1 /*
2  * Pixel and vertex shaders implementation using ARB_vertex_program
3  * and ARB_fragment_program GL extensions.
4  *
5  * Copyright 2002-2003 Jason Edmeades
6  * Copyright 2002-2003 Raphael Junqueira
7  * Copyright 2004 Christian Costa
8  * Copyright 2005 Oliver Stieber
9  * Copyright 2006 Ivan Gyurdiev
10  * Copyright 2006 Jason Green
11  * Copyright 2006 Henri Verbeet
12  * Copyright 2007-2011, 2013 Stefan Dösinger for CodeWeavers
13  * Copyright 2009 Henri Verbeet for CodeWeavers
14  *
15  * This library is free software; you can redistribute it and/or
16  * modify it under the terms of the GNU Lesser General Public
17  * License as published by the Free Software Foundation; either
18  * version 2.1 of the License, or (at your option) any later version.
19  *
20  * This library is distributed in the hope that it will be useful,
21  * but WITHOUT ANY WARRANTY; without even the implied warranty of
22  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
23  * Lesser General Public License for more details.
24  *
25  * You should have received a copy of the GNU Lesser General Public
26  * License along with this library; if not, write to the Free Software
27  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
28  */
29 
30 #include "wined3d_private.h"
31 
32 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
33 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
34 WINE_DECLARE_DEBUG_CHANNEL(d3d);
35 
36 static BOOL shader_is_pshader_version(enum wined3d_shader_type type)
37 {
38     return type == WINED3D_SHADER_TYPE_PIXEL;
39 }
40 
41 static BOOL shader_is_vshader_version(enum wined3d_shader_type type)
42 {
43     return type == WINED3D_SHADER_TYPE_VERTEX;
44 }
45 
46 /* Extract a line. Note that this modifies the source string. */
47 static char *get_line(char **ptr)
48 {
49     char *p, *q;
50 
51     p = *ptr;
52     if (!(q = strstr(p, "\n")))
53     {
54         if (!*p) return NULL;
55         *ptr += strlen(p);
56         return p;
57     }
58     *q = '\0';
59     *ptr = q + 1;
60 
61     return p;
62 }
63 
64 static void shader_arb_dump_program_source(const char *source)
65 {
66     ULONG source_size;
67     char *ptr, *line, *tmp;
68 
69     source_size = strlen(source) + 1;
70     tmp = HeapAlloc(GetProcessHeap(), 0, source_size);
71     if (!tmp)
72     {
73         ERR("Failed to allocate %u bytes for shader source.\n", source_size);
74         return;
75     }
76     memcpy(tmp, source, source_size);
77 
78     ptr = tmp;
79     while ((line = get_line(&ptr))) FIXME("    %s\n", line);
80     FIXME("\n");
81 
82     HeapFree(GetProcessHeap(), 0, tmp);
83 }
84 
85 enum arb_helper_value
86 {
87     ARB_ZERO,
88     ARB_ONE,
89     ARB_TWO,
90     ARB_0001,
91     ARB_EPS,
92 
93     ARB_VS_REL_OFFSET
94 };
95 
96 static const char *arb_get_helper_value(enum wined3d_shader_type shader, enum arb_helper_value value)
97 {
98     if (shader == WINED3D_SHADER_TYPE_GEOMETRY)
99     {
100         ERR("Geometry shaders are unsupported\n");
101         return "bad";
102     }
103 
104     if (shader == WINED3D_SHADER_TYPE_PIXEL)
105     {
106         switch (value)
107         {
108             case ARB_ZERO: return "ps_helper_const.x";
109             case ARB_ONE: return "ps_helper_const.y";
110             case ARB_TWO: return "coefmul.x";
111             case ARB_0001: return "ps_helper_const.xxxy";
112             case ARB_EPS: return "ps_helper_const.z";
113             default: break;
114         }
115     }
116     else
117     {
118         switch (value)
119         {
120             case ARB_ZERO: return "helper_const.x";
121             case ARB_ONE: return "helper_const.y";
122             case ARB_TWO: return "helper_const.z";
123             case ARB_EPS: return "helper_const.w";
124             case ARB_0001: return "helper_const.xxxy";
125             case ARB_VS_REL_OFFSET: return "rel_addr_const.y";
126         }
127     }
128     FIXME("Unmanaged %s shader helper constant requested: %u\n",
