wined3d: Create clip coords only if they are used.
[wine] / dlls / wined3d / glsl_shader.c
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  * Copyright 2009 Henri Verbeet for CodeWeavers
8  *
9  * This library is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU Lesser General Public
11  * License as published by the Free Software Foundation; either
12  * version 2.1 of the License, or (at your option) any later version.
13  *
14  * This library is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17  * Lesser General Public License for more details.
18  *
19  * You should have received a copy of the GNU Lesser General Public
20  * License along with this library; if not, write to the Free Software
21  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
22  */
23
24 /*
25  * D3D shader asm has swizzles on source parameters, and write masks for
26  * destination parameters. GLSL uses swizzles for both. The result of this is
27  * that for example "mov dst.xw, src.zyxw" becomes "dst.xw = src.zw" in GLSL.
28  * Ie, to generate a proper GLSL source swizzle, we need to take the D3D write
29  * mask for the destination parameter into account.
30  */
31
32 #include "config.h"
33 #include <limits.h>
34 #include <stdio.h>
35 #include "wined3d_private.h"
36
37 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
38 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
39 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
40 WINE_DECLARE_DEBUG_CHANNEL(d3d);
41
42 #define GLINFO_LOCATION      (*gl_info)
43
44 #define WINED3D_GLSL_SAMPLE_PROJECTED   0x1
45 #define WINED3D_GLSL_SAMPLE_RECT        0x2
46 #define WINED3D_GLSL_SAMPLE_LOD         0x4
47 #define WINED3D_GLSL_SAMPLE_GRAD        0x8
48
49 typedef struct {
50     char reg_name[150];
51     char mask_str[6];
52 } glsl_dst_param_t;
53
54 typedef struct {
55     char reg_name[150];
56     char param_str[200];
57 } glsl_src_param_t;
58
59 typedef struct {
60     const char *name;
61     DWORD coord_mask;
62 } glsl_sample_function_t;
63
64 enum heap_node_op
65 {
66     HEAP_NODE_TRAVERSE_LEFT,
67     HEAP_NODE_TRAVERSE_RIGHT,
68     HEAP_NODE_POP,
69 };
70
71 struct constant_entry
72 {
73     unsigned int idx;
74     unsigned int version;
75 };
76
77 struct constant_heap
78 {
79     struct constant_entry *entries;
80     unsigned int *positions;
81     unsigned int size;
82 };
83
84 /* GLSL shader private data */
85 struct shader_glsl_priv {
86     struct wined3d_shader_buffer shader_buffer;
87     struct wine_rb_tree program_lookup;
88     struct glsl_shader_prog_link *glsl_program;
89     struct constant_heap vconst_heap;
90     struct constant_heap pconst_heap;
91     unsigned char *stack;
92     GLhandleARB depth_blt_program[tex_type_count];
93     UINT next_constant_version;
94 };
95
96 /* Struct to maintain data about a linked GLSL program */
97 struct glsl_shader_prog_link {
98     struct wine_rb_entry        program_lookup_entry;
99     struct list                 vshader_entry;
100     struct list                 pshader_entry;
101     GLhandleARB                 programId;
102     GLint                       *vuniformF_locations;
103     GLint                       *puniformF_locations;
104     GLint                       vuniformI_locations[MAX_CONST_I];
105     GLint                       puniformI_locations[MAX_CONST_I];
106     GLint                       posFixup_location;
107     GLint                       np2Fixup_location;
108     GLint                       bumpenvmat_location[MAX_TEXTURES];
109     GLint                       luminancescale_location[MAX_TEXTURES];
110     GLint                       luminanceoffset_location[MAX_TEXTURES];
111     GLint                       ycorrection_location;
112     GLenum                      vertex_color_clamp;
113     IWineD3DVertexShader        *vshader;
114     IWineD3DPixelShader         *pshader;
115     struct vs_compile_args      vs_args;
116     struct ps_compile_args      ps_args;
117     UINT                        constant_version;
118     const struct ps_np2fixup_info *np2Fixup_info;
119 };
120
121 typedef struct {
122     IWineD3DVertexShader        *vshader;
123     IWineD3DPixelShader         *pshader;
124     struct ps_compile_args      ps_args;
125     struct vs_compile_args      vs_args;
126 } glsl_program_key_t;
127
128 struct shader_glsl_ctx_priv {
129     const struct vs_compile_args    *cur_vs_args;
130     const struct ps_compile_args    *cur_ps_args;
131     struct ps_np2fixup_info         *cur_np2fixup_info;
132 };
133
134 struct glsl_ps_compiled_shader
135 {
136     struct ps_compile_args          args;
137     struct ps_np2fixup_info         np2fixup;
138     GLhandleARB                     prgId;
139 };
140
141 struct glsl_pshader_private
142 {
143     struct glsl_ps_compiled_shader  *gl_shaders;
144     UINT                            num_gl_shaders, shader_array_size;
145 };
146
147 struct glsl_vs_compiled_shader
148 {
149     struct vs_compile_args          args;
150     GLhandleARB                     prgId;
151 };
152
153 struct glsl_vshader_private
154 {
155     struct glsl_vs_compiled_shader  *gl_shaders;
156     UINT                            num_gl_shaders, shader_array_size;
157 };
158
159 /* Extract a line from the info log.
160  * Note that this modifies the source string. */
161 static char *get_info_log_line(char **ptr)
162 {
163     char *p, *q;
164
165     p = *ptr;
166     if (!(q = strstr(p, "\n")))
167     {
168         if (!*p) return NULL;
169         *ptr += strlen(p);
170         return p;
171     }
172     *q = '\0';
173     *ptr = q + 1;
174
175     return p;
176 }
177
178 /** Prints the GLSL info log which will contain error messages if they exist */
179 /* GL locking is done by the caller */
180 static void print_glsl_info_log(const struct wined3d_gl_info *gl_info, GLhandleARB obj)
181 {
182     int infologLength = 0;
183     char *infoLog;
184     unsigned int i;
185     BOOL is_spam;
186
187     static const char * const spam[] =
188     {
189         "Vertex shader was successfully compiled to run on hardware.\n",    /* fglrx          */
190         "Fragment shader was successfully compiled to run on hardware.\n",  /* fglrx, with \n */
191         "Fragment shader was successfully compiled to run on hardware.",    /* fglrx, no \n   */
192         "Fragment shader(s) linked, vertex shader(s) linked. \n ",          /* fglrx, with \n */
193         "Fragment shader(s) linked, vertex shader(s) linked.",              /* fglrx, no \n   */
194         "Vertex shader(s) linked, no fragment shader(s) defined. \n ",      /* fglrx, with \n */
195         "Vertex shader(s) linked, no fragment shader(s) defined.",          /* fglrx, no \n   */
196         "Fragment shader(s) linked, no vertex shader(s) defined. \n ",      /* fglrx, with \n */
197         "Fragment shader(s) linked, no vertex shader(s) defined.",          /* fglrx, no \n   */
198     };
199
200     if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
201
202     GL_EXTCALL(glGetObjectParameterivARB(obj,
203                GL_OBJECT_INFO_LOG_LENGTH_ARB,
204                &infologLength));
205
206     /* A size of 1 is just a null-terminated string, so the log should be bigger than
207      * that if there are errors. */
208     if (infologLength > 1)
209     {
210         char *ptr, *line;
211
212         /* Fglrx doesn't terminate the string properly, but it tells us the proper length.
213          * So use HEAP_ZERO_MEMORY to avoid uninitialized bytes
214          */
215         infoLog = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, infologLength);
216         GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
217         is_spam = FALSE;
218
219         for(i = 0; i < sizeof(spam) / sizeof(spam[0]); i++) {
220             if(strcmp(infoLog, spam[i]) == 0) {
221                 is_spam = TRUE;
222                 break;
223             }
224         }
225
226         ptr = infoLog;
227         if (is_spam)
228         {
229             TRACE("Spam received from GLSL shader #%u:\n", obj);
230             while ((line = get_info_log_line(&ptr))) TRACE("    %s\n", line);
231         }
232         else
233         {
234             FIXME("Error received from GLSL shader #%u:\n", obj);
235             while ((line = get_info_log_line(&ptr))) FIXME("    %s\n", line);
236         }
237         HeapFree(GetProcessHeap(), 0, infoLog);
238     }
239 }
240
241 /**
242  * Loads (pixel shader) samplers
243  */
244 /* GL locking is done by the caller */
245 static void shader_glsl_load_psamplers(const struct wined3d_gl_info *gl_info,
246         DWORD *tex_unit_map, GLhandleARB programId)
247 {
248     GLint name_loc;
249     int i;
250     char sampler_name[20];
251
252     for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
253         snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
254         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
255         if (name_loc != -1) {
256             DWORD mapped_unit = tex_unit_map[i];
257             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.fragment_samplers)
258             {
259                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
260                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
261                 checkGLcall("glUniform1iARB");
262             } else {
263                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
264             }
265         }
266     }
267 }
268
269 /* GL locking is done by the caller */
270 static void shader_glsl_load_vsamplers(const struct wined3d_gl_info *gl_info,
271         DWORD *tex_unit_map, GLhandleARB programId)
272 {
273     GLint name_loc;
274     char sampler_name[20];
275     int i;
276
277     for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
278         snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
279         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
280         if (name_loc != -1) {
281             DWORD mapped_unit = tex_unit_map[MAX_FRAGMENT_SAMPLERS + i];
282             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.combined_samplers)
283             {
284                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
285                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
286                 checkGLcall("glUniform1iARB");
287             } else {
288                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
289             }
290         }
291     }
292 }
293
294 /* GL locking is done by the caller */
295 static inline void walk_constant_heap(const struct wined3d_gl_info *gl_info, const float *constants,
296         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
297 {
298     int stack_idx = 0;
299     unsigned int heap_idx = 1;
300     unsigned int idx;
301
302     if (heap->entries[heap_idx].version <= version) return;
303
304     idx = heap->entries[heap_idx].idx;
305     if (constant_locations[idx] != -1) GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
306     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
307
308     while (stack_idx >= 0)
309     {
310         /* Note that we fall through to the next case statement. */
311         switch(stack[stack_idx])
312         {
313             case HEAP_NODE_TRAVERSE_LEFT:
314             {
315                 unsigned int left_idx = heap_idx << 1;
316                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
317                 {
318                     heap_idx = left_idx;
319                     idx = heap->entries[heap_idx].idx;
320                     if (constant_locations[idx] != -1)
321                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
322
323                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
324                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
325                     break;
326                 }
327             }
328
329             case HEAP_NODE_TRAVERSE_RIGHT:
330             {
331                 unsigned int right_idx = (heap_idx << 1) + 1;
332                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
333                 {
334                     heap_idx = right_idx;
335                     idx = heap->entries[heap_idx].idx;
336                     if (constant_locations[idx] != -1)
337                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
338
339                     stack[stack_idx++] = HEAP_NODE_POP;
340                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
341                     break;
342                 }
343             }
344
345             case HEAP_NODE_POP:
346             {
347                 heap_idx >>= 1;
348                 --stack_idx;
349                 break;
350             }
351         }
352     }
353     checkGLcall("walk_constant_heap()");
354 }
355
356 /* GL locking is done by the caller */
357 static inline void apply_clamped_constant(const struct wined3d_gl_info *gl_info, GLint location, const GLfloat *data)
358 {
359     GLfloat clamped_constant[4];
360
361     if (location == -1) return;
362
363     clamped_constant[0] = data[0] < -1.0f ? -1.0f : data[0] > 1.0f ? 1.0f : data[0];
364     clamped_constant[1] = data[1] < -1.0f ? -1.0f : data[1] > 1.0f ? 1.0f : data[1];
365     clamped_constant[2] = data[2] < -1.0f ? -1.0f : data[2] > 1.0f ? 1.0f : data[2];
366     clamped_constant[3] = data[3] < -1.0f ? -1.0f : data[3] > 1.0f ? 1.0f : data[3];
367
368     GL_EXTCALL(glUniform4fvARB(location, 1, clamped_constant));
369 }
370
371 /* GL locking is done by the caller */
372 static inline void walk_constant_heap_clamped(const struct wined3d_gl_info *gl_info, const float *constants,
373         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
374 {
375     int stack_idx = 0;
376     unsigned int heap_idx = 1;
377     unsigned int idx;
378
379     if (heap->entries[heap_idx].version <= version) return;
380
381     idx = heap->entries[heap_idx].idx;
382     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
383     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
384
385     while (stack_idx >= 0)
386     {
387         /* Note that we fall through to the next case statement. */
388         switch(stack[stack_idx])
389         {
390             case HEAP_NODE_TRAVERSE_LEFT:
391             {
392                 unsigned int left_idx = heap_idx << 1;
393                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
394                 {
395                     heap_idx = left_idx;
396                     idx = heap->entries[heap_idx].idx;
397                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
398
399                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
400                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
401                     break;
402                 }
403             }
404
405             case HEAP_NODE_TRAVERSE_RIGHT:
406             {
407                 unsigned int right_idx = (heap_idx << 1) + 1;
408                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
409                 {
410                     heap_idx = right_idx;
411                     idx = heap->entries[heap_idx].idx;
412                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
413
414                     stack[stack_idx++] = HEAP_NODE_POP;
415                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
416                     break;
417                 }
418             }
419
420             case HEAP_NODE_POP:
421             {
422                 heap_idx >>= 1;
423                 --stack_idx;
424                 break;
425             }
426         }
427     }
428     checkGLcall("walk_constant_heap_clamped()");
429 }
430
431 /* Loads floating point constants (aka uniforms) into the currently set GLSL program. */
432 /* GL locking is done by the caller */
433 static void shader_glsl_load_constantsF(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
434         const float *constants, const GLint *constant_locations, const struct constant_heap *heap,
435         unsigned char *stack, UINT version)
436 {
437     const local_constant *lconst;
438
439     /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
440     if (This->baseShader.reg_maps.shader_version.major == 1
441             && shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type))
442         walk_constant_heap_clamped(gl_info, constants, constant_locations, heap, stack, version);
443     else
444         walk_constant_heap(gl_info, constants, constant_locations, heap, stack, version);
445
446     if (!This->baseShader.load_local_constsF)
447     {
448         TRACE("No need to load local float constants for this shader\n");
449         return;
450     }
451
452     /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
453     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry)
454     {
455         GLint location = constant_locations[lconst->idx];
456         /* We found this uniform name in the program - go ahead and send the data */
457         if (location != -1) GL_EXTCALL(glUniform4fvARB(location, 1, (const GLfloat *)lconst->value));
458     }
459     checkGLcall("glUniform4fvARB()");
460 }
461
462 /* Loads integer constants (aka uniforms) into the currently set GLSL program. */
463 /* GL locking is done by the caller */
464 static void shader_glsl_load_constantsI(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
465         const GLint locations[MAX_CONST_I], const int *constants, WORD constants_set)
466 {
467     unsigned int i;
468     struct list* ptr;
469
470     for (i = 0; constants_set; constants_set >>= 1, ++i)
471     {
472         if (!(constants_set & 1)) continue;
473
474         TRACE_(d3d_constants)("Loading constants %u: %i, %i, %i, %i\n",
475                 i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
476
477         /* We found this uniform name in the program - go ahead and send the data */
478         GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
479         checkGLcall("glUniform4ivARB");
480     }
481
482     /* Load immediate constants */
483     ptr = list_head(&This->baseShader.constantsI);
484     while (ptr) {
485         const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
486         unsigned int idx = lconst->idx;
487         const GLint *values = (const GLint *)lconst->value;
488
489         TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
490             values[0], values[1], values[2], values[3]);
491
492         /* We found this uniform name in the program - go ahead and send the data */
493         GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
494         checkGLcall("glUniform4ivARB");
495         ptr = list_next(&This->baseShader.constantsI, ptr);
496     }
497 }
498
499 /* Loads boolean constants (aka uniforms) into the currently set GLSL program. */
500 /* GL locking is done by the caller */
501 static void shader_glsl_load_constantsB(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
502         GLhandleARB programId, const BOOL *constants, WORD constants_set)
503 {
504     GLint tmp_loc;
505     unsigned int i;
506     char tmp_name[8];
507     const char *prefix;
508     struct list* ptr;
509
510     switch (This->baseShader.reg_maps.shader_version.type)
511     {
512         case WINED3D_SHADER_TYPE_VERTEX:
513             prefix = "VB";
514             break;
515
516         case WINED3D_SHADER_TYPE_GEOMETRY:
517             prefix = "GB";
518             break;
519
520         case WINED3D_SHADER_TYPE_PIXEL:
521             prefix = "PB";
522             break;
523
524         default:
525             FIXME("Unknown shader type %#x.\n",
526                     This->baseShader.reg_maps.shader_version.type);
527             prefix = "UB";
528             break;
529     }
530
531     /* TODO: Benchmark and see if it would be beneficial to store the
532      * locations of the constants to avoid looking up each time */
533     for (i = 0; constants_set; constants_set >>= 1, ++i)
534     {
535         if (!(constants_set & 1)) continue;
536
537         TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
538
539         /* TODO: Benchmark and see if it would be beneficial to store the
540          * locations of the constants to avoid looking up each time */
541         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
542         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
543         if (tmp_loc != -1)
544         {
545             /* We found this uniform name in the program - go ahead and send the data */
546             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
547             checkGLcall("glUniform1ivARB");
548         }
549     }
550
551     /* Load immediate constants */
552     ptr = list_head(&This->baseShader.constantsB);
553     while (ptr) {
554         const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
555         unsigned int idx = lconst->idx;
556         const GLint *values = (const GLint *)lconst->value;
557
558         TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
559
560         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
561         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
562         if (tmp_loc != -1) {
563             /* We found this uniform name in the program - go ahead and send the data */
564             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
565             checkGLcall("glUniform1ivARB");
566         }
567         ptr = list_next(&This->baseShader.constantsB, ptr);
568     }
569 }
570
571 static void reset_program_constant_version(struct wine_rb_entry *entry, void *context)
572 {
573     WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry)->constant_version = 0;
574 }
575
576 /**
577  * Loads the texture dimensions for NP2 fixup into the currently set GLSL program.
578  */
579 /* GL locking is done by the caller (state handler) */
580 static void shader_glsl_load_np2fixup_constants(
581     IWineD3DDevice* device,
582     char usePixelShader,
583     char useVertexShader) {
584
585     const IWineD3DDeviceImpl* deviceImpl = (const IWineD3DDeviceImpl*) device;
586     const struct glsl_shader_prog_link* prog = ((struct shader_glsl_priv *)(deviceImpl->shader_priv))->glsl_program;
587
588     if (!prog) {
589         /* No GLSL program set - nothing to do. */
590         return;
591     }
592
593     if (!usePixelShader) {
594         /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
595         return;
596     }
597
598     if (prog->ps_args.np2_fixup && -1 != prog->np2Fixup_location) {
599         const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
600         const IWineD3DStateBlockImpl* stateBlock = (const IWineD3DStateBlockImpl*) deviceImpl->stateBlock;
601         UINT i;
602         UINT fixup = prog->ps_args.np2_fixup;
603         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
604
605         for (i = 0; fixup; fixup >>= 1, ++i) {
606             const unsigned char idx = prog->np2Fixup_info->idx[i];
607             const IWineD3DBaseTextureImpl* const tex = (const IWineD3DBaseTextureImpl*) stateBlock->textures[i];
608             GLfloat* tex_dim = &np2fixup_constants[(idx >> 1) * 4];
609
610             if (!tex) {
611                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
612                 continue;
613             }
614
615             if (idx % 2) {
616                 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
617             } else {
618                 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
619             }
620         }
621
622         GL_EXTCALL(glUniform4fvARB(prog->np2Fixup_location, prog->np2Fixup_info->num_consts, np2fixup_constants));
623     }
624 }
625
626 /**
627  * Loads the app-supplied constants into the currently set GLSL program.
628  */
629 /* GL locking is done by the caller (state handler) */
630 static void shader_glsl_load_constants(const struct wined3d_context *context,
631         char usePixelShader, char useVertexShader)
632 {
633     IWineD3DDeviceImpl *device = ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice;
634     const struct wined3d_gl_info *gl_info = context->gl_info;
635     IWineD3DStateBlockImpl* stateBlock = device->stateBlock;
636     struct shader_glsl_priv *priv = device->shader_priv;
637
638     GLhandleARB programId;
639     struct glsl_shader_prog_link *prog = priv->glsl_program;
640     UINT constant_version;
641     int i;
642
643     if (!prog) {
644         /* No GLSL program set - nothing to do. */
645         return;
646     }
647     programId = prog->programId;
648     constant_version = prog->constant_version;
649
650     if (useVertexShader) {
651         IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
652
653         /* Load DirectX 9 float constants/uniforms for vertex shader */
654         shader_glsl_load_constantsF(vshader, gl_info, stateBlock->vertexShaderConstantF,
655                 prog->vuniformF_locations, &priv->vconst_heap, priv->stack, constant_version);
656
657         /* Load DirectX 9 integer constants/uniforms for vertex shader */
658         shader_glsl_load_constantsI(vshader, gl_info, prog->vuniformI_locations, stateBlock->vertexShaderConstantI,
659                 stateBlock->changed.vertexShaderConstantsI & vshader->baseShader.reg_maps.integer_constants);
660
661         /* Load DirectX 9 boolean constants/uniforms for vertex shader */
662         shader_glsl_load_constantsB(vshader, gl_info, programId, stateBlock->vertexShaderConstantB,
663                 stateBlock->changed.vertexShaderConstantsB & vshader->baseShader.reg_maps.boolean_constants);
664
665         /* Upload the position fixup params */
666         GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, &device->posFixup[0]));
667         checkGLcall("glUniform4fvARB");
668     }
669
670     if (usePixelShader) {
671
672         IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
673
674         /* Load DirectX 9 float constants/uniforms for pixel shader */
675         shader_glsl_load_constantsF(pshader, gl_info, stateBlock->pixelShaderConstantF,
676                 prog->puniformF_locations, &priv->pconst_heap, priv->stack, constant_version);
677
678         /* Load DirectX 9 integer constants/uniforms for pixel shader */
679         shader_glsl_load_constantsI(pshader, gl_info, prog->puniformI_locations, stateBlock->pixelShaderConstantI,
680                 stateBlock->changed.pixelShaderConstantsI & pshader->baseShader.reg_maps.integer_constants);
681
682         /* Load DirectX 9 boolean constants/uniforms for pixel shader */
683         shader_glsl_load_constantsB(pshader, gl_info, programId, stateBlock->pixelShaderConstantB,
684                 stateBlock->changed.pixelShaderConstantsB & pshader->baseShader.reg_maps.boolean_constants);
685
686         /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
687          * It can't be 0 for a valid texbem instruction.
