wined3d: Test more thoroughly for post-pixelshader blending support, try on more...
[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 2009-2011 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 WINED3D_GLSL_SAMPLE_PROJECTED   0x1
43 #define WINED3D_GLSL_SAMPLE_RECT        0x2
44 #define WINED3D_GLSL_SAMPLE_LOD         0x4
45 #define WINED3D_GLSL_SAMPLE_GRAD        0x8
46
47 struct glsl_dst_param
48 {
49     char reg_name[150];
50     char mask_str[6];
51 };
52
53 struct glsl_src_param
54 {
55     char reg_name[150];
56     char param_str[200];
57 };
58
59 struct glsl_sample_function
60 {
61     const char *name;
62     DWORD coord_mask;
63 };
64
65 enum heap_node_op
66 {
67     HEAP_NODE_TRAVERSE_LEFT,
68     HEAP_NODE_TRAVERSE_RIGHT,
69     HEAP_NODE_POP,
70 };
71
72 struct constant_entry
73 {
74     unsigned int idx;
75     unsigned int version;
76 };
77
78 struct constant_heap
79 {
80     struct constant_entry *entries;
81     unsigned int *positions;
82     unsigned int size;
83 };
84
85 /* GLSL shader private data */
86 struct shader_glsl_priv {
87     struct wined3d_shader_buffer shader_buffer;
88     struct wine_rb_tree program_lookup;
89     struct glsl_shader_prog_link *glsl_program;
90     struct constant_heap vconst_heap;
91     struct constant_heap pconst_heap;
92     unsigned char *stack;
93     GLhandleARB depth_blt_program_full[tex_type_count];
94     GLhandleARB depth_blt_program_masked[tex_type_count];
95     UINT next_constant_version;
96 };
97
98 /* Struct to maintain data about a linked GLSL program */
99 struct glsl_shader_prog_link {
100     struct wine_rb_entry        program_lookup_entry;
101     struct list                 vshader_entry;
102     struct list                 pshader_entry;
103     GLhandleARB                 programId;
104     GLint                       *vuniformF_locations;
105     GLint                       *puniformF_locations;
106     GLint                       vuniformI_locations[MAX_CONST_I];
107     GLint                       puniformI_locations[MAX_CONST_I];
108     GLint                       posFixup_location;
109     GLint                       np2Fixup_location;
110     GLint                       bumpenvmat_location[MAX_TEXTURES];
111     GLint                       luminancescale_location[MAX_TEXTURES];
112     GLint                       luminanceoffset_location[MAX_TEXTURES];
113     GLint                       ycorrection_location;
114     GLenum                      vertex_color_clamp;
115     const struct wined3d_shader *vshader;
116     const struct wined3d_shader *pshader;
117     struct vs_compile_args      vs_args;
118     struct ps_compile_args      ps_args;
119     UINT                        constant_version;
120     const struct ps_np2fixup_info *np2Fixup_info;
121 };
122
123 struct glsl_program_key
124 {
125     const struct wined3d_shader *vshader;
126     const struct wined3d_shader *pshader;
127     struct ps_compile_args      ps_args;
128     struct vs_compile_args      vs_args;
129 };
130
131 struct shader_glsl_ctx_priv {
132     const struct vs_compile_args    *cur_vs_args;
133     const struct ps_compile_args    *cur_ps_args;
134     struct ps_np2fixup_info         *cur_np2fixup_info;
135 };
136
137 struct glsl_ps_compiled_shader
138 {
139     struct ps_compile_args          args;
140     struct ps_np2fixup_info         np2fixup;
141     GLhandleARB                     prgId;
142 };
143
144 struct glsl_pshader_private
145 {
146     struct glsl_ps_compiled_shader  *gl_shaders;
147     UINT                            num_gl_shaders, shader_array_size;
148 };
149
150 struct glsl_vs_compiled_shader
151 {
152     struct vs_compile_args          args;
153     GLhandleARB                     prgId;
154 };
155
156 struct glsl_vshader_private
157 {
158     struct glsl_vs_compiled_shader  *gl_shaders;
159     UINT                            num_gl_shaders, shader_array_size;
160 };
161
162 static const char *debug_gl_shader_type(GLenum type)
163 {
164     switch (type)
165     {
166 #define WINED3D_TO_STR(u) case u: return #u
167         WINED3D_TO_STR(GL_VERTEX_SHADER_ARB);
168         WINED3D_TO_STR(GL_GEOMETRY_SHADER_ARB);
169         WINED3D_TO_STR(GL_FRAGMENT_SHADER_ARB);
170 #undef WINED3D_TO_STR
171         default:
172             return wine_dbg_sprintf("UNKNOWN(%#x)", type);
173     }
174 }
175
176 /* Extract a line from the info log.
177  * Note that this modifies the source string. */
178 static char *get_info_log_line(char **ptr)
179 {
180     char *p, *q;
181
182     p = *ptr;
183     if (!(q = strstr(p, "\n")))
184     {
185         if (!*p) return NULL;
186         *ptr += strlen(p);
187         return p;
188     }
189     *q = '\0';
190     *ptr = q + 1;
191
192     return p;
193 }
194
195 /** Prints the GLSL info log which will contain error messages if they exist */
196 /* GL locking is done by the caller */
197 static void print_glsl_info_log(const struct wined3d_gl_info *gl_info, GLhandleARB obj)
198 {
199     int infologLength = 0;
200     char *infoLog;
201
202     if (!WARN_ON(d3d_shader) && !FIXME_ON(d3d_shader))
203         return;
204
205     GL_EXTCALL(glGetObjectParameterivARB(obj,
206                GL_OBJECT_INFO_LOG_LENGTH_ARB,
207                &infologLength));
208
209     /* A size of 1 is just a null-terminated string, so the log should be bigger than
210      * that if there are errors. */
211     if (infologLength > 1)
212     {
213         char *ptr, *line;
214
215         infoLog = HeapAlloc(GetProcessHeap(), 0, infologLength);
216         /* The info log is supposed to be zero-terminated, but at least some
217          * versions of fglrx don't terminate the string properly. The reported
218          * length does include the terminator, so explicitly set it to zero
219          * here. */
220         infoLog[infologLength - 1] = 0;
221         GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
222
223         ptr = infoLog;
224         if (gl_info->quirks & WINED3D_QUIRK_INFO_LOG_SPAM)
225         {
226             WARN("Info log received from GLSL shader #%u:\n", obj);
227             while ((line = get_info_log_line(&ptr))) WARN("    %s\n", line);
228         }
229         else
230         {
231             FIXME("Info log received from GLSL shader #%u:\n", obj);
232             while ((line = get_info_log_line(&ptr))) FIXME("    %s\n", line);
233         }
234         HeapFree(GetProcessHeap(), 0, infoLog);
235     }
236 }
237
238 /* GL locking is done by the caller. */
239 static void shader_glsl_compile(const struct wined3d_gl_info *gl_info, GLhandleARB shader, const char *src)
240 {
241     TRACE("Compiling shader object %u.\n", shader);
242     GL_EXTCALL(glShaderSourceARB(shader, 1, &src, NULL));
243     checkGLcall("glShaderSourceARB");
244     GL_EXTCALL(glCompileShaderARB(shader));
245     checkGLcall("glCompileShaderARB");
246     print_glsl_info_log(gl_info, shader);
247 }
248
249 /* GL locking is done by the caller. */
250 static void shader_glsl_dump_program_source(const struct wined3d_gl_info *gl_info, GLhandleARB program)
251 {
252     GLint i, object_count, source_size = -1;
253     GLhandleARB *objects;
254     char *source = NULL;
255
256     GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_ATTACHED_OBJECTS_ARB, &object_count));
257     objects = HeapAlloc(GetProcessHeap(), 0, object_count * sizeof(*objects));
258     if (!objects)
259     {
260         ERR("Failed to allocate object array memory.\n");
261         return;
262     }
263
264     GL_EXTCALL(glGetAttachedObjectsARB(program, object_count, NULL, objects));
265     for (i = 0; i < object_count; ++i)
266     {
267         char *ptr, *line;
268         GLint tmp;
269
270         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_SHADER_SOURCE_LENGTH_ARB, &tmp));
271
272         if (source_size < tmp)
273         {
274             HeapFree(GetProcessHeap(), 0, source);
275
276             source = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, tmp);
277             if (!source)
278             {
279                 ERR("Failed to allocate %d bytes for shader source.\n", tmp);
280                 HeapFree(GetProcessHeap(), 0, objects);
281                 return;
282             }
283             source_size = tmp;
284         }
285
286         FIXME("Object %u:\n", objects[i]);
287         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_SUBTYPE_ARB, &tmp));
288         FIXME("    GL_OBJECT_SUBTYPE_ARB: %s.\n", debug_gl_shader_type(tmp));
289         GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_COMPILE_STATUS_ARB, &tmp));
290         FIXME("    GL_OBJECT_COMPILE_STATUS_ARB: %d.\n", tmp);
291         FIXME("\n");
292
293         ptr = source;
294         GL_EXTCALL(glGetShaderSourceARB(objects[i], source_size, NULL, source));
295         while ((line = get_info_log_line(&ptr))) FIXME("    %s\n", line);
296         FIXME("\n");
297     }
298
299     HeapFree(GetProcessHeap(), 0, source);
300     HeapFree(GetProcessHeap(), 0, objects);
301 }
302
303 /* GL locking is done by the caller. */
304 static void shader_glsl_validate_link(const struct wined3d_gl_info *gl_info, GLhandleARB program)
305 {
306     GLint tmp;
307
308     if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
309
310     GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_TYPE_ARB, &tmp));
311     if (tmp == GL_PROGRAM_OBJECT_ARB)
312     {
313         GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_LINK_STATUS_ARB, &tmp));
314         if (!tmp)
315         {
316             FIXME("Program %u link status invalid.\n", program);
317             shader_glsl_dump_program_source(gl_info, program);
318         }
319     }
320
321     print_glsl_info_log(gl_info, program);
322 }
323
324 /**
325  * Loads (pixel shader) samplers
326  */
327 /* GL locking is done by the caller */
328 static void shader_glsl_load_psamplers(const struct wined3d_gl_info *gl_info,
329         const DWORD *tex_unit_map, GLhandleARB programId)
330 {
331     GLint name_loc;
332     int i;
333     char sampler_name[20];
334
335     for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
336         snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
337         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
338         if (name_loc != -1) {
339             DWORD mapped_unit = tex_unit_map[i];
340             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.fragment_samplers)
341             {
342                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
343                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
344                 checkGLcall("glUniform1iARB");
345             } else {
346                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
347             }
348         }
349     }
350 }
351
352 /* GL locking is done by the caller */
353 static void shader_glsl_load_vsamplers(const struct wined3d_gl_info *gl_info,
354         const DWORD *tex_unit_map, GLhandleARB programId)
355 {
356     GLint name_loc;
357     char sampler_name[20];
358     int i;
359
360     for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
361         snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
362         name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
363         if (name_loc != -1) {
364             DWORD mapped_unit = tex_unit_map[MAX_FRAGMENT_SAMPLERS + i];
365             if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.combined_samplers)
366             {
367                 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
368                 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
369                 checkGLcall("glUniform1iARB");
370             } else {
371                 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
372             }
373         }
374     }
375 }
376
377 /* GL locking is done by the caller */
378 static inline void walk_constant_heap(const struct wined3d_gl_info *gl_info, const float *constants,
379         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
380 {
381     int stack_idx = 0;
382     unsigned int heap_idx = 1;
383     unsigned int idx;
384
385     if (heap->entries[heap_idx].version <= version) return;
386
387     idx = heap->entries[heap_idx].idx;
388     if (constant_locations[idx] != -1) GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
389     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
390
391     while (stack_idx >= 0)
392     {
393         /* Note that we fall through to the next case statement. */
394         switch(stack[stack_idx])
395         {
396             case HEAP_NODE_TRAVERSE_LEFT:
397             {
398                 unsigned int left_idx = heap_idx << 1;
399                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
400                 {
401                     heap_idx = left_idx;
402                     idx = heap->entries[heap_idx].idx;
403                     if (constant_locations[idx] != -1)
404                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
405
406                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
407                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
408                     break;
409                 }
410             }
411
412             case HEAP_NODE_TRAVERSE_RIGHT:
413             {
414                 unsigned int right_idx = (heap_idx << 1) + 1;
415                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
416                 {
417                     heap_idx = right_idx;
418                     idx = heap->entries[heap_idx].idx;
419                     if (constant_locations[idx] != -1)
420                         GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
421
422                     stack[stack_idx++] = HEAP_NODE_POP;
423                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
424                     break;
425                 }
426             }
427
428             case HEAP_NODE_POP:
429             {
430                 heap_idx >>= 1;
431                 --stack_idx;
432                 break;
433             }
434         }
435     }
436     checkGLcall("walk_constant_heap()");
437 }
438
439 /* GL locking is done by the caller */
440 static inline void apply_clamped_constant(const struct wined3d_gl_info *gl_info, GLint location, const GLfloat *data)
441 {
442     GLfloat clamped_constant[4];
443
444     if (location == -1) return;
445
446     clamped_constant[0] = data[0] < -1.0f ? -1.0f : data[0] > 1.0f ? 1.0f : data[0];
447     clamped_constant[1] = data[1] < -1.0f ? -1.0f : data[1] > 1.0f ? 1.0f : data[1];
448     clamped_constant[2] = data[2] < -1.0f ? -1.0f : data[2] > 1.0f ? 1.0f : data[2];
449     clamped_constant[3] = data[3] < -1.0f ? -1.0f : data[3] > 1.0f ? 1.0f : data[3];
450
451     GL_EXTCALL(glUniform4fvARB(location, 1, clamped_constant));
452 }
453
454 /* GL locking is done by the caller */
455 static inline void walk_constant_heap_clamped(const struct wined3d_gl_info *gl_info, const float *constants,
456         const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
457 {
458     int stack_idx = 0;
459     unsigned int heap_idx = 1;
460     unsigned int idx;
461
462     if (heap->entries[heap_idx].version <= version) return;
463
464     idx = heap->entries[heap_idx].idx;
465     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
466     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
467
468     while (stack_idx >= 0)
469     {
470         /* Note that we fall through to the next case statement. */
471         switch(stack[stack_idx])
472         {
473             case HEAP_NODE_TRAVERSE_LEFT:
474             {
475                 unsigned int left_idx = heap_idx << 1;
476                 if (left_idx < heap->size && heap->entries[left_idx].version > version)
477                 {
478                     heap_idx = left_idx;
479                     idx = heap->entries[heap_idx].idx;
480                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
481
482                     stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
483                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
484                     break;
485                 }
486             }
487
488             case HEAP_NODE_TRAVERSE_RIGHT:
489             {
490                 unsigned int right_idx = (heap_idx << 1) + 1;
491                 if (right_idx < heap->size && heap->entries[right_idx].version > version)
492                 {
493                     heap_idx = right_idx;
494                     idx = heap->entries[heap_idx].idx;
495                     apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
496
497                     stack[stack_idx++] = HEAP_NODE_POP;
498                     stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
499                     break;
500                 }
501             }
502
503             case HEAP_NODE_POP:
504             {
505                 heap_idx >>= 1;
506                 --stack_idx;
507                 break;
508             }
509         }
510     }
511     checkGLcall("walk_constant_heap_clamped()");
512 }
513
514 /* Loads floating point constants (aka uniforms) into the currently set GLSL program. */
515 /* GL locking is done by the caller */
516 static void shader_glsl_load_constantsF(const struct wined3d_shader *shader, const struct wined3d_gl_info *gl_info,
517         const float *constants, const GLint *constant_locations, const struct constant_heap *heap,
518         unsigned char *stack, UINT version)
519 {
520     const struct wined3d_shader_lconst *lconst;
521
522     /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
523     if (shader->reg_maps.shader_version.major == 1
524             && shader_is_pshader_version(shader->reg_maps.shader_version.type))
525         walk_constant_heap_clamped(gl_info, constants, constant_locations, heap, stack, version);
526     else
527         walk_constant_heap(gl_info, constants, constant_locations, heap, stack, version);
528
529     if (!shader->load_local_constsF)
530     {
531         TRACE("No need to load local float constants for this shader\n");
532         return;
533     }
534
535     /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
536     LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
537     {
538         GLint location = constant_locations[lconst->idx];
539         /* We found this uniform name in the program - go ahead and send the data */
540         if (location != -1) GL_EXTCALL(glUniform4fvARB(location, 1, (const GLfloat *)lconst->value));
541     }
542     checkGLcall("glUniform4fvARB()");
543 }
544
545 /* Loads integer constants (aka uniforms) into the currently set GLSL program. */
546 /* GL locking is done by the caller */
547 static void shader_glsl_load_constantsI(const struct wined3d_shader *shader, const struct wined3d_gl_info *gl_info,
548         const GLint locations[MAX_CONST_I], const int *constants, WORD constants_set)
549 {
550     unsigned int i;
551     struct list* ptr;
552
553     for (i = 0; constants_set; constants_set >>= 1, ++i)
554     {
555         if (!(constants_set & 1)) continue;
556
557         TRACE_(d3d_constants)("Loading constants %u: %i, %i, %i, %i\n",
558                 i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
559
560         /* We found this uniform name in the program - go ahead and send the data */
561         GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
562         checkGLcall("glUniform4ivARB");
563     }
564
565     /* Load immediate constants */
566     ptr = list_head(&shader->constantsI);
567     while (ptr)
568     {
569         const struct wined3d_shader_lconst *lconst = LIST_ENTRY(ptr, const struct wined3d_shader_lconst, entry);
570         unsigned int idx = lconst->idx;
571         const GLint *values = (const GLint *)lconst->value;
572
573         TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
574             values[0], values[1], values[2], values[3]);
575
576         /* We found this uniform name in the program - go ahead and send the data */
577         GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
578         checkGLcall("glUniform4ivARB");
579         ptr = list_next(&shader->constantsI, ptr);
580     }
581 }
582
583 /* Loads boolean constants (aka uniforms) into the currently set GLSL program. */
584 /* GL locking is done by the caller */
585 static void shader_glsl_load_constantsB(const struct wined3d_shader *shader, const struct wined3d_gl_info *gl_info,
586         GLhandleARB programId, const BOOL *constants, WORD constants_set)
587 {
588     GLint tmp_loc;
589     unsigned int i;
590     char tmp_name[8];
591     const char *prefix;
592     struct list* ptr;
593
594     switch (shader->reg_maps.shader_version.type)
595     {
596         case WINED3D_SHADER_TYPE_VERTEX:
597             prefix = "VB";
598             break;
599
600         case WINED3D_SHADER_TYPE_GEOMETRY:
601             prefix = "GB";
602             break;
603
604         case WINED3D_SHADER_TYPE_PIXEL:
605             prefix = "PB";
606             break;
607
608         default:
609             FIXME("Unknown shader type %#x.\n",
610                     shader->reg_maps.shader_version.type);
611             prefix = "UB";
612             break;
613     }
614
615     /* TODO: Benchmark and see if it would be beneficial to store the
616      * locations of the constants to avoid looking up each time */
617     for (i = 0; constants_set; constants_set >>= 1, ++i)
618     {
619         if (!(constants_set & 1)) continue;
620
621         TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
622
623         /* TODO: Benchmark and see if it would be beneficial to store the
624          * locations of the constants to avoid looking up each time */
625         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
626         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
627         if (tmp_loc != -1)
628         {
629             /* We found this uniform name in the program - go ahead and send the data */
630             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
631             checkGLcall("glUniform1ivARB");
632         }
633     }
634
635     /* Load immediate constants */
636     ptr = list_head(&shader->constantsB);
637     while (ptr)
638     {
639         const struct wined3d_shader_lconst *lconst = LIST_ENTRY(ptr, const struct wined3d_shader_lconst, entry);
640         unsigned int idx = lconst->idx;
641         const GLint *values = (const GLint *)lconst->value;
642
643         TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
644
645         snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
646         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
647         if (tmp_loc != -1) {
648             /* We found this uniform name in the program - go ahead and send the data */
649             GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
650             checkGLcall("glUniform1ivARB");
651         }
652         ptr = list_next(&shader->constantsB, ptr);
653     }
654 }
655
656 static void reset_program_constant_version(struct wine_rb_entry *entry, void *context)
657 {
658     WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry)->constant_version = 0;
659 }
660
661 /**
662  * Loads the texture dimensions for NP2 fixup into the currently set GLSL program.
663  */
664 /* GL locking is done by the caller (state handler) */
665 static void shader_glsl_load_np2fixup_constants(void *shader_priv,
666         const struct wined3d_gl_info *gl_info, const struct wined3d_state *state)
667 {
668     struct shader_glsl_priv *glsl_priv = shader_priv;
669     const struct glsl_shader_prog_link *prog = glsl_priv->glsl_program;
670
671     /* No GLSL program set - nothing to do. */
672     if (!prog) return;
673
674     /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
675     if (!use_ps(state)) return;
676
677     if (prog->ps_args.np2_fixup && prog->np2Fixup_location != -1)
678     {
679         UINT i;
680         UINT fixup = prog->ps_args.np2_fixup;
681         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
682
683         for (i = 0; fixup; fixup >>= 1, ++i)
684         {
685             const struct wined3d_texture *tex = state->textures[i];
686             const unsigned char idx = prog->np2Fixup_info->idx[i];
687             GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
688
689             if (!tex)
690             {
691                 ERR("Nonexistent texture is flagged for NP2 texcoord fixup.\n");
692                 continue;
693             }
694
695             if (idx % 2)
696             {
697                 tex_dim[2] = tex->pow2_matrix[0];
698                 tex_dim[3] = tex->pow2_matrix[5];
699             }
700             else
701             {
702                 tex_dim[0] = tex->pow2_matrix[0];
703                 tex_dim[1] = tex->pow2_matrix[5];
704             }
705         }
706
707         GL_EXTCALL(glUniform4fvARB(prog->np2Fixup_location, prog->np2Fixup_info->num_consts, np2fixup_constants));
708     }
709 }
710
711 /**
712  * Loads the app-supplied constants into the currently set GLSL program.
