl_info, (IWineD3DStateBlock*)This->stateBlock, programId);
3226 }
3227 if(pshader_id) {
3228 /* Load pixel shader samplers */
3229 shader_glsl_load_psamplers(gl_info, (IWineD3DStateBlock*)This->stateBlock, programId);
3230 }
3231 }
3232
3233 static GLhandleARB create_glsl_blt_shader(WineD3D_GL_Info *gl_info) {
3234 GLhandleARB program_id;
3235 GLhandleARB vshader_id, pshader_id;
3236 const char *blt_vshader[] = {
3237 "void main(void)\n"
3238 "{\n"
3239 " gl_Position = gl_Vertex;\n"
3240 " gl_FrontColor = vec4(1.0);\n"
3241 " gl_TexCoord[0].x = (gl_Vertex.x * 0.5) + 0.5;\n"
3242 " gl_TexCoord[0].y = (-gl_Vertex.y * 0.5) + 0.5;\n"
3243 "}\n"
3244 };
3245
3246 const char *blt_pshader[] = {
3247 "uniform sampler2D sampler;\n"
3248 "void main(void)\n"
3249 "{\n"
3250 " gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
3251 "}\n"
3252 };
3253
3254 vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3255 GL_EXTCALL(glShaderSourceARB(vshader_id, 1, blt_vshader, NULL));
3256 GL_EXTCALL(glCompileShaderARB(vshader_id));
3257
3258 pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3259 GL_EXTCALL(glShaderSourceARB(pshader_id, 1, blt_pshader, NULL));
3260 GL_EXTCALL(glCompileShaderARB(pshader_id));
3261
3262 program_id = GL_EXTCALL(glCreateProgramObjectARB());
3263 GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
3264 GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
3265 GL_EXTCALL(glLinkProgramARB(program_id));
3266
3267 print_glsl_info_log(&GLINFO_LOCATION, program_id);
3268
3269 /* Once linked we can mark the shaders for deletion. They will be deleted once the program
3270 * is destroyed
3271 */
3272 GL_EXTCALL(glDeleteObjectARB(vshader_id));
3273 GL_EXTCALL(glDeleteObjectARB(pshader_id));
3274 return program_id;
3275 }
3276
3277 static void shader_glsl_select(IWineD3DDevice *iface, BOOL usePS, BOOL useVS) {
3278 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3279 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3280 GLhandleARB program_id = 0;
3281
3282 if (useVS || usePS) set_glsl_shader_program(iface, usePS, useVS);
3283 else This->stateBlock->glsl_program = NULL;
3284
3285 program_id = This->stateBlock->glsl_program ? This->stateBlock->glsl_program->programId : 0;
3286 if (program_id) TRACE("Using GLSL program %u\n", program_id);
3287 GL_EXTCALL(glUseProgramObjectARB(program_id));
3288 checkGLcall("glUseProgramObjectARB");
3289 }
3290
3291 static void shader_glsl_select_depth_blt(IWineD3DDevice *iface) {
3292 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3293 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3294 struct shader_glsl_priv *priv = (struct shader_glsl_priv *) This->shader_priv;
3295 static GLhandleARB loc = -1;
3296
3297 if (!priv->depth_blt_glsl_program_id) {
3298 priv->depth_blt_glsl_program_id = create_glsl_blt_shader(gl_info);
3299 loc = GL_EXTCALL(glGetUniformLocationARB(priv->depth_blt_glsl_program_id, "sampler"));
3300 }
3301
3302 GL_EXTCALL(glUseProgramObjectARB(priv->depth_blt_glsl_program_id));
3303 GL_EXTCALL(glUniform1iARB(loc, 0));
3304 }
3305
3306 static void shader_glsl_destroy_depth_blt(IWineD3DDevice *iface) {
3307 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3308 struct shader_glsl_priv *priv = (struct shader_glsl_priv *) This->shader_priv;
3309 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3310
3311 if(priv->depth_blt_glsl_program_id) {
3312 GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_glsl_program_id));
3313 priv->depth_blt_glsl_program_id = 0;
3314 }
3315 }
3316
3317 static void shader_glsl_cleanup(IWineD3DDevice *iface) {
3318 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3319 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3320 GL_EXTCALL(glUseProgramObjectARB(0));
3321 }
3322
3323 static void shader_glsl_destroy(IWineD3DBaseShader *iface) {
3324 struct list *linked_programs;
3325 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) iface;
3326 WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *) This->baseShader.device)->adapter->gl_info;