688          */
689         for(i = 0; i < MAX_TEXTURES; i++) {
690             const float *data;
691
692             if(prog->bumpenvmat_location[i] == -1) continue;
693
694             data = (const float *)&stateBlock->textureState[i][WINED3DTSS_BUMPENVMAT00];
695             GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
696             checkGLcall("glUniformMatrix2fvARB");
697
698             /* texbeml needs the luminance scale and offset too. If texbeml is used, needsbumpmat
699              * is set too, so we can check that in the needsbumpmat check
700              */
701             if(prog->luminancescale_location[i] != -1) {
702                 const GLfloat *scale = (const GLfloat *)&stateBlock->textureState[i][WINED3DTSS_BUMPENVLSCALE];
703                 const GLfloat *offset = (const GLfloat *)&stateBlock->textureState[i][WINED3DTSS_BUMPENVLOFFSET];
704
705                 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
706                 checkGLcall("glUniform1fvARB");
707                 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
708                 checkGLcall("glUniform1fvARB");
709             }
710         }
711
712         if(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
713             float correction_params[4];
714
715             if (context->render_offscreen)
716             {
717                 correction_params[0] = 0.0f;
718                 correction_params[1] = 1.0f;
719             } else {
720                 /* position is window relative, not viewport relative */
721                 correction_params[0] = ((IWineD3DSurfaceImpl *)context->current_rt)->currentDesc.Height;
722                 correction_params[1] = -1.0f;
723             }
724             GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
725         }
726     }
727
728     if (priv->next_constant_version == UINT_MAX)
729     {
730         TRACE("Max constant version reached, resetting to 0.\n");
731         wine_rb_for_each_entry(&priv->program_lookup, reset_program_constant_version, NULL);
732         priv->next_constant_version = 1;
733     }
734     else
735     {
736         prog->constant_version = priv->next_constant_version++;
737     }
738 }
739
740 static inline void update_heap_entry(struct constant_heap *heap, unsigned int idx,
741         unsigned int heap_idx, DWORD new_version)
742 {
743     struct constant_entry *entries = heap->entries;
744     unsigned int *positions = heap->positions;
745     unsigned int parent_idx;
746
747     while (heap_idx > 1)
748     {
749         parent_idx = heap_idx >> 1;
750
751         if (new_version <= entries[parent_idx].version) break;
752
753         entries[heap_idx] = entries[parent_idx];
754         positions[entries[parent_idx].idx] = heap_idx;
755         heap_idx = parent_idx;
756     }
757
758     entries[heap_idx].version = new_version;
759     entries[heap_idx].idx = idx;
760     positions[idx] = heap_idx;
761 }
762
763 static void shader_glsl_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
764 {
765     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
766     struct shader_glsl_priv *priv = This->shader_priv;
767     struct constant_heap *heap = &priv->vconst_heap;
768     UINT i;
769
770     for (i = start; i < count + start; ++i)
771     {
772         if (!This->stateBlock->changed.vertexShaderConstantsF[i])
773             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
774         else
775             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
776     }
777 }
778
779 static void shader_glsl_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
780 {
781     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
782     struct shader_glsl_priv *priv = This->shader_priv;
783     struct constant_heap *heap = &priv->pconst_heap;
784     UINT i;
785
786     for (i = start; i < count + start; ++i)
787     {
788         if (!This->stateBlock->changed.pixelShaderConstantsF[i])
789             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
790         else
791             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
792     }
793 }
794
795 static unsigned int vec4_varyings(DWORD shader_major, const struct wined3d_gl_info *gl_info)
796 {
797     unsigned int ret = gl_info->limits.glsl_varyings / 4;
798     /* 4.0 shaders do not write clip coords because d3d10 does not support user clipplanes */
799     if(shader_major > 3) return ret;
800
801     /* 3.0 shaders may need an extra varying for the clip coord on some cards(mostly dx10 ones) */
802     if (gl_info->quirks & WINED3D_QUIRK_GLSL_CLIP_VARYING) ret -= 1;
803     return ret;
804 }
805
806 /** Generate the variable & register declarations for the GLSL output target */
807 static void shader_generate_glsl_declarations(const struct wined3d_context *context,
808         struct wined3d_shader_buffer *buffer, IWineD3DBaseShader *iface,
809         const shader_reg_maps *reg_maps, struct shader_glsl_ctx_priv *ctx_priv)
810 {
811     IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
812     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
813     const struct ps_compile_args *ps_args = ctx_priv->cur_ps_args;
814     const struct wined3d_gl_info *gl_info = context->gl_info;
815     unsigned int i, extra_constants_needed = 0;
816     const local_constant *lconst;
817     DWORD map;
818
819     /* There are some minor differences between pixel and vertex shaders */
820     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
821     char prefix = pshader ? 'P' : 'V';
822
823     /* Prototype the subroutines */
824     for (i = 0, map = reg_maps->labels; map; map >>= 1, ++i)
825     {
826         if (map & 1) shader_addline(buffer, "void subroutine%u();\n", i);
827     }
828
829     /* Declare the constants (aka uniforms) */
830     if (This->baseShader.limits.constant_float > 0) {
831         unsigned max_constantsF;
832         /* Unless the shader uses indirect addressing, always declare the maximum array size and ignore that we need some
833          * uniforms privately. E.g. if GL supports 256 uniforms, and we need 2 for the pos fixup and immediate values, still
834          * declare VC[256]. If the shader needs more uniforms than we have it won't work in any case. If it uses less, the
835          * compiler will figure out which uniforms are really used and strip them out. This allows a shader to use c255 on
836          * a dx9 card, as long as it doesn't also use all the other constants.
837          *
838          * If the shader uses indirect addressing the compiler must assume that all declared uniforms are used. In this case,
839          * declare only the amount that we're assured to have.
840          *
841          * Thus we run into problems in these two cases:
842          * 1) The shader really uses more uniforms than supported
843          * 2) The shader uses indirect addressing, less constants than supported, but uses a constant index > #supported consts
844          */
845         if (pshader)
846         {
847             /* No indirect addressing here. */
848             max_constantsF = gl_info->limits.glsl_ps_float_constants;
849         }
850         else
851         {
852             if(This->baseShader.reg_maps.usesrelconstF) {
853                 /* Subtract the other potential uniforms from the max available (bools, ints, and 1 row of projection matrix).
854                  * Subtract another uniform for immediate values, which have to be loaded via uniform by the driver as well.
855                  * The shader code only uses 0.5, 2.0, 1.0, 128 and -128 in vertex shader code, so one vec4 should be enough
856                  * (Unfortunately the Nvidia driver doesn't store 128 and -128 in one float).
857                  *
858                  * Writing gl_ClipVertex requires one uniform for each clipplane as well.
859                  */
860                 max_constantsF = gl_info->limits.glsl_vs_float_constants - 3;
861                 if(ctx_priv->cur_vs_args->clip_enabled)
862                 {
863                     max_constantsF -= gl_info->limits.clipplanes;
864                 }
865                 max_constantsF -= count_bits(This->baseShader.reg_maps.integer_constants);
866                 /* Strictly speaking a bool only uses one scalar, but the nvidia(Linux) compiler doesn't pack them properly,
867                  * so each scalar requires a full vec4. We could work around this by packing the booleans ourselves, but
868                  * for now take this into account when calculating the number of available constants
869                  */
870                 max_constantsF -= count_bits(This->baseShader.reg_maps.boolean_constants);
871                 /* Set by driver quirks in directx.c */
872                 max_constantsF -= gl_info->reserved_glsl_constants;
873             }
874             else
875             {
876                 max_constantsF = gl_info->limits.glsl_vs_float_constants;
877             }
878         }
879         max_constantsF = min(This->baseShader.limits.constant_float, max_constantsF);
880         shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
881     }
882
883     /* Always declare the full set of constants, the compiler can remove the unused ones because d3d doesn't(yet)
884      * support indirect int and bool constant addressing. This avoids problems if the app uses e.g. i0 and i9.
885      */
886     if (This->baseShader.limits.constant_int > 0 && This->baseShader.reg_maps.integer_constants)
887         shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, This->baseShader.limits.constant_int);
888
889     if (This->baseShader.limits.constant_bool > 0 && This->baseShader.reg_maps.boolean_constants)
890         shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, This->baseShader.limits.constant_bool);
891
892     if(!pshader) {
893         shader_addline(buffer, "uniform vec4 posFixup;\n");
894         /* Predeclaration; This function is added at link time based on the pixel shader.
895          * VS 3.0 shaders have an array OUT[] the shader writes to, earlier versions don't have
896          * that. We know the input to the reorder function at vertex shader compile time, so
897          * we can deal with that. The reorder function for a 1.x and 2.x vertex shader can just
898          * read gl_FrontColor. The output depends on the pixel shader. The reorder function for a
899          * 1.x and 2.x pshader or for fixed function will write gl_FrontColor, and for a 3.0 shader
900          * it will write to the varying array. Here we depend on the shader optimizer on sorting that
901          * out. The nvidia driver only does that if the parameter is inout instead of out, hence the
902          * inout.
903          */
904         if (reg_maps->shader_version.major >= 3)
905         {
906             shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
907         } else {
908             shader_addline(buffer, "void order_ps_input();\n");
909         }
910     } else {
911         for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
912         {
913             if (!(map & 1)) continue;
914
915             shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
916
917             if (reg_maps->luminanceparams & (1 << i))
918             {
919                 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
920                 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
921                 extra_constants_needed++;
922             }
923
924             extra_constants_needed++;
925         }
926
927         if (ps_args->srgb_correction)
928         {
929             shader_addline(buffer, "const vec4 srgb_const0 = vec4(%.8e, %.8e, %.8e, %.8e);\n",
930                     srgb_pow, srgb_mul_high, srgb_sub_high, srgb_mul_low);
931             shader_addline(buffer, "const vec4 srgb_const1 = vec4(%.8e, 0.0, 0.0, 0.0);\n",
932                     srgb_cmp);
933         }
934         if (reg_maps->vpos || reg_maps->usesdsy)
935         {
936             if (This->baseShader.limits.constant_float + extra_constants_needed
937                     + 1 < gl_info->limits.glsl_ps_float_constants)
938             {
939                 shader_addline(buffer, "uniform vec4 ycorrection;\n");
940                 ((IWineD3DPixelShaderImpl *) This)->vpos_uniform = 1;
941                 extra_constants_needed++;
942             } else {
943                 /* This happens because we do not have proper tracking of the constant registers that are
944                  * actually used, only the max limit of the shader version
945                  */
946                 FIXME("Cannot find a free uniform for vpos correction params\n");
947                 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
948                         context->render_offscreen ? 0.0f : ((IWineD3DSurfaceImpl *)device->render_targets[0])->currentDesc.Height,
949                         context->render_offscreen ? 1.0f : -1.0f);
950             }
951             shader_addline(buffer, "vec4 vpos;\n");
952         }
953     }
954
955     /* Declare texture samplers */
956     for (i = 0; i < This->baseShader.limits.sampler; i++) {
957         if (reg_maps->sampler_type[i])
958         {
959             switch (reg_maps->sampler_type[i])
960             {
961                 case WINED3DSTT_1D:
962                     shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
963                     break;
964                 case WINED3DSTT_2D:
965                     if(device->stateBlock->textures[i] &&
966                        IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[i]) == GL_TEXTURE_RECTANGLE_ARB) {
967                         shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
968                     } else {
969                         shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
970                     }
971                     break;
972                 case WINED3DSTT_CUBE:
973                     shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
974                     break;
975                 case WINED3DSTT_VOLUME:
976                     shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
977                     break;
978                 default:
979                     shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
980                     FIXME("Unrecognized sampler type: %#x\n", reg_maps->sampler_type[i]);
981                     break;
982             }
983         }
984     }
985
986     /* Declare uniforms for NP2 texcoord fixup:
987      * This is NOT done inside the loop that declares the texture samplers since the NP2 fixup code
988      * is currently only used for the GeforceFX series and when forcing the ARB_npot extension off.
989      * Modern cards just skip the code anyway, so put it inside a separate loop. */
990     if (pshader && ps_args->np2_fixup) {
991
992         struct ps_np2fixup_info* const fixup = ctx_priv->cur_np2fixup_info;
993         UINT cur = 0;
994
995         /* NP2/RECT textures in OpenGL use texcoords in the range [0,width]x[0,height]
996          * while D3D has them in the (normalized) [0,1]x[0,1] range.
997          * samplerNP2Fixup stores texture dimensions and is updated through
998          * shader_glsl_load_np2fixup_constants when the sampler changes. */
999
1000         for (i = 0; i < This->baseShader.limits.sampler; ++i) {
1001             if (reg_maps->sampler_type[i]) {
1002                 if (!(ps_args->np2_fixup & (1 << i))) continue;
1003
1004                 if (WINED3DSTT_2D != reg_maps->sampler_type[i]) {
1005                     FIXME("Non-2D texture is flagged for NP2 texcoord fixup.\n");
1006                     continue;
1007                 }
1008
1009                 fixup->idx[i] = cur++;
1010             }
1011         }
1012
1013         fixup->num_consts = (cur + 1) >> 1;
1014         shader_addline(buffer, "uniform vec4 %csamplerNP2Fixup[%u];\n", prefix, fixup->num_consts);
1015     }
1016
1017     /* Declare address variables */
1018     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
1019     {
1020         if (map & 1) shader_addline(buffer, "ivec4 A%u;\n", i);
1021     }
1022
1023     /* Declare texture coordinate temporaries and initialize them */
1024     for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
1025     {
1026         if (map & 1) shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
1027     }
1028
1029     /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
1030      * helper function shader that is linked in at link time
1031      */
1032     if (pshader && reg_maps->shader_version.major >= 3)
1033     {
1034         if (use_vs(device->stateBlock))
1035         {
1036             shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1037         } else {
1038             /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
1039              * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
1040              * pixel shader that reads the fixed function color into the packed input registers.
1041              */
1042             shader_addline(buffer, "vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1043         }
1044     }
1045
1046     /* Declare output register temporaries */
1047     if(This->baseShader.limits.packed_output) {
1048         shader_addline(buffer, "vec4 OUT[%u];\n", This->baseShader.limits.packed_output);
1049     }
1050
1051     /* Declare temporary variables */
1052     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
1053     {
1054         if (map & 1) shader_addline(buffer, "vec4 R%u;\n", i);
1055     }
1056
1057     /* Declare attributes */
1058     if (reg_maps->shader_version.type == WINED3D_SHADER_TYPE_VERTEX)
1059     {
1060         for (i = 0, map = reg_maps->input_registers; map; map >>= 1, ++i)
1061         {
1062             if (map & 1) shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
1063         }
1064     }
1065
1066     /* Declare loop registers aLx */
1067     for (i = 0; i < reg_maps->loop_depth; i++) {
1068         shader_addline(buffer, "int aL%u;\n", i);
1069         shader_addline(buffer, "int tmpInt%u;\n", i);
1070     }
1071
1072     /* Temporary variables for matrix operations */
1073     shader_addline(buffer, "vec4 tmp0;\n");
1074     shader_addline(buffer, "vec4 tmp1;\n");
1075
1076     /* Local constants use a different name so they can be loaded once at shader link time
1077      * They can't be hardcoded into the shader text via LC = {x, y, z, w}; because the
1078      * float -> string conversion can cause precision loss.
1079      */
1080     if(!This->baseShader.load_local_constsF) {
1081         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
1082             shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
1083         }
1084     }
1085
1086     /* Start the main program */
1087     shader_addline(buffer, "void main() {\n");
1088     if(pshader && reg_maps->vpos) {
1089         /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
1090          * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
1091          * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
1092          * precision troubles when we just substract 0.5.
1093          *
1094          * To deal with that just floor() the position. This will eliminate the fraction on all cards.
1095          *
1096          * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
1097          *
1098          * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
1099          * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
1100          * coordinates specify the pixel centers instead of the pixel corners. This code will behave
1101          * correctly on drivers that returns integer values.