713  */
714 /* GL locking is done by the caller (state handler) */
715 static void shader_glsl_load_constants(const struct wined3d_context *context,
716         char usePixelShader, char useVertexShader)
717 {
718     const struct wined3d_gl_info *gl_info = context->gl_info;
719     struct wined3d_device *device = context->swapchain->device;
720     struct wined3d_stateblock *stateBlock = device->stateBlock;
721     struct shader_glsl_priv *priv = device->shader_priv;
722     float position_fixup[4];
723
724     GLhandleARB programId;
725     struct glsl_shader_prog_link *prog = priv->glsl_program;
726     UINT constant_version;
727     int i;
728
729     if (!prog) {
730         /* No GLSL program set - nothing to do. */
731         return;
732     }
733     programId = prog->programId;
734     constant_version = prog->constant_version;
735
736     if (useVertexShader)
737     {
738         const struct wined3d_shader *vshader = stateBlock->state.vertex_shader;
739
740         /* Load DirectX 9 float constants/uniforms for vertex shader */
741         shader_glsl_load_constantsF(vshader, gl_info, stateBlock->state.vs_consts_f,
742                 prog->vuniformF_locations, &priv->vconst_heap, priv->stack, constant_version);
743
744         /* Load DirectX 9 integer constants/uniforms for vertex shader */
745         shader_glsl_load_constantsI(vshader, gl_info, prog->vuniformI_locations, stateBlock->state.vs_consts_i,
746                 stateBlock->changed.vertexShaderConstantsI & vshader->reg_maps.integer_constants);
747
748         /* Load DirectX 9 boolean constants/uniforms for vertex shader */
749         shader_glsl_load_constantsB(vshader, gl_info, programId, stateBlock->state.vs_consts_b,
750                 stateBlock->changed.vertexShaderConstantsB & vshader->reg_maps.boolean_constants);
751
752         /* Upload the position fixup params */
753         shader_get_position_fixup(context, &stateBlock->state, position_fixup);
754         GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, position_fixup));
755         checkGLcall("glUniform4fvARB");
756     }
757
758     if (usePixelShader)
759     {
760         const struct wined3d_shader *pshader = stateBlock->state.pixel_shader;
761
762         /* Load DirectX 9 float constants/uniforms for pixel shader */
763         shader_glsl_load_constantsF(pshader, gl_info, stateBlock->state.ps_consts_f,
764                 prog->puniformF_locations, &priv->pconst_heap, priv->stack, constant_version);
765
766         /* Load DirectX 9 integer constants/uniforms for pixel shader */
767         shader_glsl_load_constantsI(pshader, gl_info, prog->puniformI_locations, stateBlock->state.ps_consts_i,
768                 stateBlock->changed.pixelShaderConstantsI & pshader->reg_maps.integer_constants);
769
770         /* Load DirectX 9 boolean constants/uniforms for pixel shader */
771         shader_glsl_load_constantsB(pshader, gl_info, programId, stateBlock->state.ps_consts_b,
772                 stateBlock->changed.pixelShaderConstantsB & pshader->reg_maps.boolean_constants);
773
774         /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
775          * It can't be 0 for a valid texbem instruction.
776          */
777         for(i = 0; i < MAX_TEXTURES; i++) {
778             const float *data;
779
780             if(prog->bumpenvmat_location[i] == -1) continue;
781
782             data = (const float *)&stateBlock->state.texture_states[i][WINED3DTSS_BUMPENVMAT00];
783             GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
784             checkGLcall("glUniformMatrix2fvARB");
785
786             /* texbeml needs the luminance scale and offset too. If texbeml
787              * is used, needsbumpmat is set too, so we can check that in the
788              * needsbumpmat check. */
789             if (prog->luminancescale_location[i] != -1)
790             {
791                 const GLfloat *scale = (const GLfloat *)&stateBlock->state.texture_states[i][WINED3DTSS_BUMPENVLSCALE];
792                 const GLfloat *offset = (const GLfloat *)&stateBlock->state.texture_states[i][WINED3DTSS_BUMPENVLOFFSET];
793
794                 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
795                 checkGLcall("glUniform1fvARB");
796                 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
797                 checkGLcall("glUniform1fvARB");
798             }
799         }
800
801         if (prog->ycorrection_location != -1)
802         {
803             float correction_params[4];
804
805             if (context->render_offscreen)
806             {
807                 correction_params[0] = 0.0f;
808                 correction_params[1] = 1.0f;
809             } else {
810                 /* position is window relative, not viewport relative */
811                 correction_params[0] = (float) context->current_rt->resource.height;
812                 correction_params[1] = -1.0f;
813             }
814             GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
815         }
816     }
817
818     if (priv->next_constant_version == UINT_MAX)
819     {
820         TRACE("Max constant version reached, resetting to 0.\n");
821         wine_rb_for_each_entry(&priv->program_lookup, reset_program_constant_version, NULL);
822         priv->next_constant_version = 1;
823     }
824     else
825     {
826         prog->constant_version = priv->next_constant_version++;
827     }
828 }
829
830 static void update_heap_entry(const struct constant_heap *heap, unsigned int idx,
831         unsigned int heap_idx, DWORD new_version)
832 {
833     struct constant_entry *entries = heap->entries;
834     unsigned int *positions = heap->positions;
835     unsigned int parent_idx;
836
837     while (heap_idx > 1)
838     {
839         parent_idx = heap_idx >> 1;
840
841         if (new_version <= entries[parent_idx].version) break;
842
843         entries[heap_idx] = entries[parent_idx];
844         positions[entries[parent_idx].idx] = heap_idx;
845         heap_idx = parent_idx;
846     }
847
848     entries[heap_idx].version = new_version;
849     entries[heap_idx].idx = idx;
850     positions[idx] = heap_idx;
851 }
852
853 static void shader_glsl_update_float_vertex_constants(struct wined3d_device *device, UINT start, UINT count)
854 {
855     struct shader_glsl_priv *priv = device->shader_priv;
856     struct constant_heap *heap = &priv->vconst_heap;
857     UINT i;
858
859     for (i = start; i < count + start; ++i)
860     {
861         if (!device->stateBlock->changed.vertexShaderConstantsF[i])
862             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
863         else
864             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
865     }
866 }
867
868 static void shader_glsl_update_float_pixel_constants(struct wined3d_device *device, UINT start, UINT count)
869 {
870     struct shader_glsl_priv *priv = device->shader_priv;
871     struct constant_heap *heap = &priv->pconst_heap;
872     UINT i;
873
874     for (i = start; i < count + start; ++i)
875     {
876         if (!device->stateBlock->changed.pixelShaderConstantsF[i])
877             update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
878         else
879             update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
880     }
881 }
882
883 static unsigned int vec4_varyings(DWORD shader_major, const struct wined3d_gl_info *gl_info)
884 {
885     unsigned int ret = gl_info->limits.glsl_varyings / 4;
886     /* 4.0 shaders do not write clip coords because d3d10 does not support user clipplanes */
887     if(shader_major > 3) return ret;
888
889     /* 3.0 shaders may need an extra varying for the clip coord on some cards(mostly dx10 ones) */
890     if (gl_info->quirks & WINED3D_QUIRK_GLSL_CLIP_VARYING) ret -= 1;
891     return ret;
892 }
893
894 /** Generate the variable & register declarations for the GLSL output target */
895 static void shader_generate_glsl_declarations(const struct wined3d_context *context,
896         struct wined3d_shader_buffer *buffer, const struct wined3d_shader *shader,
897         const struct wined3d_shader_reg_maps *reg_maps, const struct shader_glsl_ctx_priv *ctx_priv)
898 {
899     const struct wined3d_state *state = &shader->device->stateBlock->state;
900     const struct ps_compile_args *ps_args = ctx_priv->cur_ps_args;
901     const struct wined3d_gl_info *gl_info = context->gl_info;
902     const struct wined3d_fb_state *fb = &shader->device->fb;
903     unsigned int i, extra_constants_needed = 0;
904     const struct wined3d_shader_lconst *lconst;
905     DWORD map;
906
907     /* There are some minor differences between pixel and vertex shaders */
908     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
909     char prefix = pshader ? 'P' : 'V';
910
911     /* Prototype the subroutines */
912     for (i = 0, map = reg_maps->labels; map; map >>= 1, ++i)
913     {
914         if (map & 1) shader_addline(buffer, "void subroutine%u();\n", i);
915     }
916
917     /* Declare the constants (aka uniforms) */
918     if (shader->limits.constant_float > 0)
919     {
920         unsigned max_constantsF;
921         /* Unless the shader uses indirect addressing, always declare the maximum array size and ignore that we need some
922          * uniforms privately. E.g. if GL supports 256 uniforms, and we need 2 for the pos fixup and immediate values, still
923          * declare VC[256]. If the shader needs more uniforms than we have it won't work in any case. If it uses less, the
924          * compiler will figure out which uniforms are really used and strip them out. This allows a shader to use c255 on
925          * a dx9 card, as long as it doesn't also use all the other constants.
926          *
927          * If the shader uses indirect addressing the compiler must assume that all declared uniforms are used. In this case,
928          * declare only the amount that we're assured to have.
929          *
930          * Thus we run into problems in these two cases:
931          * 1) The shader really uses more uniforms than supported
932          * 2) The shader uses indirect addressing, less constants than supported, but uses a constant index > #supported consts
933          */
934         if (pshader)
935         {
936             /* No indirect addressing here. */
937             max_constantsF = gl_info->limits.glsl_ps_float_constants;
938         }
939         else
940         {
941             if (reg_maps->usesrelconstF)
942             {
943                 /* Subtract the other potential uniforms from the max
944                  * available (bools, ints, and 1 row of projection matrix).
945                  * Subtract another uniform for immediate values, which have
946                  * to be loaded via uniform by the driver as well. The shader
947                  * code only uses 0.5, 2.0, 1.0, 128 and -128 in vertex
948                  * shader code, so one vec4 should be enough. (Unfortunately
949                  * the Nvidia driver doesn't store 128 and -128 in one float).
950                  *
951                  * Writing gl_ClipVertex requires one uniform for each
952                  * clipplane as well. */
953                 max_constantsF = gl_info->limits.glsl_vs_float_constants - 3;
954                 if(ctx_priv->cur_vs_args->clip_enabled)
955                 {
956                     max_constantsF -= gl_info->limits.clipplanes;
957                 }
958                 max_constantsF -= count_bits(reg_maps->integer_constants);
959                 /* Strictly speaking a bool only uses one scalar, but the nvidia(Linux) compiler doesn't pack them properly,
960                  * so each scalar requires a full vec4. We could work around this by packing the booleans ourselves, but
961                  * for now take this into account when calculating the number of available constants
962                  */
963                 max_constantsF -= count_bits(reg_maps->boolean_constants);
964                 /* Set by driver quirks in directx.c */
965                 max_constantsF -= gl_info->reserved_glsl_constants;
966             }
967             else
968             {
969                 max_constantsF = gl_info->limits.glsl_vs_float_constants;
970             }
971         }
972         max_constantsF = min(shader->limits.constant_float, max_constantsF);
973         shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
974     }
975
976     /* Always declare the full set of constants, the compiler can remove the
977      * unused ones because d3d doesn't (yet) support indirect int and bool
978      * constant addressing. This avoids problems if the app uses e.g. i0 and i9. */
979     if (shader->limits.constant_int > 0 && reg_maps->integer_constants)
980         shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, shader->limits.constant_int);
981
982     if (shader->limits.constant_bool > 0 && reg_maps->boolean_constants)
983         shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, shader->limits.constant_bool);
984
985     if (!pshader)
986     {
987         shader_addline(buffer, "uniform vec4 posFixup;\n");
988         shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
989     }
990     else
991     {
992         for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
993         {
994             if (!(map & 1)) continue;
995
996             shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
997
998             if (reg_maps->luminanceparams & (1 << i))
999             {
1000                 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
1001                 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
1002                 extra_constants_needed++;
1003             }
1004
1005             extra_constants_needed++;
1006         }
1007
1008         if (ps_args->srgb_correction)
1009         {
1010             shader_addline(buffer, "const vec4 srgb_const0 = vec4(%.8e, %.8e, %.8e, %.8e);\n",
1011                     srgb_pow, srgb_mul_high, srgb_sub_high, srgb_mul_low);
1012             shader_addline(buffer, "const vec4 srgb_const1 = vec4(%.8e, 0.0, 0.0, 0.0);\n",
1013                     srgb_cmp);
1014         }
1015         if (reg_maps->vpos || reg_maps->usesdsy)
1016         {
1017             if (shader->limits.constant_float + extra_constants_needed
1018                     + 1 < gl_info->limits.glsl_ps_float_constants)
1019             {
1020                 shader_addline(buffer, "uniform vec4 ycorrection;\n");
1021                 extra_constants_needed++;
1022             }
1023             else
1024             {
1025                 /* This happens because we do not have proper tracking of the constant registers that are
1026                  * actually used, only the max limit of the shader version
1027                  */
1028                 FIXME("Cannot find a free uniform for vpos correction params\n");
1029                 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
1030                         context->render_offscreen ? 0.0f : fb->render_targets[0]->resource.height,
1031                         context->render_offscreen ? 1.0f : -1.0f);
1032             }
1033             shader_addline(buffer, "vec4 vpos;\n");
1034         }
1035     }
1036
1037     /* Declare texture samplers */
1038     for (i = 0; i < shader->limits.sampler; ++i)
1039     {
1040         if (reg_maps->sampler_type[i])
1041         {
1042             const struct wined3d_texture *texture;
1043
1044             switch (reg_maps->sampler_type[i])
1045             {
1046                 case WINED3DSTT_1D:
1047                     if (pshader && ps_args->shadow & (1 << i))
1048                         shader_addline(buffer, "uniform sampler1DShadow %csampler%u;\n", prefix, i);
1049                     else
1050                         shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
1051                     break;
1052                 case WINED3DSTT_2D:
1053                     texture = state->textures[i];
1054                     if (pshader && ps_args->shadow & (1 << i))
1055                     {
1056                         if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
1057                             shader_addline(buffer, "uniform sampler2DRectShadow %csampler%u;\n", prefix, i);
1058                         else
1059                             shader_addline(buffer, "uniform sampler2DShadow %csampler%u;\n", prefix, i);
1060                     }
1061                     else
1062                     {
1063                         if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
1064                             shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
1065                         else
1066                             shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
1067                     }
1068                     break;
1069                 case WINED3DSTT_CUBE:
1070                     if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported Cube shadow sampler.\n");
1071                     shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
1072                     break;
1073                 case WINED3DSTT_VOLUME:
1074                     if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported 3D shadow sampler.\n");
1075                     shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
1076                     break;
1077                 default:
1078                     shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
1079                     FIXME("Unrecognized sampler type: %#x\n", reg_maps->sampler_type[i]);
1080                     break;
1081             }
1082         }
1083     }
1084
1085     /* Declare uniforms for NP2 texcoord fixup:
1086      * This is NOT done inside the loop that declares the texture samplers since the NP2 fixup code
1087      * is currently only used for the GeforceFX series and when forcing the ARB_npot extension off.
1088      * Modern cards just skip the code anyway, so put it inside a separate loop. */
1089     if (pshader && ps_args->np2_fixup) {
1090
1091         struct ps_np2fixup_info* const fixup = ctx_priv->cur_np2fixup_info;
1092         UINT cur = 0;
1093
1094         /* NP2/RECT textures in OpenGL use texcoords in the range [0,width]x[0,height]
1095          * while D3D has them in the (normalized) [0,1]x[0,1] range.
1096          * samplerNP2Fixup stores texture dimensions and is updated through
1097          * shader_glsl_load_np2fixup_constants when the sampler changes. */
1098
1099         for (i = 0; i < shader->limits.sampler; ++i)
1100         {
1101             if (reg_maps->sampler_type[i])
1102             {
1103                 if (!(ps_args->np2_fixup & (1 << i))) continue;
1104
1105                 if (WINED3DSTT_2D != reg_maps->sampler_type[i]) {
1106                     FIXME("Non-2D texture is flagged for NP2 texcoord fixup.\n");
1107                     continue;
1108                 }
1109
1110                 fixup->idx[i] = cur++;
1111             }
1112         }
1113
1114         fixup->num_consts = (cur + 1) >> 1;
1115         shader_addline(buffer, "uniform vec4 %csamplerNP2Fixup[%u];\n", prefix, fixup->num_consts);
1116     }
1117
1118     /* Declare address variables */
1119     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
1120     {
1121         if (map & 1) shader_addline(buffer, "ivec4 A%u;\n", i);
1122     }
1123
1124     /* Declare texture coordinate temporaries and initialize them */
1125     for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
1126     {
1127         if (map & 1) shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
1128     }
1129
1130     /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
1131      * helper function shader that is linked in at link time
1132      */
1133     if (pshader && reg_maps->shader_version.major >= 3)
1134     {
1135         if (use_vs(state))
1136         {
1137             shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1138         } else {
1139             /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
1140              * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
1141              * pixel shader that reads the fixed function color into the packed input registers.
1142              */
1143             shader_addline(buffer, "vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1144         }
1145     }
1146
1147     /* Declare output register temporaries */
1148     if (shader->limits.packed_output)
1149         shader_addline(buffer, "vec4 OUT[%u];\n", shader->limits.packed_output);
1150
1151     /* Declare temporary variables */
1152     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
1153     {
1154         if (map & 1) shader_addline(buffer, "vec4 R%u;\n", i);
1155     }
1156
1157     /* Declare attributes */
1158     if (reg_maps->shader_version.type == WINED3D_SHADER_TYPE_VERTEX)
1159     {
1160         for (i = 0, map = reg_maps->input_registers; map; map >>= 1, ++i)
1161         {
1162             if (map & 1) shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
1163         }
1164     }
1165
1166     /* Declare loop registers aLx */
1167     for (i = 0; i < reg_maps->loop_depth; i++) {
1168         shader_addline(buffer, "int aL%u;\n", i);
1169         shader_addline(buffer, "int tmpInt%u;\n", i);
1170     }
1171
1172     /* Temporary variables for matrix operations */
1173     shader_addline(buffer, "vec4 tmp0;\n");
1174     shader_addline(buffer, "vec4 tmp1;\n");
1175
1176     /* Local constants use a different name so they can be loaded once at shader link time
1177      * They can't be hardcoded into the shader text via LC = {x, y, z, w}; because the
1178      * float -> string conversion can cause precision loss.
1179      */
1180     if (!shader->load_local_constsF)
1181     {
1182         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
1183         {
1184             shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
1185         }
1186     }
1187
1188     /* Start the main program */
1189     shader_addline(buffer, "void main() {\n");
1190     if(pshader && reg_maps->vpos) {
1191         /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
1192          * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
1193          * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
1194          * precision troubles when we just subtract 0.5.
1195          *
1196          * To deal with that just floor() the position. This will eliminate the fraction on all cards.
1197          *
1198          * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
1199          *
1200          * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
1201          * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
1202          * coordinates specify the pixel centers instead of the pixel corners. This code will behave
1203          * correctly on drivers that returns integer values.