3327
3328 /* Note: Do not use QueryInterface here to find out which shader type this is because this code
3329 * can be called from IWineD3DBaseShader::Release
3330 */
3331 char pshader = shader_is_pshader_version(This->baseShader.hex_version);
3332
3333 if(This->baseShader.prgId == 0) return;
3334 linked_programs = &This->baseShader.linked_programs;
3335
3336 TRACE("Deleting linked programs\n");
3337 if (linked_programs->next) {
3338 struct glsl_shader_prog_link *entry, *entry2;
3339
3340 if(pshader) {
3341 LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
3342 delete_glsl_program_entry(This->baseShader.device, entry);
3343 }
3344 } else {
3345 LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
3346 delete_glsl_program_entry(This->baseShader.device, entry);
3347 }
3348 }
3349 }
3350
3351 TRACE("Deleting shader object %u\n", This->baseShader.prgId);
3352 GL_EXTCALL(glDeleteObjectARB(This->baseShader.prgId));
3353 checkGLcall("glDeleteObjectARB");
3354 This->baseShader.prgId = 0;
3355 This->baseShader.is_compiled = FALSE;
3356 }
3357
3358 static unsigned int glsl_program_key_hash(void *key) {
3359 glsl_program_key_t *k = (glsl_program_key_t *)key;
3360
3361 unsigned int hash = k->vshader | k->pshader << 16;
3362 hash += ~(hash << 15);
3363 hash ^= (hash >> 10);
3364 hash += (hash << 3);
3365 hash ^= (hash >> 6);
3366 hash += ~(hash << 11);
3367 hash ^= (hash >> 16);
3368
3369 return hash;
3370 }
3371
3372 static BOOL glsl_program_key_compare(void *keya, void *keyb) {
3373 glsl_program_key_t *ka = (glsl_program_key_t *)keya;
3374 glsl_program_key_t *kb = (glsl_program_key_t *)keyb;
3375
3376 return ka->vshader == kb->vshader && ka->pshader == kb->pshader;
3377 }
3378
3379 static HRESULT shader_glsl_alloc(IWineD3DDevice *iface) {
3380 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3381 This->shader_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
3382 This->glsl_program_lookup = hash_table_create(&glsl_program_key_hash, &glsl_program_key_compare);
3383 return WINED3D_OK;
3384 }
3385
3386 static void shader_glsl_free(IWineD3DDevice *iface) {
3387 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3388 HeapFree(GetProcessHeap(), 0, This->shader_priv);
3389 This->shader_priv = NULL;
3390 }
3391
3392 static BOOL shader_glsl_dirty_const(IWineD3DDevice *iface) {
3393 /* TODO: GL_EXT_bindable_uniform can be used to share constants across shaders */
3394 return FALSE;
3395 }
3396
3397 static void shader_glsl_generate_pshader(IWineD3DPixelShader *iface, SHADER_BUFFER *buffer) {
3398 IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
3399 shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3400 CONST DWORD *function = This->baseShader.function;
3401 const char *fragcolor;
3402 WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3403
3404 /* Create the hw GLSL shader object and assign it as the baseShader.prgId */
3405 GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3406
3407 if (GL_SUPPORT(ARB_DRAW_BUFFERS)) {
3408 shader_addline(buffer, "#extension GL_ARB_draw_buffers : enable\n");
3409 }
3410 if (GL_SUPPORT(ARB_TEXTURE_RECTANGLE)) {
3411 /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
3412 * drivers write a warning if we don't do so
3413 */
3414 shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
3415 }
3416
3417 /* Base Declarations */
3418 shader_generate_glsl_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION);
3419
3420 /* Pack 3.0 inputs */
3421 if (This->baseShader.hex_version >= WINED3DPS_VERSION(3,0)) {
3422
3423 if(((IWineD3DDeviceImpl *) This->baseShader.device)->strided_streams.u.s.position_transformed) {
3424 This->vertexprocessing = pretransformed;
3425 pshader_glsl_input_pack(buffer, This->semantics_in, iface);
3426 } else if(!use_vs((IWineD3DDeviceImpl *) This->baseShader.device)) {
3427 This->vertexprocessing = fixedfunction;
3428 pshader_glsl_input_pack(buffer, This->semantics_in, iface);
3429 } else {
3430 This->vertexprocessing = vertexshader;
3431 }
3432 }
3433
3434 /* Base Shader Body */