1102          */
1103         shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
1104     }
1105 }
1106
1107 /*****************************************************************************
1108  * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
1109  *
1110  * For more information, see http://wiki.winehq.org/DirectX-Shaders
1111  ****************************************************************************/
1112
1113 /* Prototypes */
1114 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1115         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src);
1116
1117 /** Used for opcode modifiers - They multiply the result by the specified amount */
1118 static const char * const shift_glsl_tab[] = {
1119     "",           /*  0 (none) */
1120     "2.0 * ",     /*  1 (x2)   */
1121     "4.0 * ",     /*  2 (x4)   */
1122     "8.0 * ",     /*  3 (x8)   */
1123     "16.0 * ",    /*  4 (x16)  */
1124     "32.0 * ",    /*  5 (x32)  */
1125     "",           /*  6 (x64)  */
1126     "",           /*  7 (x128) */
1127     "",           /*  8 (d256) */
1128     "",           /*  9 (d128) */
1129     "",           /* 10 (d64)  */
1130     "",           /* 11 (d32)  */
1131     "0.0625 * ",  /* 12 (d16)  */
1132     "0.125 * ",   /* 13 (d8)   */
1133     "0.25 * ",    /* 14 (d4)   */
1134     "0.5 * "      /* 15 (d2)   */
1135 };
1136
1137 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
1138 static void shader_glsl_gen_modifier(DWORD src_modifier, const char *in_reg, const char *in_regswizzle, char *out_str)
1139 {
1140     out_str[0] = 0;
1141
1142     switch (src_modifier)
1143     {
1144     case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
1145     case WINED3DSPSM_DW:
1146     case WINED3DSPSM_NONE:
1147         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1148         break;
1149     case WINED3DSPSM_NEG:
1150         sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
1151         break;
1152     case WINED3DSPSM_NOT:
1153         sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
1154         break;
1155     case WINED3DSPSM_BIAS:
1156         sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1157         break;
1158     case WINED3DSPSM_BIASNEG:
1159         sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1160         break;
1161     case WINED3DSPSM_SIGN:
1162         sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1163         break;
1164     case WINED3DSPSM_SIGNNEG:
1165         sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1166         break;
1167     case WINED3DSPSM_COMP:
1168         sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
1169         break;
1170     case WINED3DSPSM_X2:
1171         sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
1172         break;
1173     case WINED3DSPSM_X2NEG:
1174         sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
1175         break;
1176     case WINED3DSPSM_ABS:
1177         sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
1178         break;
1179     case WINED3DSPSM_ABSNEG:
1180         sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
1181         break;
1182     default:
1183         FIXME("Unhandled modifier %u\n", src_modifier);
1184         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1185     }
1186 }
1187
1188 /** Writes the GLSL variable name that corresponds to the register that the
1189  * DX opcode parameter is trying to access */
1190 static void shader_glsl_get_register_name(const struct wined3d_shader_register *reg,
1191         char *register_name, BOOL *is_color, const struct wined3d_shader_instruction *ins)
1192 {
1193     /* oPos, oFog and oPts in D3D */
1194     static const char * const hwrastout_reg_names[] = { "gl_Position", "gl_FogFragCoord", "gl_PointSize" };
1195
1196     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1197     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1198     const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
1199     char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
1200
1201     *is_color = FALSE;
1202
1203     switch (reg->type)
1204     {
1205         case WINED3DSPR_TEMP:
1206             sprintf(register_name, "R%u", reg->idx);
1207             break;
1208
1209         case WINED3DSPR_INPUT:
1210             /* vertex shaders */
1211             if (!pshader)
1212             {
1213                 struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1214                 if (priv->cur_vs_args->swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1215                 sprintf(register_name, "attrib%u", reg->idx);
1216                 break;
1217             }
1218
1219             /* pixel shaders >= 3.0 */
1220             if (This->baseShader.reg_maps.shader_version.major >= 3)
1221             {
1222                 DWORD idx = ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg->idx];
1223                 unsigned int in_count = vec4_varyings(This->baseShader.reg_maps.shader_version.major, gl_info);
1224
1225                 if (reg->rel_addr)
1226                 {
1227                     glsl_src_param_t rel_param;
1228
1229                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1230
1231                     /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
1232                      * operation there */
1233                     if (idx)
1234                     {
1235                         if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count)
1236                         {
1237                             sprintf(register_name,
1238                                     "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
1239                                     rel_param.param_str, idx, in_count - 1, rel_param.param_str, idx, in_count,
1240                                     rel_param.param_str, idx);
1241                         }
1242                         else
1243                         {
1244                             sprintf(register_name, "IN[%s + %u]", rel_param.param_str, idx);
1245                         }
1246                     }
1247                     else
1248                     {
1249                         if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count)
1250                         {
1251                             sprintf(register_name, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
1252                                     rel_param.param_str, in_count - 1, rel_param.param_str, in_count,
1253                                     rel_param.param_str);
1254                         }
1255                         else
1256                         {
1257                             sprintf(register_name, "IN[%s]", rel_param.param_str);
1258                         }
1259                     }
1260                 }
1261                 else
1262                 {
1263                     if (idx == in_count) sprintf(register_name, "gl_Color");
1264                     else if (idx == in_count + 1) sprintf(register_name, "gl_SecondaryColor");
1265                     else sprintf(register_name, "IN[%u]", idx);
1266                 }
1267             }
1268             else
1269             {
1270                 if (reg->idx == 0) strcpy(register_name, "gl_Color");
1271                 else strcpy(register_name, "gl_SecondaryColor");
1272                 break;
1273             }
1274             break;
1275
1276         case WINED3DSPR_CONST:
1277             {
1278                 const char prefix = pshader ? 'P' : 'V';
1279
1280                 /* Relative addressing */
1281                 if (reg->rel_addr)
1282                 {
1283                     glsl_src_param_t rel_param;
1284                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1285                     if (reg->idx) sprintf(register_name, "%cC[%s + %u]", prefix, rel_param.param_str, reg->idx);
1286                     else sprintf(register_name, "%cC[%s]", prefix, rel_param.param_str);
1287                 }
1288                 else
1289                 {
1290                     if (shader_constant_is_local(This, reg->idx))
1291                         sprintf(register_name, "%cLC%u", prefix, reg->idx);
1292                     else
1293                         sprintf(register_name, "%cC[%u]", prefix, reg->idx);
1294                 }
1295             }
1296             break;
1297
1298         case WINED3DSPR_CONSTINT:
1299             if (pshader) sprintf(register_name, "PI[%u]", reg->idx);
1300             else sprintf(register_name, "VI[%u]", reg->idx);
1301             break;
1302
1303         case WINED3DSPR_CONSTBOOL:
1304             if (pshader) sprintf(register_name, "PB[%u]", reg->idx);
1305             else sprintf(register_name, "VB[%u]", reg->idx);
1306             break;
1307
1308         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1309             if (pshader) sprintf(register_name, "T%u", reg->idx);
1310             else sprintf(register_name, "A%u", reg->idx);
1311             break;
1312
1313         case WINED3DSPR_LOOP:
1314             sprintf(register_name, "aL%u", This->baseShader.cur_loop_regno - 1);
1315             break;
1316
1317         case WINED3DSPR_SAMPLER:
1318             if (pshader) sprintf(register_name, "Psampler%u", reg->idx);
1319             else sprintf(register_name, "Vsampler%u", reg->idx);
1320             break;
1321
1322         case WINED3DSPR_COLOROUT:
1323             if (reg->idx >= gl_info->limits.buffers)
1324                 WARN("Write to render target %u, only %d supported.\n", reg->idx, gl_info->limits.buffers);
1325
1326             sprintf(register_name, "gl_FragData[%u]", reg->idx);
1327             break;
1328
1329         case WINED3DSPR_RASTOUT:
1330             sprintf(register_name, "%s", hwrastout_reg_names[reg->idx]);
1331             break;
1332
1333         case WINED3DSPR_DEPTHOUT:
1334             sprintf(register_name, "gl_FragDepth");
1335             break;
1336
1337         case WINED3DSPR_ATTROUT:
1338             if (reg->idx == 0) sprintf(register_name, "gl_FrontColor");
1339             else sprintf(register_name, "gl_FrontSecondaryColor");
1340             break;
1341
1342         case WINED3DSPR_TEXCRDOUT:
1343             /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
1344             if (This->baseShader.reg_maps.shader_version.major >= 3) sprintf(register_name, "OUT[%u]", reg->idx);
1345             else sprintf(register_name, "gl_TexCoord[%u]", reg->idx);
1346             break;
1347
1348         case WINED3DSPR_MISCTYPE:
1349             if (reg->idx == 0)
1350             {
1351                 /* vPos */
1352                 sprintf(register_name, "vpos");
1353             }
1354             else if (reg->idx == 1)
1355             {
1356                 /* Note that gl_FrontFacing is a bool, while vFace is
1357                  * a float for which the sign determines front/back */
1358                 sprintf(register_name, "(gl_FrontFacing ? 1.0 : -1.0)");
1359             }
1360             else
1361             {
1362                 FIXME("Unhandled misctype register %d\n", reg->idx);
1363                 sprintf(register_name, "unrecognized_register");
1364             }
1365             break;
1366
1367         case WINED3DSPR_IMMCONST:
1368             switch (reg->immconst_type)
1369             {
1370                 case WINED3D_IMMCONST_FLOAT:
1371                     sprintf(register_name, "%.8e", *(const float *)reg->immconst_data);
1372                     break;
1373
1374                 case WINED3D_IMMCONST_FLOAT4:
1375                     sprintf(register_name, "vec4(%.8e, %.8e, %.8e, %.8e)",
1376                             *(const float *)&reg->immconst_data[0], *(const float *)&reg->immconst_data[1],
1377                             *(const float *)&reg->immconst_data[2], *(const float *)&reg->immconst_data[3]);
1378                     break;
1379
1380                 default:
1381                     FIXME("Unhandled immconst type %#x\n", reg->immconst_type);
1382                     sprintf(register_name, "<unhandled_immconst_type %#x>", reg->immconst_type);
1383             }
1384             break;
1385
1386         default:
1387             FIXME("Unhandled register name Type(%d)\n", reg->type);
1388             sprintf(register_name, "unrecognized_register");
1389             break;
1390     }
1391 }
1392
1393 static void shader_glsl_write_mask_to_str(DWORD write_mask, char *str)
1394 {
1395     *str++ = '.';
1396     if (write_mask & WINED3DSP_WRITEMASK_0) *str++ = 'x';
1397     if (write_mask & WINED3DSP_WRITEMASK_1) *str++ = 'y';
1398     if (write_mask & WINED3DSP_WRITEMASK_2) *str++ = 'z';
1399     if (write_mask & WINED3DSP_WRITEMASK_3) *str++ = 'w';
1400     *str = '\0';
1401 }
1402
1403 /* Get the GLSL write mask for the destination register */
1404 static DWORD shader_glsl_get_write_mask(const struct wined3d_shader_dst_param *param, char *write_mask)
1405 {
1406     DWORD mask = param->write_mask;
1407
1408     if (shader_is_scalar(&param->reg))
1409     {
1410         mask = WINED3DSP_WRITEMASK_0;
1411         *write_mask = '\0';
1412     }
1413     else
1414     {
1415         shader_glsl_write_mask_to_str(mask, write_mask);
1416     }
1417
1418     return mask;
1419 }
1420
1421 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1422     unsigned int size = 0;
1423
1424     if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1425     if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1426     if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1427     if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1428
1429     return size;
1430 }
1431
1432 static void shader_glsl_swizzle_to_str(const DWORD swizzle, BOOL fixup, DWORD mask, char *str)
1433 {
1434     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1435      * but addressed as "rgba". To fix this we need to swap the register's x
1436      * and z components. */
1437     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1438
1439     *str++ = '.';
1440     /* swizzle bits fields: wwzzyyxx */
1441     if (mask & WINED3DSP_WRITEMASK_0) *str++ = swizzle_chars[swizzle & 0x03];
1442     if (mask & WINED3DSP_WRITEMASK_1) *str++ = swizzle_chars[(swizzle >> 2) & 0x03];
1443     if (mask & WINED3DSP_WRITEMASK_2) *str++ = swizzle_chars[(swizzle >> 4) & 0x03];
1444     if (mask & WINED3DSP_WRITEMASK_3) *str++ = swizzle_chars[(swizzle >> 6) & 0x03];
1445     *str = '\0';
1446 }
1447
1448 static void shader_glsl_get_swizzle(const struct wined3d_shader_src_param *param,
1449         BOOL fixup, DWORD mask, char *swizzle_str)
1450 {
1451     if (shader_is_scalar(&param->reg))
1452         *swizzle_str = '\0';
1453     else
1454         shader_glsl_swizzle_to_str(param->swizzle, fixup, mask, swizzle_str);
1455 }
1456
1457 /* From a given parameter token, generate the corresponding GLSL string.
1458  * Also, return the actual register name and swizzle in case the
1459  * caller needs this information as well. */
1460 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1461         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src)
1462 {
1463     BOOL is_color = FALSE;
1464     char swizzle_str[6];
1465
1466     glsl_src->reg_name[0] = '\0';
1467     glsl_src->param_str[0] = '\0';
1468     swizzle_str[0] = '\0';
1469
1470     shader_glsl_get_register_name(&wined3d_src->reg, glsl_src->reg_name, &is_color, ins);
1471     shader_glsl_get_swizzle(wined3d_src, is_color, mask, swizzle_str);
1472     shader_glsl_gen_modifier(wined3d_src->modifiers, glsl_src->reg_name, swizzle_str, glsl_src->param_str);
1473 }
1474
1475 /* From a given parameter token, generate the corresponding GLSL string.
1476  * Also, return the actual register name and swizzle in case the
1477  * caller needs this information as well. */
1478 static DWORD shader_glsl_add_dst_param(const struct wined3d_shader_instruction *ins,
1479         const struct wined3d_shader_dst_param *wined3d_dst, glsl_dst_param_t *glsl_dst)
1480 {
1481     BOOL is_color = FALSE;
1482
1483     glsl_dst->mask_str[0] = '\0';
1484     glsl_dst->reg_name[0] = '\0';
1485
1486     shader_glsl_get_register_name(&wined3d_dst->reg, glsl_dst->reg_name, &is_color, ins);
1487     return shader_glsl_get_write_mask(wined3d_dst, glsl_dst->mask_str);
1488 }
1489
1490 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1491 static DWORD shader_glsl_append_dst_ext(struct wined3d_shader_buffer *buffer,
1492         const struct wined3d_shader_instruction *ins, const struct wined3d_shader_dst_param *dst)
1493 {
1494     glsl_dst_param_t glsl_dst;
1495     DWORD mask;
1496
1497     mask = shader_glsl_add_dst_param(ins, dst, &glsl_dst);
1498     if (mask) shader_addline(buffer, "%s%s = %s(", glsl_dst.reg_name, glsl_dst.mask_str, shift_glsl_tab[dst->shift]);
1499
1500     return mask;
1501 }
1502
1503 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1504 static DWORD shader_glsl_append_dst(struct wined3d_shader_buffer *buffer, const struct wined3d_shader_instruction *ins)
1505 {
1506     return shader_glsl_append_dst_ext(buffer, ins, &ins->dst[0]);
1507 }
1508
1509 /** Process GLSL instruction modifiers */
1510 static void shader_glsl_add_instruction_modifiers(const struct wined3d_shader_instruction *ins)
1511 {
1512     glsl_dst_param_t dst_param;
1513     DWORD modifiers;
1514
1515     if (!ins->dst_count) return;
1516
1517     modifiers = ins->dst[0].modifiers;
1518     if (!modifiers) return;
1519
1520     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
1521
1522     if (modifiers & WINED3DSPDM_SATURATE)
1523     {
1524         /* _SAT means to clamp the value of the register to between 0 and 1 */
1525         shader_addline(ins->ctx->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1526                 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1527     }
1528
1529     if (modifiers & WINED3DSPDM_MSAMPCENTROID)
1530     {
1531         FIXME("_centroid modifier not handled\n");
1532     }
1533
1534     if (modifiers & WINED3DSPDM_PARTIALPRECISION)
1535     {
1536         /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1537     }
1538 }
1539
1540 static inline const char *shader_get_comp_op(DWORD op)
1541 {
1542     switch (op) {
1543         case COMPARISON_GT: return ">";
1544         case COMPARISON_EQ: return "==";
1545         case COMPARISON_GE: return ">=";
1546         case COMPARISON_LT: return "<";
1547         case COMPARISON_NE: return "!=";
1548         case COMPARISON_LE: return "<=";
1549         default:
1550             FIXME("Unrecognized comparison value: %u\n", op);
1551             return "(\?\?)";
1552     }
1553 }
1554
1555 static void shader_glsl_get_sample_function(DWORD sampler_type, DWORD flags, glsl_sample_function_t *sample_function)
1556 {
1557     BOOL projected = flags & WINED3D_GLSL_SAMPLE_PROJECTED;
1558     BOOL texrect = flags & WINED3D_GLSL_SAMPLE_RECT;
1559     BOOL lod = flags & WINED3D_GLSL_SAMPLE_LOD;
1560     BOOL grad = flags & WINED3D_GLSL_SAMPLE_GRAD;
1561
1562     /* Note that there's no such thing as a projected cube texture. */
1563     switch(sampler_type) {
1564         case WINED3DSTT_1D:
1565             if(lod) {
1566                 sample_function->name = projected ? "texture1DProjLod" : "texture1DLod";
1567             } else  if(grad) {
1568                 sample_function->name = projected ? "texture1DProjGradARB" : "texture1DGradARB";
1569             } else {
1570                 sample_function->name = projected ? "texture1DProj" : "texture1D";
1571             }
1572             sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1573             break;
1574         case WINED3DSTT_2D:
1575             if(texrect) {
1576                 if(lod) {
1577                     sample_function->name = projected ? "texture2DRectProjLod" : "texture2DRectLod";
1578                 } else  if(grad) {
1579                     /* What good are texrect grad functions? I don't know, but GL_EXT_gpu_shader4 defines them.
1580                     * There is no GL_ARB_shader_texture_lod spec yet, so I don't know if they're defined there
1581                      */
1582                     sample_function->name = projected ? "shadow2DRectProjGradARB" : "shadow2DRectGradARB";
1583                 } else {
1584                     sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1585                 }
1586             } else {
1587                 if(lod) {
1588                     sample_function->name = projected ? "texture2DProjLod" : "texture2DLod";
1589                 } else  if(grad) {
1590                     sample_function->name = projected ? "texture2DProjGradARB" : "texture2DGradARB";
1591                 } else {
1592                     sample_function->name = projected ? "texture2DProj" : "texture2D";
1593                 }
1594             }
1595             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1596             break;
1597         case WINED3DSTT_CUBE:
1598             if(lod) {
1599                 sample_function->name = "textureCubeLod";
1600             } else if(grad) {
1601                 sample_function->name = "textureCubeGradARB";
1602             } else {
1603                 sample_function->name = "textureCube";
1604             }
1605             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1606             break;
1607         case WINED3DSTT_VOLUME:
1608             if(lod) {
1609                 sample_function->name = projected ? "texture3DProjLod" : "texture3DLod";
1610             } else  if(grad) {
1611                 sample_function->name = projected ? "texture3DProjGradARB" : "texture3DGradARB";
1612             } else {
1613                 sample_function->name = projected ? "texture3DProj" : "texture3D";
1614             }
1615             sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1616             break;
1617         default:
1618             sample_function->name = "";
1619             sample_function->coord_mask = 0;
1620             FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1621             break;
1622     }
1623 }
1624
1625 static void shader_glsl_append_fixup_arg(char *arguments, const char *reg_name,
1626         BOOL sign_fixup, enum fixup_channel_source channel_source)
1627 {
1628     switch(channel_source)
1629     {
1630         case CHANNEL_SOURCE_ZERO:
1631             strcat(arguments, "0.0");
1632             break;
1633
1634         case CHANNEL_SOURCE_ONE:
1635             strcat(arguments, "1.0");
1636             break;
1637
1638         case CHANNEL_SOURCE_X:
1639             strcat(arguments, reg_name);
1640             strcat(arguments, ".x");
1641             break;
1642
1643         case CHANNEL_SOURCE_Y:
1644             strcat(arguments, reg_name);
1645             strcat(arguments, ".y");
1646             break;
1647
1648         case CHANNEL_SOURCE_Z:
1649             strcat(arguments, reg_name);
1650             strcat(arguments, ".z");
1651             break;
1652
1653         case CHANNEL_SOURCE_W:
1654             strcat(arguments, reg_name);
1655             strcat(arguments, ".w");
1656             break;
1657
1658         default:
1659             FIXME("Unhandled channel source %#x\n", channel_source);
1660             strcat(arguments, "undefined");
1661             break;
1662     }
1663
1664     if (sign_fixup) strcat(arguments, " * 2.0 - 1.0");
1665 }
1666
1667 static void shader_glsl_color_correction(const struct wined3d_shader_instruction *ins, struct color_fixup_desc fixup)
1668 {
1669     struct wined3d_shader_dst_param dst;
1670     unsigned int mask_size, remaining;
1671     glsl_dst_param_t dst_param;
1672     char arguments[256];
1673     DWORD mask;
1674
1675     mask = 0;
1676     if (fixup.x_sign_fixup || fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1677     if (fixup.y_sign_fixup || fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1678     if (fixup.z_sign_fixup || fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1679     if (fixup.w_sign_fixup || fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1680     mask &= ins->dst[0].write_mask;
1681
1682     if (!mask) return; /* Nothing to do */
1683
1684     if (is_yuv_fixup(fixup))
1685     {
1686         enum yuv_fixup yuv_fixup = get_yuv_fixup(fixup);
1687         FIXME("YUV fixup (%#x) not supported\n", yuv_fixup);
1688         return;
1689     }
1690
1691     mask_size = shader_glsl_get_write_mask_size(mask);
1692
1693     dst = ins->dst[0];
1694     dst.write_mask = mask;
1695     shader_glsl_add_dst_param(ins, &dst, &dst_param);
1696
1697     arguments[0] = '\0';
1698     remaining = mask_size;
1699     if (mask & WINED3DSP_WRITEMASK_0)
1700     {
1701         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.x_sign_fixup, fixup.x_source);
1702         if (--remaining) strcat(arguments, ", ");
1703     }
1704     if (mask & WINED3DSP_WRITEMASK_1)
1705     {
1706         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.y_sign_fixup, fixup.y_source);
1707         if (--remaining) strcat(arguments, ", ");
1708     }
1709     if (mask & WINED3DSP_WRITEMASK_2)
1710     {
1711         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.z_sign_fixup, fixup.z_source);
1712         if (--remaining) strcat(arguments, ", ");
1713     }
1714     if (mask & WINED3DSP_WRITEMASK_3)
1715     {
1716         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.w_sign_fixup, fixup.w_source);
1717         if (--remaining) strcat(arguments, ", ");
1718     }
1719
1720     if (mask_size > 1)
1721     {
1722         shader_addline(ins->ctx->buffer, "%s%s = vec%u(%s);\n",
1723                 dst_param.reg_name, dst_param.mask_str, mask_size, arguments);
1724     }
1725     else
1726     {
1727         shader_addline(ins->ctx->buffer, "%s%s = %s;\n", dst_param.reg_name, dst_param.mask_str, arguments);
1728     }
1729 }
1730
1731 static void PRINTF_ATTR(8, 9) shader_glsl_gen_sample_code(const struct wined3d_shader_instruction *ins,
1732         DWORD sampler, const glsl_sample_function_t *sample_function, DWORD swizzle,
1733         const char *dx, const char *dy,
1734         const char *bias, const char *coord_reg_fmt, ...)