1204          */
1205         shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
1206     }
1207 }
1208
1209 /*****************************************************************************
1210  * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
1211  *
1212  * For more information, see http://wiki.winehq.org/DirectX-Shaders
1213  ****************************************************************************/
1214
1215 /* Prototypes */
1216 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1217         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, struct glsl_src_param *glsl_src);
1218
1219 /** Used for opcode modifiers - They multiply the result by the specified amount */
1220 static const char * const shift_glsl_tab[] = {
1221     "",           /*  0 (none) */
1222     "2.0 * ",     /*  1 (x2)   */
1223     "4.0 * ",     /*  2 (x4)   */
1224     "8.0 * ",     /*  3 (x8)   */
1225     "16.0 * ",    /*  4 (x16)  */
1226     "32.0 * ",    /*  5 (x32)  */
1227     "",           /*  6 (x64)  */
1228     "",           /*  7 (x128) */
1229     "",           /*  8 (d256) */
1230     "",           /*  9 (d128) */
1231     "",           /* 10 (d64)  */
1232     "",           /* 11 (d32)  */
1233     "0.0625 * ",  /* 12 (d16)  */
1234     "0.125 * ",   /* 13 (d8)   */
1235     "0.25 * ",    /* 14 (d4)   */
1236     "0.5 * "      /* 15 (d2)   */
1237 };
1238
1239 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
1240 static void shader_glsl_gen_modifier(DWORD src_modifier, const char *in_reg, const char *in_regswizzle, char *out_str)
1241 {
1242     out_str[0] = 0;
1243
1244     switch (src_modifier)
1245     {
1246     case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
1247     case WINED3DSPSM_DW:
1248     case WINED3DSPSM_NONE:
1249         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1250         break;
1251     case WINED3DSPSM_NEG:
1252         sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
1253         break;
1254     case WINED3DSPSM_NOT:
1255         sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
1256         break;
1257     case WINED3DSPSM_BIAS:
1258         sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1259         break;
1260     case WINED3DSPSM_BIASNEG:
1261         sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1262         break;
1263     case WINED3DSPSM_SIGN:
1264         sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1265         break;
1266     case WINED3DSPSM_SIGNNEG:
1267         sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1268         break;
1269     case WINED3DSPSM_COMP:
1270         sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
1271         break;
1272     case WINED3DSPSM_X2:
1273         sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
1274         break;
1275     case WINED3DSPSM_X2NEG:
1276         sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
1277         break;
1278     case WINED3DSPSM_ABS:
1279         sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
1280         break;
1281     case WINED3DSPSM_ABSNEG:
1282         sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
1283         break;
1284     default:
1285         FIXME("Unhandled modifier %u\n", src_modifier);
1286         sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1287     }
1288 }
1289
1290 /** Writes the GLSL variable name that corresponds to the register that the
1291  * DX opcode parameter is trying to access */
1292 static void shader_glsl_get_register_name(const struct wined3d_shader_register *reg,
1293         char *register_name, BOOL *is_color, const struct wined3d_shader_instruction *ins)
1294 {
1295     /* oPos, oFog and oPts in D3D */
1296     static const char * const hwrastout_reg_names[] = {"OUT[10]", "OUT[11].x", "OUT[11].y"};
1297
1298     const struct wined3d_shader *shader = ins->ctx->shader;
1299     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1300     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
1301     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1302
1303     *is_color = FALSE;
1304
1305     switch (reg->type)
1306     {
1307         case WINED3DSPR_TEMP:
1308             sprintf(register_name, "R%u", reg->idx);
1309             break;
1310
1311         case WINED3DSPR_INPUT:
1312             /* vertex shaders */
1313             if (!pshader)
1314             {
1315                 struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1316                 if (priv->cur_vs_args->swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1317                 sprintf(register_name, "attrib%u", reg->idx);
1318                 break;
1319             }
1320
1321             /* pixel shaders >= 3.0 */
1322             if (reg_maps->shader_version.major >= 3)
1323             {
1324                 DWORD idx = shader->u.ps.input_reg_map[reg->idx];
1325                 unsigned int in_count = vec4_varyings(reg_maps->shader_version.major, gl_info);
1326
1327                 if (reg->rel_addr)
1328                 {
1329                     struct glsl_src_param rel_param;
1330
1331                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1332
1333                     /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
1334                      * operation there */
1335                     if (idx)
1336                     {
1337                         if (shader->u.ps.declared_in_count > in_count)
1338                         {
1339                             sprintf(register_name,
1340                                     "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
1341                                     rel_param.param_str, idx, in_count - 1, rel_param.param_str, idx, in_count,
1342                                     rel_param.param_str, idx);
1343                         }
1344                         else
1345                         {
1346                             sprintf(register_name, "IN[%s + %u]", rel_param.param_str, idx);
1347                         }
1348                     }
1349                     else
1350                     {
1351                         if (shader->u.ps.declared_in_count > in_count)
1352                         {
1353                             sprintf(register_name, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
1354                                     rel_param.param_str, in_count - 1, rel_param.param_str, in_count,
1355                                     rel_param.param_str);
1356                         }
1357                         else
1358                         {
1359                             sprintf(register_name, "IN[%s]", rel_param.param_str);
1360                         }
1361                     }
1362                 }
1363                 else
1364                 {
1365                     if (idx == in_count) sprintf(register_name, "gl_Color");
1366                     else if (idx == in_count + 1) sprintf(register_name, "gl_SecondaryColor");
1367                     else sprintf(register_name, "IN[%u]", idx);
1368                 }
1369             }
1370             else
1371             {
1372                 if (!reg->idx) strcpy(register_name, "gl_Color");
1373                 else strcpy(register_name, "gl_SecondaryColor");
1374                 break;
1375             }
1376             break;
1377
1378         case WINED3DSPR_CONST:
1379             {
1380                 const char prefix = pshader ? 'P' : 'V';
1381
1382                 /* Relative addressing */
1383                 if (reg->rel_addr)
1384                 {
1385                     struct glsl_src_param rel_param;
1386                     shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1387                     if (reg->idx) sprintf(register_name, "%cC[%s + %u]", prefix, rel_param.param_str, reg->idx);
1388                     else sprintf(register_name, "%cC[%s]", prefix, rel_param.param_str);
1389                 }
1390                 else
1391                 {
1392                     if (shader_constant_is_local(shader, reg->idx))
1393                         sprintf(register_name, "%cLC%u", prefix, reg->idx);
1394                     else
1395                         sprintf(register_name, "%cC[%u]", prefix, reg->idx);
1396                 }
1397             }
1398             break;
1399
1400         case WINED3DSPR_CONSTINT:
1401             if (pshader) sprintf(register_name, "PI[%u]", reg->idx);
1402             else sprintf(register_name, "VI[%u]", reg->idx);
1403             break;
1404
1405         case WINED3DSPR_CONSTBOOL:
1406             if (pshader) sprintf(register_name, "PB[%u]", reg->idx);
1407             else sprintf(register_name, "VB[%u]", reg->idx);
1408             break;
1409
1410         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1411             if (pshader) sprintf(register_name, "T%u", reg->idx);
1412             else sprintf(register_name, "A%u", reg->idx);
1413             break;
1414
1415         case WINED3DSPR_LOOP:
1416             sprintf(register_name, "aL%u", ins->ctx->loop_state->current_reg - 1);
1417             break;
1418
1419         case WINED3DSPR_SAMPLER:
1420             if (pshader) sprintf(register_name, "Psampler%u", reg->idx);
1421             else sprintf(register_name, "Vsampler%u", reg->idx);
1422             break;
1423
1424         case WINED3DSPR_COLOROUT:
1425             if (reg->idx >= gl_info->limits.buffers)
1426                 WARN("Write to render target %u, only %d supported.\n", reg->idx, gl_info->limits.buffers);
1427
1428             sprintf(register_name, "gl_FragData[%u]", reg->idx);
1429             break;
1430
1431         case WINED3DSPR_RASTOUT:
1432             sprintf(register_name, "%s", hwrastout_reg_names[reg->idx]);
1433             break;
1434
1435         case WINED3DSPR_DEPTHOUT:
1436             sprintf(register_name, "gl_FragDepth");
1437             break;
1438
1439         case WINED3DSPR_ATTROUT:
1440             if (!reg->idx) sprintf(register_name, "OUT[8]");
1441             else sprintf(register_name, "OUT[9]");
1442             break;
1443
1444         case WINED3DSPR_TEXCRDOUT:
1445             /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
1446             sprintf(register_name, "OUT[%u]", reg->idx);
1447             break;
1448
1449         case WINED3DSPR_MISCTYPE:
1450             if (!reg->idx)
1451             {
1452                 /* vPos */
1453                 sprintf(register_name, "vpos");
1454             }
1455             else if (reg->idx == 1)
1456             {
1457                 /* Note that gl_FrontFacing is a bool, while vFace is
1458                  * a float for which the sign determines front/back */
1459                 sprintf(register_name, "(gl_FrontFacing ? 1.0 : -1.0)");
1460             }
1461             else
1462             {
1463                 FIXME("Unhandled misctype register %d\n", reg->idx);
1464                 sprintf(register_name, "unrecognized_register");
1465             }
1466             break;
1467
1468         case WINED3DSPR_IMMCONST:
1469             switch (reg->immconst_type)
1470             {
1471                 case WINED3D_IMMCONST_SCALAR:
1472                     sprintf(register_name, "%.8e", *(const float *)reg->immconst_data);
1473                     break;
1474
1475                 case WINED3D_IMMCONST_VEC4:
1476                     sprintf(register_name, "vec4(%.8e, %.8e, %.8e, %.8e)",
1477                             *(const float *)&reg->immconst_data[0], *(const float *)&reg->immconst_data[1],
1478                             *(const float *)&reg->immconst_data[2], *(const float *)&reg->immconst_data[3]);
1479                     break;
1480
1481                 default:
1482                     FIXME("Unhandled immconst type %#x\n", reg->immconst_type);
1483                     sprintf(register_name, "<unhandled_immconst_type %#x>", reg->immconst_type);
1484             }
1485             break;
1486
1487         default:
1488             FIXME("Unhandled register name Type(%d)\n", reg->type);
1489             sprintf(register_name, "unrecognized_register");
1490             break;
1491     }
1492 }
1493
1494 static void shader_glsl_write_mask_to_str(DWORD write_mask, char *str)
1495 {
1496     *str++ = '.';
1497     if (write_mask & WINED3DSP_WRITEMASK_0) *str++ = 'x';
1498     if (write_mask & WINED3DSP_WRITEMASK_1) *str++ = 'y';
1499     if (write_mask & WINED3DSP_WRITEMASK_2) *str++ = 'z';
1500     if (write_mask & WINED3DSP_WRITEMASK_3) *str++ = 'w';
1501     *str = '\0';
1502 }
1503
1504 /* Get the GLSL write mask for the destination register */
1505 static DWORD shader_glsl_get_write_mask(const struct wined3d_shader_dst_param *param, char *write_mask)
1506 {
1507     DWORD mask = param->write_mask;
1508
1509     if (shader_is_scalar(&param->reg))
1510     {
1511         mask = WINED3DSP_WRITEMASK_0;
1512         *write_mask = '\0';
1513     }
1514     else
1515     {
1516         shader_glsl_write_mask_to_str(mask, write_mask);
1517     }
1518
1519     return mask;
1520 }
1521
1522 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1523     unsigned int size = 0;
1524
1525     if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1526     if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1527     if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1528     if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1529
1530     return size;
1531 }
1532
1533 static void shader_glsl_swizzle_to_str(const DWORD swizzle, BOOL fixup, DWORD mask, char *str)
1534 {
1535     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1536      * but addressed as "rgba". To fix this we need to swap the register's x
1537      * and z components. */
1538     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1539
1540     *str++ = '.';
1541     /* swizzle bits fields: wwzzyyxx */
1542     if (mask & WINED3DSP_WRITEMASK_0) *str++ = swizzle_chars[swizzle & 0x03];
1543     if (mask & WINED3DSP_WRITEMASK_1) *str++ = swizzle_chars[(swizzle >> 2) & 0x03];
1544     if (mask & WINED3DSP_WRITEMASK_2) *str++ = swizzle_chars[(swizzle >> 4) & 0x03];
1545     if (mask & WINED3DSP_WRITEMASK_3) *str++ = swizzle_chars[(swizzle >> 6) & 0x03];
1546     *str = '\0';
1547 }
1548
1549 static void shader_glsl_get_swizzle(const struct wined3d_shader_src_param *param,
1550         BOOL fixup, DWORD mask, char *swizzle_str)
1551 {
1552     if (shader_is_scalar(&param->reg))
1553         *swizzle_str = '\0';
1554     else
1555         shader_glsl_swizzle_to_str(param->swizzle, fixup, mask, swizzle_str);
1556 }
1557
1558 /* From a given parameter token, generate the corresponding GLSL string.
1559  * Also, return the actual register name and swizzle in case the
1560  * caller needs this information as well. */
1561 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1562         const struct wined3d_shader_src_param *wined3d_src, DWORD mask, struct glsl_src_param *glsl_src)
1563 {
1564     BOOL is_color = FALSE;
1565     char swizzle_str[6];
1566
1567     glsl_src->reg_name[0] = '\0';
1568     glsl_src->param_str[0] = '\0';
1569     swizzle_str[0] = '\0';
1570
1571     shader_glsl_get_register_name(&wined3d_src->reg, glsl_src->reg_name, &is_color, ins);
1572     shader_glsl_get_swizzle(wined3d_src, is_color, mask, swizzle_str);
1573     shader_glsl_gen_modifier(wined3d_src->modifiers, glsl_src->reg_name, swizzle_str, glsl_src->param_str);
1574 }
1575
1576 /* From a given parameter token, generate the corresponding GLSL string.
1577  * Also, return the actual register name and swizzle in case the
1578  * caller needs this information as well. */
1579 static DWORD shader_glsl_add_dst_param(const struct wined3d_shader_instruction *ins,
1580         const struct wined3d_shader_dst_param *wined3d_dst, struct glsl_dst_param *glsl_dst)
1581 {
1582     BOOL is_color = FALSE;
1583
1584     glsl_dst->mask_str[0] = '\0';
1585     glsl_dst->reg_name[0] = '\0';
1586
1587     shader_glsl_get_register_name(&wined3d_dst->reg, glsl_dst->reg_name, &is_color, ins);
1588     return shader_glsl_get_write_mask(wined3d_dst, glsl_dst->mask_str);
1589 }
1590
1591 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1592 static DWORD shader_glsl_append_dst_ext(struct wined3d_shader_buffer *buffer,
1593         const struct wined3d_shader_instruction *ins, const struct wined3d_shader_dst_param *dst)
1594 {
1595     struct glsl_dst_param glsl_dst;
1596     DWORD mask;
1597
1598     mask = shader_glsl_add_dst_param(ins, dst, &glsl_dst);
1599     if (mask) shader_addline(buffer, "%s%s = %s(", glsl_dst.reg_name, glsl_dst.mask_str, shift_glsl_tab[dst->shift]);
1600
1601     return mask;
1602 }
1603
1604 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1605 static DWORD shader_glsl_append_dst(struct wined3d_shader_buffer *buffer, const struct wined3d_shader_instruction *ins)
1606 {
1607     return shader_glsl_append_dst_ext(buffer, ins, &ins->dst[0]);
1608 }
1609
1610 /** Process GLSL instruction modifiers */
1611 static void shader_glsl_add_instruction_modifiers(const struct wined3d_shader_instruction *ins)
1612 {
1613     struct glsl_dst_param dst_param;
1614     DWORD modifiers;
1615
1616     if (!ins->dst_count) return;
1617
1618     modifiers = ins->dst[0].modifiers;
1619     if (!modifiers) return;
1620
1621     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
1622
1623     if (modifiers & WINED3DSPDM_SATURATE)
1624     {
1625         /* _SAT means to clamp the value of the register to between 0 and 1 */
1626         shader_addline(ins->ctx->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1627                 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1628     }
1629
1630     if (modifiers & WINED3DSPDM_MSAMPCENTROID)
1631     {
1632         FIXME("_centroid modifier not handled\n");
1633     }
1634
1635     if (modifiers & WINED3DSPDM_PARTIALPRECISION)
1636     {
1637         /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1638     }
1639 }
1640
1641 static inline const char *shader_get_comp_op(DWORD op)
1642 {
1643     switch (op) {
1644         case COMPARISON_GT: return ">";
1645         case COMPARISON_EQ: return "==";
1646         case COMPARISON_GE: return ">=";
1647         case COMPARISON_LT: return "<";
1648         case COMPARISON_NE: return "!=";
1649         case COMPARISON_LE: return "<=";
1650         default:
1651             FIXME("Unrecognized comparison value: %u\n", op);
1652             return "(\?\?)";
1653     }
1654 }
1655
1656 static void shader_glsl_get_sample_function(const struct wined3d_shader_context *ctx,
1657         DWORD sampler_idx, DWORD flags, struct glsl_sample_function *sample_function)
1658 {
1659     WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ctx->reg_maps->sampler_type[sampler_idx];
1660     const struct wined3d_gl_info *gl_info = ctx->gl_info;
1661     BOOL shadow = shader_is_pshader_version(ctx->reg_maps->shader_version.type)
1662             && (((const struct shader_glsl_ctx_priv *)ctx->backend_data)->cur_ps_args->shadow & (1 << sampler_idx));
1663     BOOL projected = flags & WINED3D_GLSL_SAMPLE_PROJECTED;
1664     BOOL texrect = flags & WINED3D_GLSL_SAMPLE_RECT;
1665     BOOL lod = flags & WINED3D_GLSL_SAMPLE_LOD;
1666     BOOL grad = flags & WINED3D_GLSL_SAMPLE_GRAD;
1667
1668     /* Note that there's no such thing as a projected cube texture. */
1669     switch(sampler_type) {
1670         case WINED3DSTT_1D:
1671             if (shadow)
1672             {
1673                 if (lod)
1674                 {
1675                     sample_function->name = projected ? "shadow1DProjLod" : "shadow1DLod";
1676                 }
1677                 else if (grad)
1678                 {
1679                     if (gl_info->supported[EXT_GPU_SHADER4])
1680                         sample_function->name = projected ? "shadow1DProjGrad" : "shadow1DGrad";
1681                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1682                         sample_function->name = projected ? "shadow1DProjGradARB" : "shadow1DGradARB";
1683                     else
1684                     {
1685                         FIXME("Unsupported 1D shadow grad function.\n");
1686                         sample_function->name = "unsupported1DGrad";
1687                     }
1688                 }
1689                 else
1690                 {
1691                     sample_function->name = projected ? "shadow1DProj" : "shadow1D";
1692                 }
1693                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1694             }
1695             else
1696             {
1697                 if (lod)
1698                 {
1699                     sample_function->name = projected ? "texture1DProjLod" : "texture1DLod";
1700                 }
1701                 else if (grad)
1702                 {
1703                     if (gl_info->supported[EXT_GPU_SHADER4])
1704                         sample_function->name = projected ? "texture1DProjGrad" : "texture1DGrad";
1705                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1706                         sample_function->name = projected ? "texture1DProjGradARB" : "texture1DGradARB";
1707                     else
1708                     {
1709                         FIXME("Unsupported 1D grad function.\n");
1710                         sample_function->name = "unsupported1DGrad";
1711                     }
1712                 }
1713                 else
1714                 {
1715                     sample_function->name = projected ? "texture1DProj" : "texture1D";
1716                 }
1717                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1718             }
1719             break;
1720
1721         case WINED3DSTT_2D:
1722             if (shadow)
1723             {
1724                 if (texrect)
1725                 {
1726                     if (lod)
1727                     {
1728                         sample_function->name = projected ? "shadow2DRectProjLod" : "shadow2DRectLod";
1729                     }
1730                     else if (grad)
1731                     {
1732                         if (gl_info->supported[EXT_GPU_SHADER4])
1733                             sample_function->name = projected ? "shadow2DRectProjGrad" : "shadow2DRectGrad";
1734                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1735                             sample_function->name = projected ? "shadow2DRectProjGradARB" : "shadow2DRectGradARB";
1736                         else
1737                         {
1738                             FIXME("Unsupported RECT shadow grad function.\n");
1739                             sample_function->name = "unsupported2DRectGrad";
1740                         }
1741                     }
1742                     else
1743                     {
1744                         sample_function->name = projected ? "shadow2DRectProj" : "shadow2DRect";
1745                     }
1746                 }
1747                 else
1748                 {
1749                     if (lod)
1750                     {
1751                         sample_function->name = projected ? "shadow2DProjLod" : "shadow2DLod";
1752                     }
1753                     else if (grad)
1754                     {
1755                         if (gl_info->supported[EXT_GPU_SHADER4])
1756                             sample_function->name = projected ? "shadow2DProjGrad" : "shadow2DGrad";
1757                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1758                             sample_function->name = projected ? "shadow2DProjGradARB" : "shadow2DGradARB";
1759                         else
1760                         {
1761                             FIXME("Unsupported 2D shadow grad function.\n");
1762                             sample_function->name = "unsupported2DGrad";
1763                         }
1764                     }
1765                     else
1766                     {
1767                         sample_function->name = projected ? "shadow2DProj" : "shadow2D";
1768                     }
1769                 }
1770                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1771             }
1772             else
1773             {
1774                 if (texrect)
1775                 {
1776                     if (lod)
1777                     {
1778                         sample_function->name = projected ? "texture2DRectProjLod" : "texture2DRectLod";
1779                     }
1780                     else if (grad)
1781                     {
1782                         if (gl_info->supported[EXT_GPU_SHADER4])
1783                             sample_function->name = projected ? "texture2DRectProjGrad" : "texture2DRectGrad";
1784                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1785                             sample_function->name = projected ? "texture2DRectProjGradARB" : "texture2DRectGradARB";
1786                         else
1787                         {
1788                             FIXME("Unsupported RECT grad function.\n");
1789                             sample_function->name = "unsupported2DRectGrad";
1790                         }
1791                     }
1792                     else
1793                     {
1794                         sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1795                     }
1796                 }
1797                 else
1798                 {
1799                     if (lod)
1800                     {
1801                         sample_function->name = projected ? "texture2DProjLod" : "texture2DLod";
1802                     }
1803                     else if (grad)
1804                     {
1805                         if (gl_info->supported[EXT_GPU_SHADER4])
1806                             sample_function->name = projected ? "texture2DProjGrad" : "texture2DGrad";
1807                         else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1808                             sample_function->name = projected ? "texture2DProjGradARB" : "texture2DGradARB";
1809                         else
1810                         {
1811                             FIXME("Unsupported 2D grad function.\n");
1812                             sample_function->name = "unsupported2DGrad";
1813                         }
1814                     }
1815                     else
1816                     {
1817                         sample_function->name = projected ? "texture2DProj" : "texture2D";
1818                     }
1819                 }
1820                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1821             }
1822             break;
1823
1824         case WINED3DSTT_CUBE:
1825             if (shadow)
1826             {
1827                 FIXME("Unsupported Cube shadow function.\n");
1828                 sample_function->name = "unsupportedCubeShadow";
1829                 sample_function->coord_mask = 0;
1830             }
1831             else
1832             {
1833                 if (lod)
1834                 {
1835                     sample_function->name = "textureCubeLod";
1836                 }
1837                 else if (grad)
1838                 {
1839                     if (gl_info->supported[EXT_GPU_SHADER4])
1840                         sample_function->name = "textureCubeGrad";
1841                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1842                         sample_function->name = "textureCubeGradARB";
1843                     else
1844                     {
1845                         FIXME("Unsupported Cube grad function.\n");
1846                         sample_function->name = "unsupportedCubeGrad";
1847                     }
1848                 }
1849                 else
1850                 {
1851                     sample_function->name = "textureCube";
1852                 }
1853                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1854             }
1855             break;
1856
1857         case WINED3DSTT_VOLUME:
1858             if (shadow)
1859             {
1860                 FIXME("Unsupported 3D shadow function.\n");
1861                 sample_function->name = "unsupported3DShadow";
1862                 sample_function->coord_mask = 0;
1863             }
1864             else
1865             {
1866                 if (lod)
1867                 {
1868                     sample_function->name = projected ? "texture3DProjLod" : "texture3DLod";
1869                 }
1870                 else  if (grad)
1871                 {
1872                     if (gl_info->supported[EXT_GPU_SHADER4])
1873                         sample_function->name = projected ? "texture3DProjGrad" : "texture3DGrad";
1874                     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1875                         sample_function->name = projected ? "texture3DProjGradARB" : "texture3DGradARB";
1876                     else
1877                     {
1878                         FIXME("Unsupported 3D grad function.\n");
1879                         sample_function->name = "unsupported3DGrad";
1880                     }
1881                 }
1882                 else
1883                 {
1884                     sample_function->name = projected ? "texture3DProj" : "texture3D";
1885                 }
1886                 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1887             }
1888             break;
1889
1890         default:
1891             sample_function->name = "";
1892             sample_function->coord_mask = 0;
1893             FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1894             break;
1895     }
1896 }
1897
1898 static void shader_glsl_append_fixup_arg(char *arguments, const char *reg_name,
1899         BOOL sign_fixup, enum fixup_channel_source channel_source)
1900 {
1901     switch(channel_source)
1902     {
1903         case CHANNEL_SOURCE_ZERO:
1904             strcat(arguments, "0.0");
1905             break;
1906
1907         case CHANNEL_SOURCE_ONE:
1908             strcat(arguments, "1.0");
1909             break;
1910
1911         case CHANNEL_SOURCE_X:
1912             strcat(arguments, reg_name);
1913             strcat(arguments, ".x");
1914             break;
1915
1916         case CHANNEL_SOURCE_Y:
1917             strcat(arguments, reg_name);
1918             strcat(arguments, ".y");
1919             break;
1920
1921         case CHANNEL_SOURCE_Z:
1922             strcat(arguments, reg_name);
1923             strcat(arguments, ".z");
1924             break;
1925
1926         case CHANNEL_SOURCE_W:
1927             strcat(arguments, reg_name);
1928             strcat(arguments, ".w");
1929             break;
1930
1931         default:
1932             FIXME("Unhandled channel source %#x\n", channel_source);
1933             strcat(arguments, "undefined");
1934             break;
1935     }
1936
1937     if (sign_fixup) strcat(arguments, " * 2.0 - 1.0");
1938 }
1939
1940 static void shader_glsl_color_correction(const struct wined3d_shader_instruction *ins, struct color_fixup_desc fixup)
1941 {
1942     struct wined3d_shader_dst_param dst;
1943     unsigned int mask_size, remaining;
1944     struct glsl_dst_param dst_param;
1945     char arguments[256];
1946     DWORD mask;
1947
1948     mask = 0;
1949     if (fixup.x_sign_fixup || fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1950     if (fixup.y_sign_fixup || fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1951     if (fixup.z_sign_fixup || fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1952     if (fixup.w_sign_fixup || fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1953     mask &= ins->dst[0].write_mask;
1954
1955     if (!mask) return; /* Nothing to do */
1956
1957     if (is_complex_fixup(fixup))
1958     {
1959         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1960         FIXME("Complex fixup (%#x) not supported\n",complex_fixup);
1961         return;
1962     }
1963
1964     mask_size = shader_glsl_get_write_mask_size(mask);
1965
1966     dst = ins->dst[0];
1967     dst.write_mask = mask;
1968     shader_glsl_add_dst_param(ins, &dst, &dst_param);
1969
1970     arguments[0] = '\0';
1971     remaining = mask_size;
1972     if (mask & WINED3DSP_WRITEMASK_0)
1973     {
1974         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.x_sign_fixup, fixup.x_source);
1975         if (--remaining) strcat(arguments, ", ");
1976     }
1977     if (mask & WINED3DSP_WRITEMASK_1)
1978     {
1979         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.y_sign_fixup, fixup.y_source);
1980         if (--remaining) strcat(arguments, ", ");
1981     }
1982     if (mask & WINED3DSP_WRITEMASK_2)
1983     {
1984         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.z_sign_fixup, fixup.z_source);
1985         if (--remaining) strcat(arguments, ", ");
1986     }
1987     if (mask & WINED3DSP_WRITEMASK_3)
1988     {
1989         shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.w_sign_fixup, fixup.w_source);
1990         if (--remaining) strcat(arguments, ", ");
1991     }
1992
1993     if (mask_size > 1)
1994     {
1995         shader_addline(ins->ctx->buffer, "%s%s = vec%u(%s);\n",
1996                 dst_param.reg_name, dst_param.mask_str, mask_size, arguments);
1997     }
1998     else
1999     {
2000         shader_addline(ins->ctx->buffer, "%s%s = %s;\n", dst_param.reg_name, dst_param.mask_str, arguments);
2001     }
2002 }
2003
2004 static void PRINTF_ATTR(8, 9) shader_glsl_gen_sample_code(const struct wined3d_shader_instruction *ins,
2005         DWORD sampler, const struct glsl_sample_function *sample_function, DWORD swizzle,
2006         const char *dx, const char *dy, const char *bias, const char *coord_reg_fmt, ...)