3435 shader_generate_main( (IWineD3DBaseShader*) This, buffer, reg_maps, function);
3436
3437 /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
3438 if (This->baseShader.hex_version < WINED3DPS_VERSION(2,0)) {
3439 /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
3440 if(GL_SUPPORT(ARB_DRAW_BUFFERS))
3441 shader_addline(buffer, "gl_FragData[0] = R0;\n");
3442 else
3443 shader_addline(buffer, "gl_FragColor = R0;\n");
3444 }
3445
3446 if(GL_SUPPORT(ARB_DRAW_BUFFERS)) {
3447 fragcolor = "gl_FragData[0]";
3448 } else {
3449 fragcolor = "gl_FragColor";
3450 }
3451 if(This->srgb_enabled) {
3452 shader_addline(buffer, "tmp0.xyz = pow(%s.xyz, vec3(%f, %f, %f)) * vec3(%f, %f, %f) - vec3(%f, %f, %f);\n",
3453 fragcolor, srgb_pow, srgb_pow, srgb_pow, srgb_mul_high, srgb_mul_high, srgb_mul_high,
3454 srgb_sub_high, srgb_sub_high, srgb_sub_high);
3455 shader_addline(buffer, "tmp1.xyz = %s.xyz * srgb_mul_low.xyz;\n", fragcolor);
3456 shader_addline(buffer, "%s.x = %s.x < srgb_comparison.x ? tmp1.x : tmp0.x;\n", fragcolor, fragcolor);
3457 shader_addline(buffer, "%s.y = %s.y < srgb_comparison.y ? tmp1.y : tmp0.y;\n", fragcolor, fragcolor);
3458 shader_addline(buffer, "%s.z = %s.z < srgb_comparison.z ? tmp1.z : tmp0.z;\n", fragcolor, fragcolor);
3459 shader_addline(buffer, "%s = clamp(%s, 0.0, 1.0);\n", fragcolor, fragcolor);
3460 }
3461 /* Pixel shader < 3.0 do not replace the fog stage.
3462 * This implements linear fog computation and blending.
3463 * TODO: non linear fog
3464 * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
3465 * -1/(e-s) and e/(e-s) respectively.
3466 */
3467 if(This->baseShader.hex_version < WINED3DPS_VERSION(3,0)) {
3468 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * gl_Fog.start + gl_Fog.end, 0.0, 1.0);\n");
3469 shader_addline(buffer, "%s.xyz = mix(gl_Fog.color.xyz, %s.xyz, Fog);\n", fragcolor, fragcolor);
3470 }
3471
3472 shader_addline(buffer, "}\n");
3473
3474 TRACE("Compiling shader object %u\n", shader_obj);
3475 GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
3476 GL_EXTCALL(glCompileShaderARB(shader_obj));
3477 print_glsl_info_log(&GLINFO_LOCATION, shader_obj);
3478
3479 /* Store the shader object */
3480 This->baseShader.prgId = shader_obj;
3481 }
3482
3483 static void shader_glsl_generate_vshader(IWineD3DVertexShader *iface, SHADER_BUFFER *buffer) {
3484 IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
3485 shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3486 CONST DWORD *function = This->baseShader.function;
3487 WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3488
3489 /* Create the hw GLSL shader program and assign it as the baseShader.prgId */
3490 GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3491
3492 /* Base Declarations */
3493 shader_generate_glsl_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION);
3494
3495 /* Base Shader Body */
3496 shader_generate_main( (IWineD3DBaseShader*) This, buffer, reg_maps, function);
3497
3498 /* Unpack 3.0 outputs */
3499 if (This->baseShader.hex_version >= WINED3DVS_VERSION(3,0)) {
3500 shader_addline(buffer, "order_ps_input(OUT);\n");
3501 } else {
3502 shader_addline(buffer, "order_ps_input();\n");
3503 }
3504
3505 /* If this shader doesn't use fog copy the z coord to the fog coord so that we can use table fog */
3506 if (!reg_maps->fog)
3507 shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
3508
3509 /* Write the final position.
3510 *
3511 * OpenGL coordinates specify the center of the pixel while d3d coords specify
3512 * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3513 * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3514 * contains 1.0 to allow a mad.
3515 */
3516 shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
3517 shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
3518
3519 /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3520 *
3521 * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
3522 * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
3523 * which is the same as z = z / 2 - w.