1735 {
1736     const char *sampler_base;
1737     char dst_swizzle[6];
1738     struct color_fixup_desc fixup;
1739     BOOL np2_fixup = FALSE;
1740     va_list args;
1741
1742     shader_glsl_swizzle_to_str(swizzle, FALSE, ins->dst[0].write_mask, dst_swizzle);
1743
1744     if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1745     {
1746         const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1747         fixup = priv->cur_ps_args->color_fixup[sampler];
1748         sampler_base = "Psampler";
1749
1750         if(priv->cur_ps_args->np2_fixup & (1 << sampler)) {
1751             if(bias) {
1752                 FIXME("Biased sampling from NP2 textures is unsupported\n");
1753             } else {
1754                 np2_fixup = TRUE;
1755             }
1756         }
1757     } else {
1758         sampler_base = "Vsampler";
1759         fixup = COLOR_FIXUP_IDENTITY; /* FIXME: Vshader color fixup */
1760     }
1761
1762     shader_glsl_append_dst(ins->ctx->buffer, ins);
1763
1764     shader_addline(ins->ctx->buffer, "%s(%s%u, ", sample_function->name, sampler_base, sampler);
1765
1766     va_start(args, coord_reg_fmt);
1767     shader_vaddline(ins->ctx->buffer, coord_reg_fmt, args);
1768     va_end(args);
1769
1770     if(bias) {
1771         shader_addline(ins->ctx->buffer, ", %s)%s);\n", bias, dst_swizzle);
1772     } else {
1773         if (np2_fixup) {
1774             const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1775             const unsigned char idx = priv->cur_np2fixup_info->idx[sampler];
1776
1777             shader_addline(ins->ctx->buffer, " * PsamplerNP2Fixup[%u].%s)%s);\n", idx >> 1,
1778                            (idx % 2) ? "zw" : "xy", dst_swizzle);
1779         } else if(dx && dy) {
1780             shader_addline(ins->ctx->buffer, ", %s, %s)%s);\n", dx, dy, dst_swizzle);
1781         } else {
1782             shader_addline(ins->ctx->buffer, ")%s);\n", dst_swizzle);
1783         }
1784     }
1785
1786     if(!is_identity_fixup(fixup)) {
1787         shader_glsl_color_correction(ins, fixup);
1788     }
1789 }
1790
1791 /*****************************************************************************
1792  * Begin processing individual instruction opcodes
1793  ****************************************************************************/
1794
1795 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
1796 static void shader_glsl_arith(const struct wined3d_shader_instruction *ins)
1797 {
1798     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1799     glsl_src_param_t src0_param;
1800     glsl_src_param_t src1_param;
1801     DWORD write_mask;
1802     char op;
1803
1804     /* Determine the GLSL operator to use based on the opcode */
1805     switch (ins->handler_idx)
1806     {
1807         case WINED3DSIH_MUL: op = '*'; break;
1808         case WINED3DSIH_ADD: op = '+'; break;
1809         case WINED3DSIH_SUB: op = '-'; break;
1810         default:
1811             op = ' ';
1812             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
1813             break;
1814     }
1815
1816     write_mask = shader_glsl_append_dst(buffer, ins);
1817     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
1818     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
1819     shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
1820 }
1821
1822 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
1823 static void shader_glsl_mov(const struct wined3d_shader_instruction *ins)
1824 {
1825     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1826     glsl_src_param_t src0_param;
1827     DWORD write_mask;
1828
1829     write_mask = shader_glsl_append_dst(buffer, ins);
1830     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
1831
1832     /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
1833      * shader versions WINED3DSIO_MOVA is used for this. */
1834     if (ins->ctx->reg_maps->shader_version.major == 1
1835             && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
1836             && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1837     {
1838         /* This is a simple floor() */
1839         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1840         if (mask_size > 1) {
1841             shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
1842         } else {
1843             shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
1844         }
1845     }
1846     else if(ins->handler_idx == WINED3DSIH_MOVA)
1847     {
1848         /* We need to *round* to the nearest int here. */
1849         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1850         if (mask_size > 1) {
1851             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);
1852         } else {
1853             shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n", src0_param.param_str, src0_param.param_str);
1854         }
1855     } else {
1856         shader_addline(buffer, "%s);\n", src0_param.param_str);
1857     }
1858 }
1859
1860 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
1861 static void shader_glsl_dot(const struct wined3d_shader_instruction *ins)
1862 {
1863     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1864     glsl_src_param_t src0_param;
1865     glsl_src_param_t src1_param;
1866     DWORD dst_write_mask, src_write_mask;
1867     unsigned int dst_size = 0;
1868
1869     dst_write_mask = shader_glsl_append_dst(buffer, ins);
1870     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1871
1872     /* dp3 works on vec3, dp4 on vec4 */
1873     if (ins->handler_idx == WINED3DSIH_DP4)
1874     {
1875         src_write_mask = WINED3DSP_WRITEMASK_ALL;
1876     } else {
1877         src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1878     }
1879
1880     shader_glsl_add_src_param(ins, &ins->src[0], src_write_mask, &src0_param);
1881     shader_glsl_add_src_param(ins, &ins->src[1], src_write_mask, &src1_param);
1882
1883     if (dst_size > 1) {
1884         shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1885     } else {
1886         shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
1887     }
1888 }
1889
1890 /* Note that this instruction has some restrictions. The destination write mask
1891  * can't contain the w component, and the source swizzles have to be .xyzw */
1892 static void shader_glsl_cross(const struct wined3d_shader_instruction *ins)
1893 {
1894     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1895     glsl_src_param_t src0_param;
1896     glsl_src_param_t src1_param;
1897     char dst_mask[6];
1898
1899     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
1900     shader_glsl_append_dst(ins->ctx->buffer, ins);
1901     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
1902     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
1903     shader_addline(ins->ctx->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
1904 }
1905
1906 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
1907  * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
1908  * GLSL uses the value as-is. */
1909 static void shader_glsl_pow(const struct wined3d_shader_instruction *ins)
1910 {
1911     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1912     glsl_src_param_t src0_param;
1913     glsl_src_param_t src1_param;
1914     DWORD dst_write_mask;
1915     unsigned int dst_size;
1916
1917     dst_write_mask = shader_glsl_append_dst(buffer, ins);
1918     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1919
1920     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
1921     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
1922
1923     if (dst_size > 1) {
1924         shader_addline(buffer, "vec%d(pow(abs(%s), %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1925     } else {
1926         shader_addline(buffer, "pow(abs(%s), %s));\n", src0_param.param_str, src1_param.param_str);
1927     }
1928 }
1929
1930 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
1931  * Src0 is a scalar. Note that D3D uses the absolute of src0, while
1932  * GLSL uses the value as-is. */
1933 static void shader_glsl_log(const struct wined3d_shader_instruction *ins)
1934 {
1935     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1936     glsl_src_param_t src0_param;
1937     DWORD dst_write_mask;
1938     unsigned int dst_size;
1939
1940     dst_write_mask = shader_glsl_append_dst(buffer, ins);
1941     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1942
1943     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
1944
1945     if (dst_size > 1) {
1946         shader_addline(buffer, "vec%d(log2(abs(%s))));\n", dst_size, src0_param.param_str);
1947     } else {
1948         shader_addline(buffer, "log2(abs(%s)));\n", src0_param.param_str);
1949     }
1950 }
1951
1952 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
1953 static void shader_glsl_map2gl(const struct wined3d_shader_instruction *ins)
1954 {
1955     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1956     glsl_src_param_t src_param;
1957     const char *instruction;
1958     DWORD write_mask;
1959     unsigned i;
1960
1961     /* Determine the GLSL function to use based on the opcode */
1962     /* TODO: Possibly make this a table for faster lookups */
1963     switch (ins->handler_idx)
1964     {
1965         case WINED3DSIH_MIN: instruction = "min"; break;
1966         case WINED3DSIH_MAX: instruction = "max"; break;
1967         case WINED3DSIH_ABS: instruction = "abs"; break;
1968         case WINED3DSIH_FRC: instruction = "fract"; break;
1969         case WINED3DSIH_NRM: instruction = "normalize"; break;
1970         case WINED3DSIH_EXP: instruction = "exp2"; break;
1971         case WINED3DSIH_DSX: instruction = "dFdx"; break;
1972         case WINED3DSIH_DSY: instruction = "ycorrection.y * dFdy"; break;
1973         default: instruction = "";
1974             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
1975             break;
1976     }
1977
1978     write_mask = shader_glsl_append_dst(buffer, ins);
1979
1980     shader_addline(buffer, "%s(", instruction);
1981
1982     if (ins->src_count)
1983     {
1984         shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
1985         shader_addline(buffer, "%s", src_param.param_str);
1986         for (i = 1; i < ins->src_count; ++i)
1987         {
1988             shader_glsl_add_src_param(ins, &ins->src[i], write_mask, &src_param);
1989             shader_addline(buffer, ", %s", src_param.param_str);
1990         }
1991     }
1992
1993     shader_addline(buffer, "));\n");
1994 }
1995
1996 /** Process the WINED3DSIO_EXPP instruction in GLSL:
1997  * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
1998  *   dst.x = 2^(floor(src))
1999  *   dst.y = src - floor(src)
2000  *   dst.z = 2^src   (partial precision is allowed, but optional)
2001  *   dst.w = 1.0;
2002  * For 2.0 shaders, just do this (honoring writemask and swizzle):
2003  *   dst = 2^src;    (partial precision is allowed, but optional)
2004  */
2005 static void shader_glsl_expp(const struct wined3d_shader_instruction *ins)
2006 {
2007     glsl_src_param_t src_param;
2008
2009     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src_param);
2010
2011     if (ins->ctx->reg_maps->shader_version.major < 2)
2012     {
2013         char dst_mask[6];
2014
2015         shader_addline(ins->ctx->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
2016         shader_addline(ins->ctx->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
2017         shader_addline(ins->ctx->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
2018         shader_addline(ins->ctx->buffer, "tmp0.w = 1.0;\n");
2019
2020         shader_glsl_append_dst(ins->ctx->buffer, ins);
2021         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2022         shader_addline(ins->ctx->buffer, "tmp0%s);\n", dst_mask);
2023     } else {
2024         DWORD write_mask;
2025         unsigned int mask_size;
2026
2027         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2028         mask_size = shader_glsl_get_write_mask_size(write_mask);
2029
2030         if (mask_size > 1) {
2031             shader_addline(ins->ctx->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
2032         } else {
2033             shader_addline(ins->ctx->buffer, "exp2(%s));\n", src_param.param_str);
2034         }
2035     }
2036 }
2037
2038 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
2039 static void shader_glsl_rcp(const struct wined3d_shader_instruction *ins)
2040 {
2041     glsl_src_param_t src_param;
2042     DWORD write_mask;
2043     unsigned int mask_size;
2044
2045     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2046     mask_size = shader_glsl_get_write_mask_size(write_mask);
2047     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2048
2049     if (mask_size > 1) {
2050         shader_addline(ins->ctx->buffer, "vec%d(1.0 / %s));\n", mask_size, src_param.param_str);
2051     } else {
2052         shader_addline(ins->ctx->buffer, "1.0 / %s);\n", src_param.param_str);
2053     }
2054 }
2055
2056 static void shader_glsl_rsq(const struct wined3d_shader_instruction *ins)
2057 {
2058     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2059     glsl_src_param_t src_param;
2060     DWORD write_mask;
2061     unsigned int mask_size;
2062
2063     write_mask = shader_glsl_append_dst(buffer, ins);
2064     mask_size = shader_glsl_get_write_mask_size(write_mask);
2065
2066     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2067
2068     if (mask_size > 1) {
2069         shader_addline(buffer, "vec%d(inversesqrt(%s)));\n", mask_size, src_param.param_str);
2070     } else {
2071         shader_addline(buffer, "inversesqrt(%s));\n", src_param.param_str);
2072     }
2073 }
2074
2075 /** Process signed comparison opcodes in GLSL. */
2076 static void shader_glsl_compare(const struct wined3d_shader_instruction *ins)
2077 {
2078     glsl_src_param_t src0_param;
2079     glsl_src_param_t src1_param;
2080     DWORD write_mask;
2081     unsigned int mask_size;
2082
2083     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2084     mask_size = shader_glsl_get_write_mask_size(write_mask);
2085     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2086     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2087
2088     if (mask_size > 1) {
2089         const char *compare;
2090
2091         switch(ins->handler_idx)
2092         {
2093             case WINED3DSIH_SLT: compare = "lessThan"; break;
2094             case WINED3DSIH_SGE: compare = "greaterThanEqual"; break;
2095             default: compare = "";
2096                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2097         }
2098
2099         shader_addline(ins->ctx->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
2100                 src0_param.param_str, src1_param.param_str);
2101     } else {
2102         switch(ins->handler_idx)
2103         {
2104             case WINED3DSIH_SLT:
2105                 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
2106                  * to return 0.0 but step returns 1.0 because step is not < x
2107                  * An alternative is a bvec compare padded with an unused second component.
2108                  * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
2109                  * issue. Playing with not() is not possible either because not() does not accept
2110                  * a scalar.
2111                  */
2112                 shader_addline(ins->ctx->buffer, "(%s < %s) ? 1.0 : 0.0);\n",
2113                         src0_param.param_str, src1_param.param_str);
2114                 break;
2115             case WINED3DSIH_SGE:
2116                 /* Here we can use the step() function and safe a conditional */
2117                 shader_addline(ins->ctx->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
2118                 break;
2119             default:
2120                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2121         }
2122
2123     }
2124 }
2125
2126 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
2127 static void shader_glsl_cmp(const struct wined3d_shader_instruction *ins)
2128 {
2129     glsl_src_param_t src0_param;
2130     glsl_src_param_t src1_param;
2131     glsl_src_param_t src2_param;
2132     DWORD write_mask, cmp_channel = 0;
2133     unsigned int i, j;
2134     char mask_char[6];
2135     BOOL temp_destination = FALSE;
2136
2137     if (shader_is_scalar(&ins->src[0].reg))
2138     {
2139         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2140
2141         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2142         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2143         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2144
2145         shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2146                        src0_param.param_str, src1_param.param_str, src2_param.param_str);
2147     } else {
2148         DWORD dst_mask = ins->dst[0].write_mask;
2149         struct wined3d_shader_dst_param dst = ins->dst[0];
2150
2151         /* Cycle through all source0 channels */
2152         for (i=0; i<4; i++) {
2153             write_mask = 0;
2154             /* Find the destination channels which use the current source0 channel */
2155             for (j=0; j<4; j++) {
2156                 if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2157                 {
2158                     write_mask |= WINED3DSP_WRITEMASK_0 << j;
2159                     cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2160                 }
2161             }
2162             dst.write_mask = dst_mask & write_mask;
2163
2164             /* Splitting the cmp instruction up in multiple lines imposes a problem:
2165             * The first lines may overwrite source parameters of the following lines.
2166             * Deal with that by using a temporary destination register if needed
2167             */
2168             if ((ins->src[0].reg.idx == ins->dst[0].reg.idx
2169                     && ins->src[0].reg.type == ins->dst[0].reg.type)
2170                     || (ins->src[1].reg.idx == ins->dst[0].reg.idx
2171                     && ins->src[1].reg.type == ins->dst[0].reg.type)
2172                     || (ins->src[2].reg.idx == ins->dst[0].reg.idx
2173                     && ins->src[2].reg.type == ins->dst[0].reg.type))
2174             {
2175                 write_mask = shader_glsl_get_write_mask(&dst, mask_char);
2176                 if (!write_mask) continue;
2177                 shader_addline(ins->ctx->buffer, "tmp0%s = (", mask_char);
2178                 temp_destination = TRUE;
2179             } else {
2180                 write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2181                 if (!write_mask) continue;
2182             }
2183
2184             shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2185             shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2186             shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2187
2188             shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2189                         src0_param.param_str, src1_param.param_str, src2_param.param_str);
2190         }
2191
2192         if(temp_destination) {
2193             shader_glsl_get_write_mask(&ins->dst[0], mask_char);
2194             shader_glsl_append_dst(ins->ctx->buffer, ins);
2195             shader_addline(ins->ctx->buffer, "tmp0%s);\n", mask_char);
2196         }
2197     }
2198
2199 }
2200
2201 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
2202 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
2203  * the compare is done per component of src0. */
2204 static void shader_glsl_cnd(const struct wined3d_shader_instruction *ins)
2205 {
2206     struct wined3d_shader_dst_param dst;
2207     glsl_src_param_t src0_param;
2208     glsl_src_param_t src1_param;
2209     glsl_src_param_t src2_param;
2210     DWORD write_mask, cmp_channel = 0;
2211     unsigned int i, j;
2212     DWORD dst_mask;
2213     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2214             ins->ctx->reg_maps->shader_version.minor);
2215
2216     if (shader_version < WINED3D_SHADER_VERSION(1, 4))
2217     {
2218         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2219         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2220         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2221         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2222
2223         /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
2224         if (ins->coissue)
2225         {
2226             shader_addline(ins->ctx->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
2227         } else {
2228             shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2229                     src0_param.param_str, src1_param.param_str, src2_param.param_str);
2230         }
2231         return;
2232     }
2233     /* Cycle through all source0 channels */
2234     dst_mask = ins->dst[0].write_mask;
2235     dst = ins->dst[0];
2236     for (i=0; i<4; i++) {
2237         write_mask = 0;
2238         /* Find the destination channels which use the current source0 channel */
2239         for (j=0; j<4; j++) {
2240             if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2241             {
2242                 write_mask |= WINED3DSP_WRITEMASK_0 << j;
2243                 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2244             }
2245         }
2246
2247         dst.write_mask = dst_mask & write_mask;
2248         write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2249         if (!write_mask) continue;
2250
2251         shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2252         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2253         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2254
2255         shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2256                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2257     }
2258 }
2259
2260 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
2261 static void shader_glsl_mad(const struct wined3d_shader_instruction *ins)
2262 {
2263     glsl_src_param_t src0_param;
2264     glsl_src_param_t src1_param;
2265     glsl_src_param_t src2_param;
2266     DWORD write_mask;
2267
2268     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2269     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2270     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2271     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2272     shader_addline(ins->ctx->buffer, "(%s * %s) + %s);\n",
2273             src0_param.param_str, src1_param.param_str, src2_param.param_str);
2274 }
2275
2276 /* Handles transforming all WINED3DSIO_M?x? opcodes for
2277    Vertex shaders to GLSL codes */
2278 static void shader_glsl_mnxn(const struct wined3d_shader_instruction *ins)
2279 {
2280     int i;
2281     int nComponents = 0;
2282     struct wined3d_shader_dst_param tmp_dst = {{0}};
2283     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2284     struct wined3d_shader_instruction tmp_ins;
2285
2286     memset(&tmp_ins, 0, sizeof(tmp_ins));
2287
2288     /* Set constants for the temporary argument */
2289     tmp_ins.ctx = ins->ctx;
2290     tmp_ins.dst_count = 1;
2291     tmp_ins.dst = &tmp_dst;
2292     tmp_ins.src_count = 2;
2293     tmp_ins.src = tmp_src;
2294
2295     switch(ins->handler_idx)
2296     {
2297         case WINED3DSIH_M4x4:
2298             nComponents = 4;
2299             tmp_ins.handler_idx = WINED3DSIH_DP4;
2300             break;
2301         case WINED3DSIH_M4x3:
2302             nComponents = 3;
2303             tmp_ins.handler_idx = WINED3DSIH_DP4;
2304             break;
2305         case WINED3DSIH_M3x4:
2306             nComponents = 4;
2307             tmp_ins.handler_idx = WINED3DSIH_DP3;
2308             break;
2309         case WINED3DSIH_M3x3:
2310             nComponents = 3;
2311             tmp_ins.handler_idx = WINED3DSIH_DP3;
2312             break;
2313         case WINED3DSIH_M3x2:
2314             nComponents = 2;
2315             tmp_ins.handler_idx = WINED3DSIH_DP3;
2316             break;
2317         default:
2318             break;
2319     }
2320
2321     tmp_dst = ins->dst[0];
2322     tmp_src[0] = ins->src[0];
2323     tmp_src[1] = ins->src[1];
2324     for (i = 0; i < nComponents; ++i)
2325     {
2326         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2327         shader_glsl_dot(&tmp_ins);
2328         ++tmp_src[1].reg.idx;
2329     }
2330 }
2331
2332 /**
2333     The LRP instruction performs a component-wise linear interpolation
2334     between the second and third operands using the first operand as the
2335     blend factor.  Equation:  (dst = src2 + src0 * (src1 - src2))
2336     This is equivalent to mix(src2, src1, src0);
2337 */
2338 static void shader_glsl_lrp(const struct wined3d_shader_instruction *ins)
2339 {
2340     glsl_src_param_t src0_param;
2341     glsl_src_param_t src1_param;
2342     glsl_src_param_t src2_param;
2343     DWORD write_mask;
2344
2345     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2346
2347     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2348     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2349     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2350
2351     shader_addline(ins->ctx->buffer, "mix(%s, %s, %s));\n",
2352             src2_param.param_str, src1_param.param_str, src0_param.param_str);
2353 }
2354
2355 /** Process the WINED3DSIO_LIT instruction in GLSL:
2356  * dst.x = dst.w = 1.0
2357  * dst.y = (src0.x > 0) ? src0.x
2358  * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
2359  *                                        where src.w is clamped at +- 128
2360  */
2361 static void shader_glsl_lit(const struct wined3d_shader_instruction *ins)
2362 {
2363     glsl_src_param_t src0_param;
2364     glsl_src_param_t src1_param;
2365     glsl_src_param_t src3_param;
2366     char dst_mask[6];
2367
2368     shader_glsl_append_dst(ins->ctx->buffer, ins);
2369     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2370
2371     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2372     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src1_param);
2373     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src3_param);
2374
2375     /* The sdk specifies the instruction like this
2376      * dst.x = 1.0;
2377      * if(src.x > 0.0) dst.y = src.x
2378      * else dst.y = 0.0.
2379      * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
2380      * else dst.z = 0.0;
2381      * dst.w = 1.0;
2382      *
2383      * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
2384      * dst.x = 1.0                                  ... No further explanation needed
2385      * dst.y = max(src.y, 0.0);                     ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
2386      * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0;   ... 0 ^ power is 0, and otherwise we use y anyway
2387      * dst.w = 1.0.                                 ... Nothing fancy.
2388      *
2389      * So we still have one conditional in there. So do this:
2390      * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
2391      *
2392      * 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),
2393      * 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.
2394      * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to
2395      */
2396     shader_addline(ins->ctx->buffer,
2397             "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",
2398             src0_param.param_str, src1_param.param_str, src0_param.param_str, src3_param.param_str, dst_mask);
2399 }
2400
2401 /** Process the WINED3DSIO_DST instruction in GLSL:
2402  * dst.x = 1.0
2403  * dst.y = src0.x * src0.y
2404  * dst.z = src0.z
2405  * dst.w = src1.w
2406  */
2407 static void shader_glsl_dst(const struct wined3d_shader_instruction *ins)
2408 {
2409     glsl_src_param_t src0y_param;
2410     glsl_src_param_t src0z_param;
2411     glsl_src_param_t src1y_param;
2412     glsl_src_param_t src1w_param;
2413     char dst_mask[6];
2414
2415     shader_glsl_append_dst(ins->ctx->buffer, ins);
2416     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2417
2418     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src0y_param);
2419     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &src0z_param);
2420     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_1, &src1y_param);
2421     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_3, &src1w_param);
2422
2423     shader_addline(ins->ctx->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
2424             src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
2425 }
2426
2427 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
2428  * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
2429  * can handle it.  But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
2430  *
2431  * dst.x = cos(src0.?)
2432  * dst.y = sin(src0.?)
2433  * dst.z = dst.z
2434  * dst.w = dst.w
2435  */
2436 static void shader_glsl_sincos(const struct wined3d_shader_instruction *ins)
2437 {
2438     glsl_src_param_t src0_param;
2439     DWORD write_mask;
2440
2441     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2442     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2443
2444     switch (write_mask) {
2445         case WINED3DSP_WRITEMASK_0:
2446             shader_addline(ins->ctx->buffer, "cos(%s));\n", src0_param.param_str);
2447             break;
2448
2449         case WINED3DSP_WRITEMASK_1:
2450             shader_addline(ins->ctx->buffer, "sin(%s));\n", src0_param.param_str);
2451             break;
2452
2453         case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
2454             shader_addline(ins->ctx->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
2455             break;
2456
2457         default:
2458             ERR("Write mask should be .x, .y or .xy\n");
2459             break;
2460     }
2461 }
2462
2463 /* sgn in vs_2_0 has 2 extra parameters(registers for temporary storage) which we don't use
2464  * here. But those extra parameters require a dedicated function for sgn, since map2gl would
2465  * generate invalid code
2466  */
2467 static void shader_glsl_sgn(const struct wined3d_shader_instruction *ins)
2468 {
2469     glsl_src_param_t src0_param;
2470     DWORD write_mask;
2471
2472     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2473     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2474
2475     shader_addline(ins->ctx->buffer, "sign(%s));\n", src0_param.param_str);
2476 }
2477
2478 /** Process the WINED3DSIO_LOOP instruction in GLSL:
2479  * Start a for() loop where src1.y is the initial value of aL,
2480  *  increment aL by src1.z for a total of src1.x iterations.