129           shader == WINED3D_SHADER_TYPE_PIXEL ? "pixel" : "vertex", value);
130     switch (value)
131     {
132         case ARB_ZERO: return "0.0";
133         case ARB_ONE: return "1.0";
134         case ARB_TWO: return "2.0";
135         case ARB_0001: return "{0.0, 0.0, 0.0, 1.0}";
136         case ARB_EPS: return "1e-8";
137         default: return "bad";
138     }
139 }
140 
141 static inline BOOL ffp_clip_emul(const struct wined3d_context *context)
142 {
143     return context->lowest_disabled_stage < 7;
144 }
145 
146 /* ARB_program_shader private data */
147 
148 struct control_frame
149 {
150     struct                          list entry;
151     enum
152     {
153         IF,
154         IFC,
155         LOOP,
156         REP
157     } type;
158     BOOL                            muting;
159     BOOL                            outer_loop;
160     union
161     {
162         unsigned int                loop;
163         unsigned int                ifc;
164     } no;
165     struct wined3d_shader_loop_control loop_control;
166     BOOL                            had_else;
167 };
168 
169 struct arb_ps_np2fixup_info
170 {
171     struct ps_np2fixup_info         super;
172     /* For ARB we need an offset value:
173      * With both GLSL and ARB mode the NP2 fixup information (the texture dimensions) are stored in a
174      * consecutive way (GLSL uses a uniform array). Since ARB doesn't know the notion of a "standalone"
175      * array we need an offset to the index inside the program local parameter array. */
176     UINT                            offset;
177 };
178 
179 struct arb_ps_compile_args
180 {
181     struct ps_compile_args          super;
182     WORD                            bools;
183     WORD                            clip;  /* only a boolean, use a WORD for alignment */
184     unsigned char                   loop_ctrl[MAX_CONST_I][3];
185 };
186 
187 struct stb_const_desc
188 {
189     unsigned char           texunit;
190     UINT                    const_num;
191 };
192 
193 struct arb_ps_compiled_shader
194 {
195     struct arb_ps_compile_args      args;
196     struct arb_ps_np2fixup_info     np2fixup_info;
197     struct stb_const_desc           bumpenvmatconst[MAX_TEXTURES];
198     struct stb_const_desc           luminanceconst[MAX_TEXTURES];
199     UINT                            int_consts[MAX_CONST_I];
200     GLuint                          prgId;
201     UINT                            ycorrection;
202     unsigned char                   numbumpenvmatconsts;
203     char                            num_int_consts;
204 };
205 
206 struct arb_vs_compile_args
207 {
208     struct vs_compile_args          super;
209     union
210     {
211         struct
212         {
213             WORD                    bools;
214             unsigned char           clip_texcoord;
215             unsigned char           clipplane_mask;
216         }                           boolclip;
217         DWORD                       boolclip_compare;
218     } clip;
219     DWORD                           ps_signature;
220     union
221     {
222         unsigned char               samplers[4];
223         DWORD                       samplers_compare;
224     } vertex;
225     unsigned char                   loop_ctrl[MAX_CONST_I][3];
226 };
227 
228 struct arb_vs_compiled_shader
229 {
230     struct arb_vs_compile_args      args;
231     GLuint                          prgId;
232     UINT                            int_consts[MAX_CONST_I];
233     char                            num_int_consts;
234     char                            need_color_unclamp;
235     UINT                            pos_fixup;
236 };
237 
238 struct recorded_instruction
239 {
240     struct wined3d_shader_instruction ins;
241     struct list entry;
242 };
243 
244 struct shader_arb_ctx_priv
245 {
246     char addr_reg[20];
247     enum
248     {
249         /* plain GL_ARB_vertex_program or GL_ARB_fragment_program */
250         ARB,
251         /* GL_NV_vertex_progam2_option or GL_NV_fragment_program_option */