2007 {
2008     const char *sampler_base;
2009     char dst_swizzle[6];
2010     struct color_fixup_desc fixup;
2011     BOOL np2_fixup = FALSE;
2012     va_list args;
2013
2014     shader_glsl_swizzle_to_str(swizzle, FALSE, ins->dst[0].write_mask, dst_swizzle);
2015
2016     if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
2017     {
2018         const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2019         fixup = priv->cur_ps_args->color_fixup[sampler];
2020         sampler_base = "Psampler";
2021
2022         if(priv->cur_ps_args->np2_fixup & (1 << sampler)) {
2023             if(bias) {
2024                 FIXME("Biased sampling from NP2 textures is unsupported\n");
2025             } else {
2026                 np2_fixup = TRUE;
2027             }
2028         }
2029     } else {
2030         sampler_base = "Vsampler";
2031         fixup = COLOR_FIXUP_IDENTITY; /* FIXME: Vshader color fixup */
2032     }
2033
2034     shader_glsl_append_dst(ins->ctx->buffer, ins);
2035
2036     shader_addline(ins->ctx->buffer, "%s(%s%u, ", sample_function->name, sampler_base, sampler);
2037
2038     va_start(args, coord_reg_fmt);
2039     shader_vaddline(ins->ctx->buffer, coord_reg_fmt, args);
2040     va_end(args);
2041
2042     if(bias) {
2043         shader_addline(ins->ctx->buffer, ", %s)%s);\n", bias, dst_swizzle);
2044     } else {
2045         if (np2_fixup) {
2046             const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2047             const unsigned char idx = priv->cur_np2fixup_info->idx[sampler];
2048
2049             shader_addline(ins->ctx->buffer, " * PsamplerNP2Fixup[%u].%s)%s);\n", idx >> 1,
2050                            (idx % 2) ? "zw" : "xy", dst_swizzle);
2051         } else if(dx && dy) {
2052             shader_addline(ins->ctx->buffer, ", %s, %s)%s);\n", dx, dy, dst_swizzle);
2053         } else {
2054             shader_addline(ins->ctx->buffer, ")%s);\n", dst_swizzle);
2055         }
2056     }
2057
2058     if(!is_identity_fixup(fixup)) {
2059         shader_glsl_color_correction(ins, fixup);
2060     }
2061 }
2062
2063 /*****************************************************************************
2064  * Begin processing individual instruction opcodes
2065  ****************************************************************************/
2066
2067 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
2068 static void shader_glsl_arith(const struct wined3d_shader_instruction *ins)
2069 {
2070     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2071     struct glsl_src_param src0_param;
2072     struct glsl_src_param src1_param;
2073     DWORD write_mask;
2074     char op;
2075
2076     /* Determine the GLSL operator to use based on the opcode */
2077     switch (ins->handler_idx)
2078     {
2079         case WINED3DSIH_MUL: op = '*'; break;
2080         case WINED3DSIH_ADD: op = '+'; break;
2081         case WINED3DSIH_SUB: op = '-'; break;
2082         default:
2083             op = ' ';
2084             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2085             break;
2086     }
2087
2088     write_mask = shader_glsl_append_dst(buffer, ins);
2089     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2090     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2091     shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
2092 }
2093
2094 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
2095 static void shader_glsl_mov(const struct wined3d_shader_instruction *ins)
2096 {
2097     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
2098     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2099     struct glsl_src_param src0_param;
2100     DWORD write_mask;
2101
2102     write_mask = shader_glsl_append_dst(buffer, ins);
2103     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2104
2105     /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
2106      * shader versions WINED3DSIO_MOVA is used for this. */
2107     if (ins->ctx->reg_maps->shader_version.major == 1
2108             && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
2109             && ins->dst[0].reg.type == WINED3DSPR_ADDR)
2110     {
2111         /* This is a simple floor() */
2112         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2113         if (mask_size > 1) {
2114             shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
2115         } else {
2116             shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
2117         }
2118     }
2119     else if(ins->handler_idx == WINED3DSIH_MOVA)
2120     {
2121         /* We need to *round* to the nearest int here. */
2122         unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2123
2124         if (gl_info->supported[EXT_GPU_SHADER4])
2125         {
2126             if (mask_size > 1)
2127                 shader_addline(buffer, "ivec%d(round(%s)));\n", mask_size, src0_param.param_str);
2128             else
2129                 shader_addline(buffer, "int(round(%s)));\n", src0_param.param_str);
2130         }
2131         else
2132         {
2133             if (mask_size > 1)
2134                 shader_addline(buffer, "ivec%d(floor(abs(%s) + vec%d(0.5)) * sign(%s)));\n",
2135                         mask_size, src0_param.param_str, mask_size, src0_param.param_str);
2136             else
2137                 shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n",
2138                         src0_param.param_str, src0_param.param_str);
2139         }
2140     }
2141     else
2142     {
2143         shader_addline(buffer, "%s);\n", src0_param.param_str);
2144     }
2145 }
2146
2147 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
2148 static void shader_glsl_dot(const struct wined3d_shader_instruction *ins)
2149 {
2150     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2151     struct glsl_src_param src0_param;
2152     struct glsl_src_param src1_param;
2153     DWORD dst_write_mask, src_write_mask;
2154     unsigned int dst_size = 0;
2155
2156     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2157     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2158
2159     /* dp3 works on vec3, dp4 on vec4 */
2160     if (ins->handler_idx == WINED3DSIH_DP4)
2161     {
2162         src_write_mask = WINED3DSP_WRITEMASK_ALL;
2163     } else {
2164         src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2165     }
2166
2167     shader_glsl_add_src_param(ins, &ins->src[0], src_write_mask, &src0_param);
2168     shader_glsl_add_src_param(ins, &ins->src[1], src_write_mask, &src1_param);
2169
2170     if (dst_size > 1) {
2171         shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
2172     } else {
2173         shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
2174     }
2175 }
2176
2177 /* Note that this instruction has some restrictions. The destination write mask
2178  * can't contain the w component, and the source swizzles have to be .xyzw */
2179 static void shader_glsl_cross(const struct wined3d_shader_instruction *ins)
2180 {
2181     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2182     struct glsl_src_param src0_param;
2183     struct glsl_src_param src1_param;
2184     char dst_mask[6];
2185
2186     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2187     shader_glsl_append_dst(ins->ctx->buffer, ins);
2188     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2189     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
2190     shader_addline(ins->ctx->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
2191 }
2192
2193 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
2194  * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
2195  * GLSL uses the value as-is. */
2196 static void shader_glsl_pow(const struct wined3d_shader_instruction *ins)
2197 {
2198     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2199     struct glsl_src_param src0_param;
2200     struct glsl_src_param src1_param;
2201     DWORD dst_write_mask;
2202     unsigned int dst_size;
2203
2204     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2205     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2206
2207     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2208     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2209
2210     if (dst_size > 1)
2211     {
2212         shader_addline(buffer, "vec%u(%s == 0.0 ? 1.0 : pow(abs(%s), %s)));\n",
2213                 dst_size, src1_param.param_str, src0_param.param_str, src1_param.param_str);
2214     }
2215     else
2216     {
2217         shader_addline(buffer, "%s == 0.0 ? 1.0 : pow(abs(%s), %s));\n",
2218                 src1_param.param_str, src0_param.param_str, src1_param.param_str);
2219     }
2220 }
2221
2222 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
2223  * Src0 is a scalar. Note that D3D uses the absolute of src0, while
2224  * GLSL uses the value as-is. */
2225 static void shader_glsl_log(const struct wined3d_shader_instruction *ins)
2226 {
2227     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2228     struct glsl_src_param src0_param;
2229     DWORD dst_write_mask;
2230     unsigned int dst_size;
2231
2232     dst_write_mask = shader_glsl_append_dst(buffer, ins);
2233     dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2234
2235     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2236
2237     if (dst_size > 1)
2238     {
2239         shader_addline(buffer, "vec%u(log2(abs(%s))));\n",
2240                 dst_size, src0_param.param_str);
2241     }
2242     else
2243     {
2244         shader_addline(buffer, "log2(abs(%s)));\n",
2245                 src0_param.param_str);
2246     }
2247 }
2248
2249 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
2250 static void shader_glsl_map2gl(const struct wined3d_shader_instruction *ins)
2251 {
2252     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2253     struct glsl_src_param src_param;
2254     const char *instruction;
2255     DWORD write_mask;
2256     unsigned i;
2257
2258     /* Determine the GLSL function to use based on the opcode */
2259     /* TODO: Possibly make this a table for faster lookups */
2260     switch (ins->handler_idx)
2261     {
2262         case WINED3DSIH_MIN: instruction = "min"; break;
2263         case WINED3DSIH_MAX: instruction = "max"; break;
2264         case WINED3DSIH_ABS: instruction = "abs"; break;
2265         case WINED3DSIH_FRC: instruction = "fract"; break;
2266         case WINED3DSIH_EXP: instruction = "exp2"; break;
2267         case WINED3DSIH_DSX: instruction = "dFdx"; break;
2268         case WINED3DSIH_DSY: instruction = "ycorrection.y * dFdy"; break;
2269         default: instruction = "";
2270             FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2271             break;
2272     }
2273
2274     write_mask = shader_glsl_append_dst(buffer, ins);
2275
2276     shader_addline(buffer, "%s(", instruction);
2277
2278     if (ins->src_count)
2279     {
2280         shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2281         shader_addline(buffer, "%s", src_param.param_str);
2282         for (i = 1; i < ins->src_count; ++i)
2283         {
2284             shader_glsl_add_src_param(ins, &ins->src[i], write_mask, &src_param);
2285             shader_addline(buffer, ", %s", src_param.param_str);
2286         }
2287     }
2288
2289     shader_addline(buffer, "));\n");
2290 }
2291
2292 static void shader_glsl_nrm(const struct wined3d_shader_instruction *ins)
2293 {
2294     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2295     struct glsl_src_param src_param;
2296     unsigned int mask_size;
2297     DWORD write_mask;
2298     char dst_mask[6];
2299
2300     write_mask = shader_glsl_get_write_mask(ins->dst, dst_mask);
2301     mask_size = shader_glsl_get_write_mask_size(write_mask);
2302     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2303
2304     shader_addline(buffer, "tmp0.x = dot(%s, %s);\n",
2305             src_param.param_str, src_param.param_str);
2306     shader_glsl_append_dst(buffer, ins);
2307
2308     if (mask_size > 1)
2309     {
2310         shader_addline(buffer, "tmp0.x == 0.0 ? vec%u(0.0) : (%s * inversesqrt(tmp0.x)));\n",
2311                 mask_size, src_param.param_str);
2312     }
2313     else
2314     {
2315         shader_addline(buffer, "tmp0.x == 0.0 ? 0.0 : (%s * inversesqrt(tmp0.x)));\n",
2316                 src_param.param_str);
2317     }
2318 }
2319
2320 /** Process the WINED3DSIO_EXPP instruction in GLSL:
2321  * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
2322  *   dst.x = 2^(floor(src))
2323  *   dst.y = src - floor(src)
2324  *   dst.z = 2^src   (partial precision is allowed, but optional)
2325  *   dst.w = 1.0;
2326  * For 2.0 shaders, just do this (honoring writemask and swizzle):
2327  *   dst = 2^src;    (partial precision is allowed, but optional)
2328  */
2329 static void shader_glsl_expp(const struct wined3d_shader_instruction *ins)
2330 {
2331     struct glsl_src_param src_param;
2332
2333     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src_param);
2334
2335     if (ins->ctx->reg_maps->shader_version.major < 2)
2336     {
2337         char dst_mask[6];
2338
2339         shader_addline(ins->ctx->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
2340         shader_addline(ins->ctx->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
2341         shader_addline(ins->ctx->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
2342         shader_addline(ins->ctx->buffer, "tmp0.w = 1.0;\n");
2343
2344         shader_glsl_append_dst(ins->ctx->buffer, ins);
2345         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2346         shader_addline(ins->ctx->buffer, "tmp0%s);\n", dst_mask);
2347     } else {
2348         DWORD write_mask;
2349         unsigned int mask_size;
2350
2351         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2352         mask_size = shader_glsl_get_write_mask_size(write_mask);
2353
2354         if (mask_size > 1) {
2355             shader_addline(ins->ctx->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
2356         } else {
2357             shader_addline(ins->ctx->buffer, "exp2(%s));\n", src_param.param_str);
2358         }
2359     }
2360 }
2361
2362 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
2363 static void shader_glsl_rcp(const struct wined3d_shader_instruction *ins)
2364 {
2365     struct glsl_src_param src_param;
2366     DWORD write_mask;
2367     unsigned int mask_size;
2368
2369     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2370     mask_size = shader_glsl_get_write_mask_size(write_mask);
2371     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2372
2373     if (mask_size > 1)
2374     {
2375         shader_addline(ins->ctx->buffer, "vec%u(1.0 / %s));\n",
2376                 mask_size, src_param.param_str);
2377     }
2378     else
2379     {
2380         shader_addline(ins->ctx->buffer, "1.0 / %s);\n",
2381                 src_param.param_str);
2382     }
2383 }
2384
2385 static void shader_glsl_rsq(const struct wined3d_shader_instruction *ins)
2386 {
2387     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2388     struct glsl_src_param src_param;
2389     DWORD write_mask;
2390     unsigned int mask_size;
2391
2392     write_mask = shader_glsl_append_dst(buffer, ins);
2393     mask_size = shader_glsl_get_write_mask_size(write_mask);
2394
2395     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2396
2397     if (mask_size > 1)
2398     {
2399         shader_addline(buffer, "vec%u(inversesqrt(abs(%s))));\n",
2400                 mask_size, src_param.param_str);
2401     }
2402     else
2403     {
2404         shader_addline(buffer, "inversesqrt(abs(%s)));\n",
2405                 src_param.param_str);
2406     }
2407 }
2408
2409 /** Process signed comparison opcodes in GLSL. */
2410 static void shader_glsl_compare(const struct wined3d_shader_instruction *ins)
2411 {
2412     struct glsl_src_param src0_param;
2413     struct glsl_src_param src1_param;
2414     DWORD write_mask;
2415     unsigned int mask_size;
2416
2417     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2418     mask_size = shader_glsl_get_write_mask_size(write_mask);
2419     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2420     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2421
2422     if (mask_size > 1) {
2423         const char *compare;
2424
2425         switch(ins->handler_idx)
2426         {
2427             case WINED3DSIH_SLT: compare = "lessThan"; break;
2428             case WINED3DSIH_SGE: compare = "greaterThanEqual"; break;
2429             default: compare = "";
2430                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2431         }
2432
2433         shader_addline(ins->ctx->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
2434                 src0_param.param_str, src1_param.param_str);
2435     } else {
2436         switch(ins->handler_idx)
2437         {
2438             case WINED3DSIH_SLT:
2439                 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
2440                  * to return 0.0 but step returns 1.0 because step is not < x
2441                  * An alternative is a bvec compare padded with an unused second component.
2442                  * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
2443                  * issue. Playing with not() is not possible either because not() does not accept
2444                  * a scalar.
2445                  */
2446                 shader_addline(ins->ctx->buffer, "(%s < %s) ? 1.0 : 0.0);\n",
2447                         src0_param.param_str, src1_param.param_str);
2448                 break;
2449             case WINED3DSIH_SGE:
2450                 /* Here we can use the step() function and safe a conditional */
2451                 shader_addline(ins->ctx->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
2452                 break;
2453             default:
2454                 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2455         }
2456
2457     }
2458 }
2459
2460 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
2461 static void shader_glsl_cmp(const struct wined3d_shader_instruction *ins)
2462 {
2463     struct glsl_src_param src0_param;
2464     struct glsl_src_param src1_param;
2465     struct glsl_src_param src2_param;
2466     DWORD write_mask, cmp_channel = 0;
2467     unsigned int i, j;
2468     char mask_char[6];
2469     BOOL temp_destination = FALSE;
2470
2471     if (shader_is_scalar(&ins->src[0].reg))
2472     {
2473         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2474
2475         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2476         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2477         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2478
2479         shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2480                        src0_param.param_str, src1_param.param_str, src2_param.param_str);
2481     } else {
2482         DWORD dst_mask = ins->dst[0].write_mask;
2483         struct wined3d_shader_dst_param dst = ins->dst[0];
2484
2485         /* Cycle through all source0 channels */
2486         for (i=0; i<4; i++) {
2487             write_mask = 0;
2488             /* Find the destination channels which use the current source0 channel */
2489             for (j=0; j<4; j++) {
2490                 if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2491                 {
2492                     write_mask |= WINED3DSP_WRITEMASK_0 << j;
2493                     cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2494                 }
2495             }
2496             dst.write_mask = dst_mask & write_mask;
2497
2498             /* Splitting the cmp instruction up in multiple lines imposes a problem:
2499             * The first lines may overwrite source parameters of the following lines.