2481  *  Need to use a temporary variable for this operation.
2482  */
2483 /* FIXME: I don't think nested loops will work correctly this way. */
2484 static void shader_glsl_loop(const struct wined3d_shader_instruction *ins)
2485 {
2486     glsl_src_param_t src1_param;
2487     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2488     const DWORD *control_values = NULL;
2489     const local_constant *constant;
2490
2491     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2492
2493     /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2494      * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2495      * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2496      * addressing.
2497      */
2498     if (ins->src[1].reg.type == WINED3DSPR_CONSTINT)
2499     {
2500         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
2501             if (constant->idx == ins->src[1].reg.idx)
2502             {
2503                 control_values = constant->value;
2504                 break;
2505             }
2506         }
2507     }
2508
2509     if (control_values)
2510     {
2511         struct wined3d_shader_loop_control loop_control;
2512         loop_control.count = control_values[0];
2513         loop_control.start = control_values[1];
2514         loop_control.step = (int)control_values[2];
2515
2516         if (loop_control.step > 0)
2517         {
2518             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u < (%u * %d + %u); aL%u += %d) {\n",
2519                     shader->baseShader.cur_loop_depth, loop_control.start,
2520                     shader->baseShader.cur_loop_depth, loop_control.count, loop_control.step, loop_control.start,
2521                     shader->baseShader.cur_loop_depth, loop_control.step);
2522         }
2523         else if (loop_control.step < 0)
2524         {
2525             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u > (%u * %d + %u); aL%u += %d) {\n",
2526                     shader->baseShader.cur_loop_depth, loop_control.start,
2527                     shader->baseShader.cur_loop_depth, loop_control.count, loop_control.step, loop_control.start,
2528                     shader->baseShader.cur_loop_depth, loop_control.step);
2529         }
2530         else
2531         {
2532             shader_addline(ins->ctx->buffer, "for (aL%u = %u, tmpInt%u = 0; tmpInt%u < %u; tmpInt%u++) {\n",
2533                     shader->baseShader.cur_loop_depth, loop_control.start, shader->baseShader.cur_loop_depth,
2534                     shader->baseShader.cur_loop_depth, loop_control.count,
2535                     shader->baseShader.cur_loop_depth);
2536         }
2537     } else {
2538         shader_addline(ins->ctx->buffer,
2539                 "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2540                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno,
2541                 src1_param.reg_name, shader->baseShader.cur_loop_depth, src1_param.reg_name,
2542                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno, src1_param.reg_name);
2543     }
2544
2545     shader->baseShader.cur_loop_depth++;
2546     shader->baseShader.cur_loop_regno++;
2547 }
2548
2549 static void shader_glsl_end(const struct wined3d_shader_instruction *ins)
2550 {
2551     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2552
2553     shader_addline(ins->ctx->buffer, "}\n");
2554
2555     if (ins->handler_idx == WINED3DSIH_ENDLOOP)
2556     {
2557         shader->baseShader.cur_loop_depth--;
2558         shader->baseShader.cur_loop_regno--;
2559     }
2560
2561     if (ins->handler_idx == WINED3DSIH_ENDREP)
2562     {
2563         shader->baseShader.cur_loop_depth--;
2564     }
2565 }
2566
2567 static void shader_glsl_rep(const struct wined3d_shader_instruction *ins)
2568 {
2569     IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2570     glsl_src_param_t src0_param;
2571     const DWORD *control_values = NULL;
2572     const local_constant *constant;
2573
2574     /* Try to hardcode local values to help the GLSL compiler to unroll and optimize the loop */
2575     if (ins->src[0].reg.type == WINED3DSPR_CONSTINT)
2576     {
2577         LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry)
2578         {
2579             if (constant->idx == ins->src[0].reg.idx)
2580             {
2581                 control_values = constant->value;
2582                 break;
2583             }
2584         }
2585     }
2586
2587     if(control_values) {
2588         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %d; tmpInt%d++) {\n",
2589                        shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2590                        control_values[0], shader->baseShader.cur_loop_depth);
2591     } else {
2592         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2593         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2594                 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2595                 src0_param.param_str, shader->baseShader.cur_loop_depth);
2596     }
2597     shader->baseShader.cur_loop_depth++;
2598 }
2599
2600 static void shader_glsl_if(const struct wined3d_shader_instruction *ins)
2601 {
2602     glsl_src_param_t src0_param;
2603
2604     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2605     shader_addline(ins->ctx->buffer, "if (%s) {\n", src0_param.param_str);
2606 }
2607
2608 static void shader_glsl_ifc(const struct wined3d_shader_instruction *ins)
2609 {
2610     glsl_src_param_t src0_param;
2611     glsl_src_param_t src1_param;
2612
2613     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2614     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2615
2616     shader_addline(ins->ctx->buffer, "if (%s %s %s) {\n",
2617             src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2618 }
2619
2620 static void shader_glsl_else(const struct wined3d_shader_instruction *ins)
2621 {
2622     shader_addline(ins->ctx->buffer, "} else {\n");
2623 }
2624
2625 static void shader_glsl_break(const struct wined3d_shader_instruction *ins)
2626 {
2627     shader_addline(ins->ctx->buffer, "break;\n");
2628 }
2629
2630 /* FIXME: According to MSDN the compare is done per component. */
2631 static void shader_glsl_breakc(const struct wined3d_shader_instruction *ins)
2632 {
2633     glsl_src_param_t src0_param;
2634     glsl_src_param_t src1_param;
2635
2636     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2637     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2638
2639     shader_addline(ins->ctx->buffer, "if (%s %s %s) break;\n",
2640             src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2641 }
2642
2643 static void shader_glsl_label(const struct wined3d_shader_instruction *ins)
2644 {
2645     shader_addline(ins->ctx->buffer, "}\n");
2646     shader_addline(ins->ctx->buffer, "void subroutine%u () {\n",  ins->src[0].reg.idx);
2647 }
2648
2649 static void shader_glsl_call(const struct wined3d_shader_instruction *ins)
2650 {
2651     shader_addline(ins->ctx->buffer, "subroutine%u();\n", ins->src[0].reg.idx);
2652 }
2653
2654 static void shader_glsl_callnz(const struct wined3d_shader_instruction *ins)
2655 {
2656     glsl_src_param_t src1_param;
2657
2658     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2659     shader_addline(ins->ctx->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, ins->src[0].reg.idx);
2660 }
2661
2662 static void shader_glsl_ret(const struct wined3d_shader_instruction *ins)
2663 {
2664     /* No-op. The closing } is written when a new function is started, and at the end of the shader. This
2665      * function only suppresses the unhandled instruction warning
2666      */
2667 }
2668
2669 /*********************************************
2670  * Pixel Shader Specific Code begins here
2671  ********************************************/
2672 static void shader_glsl_tex(const struct wined3d_shader_instruction *ins)
2673 {
2674     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2675     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2676     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2677             ins->ctx->reg_maps->shader_version.minor);
2678     glsl_sample_function_t sample_function;
2679     DWORD sample_flags = 0;
2680     WINED3DSAMPLER_TEXTURE_TYPE sampler_type;
2681     DWORD sampler_idx;
2682     DWORD mask = 0, swizzle;
2683
2684     /* 1.0-1.4: Use destination register as sampler source.
2685      * 2.0+: Use provided sampler source. */
2686     if (shader_version < WINED3D_SHADER_VERSION(2,0)) sampler_idx = ins->dst[0].reg.idx;
2687     else sampler_idx = ins->src[1].reg.idx;
2688     sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
2689
2690     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2691     {
2692         DWORD flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2693
2694         /* Projected cube textures don't make a lot of sense, the resulting coordinates stay the same. */
2695         if (flags & WINED3DTTFF_PROJECTED && sampler_type != WINED3DSTT_CUBE) {
2696             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2697             switch (flags & ~WINED3DTTFF_PROJECTED) {
2698                 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2699                 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
2700                 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
2701                 case WINED3DTTFF_COUNT4:
2702                 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
2703             }
2704         }
2705     }
2706     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
2707     {
2708         DWORD src_mod = ins->src[0].modifiers;
2709
2710         if (src_mod == WINED3DSPSM_DZ) {
2711             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2712             mask = WINED3DSP_WRITEMASK_2;
2713         } else if (src_mod == WINED3DSPSM_DW) {
2714             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2715             mask = WINED3DSP_WRITEMASK_3;
2716         }
2717     } else {
2718         if (ins->flags & WINED3DSI_TEXLD_PROJECT)
2719         {
2720             /* ps 2.0 texldp instruction always divides by the fourth component. */
2721             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2722             mask = WINED3DSP_WRITEMASK_3;
2723         }
2724     }
2725
2726     if(deviceImpl->stateBlock->textures[sampler_idx] &&
2727        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2728         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
2729     }
2730
2731     shader_glsl_get_sample_function(sampler_type, sample_flags, &sample_function);
2732     mask |= sample_function.coord_mask;
2733
2734     if (shader_version < WINED3D_SHADER_VERSION(2,0)) swizzle = WINED3DSP_NOSWIZZLE;
2735     else swizzle = ins->src[1].swizzle;
2736
2737     /* 1.0-1.3: Use destination register as coordinate source.
2738        1.4+: Use provided coordinate source register. */
2739     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2740     {
2741         char coord_mask[6];
2742         shader_glsl_write_mask_to_str(mask, coord_mask);
2743         shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
2744                 "T%u%s", sampler_idx, coord_mask);
2745     } else {
2746         glsl_src_param_t coord_param;
2747         shader_glsl_add_src_param(ins, &ins->src[0], mask, &coord_param);
2748         if (ins->flags & WINED3DSI_TEXLD_BIAS)
2749         {
2750             glsl_src_param_t bias;
2751             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &bias);
2752             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, bias.param_str,
2753                     "%s", coord_param.param_str);
2754         } else {
2755             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
2756                     "%s", coord_param.param_str);
2757         }
2758     }
2759 }
2760
2761 static void shader_glsl_texldd(const struct wined3d_shader_instruction *ins)
2762 {
2763     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2764     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2765     const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
2766     glsl_sample_function_t sample_function;
2767     glsl_src_param_t coord_param, dx_param, dy_param;
2768     DWORD sample_flags = WINED3D_GLSL_SAMPLE_GRAD;
2769     DWORD sampler_type;
2770     DWORD sampler_idx;
2771     DWORD swizzle = ins->src[1].swizzle;
2772
2773     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD])
2774     {
2775         FIXME("texldd used, but not supported by hardware. Falling back to regular tex\n");
2776         return shader_glsl_tex(ins);
2777     }
2778
2779     sampler_idx = ins->src[1].reg.idx;
2780     sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
2781     if(deviceImpl->stateBlock->textures[sampler_idx] &&
2782        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2783         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
2784     }
2785
2786     shader_glsl_get_sample_function(sampler_type, sample_flags, &sample_function);
2787     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
2788     shader_glsl_add_src_param(ins, &ins->src[2], sample_function.coord_mask, &dx_param);
2789     shader_glsl_add_src_param(ins, &ins->src[3], sample_function.coord_mask, &dy_param);
2790
2791     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, dx_param.param_str, dy_param.param_str, NULL,
2792                                 "%s", coord_param.param_str);
2793 }
2794
2795 static void shader_glsl_texldl(const struct wined3d_shader_instruction *ins)
2796 {
2797     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2798     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2799     const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
2800     glsl_sample_function_t sample_function;
2801     glsl_src_param_t coord_param, lod_param;
2802     DWORD sample_flags = WINED3D_GLSL_SAMPLE_LOD;
2803     DWORD sampler_type;
2804     DWORD sampler_idx;
2805     DWORD swizzle = ins->src[1].swizzle;
2806
2807     sampler_idx = ins->src[1].reg.idx;
2808     sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
2809     if(deviceImpl->stateBlock->textures[sampler_idx] &&
2810        IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2811         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
2812     }
2813     shader_glsl_get_sample_function(sampler_type, sample_flags, &sample_function);
2814     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
2815
2816     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &lod_param);
2817
2818     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD]
2819             && shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
2820     {
2821         /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
2822          * However, they seem to work just fine in fragment shaders as well. */
2823         WARN("Using %s in fragment shader.\n", sample_function.name);
2824     }
2825     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, lod_param.param_str,
2826             "%s", coord_param.param_str);
2827 }
2828
2829 static void shader_glsl_texcoord(const struct wined3d_shader_instruction *ins)
2830 {
2831     /* FIXME: Make this work for more than just 2D textures */
2832     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2833     DWORD write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2834
2835     if (!(ins->ctx->reg_maps->shader_version.major == 1 && ins->ctx->reg_maps->shader_version.minor == 4))
2836     {
2837         char dst_mask[6];
2838
2839         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2840         shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n",
2841                 ins->dst[0].reg.idx, dst_mask);
2842     } else {
2843         DWORD reg = ins->src[0].reg.idx;
2844         DWORD src_mod = ins->src[0].modifiers;
2845         char dst_swizzle[6];
2846
2847         shader_glsl_get_swizzle(&ins->src[0], FALSE, write_mask, dst_swizzle);
2848
2849         if (src_mod == WINED3DSPSM_DZ) {
2850             glsl_src_param_t div_param;
2851             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2852             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &div_param);
2853
2854             if (mask_size > 1) {
2855                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2856             } else {
2857                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2858             }
2859         } else if (src_mod == WINED3DSPSM_DW) {
2860             glsl_src_param_t div_param;
2861             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2862             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &div_param);
2863
2864             if (mask_size > 1) {
2865                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2866             } else {
2867                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2868             }
2869         } else {
2870             shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
2871         }
2872     }
2873 }
2874
2875 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
2876  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2877  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2878 static void shader_glsl_texdp3tex(const struct wined3d_shader_instruction *ins)
2879 {
2880     glsl_src_param_t src0_param;
2881     glsl_sample_function_t sample_function;
2882     DWORD sampler_idx = ins->dst[0].reg.idx;
2883     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2884     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
2885     UINT mask_size;
2886
2887     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2888
2889     /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
2890      * scalar, and projected sampling would require 4.
2891      *
2892      * It is a dependent read - not valid with conditional NP2 textures
2893      */
2894     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
2895     mask_size = shader_glsl_get_write_mask_size(sample_function.coord_mask);
2896
2897     switch(mask_size)
2898     {
2899         case 1:
2900             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
2901                     "dot(gl_TexCoord[%u].xyz, %s)", sampler_idx, src0_param.param_str);
2902             break;
2903
2904         case 2:
2905             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
2906                     "vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0)", sampler_idx, src0_param.param_str);
2907             break;
2908
2909         case 3:
2910             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
2911                     "vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0)", sampler_idx, src0_param.param_str);
2912             break;
2913
2914         default:
2915             FIXME("Unexpected mask size %u\n", mask_size);
2916             break;
2917     }
2918 }
2919
2920 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
2921  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2922 static void shader_glsl_texdp3(const struct wined3d_shader_instruction *ins)
2923 {
2924     glsl_src_param_t src0_param;
2925     DWORD dstreg = ins->dst[0].reg.idx;
2926     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2927     DWORD dst_mask;
2928     unsigned int mask_size;
2929
2930     dst_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2931     mask_size = shader_glsl_get_write_mask_size(dst_mask);
2932     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2933
2934     if (mask_size > 1) {
2935         shader_addline(ins->ctx->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
2936     } else {
2937         shader_addline(ins->ctx->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
2938     }
2939 }
2940
2941 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
2942  * Calculate the depth as dst.x / dst.y   */
2943 static void shader_glsl_texdepth(const struct wined3d_shader_instruction *ins)
2944 {
2945     glsl_dst_param_t dst_param;
2946
2947     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
2948
2949     /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
2950      * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
2951      * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
2952      * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
2953      * >= 1.0 or < 0.0
2954      */
2955     shader_addline(ins->ctx->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n",
2956             dst_param.reg_name, dst_param.reg_name);
2957 }
2958
2959 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
2960  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2961  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2962  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2963  */
2964 static void shader_glsl_texm3x2depth(const struct wined3d_shader_instruction *ins)
2965 {
2966     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2967     DWORD dstreg = ins->dst[0].reg.idx;
2968     glsl_src_param_t src0_param;
2969
2970     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2971
2972     shader_addline(ins->ctx->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
2973     shader_addline(ins->ctx->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
2974 }
2975
2976 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
2977  * Calculate the 1st of a 2-row matrix multiplication. */
2978 static void shader_glsl_texm3x2pad(const struct wined3d_shader_instruction *ins)
2979 {
2980     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2981     DWORD reg = ins->dst[0].reg.idx;
2982     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2983     glsl_src_param_t src0_param;
2984
2985     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2986     shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2987 }
2988
2989 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
2990  * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
2991 static void shader_glsl_texm3x3pad(const struct wined3d_shader_instruction *ins)
2992 {
2993     IWineD3DPixelShaderImpl *shader = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
2994     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2995     DWORD reg = ins->dst[0].reg.idx;
2996     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2997     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2998     glsl_src_param_t src0_param;
2999
3000     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3001     shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + current_state->current_row, reg, src0_param.param_str);
3002     current_state->texcoord_w[current_state->current_row++] = reg;
3003 }
3004
3005 static void shader_glsl_texm3x2tex(const struct wined3d_shader_instruction *ins)
3006 {
3007     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3008     DWORD reg = ins->dst[0].reg.idx;
3009     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3010     glsl_src_param_t src0_param;
3011     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[reg];
3012     glsl_sample_function_t sample_function;
3013
3014     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3015     shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3016
3017     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
3018
3019     /* Sample the texture using the calculated coordinates */
3020     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xy");
3021 }
3022
3023 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
3024  * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
3025 static void shader_glsl_texm3x3tex(const struct wined3d_shader_instruction *ins)
3026 {
3027     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3028     glsl_src_param_t src0_param;
3029     DWORD reg = ins->dst[0].reg.idx;
3030     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
3031     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
3032     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[reg];
3033     glsl_sample_function_t sample_function;
3034
3035     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3036     shader_addline(ins->ctx->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3037
3038     /* Dependent read, not valid with conditional NP2 */
3039     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
3040
3041     /* Sample the texture using the calculated coordinates */
3042     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3043
3044     current_state->current_row = 0;
3045 }
3046
3047 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
3048  * Perform the 3rd row of a 3x3 matrix multiply */
3049 static void shader_glsl_texm3x3(const struct wined3d_shader_instruction *ins)
3050 {
3051     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3052     glsl_src_param_t src0_param;
3053     char dst_mask[6];
3054     DWORD reg = ins->dst[0].reg.idx;
3055     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
3056     SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
3057
3058     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3059
3060     shader_glsl_append_dst(ins->ctx->buffer, ins);
3061     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3062     shader_addline(ins->ctx->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
3063
3064     current_state->current_row = 0;
3065 }
3066
3067 /* Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
3068  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3069 static void shader_glsl_texm3x3spec(const struct wined3d_shader_instruction *ins)
3070 {
3071     IWineD3DPixelShaderImpl *shader = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
3072     DWORD reg = ins->dst[0].reg.idx;
3073     glsl_src_param_t src0_param;
3074     glsl_src_param_t src1_param;
3075     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3076     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3077     WINED3DSAMPLER_TEXTURE_TYPE stype = ins->ctx->reg_maps->sampler_type[reg];
3078     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3079     glsl_sample_function_t sample_function;
3080
3081     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3082     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
3083
3084     /* Perform the last matrix multiply operation */
3085     shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3086     /* Reflection calculation */
3087     shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
3088
3089     /* Dependent read, not valid with conditional NP2 */
3090     shader_glsl_get_sample_function(stype, 0, &sample_function);
3091
3092     /* Sample the texture */
3093     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3094
3095     current_state->current_row = 0;
3096 }
3097
3098 /* Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
3099  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3100 static void shader_glsl_texm3x3vspec(const struct wined3d_shader_instruction *ins)
3101 {
3102     IWineD3DPixelShaderImpl *shader = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
3103     DWORD reg = ins->dst[0].reg.idx;
3104     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3105     SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3106     glsl_src_param_t src0_param;
3107     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3108     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[reg];
3109     glsl_sample_function_t sample_function;
3110
3111     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3112
3113     /* Perform the last matrix multiply operation */
3114     shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
3115
3116     /* Construct the eye-ray vector from w coordinates */
3117     shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
3118             current_state->texcoord_w[0], current_state->texcoord_w[1], reg);
3119     shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
3120
3121     /* Dependent read, not valid with conditional NP2 */
3122     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
3123
3124     /* Sample the texture using the calculated coordinates */
3125     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3126
3127     current_state->current_row = 0;
3128 }
3129
3130 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
3131  * Apply a fake bump map transform.
3132  * texbem is pshader <= 1.3 only, this saves a few version checks
3133  */
3134 static void shader_glsl_texbem(const struct wined3d_shader_instruction *ins)
3135 {
3136     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->ctx->shader;
3137     IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
3138     glsl_sample_function_t sample_function;
3139     glsl_src_param_t coord_param;
3140     WINED3DSAMPLER_TEXTURE_TYPE sampler_type;
3141     DWORD sampler_idx;
3142     DWORD mask;
3143     DWORD flags;
3144     char coord_mask[6];
3145
3146     sampler_idx = ins->dst[0].reg.idx;
3147     flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
3148
3149     sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3150     /* Dependent read, not valid with conditional NP2 */
3151     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
3152     mask = sample_function.coord_mask;
3153
3154     shader_glsl_write_mask_to_str(mask, coord_mask);
3155
3156     /* with projective textures, texbem only divides the static texture coord, not the displacement,
3157          * so we can't let the GL handle this.