252         NV2,
253         /* GL_NV_vertex_program3 or GL_NV_fragment_program2 */
254         NV3
255     } target_version;
256 
257     const struct arb_vs_compile_args    *cur_vs_args;
258     const struct arb_ps_compile_args    *cur_ps_args;
259     const struct arb_ps_compiled_shader *compiled_fprog;
260     const struct arb_vs_compiled_shader *compiled_vprog;
261     struct arb_ps_np2fixup_info         *cur_np2fixup_info;
262     struct list                         control_frames;
263     struct list                         record;
264     BOOL                                recording;
265     BOOL                                muted;
266     unsigned int                        num_loops, loop_depth, num_ifcs;
267     int                                 aL;
268     BOOL                                ps_post_process;
269 
270     unsigned int                        vs_clipplanes;
271     BOOL                                footer_written;
272     BOOL                                in_main_func;
273 
274     /* For 3.0 vertex shaders */
275     const char                          *vs_output[MAX_REG_OUTPUT];
276     /* For 2.x and earlier vertex shaders */
277     const char                          *texcrd_output[8], *color_output[2], *fog_output;
278 
279     /* 3.0 pshader input for compatibility with fixed function */
280     const char                          *ps_input[MAX_REG_INPUT];
281 };
282 
283 struct ps_signature
284 {
285     struct wined3d_shader_signature_element *sig;
286     DWORD                               idx;
287     struct wine_rb_entry                entry;
288 };
289 
290 struct arb_pshader_private {
291     struct arb_ps_compiled_shader   *gl_shaders;
292     UINT                            num_gl_shaders, shader_array_size;
293     DWORD                           input_signature_idx;
294     DWORD                           clipplane_emulation;
295     BOOL                            clamp_consts;
296 };
297 
298 struct arb_vshader_private {
299     struct arb_vs_compiled_shader   *gl_shaders;
300     UINT                            num_gl_shaders, shader_array_size;
301     UINT rel_offset;
302 };
303 
304 struct shader_arb_priv
305 {
306     GLuint                  current_vprogram_id;
307     GLuint                  current_fprogram_id;
308     const struct arb_ps_compiled_shader *compiled_fprog;
309     const struct arb_vs_compiled_shader *compiled_vprog;
310     GLuint                  depth_blt_vprogram_id;
311     GLuint                  depth_blt_fprogram_id_full[tex_type_count];
312     GLuint                  depth_blt_fprogram_id_masked[tex_type_count];
313     BOOL                    use_arbfp_fixed_func;
314     struct wine_rb_tree     fragment_shaders;
315     BOOL                    last_ps_const_clamped;
316     BOOL                    last_vs_color_unclamp;
317 
318     struct wine_rb_tree     signature_tree;
319     DWORD ps_sig_number;
320 
321     unsigned int highest_dirty_ps_const, highest_dirty_vs_const;
322     char *vshader_const_dirty, *pshader_const_dirty;
323     const struct wined3d_context *last_context;
324 
325     const struct wined3d_vertex_pipe_ops *vertex_pipe;
326     const struct fragment_pipeline *fragment_pipe;
327     BOOL ffp_proj_control;
328 };
329 
330 /* Context activation for state handlers is done by the caller. */
331 
332 static BOOL need_rel_addr_const(const struct arb_vshader_private *shader_data,
333         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
334 {
335     if (shader_data->rel_offset) return TRUE;
336     if (!reg_maps->usesmova) return FALSE;
337     return !gl_info->supported[NV_VERTEX_PROGRAM2_OPTION];
338 }
339 
340 /* Returns TRUE if result.clip from GL_NV_vertex_program2 should be used and FALSE otherwise */
341 static inline BOOL use_nv_clip(const struct wined3d_gl_info *gl_info)
342 {
343     return gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
344             && !(gl_info->quirks & WINED3D_QUIRK_NV_CLIP_BROKEN);