2500             * Deal with that by using a temporary destination register if needed
2501             */
2502             if ((ins->src[0].reg.idx == ins->dst[0].reg.idx
2503                     && ins->src[0].reg.type == ins->dst[0].reg.type)
2504                     || (ins->src[1].reg.idx == ins->dst[0].reg.idx
2505                     && ins->src[1].reg.type == ins->dst[0].reg.type)
2506                     || (ins->src[2].reg.idx == ins->dst[0].reg.idx
2507                     && ins->src[2].reg.type == ins->dst[0].reg.type))
2508             {
2509                 write_mask = shader_glsl_get_write_mask(&dst, mask_char);
2510                 if (!write_mask) continue;
2511                 shader_addline(ins->ctx->buffer, "tmp0%s = (", mask_char);
2512                 temp_destination = TRUE;
2513             } else {
2514                 write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2515                 if (!write_mask) continue;
2516             }
2517
2518             shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2519             shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2520             shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2521
2522             shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2523                         src0_param.param_str, src1_param.param_str, src2_param.param_str);
2524         }
2525
2526         if(temp_destination) {
2527             shader_glsl_get_write_mask(&ins->dst[0], mask_char);
2528             shader_glsl_append_dst(ins->ctx->buffer, ins);
2529             shader_addline(ins->ctx->buffer, "tmp0%s);\n", mask_char);
2530         }
2531     }
2532
2533 }
2534
2535 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
2536 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
2537  * the compare is done per component of src0. */
2538 static void shader_glsl_cnd(const struct wined3d_shader_instruction *ins)
2539 {
2540     struct wined3d_shader_dst_param dst;
2541     struct glsl_src_param src0_param;
2542     struct glsl_src_param src1_param;
2543     struct glsl_src_param src2_param;
2544     DWORD write_mask, cmp_channel = 0;
2545     unsigned int i, j;
2546     DWORD dst_mask;
2547     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2548             ins->ctx->reg_maps->shader_version.minor);
2549
2550     if (shader_version < WINED3D_SHADER_VERSION(1, 4))
2551     {
2552         write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2553         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2554         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2555         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2556
2557         /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
2558         if (ins->coissue)
2559         {
2560             shader_addline(ins->ctx->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
2561         } else {
2562             shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2563                     src0_param.param_str, src1_param.param_str, src2_param.param_str);
2564         }
2565         return;
2566     }
2567     /* Cycle through all source0 channels */
2568     dst_mask = ins->dst[0].write_mask;
2569     dst = ins->dst[0];
2570     for (i=0; i<4; i++) {
2571         write_mask = 0;
2572         /* Find the destination channels which use the current source0 channel */
2573         for (j=0; j<4; j++) {
2574             if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2575             {
2576                 write_mask |= WINED3DSP_WRITEMASK_0 << j;
2577                 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2578             }
2579         }
2580
2581         dst.write_mask = dst_mask & write_mask;
2582         write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2583         if (!write_mask) continue;
2584
2585         shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2586         shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2587         shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2588
2589         shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2590                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2591     }
2592 }
2593
2594 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
2595 static void shader_glsl_mad(const struct wined3d_shader_instruction *ins)
2596 {
2597     struct glsl_src_param src0_param;
2598     struct glsl_src_param src1_param;
2599     struct glsl_src_param src2_param;
2600     DWORD write_mask;
2601
2602     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2603     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2604     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2605     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2606     shader_addline(ins->ctx->buffer, "(%s * %s) + %s);\n",
2607             src0_param.param_str, src1_param.param_str, src2_param.param_str);
2608 }
2609
2610 /* Handles transforming all WINED3DSIO_M?x? opcodes for
2611    Vertex shaders to GLSL codes */
2612 static void shader_glsl_mnxn(const struct wined3d_shader_instruction *ins)
2613 {
2614     int i;
2615     int nComponents = 0;
2616     struct wined3d_shader_dst_param tmp_dst = {{0}};
2617     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2618     struct wined3d_shader_instruction tmp_ins;
2619
2620     memset(&tmp_ins, 0, sizeof(tmp_ins));
2621
2622     /* Set constants for the temporary argument */
2623     tmp_ins.ctx = ins->ctx;
2624     tmp_ins.dst_count = 1;
2625     tmp_ins.dst = &tmp_dst;
2626     tmp_ins.src_count = 2;
2627     tmp_ins.src = tmp_src;
2628
2629     switch(ins->handler_idx)
2630     {
2631         case WINED3DSIH_M4x4:
2632             nComponents = 4;
2633             tmp_ins.handler_idx = WINED3DSIH_DP4;
2634             break;
2635         case WINED3DSIH_M4x3:
2636             nComponents = 3;
2637             tmp_ins.handler_idx = WINED3DSIH_DP4;
2638             break;
2639         case WINED3DSIH_M3x4:
2640             nComponents = 4;
2641             tmp_ins.handler_idx = WINED3DSIH_DP3;
2642             break;
2643         case WINED3DSIH_M3x3:
2644             nComponents = 3;
2645             tmp_ins.handler_idx = WINED3DSIH_DP3;
2646             break;
2647         case WINED3DSIH_M3x2:
2648             nComponents = 2;
2649             tmp_ins.handler_idx = WINED3DSIH_DP3;
2650             break;
2651         default:
2652             break;
2653     }
2654
2655     tmp_dst = ins->dst[0];
2656     tmp_src[0] = ins->src[0];
2657     tmp_src[1] = ins->src[1];
2658     for (i = 0; i < nComponents; ++i)
2659     {
2660         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2661         shader_glsl_dot(&tmp_ins);
2662         ++tmp_src[1].reg.idx;
2663     }
2664 }
2665
2666 /**
2667     The LRP instruction performs a component-wise linear interpolation
2668     between the second and third operands using the first operand as the
2669     blend factor.  Equation:  (dst = src2 + src0 * (src1 - src2))
2670     This is equivalent to mix(src2, src1, src0);
2671 */
2672 static void shader_glsl_lrp(const struct wined3d_shader_instruction *ins)
2673 {
2674     struct glsl_src_param src0_param;
2675     struct glsl_src_param src1_param;
2676     struct glsl_src_param src2_param;
2677     DWORD write_mask;
2678
2679     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2680
2681     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2682     shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2683     shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2684
2685     shader_addline(ins->ctx->buffer, "mix(%s, %s, %s));\n",
2686             src2_param.param_str, src1_param.param_str, src0_param.param_str);
2687 }
2688
2689 /** Process the WINED3DSIO_LIT instruction in GLSL:
2690  * dst.x = dst.w = 1.0
2691  * dst.y = (src0.x > 0) ? src0.x
2692  * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
2693  *                                        where src.w is clamped at +- 128
2694  */
2695 static void shader_glsl_lit(const struct wined3d_shader_instruction *ins)
2696 {
2697     struct glsl_src_param src0_param;
2698     struct glsl_src_param src1_param;
2699     struct glsl_src_param src3_param;
2700     char dst_mask[6];
2701
2702     shader_glsl_append_dst(ins->ctx->buffer, ins);
2703     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2704
2705     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2706     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src1_param);
2707     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src3_param);
2708
2709     /* The sdk specifies the instruction like this
2710      * dst.x = 1.0;
2711      * if(src.x > 0.0) dst.y = src.x
2712      * else dst.y = 0.0.
2713      * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
2714      * else dst.z = 0.0;
2715      * dst.w = 1.0;
2716      * (where power = src.w clamped between -128 and 128)
2717      *
2718      * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
2719      * dst.x = 1.0                                  ... No further explanation needed
2720      * dst.y = max(src.y, 0.0);                     ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
2721      * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0;   ... 0 ^ power is 0, and otherwise we use y anyway
2722      * dst.w = 1.0.                                 ... Nothing fancy.
2723      *
2724      * So we still have one conditional in there. So do this:
2725      * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
2726      *
2727      * 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),
2728      * 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.
2729      * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to.
2730      *
2731      * Unfortunately pow(0.0 ^ 0.0) returns NaN on most GPUs, but lit with src.y = 0 and src.w = 0 returns
2732      * a non-NaN value in dst.z. What we return doesn't matter, as long as it is not NaN. Return 0, which is
2733      * what all Windows HW drivers and GL_ARB_vertex_program's LIT do.
2734      */
2735     shader_addline(ins->ctx->buffer,
2736             "vec4(1.0, max(%s, 0.0), %s == 0.0 ? 0.0 : "
2737             "pow(max(0.0, %s) * step(0.0, %s), clamp(%s, -128.0, 128.0)), 1.0)%s);\n",
2738             src0_param.param_str, src3_param.param_str, src1_param.param_str,
2739             src0_param.param_str, src3_param.param_str, dst_mask);
2740 }
2741
2742 /** Process the WINED3DSIO_DST instruction in GLSL:
2743  * dst.x = 1.0
2744  * dst.y = src0.x * src0.y
2745  * dst.z = src0.z
2746  * dst.w = src1.w
2747  */
2748 static void shader_glsl_dst(const struct wined3d_shader_instruction *ins)
2749 {
2750     struct glsl_src_param src0y_param;
2751     struct glsl_src_param src0z_param;
2752     struct glsl_src_param src1y_param;
2753     struct glsl_src_param src1w_param;
2754     char dst_mask[6];
2755
2756     shader_glsl_append_dst(ins->ctx->buffer, ins);
2757     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2758
2759     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src0y_param);
2760     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &src0z_param);
2761     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_1, &src1y_param);
2762     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_3, &src1w_param);
2763
2764     shader_addline(ins->ctx->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
2765             src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
2766 }
2767
2768 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
2769  * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
2770  * can handle it.  But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
2771  *
2772  * dst.x = cos(src0.?)
2773  * dst.y = sin(src0.?)
2774  * dst.z = dst.z
2775  * dst.w = dst.w
2776  */
2777 static void shader_glsl_sincos(const struct wined3d_shader_instruction *ins)
2778 {
2779     struct glsl_src_param src0_param;
2780     DWORD write_mask;
2781
2782     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2783     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2784
2785     switch (write_mask) {
2786         case WINED3DSP_WRITEMASK_0:
2787             shader_addline(ins->ctx->buffer, "cos(%s));\n", src0_param.param_str);
2788             break;
2789
2790         case WINED3DSP_WRITEMASK_1:
2791             shader_addline(ins->ctx->buffer, "sin(%s));\n", src0_param.param_str);
2792             break;
2793
2794         case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
2795             shader_addline(ins->ctx->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
2796             break;
2797
2798         default:
2799             ERR("Write mask should be .x, .y or .xy\n");
2800             break;
2801     }
2802 }
2803
2804 /* sgn in vs_2_0 has 2 extra parameters(registers for temporary storage) which we don't use
2805  * here. But those extra parameters require a dedicated function for sgn, since map2gl would
2806  * generate invalid code
2807  */
2808 static void shader_glsl_sgn(const struct wined3d_shader_instruction *ins)
2809 {
2810     struct glsl_src_param src0_param;
2811     DWORD write_mask;
2812
2813     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2814     shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2815
2816     shader_addline(ins->ctx->buffer, "sign(%s));\n", src0_param.param_str);
2817 }
2818
2819 /** Process the WINED3DSIO_LOOP instruction in GLSL:
2820  * Start a for() loop where src1.y is the initial value of aL,
2821  *  increment aL by src1.z for a total of src1.x iterations.
2822  *  Need to use a temporary variable for this operation.
2823  */
2824 /* FIXME: I don't think nested loops will work correctly this way. */
2825 static void shader_glsl_loop(const struct wined3d_shader_instruction *ins)
2826 {
2827     struct wined3d_shader_loop_state *loop_state = ins->ctx->loop_state;
2828     const struct wined3d_shader *shader = ins->ctx->shader;
2829     const struct wined3d_shader_lconst *constant;
2830     struct glsl_src_param src1_param;
2831     const DWORD *control_values = NULL;
2832
2833     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2834
2835     /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2836      * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2837      * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2838      * addressing.
2839      */
2840     if (ins->src[1].reg.type == WINED3DSPR_CONSTINT)
2841     {
2842         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
2843         {
2844             if (constant->idx == ins->src[1].reg.idx)
2845             {
2846                 control_values = constant->value;
2847                 break;
2848             }
2849         }
2850     }
2851
2852     if (control_values)
2853     {
2854         struct wined3d_shader_loop_control loop_control;
2855         loop_control.count = control_values[0];
2856         loop_control.start = control_values[1];
2857         loop_control.step = (int)control_values[2];
2858
2859         if (loop_control.step > 0)
2860         {
2861             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u < (%u * %d + %u); aL%u += %d) {\n",
2862                     loop_state->current_depth, loop_control.start,
2863                     loop_state->current_depth, loop_control.count, loop_control.step, loop_control.start,
2864                     loop_state->current_depth, loop_control.step);
2865         }
2866         else if (loop_control.step < 0)
2867         {
2868             shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u > (%u * %d + %u); aL%u += %d) {\n",
2869                     loop_state->current_depth, loop_control.start,
2870                     loop_state->current_depth, loop_control.count, loop_control.step, loop_control.start,
2871                     loop_state->current_depth, loop_control.step);
2872         }
2873         else
2874         {
2875             shader_addline(ins->ctx->buffer, "for (aL%u = %u, tmpInt%u = 0; tmpInt%u < %u; tmpInt%u++) {\n",
2876                     loop_state->current_depth, loop_control.start, loop_state->current_depth,
2877                     loop_state->current_depth, loop_control.count,
2878                     loop_state->current_depth);
2879         }
2880     } else {
2881         shader_addline(ins->ctx->buffer,
2882                 "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2883                 loop_state->current_depth, loop_state->current_reg,
2884                 src1_param.reg_name, loop_state->current_depth, src1_param.reg_name,
2885                 loop_state->current_depth, loop_state->current_reg, src1_param.reg_name);
2886     }
2887
2888     ++loop_state->current_depth;
2889     ++loop_state->current_reg;
2890 }
2891
2892 static void shader_glsl_end(const struct wined3d_shader_instruction *ins)
2893 {
2894     struct wined3d_shader_loop_state *loop_state = ins->ctx->loop_state;
2895
2896     shader_addline(ins->ctx->buffer, "}\n");
2897
2898     if (ins->handler_idx == WINED3DSIH_ENDLOOP)
2899     {
2900         --loop_state->current_depth;
2901         --loop_state->current_reg;
2902     }
2903
2904     if (ins->handler_idx == WINED3DSIH_ENDREP)
2905     {
2906         --loop_state->current_depth;
2907     }
2908 }
2909
2910 static void shader_glsl_rep(const struct wined3d_shader_instruction *ins)
2911 {
2912     const struct wined3d_shader *shader = ins->ctx->shader;
2913     struct wined3d_shader_loop_state *loop_state = ins->ctx->loop_state;
2914     const struct wined3d_shader_lconst *constant;
2915     struct glsl_src_param src0_param;
2916     const DWORD *control_values = NULL;
2917
2918     /* Try to hardcode local values to help the GLSL compiler to unroll and optimize the loop */
2919     if (ins->src[0].reg.type == WINED3DSPR_CONSTINT)
2920     {
2921         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
2922         {
2923             if (constant->idx == ins->src[0].reg.idx)
2924             {
2925                 control_values = constant->value;
2926                 break;
2927             }
2928         }
2929     }
2930
2931     if (control_values)
2932     {
2933         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %d; tmpInt%d++) {\n",
2934                 loop_state->current_depth, loop_state->current_depth,
2935                 control_values[0], loop_state->current_depth);
2936     }
2937     else
2938     {
2939         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2940         shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2941                 loop_state->current_depth, loop_state->current_depth,
2942                 src0_param.param_str, loop_state->current_depth);
2943     }
2944
2945     ++loop_state->current_depth;
2946 }
2947
2948 static void shader_glsl_if(const struct wined3d_shader_instruction *ins)
2949 {
2950     struct glsl_src_param src0_param;
2951
2952     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2953     shader_addline(ins->ctx->buffer, "if (%s) {\n", src0_param.param_str);
2954 }
2955
2956 static void shader_glsl_ifc(const struct wined3d_shader_instruction *ins)
2957 {
2958     struct glsl_src_param src0_param;
2959     struct glsl_src_param src1_param;
2960
2961     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2962     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2963
2964     shader_addline(ins->ctx->buffer, "if (%s %s %s) {\n",
2965             src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2966 }
2967
2968 static void shader_glsl_else(const struct wined3d_shader_instruction *ins)
2969 {
2970     shader_addline(ins->ctx->buffer, "} else {\n");
2971 }
2972
2973 static void shader_glsl_break(const struct wined3d_shader_instruction *ins)
2974 {
2975     shader_addline(ins->ctx->buffer, "break;\n");
2976 }
2977
2978 /* FIXME: According to MSDN the compare is done per component. */
2979 static void shader_glsl_breakc(const struct wined3d_shader_instruction *ins)
2980 {
2981     struct glsl_src_param src0_param;
2982     struct glsl_src_param src1_param;
2983
2984     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2985     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2986
2987     shader_addline(ins->ctx->buffer, "if (%s %s %s) break;\n",
2988             src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2989 }
2990
2991 static void shader_glsl_label(const struct wined3d_shader_instruction *ins)
2992 {
2993     shader_addline(ins->ctx->buffer, "}\n");
2994     shader_addline(ins->ctx->buffer, "void subroutine%u () {\n",  ins->src[0].reg.idx);
2995 }
2996
2997 static void shader_glsl_call(const struct wined3d_shader_instruction *ins)
2998 {
2999     shader_addline(ins->ctx->buffer, "subroutine%u();\n", ins->src[0].reg.idx);
3000 }
3001
3002 static void shader_glsl_callnz(const struct wined3d_shader_instruction *ins)
3003 {
3004     struct glsl_src_param src1_param;
3005
3006     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
3007     shader_addline(ins->ctx->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, ins->src[0].reg.idx);
3008 }
3009
3010 static void shader_glsl_ret(const struct wined3d_shader_instruction *ins)
3011 {
3012     /* No-op. The closing } is written when a new function is started, and at the end of the shader. This
3013      * function only suppresses the unhandled instruction warning
3014      */
3015 }
3016
3017 /*********************************************
3018  * Pixel Shader Specific Code begins here
3019  ********************************************/
3020 static void shader_glsl_tex(const struct wined3d_shader_instruction *ins)
3021 {
3022     const struct wined3d_shader *shader = ins->ctx->shader;
3023     struct wined3d_device *device = shader->device;
3024     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
3025             ins->ctx->reg_maps->shader_version.minor);
3026     struct glsl_sample_function sample_function;
3027     const struct wined3d_texture *texture;
3028     DWORD sample_flags = 0;
3029     DWORD sampler_idx;
3030     DWORD mask = 0, swizzle;
3031
3032     /* 1.0-1.4: Use destination register as sampler source.
3033      * 2.0+: Use provided sampler source. */
3034     if (shader_version < WINED3D_SHADER_VERSION(2,0)) sampler_idx = ins->dst[0].reg.idx;
3035     else sampler_idx = ins->src[1].reg.idx;
3036     texture = device->stateBlock->state.textures[sampler_idx];
3037
3038     if (shader_version < WINED3D_SHADER_VERSION(1,4))
3039     {
3040         const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
3041         DWORD flags = (priv->cur_ps_args->tex_transform >> sampler_idx * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
3042                 & WINED3D_PSARGS_TEXTRANSFORM_MASK;
3043         WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3044
3045         /* Projected cube textures don't make a lot of sense, the resulting coordinates stay the same. */
3046         if (flags & WINED3D_PSARGS_PROJECTED && sampler_type != WINED3DSTT_CUBE) {
3047             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3048             switch (flags & ~WINED3D_PSARGS_PROJECTED) {
3049                 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3050                 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
3051                 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
3052                 case WINED3DTTFF_COUNT4:
3053                 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
3054             }
3055         }
3056     }
3057     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
3058     {
3059         DWORD src_mod = ins->src[0].modifiers;
3060
3061         if (src_mod == WINED3DSPSM_DZ) {
3062             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3063             mask = WINED3DSP_WRITEMASK_2;
3064         } else if (src_mod == WINED3DSPSM_DW) {
3065             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3066             mask = WINED3DSP_WRITEMASK_3;
3067         }
3068     } else {
3069         if (ins->flags & WINED3DSI_TEXLD_PROJECT)
3070         {
3071             /* ps 2.0 texldp instruction always divides by the fourth component. */
3072             sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3073             mask = WINED3DSP_WRITEMASK_3;
3074         }
3075     }
3076
3077     if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
3078         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3079
3080     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3081     mask |= sample_function.coord_mask;
3082
3083     if (shader_version < WINED3D_SHADER_VERSION(2,0)) swizzle = WINED3DSP_NOSWIZZLE;
3084     else swizzle = ins->src[1].swizzle;
3085
3086     /* 1.0-1.3: Use destination register as coordinate source.
3087        1.4+: Use provided coordinate source register. */
3088     if (shader_version < WINED3D_SHADER_VERSION(1,4))
3089     {
3090         char coord_mask[6];
3091         shader_glsl_write_mask_to_str(mask, coord_mask);
3092         shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3093                 "T%u%s", sampler_idx, coord_mask);
3094     }
3095     else
3096     {
3097         struct glsl_src_param coord_param;
3098         shader_glsl_add_src_param(ins, &ins->src[0], mask, &coord_param);
3099         if (ins->flags & WINED3DSI_TEXLD_BIAS)
3100         {
3101             struct glsl_src_param bias;
3102             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &bias);
3103             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, bias.param_str,
3104                     "%s", coord_param.param_str);
3105         } else {
3106             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3107                     "%s", coord_param.param_str);
3108         }
3109     }
3110 }
3111
3112 static void shader_glsl_texldd(const struct wined3d_shader_instruction *ins)
3113 {
3114     const struct wined3d_shader *shader = ins->ctx->shader;
3115     struct wined3d_device *device = shader->device;
3116     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3117     struct glsl_src_param coord_param, dx_param, dy_param;
3118     DWORD sample_flags = WINED3D_GLSL_SAMPLE_GRAD;
3119     struct glsl_sample_function sample_function;
3120     DWORD sampler_idx;
3121     DWORD swizzle = ins->src[1].swizzle;
3122     const struct wined3d_texture *texture;
3123
3124     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4])
3125     {
3126         FIXME("texldd used, but not supported by hardware. Falling back to regular tex\n");
3127         shader_glsl_tex(ins);
3128         return;
3129     }
3130
3131     sampler_idx = ins->src[1].reg.idx;
3132     texture = device->stateBlock->state.textures[sampler_idx];
3133     if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
3134         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3135
3136     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3137     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3138     shader_glsl_add_src_param(ins, &ins->src[2], sample_function.coord_mask, &dx_param);
3139     shader_glsl_add_src_param(ins, &ins->src[3], sample_function.coord_mask, &dy_param);
3140
3141     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, dx_param.param_str, dy_param.param_str, NULL,
3142                                 "%s", coord_param.param_str);
3143 }
3144
3145 static void shader_glsl_texldl(const struct wined3d_shader_instruction *ins)
3146 {
3147     const struct wined3d_shader *shader = ins->ctx->shader;
3148     struct wined3d_device *device = shader->device;
3149     const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3150     struct glsl_src_param coord_param, lod_param;
3151     DWORD sample_flags = WINED3D_GLSL_SAMPLE_LOD;
3152     struct glsl_sample_function sample_function;
3153     DWORD sampler_idx;
3154     DWORD swizzle = ins->src[1].swizzle;
3155     const struct wined3d_texture *texture;
3156
3157     sampler_idx = ins->src[1].reg.idx;
3158     texture = device->stateBlock->state.textures[sampler_idx];
3159     if (texture && texture->target == GL_TEXTURE_RECTANGLE_ARB)
3160         sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3161
3162     shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3163     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3164
3165     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &lod_param);
3166
3167     if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4]
3168             && shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
3169     {
3170         /* Plain GLSL only supports Lod sampling functions in vertex shaders.
3171          * However, the NVIDIA drivers allow them in fragment shaders as well,
3172          * even without the appropriate extension. */
3173         WARN("Using %s in fragment shader.\n", sample_function.name);
3174     }
3175     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, lod_param.param_str,
3176             "%s", coord_param.param_str);
3177 }
3178
3179 static void shader_glsl_texcoord(const struct wined3d_shader_instruction *ins)
3180 {
3181     /* FIXME: Make this work for more than just 2D textures */
3182     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3183     DWORD write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3184
3185     if (!(ins->ctx->reg_maps->shader_version.major == 1 && ins->ctx->reg_maps->shader_version.minor == 4))
3186     {
3187         char dst_mask[6];
3188
3189         shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3190         shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n",
3191                 ins->dst[0].reg.idx, dst_mask);
3192     } else {
3193         DWORD reg = ins->src[0].reg.idx;
3194         DWORD src_mod = ins->src[0].modifiers;
3195         char dst_swizzle[6];
3196
3197         shader_glsl_get_swizzle(&ins->src[0], FALSE, write_mask, dst_swizzle);
3198
3199         if (src_mod == WINED3DSPSM_DZ)
3200         {
3201             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3202             struct glsl_src_param div_param;
3203
3204             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &div_param);
3205
3206             if (mask_size > 1) {
3207                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3208             } else {
3209                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3210             }
3211         }
3212         else if (src_mod == WINED3DSPSM_DW)
3213         {
3214             unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3215             struct glsl_src_param div_param;
3216
3217             shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &div_param);
3218
3219             if (mask_size > 1) {
3220                 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3221             } else {
3222                 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3223             }
3224         } else {
3225             shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
3226         }
3227     }
3228 }
3229
3230 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
3231  * Take a 3-component dot product of the TexCoord[dstreg] and src,
3232  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
3233 static void shader_glsl_texdp3tex(const struct wined3d_shader_instruction *ins)
3234 {
3235     DWORD sampler_idx = ins->dst[0].reg.idx;
3236     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3237     struct glsl_sample_function sample_function;
3238     struct glsl_src_param src0_param;
3239     UINT mask_size;
3240
3241     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3242
3243     /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
3244      * scalar, and projected sampling would require 4.