3158          */
3159     if (flags & WINED3DTTFF_PROJECTED) {
3160         DWORD div_mask=0;
3161         char coord_div_mask[3];
3162         switch (flags & ~WINED3DTTFF_PROJECTED) {
3163             case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3164             case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
3165             case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
3166             case WINED3DTTFF_COUNT4:
3167             case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
3168         }
3169         shader_glsl_write_mask_to_str(div_mask, coord_div_mask);
3170         shader_addline(ins->ctx->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
3171     }
3172
3173     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &coord_param);
3174
3175     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3176             "T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s", sampler_idx, coord_mask, sampler_idx,
3177             coord_param.param_str, coord_mask);
3178
3179     if (ins->handler_idx == WINED3DSIH_TEXBEML)
3180     {
3181         glsl_src_param_t luminance_param;
3182         glsl_dst_param_t dst_param;
3183
3184         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &luminance_param);
3185         shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3186
3187         shader_addline(ins->ctx->buffer, "%s%s *= (%s * luminancescale%d + luminanceoffset%d);\n",
3188                 dst_param.reg_name, dst_param.mask_str,
3189                 luminance_param.param_str, sampler_idx, sampler_idx);
3190     }
3191 }
3192
3193 static void shader_glsl_bem(const struct wined3d_shader_instruction *ins)
3194 {
3195     glsl_src_param_t src0_param, src1_param;
3196     DWORD sampler_idx = ins->dst[0].reg.idx;
3197
3198     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3199     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3200
3201     shader_glsl_append_dst(ins->ctx->buffer, ins);
3202     shader_addline(ins->ctx->buffer, "%s + bumpenvmat%d * %s);\n",
3203             src0_param.param_str, sampler_idx, src1_param.param_str);
3204 }
3205
3206 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
3207  * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
3208 static void shader_glsl_texreg2ar(const struct wined3d_shader_instruction *ins)
3209 {
3210     glsl_src_param_t src0_param;
3211     DWORD sampler_idx = ins->dst[0].reg.idx;
3212     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3213     glsl_sample_function_t sample_function;
3214
3215     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3216
3217     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
3218     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3219             "%s.wx", src0_param.reg_name);
3220 }
3221
3222 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
3223  * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
3224 static void shader_glsl_texreg2gb(const struct wined3d_shader_instruction *ins)
3225 {
3226     glsl_src_param_t src0_param;
3227     DWORD sampler_idx = ins->dst[0].reg.idx;
3228     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3229     glsl_sample_function_t sample_function;
3230
3231     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3232
3233     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
3234     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3235             "%s.yz", src0_param.reg_name);
3236 }
3237
3238 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
3239  * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
3240 static void shader_glsl_texreg2rgb(const struct wined3d_shader_instruction *ins)
3241 {
3242     glsl_src_param_t src0_param;
3243     DWORD sampler_idx = ins->dst[0].reg.idx;
3244     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3245     glsl_sample_function_t sample_function;
3246
3247     /* Dependent read, not valid with conditional NP2 */
3248     shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
3249     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &src0_param);
3250
3251     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3252             "%s", src0_param.param_str);
3253 }
3254
3255 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
3256  * If any of the first 3 components are < 0, discard this pixel */
3257 static void shader_glsl_texkill(const struct wined3d_shader_instruction *ins)
3258 {
3259     glsl_dst_param_t dst_param;
3260
3261     /* The argument is a destination parameter, and no writemasks are allowed */
3262     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3263     if (ins->ctx->reg_maps->shader_version.major >= 2)
3264     {
3265         /* 2.0 shaders compare all 4 components in texkill */
3266         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
3267     } else {
3268         /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
3269          * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
3270          * 4 components are defined, only the first 3 are used
3271          */
3272         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
3273     }
3274 }
3275
3276 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
3277  * dst = dot2(src0, src1) + src2 */
3278 static void shader_glsl_dp2add(const struct wined3d_shader_instruction *ins)
3279 {
3280     glsl_src_param_t src0_param;
3281     glsl_src_param_t src1_param;
3282     glsl_src_param_t src2_param;
3283     DWORD write_mask;
3284     unsigned int mask_size;
3285
3286     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3287     mask_size = shader_glsl_get_write_mask_size(write_mask);
3288
3289     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3290     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3291     shader_glsl_add_src_param(ins, &ins->src[2], WINED3DSP_WRITEMASK_0, &src2_param);
3292
3293     if (mask_size > 1) {
3294         shader_addline(ins->ctx->buffer, "vec%d(dot(%s, %s) + %s));\n",
3295                 mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
3296     } else {
3297         shader_addline(ins->ctx->buffer, "dot(%s, %s) + %s);\n",
3298                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
3299     }
3300 }
3301
3302 static void shader_glsl_input_pack(IWineD3DPixelShader *iface, struct wined3d_shader_buffer *buffer,
3303         const struct wined3d_shader_signature_element *input_signature, const struct shader_reg_maps *reg_maps,
3304         enum vertexprocessing_mode vertexprocessing)
3305 {
3306     unsigned int i;
3307     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
3308     WORD map = reg_maps->input_registers;
3309
3310     for (i = 0; map; map >>= 1, ++i)
3311     {
3312         const char *semantic_name;
3313         UINT semantic_idx;
3314         char reg_mask[6];
3315
3316         /* Unused */
3317         if (!(map & 1)) continue;
3318
3319         semantic_name = input_signature[i].semantic_name;
3320         semantic_idx = input_signature[i].semantic_idx;
3321         shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3322
3323         if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3324         {
3325             if (semantic_idx < 8 && vertexprocessing == pretransformed)
3326                 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
3327                         This->input_reg_map[i], reg_mask, semantic_idx, reg_mask);
3328             else
3329                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3330                         This->input_reg_map[i], reg_mask, reg_mask);
3331         }
3332         else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3333         {
3334             if (semantic_idx == 0)
3335                 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
3336                         This->input_reg_map[i], reg_mask, reg_mask);
3337             else if (semantic_idx == 1)
3338                 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
3339                         This->input_reg_map[i], reg_mask, reg_mask);
3340             else
3341                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3342                         This->input_reg_map[i], reg_mask, reg_mask);
3343         }
3344         else
3345         {
3346             shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3347                     This->input_reg_map[i], reg_mask, reg_mask);
3348         }
3349     }
3350 }
3351
3352 /*********************************************
3353  * Vertex Shader Specific Code begins here
3354  ********************************************/
3355
3356 static void add_glsl_program_entry(struct shader_glsl_priv *priv, struct glsl_shader_prog_link *entry) {
3357     glsl_program_key_t key;
3358
3359     key.vshader = entry->vshader;
3360     key.pshader = entry->pshader;
3361     key.vs_args = entry->vs_args;
3362     key.ps_args = entry->ps_args;
3363
3364     if (wine_rb_put(&priv->program_lookup, &key, &entry->program_lookup_entry) == -1)
3365     {
3366         ERR("Failed to insert program entry.\n");
3367     }
3368 }
3369
3370 static struct glsl_shader_prog_link *get_glsl_program_entry(struct shader_glsl_priv *priv,
3371         IWineD3DVertexShader *vshader, IWineD3DPixelShader *pshader, struct vs_compile_args *vs_args,
3372         struct ps_compile_args *ps_args) {
3373     struct wine_rb_entry *entry;
3374     glsl_program_key_t key;
3375
3376     key.vshader = vshader;
3377     key.pshader = pshader;
3378     key.vs_args = *vs_args;
3379     key.ps_args = *ps_args;
3380
3381     entry = wine_rb_get(&priv->program_lookup, &key);
3382     return entry ? WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry) : NULL;
3383 }
3384
3385 /* GL locking is done by the caller */
3386 static void delete_glsl_program_entry(struct shader_glsl_priv *priv, const struct wined3d_gl_info *gl_info,
3387         struct glsl_shader_prog_link *entry)
3388 {
3389     glsl_program_key_t key;
3390
3391     key.vshader = entry->vshader;
3392     key.pshader = entry->pshader;
3393     key.vs_args = entry->vs_args;
3394     key.ps_args = entry->ps_args;
3395     wine_rb_remove(&priv->program_lookup, &key);
3396
3397     GL_EXTCALL(glDeleteObjectARB(entry->programId));
3398     if (entry->vshader) list_remove(&entry->vshader_entry);
3399     if (entry->pshader) list_remove(&entry->pshader_entry);
3400     HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
3401     HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
3402     HeapFree(GetProcessHeap(), 0, entry);
3403 }
3404
3405 static void handle_ps3_input(struct wined3d_shader_buffer *buffer, const struct wined3d_gl_info *gl_info, const DWORD *map,
3406         const struct wined3d_shader_signature_element *input_signature, const struct shader_reg_maps *reg_maps_in,
3407         const struct wined3d_shader_signature_element *output_signature, const struct shader_reg_maps *reg_maps_out)
3408 {
3409     unsigned int i, j;
3410     const char *semantic_name_in, *semantic_name_out;
3411     UINT semantic_idx_in, semantic_idx_out;
3412     DWORD *set;
3413     DWORD in_idx;
3414     unsigned int in_count = vec4_varyings(3, gl_info);
3415     char reg_mask[6], reg_mask_out[6];
3416     char destination[50];
3417     WORD input_map, output_map;
3418
3419     set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
3420
3421     if (!output_signature)
3422     {
3423         /* Save gl_FrontColor & gl_FrontSecondaryColor before overwriting them. */
3424         shader_addline(buffer, "vec4 front_color = gl_FrontColor;\n");
3425         shader_addline(buffer, "vec4 front_secondary_color = gl_FrontSecondaryColor;\n");
3426     }
3427
3428     input_map = reg_maps_in->input_registers;
3429     for (i = 0; input_map; input_map >>= 1, ++i)
3430     {
3431         if (!(input_map & 1)) continue;
3432
3433         in_idx = map[i];
3434         if (in_idx >= (in_count + 2)) {
3435             FIXME("More input varyings declared than supported, expect issues\n");
3436             continue;
3437         }
3438         else if (map[i] == ~0U)
3439         {
3440             /* Declared, but not read register */
3441             continue;
3442         }
3443
3444         if (in_idx == in_count) {
3445             sprintf(destination, "gl_FrontColor");
3446         } else if (in_idx == in_count + 1) {
3447             sprintf(destination, "gl_FrontSecondaryColor");
3448         } else {
3449             sprintf(destination, "IN[%u]", in_idx);
3450         }
3451
3452         semantic_name_in = input_signature[i].semantic_name;
3453         semantic_idx_in = input_signature[i].semantic_idx;
3454         set[map[i]] = input_signature[i].mask;
3455         shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3456
3457         if (!output_signature)
3458         {
3459             if (shader_match_semantic(semantic_name_in, WINED3DDECLUSAGE_COLOR))
3460             {
3461                 if (semantic_idx_in == 0)
3462                     shader_addline(buffer, "%s%s = front_color%s;\n",
3463                             destination, reg_mask, reg_mask);
3464                 else if (semantic_idx_in == 1)
3465                     shader_addline(buffer, "%s%s = front_secondary_color%s;\n",
3466                             destination, reg_mask, reg_mask);
3467                 else
3468                     shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3469                             destination, reg_mask, reg_mask);
3470             }
3471             else if (shader_match_semantic(semantic_name_in, WINED3DDECLUSAGE_TEXCOORD))
3472             {
3473                 if (semantic_idx_in < 8)
3474                 {
3475                     shader_addline(buffer, "%s%s = gl_TexCoord[%u]%s;\n",
3476                             destination, reg_mask, semantic_idx_in, reg_mask);
3477                 }
3478                 else
3479                 {
3480                     shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3481                             destination, reg_mask, reg_mask);
3482                 }
3483             }
3484             else if (shader_match_semantic(semantic_name_in, WINED3DDECLUSAGE_FOG))
3485             {
3486                 shader_addline(buffer, "%s%s = vec4(gl_FogFragCoord, 0.0, 0.0, 0.0)%s;\n",
3487                         destination, reg_mask, reg_mask);
3488             }
3489             else
3490             {
3491                 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3492                         destination, reg_mask, reg_mask);
3493             }
3494         } else {
3495             BOOL found = FALSE;
3496
3497             output_map = reg_maps_out->output_registers;
3498             for (j = 0; output_map; output_map >>= 1, ++j)
3499             {
3500                 if (!(output_map & 1)) continue;
3501
3502                 semantic_name_out = output_signature[j].semantic_name;
3503                 semantic_idx_out = output_signature[j].semantic_idx;
3504                 shader_glsl_write_mask_to_str(output_signature[j].mask, reg_mask_out);
3505
3506                 if (semantic_idx_in == semantic_idx_out
3507                         && !strcmp(semantic_name_in, semantic_name_out))
3508                 {
3509                     shader_addline(buffer, "%s%s = OUT[%u]%s;\n",
3510                             destination, reg_mask, j, reg_mask);
3511                     found = TRUE;
3512                 }
3513             }
3514             if(!found) {
3515                 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3516                                destination, reg_mask, reg_mask);
3517             }
3518         }
3519     }
3520
3521     /* This is solely to make the compiler / linker happy and avoid warning about undefined
3522      * varyings. It shouldn't result in any real code executed on the GPU, since all read
3523      * input varyings are assigned above, if the optimizer works properly.
3524      */
3525     for(i = 0; i < in_count + 2; i++) {
3526         if (set[i] && set[i] != WINED3DSP_WRITEMASK_ALL)
3527         {
3528             unsigned int size = 0;
3529             memset(reg_mask, 0, sizeof(reg_mask));
3530             if(!(set[i] & WINED3DSP_WRITEMASK_0)) {
3531                 reg_mask[size] = 'x';
3532                 size++;
3533             }
3534             if(!(set[i] & WINED3DSP_WRITEMASK_1)) {
3535                 reg_mask[size] = 'y';
3536                 size++;
3537             }
3538             if(!(set[i] & WINED3DSP_WRITEMASK_2)) {
3539                 reg_mask[size] = 'z';
3540                 size++;
3541             }
3542             if(!(set[i] & WINED3DSP_WRITEMASK_3)) {
3543                 reg_mask[size] = 'w';
3544                 size++;
3545             }
3546
3547             if (i == in_count) {
3548                 sprintf(destination, "gl_FrontColor");
3549             } else if (i == in_count + 1) {
3550                 sprintf(destination, "gl_FrontSecondaryColor");
3551             } else {
3552                 sprintf(destination, "IN[%u]", i);
3553             }
3554
3555             if (size == 1) {
3556                 shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
3557             } else {
3558                 shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
3559             }
3560         }
3561     }
3562
3563     HeapFree(GetProcessHeap(), 0, set);
3564 }
3565
3566 /* GL locking is done by the caller */
3567 static GLhandleARB generate_param_reorder_function(struct wined3d_shader_buffer *buffer,
3568         IWineD3DVertexShader *vertexshader, IWineD3DPixelShader *pixelshader, const struct wined3d_gl_info *gl_info)
3569 {
3570     GLhandleARB ret = 0;
3571     IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) vertexshader;
3572     IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pixelshader;
3573     IWineD3DDeviceImpl *device;
3574     DWORD vs_major = vs->baseShader.reg_maps.shader_version.major;
3575     DWORD ps_major = ps ? ps->baseShader.reg_maps.shader_version.major : 0;
3576     unsigned int i;
3577     const char *semantic_name;
3578     UINT semantic_idx;
3579     char reg_mask[6];
3580     const struct wined3d_shader_signature_element *output_signature;
3581
3582     shader_buffer_clear(buffer);
3583
3584     shader_addline(buffer, "#version 120\n");
3585
3586     if(vs_major < 3 && ps_major < 3) {
3587         /* That one is easy: The vertex shader writes to the builtin varyings, the pixel shader reads from them.
3588          * Take care about the texcoord .w fixup though if we're using the fixed function fragment pipeline
3589          */
3590         device = (IWineD3DDeviceImpl *) vs->baseShader.device;
3591         if ((gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W)
3592                 && ps_major == 0 && vs_major > 0 && !device->frag_pipe->ffp_proj_control)
3593         {
3594             shader_addline(buffer, "void order_ps_input() {\n");
3595             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
3596                 if(vs->baseShader.reg_maps.texcoord_mask[i] != 0 &&
3597                    vs->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
3598                     shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", i);
3599                 }
3600             }
3601             shader_addline(buffer, "}\n");
3602         } else {
3603             shader_addline(buffer, "void order_ps_input() { /* do nothing */ }\n");
3604         }
3605     } else if(ps_major < 3 && vs_major >= 3) {
3606         WORD map = vs->baseShader.reg_maps.output_registers;
3607
3608         /* The vertex shader writes to its own varyings, the pixel shader needs them in the builtin ones */
3609         output_signature = vs->output_signature;
3610
3611         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3612         for (i = 0; map; map >>= 1, ++i)
3613         {
3614             DWORD write_mask;
3615
3616             if (!(map & 1)) continue;
3617
3618             semantic_name = output_signature[i].semantic_name;
3619             semantic_idx = output_signature[i].semantic_idx;
3620             write_mask = output_signature[i].mask;
3621             shader_glsl_write_mask_to_str(write_mask, reg_mask);
3622
3623             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3624             {
3625                 if (semantic_idx == 0)
3626                     shader_addline(buffer, "gl_FrontColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3627                 else if (semantic_idx == 1)
3628                     shader_addline(buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3629             }
3630             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3631             {
3632                 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3633             }
3634             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3635             {
3636                 if (semantic_idx < 8)
3637                 {
3638                     if (!(gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) || ps_major > 0)
3639                         write_mask |= WINED3DSP_WRITEMASK_3;
3640
3641                     shader_addline(buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
3642                             semantic_idx, reg_mask, i, reg_mask);
3643                     if (!(write_mask & WINED3DSP_WRITEMASK_3))
3644                         shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", semantic_idx);
3645                 }
3646             }
3647             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3648             {
3649                 shader_addline(buffer, "gl_PointSize = OUT[%u].x;\n", i);
3650             }
3651             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3652             {
3653                 shader_addline(buffer, "gl_FogFragCoord = OUT[%u].%c;\n", i, reg_mask[1]);
3654             }
3655         }
3656         shader_addline(buffer, "}\n");
3657
3658     } else if(ps_major >= 3 && vs_major >= 3) {
3659         WORD map = vs->baseShader.reg_maps.output_registers;
3660
3661         output_signature = vs->output_signature;
3662
3663         /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3664         shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(3, gl_info));
3665         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3666
3667         /* First, sort out position and point size. Those are not passed to the pixel shader */
3668         for (i = 0; map; map >>= 1, ++i)
3669         {
3670             if (!(map & 1)) continue;
3671
3672             semantic_name = output_signature[i].semantic_name;
3673             shader_glsl_write_mask_to_str(output_signature[i].mask, reg_mask);
3674
3675             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3676             {
3677                 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3678             }
3679             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3680             {
3681                 shader_addline(buffer, "gl_PointSize = OUT[%u].x;\n", i);
3682             }
3683         }
3684
3685         /* Then, fix the pixel shader input */
3686         handle_ps3_input(buffer, gl_info, ps->input_reg_map, ps->input_signature,
3687                 &ps->baseShader.reg_maps, output_signature, &vs->baseShader.reg_maps);
3688
3689         shader_addline(buffer, "}\n");
3690     } else if(ps_major >= 3 && vs_major < 3) {
3691         shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(3, gl_info));
3692         shader_addline(buffer, "void order_ps_input() {\n");
3693         /* The vertex shader wrote to the builtin varyings. There is no need to figure out position and
3694          * point size, but we depend on the optimizers kindness to find out that the pixel shader doesn't
3695          * read gl_TexCoord and gl_ColorX, otherwise we'll run out of varyings
3696          */
3697         handle_ps3_input(buffer, gl_info, ps->input_reg_map, ps->input_signature,
3698                 &ps->baseShader.reg_maps, NULL, NULL);
3699         shader_addline(buffer, "}\n");
3700     } else {
3701         ERR("Unexpected vertex and pixel shader version condition: vs: %d, ps: %d\n", vs_major, ps_major);
3702     }
3703
3704     ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3705     checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3706     GL_EXTCALL(glShaderSourceARB(ret, 1, (const char**)&buffer->buffer, NULL));
3707     checkGLcall("glShaderSourceARB(ret, 1, &buffer->buffer, NULL)");
3708     GL_EXTCALL(glCompileShaderARB(ret));
3709     checkGLcall("glCompileShaderARB(ret)");
3710
3711     return ret;
3712 }
3713
3714 /* GL locking is done by the caller */
3715 static void hardcode_local_constants(IWineD3DBaseShaderImpl *shader, const struct wined3d_gl_info *gl_info,
3716         GLhandleARB programId, char prefix)
3717 {
3718     const local_constant *lconst;
3719     GLint tmp_loc;
3720     const float *value;
3721     char glsl_name[8];
3722
3723     LIST_FOR_EACH_ENTRY(lconst, &shader->baseShader.constantsF, local_constant, entry) {
3724         value = (const float *)lconst->value;
3725         snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
3726         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3727         GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
3728     }
3729     checkGLcall("Hardcoding local constants");
3730 }
3731
3732 /* GL locking is done by the caller */
3733 static GLuint shader_glsl_generate_pshader(const struct wined3d_context *context,
3734         struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *This,
3735         const struct ps_compile_args *args, struct ps_np2fixup_info *np2fixup_info)
3736 {
3737     const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3738     const struct wined3d_gl_info *gl_info = context->gl_info;
3739     CONST DWORD *function = This->baseShader.function;
3740     struct shader_glsl_ctx_priv priv_ctx;
3741
3742     /* Create the hw GLSL shader object and assign it as the shader->prgId */
3743     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3744
3745     memset(&priv_ctx, 0, sizeof(priv_ctx));
3746     priv_ctx.cur_ps_args = args;
3747     priv_ctx.cur_np2fixup_info = np2fixup_info;
3748
3749     shader_addline(buffer, "#version 120\n");
3750
3751     if (gl_info->supported[ARB_SHADER_TEXTURE_LOD] && reg_maps->usestexldd)
3752     {
3753         shader_addline(buffer, "#extension GL_ARB_shader_texture_lod : enable\n");
3754     }
3755     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
3756     {
3757         /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
3758          * drivers write a warning if we don't do so
3759          */
3760         shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
3761     }
3762
3763     /* Base Declarations */
3764     shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShader *)This, reg_maps, &priv_ctx);
3765
3766     /* Pack 3.0 inputs */
3767     if (reg_maps->shader_version.major >= 3 && args->vp_mode != vertexshader)
3768     {
3769         shader_glsl_input_pack((IWineD3DPixelShader *) This, buffer, This->input_signature, reg_maps, args->vp_mode);
3770     }
3771
3772     /* Base Shader Body */
3773     shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
3774
3775     /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
3776     if (reg_maps->shader_version.major < 2)
3777     {
3778         /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
3779         shader_addline(buffer, "gl_FragData[0] = R0;\n");
3780     }
3781
3782     if (args->srgb_correction)
3783     {
3784         shader_addline(buffer, "tmp0.xyz = pow(gl_FragData[0].xyz, vec3(srgb_const0.x));\n");
3785         shader_addline(buffer, "tmp0.xyz = tmp0.xyz * vec3(srgb_const0.y) - vec3(srgb_const0.z);\n");
3786         shader_addline(buffer, "tmp1.xyz = gl_FragData[0].xyz * vec3(srgb_const0.w);\n");
3787         shader_addline(buffer, "bvec3 srgb_compare = lessThan(gl_FragData[0].xyz, vec3(srgb_const1.x));\n");
3788         shader_addline(buffer, "gl_FragData[0].xyz = mix(tmp0.xyz, tmp1.xyz, vec3(srgb_compare));\n");
3789         shader_addline(buffer, "gl_FragData[0] = clamp(gl_FragData[0], 0.0, 1.0);\n");
3790     }
3791     /* Pixel shader < 3.0 do not replace the fog stage.