345 }
346 
347 static BOOL need_helper_const(const struct arb_vshader_private *shader_data,
348         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
349 {
350     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) return TRUE;
351     if (!gl_info->supported[NV_VERTEX_PROGRAM]) return TRUE; /* Need to init colors. */
352     if (gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT) return TRUE; /* Load the immval offset. */
353     if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) return TRUE; /* Have to init texcoords. */
354     if (!use_nv_clip(gl_info)) return TRUE; /* Init the clip texcoord */
355     if (reg_maps->usesnrm) return TRUE; /* 0.0 */
356     if (reg_maps->usespow) return TRUE; /* EPS, 0.0 and 1.0 */
357     if (reg_maps->fog) return TRUE; /* Clamping fog coord, 0.0 and 1.0 */
358     return FALSE;
359 }
360 
361 static unsigned int reserved_vs_const(const struct arb_vshader_private *shader_data,
362         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
363 {
364     unsigned int ret = 1;
365     /* We use one PARAM for the pos fixup, and in some cases one to load
366      * some immediate values into the shader. */
367     if (need_helper_const(shader_data, reg_maps, gl_info)) ++ret;
368     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) ++ret;
369     return ret;
370 }
371 
372 /* Loads floating point constants into the currently set ARB_vertex/fragment_program.
373  * When constant_list == NULL, it will load all the constants.
374  *
375  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
376  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
377  */
378 /* Context activation is done by the caller. */
379 static unsigned int shader_arb_load_constantsF(const struct wined3d_shader *shader,
380         const struct wined3d_gl_info *gl_info, GLuint target_type, unsigned int max_constants,
381         const float *constants, char *dirty_consts)
382 {
383     struct wined3d_shader_lconst *lconst;
384     DWORD i, j;
385     unsigned int ret;
386 
387     if (TRACE_ON(d3d_constants))
388     {
389         for(i = 0; i < max_constants; i++) {
390             if(!dirty_consts[i]) continue;
391             TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
392                         constants[i * 4 + 0], constants[i * 4 + 1],
393                         constants[i * 4 + 2], constants[i * 4 + 3]);
394         }
395     }
396 
397     i = 0;
398 
399     /* In 1.X pixel shaders constants are implicitly clamped in the range [-1;1] */
400     if (target_type == GL_FRAGMENT_PROGRAM_ARB && shader->reg_maps.shader_version.major == 1)
401     {
402         float lcl_const[4];
403         /* ps 1.x supports only 8 constants, clamp only those. When switching between 1.x and higher
404          * shaders, the first 8 constants are marked dirty for reload
405          */
406         for(; i < min(8, max_constants); i++) {
407             if(!dirty_consts[i]) continue;
408             dirty_consts[i] = 0;
409 
410             j = 4 * i;
411             if (constants[j + 0] > 1.0f) lcl_const[0] = 1.0f;
412             else if (constants[j + 0] < -1.0f) lcl_const[0] = -1.0f;
413             else lcl_const[0] = constants[j + 0];
414 
415             if (constants[j + 1] > 1.0f) lcl_const[1] = 1.0f;
416             else if (constants[j + 1] < -1.0f) lcl_const[1] = -1.0f;
417             else lcl_const[1] = constants[j + 1];
418 
419             if (constants[j + 2] > 1.0f) lcl_const[2] = 1.0f;
420             else if (constants[j + 2] < -1.0f) lcl_const[2] = -1.0f;
421             else lcl_const[2] = constants[j + 2];
422 
423             if (constants[j + 3] > 1.0f) lcl_const[3] = 1.0f;
424             else if (constants[j + 3] < -1.0f) lcl_const[3] = -1.0f;
425             else lcl_const[3] = constants[j + 3];
426 
427             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, lcl_const));
428         }
429 
430         /* If further constants are dirty, reload them without clamping.