3524 */
3525 shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
3526
3527 shader_addline(buffer, "}\n");
3528
3529 TRACE("Compiling shader object %u\n", shader_obj);
3530 GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
3531 GL_EXTCALL(glCompileShaderARB(shader_obj));
3532 print_glsl_info_log(&GLINFO_LOCATION, shader_obj);
3533
3534 /* Store the shader object */
3535 This->baseShader.prgId = shader_obj;
3536 }
3537
3538 static void shader_glsl_get_caps(WINED3DDEVTYPE devtype, WineD3D_GL_Info *gl_info, struct shader_caps *pCaps) {
3539 /* We don't have a GLSL fixed function pipeline yet, so let the none backend set its caps,
3540 * then overwrite the shader specific ones
3541 */
3542 none_shader_backend.shader_get_caps(devtype, gl_info, pCaps);
3543
3544 /* Nvidia Geforce6/7 or Ati R4xx/R5xx cards with GLSL support, support VS 3.0 but older Nvidia/Ati
3545 * models with GLSL support only support 2.0. In case of nvidia we can detect VS 2.0 support using
3546 * vs_nv_version which is based on NV_vertex_program.
3547 * For Ati cards there's no way using glsl (it abstracts the lowlevel info away) and also not
3548 * using ARB_vertex_program. It is safe to assume that when a card supports pixel shader 2.0 it
3549 * supports vertex shader 2.0 too and the way around. We can detect ps2.0 using the maximum number
3550 * of native instructions, so use that here. For more info see the pixel shader versioning code below.
3551 */
3552 if((GLINFO_LOCATION.vs_nv_version == VS_VERSION_20) || (GLINFO_LOCATION.ps_arb_max_instructions <= 512))
3553 pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
3554 else
3555 pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
3556 TRACE_(d3d_caps)("Hardware vertex shader version %d.%d enabled (GLSL)\n", (pCaps->VertexShaderVersion >> 8) & 0xff, pCaps->VertexShaderVersion & 0xff);
3557 pCaps->MaxVertexShaderConst = GL_LIMITS(vshader_constantsF);
3558
3559 /* Older DX9-class videocards (GeforceFX / Radeon >9500/X*00) only support pixel shader 2.0/2.0a/2.0b.
3560 * In OpenGL the extensions related to GLSL abstract lowlevel GL info away which is needed
3561 * to distinguish between 2.0 and 3.0 (and 2.0a/2.0b). In case of Nvidia we use their fragment
3562 * program extensions. On other hardware including ATI GL_ARB_fragment_program offers the info
3563 * in max native instructions. Intel and others also offer the info in this extension but they
3564 * don't support GLSL (at least on Windows).
3565 *
3566 * PS2.0 requires at least 96 instructions, 2.0a/2.0b go up to 512. Assume that if the number
3567 * of instructions is 512 or less we have to do with ps2.0 hardware.
3568 * 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.
3569 */
3570 if((GLINFO_LOCATION.ps_nv_version == PS_VERSION_20) || (GLINFO_LOCATION.ps_arb_max_instructions <= 512))
3571 pCaps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
3572 else
3573 pCaps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
3574
3575 /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
3576 * Direct3D minimum requirement.
3577 *
3578 * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
3579 * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
3580 *
3581 * The problem is that the refrast clamps temporary results in the shader to
3582 * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
3583 * then applications may miss the clamping behavior. On the other hand, if it is smaller,
3584 * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
3585 * offer a way to query this.
3586 */
3587 pCaps->PixelShader1xMaxValue = 8.0;
3588 TRACE_(d3d_caps)("Hardware pixel shader version %d.%d enabled (GLSL)\n", (pCaps->PixelShaderVersion >> 8) & 0xff, pCaps->PixelShaderVersion & 0xff);
3589 }
3590
3591 static void shader_glsl_load_init(void) {}
3592
3593 static void shader_glsl_fragment_enable(IWineD3DDevice *iface, BOOL enable) {
3594 none_shader_backend.shader_fragment_enable(iface, enable);
3595 }
3596
3597 const shader_backend_t glsl_shader_backend = {
3598 &shader_glsl_select,
3599 &shader_glsl_select_depth_blt,
3600 &shader_glsl_destroy_depth_blt,
3601 &shader_glsl_load_constants,
3602 &shader_glsl_cleanup,
3603 &shader_glsl_color_correction,
3604 &shader_glsl_destroy,
3605 &shader_glsl_alloc,
3606 &shader_glsl_free,
3607 &shader_glsl_dirty_const,
3608 &shader_glsl_generate_pshader,
3609 &shader_glsl_generate_vshader,
3610 &shader_glsl_get_caps,
3611 &shader_glsl_load_init,
3612 &shader_glsl_fragment_enable,
3613 FFPStateTable
3614 };
Windows API implementation