3245      *
3246      * It is a dependent read - not valid with conditional NP2 textures
3247      */
3248     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3249     mask_size = shader_glsl_get_write_mask_size(sample_function.coord_mask);
3250
3251     switch(mask_size)
3252     {
3253         case 1:
3254             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3255                     "dot(gl_TexCoord[%u].xyz, %s)", sampler_idx, src0_param.param_str);
3256             break;
3257
3258         case 2:
3259             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3260                     "vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0)", sampler_idx, src0_param.param_str);
3261             break;
3262
3263         case 3:
3264             shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3265                     "vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0)", sampler_idx, src0_param.param_str);
3266             break;
3267
3268         default:
3269             FIXME("Unexpected mask size %u\n", mask_size);
3270             break;
3271     }
3272 }
3273
3274 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
3275  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
3276 static void shader_glsl_texdp3(const struct wined3d_shader_instruction *ins)
3277 {
3278     DWORD dstreg = ins->dst[0].reg.idx;
3279     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3280     struct glsl_src_param src0_param;
3281     DWORD dst_mask;
3282     unsigned int mask_size;
3283
3284     dst_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3285     mask_size = shader_glsl_get_write_mask_size(dst_mask);
3286     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3287
3288     if (mask_size > 1) {
3289         shader_addline(ins->ctx->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
3290     } else {
3291         shader_addline(ins->ctx->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
3292     }
3293 }
3294
3295 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
3296  * Calculate the depth as dst.x / dst.y   */
3297 static void shader_glsl_texdepth(const struct wined3d_shader_instruction *ins)
3298 {
3299     struct glsl_dst_param dst_param;
3300
3301     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3302
3303     /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
3304      * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
3305      * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
3306      * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
3307      * >= 1.0 or < 0.0
3308      */
3309     shader_addline(ins->ctx->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n",
3310             dst_param.reg_name, dst_param.reg_name);
3311 }
3312
3313 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
3314  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
3315  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
3316  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
3317  */
3318 static void shader_glsl_texm3x2depth(const struct wined3d_shader_instruction *ins)
3319 {
3320     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3321     DWORD dstreg = ins->dst[0].reg.idx;
3322     struct glsl_src_param src0_param;
3323
3324     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3325
3326     shader_addline(ins->ctx->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
3327     shader_addline(ins->ctx->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
3328 }
3329
3330 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
3331  * Calculate the 1st of a 2-row matrix multiplication. */
3332 static void shader_glsl_texm3x2pad(const struct wined3d_shader_instruction *ins)
3333 {
3334     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3335     DWORD reg = ins->dst[0].reg.idx;
3336     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3337     struct glsl_src_param src0_param;
3338
3339     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3340     shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3341 }
3342
3343 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
3344  * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
3345 static void shader_glsl_texm3x3pad(const struct wined3d_shader_instruction *ins)
3346 {
3347     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3348     DWORD reg = ins->dst[0].reg.idx;
3349     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3350     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3351     struct glsl_src_param src0_param;
3352
3353     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3354     shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + tex_mx->current_row, reg, src0_param.param_str);
3355     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
3356 }
3357
3358 static void shader_glsl_texm3x2tex(const struct wined3d_shader_instruction *ins)
3359 {
3360     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3361     DWORD reg = ins->dst[0].reg.idx;
3362     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3363     struct glsl_sample_function sample_function;
3364     struct glsl_src_param src0_param;
3365
3366     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3367     shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3368
3369     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3370
3371     /* Sample the texture using the calculated coordinates */
3372     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xy");
3373 }
3374
3375 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
3376  * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
3377 static void shader_glsl_texm3x3tex(const struct wined3d_shader_instruction *ins)
3378 {
3379     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3380     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3381     struct glsl_sample_function sample_function;
3382     struct glsl_src_param src0_param;
3383     DWORD reg = ins->dst[0].reg.idx;
3384
3385     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3386     shader_addline(ins->ctx->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3387
3388     /* Dependent read, not valid with conditional NP2 */
3389     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3390
3391     /* Sample the texture using the calculated coordinates */
3392     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3393
3394     tex_mx->current_row = 0;
3395 }
3396
3397 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
3398  * Perform the 3rd row of a 3x3 matrix multiply */
3399 static void shader_glsl_texm3x3(const struct wined3d_shader_instruction *ins)
3400 {
3401     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3402     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3403     struct glsl_src_param src0_param;
3404     char dst_mask[6];
3405     DWORD reg = ins->dst[0].reg.idx;
3406
3407     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3408
3409     shader_glsl_append_dst(ins->ctx->buffer, ins);
3410     shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3411     shader_addline(ins->ctx->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
3412
3413     tex_mx->current_row = 0;
3414 }
3415
3416 /* Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
3417  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3418 static void shader_glsl_texm3x3spec(const struct wined3d_shader_instruction *ins)
3419 {
3420     struct glsl_src_param src0_param;
3421     struct glsl_src_param src1_param;
3422     DWORD reg = ins->dst[0].reg.idx;
3423     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3424     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3425     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3426     struct glsl_sample_function sample_function;
3427
3428     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3429     shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
3430
3431     /* Perform the last matrix multiply operation */
3432     shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3433     /* Reflection calculation */
3434     shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
3435
3436     /* Dependent read, not valid with conditional NP2 */
3437     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3438
3439     /* Sample the texture */
3440     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3441
3442     tex_mx->current_row = 0;
3443 }
3444
3445 /* Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
3446  * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3447 static void shader_glsl_texm3x3vspec(const struct wined3d_shader_instruction *ins)
3448 {
3449     DWORD reg = ins->dst[0].reg.idx;
3450     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3451     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3452     DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3453     struct glsl_sample_function sample_function;
3454     struct glsl_src_param src0_param;
3455
3456     shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3457
3458     /* Perform the last matrix multiply operation */
3459     shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
3460
3461     /* Construct the eye-ray vector from w coordinates */
3462     shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
3463             tex_mx->texcoord_w[0], tex_mx->texcoord_w[1], reg);
3464     shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
3465
3466     /* Dependent read, not valid with conditional NP2 */
3467     shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3468
3469     /* Sample the texture using the calculated coordinates */
3470     shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3471
3472     tex_mx->current_row = 0;
3473 }
3474
3475 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
3476  * Apply a fake bump map transform.
3477  * texbem is pshader <= 1.3 only, this saves a few version checks
3478  */
3479 static void shader_glsl_texbem(const struct wined3d_shader_instruction *ins)
3480 {
3481     const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
3482     struct glsl_sample_function sample_function;
3483     struct glsl_src_param coord_param;
3484     DWORD sampler_idx;
3485     DWORD mask;
3486     DWORD flags;
3487     char coord_mask[6];
3488
3489     sampler_idx = ins->dst[0].reg.idx;
3490     flags = (priv->cur_ps_args->tex_transform >> sampler_idx * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
3491             & WINED3D_PSARGS_TEXTRANSFORM_MASK;
3492
3493     /* Dependent read, not valid with conditional NP2 */
3494     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3495     mask = sample_function.coord_mask;
3496
3497     shader_glsl_write_mask_to_str(mask, coord_mask);
3498
3499     /* with projective textures, texbem only divides the static texture coord, not the displacement,
3500          * so we can't let the GL handle this.
3501          */
3502     if (flags & WINED3D_PSARGS_PROJECTED) {
3503         DWORD div_mask=0;
3504         char coord_div_mask[3];
3505         switch (flags & ~WINED3D_PSARGS_PROJECTED) {
3506             case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3507             case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
3508             case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
3509             case WINED3DTTFF_COUNT4:
3510             case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
3511         }
3512         shader_glsl_write_mask_to_str(div_mask, coord_div_mask);
3513         shader_addline(ins->ctx->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
3514     }
3515
3516     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &coord_param);
3517
3518     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3519             "T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s", sampler_idx, coord_mask, sampler_idx,
3520             coord_param.param_str, coord_mask);
3521
3522     if (ins->handler_idx == WINED3DSIH_TEXBEML)
3523     {
3524         struct glsl_src_param luminance_param;
3525         struct glsl_dst_param dst_param;
3526
3527         shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &luminance_param);
3528         shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3529
3530         shader_addline(ins->ctx->buffer, "%s%s *= (%s * luminancescale%d + luminanceoffset%d);\n",
3531                 dst_param.reg_name, dst_param.mask_str,
3532                 luminance_param.param_str, sampler_idx, sampler_idx);
3533     }
3534 }
3535
3536 static void shader_glsl_bem(const struct wined3d_shader_instruction *ins)
3537 {
3538     struct glsl_src_param src0_param, src1_param;
3539     DWORD sampler_idx = ins->dst[0].reg.idx;
3540
3541     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3542     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3543
3544     shader_glsl_append_dst(ins->ctx->buffer, ins);
3545     shader_addline(ins->ctx->buffer, "%s + bumpenvmat%d * %s);\n",
3546             src0_param.param_str, sampler_idx, src1_param.param_str);
3547 }
3548
3549 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
3550  * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
3551 static void shader_glsl_texreg2ar(const struct wined3d_shader_instruction *ins)
3552 {
3553     struct glsl_sample_function sample_function;
3554     struct glsl_src_param src0_param;
3555     DWORD sampler_idx = ins->dst[0].reg.idx;
3556
3557     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3558
3559     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3560     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3561             "%s.wx", src0_param.reg_name);
3562 }
3563
3564 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
3565  * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
3566 static void shader_glsl_texreg2gb(const struct wined3d_shader_instruction *ins)
3567 {
3568     struct glsl_sample_function sample_function;
3569     struct glsl_src_param src0_param;
3570     DWORD sampler_idx = ins->dst[0].reg.idx;
3571
3572     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3573
3574     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3575     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3576             "%s.yz", src0_param.reg_name);
3577 }
3578
3579 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
3580  * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
3581 static void shader_glsl_texreg2rgb(const struct wined3d_shader_instruction *ins)
3582 {
3583     struct glsl_sample_function sample_function;
3584     struct glsl_src_param src0_param;
3585     DWORD sampler_idx = ins->dst[0].reg.idx;
3586
3587     /* Dependent read, not valid with conditional NP2 */
3588     shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3589     shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &src0_param);
3590
3591     shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3592             "%s", src0_param.param_str);
3593 }
3594
3595 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
3596  * If any of the first 3 components are < 0, discard this pixel */
3597 static void shader_glsl_texkill(const struct wined3d_shader_instruction *ins)
3598 {
3599     struct glsl_dst_param dst_param;
3600
3601     /* The argument is a destination parameter, and no writemasks are allowed */
3602     shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3603     if (ins->ctx->reg_maps->shader_version.major >= 2)
3604     {
3605         /* 2.0 shaders compare all 4 components in texkill */
3606         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
3607     } else {
3608         /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
3609          * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
3610          * 4 components are defined, only the first 3 are used
3611          */
3612         shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
3613     }
3614 }
3615
3616 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
3617  * dst = dot2(src0, src1) + src2 */
3618 static void shader_glsl_dp2add(const struct wined3d_shader_instruction *ins)
3619 {
3620     struct glsl_src_param src0_param;
3621     struct glsl_src_param src1_param;
3622     struct glsl_src_param src2_param;
3623     DWORD write_mask;
3624     unsigned int mask_size;
3625
3626     write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3627     mask_size = shader_glsl_get_write_mask_size(write_mask);
3628
3629     shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3630     shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3631     shader_glsl_add_src_param(ins, &ins->src[2], WINED3DSP_WRITEMASK_0, &src2_param);
3632
3633     if (mask_size > 1) {
3634         shader_addline(ins->ctx->buffer, "vec%d(dot(%s, %s) + %s));\n",
3635                 mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
3636     } else {
3637         shader_addline(ins->ctx->buffer, "dot(%s, %s) + %s);\n",
3638                 src0_param.param_str, src1_param.param_str, src2_param.param_str);
3639     }
3640 }
3641
3642 static void shader_glsl_input_pack(const struct wined3d_shader *shader, struct wined3d_shader_buffer *buffer,
3643         const struct wined3d_shader_signature_element *input_signature,
3644         const struct wined3d_shader_reg_maps *reg_maps,
3645         enum vertexprocessing_mode vertexprocessing)
3646 {
3647     WORD map = reg_maps->input_registers;
3648     unsigned int i;
3649
3650     for (i = 0; map; map >>= 1, ++i)
3651     {
3652         const char *semantic_name;
3653         UINT semantic_idx;
3654         char reg_mask[6];
3655
3656         /* Unused */
3657         if (!(map & 1)) continue;
3658
3659         semantic_name = input_signature[i].semantic_name;
3660         semantic_idx = input_signature[i].semantic_idx;
3661         shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3662
3663         if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3664         {
3665             if (semantic_idx < 8 && vertexprocessing == pretransformed)
3666                 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
3667                         shader->u.ps.input_reg_map[i], reg_mask, semantic_idx, reg_mask);
3668             else
3669                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3670                         shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3671         }
3672         else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3673         {
3674             if (!semantic_idx)
3675                 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
3676                         shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3677             else if (semantic_idx == 1)
3678                 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
3679                         shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3680             else
3681                 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3682                         shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3683         }
3684         else
3685         {
3686             shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3687                     shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
3688         }
3689     }
3690 }
3691
3692 /*********************************************
3693  * Vertex Shader Specific Code begins here
3694  ********************************************/
3695
3696 static void add_glsl_program_entry(struct shader_glsl_priv *priv, struct glsl_shader_prog_link *entry)
3697 {
3698     struct glsl_program_key key;
3699
3700     key.vshader = entry->vshader;
3701     key.pshader = entry->pshader;
3702     key.vs_args = entry->vs_args;
3703     key.ps_args = entry->ps_args;
3704
3705     if (wine_rb_put(&priv->program_lookup, &key, &entry->program_lookup_entry) == -1)
3706     {
3707         ERR("Failed to insert program entry.\n");
3708     }
3709 }
3710
3711 static struct glsl_shader_prog_link *get_glsl_program_entry(const struct shader_glsl_priv *priv,
3712         const struct wined3d_shader *vshader, const struct wined3d_shader *pshader,
3713         const struct vs_compile_args *vs_args, const struct ps_compile_args *ps_args)
3714 {
3715     struct wine_rb_entry *entry;
3716     struct glsl_program_key key;
3717
3718     key.vshader = vshader;
3719     key.pshader = pshader;
3720     key.vs_args = *vs_args;
3721     key.ps_args = *ps_args;
3722
3723     entry = wine_rb_get(&priv->program_lookup, &key);
3724     return entry ? WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry) : NULL;
3725 }
3726
3727 /* GL locking is done by the caller */
3728 static void delete_glsl_program_entry(struct shader_glsl_priv *priv, const struct wined3d_gl_info *gl_info,
3729         struct glsl_shader_prog_link *entry)
3730 {
3731     struct glsl_program_key key;
3732
3733     key.vshader = entry->vshader;
3734     key.pshader = entry->pshader;
3735     key.vs_args = entry->vs_args;
3736     key.ps_args = entry->ps_args;
3737     wine_rb_remove(&priv->program_lookup, &key);
3738
3739     GL_EXTCALL(glDeleteObjectARB(entry->programId));
3740     if (entry->vshader) list_remove(&entry->vshader_entry);
3741     if (entry->pshader) list_remove(&entry->pshader_entry);
3742     HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
3743     HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
3744     HeapFree(GetProcessHeap(), 0, entry);
3745 }
3746
3747 static void handle_ps3_input(struct wined3d_shader_buffer *buffer,
3748         const struct wined3d_gl_info *gl_info, const DWORD *map,
3749         const struct wined3d_shader_signature_element *input_signature,
3750         const struct wined3d_shader_reg_maps *reg_maps_in,
3751         const struct wined3d_shader_signature_element *output_signature,
3752         const struct wined3d_shader_reg_maps *reg_maps_out)
3753 {
3754     unsigned int i, j;
3755     const char *semantic_name_in;
3756     UINT semantic_idx_in;
3757     DWORD *set;
3758     DWORD in_idx;
3759     unsigned int in_count = vec4_varyings(3, gl_info);
3760     char reg_mask[6];
3761     char destination[50];
3762     WORD input_map, output_map;
3763
3764     set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
3765
3766     input_map = reg_maps_in->input_registers;
3767     for (i = 0; input_map; input_map >>= 1, ++i)
3768     {
3769         if (!(input_map & 1)) continue;
3770
3771         in_idx = map[i];
3772         /* Declared, but not read register */
3773         if (in_idx == ~0U) continue;
3774         if (in_idx >= (in_count + 2))
3775         {
3776             FIXME("More input varyings declared than supported, expect issues.\n");
3777             continue;
3778         }
3779
3780         if (in_idx == in_count) {
3781             sprintf(destination, "gl_FrontColor");
3782         } else if (in_idx == in_count + 1) {
3783             sprintf(destination, "gl_FrontSecondaryColor");
3784         } else {
3785             sprintf(destination, "IN[%u]", in_idx);
3786         }
3787
3788         semantic_name_in = input_signature[i].semantic_name;
3789         semantic_idx_in = input_signature[i].semantic_idx;
3790         set[in_idx] = ~0U;
3791
3792         output_map = reg_maps_out->output_registers;
3793         for (j = 0; output_map; output_map >>= 1, ++j)
3794         {
3795             DWORD mask;
3796
3797             if (!(output_map & 1)
3798                     || semantic_idx_in != output_signature[j].semantic_idx
3799                     || strcmp(semantic_name_in, output_signature[j].semantic_name)
3800                     || !(mask = input_signature[i].mask & output_signature[j].mask))
3801                 continue;
3802
3803             set[in_idx] = mask;
3804             shader_glsl_write_mask_to_str(mask, reg_mask);
3805
3806             shader_addline(buffer, "%s%s = OUT[%u]%s;\n",
3807                     destination, reg_mask, j, reg_mask);
3808         }
3809     }
3810
3811     for (i = 0; i < in_count + 2; ++i)
3812     {
3813         unsigned int size;
3814
3815         if (!set[i] || set[i] == WINED3DSP_WRITEMASK_ALL)
3816             continue;
3817
3818         if (set[i] == ~0U) set[i] = 0;
3819
3820         size = 0;
3821         if (!(set[i] & WINED3DSP_WRITEMASK_0)) reg_mask[size++] = 'x';
3822         if (!(set[i] & WINED3DSP_WRITEMASK_1)) reg_mask[size++] = 'y';
3823         if (!(set[i] & WINED3DSP_WRITEMASK_2)) reg_mask[size++] = 'z';
3824         if (!(set[i] & WINED3DSP_WRITEMASK_3)) reg_mask[size++] = 'w';
3825         reg_mask[size] = '\0';
3826
3827         if (i == in_count) sprintf(destination, "gl_FrontColor");
3828         else if (i == in_count + 1) sprintf(destination, "gl_FrontSecondaryColor");
3829         else sprintf(destination, "IN[%u]", i);
3830
3831         if (size == 1) shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
3832         else shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
3833     }
3834
3835     HeapFree(GetProcessHeap(), 0, set);
3836 }
3837
3838 /* GL locking is done by the caller */
3839 static GLhandleARB generate_param_reorder_function(struct wined3d_shader_buffer *buffer,
3840         const struct wined3d_shader *vs, const struct wined3d_shader *ps,
3841         const struct wined3d_gl_info *gl_info)
3842 {
3843     GLhandleARB ret = 0;
3844     DWORD ps_major = ps ? ps->reg_maps.shader_version.major : 0;
3845     unsigned int i;
3846     const char *semantic_name;
3847     UINT semantic_idx;
3848     char reg_mask[6];
3849     const struct wined3d_shader_signature_element *output_signature = vs->output_signature;
3850     WORD map = vs->reg_maps.output_registers;
3851
3852     shader_buffer_clear(buffer);
3853
3854     shader_addline(buffer, "#version 120\n");
3855
3856     if (ps_major < 3)
3857     {
3858         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3859
3860         for (i = 0; map; map >>= 1, ++i)
3861         {
3862             DWORD write_mask;
3863
3864             if (!(map & 1)) continue;
3865
3866             semantic_name = output_signature[i].semantic_name;
3867             semantic_idx = output_signature[i].semantic_idx;
3868             write_mask = output_signature[i].mask;
3869             shader_glsl_write_mask_to_str(write_mask, reg_mask);
3870
3871             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3872             {
3873                 if (!semantic_idx)
3874                     shader_addline(buffer, "gl_FrontColor%s = OUT[%u]%s;\n",
3875                             reg_mask, i, reg_mask);
3876                 else if (semantic_idx == 1)
3877                     shader_addline(buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n",
3878                             reg_mask, i, reg_mask);
3879             }
3880             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3881             {
3882                 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n",
3883                         reg_mask, i, reg_mask);
3884             }
3885             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3886             {
3887                 if (semantic_idx < 8)
3888                 {
3889                     if (!(gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) || ps_major > 0)
3890                         write_mask |= WINED3DSP_WRITEMASK_3;
3891
3892                     shader_addline(buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
3893                             semantic_idx, reg_mask, i, reg_mask);
3894                     if (!(write_mask & WINED3DSP_WRITEMASK_3))
3895                         shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", semantic_idx);
3896                 }
3897             }
3898             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3899             {
3900                 shader_addline(buffer, "gl_PointSize = OUT[%u].%c;\n", i, reg_mask[1]);
3901             }
3902             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3903             {
3904                 shader_addline(buffer, "gl_FogFragCoord = OUT[%u].%c;\n", i, reg_mask[1]);
3905             }
3906         }
3907         shader_addline(buffer, "}\n");
3908
3909     }
3910     else
3911     {
3912         /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3913         shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(3, gl_info));
3914         shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3915
3916         /* First, sort out position and point size. Those are not passed to the pixel shader */
3917         for (i = 0; map; map >>= 1, ++i)
3918         {
3919             if (!(map & 1)) continue;
3920
3921             semantic_name = output_signature[i].semantic_name;
3922             shader_glsl_write_mask_to_str(output_signature[i].mask, reg_mask);
3923
3924             if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3925             {
3926                 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n",
3927                         reg_mask, i, reg_mask);
3928             }
3929             else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3930             {
3931                 shader_addline(buffer, "gl_PointSize = OUT[%u].%c;\n", i, reg_mask[1]);
3932             }
3933         }
3934
3935         /* Then, fix the pixel shader input */
3936         handle_ps3_input(buffer, gl_info, ps->u.ps.input_reg_map, ps->input_signature,
3937                 &ps->reg_maps, output_signature, &vs->reg_maps);
3938
3939         shader_addline(buffer, "}\n");
3940     }
3941
3942     ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3943     checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3944     shader_glsl_compile(gl_info, ret, buffer->buffer);
3945
3946     return ret;
3947 }
3948
3949 /* GL locking is done by the caller */
3950 static void hardcode_local_constants(const struct wined3d_shader *shader,
3951         const struct wined3d_gl_info *gl_info, GLhandleARB programId, char prefix)
3952 {
3953     const struct wined3d_shader_lconst *lconst;
3954     GLint tmp_loc;
3955     const float *value;
3956     char glsl_name[8];
3957
3958     LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
3959     {
3960         value = (const float *)lconst->value;
3961         snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
3962         tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3963         GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
3964     }
3965     checkGLcall("Hardcoding local constants");
3966 }
3967
3968 /* GL locking is done by the caller */
3969 static GLuint shader_glsl_generate_pshader(const struct wined3d_context *context,
3970         struct wined3d_shader_buffer *buffer, const struct wined3d_shader *shader,
3971         const struct ps_compile_args *args, struct ps_np2fixup_info *np2fixup_info)
3972 {
3973     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3974     const struct wined3d_gl_info *gl_info = context->gl_info;
3975     const DWORD *function = shader->function;
3976     struct shader_glsl_ctx_priv priv_ctx;
3977
3978     /* Create the hw GLSL shader object and assign it as the shader->prgId */
3979     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3980
3981     memset(&priv_ctx, 0, sizeof(priv_ctx));
3982     priv_ctx.cur_ps_args = args;
3983     priv_ctx.cur_np2fixup_info = np2fixup_info;
3984
3985     shader_addline(buffer, "#version 120\n");
3986
3987     if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
3988     {
3989         shader_addline(buffer, "#extension GL_ARB_shader_texture_lod : enable\n");
3990     }
3991     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
3992     {
3993         /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
3994          * drivers write a warning if we don't do so
3995          */
3996         shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
3997     }
3998     if (gl_info->supported[EXT_GPU_SHADER4])
3999     {
4000         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
4001     }
4002
4003     /* Base Declarations */
4004     shader_generate_glsl_declarations(context, buffer, shader, reg_maps, &priv_ctx);
4005
4006     /* Pack 3.0 inputs */
4007     if (reg_maps->shader_version.major >= 3 && args->vp_mode != vertexshader)
4008         shader_glsl_input_pack(shader, buffer, shader->input_signature, reg_maps, args->vp_mode);
4009
4010     /* Base Shader Body */
4011     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4012
4013     /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
4014     if (reg_maps->shader_version.major < 2)
4015     {
4016         /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
4017         shader_addline(buffer, "gl_FragData[0] = R0;\n");
4018     }
4019
4020     if (args->srgb_correction)
4021     {
4022         shader_addline(buffer, "tmp0.xyz = pow(gl_FragData[0].xyz, vec3(srgb_const0.x));\n");
4023         shader_addline(buffer, "tmp0.xyz = tmp0.xyz * vec3(srgb_const0.y) - vec3(srgb_const0.z);\n");
4024         shader_addline(buffer, "tmp1.xyz = gl_FragData[0].xyz * vec3(srgb_const0.w);\n");
4025         shader_addline(buffer, "bvec3 srgb_compare = lessThan(gl_FragData[0].xyz, vec3(srgb_const1.x));\n");
4026         shader_addline(buffer, "gl_FragData[0].xyz = mix(tmp0.xyz, tmp1.xyz, vec3(srgb_compare));\n");
4027         shader_addline(buffer, "gl_FragData[0] = clamp(gl_FragData[0], 0.0, 1.0);\n");
4028     }
4029     /* Pixel shader < 3.0 do not replace the fog stage.