3792      * This implements linear fog computation and blending.
3793      * TODO: non linear fog
3794      * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
3795      * -1/(e-s) and e/(e-s) respectively.
3796      */
3797     if (reg_maps->shader_version.major < 3)
3798     {
3799         switch(args->fog) {
3800             case FOG_OFF: break;
3801             case FOG_LINEAR:
3802                 shader_addline(buffer, "float fogstart = -1.0 / (gl_Fog.end - gl_Fog.start);\n");
3803                 shader_addline(buffer, "float fogend = gl_Fog.end * -fogstart;\n");
3804                 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * fogstart + fogend, 0.0, 1.0);\n");
3805                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
3806                 break;
3807             case FOG_EXP:
3808                 /* Fog = e^(-gl_Fog.density * gl_FogFragCoord) */
3809                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_FogFragCoord);\n");
3810                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
3811                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
3812                 break;
3813             case FOG_EXP2:
3814                 /* Fog = e^(-(gl_Fog.density * gl_FogFragCoord)^2) */
3815                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_Fog.density * gl_FogFragCoord * gl_FogFragCoord);\n");
3816                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
3817                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
3818                 break;
3819         }
3820     }
3821
3822     shader_addline(buffer, "}\n");
3823
3824     TRACE("Compiling shader object %u\n", shader_obj);
3825     GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
3826     GL_EXTCALL(glCompileShaderARB(shader_obj));
3827     print_glsl_info_log(gl_info, shader_obj);
3828
3829     /* Store the shader object */
3830     return shader_obj;
3831 }
3832
3833 /* GL locking is done by the caller */
3834 static GLuint shader_glsl_generate_vshader(const struct wined3d_context *context,
3835         struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *This,
3836         const struct vs_compile_args *args)
3837 {
3838     const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3839     const struct wined3d_gl_info *gl_info = context->gl_info;
3840     CONST DWORD *function = This->baseShader.function;
3841     struct shader_glsl_ctx_priv priv_ctx;
3842
3843     /* Create the hw GLSL shader program and assign it as the shader->prgId */
3844     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3845
3846     shader_addline(buffer, "#version 120\n");
3847
3848     memset(&priv_ctx, 0, sizeof(priv_ctx));
3849     priv_ctx.cur_vs_args = args;
3850
3851     /* Base Declarations */
3852     shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShader *)This, reg_maps, &priv_ctx);
3853
3854     /* Base Shader Body */
3855     shader_generate_main((IWineD3DBaseShader*)This, buffer, reg_maps, function, &priv_ctx);
3856
3857     /* Unpack 3.0 outputs */
3858     if (reg_maps->shader_version.major >= 3) shader_addline(buffer, "order_ps_input(OUT);\n");
3859     else shader_addline(buffer, "order_ps_input();\n");
3860
3861     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3862      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3863      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3864      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3865      */
3866     if(args->fog_src == VS_FOG_Z) {
3867         shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
3868     } else if (!reg_maps->fog) {
3869         shader_addline(buffer, "gl_FogFragCoord = 0.0;\n");
3870     }
3871
3872     /* Write the final position.
3873      *
3874      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3875      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3876      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3877      * contains 1.0 to allow a mad.
3878      */
3879     shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
3880     shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
3881     if(args->clip_enabled) {
3882         shader_addline(buffer, "gl_ClipVertex = gl_Position;\n");
3883     }
3884
3885     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3886      *
3887      * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
3888      * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
3889      * which is the same as z = z * 2 - w.
3890      */
3891     shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
3892
3893     shader_addline(buffer, "}\n");
3894
3895     TRACE("Compiling shader object %u\n", shader_obj);
3896     GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
3897     GL_EXTCALL(glCompileShaderARB(shader_obj));
3898     print_glsl_info_log(gl_info, shader_obj);
3899
3900     return shader_obj;
3901 }
3902
3903 static GLhandleARB find_glsl_pshader(const struct wined3d_context *context,
3904         struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *shader,
3905         const struct ps_compile_args *args, const struct ps_np2fixup_info **np2fixup_info)
3906 {
3907     UINT i;
3908     DWORD new_size;
3909     struct glsl_ps_compiled_shader *new_array;
3910     struct glsl_pshader_private    *shader_data;
3911     struct ps_np2fixup_info        *np2fixup = NULL;
3912     GLhandleARB ret;
3913
3914     if (!shader->baseShader.backend_data)
3915     {
3916         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
3917         if (!shader->baseShader.backend_data)
3918         {
3919             ERR("Failed to allocate backend data.\n");
3920             return 0;
3921         }
3922     }
3923     shader_data = shader->baseShader.backend_data;
3924
3925     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
3926      * so a linear search is more performant than a hashmap or a binary search
3927      * (cache coherency etc)
3928      */
3929     for(i = 0; i < shader_data->num_gl_shaders; i++) {
3930         if(memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)) == 0) {
3931             if(args->np2_fixup) *np2fixup_info = &shader_data->gl_shaders[i].np2fixup;
3932             return shader_data->gl_shaders[i].prgId;
3933         }
3934     }
3935
3936     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
3937     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
3938         if (shader_data->num_gl_shaders)
3939         {
3940             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
3941             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
3942                                     new_size * sizeof(*shader_data->gl_shaders));
3943         } else {
3944             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
3945             new_size = 1;
3946         }
3947
3948         if(!new_array) {
3949             ERR("Out of memory\n");
3950             return 0;
3951         }
3952         shader_data->gl_shaders = new_array;
3953         shader_data->shader_array_size = new_size;
3954     }
3955
3956     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
3957
3958     memset(&shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup, 0, sizeof(struct ps_np2fixup_info));
3959     if (args->np2_fixup) np2fixup = &shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup;
3960
3961     pixelshader_update_samplers(&shader->baseShader.reg_maps,
3962             ((IWineD3DDeviceImpl *)shader->baseShader.device)->stateBlock->textures);
3963
3964     shader_buffer_clear(buffer);
3965     ret = shader_glsl_generate_pshader(context, buffer, shader, args, np2fixup);
3966     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
3967     *np2fixup_info = np2fixup;
3968
3969     return ret;
3970 }
3971
3972 static inline BOOL vs_args_equal(const struct vs_compile_args *stored, const struct vs_compile_args *new,
3973                                  const DWORD use_map) {
3974     if((stored->swizzle_map & use_map) != new->swizzle_map) return FALSE;
3975     if((stored->clip_enabled) != new->clip_enabled) return FALSE;
3976     return stored->fog_src == new->fog_src;
3977 }
3978
3979 static GLhandleARB find_glsl_vshader(const struct wined3d_context *context,
3980         struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *shader,
3981         const struct vs_compile_args *args)
3982 {
3983     UINT i;
3984     DWORD new_size;
3985     struct glsl_vs_compiled_shader *new_array;
3986     DWORD use_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.use_map;
3987     struct glsl_vshader_private *shader_data;
3988     GLhandleARB ret;
3989
3990     if (!shader->baseShader.backend_data)
3991     {
3992         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
3993         if (!shader->baseShader.backend_data)
3994         {
3995             ERR("Failed to allocate backend data.\n");
3996             return 0;
3997         }
3998     }
3999     shader_data = shader->baseShader.backend_data;
4000
4001     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4002      * so a linear search is more performant than a hashmap or a binary search
4003      * (cache coherency etc)
4004      */
4005     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4006         if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map)) {
4007             return shader_data->gl_shaders[i].prgId;
4008         }
4009     }
4010
4011     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4012
4013     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4014         if (shader_data->num_gl_shaders)
4015         {
4016             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4017             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4018                                     new_size * sizeof(*shader_data->gl_shaders));
4019         } else {
4020             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4021             new_size = 1;
4022         }
4023
4024         if(!new_array) {
4025             ERR("Out of memory\n");
4026             return 0;
4027         }
4028         shader_data->gl_shaders = new_array;
4029         shader_data->shader_array_size = new_size;
4030     }
4031
4032     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4033
4034     shader_buffer_clear(buffer);
4035     ret = shader_glsl_generate_vshader(context, buffer, shader, args);
4036     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4037
4038     return ret;
4039 }
4040
4041 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
4042  * It sets the programId on the current StateBlock (because it should be called
4043  * inside of the DrawPrimitive() part of the render loop).
4044  *
4045  * If a program for the given combination does not exist, create one, and store
4046  * the program in the hash table.  If it creates a program, it will link the
4047  * given objects, too.
4048  */
4049
4050 /* GL locking is done by the caller */
4051 static void set_glsl_shader_program(const struct wined3d_context *context,
4052         IWineD3DDeviceImpl *device, BOOL use_ps, BOOL use_vs)
4053 {
4054     IWineD3DVertexShader *vshader = use_vs ? device->stateBlock->vertexShader : NULL;
4055     IWineD3DPixelShader *pshader = use_ps ? device->stateBlock->pixelShader : NULL;
4056     const struct wined3d_gl_info *gl_info = context->gl_info;
4057     struct shader_glsl_priv *priv = device->shader_priv;
4058     struct glsl_shader_prog_link *entry    = NULL;
4059     GLhandleARB programId                  = 0;
4060     GLhandleARB reorder_shader_id          = 0;
4061     unsigned int i;
4062     char glsl_name[8];
4063     struct ps_compile_args ps_compile_args;
4064     struct vs_compile_args vs_compile_args;
4065
4066     if (vshader) find_vs_compile_args((IWineD3DVertexShaderImpl *)vshader, device->stateBlock, &vs_compile_args);
4067     if (pshader) find_ps_compile_args((IWineD3DPixelShaderImpl *)pshader, device->stateBlock, &ps_compile_args);
4068
4069     entry = get_glsl_program_entry(priv, vshader, pshader, &vs_compile_args, &ps_compile_args);
4070     if (entry) {
4071         priv->glsl_program = entry;
4072         return;
4073     }
4074
4075     /* If we get to this point, then no matching program exists, so we create one */
4076     programId = GL_EXTCALL(glCreateProgramObjectARB());
4077     TRACE("Created new GLSL shader program %u\n", programId);
4078
4079     /* Create the entry */
4080     entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
4081     entry->programId = programId;
4082     entry->vshader = vshader;
4083     entry->pshader = pshader;
4084     entry->vs_args = vs_compile_args;
4085     entry->ps_args = ps_compile_args;
4086     entry->constant_version = 0;
4087     entry->np2Fixup_info = NULL;
4088     /* Add the hash table entry */
4089     add_glsl_program_entry(priv, entry);
4090
4091     /* Set the current program */
4092     priv->glsl_program = entry;
4093
4094     /* Attach GLSL vshader */
4095     if (vshader)
4096     {
4097         GLhandleARB vshader_id = find_glsl_vshader(context, &priv->shader_buffer,
4098                 (IWineD3DVertexShaderImpl *)vshader, &vs_compile_args);
4099         WORD map = ((IWineD3DBaseShaderImpl *)vshader)->baseShader.reg_maps.input_registers;
4100         char tmp_name[10];
4101
4102         reorder_shader_id = generate_param_reorder_function(&priv->shader_buffer, vshader, pshader, gl_info);
4103         TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
4104         GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
4105         checkGLcall("glAttachObjectARB");
4106         /* Flag the reorder function for deletion, then it will be freed automatically when the program
4107          * is destroyed
4108          */
4109         GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
4110
4111         TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
4112         GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
4113         checkGLcall("glAttachObjectARB");
4114
4115         /* Bind vertex attributes to a corresponding index number to match
4116          * the same index numbers as ARB_vertex_programs (makes loading
4117          * vertex attributes simpler).  With this method, we can use the
4118          * exact same code to load the attributes later for both ARB and
4119          * GLSL shaders.
4120          *
4121          * We have to do this here because we need to know the Program ID
4122          * in order to make the bindings work, and it has to be done prior
4123          * to linking the GLSL program. */
4124         for (i = 0; map; map >>= 1, ++i)
4125         {
4126             if (!(map & 1)) continue;
4127
4128             snprintf(tmp_name, sizeof(tmp_name), "attrib%u", i);
4129             GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
4130         }
4131         checkGLcall("glBindAttribLocationARB");
4132
4133         list_add_head(&((IWineD3DBaseShaderImpl *)vshader)->baseShader.linked_programs, &entry->vshader_entry);
4134     }
4135
4136     /* Attach GLSL pshader */
4137     if (pshader)
4138     {
4139         GLhandleARB pshader_id = find_glsl_pshader(context, &priv->shader_buffer,
4140                 (IWineD3DPixelShaderImpl *)pshader, &ps_compile_args, &entry->np2Fixup_info);
4141         TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
4142         GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
4143         checkGLcall("glAttachObjectARB");
4144
4145         list_add_head(&((IWineD3DBaseShaderImpl *)pshader)->baseShader.linked_programs, &entry->pshader_entry);
4146     }
4147
4148     /* Link the program */
4149     TRACE("Linking GLSL shader program %u\n", programId);
4150     GL_EXTCALL(glLinkProgramARB(programId));
4151     print_glsl_info_log(gl_info, programId);
4152
4153     entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4154             sizeof(GLhandleARB) * gl_info->limits.glsl_vs_float_constants);
4155     for (i = 0; i < gl_info->limits.glsl_vs_float_constants; ++i)
4156     {
4157         snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
4158         entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4159     }
4160     for (i = 0; i < MAX_CONST_I; ++i)
4161     {
4162         snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
4163         entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4164     }
4165     entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4166             sizeof(GLhandleARB) * gl_info->limits.glsl_ps_float_constants);
4167     for (i = 0; i < gl_info->limits.glsl_ps_float_constants; ++i)
4168     {
4169         snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
4170         entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4171     }
4172     for (i = 0; i < MAX_CONST_I; ++i)
4173     {
4174         snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
4175         entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4176     }
4177
4178     if(pshader) {
4179         char name[32];
4180
4181         for(i = 0; i < MAX_TEXTURES; i++) {
4182             sprintf(name, "bumpenvmat%u", i);
4183             entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4184             sprintf(name, "luminancescale%u", i);
4185             entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4186             sprintf(name, "luminanceoffset%u", i);
4187             entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4188         }
4189
4190         if (ps_compile_args.np2_fixup) {
4191             if (entry->np2Fixup_info) {
4192                 entry->np2Fixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "PsamplerNP2Fixup"));
4193             } else {
4194                 FIXME("NP2 texcoord fixup needed for this pixelshader, but no fixup uniform found.\n");
4195             }
4196         }
4197     }
4198
4199     entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
4200     entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
4201     checkGLcall("Find glsl program uniform locations");
4202
4203     if (pshader
4204             && ((IWineD3DPixelShaderImpl *)pshader)->baseShader.reg_maps.shader_version.major >= 3
4205             && ((IWineD3DPixelShaderImpl *)pshader)->declared_in_count > vec4_varyings(3, gl_info))
4206     {
4207         TRACE("Shader %d needs vertex color clamping disabled\n", programId);
4208         entry->vertex_color_clamp = GL_FALSE;
4209     } else {
4210         entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
4211     }
4212
4213     /* Set the shader to allow uniform loading on it */
4214     GL_EXTCALL(glUseProgramObjectARB(programId));
4215     checkGLcall("glUseProgramObjectARB(programId)");
4216
4217     /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
4218      * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
4219      * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
4220      * vertex shader with fixed function pixel processing is used we make sure that the card
4221      * supports enough samplers to allow the max number of vertex samplers with all possible
4222      * fixed function fragment processing setups. So once the program is linked these samplers
4223      * won't change.