4030      * This implements linear fog computation and blending.
4031      * TODO: non linear fog
4032      * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
4033      * -1/(e-s) and e/(e-s) respectively.
4034      */
4035     if (reg_maps->shader_version.major < 3)
4036     {
4037         switch(args->fog) {
4038             case FOG_OFF: break;
4039             case FOG_LINEAR:
4040                 shader_addline(buffer, "float fogstart = -1.0 / (gl_Fog.end - gl_Fog.start);\n");
4041                 shader_addline(buffer, "float fogend = gl_Fog.end * -fogstart;\n");
4042                 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * fogstart + fogend, 0.0, 1.0);\n");
4043                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4044                 break;
4045             case FOG_EXP:
4046                 /* Fog = e^(-gl_Fog.density * gl_FogFragCoord) */
4047                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_FogFragCoord);\n");
4048                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4049                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4050                 break;
4051             case FOG_EXP2:
4052                 /* Fog = e^(-(gl_Fog.density * gl_FogFragCoord)^2) */
4053                 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_Fog.density * gl_FogFragCoord * gl_FogFragCoord);\n");
4054                 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4055                 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4056                 break;
4057         }
4058     }
4059
4060     shader_addline(buffer, "}\n");
4061
4062     TRACE("Compiling shader object %u\n", shader_obj);
4063     shader_glsl_compile(gl_info, shader_obj, buffer->buffer);
4064
4065     /* Store the shader object */
4066     return shader_obj;
4067 }
4068
4069 /* GL locking is done by the caller */
4070 static GLuint shader_glsl_generate_vshader(const struct wined3d_context *context,
4071         struct wined3d_shader_buffer *buffer, const struct wined3d_shader *shader,
4072         const struct vs_compile_args *args)
4073 {
4074     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4075     const struct wined3d_gl_info *gl_info = context->gl_info;
4076     const DWORD *function = shader->function;
4077     struct shader_glsl_ctx_priv priv_ctx;
4078
4079     /* Create the hw GLSL shader program and assign it as the shader->prgId */
4080     GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4081
4082     shader_addline(buffer, "#version 120\n");
4083
4084     if (gl_info->supported[EXT_GPU_SHADER4])
4085         shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
4086
4087     memset(&priv_ctx, 0, sizeof(priv_ctx));
4088     priv_ctx.cur_vs_args = args;
4089
4090     /* Base Declarations */
4091     shader_generate_glsl_declarations(context, buffer, shader, reg_maps, &priv_ctx);
4092
4093     /* Base Shader Body */
4094     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4095
4096     /* Unpack outputs */
4097     shader_addline(buffer, "order_ps_input(OUT);\n");
4098
4099     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
4100      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
4101      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
4102      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
4103      */
4104     if (args->fog_src == VS_FOG_Z)
4105         shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
4106     else if (!reg_maps->fog)
4107         shader_addline(buffer, "gl_FogFragCoord = 0.0;\n");
4108
4109     /* We always store the clipplanes without y inversion */
4110     if (args->clip_enabled)
4111         shader_addline(buffer, "gl_ClipVertex = gl_Position;\n");
4112
4113     /* Write the final position.
4114      *
4115      * OpenGL coordinates specify the center of the pixel while d3d coords specify
4116      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
4117      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
4118      * contains 1.0 to allow a mad.
4119      */
4120     shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
4121     shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
4122
4123     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
4124      *
4125      * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
4126      * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
4127      * which is the same as z = z * 2 - w.
4128      */
4129     shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
4130
4131     shader_addline(buffer, "}\n");
4132
4133     TRACE("Compiling shader object %u\n", shader_obj);
4134     shader_glsl_compile(gl_info, shader_obj, buffer->buffer);
4135
4136     return shader_obj;
4137 }
4138
4139 static GLhandleARB find_glsl_pshader(const struct wined3d_context *context,
4140         struct wined3d_shader_buffer *buffer, struct wined3d_shader *shader,
4141         const struct ps_compile_args *args, const struct ps_np2fixup_info **np2fixup_info)
4142 {
4143     struct wined3d_state *state = &shader->device->stateBlock->state;
4144     UINT i;
4145     DWORD new_size;
4146     struct glsl_ps_compiled_shader *new_array;
4147     struct glsl_pshader_private    *shader_data;
4148     struct ps_np2fixup_info        *np2fixup = NULL;
4149     GLhandleARB ret;
4150
4151     if (!shader->backend_data)
4152     {
4153         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4154         if (!shader->backend_data)
4155         {
4156             ERR("Failed to allocate backend data.\n");
4157             return 0;
4158         }
4159     }
4160     shader_data = shader->backend_data;
4161
4162     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4163      * so a linear search is more performant than a hashmap or a binary search
4164      * (cache coherency etc)
4165      */
4166     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4167     {
4168         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4169         {
4170             if (args->np2_fixup) *np2fixup_info = &shader_data->gl_shaders[i].np2fixup;
4171             return shader_data->gl_shaders[i].prgId;
4172         }
4173     }
4174
4175     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4176     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4177         if (shader_data->num_gl_shaders)
4178         {
4179             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4180             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4181                                     new_size * sizeof(*shader_data->gl_shaders));
4182         } else {
4183             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4184             new_size = 1;
4185         }
4186
4187         if(!new_array) {
4188             ERR("Out of memory\n");
4189             return 0;
4190         }
4191         shader_data->gl_shaders = new_array;
4192         shader_data->shader_array_size = new_size;
4193     }
4194
4195     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4196
4197     memset(&shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup, 0, sizeof(struct ps_np2fixup_info));
4198     if (args->np2_fixup) np2fixup = &shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup;
4199
4200     pixelshader_update_samplers(&shader->reg_maps, state->textures);
4201
4202     shader_buffer_clear(buffer);
4203     ret = shader_glsl_generate_pshader(context, buffer, shader, args, np2fixup);
4204     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4205     *np2fixup_info = np2fixup;
4206
4207     return ret;
4208 }
4209
4210 static inline BOOL vs_args_equal(const struct vs_compile_args *stored, const struct vs_compile_args *new,
4211                                  const DWORD use_map) {
4212     if((stored->swizzle_map & use_map) != new->swizzle_map) return FALSE;
4213     if((stored->clip_enabled) != new->clip_enabled) return FALSE;
4214     return stored->fog_src == new->fog_src;
4215 }
4216
4217 static GLhandleARB find_glsl_vshader(const struct wined3d_context *context,
4218         struct wined3d_shader_buffer *buffer, struct wined3d_shader *shader,
4219         const struct vs_compile_args *args)
4220 {
4221     UINT i;
4222     DWORD new_size;
4223     struct glsl_vs_compiled_shader *new_array;
4224     DWORD use_map = shader->device->strided_streams.use_map;
4225     struct glsl_vshader_private *shader_data;
4226     GLhandleARB ret;
4227
4228     if (!shader->backend_data)
4229     {
4230         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4231         if (!shader->backend_data)
4232         {
4233             ERR("Failed to allocate backend data.\n");
4234             return 0;
4235         }
4236     }
4237     shader_data = shader->backend_data;
4238
4239     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4240      * so a linear search is more performant than a hashmap or a binary search
4241      * (cache coherency etc)
4242      */
4243     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4244         if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map)) {
4245             return shader_data->gl_shaders[i].prgId;
4246         }
4247     }
4248
4249     TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4250
4251     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4252         if (shader_data->num_gl_shaders)
4253         {
4254             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4255             new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4256                                     new_size * sizeof(*shader_data->gl_shaders));
4257         } else {
4258             new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4259             new_size = 1;
4260         }
4261
4262         if(!new_array) {
4263             ERR("Out of memory\n");
4264             return 0;
4265         }
4266         shader_data->gl_shaders = new_array;
4267         shader_data->shader_array_size = new_size;
4268     }
4269
4270     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4271
4272     shader_buffer_clear(buffer);
4273     ret = shader_glsl_generate_vshader(context, buffer, shader, args);
4274     shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4275
4276     return ret;
4277 }
4278
4279 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
4280  * It sets the programId on the current StateBlock (because it should be called
4281  * inside of the DrawPrimitive() part of the render loop).
4282  *
4283  * If a program for the given combination does not exist, create one, and store
4284  * the program in the hash table.  If it creates a program, it will link the
4285  * given objects, too.
4286  */
4287
4288 /* GL locking is done by the caller */
4289 static void set_glsl_shader_program(const struct wined3d_context *context,
4290         struct wined3d_device *device, BOOL use_ps, BOOL use_vs)
4291 {
4292     const struct wined3d_state *state = &device->stateBlock->state;
4293     struct wined3d_shader *vshader = use_vs ? state->vertex_shader : NULL;
4294     struct wined3d_shader *pshader = use_ps ? state->pixel_shader : NULL;
4295     const struct wined3d_gl_info *gl_info = context->gl_info;
4296     struct shader_glsl_priv *priv = device->shader_priv;
4297     struct glsl_shader_prog_link *entry    = NULL;
4298     GLhandleARB programId                  = 0;
4299     GLhandleARB reorder_shader_id          = 0;
4300     unsigned int i;
4301     char glsl_name[8];
4302     struct ps_compile_args ps_compile_args;
4303     struct vs_compile_args vs_compile_args;
4304
4305     if (vshader) find_vs_compile_args(state, vshader, &vs_compile_args);
4306     if (pshader) find_ps_compile_args(state, pshader, &ps_compile_args);
4307
4308     entry = get_glsl_program_entry(priv, vshader, pshader, &vs_compile_args, &ps_compile_args);
4309     if (entry)
4310     {
4311         priv->glsl_program = entry;
4312         return;
4313     }
4314
4315     /* If we get to this point, then no matching program exists, so we create one */
4316     programId = GL_EXTCALL(glCreateProgramObjectARB());
4317     TRACE("Created new GLSL shader program %u\n", programId);
4318
4319     /* Create the entry */
4320     entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
4321     entry->programId = programId;
4322     entry->vshader = vshader;
4323     entry->pshader = pshader;
4324     entry->vs_args = vs_compile_args;
4325     entry->ps_args = ps_compile_args;
4326     entry->constant_version = 0;
4327     entry->np2Fixup_info = NULL;
4328     /* Add the hash table entry */
4329     add_glsl_program_entry(priv, entry);
4330
4331     /* Set the current program */
4332     priv->glsl_program = entry;
4333
4334     /* Attach GLSL vshader */
4335     if (vshader)
4336     {
4337         GLhandleARB vshader_id = find_glsl_vshader(context, &priv->shader_buffer, vshader, &vs_compile_args);
4338         WORD map = vshader->reg_maps.input_registers;
4339         char tmp_name[10];
4340
4341         reorder_shader_id = generate_param_reorder_function(&priv->shader_buffer, vshader, pshader, gl_info);
4342         TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
4343         GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
4344         checkGLcall("glAttachObjectARB");
4345         /* Flag the reorder function for deletion, then it will be freed automatically when the program
4346          * is destroyed
4347          */
4348         GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
4349
4350         TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
4351         GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
4352         checkGLcall("glAttachObjectARB");
4353
4354         /* Bind vertex attributes to a corresponding index number to match
4355          * the same index numbers as ARB_vertex_programs (makes loading
4356          * vertex attributes simpler).  With this method, we can use the
4357          * exact same code to load the attributes later for both ARB and
4358          * GLSL shaders.
4359          *
4360          * We have to do this here because we need to know the Program ID
4361          * in order to make the bindings work, and it has to be done prior
4362          * to linking the GLSL program. */
4363         for (i = 0; map; map >>= 1, ++i)
4364         {
4365             if (!(map & 1)) continue;
4366
4367             snprintf(tmp_name, sizeof(tmp_name), "attrib%u", i);
4368             GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
4369         }
4370         checkGLcall("glBindAttribLocationARB");
4371
4372         list_add_head(&vshader->linked_programs, &entry->vshader_entry);
4373     }
4374
4375     /* Attach GLSL pshader */
4376     if (pshader)
4377     {
4378         GLhandleARB pshader_id = find_glsl_pshader(context, &priv->shader_buffer,
4379                 pshader, &ps_compile_args, &entry->np2Fixup_info);
4380         TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
4381         GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
4382         checkGLcall("glAttachObjectARB");
4383
4384         list_add_head(&pshader->linked_programs, &entry->pshader_entry);
4385     }
4386
4387     /* Link the program */
4388     TRACE("Linking GLSL shader program %u\n", programId);
4389     GL_EXTCALL(glLinkProgramARB(programId));
4390     shader_glsl_validate_link(gl_info, programId);
4391
4392     entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4393             sizeof(GLhandleARB) * gl_info->limits.glsl_vs_float_constants);
4394     for (i = 0; i < gl_info->limits.glsl_vs_float_constants; ++i)
4395     {
4396         snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
4397         entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4398     }
4399     for (i = 0; i < MAX_CONST_I; ++i)
4400     {
4401         snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
4402         entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4403     }
4404     entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4405             sizeof(GLhandleARB) * gl_info->limits.glsl_ps_float_constants);
4406     for (i = 0; i < gl_info->limits.glsl_ps_float_constants; ++i)
4407     {
4408         snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
4409         entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4410     }
4411     for (i = 0; i < MAX_CONST_I; ++i)
4412     {
4413         snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
4414         entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4415     }
4416
4417     if(pshader) {
4418         char name[32];
4419
4420         for(i = 0; i < MAX_TEXTURES; i++) {
4421             sprintf(name, "bumpenvmat%u", i);
4422             entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4423             sprintf(name, "luminancescale%u", i);
4424             entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4425             sprintf(name, "luminanceoffset%u", i);
4426             entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4427         }
4428
4429         if (ps_compile_args.np2_fixup) {
4430             if (entry->np2Fixup_info) {
4431                 entry->np2Fixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "PsamplerNP2Fixup"));
4432             } else {
4433                 FIXME("NP2 texcoord fixup needed for this pixelshader, but no fixup uniform found.\n");
4434             }
4435         }
4436     }
4437
4438     entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
4439     entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
4440     checkGLcall("Find glsl program uniform locations");
4441
4442     if (pshader && pshader->reg_maps.shader_version.major >= 3
4443             && pshader->u.ps.declared_in_count > vec4_varyings(3, gl_info))
4444     {
4445         TRACE("Shader %d needs vertex color clamping disabled\n", programId);
4446         entry->vertex_color_clamp = GL_FALSE;
4447     } else {
4448         entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
4449     }
4450
4451     /* Set the shader to allow uniform loading on it */
4452     GL_EXTCALL(glUseProgramObjectARB(programId));
4453     checkGLcall("glUseProgramObjectARB(programId)");
4454
4455     /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
4456      * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
4457      * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
4458      * vertex shader with fixed function pixel processing is used we make sure that the card
4459      * supports enough samplers to allow the max number of vertex samplers with all possible
4460      * fixed function fragment processing setups. So once the program is linked these samplers
4461      * won't change.
4462      */
4463     if (vshader) shader_glsl_load_vsamplers(gl_info, device->texUnitMap, programId);
4464     if (pshader) shader_glsl_load_psamplers(gl_info, device->texUnitMap, programId);
4465
4466     /* If the local constants do not have to be loaded with the environment constants,
4467      * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
4468      * later
4469      */
4470     if (pshader && !pshader->load_local_constsF)
4471         hardcode_local_constants(pshader, gl_info, programId, 'P');
4472     if (vshader && !vshader->load_local_constsF)
4473         hardcode_local_constants(vshader, gl_info, programId, 'V');
4474 }
4475
4476 /* GL locking is done by the caller */
4477 static GLhandleARB create_glsl_blt_shader(const struct wined3d_gl_info *gl_info, enum tex_types tex_type, BOOL masked)
4478 {
4479     GLhandleARB program_id;
4480     GLhandleARB vshader_id, pshader_id;
4481     const char *blt_pshader;
4482
4483     static const char *blt_vshader =
4484         "#version 120\n"
4485         "void main(void)\n"
4486         "{\n"
4487         "    gl_Position = gl_Vertex;\n"
4488         "    gl_FrontColor = vec4(1.0);\n"
4489         "    gl_TexCoord[0] = gl_MultiTexCoord0;\n"
4490         "}\n";
4491
4492     static const char * const blt_pshaders_full[tex_type_count] =
4493     {
4494         /* tex_1d */
4495         NULL,
4496         /* tex_2d */
4497         "#version 120\n"
4498         "uniform sampler2D sampler;\n"
4499         "void main(void)\n"
4500         "{\n"
4501         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4502         "}\n",
4503         /* tex_3d */
4504         NULL,
4505         /* tex_cube */
4506         "#version 120\n"
4507         "uniform samplerCube sampler;\n"
4508         "void main(void)\n"
4509         "{\n"
4510         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4511         "}\n",
4512         /* tex_rect */
4513         "#version 120\n"
4514         "#extension GL_ARB_texture_rectangle : enable\n"
4515         "uniform sampler2DRect sampler;\n"
4516         "void main(void)\n"
4517         "{\n"
4518         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4519         "}\n",
4520     };
4521
4522     static const char * const blt_pshaders_masked[tex_type_count] =
4523     {
4524         /* tex_1d */
4525         NULL,
4526         /* tex_2d */
4527         "#version 120\n"
4528         "uniform sampler2D sampler;\n"
4529         "uniform vec4 mask;\n"
4530         "void main(void)\n"
4531         "{\n"
4532         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4533         "    gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4534         "}\n",
4535         /* tex_3d */
4536         NULL,
4537         /* tex_cube */
4538         "#version 120\n"
4539         "uniform samplerCube sampler;\n"
4540         "uniform vec4 mask;\n"
4541         "void main(void)\n"
4542         "{\n"
4543         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4544         "    gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4545         "}\n",
4546         /* tex_rect */
4547         "#version 120\n"
4548         "#extension GL_ARB_texture_rectangle : enable\n"
4549         "uniform sampler2DRect sampler;\n"
4550         "uniform vec4 mask;\n"
4551         "void main(void)\n"
4552         "{\n"
4553         "    if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4554         "    gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4555         "}\n",
4556     };
4557
4558     blt_pshader = masked ? blt_pshaders_masked[tex_type] : blt_pshaders_full[tex_type];
4559     if (!blt_pshader)
4560     {
4561         FIXME("tex_type %#x not supported\n", tex_type);
4562         return 0;
4563     }
4564
4565     vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4566     shader_glsl_compile(gl_info, vshader_id, blt_vshader);
4567
4568     pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
4569     shader_glsl_compile(gl_info, pshader_id, blt_pshader);
4570
4571     program_id = GL_EXTCALL(glCreateProgramObjectARB());
4572     GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
4573     GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
4574     GL_EXTCALL(glLinkProgramARB(program_id));
4575
4576     shader_glsl_validate_link(gl_info, program_id);
4577
4578     /* Once linked we can mark the shaders for deletion. They will be deleted once the program
4579      * is destroyed
4580      */
4581     GL_EXTCALL(glDeleteObjectARB(vshader_id));
4582     GL_EXTCALL(glDeleteObjectARB(pshader_id));
4583     return program_id;
4584 }
4585
4586 /* GL locking is done by the caller */
4587 static void shader_glsl_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4588 {
4589     const struct wined3d_gl_info *gl_info = context->gl_info;
4590     struct wined3d_device *device = context->swapchain->device;
4591     struct shader_glsl_priv *priv = device->shader_priv;
4592     GLhandleARB program_id = 0;
4593     GLenum old_vertex_color_clamp, current_vertex_color_clamp;
4594
4595     old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4596
4597     if (useVS || usePS) set_glsl_shader_program(context, device, usePS, useVS);
4598     else priv->glsl_program = NULL;
4599
4600     current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4601
4602     if (old_vertex_color_clamp != current_vertex_color_clamp)
4603     {
4604         if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4605         {
4606             GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
4607             checkGLcall("glClampColorARB");
4608         }
4609         else
4610         {
4611             FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4612         }
4613     }
4614
4615     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4616     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4617     GL_EXTCALL(glUseProgramObjectARB(program_id));
4618     checkGLcall("glUseProgramObjectARB");
4619
4620     /* In case that NP2 texcoord fixup data is found for the selected program, trigger a reload of the
4621      * constants. This has to be done because it can't be guaranteed that sampler() (from state.c) is
4622      * called between selecting the shader and using it, which results in wrong fixup for some frames. */
4623     if (priv->glsl_program && priv->glsl_program->np2Fixup_info)
4624     {
4625         shader_glsl_load_np2fixup_constants(priv, gl_info, &device->stateBlock->state);
4626     }
4627 }
4628
4629 /* GL locking is done by the caller */
4630 static void shader_glsl_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4631         enum tex_types tex_type, const SIZE *ds_mask_size)
4632 {
4633     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4634     struct shader_glsl_priv *priv = shader_priv;
4635     GLhandleARB *blt_program;
4636     GLint loc;
4637
4638     blt_program = masked ? &priv->depth_blt_program_masked[tex_type] : &priv->depth_blt_program_full[tex_type];
4639     if (!*blt_program)
4640     {
4641         *blt_program = create_glsl_blt_shader(gl_info, tex_type, masked);
4642         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "sampler"));
4643         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4644         GL_EXTCALL(glUniform1iARB(loc, 0));
4645     }
4646     else
4647     {
4648         GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4649     }
4650
4651     if (masked)
4652     {
4653         loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "mask"));
4654         GL_EXTCALL(glUniform4fARB(loc, 0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy));
4655     }
4656 }
4657
4658 /* GL locking is done by the caller */
4659 static void shader_glsl_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4660 {
4661     struct shader_glsl_priv *priv = shader_priv;
4662     GLhandleARB program_id;
4663
4664     program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4665     if (program_id) TRACE("Using GLSL program %u\n", program_id);
4666
4667     GL_EXTCALL(glUseProgramObjectARB(program_id));
4668     checkGLcall("glUseProgramObjectARB");
4669 }
4670
4671 static void shader_glsl_destroy(struct wined3d_shader *shader)
4672 {
4673     struct wined3d_device *device = shader->device;
4674     struct shader_glsl_priv *priv = device->shader_priv;
4675     const struct wined3d_gl_info *gl_info;
4676     const struct list *linked_programs;
4677     struct wined3d_context *context;
4678
4679     char pshader = shader_is_pshader_version(shader->reg_maps.shader_version.type);
4680
4681     if (pshader)
4682     {
4683         struct glsl_pshader_private *shader_data = shader->backend_data;
4684
4685         if (!shader_data || !shader_data->num_gl_shaders)
4686         {
4687             HeapFree(GetProcessHeap(), 0, shader_data);
4688             shader->backend_data = NULL;
4689             return;
4690         }
4691
4692         context = context_acquire(device, NULL);
4693         gl_info = context->gl_info;
4694
4695         if (priv->glsl_program && priv->glsl_program->pshader == shader)
4696         {
4697             ENTER_GL();
4698             shader_glsl_select(context, FALSE, FALSE);
4699             LEAVE_GL();
4700         }
4701     }
4702     else
4703     {
4704         struct glsl_vshader_private *shader_data = shader->backend_data;
4705
4706         if (!shader_data || !shader_data->num_gl_shaders)
4707         {
4708             HeapFree(GetProcessHeap(), 0, shader_data);
4709             shader->backend_data = NULL;
4710             return;
4711         }
4712
4713         context = context_acquire(device, NULL);
4714         gl_info = context->gl_info;
4715
4716         if (priv->glsl_program && priv->glsl_program->vshader == shader)
4717         {
4718             ENTER_GL();
4719             shader_glsl_select(context, FALSE, FALSE);
4720             LEAVE_GL();
4721         }
4722     }
4723
4724     linked_programs = &shader->linked_programs;
4725
4726     TRACE("Deleting linked programs\n");
4727     if (linked_programs->next) {
4728         struct glsl_shader_prog_link *entry, *entry2;
4729
4730         ENTER_GL();
4731         if(pshader) {
4732             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
4733                 delete_glsl_program_entry(priv, gl_info, entry);
4734             }
4735         } else {
4736             LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
4737                 delete_glsl_program_entry(priv, gl_info, entry);
4738             }
4739         }
4740         LEAVE_GL();
4741     }
4742
4743     if (pshader)
4744     {
4745         struct glsl_pshader_private *shader_data = shader->backend_data;
4746         UINT i;
4747
4748         ENTER_GL();
4749         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4750             TRACE("deleting pshader %u\n", shader_data->gl_shaders[i].prgId);
4751             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4752             checkGLcall("glDeleteObjectARB");
4753         }
4754         LEAVE_GL();
4755         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4756     }
4757     else
4758     {
4759         struct glsl_vshader_private *shader_data = shader->backend_data;
4760         UINT i;
4761
4762         ENTER_GL();
4763         for(i = 0; i < shader_data->num_gl_shaders; i++) {
4764             TRACE("deleting vshader %u\n", shader_data->gl_shaders[i].prgId);
4765             GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4766             checkGLcall("glDeleteObjectARB");
4767         }
4768         LEAVE_GL();
4769         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4770     }
4771
4772     HeapFree(GetProcessHeap(), 0, shader->backend_data);
4773     shader->backend_data = NULL;
4774
4775     context_release(context);
4776 }
4777
4778 static int glsl_program_key_compare(const void *key, const struct wine_rb_entry *entry)
4779 {
4780     const struct glsl_program_key *k = key;
4781     const struct glsl_shader_prog_link *prog = WINE_RB_ENTRY_VALUE(entry,
4782             const struct glsl_shader_prog_link, program_lookup_entry);
4783     int cmp;
4784
4785     if (k->vshader > prog->vshader) return 1;
4786     else if (k->vshader < prog->vshader) return -1;
4787
4788     if (k->pshader > prog->pshader) return 1;
4789     else if (k->pshader < prog->pshader) return -1;
4790
4791     if (k->vshader && (cmp = memcmp(&k->vs_args, &prog->vs_args, sizeof(prog->vs_args)))) return cmp;
4792     if (k->pshader && (cmp = memcmp(&k->ps_args, &prog->ps_args, sizeof(prog->ps_args)))) return cmp;
4793
4794     return 0;
4795 }
4796
4797 static BOOL constant_heap_init(struct constant_heap *heap, unsigned int constant_count)
4798 {
4799     SIZE_T size = (constant_count + 1) * sizeof(*heap->entries) + constant_count * sizeof(*heap->positions);
4800     void *mem = HeapAlloc(GetProcessHeap(), 0, size);
4801
4802     if (!mem)
4803     {
4804         ERR("Failed to allocate memory\n");
4805         return FALSE;
4806     }
4807
4808     heap->entries = mem;
4809     heap->entries[1].version = 0;
4810     heap->positions = (unsigned int *)(heap->entries + constant_count + 1);
4811     heap->size = 1;
4812
4813     return TRUE;
4814 }
4815
4816 static void constant_heap_free(struct constant_heap *heap)
4817 {
4818     HeapFree(GetProcessHeap(), 0, heap->entries);
4819 }
4820
4821 static const struct wine_rb_functions wined3d_glsl_program_rb_functions =
4822 {
4823     wined3d_rb_alloc,
4824     wined3d_rb_realloc,
4825     wined3d_rb_free,
4826     glsl_program_key_compare,
4827 };
4828
4829 static HRESULT shader_glsl_alloc(struct wined3d_device *device)
4830 {
4831     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4832     struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
4833     SIZE_T stack_size = wined3d_log2i(max(gl_info->limits.glsl_vs_float_constants,
4834             gl_info->limits.glsl_ps_float_constants)) + 1;
4835
4836     if (!shader_buffer_init(&priv->shader_buffer))
4837     {
4838         ERR("Failed to initialize shader buffer.\n");
4839         goto fail;
4840     }
4841
4842     priv->stack = HeapAlloc(GetProcessHeap(), 0, stack_size * sizeof(*priv->stack));
4843     if (!priv->stack)
4844     {
4845         ERR("Failed to allocate memory.\n");
4846         goto fail;
4847     }
4848
4849     if (!constant_heap_init(&priv->vconst_heap, gl_info->limits.glsl_vs_float_constants))
4850     {
4851         ERR("Failed to initialize vertex shader constant heap\n");
4852         goto fail;
4853     }
4854
4855     if (!constant_heap_init(&priv->pconst_heap, gl_info->limits.glsl_ps_float_constants))
4856     {
4857         ERR("Failed to initialize pixel shader constant heap\n");
4858         goto fail;
4859     }
4860
4861     if (wine_rb_init(&priv->program_lookup, &wined3d_glsl_program_rb_functions) == -1)
4862     {
4863         ERR("Failed to initialize rbtree.\n");
4864         goto fail;
4865     }
4866
4867     priv->next_constant_version = 1;
4868
4869     device->shader_priv = priv;
4870     return WINED3D_OK;
4871
4872 fail:
4873     constant_heap_free(&priv->pconst_heap);
4874     constant_heap_free(&priv->vconst_heap);
4875     HeapFree(GetProcessHeap(), 0, priv->stack);
4876     shader_buffer_free(&priv->shader_buffer);
4877     HeapFree(GetProcessHeap(), 0, priv);
4878     return E_OUTOFMEMORY;
4879 }
4880
4881 /* Context activation is done by the caller. */
4882 static void shader_glsl_free(struct wined3d_device *device)
4883 {
4884     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4885     struct shader_glsl_priv *priv = device->shader_priv;
4886     int i;
4887
4888     ENTER_GL();
4889     for (i = 0; i < tex_type_count; ++i)
4890     {
4891         if (priv->depth_blt_program_full[i])
4892         {
4893             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_full[i]));
4894         }
4895         if (priv->depth_blt_program_masked[i])
4896         {
4897             GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_masked[i]));
4898         }
4899     }
4900     LEAVE_GL();
4901
4902     wine_rb_destroy(&priv->program_lookup, NULL, NULL);
4903     constant_heap_free(&priv->pconst_heap);
4904     constant_heap_free(&priv->vconst_heap);
4905     HeapFree(GetProcessHeap(), 0, priv->stack);
4906     shader_buffer_free(&priv->shader_buffer);
4907
4908     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4909     device->shader_priv = NULL;
4910 }
4911
4912 static void shader_glsl_context_destroyed(void *shader_priv, const struct wined3d_context *context) {}
4913
4914 static void shader_glsl_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4915 {
4916     if (gl_info->supported[EXT_GPU_SHADER4] && gl_info->supported[ARB_GEOMETRY_SHADER4]
4917             && gl_info->glsl_version >= MAKEDWORD_VERSION(1, 50))
4918     {
4919         caps->VertexShaderVersion = 4;
4920         caps->PixelShaderVersion = 4;
4921     }
4922     /* ARB_shader_texture_lod or EXT_gpu_shader4 is required for the SM3
4923      * texldd and texldl instructions. */
4924     else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD] || gl_info->supported[EXT_GPU_SHADER4])
4925     {
4926         caps->VertexShaderVersion = 3;
4927         caps->PixelShaderVersion = 3;
4928     }
4929     else
4930     {
4931         caps->VertexShaderVersion = 2;
4932         caps->PixelShaderVersion = 2;
4933     }
4934
4935     caps->MaxVertexShaderConst = gl_info->limits.glsl_vs_float_constants;
4936     caps->MaxPixelShaderConst = gl_info->limits.glsl_ps_float_constants;
4937
4938     /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
4939      * Direct3D minimum requirement.
4940      *
4941      * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
4942      * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
4943      *
4944      * The problem is that the refrast clamps temporary results in the shader to
4945      * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
4946      * then applications may miss the clamping behavior. On the other hand, if it is smaller,
4947      * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
4948      * offer a way to query this.
4949      */
4950     caps->PixelShader1xMaxValue = 8.0;
4951
4952     caps->VSClipping = TRUE;
4953
4954     TRACE_(d3d_caps)("Hardware vertex shader version %u enabled (GLSL).\n",
4955             caps->VertexShaderVersion);
4956     TRACE_(d3d_caps)("Hardware pixel shader version %u enabled (GLSL).\n",
4957             caps->PixelShaderVersion);
4958 }
4959
4960 static BOOL shader_glsl_color_fixup_supported(struct color_fixup_desc fixup)
4961 {
4962     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4963     {
4964         TRACE("Checking support for fixup:\n");
4965         dump_color_fixup_desc(fixup);
4966     }
4967
4968     /* We support everything except YUV conversions. */
4969     if (!is_complex_fixup(fixup))
4970     {
4971         TRACE("[OK]\n");
4972         return TRUE;
4973     }
4974
4975     TRACE("[FAILED]\n");
4976     return FALSE;
4977 }
4978
4979 static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4980 {
4981     /* WINED3DSIH_ABS           */ shader_glsl_map2gl,
4982     /* WINED3DSIH_ADD           */ shader_glsl_arith,
4983     /* WINED3DSIH_AND           */ NULL,
4984     /* WINED3DSIH_BEM           */ shader_glsl_bem,
4985     /* WINED3DSIH_BREAK         */ shader_glsl_break,
4986     /* WINED3DSIH_BREAKC        */ shader_glsl_breakc,
4987     /* WINED3DSIH_BREAKP        */ NULL,
4988     /* WINED3DSIH_CALL          */ shader_glsl_call,
4989     /* WINED3DSIH_CALLNZ        */ shader_glsl_callnz,
4990     /* WINED3DSIH_CMP           */ shader_glsl_cmp,
4991     /* WINED3DSIH_CND           */ shader_glsl_cnd,
4992     /* WINED3DSIH_CRS           */ shader_glsl_cross,
4993     /* WINED3DSIH_CUT           */ NULL,
4994     /* WINED3DSIH_DCL           */ NULL,
4995     /* WINED3DSIH_DEF           */ NULL,
4996     /* WINED3DSIH_DEFB          */ NULL,
4997     /* WINED3DSIH_DEFI          */ NULL,
4998     /* WINED3DSIH_DIV           */ NULL,
4999     /* WINED3DSIH_DP2ADD        */ shader_glsl_dp2add,
5000     /* WINED3DSIH_DP3           */ shader_glsl_dot,
5001     /* WINED3DSIH_DP4           */ shader_glsl_dot,
5002     /* WINED3DSIH_DST           */ shader_glsl_dst,
5003     /* WINED3DSIH_DSX           */ shader_glsl_map2gl,
5004     /* WINED3DSIH_DSY           */ shader_glsl_map2gl,
5005     /* WINED3DSIH_ELSE          */ shader_glsl_else,
5006     /* WINED3DSIH_EMIT          */ NULL,
5007     /* WINED3DSIH_ENDIF         */ shader_glsl_end,
5008     /* WINED3DSIH_ENDLOOP       */ shader_glsl_end,
5009     /* WINED3DSIH_ENDREP        */ shader_glsl_end,
5010     /* WINED3DSIH_EXP           */ shader_glsl_map2gl,
5011     /* WINED3DSIH_EXPP          */ shader_glsl_expp,
5012     /* WINED3DSIH_FRC           */ shader_glsl_map2gl,
5013     /* WINED3DSIH_FTOI          */ NULL,
5014     /* WINED3DSIH_IADD          */ NULL,
5015     /* WINED3DSIH_IEQ           */ NULL,
5016     /* WINED3DSIH_IF            */ shader_glsl_if,
5017     /* WINED3DSIH_IFC           */ shader_glsl_ifc,
5018     /* WINED3DSIH_IGE           */ NULL,
5019     /* WINED3DSIH_IMUL          */ NULL,
5020     /* WINED3DSIH_ITOF          */ NULL,
5021     /* WINED3DSIH_LABEL         */ shader_glsl_label,
5022     /* WINED3DSIH_LD            */ NULL,
5023     /* WINED3DSIH_LIT           */ shader_glsl_lit,
5024     /* WINED3DSIH_LOG           */ shader_glsl_log,
5025     /* WINED3DSIH_LOGP          */ shader_glsl_log,
5026     /* WINED3DSIH_LOOP          */ shader_glsl_loop,
5027     /* WINED3DSIH_LRP           */ shader_glsl_lrp,
5028     /* WINED3DSIH_LT            */ NULL,
5029     /* WINED3DSIH_M3x2          */ shader_glsl_mnxn,
5030     /* WINED3DSIH_M3x3          */ shader_glsl_mnxn,
5031     /* WINED3DSIH_M3x4          */ shader_glsl_mnxn,
5032     /* WINED3DSIH_M4x3          */ shader_glsl_mnxn,
5033     /* WINED3DSIH_M4x4          */ shader_glsl_mnxn,
5034     /* WINED3DSIH_MAD           */ shader_glsl_mad,
5035     /* WINED3DSIH_MAX           */ shader_glsl_map2gl,
5036     /* WINED3DSIH_MIN           */ shader_glsl_map2gl,
5037     /* WINED3DSIH_MOV           */ shader_glsl_mov,
5038     /* WINED3DSIH_MOVA          */ shader_glsl_mov,
5039     /* WINED3DSIH_MOVC          */ NULL,
5040     /* WINED3DSIH_MUL           */ shader_glsl_arith,
5041     /* WINED3DSIH_NOP           */ NULL,
5042     /* WINED3DSIH_NRM           */ shader_glsl_nrm,
5043     /* WINED3DSIH_PHASE         */ NULL,
5044     /* WINED3DSIH_POW           */ shader_glsl_pow,
5045     /* WINED3DSIH_RCP           */ shader_glsl_rcp,
5046     /* WINED3DSIH_REP           */ shader_glsl_rep,
5047     /* WINED3DSIH_RET           */ shader_glsl_ret,
5048     /* WINED3DSIH_RSQ           */ shader_glsl_rsq,
5049     /* WINED3DSIH_SAMPLE        */ NULL,
5050     /* WINED3DSIH_SAMPLE_GRAD   */ NULL,
5051     /* WINED3DSIH_SAMPLE_LOD    */ NULL,
5052     /* WINED3DSIH_SETP          */ NULL,
5053     /* WINED3DSIH_SGE           */ shader_glsl_compare,
5054     /* WINED3DSIH_SGN           */ shader_glsl_sgn,
5055     /* WINED3DSIH_SINCOS        */ shader_glsl_sincos,
5056     /* WINED3DSIH_SLT           */ shader_glsl_compare,
5057     /* WINED3DSIH_SQRT          */ NULL,
5058     /* WINED3DSIH_SUB           */ shader_glsl_arith,
5059     /* WINED3DSIH_TEX           */ shader_glsl_tex,
5060     /* WINED3DSIH_TEXBEM        */ shader_glsl_texbem,
5061     /* WINED3DSIH_TEXBEML       */ shader_glsl_texbem,
5062     /* WINED3DSIH_TEXCOORD      */ shader_glsl_texcoord,
5063     /* WINED3DSIH_TEXDEPTH      */ shader_glsl_texdepth,
5064     /* WINED3DSIH_TEXDP3        */ shader_glsl_texdp3,
5065     /* WINED3DSIH_TEXDP3TEX     */ shader_glsl_texdp3tex,
5066     /* WINED3DSIH_TEXKILL       */ shader_glsl_texkill,
5067     /* WINED3DSIH_TEXLDD        */ shader_glsl_texldd,
5068     /* WINED3DSIH_TEXLDL        */ shader_glsl_texldl,
5069     /* WINED3DSIH_TEXM3x2DEPTH  */ shader_glsl_texm3x2depth,
5070     /* WINED3DSIH_TEXM3x2PAD    */ shader_glsl_texm3x2pad,
5071     /* WINED3DSIH_TEXM3x2TEX    */ shader_glsl_texm3x2tex,
5072     /* WINED3DSIH_TEXM3x3       */ shader_glsl_texm3x3,
5073     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5074     /* WINED3DSIH_TEXM3x3PAD    */ shader_glsl_texm3x3pad,
5075     /* WINED3DSIH_TEXM3x3SPEC   */ shader_glsl_texm3x3spec,
5076     /* WINED3DSIH_TEXM3x3TEX    */ shader_glsl_texm3x3tex,
5077     /* WINED3DSIH_TEXM3x3VSPEC  */ shader_glsl_texm3x3vspec,
5078     /* WINED3DSIH_TEXREG2AR     */ shader_glsl_texreg2ar,
5079     /* WINED3DSIH_TEXREG2GB     */ shader_glsl_texreg2gb,
5080     /* WINED3DSIH_TEXREG2RGB    */ shader_glsl_texreg2rgb,
5081     /* WINED3DSIH_UTOF          */ NULL,
5082 };
5083
5084 static void shader_glsl_handle_instruction(const struct wined3d_shader_instruction *ins) {
5085     SHADER_HANDLER hw_fct;
5086
5087     /* Select handler */
5088     hw_fct = shader_glsl_instruction_handler_table[ins->handler_idx];
5089
5090     /* Unhandled opcode */
5091     if (!hw_fct)
5092     {
5093         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5094         return;
5095     }
5096     hw_fct(ins);
5097
5098     shader_glsl_add_instruction_modifiers(ins);
5099 }
5100
5101 const struct wined3d_shader_backend_ops glsl_shader_backend =
5102 {
5103     shader_glsl_handle_instruction,
5104     shader_glsl_select,
5105     shader_glsl_select_depth_blt,
5106     shader_glsl_deselect_depth_blt,
5107     shader_glsl_update_float_vertex_constants,
5108     shader_glsl_update_float_pixel_constants,
5109     shader_glsl_load_constants,
5110     shader_glsl_load_np2fixup_constants,
5111     shader_glsl_destroy,
5112     shader_glsl_alloc,
5113     shader_glsl_free,
5114     shader_glsl_context_destroyed,
5115     shader_glsl_get_caps,
5116     shader_glsl_color_fixup_supported,
5117 };