4224      */
4225     if (vshader) shader_glsl_load_vsamplers(gl_info, device->texUnitMap, programId);
4226     if (pshader) shader_glsl_load_psamplers(gl_info, device->texUnitMap, programId);
4227
4228     /* If the local constants do not have to be loaded with the environment constants,
4229      * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
4230      * later
4231      */
4232     if(pshader && !((IWineD3DPixelShaderImpl*)pshader)->baseShader.load_local_constsF) {
4233         hardcode_local_constants((IWineD3DBaseShaderImpl *) pshader, gl_info, programId, 'P');
4234     }
4235     if(vshader && !((IWineD3DVertexShaderImpl*)vshader)->baseShader.load_local_constsF) {
4236         hardcode_local_constants((IWineD3DBaseShaderImpl *) vshader, gl_info, programId, 'V');
4237     }
4238 }
4239
4240 /* GL locking is done by the caller */
4241 static GLhandleARB create_glsl_blt_shader(const struct wined3d_gl_info *gl_info, enum tex_types tex_type)
4242 {
4243     GLhandleARB program_id;
4244     GLhandleARB vshader_id, pshader_id;
4245     static const char *blt_vshader[] =
4246     {
4247         "#version 120\n"
4248         "void main(void)\n"
4249         "{\n"
4250         "    gl_Position = gl_Vertex;\n"
4251         "    gl_FrontColor = vec4(1.0);\n"
4252         "    gl_TexCoord[0] = gl_MultiTexCoord0;\n"
4253         "}\n"
4254     };
4255
4256     static const char *blt_pshaders[tex_type_count] =
4257     {
4258         /* tex_1d */
4259         NULL,
4260         /* tex_2d */
4261         "#version 120\n"
4262         "uniform sampler2D sampler;\n"
4263         "void main(void)\n"
4264         "{\n"
4265         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4266         "}\n",
4267         /* tex_3d */
4268         NULL,
4269         /* tex_cube */
4270         "#version 120\n"
4271         "uniform samplerCube sampler;\n"
4272         "void main(void)\n"
4273         "{\n"
4274         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4275         "}\n",
4276         /* tex_rect */
4277         "#version 120\n"
4278         "#extension GL_ARB_texture_rectangle : enable\n"
4279         "uniform sampler2DRect sampler;\n"
4280         "void main(void)\n"
4281         "{\n"
4282         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4283         "}\n",
4284     };
4285
4286     if (!blt_pshaders[tex_type])
4287     {
4288         FIXME("tex_type %#x not supported\n", tex_type);
4289         tex_type = tex_2d;
4290     }
4291
4292     vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4293     GL_EXTCALL(glShaderSourceARB(vshader_id, 1, blt_vshader, NULL));
4294     GL_EXTCALL(glCompileShaderARB(vshader_id));
4295
4296     pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
4297     GL_EXTCALL(glShaderSourceARB(pshader_id, 1, &blt_pshaders[tex_type], NULL));
4298     GL_EXTCALL(glCompileShaderARB(pshader_id));
4299
4300     program_id = GL_EXTCALL(glCreateProgramObjectARB());
4301     GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
4302     GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
4303     GL_EXTCALL(glLinkProgramARB(program_id));
4304
4305     print_glsl_info_log(gl_info, program_id);
4306
4307     /* Once linked we can mark the shaders for deletion. They will be deleted once the program
4308      * is destroyed
4309      */
4310     GL_EXTCALL(glDeleteObjectARB(vshader_id));
4311     GL_EXTCALL(glDeleteObjectARB(pshader_id));
4312     return program_id;
4313 }
4314
4315 /* GL locking is done by the caller */
4316 static void shader_glsl_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4317 {
4318     IWineD3DDeviceImpl *device = ((IWineD3DSurfaceImpl *)context->surface)->resource.wineD3DDevice;
4319     const struct wined3d_gl_info *gl_info = context->gl_info;
4320     struct shader_glsl_priv *priv = device->shader_priv;
4321     GLhandleARB program_id = 0;
4322     GLenum old_vertex_color_clamp, current_vertex_color_clamp;
4323
4324     old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4325
4326     if (useVS || usePS) set_glsl_shader_program(context, device, usePS, useVS);
4327     else priv->glsl_program = NULL;
4328
4329     current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4330
4331     if (old_vertex_color_clamp != current_vertex_color_clamp)
4332     {
4333         if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4334         {
4335             GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
4336             checkGLcall("glClampColorARB");
4337         }
4338         else
4339         {
4340             FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4341         }
4342     }
4343
4344     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4345     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4346     GL_EXTCALL(glUseProgramObjectARB(program_id));
4347     checkGLcall("glUseProgramObjectARB");
4348
4349     /* In case that NP2 texcoord fixup data is found for the selected program, trigger a reload of the
4350      * constants. This has to be done because it can't be guaranteed that sampler() (from state.c) is
4351      * called between selecting the shader and using it, which results in wrong fixup for some frames. */
4352     if (priv->glsl_program && priv->glsl_program->np2Fixup_info)
4353     {
4354         shader_glsl_load_np2fixup_constants((IWineD3DDevice *)device, usePS, useVS);
4355     }
4356 }
4357
4358 /* GL locking is done by the caller */
4359 static void shader_glsl_select_depth_blt(IWineD3DDevice *iface, enum tex_types tex_type) {
4360     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4361     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4362     struct shader_glsl_priv *priv = This->shader_priv;
4363     GLhandleARB *blt_program = &priv->depth_blt_program[tex_type];
4364
4365     if (!*blt_program) {
4366         GLint loc;
4367         *blt_program = create_glsl_blt_shader(gl_info, tex_type);
4368         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "sampler"));
4369         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4370         GL_EXTCALL(glUniform1iARB(loc, 0));
4371     } else {
4372         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4373     }
4374 }
4375
4376 /* GL locking is done by the caller */
4377 static void shader_glsl_deselect_depth_blt(IWineD3DDevice *iface) {
4378     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4379     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4380     struct shader_glsl_priv *priv = This->shader_priv;
4381     GLhandleARB program_id;
4382
4383     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4384     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4385
4386     GL_EXTCALL(glUseProgramObjectARB(program_id));
4387     checkGLcall("glUseProgramObjectARB");
4388 }
4389
4390 static void shader_glsl_destroy(IWineD3DBaseShader *iface) {
4391     const struct list *linked_programs;
4392     IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) iface;
4393     IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
4394     struct shader_glsl_priv *priv = device->shader_priv;
4395     const struct wined3d_gl_info *gl_info;
4396     IWineD3DPixelShaderImpl *ps = NULL;
4397     IWineD3DVertexShaderImpl *vs = NULL;
4398     struct wined3d_context *context;
4399
4400     /* Note: Do not use QueryInterface here to find out which shader type this is because this code
4401      * can be called from IWineD3DBaseShader::Release
4402      */
4403     char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
4404
4405     if(pshader) {
4406         struct glsl_pshader_private *shader_data;
4407         ps = (IWineD3DPixelShaderImpl *) This;
4408         shader_data = ps->baseShader.backend_data;
4409         if(!shader_data || shader_data->num_gl_shaders == 0)
4410         {
4411             HeapFree(GetProcessHeap(), 0, shader_data);
4412             ps->baseShader.backend_data = NULL;
4413             return;
4414         }
4415
4416         context = context_acquire(device, NULL, CTXUSAGE_RESOURCELOAD);
4417         gl_info = context->gl_info;
4418
4419         if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->pshader == iface)
4420         {
4421             ENTER_GL();
4422             shader_glsl_select(context, FALSE, FALSE);
4423             LEAVE_GL();
4424         }
4425     } else {
4426         struct glsl_vshader_private *shader_data;
4427         vs = (IWineD3DVertexShaderImpl *) This;
4428         shader_data = vs->baseShader.backend_data;
4429         if(!shader_data || shader_data->num_gl_shaders == 0)
4430         {
4431             HeapFree(GetProcessHeap(), 0, shader_data);
4432             vs->baseShader.backend_data = NULL;
4433             return;
4434         }
4435
4436         context = context_acquire(device, NULL, CTXUSAGE_RESOURCELOAD);
4437         gl_info = context->gl_info;
4438
4439         if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->vshader == iface)
4440         {
4441             ENTER_GL();
4442             shader_glsl_select(context, FALSE, FALSE);
4443             LEAVE_GL();
4444         }
4445     }
4446
4447     linked_programs = &This->baseShader.linked_programs;
4448
4449     TRACE("Deleting linked programs\n");
4450     if (linked_programs->next) {
4451         struct glsl_shader_prog_link *entry, *entry2;
4452
4453         ENTER_GL();
4454         if(pshader) {
4455             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
4456                 delete_glsl_program_entry(priv, gl_info, entry);
4457             }
4458         } else {
4459             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
4460                 delete_glsl_program_entry(priv, gl_info, entry);
4461             }
4462         }
4463         LEAVE_GL();
4464     }
4465
4466     if(pshader) {
4467         UINT i;
4468         struct glsl_pshader_private *shader_data = ps->baseShader.backend_data;
4469
4470         ENTER_GL();
4471         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4472             TRACE("deleting pshader %u\n", shader_data->gl_shaders[i].prgId);
4473             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4474             checkGLcall("glDeleteObjectARB");
4475         }
4476         LEAVE_GL();
4477         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4478         HeapFree(GetProcessHeap(), 0, shader_data);
4479         ps->baseShader.backend_data = NULL;
4480     } else {
4481         UINT i;
4482         struct glsl_vshader_private *shader_data = vs->baseShader.backend_data;
4483
4484         ENTER_GL();
4485         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4486             TRACE("deleting vshader %u\n", shader_data->gl_shaders[i].prgId);
4487             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4488             checkGLcall("glDeleteObjectARB");
4489         }
4490         LEAVE_GL();
4491         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4492         HeapFree(GetProcessHeap(), 0, shader_data);
4493         vs->baseShader.backend_data = NULL;
4494     }
4495
4496     context_release(context);
4497 }
4498
4499 static int glsl_program_key_compare(const void *key, const struct wine_rb_entry *entry)
4500 {
4501     const glsl_program_key_t *k = key;
4502     const struct glsl_shader_prog_link *prog = WINE_RB_ENTRY_VALUE(entry,
4503             const struct glsl_shader_prog_link, program_lookup_entry);
4504     int cmp;
4505
4506     if (k->vshader > prog->vshader) return 1;
4507     else if (k->vshader < prog->vshader) return -1;
4508
4509     if (k->pshader > prog->pshader) return 1;
4510     else if (k->pshader < prog->pshader) return -1;
4511
4512     if (k->vshader && (cmp = memcmp(&k->vs_args, &prog->vs_args, sizeof(prog->vs_args)))) return cmp;
4513     if (k->pshader && (cmp = memcmp(&k->ps_args, &prog->ps_args, sizeof(prog->ps_args)))) return cmp;
4514
4515     return 0;
4516 }
4517
4518 static BOOL constant_heap_init(struct constant_heap *heap, unsigned int constant_count)
4519 {
4520     SIZE_T size = (constant_count + 1) * sizeof(*heap->entries) + constant_count * sizeof(*heap->positions);
4521     void *mem = HeapAlloc(GetProcessHeap(), 0, size);
4522
4523     if (!mem)
4524     {
4525         ERR("Failed to allocate memory\n");
4526         return FALSE;
4527     }
4528
4529     heap->entries = mem;
4530     heap->entries[1].version = 0;
4531     heap->positions = (unsigned int *)(heap->entries + constant_count + 1);
4532     heap->size = 1;
4533
4534     return TRUE;
4535 }
4536
4537 static void constant_heap_free(struct constant_heap *heap)
4538 {
4539     HeapFree(GetProcessHeap(), 0, heap->entries);
4540 }
4541
4542 static const struct wine_rb_functions wined3d_glsl_program_rb_functions =
4543 {
4544     wined3d_rb_alloc,
4545     wined3d_rb_realloc,
4546     wined3d_rb_free,
4547     glsl_program_key_compare,
4548 };
4549
4550 static HRESULT shader_glsl_alloc(IWineD3DDevice *iface) {
4551     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4552     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4553     struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
4554     SIZE_T stack_size = wined3d_log2i(max(gl_info->limits.glsl_vs_float_constants,
4555             gl_info->limits.glsl_ps_float_constants)) + 1;
4556
4557     if (!shader_buffer_init(&priv->shader_buffer))
4558     {
4559         ERR("Failed to initialize shader buffer.\n");
4560         goto fail;
4561     }
4562
4563     priv->stack = HeapAlloc(GetProcessHeap(), 0, stack_size * sizeof(*priv->stack));
4564     if (!priv->stack)
4565     {
4566         ERR("Failed to allocate memory.\n");
4567         goto fail;
4568     }
4569
4570     if (!constant_heap_init(&priv->vconst_heap, gl_info->limits.glsl_vs_float_constants))
4571     {
4572         ERR("Failed to initialize vertex shader constant heap\n");
4573         goto fail;
4574     }
4575
4576     if (!constant_heap_init(&priv->pconst_heap, gl_info->limits.glsl_ps_float_constants))
4577     {
4578         ERR("Failed to initialize pixel shader constant heap\n");
4579         goto fail;
4580     }
4581
4582     if (wine_rb_init(&priv->program_lookup, &wined3d_glsl_program_rb_functions) == -1)
4583     {
4584         ERR("Failed to initialize rbtree.\n");
4585         goto fail;
4586     }
4587
4588     priv->next_constant_version = 1;
4589
4590     This->shader_priv = priv;
4591     return WINED3D_OK;
4592
4593 fail:
4594     constant_heap_free(&priv->pconst_heap);
4595     constant_heap_free(&priv->vconst_heap);
4596     HeapFree(GetProcessHeap(), 0, priv->stack);
4597     shader_buffer_free(&priv->shader_buffer);
4598     HeapFree(GetProcessHeap(), 0, priv);
4599     return E_OUTOFMEMORY;
4600 }
4601
4602 /* Context activation is done by the caller. */
4603 static void shader_glsl_free(IWineD3DDevice *iface) {
4604     IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4605     const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4606     struct shader_glsl_priv *priv = This->shader_priv;
4607     int i;
4608
4609     ENTER_GL();
4610     for (i = 0; i < tex_type_count; ++i)
4611     {
4612         if (priv->depth_blt_program[i])
4613         {
4614             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program[i]));
4615         }
4616     }
4617     LEAVE_GL();
4618
4619     wine_rb_destroy(&priv->program_lookup, NULL, NULL);
4620     constant_heap_free(&priv->pconst_heap);
4621     constant_heap_free(&priv->vconst_heap);
4622     HeapFree(GetProcessHeap(), 0, priv->stack);
4623     shader_buffer_free(&priv->shader_buffer);
4624
4625     HeapFree(GetProcessHeap(), 0, This->shader_priv);
4626     This->shader_priv = NULL;
4627 }
4628
4629 static BOOL shader_glsl_dirty_const(IWineD3DDevice *iface) {
4630     /* TODO: GL_EXT_bindable_uniform can be used to share constants across shaders */
4631     return FALSE;
4632 }
4633
4634 static void shader_glsl_get_caps(WINED3DDEVTYPE devtype,
4635         const struct wined3d_gl_info *gl_info, struct shader_caps *pCaps)
4636 {
4637     /* Nvidia Geforce6/7 or Ati R4xx/R5xx cards with GLSL support, support VS 3.0 but older Nvidia/Ati
4638      * models with GLSL support only support 2.0. In case of nvidia we can detect VS 2.0 support based
4639      * on the version of NV_vertex_program.
4640      * For Ati cards there's no way using glsl (it abstracts the lowlevel info away) and also not
4641      * using ARB_vertex_program. It is safe to assume that when a card supports pixel shader 2.0 it
4642      * supports vertex shader 2.0 too and the way around. We can detect ps2.0 using the maximum number
4643      * of native instructions, so use that here. For more info see the pixel shader versioning code below.
4644      */
4645     if ((gl_info->supported[NV_VERTEX_PROGRAM2] && !gl_info->supported[NV_VERTEX_PROGRAM3])
4646             || gl_info->limits.arb_ps_instructions <= 512)
4647         pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4648     else
4649         pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4650     TRACE_(d3d_caps)("Hardware vertex shader version %d.%d enabled (GLSL)\n", (pCaps->VertexShaderVersion >> 8) & 0xff, pCaps->VertexShaderVersion & 0xff);
4651     pCaps->MaxVertexShaderConst = gl_info->limits.glsl_vs_float_constants;
4652
4653     /* Older DX9-class videocards (GeforceFX / Radeon >9500/X*00) only support pixel shader 2.0/2.0a/2.0b.
4654      * In OpenGL the extensions related to GLSL abstract lowlevel GL info away which is needed
4655      * to distinguish between 2.0 and 3.0 (and 2.0a/2.0b). In case of Nvidia we use their fragment
4656      * program extensions. On other hardware including ATI GL_ARB_fragment_program offers the info
4657      * in max native instructions. Intel and others also offer the info in this extension but they
4658      * don't support GLSL (at least on Windows).
4659      *
4660      * PS2.0 requires at least 96 instructions, 2.0a/2.0b go up to 512. Assume that if the number
4661      * of instructions is 512 or less we have to do with ps2.0 hardware.
4662      * 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.
4663      */
4664     if ((gl_info->supported[NV_FRAGMENT_PROGRAM] && !gl_info->supported[NV_FRAGMENT_PROGRAM2])
4665             || gl_info->limits.arb_ps_instructions <= 512)
4666         pCaps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
4667     else
4668         pCaps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
4669
4670     pCaps->MaxPixelShaderConst = gl_info->limits.glsl_ps_float_constants;
4671
4672     /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
4673      * Direct3D minimum requirement.
4674      *
4675      * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
4676      * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
4677      *
4678      * The problem is that the refrast clamps temporary results in the shader to
4679      * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
4680      * then applications may miss the clamping behavior. On the other hand, if it is smaller,
4681      * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
4682      * offer a way to query this.
4683      */
4684     pCaps->PixelShader1xMaxValue = 8.0;
4685     TRACE_(d3d_caps)("Hardware pixel shader version %d.%d enabled (GLSL)\n", (pCaps->PixelShaderVersion >> 8) & 0xff, pCaps->PixelShaderVersion & 0xff);
4686
4687     pCaps->VSClipping = TRUE;
4688 }
4689
4690 static BOOL shader_glsl_color_fixup_supported(struct color_fixup_desc fixup)
4691 {
4692     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4693     {
4694         TRACE("Checking support for fixup:\n");
4695         dump_color_fixup_desc(fixup);
4696     }
4697
4698     /* We support everything except YUV conversions. */
4699     if (!is_yuv_fixup(fixup))
4700     {
4701         TRACE("[OK]\n");
4702         return TRUE;
4703     }
4704
4705     TRACE("[FAILED]\n");
4706     return FALSE;
4707 }
4708
4709 static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4710 {
4711     /* WINED3DSIH_ABS           */ shader_glsl_map2gl,
4712     /* WINED3DSIH_ADD           */ shader_glsl_arith,
4713     /* WINED3DSIH_BEM           */ shader_glsl_bem,
4714     /* WINED3DSIH_BREAK         */ shader_glsl_break,
4715     /* WINED3DSIH_BREAKC        */ shader_glsl_breakc,
4716     /* WINED3DSIH_BREAKP        */ NULL,
4717     /* WINED3DSIH_CALL          */ shader_glsl_call,
4718     /* WINED3DSIH_CALLNZ        */ shader_glsl_callnz,
4719     /* WINED3DSIH_CMP           */ shader_glsl_cmp,
4720     /* WINED3DSIH_CND           */ shader_glsl_cnd,
4721     /* WINED3DSIH_CRS           */ shader_glsl_cross,
4722     /* WINED3DSIH_DCL           */ NULL,
4723     /* WINED3DSIH_DEF           */ NULL,
4724     /* WINED3DSIH_DEFB          */ NULL,
4725     /* WINED3DSIH_DEFI          */ NULL,
4726     /* WINED3DSIH_DP2ADD        */ shader_glsl_dp2add,
4727     /* WINED3DSIH_DP3           */ shader_glsl_dot,
4728     /* WINED3DSIH_DP4           */ shader_glsl_dot,
4729     /* WINED3DSIH_DST           */ shader_glsl_dst,
4730     /* WINED3DSIH_DSX           */ shader_glsl_map2gl,
4731     /* WINED3DSIH_DSY           */ shader_glsl_map2gl,
4732     /* WINED3DSIH_ELSE          */ shader_glsl_else,
4733     /* WINED3DSIH_ENDIF         */ shader_glsl_end,
4734     /* WINED3DSIH_ENDLOOP       */ shader_glsl_end,
4735     /* WINED3DSIH_ENDREP        */ shader_glsl_end,
4736     /* WINED3DSIH_EXP           */ shader_glsl_map2gl,
4737     /* WINED3DSIH_EXPP          */ shader_glsl_expp,
4738     /* WINED3DSIH_FRC           */ shader_glsl_map2gl,
4739     /* WINED3DSIH_IF            */ shader_glsl_if,
4740     /* WINED3DSIH_IFC           */ shader_glsl_ifc,
4741     /* WINED3DSIH_LABEL         */ shader_glsl_label,
4742     /* WINED3DSIH_LIT           */ shader_glsl_lit,
4743     /* WINED3DSIH_LOG           */ shader_glsl_log,
4744     /* WINED3DSIH_LOGP          */ shader_glsl_log,
4745     /* WINED3DSIH_LOOP          */ shader_glsl_loop,
4746     /* WINED3DSIH_LRP           */ shader_glsl_lrp,
4747     /* WINED3DSIH_M3x2          */ shader_glsl_mnxn,
4748     /* WINED3DSIH_M3x3          */ shader_glsl_mnxn,
4749     /* WINED3DSIH_M3x4          */ shader_glsl_mnxn,
4750     /* WINED3DSIH_M4x3          */ shader_glsl_mnxn,
4751     /* WINED3DSIH_M4x4          */ shader_glsl_mnxn,
4752     /* WINED3DSIH_MAD           */ shader_glsl_mad,
4753     /* WINED3DSIH_MAX           */ shader_glsl_map2gl,
4754     /* WINED3DSIH_MIN           */ shader_glsl_map2gl,
4755     /* WINED3DSIH_MOV           */ shader_glsl_mov,
4756     /* WINED3DSIH_MOVA          */ shader_glsl_mov,
4757     /* WINED3DSIH_MUL           */ shader_glsl_arith,
4758     /* WINED3DSIH_NOP           */ NULL,
4759     /* WINED3DSIH_NRM           */ shader_glsl_map2gl,
4760     /* WINED3DSIH_PHASE         */ NULL,
4761     /* WINED3DSIH_POW           */ shader_glsl_pow,
4762     /* WINED3DSIH_RCP           */ shader_glsl_rcp,
4763     /* WINED3DSIH_REP           */ shader_glsl_rep,
4764     /* WINED3DSIH_RET           */ shader_glsl_ret,
4765     /* WINED3DSIH_RSQ           */ shader_glsl_rsq,
4766     /* WINED3DSIH_SETP          */ NULL,
4767     /* WINED3DSIH_SGE           */ shader_glsl_compare,
4768     /* WINED3DSIH_SGN           */ shader_glsl_sgn,
4769     /* WINED3DSIH_SINCOS        */ shader_glsl_sincos,
4770     /* WINED3DSIH_SLT           */ shader_glsl_compare,
4771     /* WINED3DSIH_SUB           */ shader_glsl_arith,
4772     /* WINED3DSIH_TEX           */ shader_glsl_tex,
4773     /* WINED3DSIH_TEXBEM        */ shader_glsl_texbem,
4774     /* WINED3DSIH_TEXBEML       */ shader_glsl_texbem,
4775     /* WINED3DSIH_TEXCOORD      */ shader_glsl_texcoord,
4776     /* WINED3DSIH_TEXDEPTH      */ shader_glsl_texdepth,
4777     /* WINED3DSIH_TEXDP3        */ shader_glsl_texdp3,
4778     /* WINED3DSIH_TEXDP3TEX     */ shader_glsl_texdp3tex,
4779     /* WINED3DSIH_TEXKILL       */ shader_glsl_texkill,
4780     /* WINED3DSIH_TEXLDD        */ shader_glsl_texldd,
4781     /* WINED3DSIH_TEXLDL        */ shader_glsl_texldl,
4782     /* WINED3DSIH_TEXM3x2DEPTH  */ shader_glsl_texm3x2depth,
4783     /* WINED3DSIH_TEXM3x2PAD    */ shader_glsl_texm3x2pad,
4784     /* WINED3DSIH_TEXM3x2TEX    */ shader_glsl_texm3x2tex,
4785     /* WINED3DSIH_TEXM3x3       */ shader_glsl_texm3x3,
4786     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
4787     /* WINED3DSIH_TEXM3x3PAD    */ shader_glsl_texm3x3pad,
4788     /* WINED3DSIH_TEXM3x3SPEC   */ shader_glsl_texm3x3spec,
4789     /* WINED3DSIH_TEXM3x3TEX    */ shader_glsl_texm3x3tex,
4790     /* WINED3DSIH_TEXM3x3VSPEC  */ shader_glsl_texm3x3vspec,
4791     /* WINED3DSIH_TEXREG2AR     */ shader_glsl_texreg2ar,
4792     /* WINED3DSIH_TEXREG2GB     */ shader_glsl_texreg2gb,
4793     /* WINED3DSIH_TEXREG2RGB    */ shader_glsl_texreg2rgb,
4794 };
4795
4796 static void shader_glsl_handle_instruction(const struct wined3d_shader_instruction *ins) {
4797     SHADER_HANDLER hw_fct;
4798
4799     /* Select handler */
4800     hw_fct = shader_glsl_instruction_handler_table[ins->handler_idx];
4801
4802     /* Unhandled opcode */
4803     if (!hw_fct)
4804     {
4805         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
4806         return;
4807     }
4808     hw_fct(ins);
4809
4810     shader_glsl_add_instruction_modifiers(ins);
4811 }
4812
4813 const shader_backend_t glsl_shader_backend = {
4814     shader_glsl_handle_instruction,
4815     shader_glsl_select,
4816     shader_glsl_select_depth_blt,
4817     shader_glsl_deselect_depth_blt,
4818     shader_glsl_update_float_vertex_constants,
4819     shader_glsl_update_float_pixel_constants,
4820     shader_glsl_load_constants,
4821     shader_glsl_load_np2fixup_constants,
4822     shader_glsl_destroy,
4823     shader_glsl_alloc,
4824     shader_glsl_free,
4825     shader_glsl_dirty_const,
4826     shader_glsl_get_caps,
4827     shader_glsl_color_fixup_supported,
4828 };