431          *
432          * The alternative is not to touch them, but then we cannot reset the dirty constant count
433          * to zero. That's bad for apps that only use PS 1.x shaders, because in that case the code
434          * above would always re-check the first 8 constants since max_constant remains at the init
435          * value
436          */
437     }
438 
439     if (gl_info->supported[EXT_GPU_PROGRAM_PARAMETERS])
440     {
441         /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
442          * or just reloading *all* constants at once
443          *
444         GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, i, max_constants, constants + (i * 4)));
445          */
446         for(; i < max_constants; i++) {
447             if(!dirty_consts[i]) continue;
448 
449             /* Find the next block of dirty constants */
450             dirty_consts[i] = 0;
451             j = i;
452             for(i++; (i < max_constants) && dirty_consts[i]; i++) {
453                 dirty_consts[i] = 0;
454             }
455 
456             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, constants + (j * 4)));
457         }
458     } else {
459         for(; i < max_constants; i++) {
460             if(dirty_consts[i]) {
461                 dirty_consts[i] = 0;
462                 GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, constants + (i * 4)));
463             }
464         }
465     }
466     checkGLcall("glProgramEnvParameter4fvARB()");
467 
468     /* Load immediate constants */
469     if (shader->load_local_constsF)
470     {
471         if (TRACE_ON(d3d_shader))
472         {
473             LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
474             {
475                 GLfloat* values = (GLfloat*)lconst->value;
476                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
477                         values[0], values[1], values[2], values[3]);
478             }
479         }
480         /* Immediate constants are clamped for 1.X shaders at loading times */
481         ret = 0;
482         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
483         {
484             dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
485             ret = max(ret, lconst->idx + 1);
486             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
487         }
488         checkGLcall("glProgramEnvParameter4fvARB()");
489         return ret; /* The loaded immediate constants need reloading for the next shader */
490     } else {
491         return 0; /* No constants are dirty now */
492     }
493 }
494 
495 /* Loads the texture dimensions for NP2 fixup into the currently set
496  * ARB_[vertex/fragment]_programs. */
497 static void shader_arb_load_np2fixup_constants(const struct arb_ps_np2fixup_info *fixup,
498         const struct wined3d_gl_info *gl_info, const struct wined3d_state *state)
499 {
500     GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
501     WORD active = fixup->super.active;
502     UINT i;
503 
504     if (!active)
505         return;
506 
507     for (i = 0; active; active >>= 1, ++i)
508     {
509         const struct wined3d_texture *tex = state->textures[i];
510         unsigned char idx = fixup->super.idx[i];
511         GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
512 
513         if (!(active & 1))
514             continue;
515 
516         if (!tex)
517         {
518             ERR("Nonexistent texture is flagged for NP2 texcoord fixup.\n");
519             continue;
520         }
521 
522         if (idx % 2)
523         {
524             tex_dim[2] = tex->pow2_matrix[0];
525             tex_dim[3] = tex->pow2_matrix[5];
526         }
527         else
528         {
529             tex_dim[0] = tex->pow2_matrix[0];
530             tex_dim[1] = tex->pow2_matrix[5];
531         }
532     }
533 
534     for (i = 0; i < fixup->super.num_consts; ++i)
535     {
536         GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
537                 fixup->offset + i, &np2fixup_constants[i * 4]));
538     }
539 }
540 
541 /* Context activation is done by the caller. */
542 static void shader_arb_ps_local_constants(const struct arb_ps_compiled_shader *gl_shader,
543         const struct wined3d_context *context, const struct wined3d_state *state, UINT rt_height)
544 {
545     const struct wined3d_gl_info *gl_info = context->gl_info;
546     unsigned char i;
547 
548     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
549     {
550         int texunit = gl_shader->bumpenvmatconst[i].texunit;
551 
552         /* The state manager takes care that this function is always called if the bump env matrix changes */
553         const float *data = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_MAT00];
554         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
555                 gl_shader->bumpenvmatconst[i].const_num, data));
556 
557         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
558         {
559             /* WINED3D_TSS_BUMPENVLSCALE and WINED3D_TSS_BUMPENVLOFFSET are next to each other.
560              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
561              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
562              * (they're WINED3D_TSS_TEXTURETRANSFORMFLAGS and WINED3D_TSS_ADDRESSW, so most likely 0 or NaN
563             */
564             const float *scale = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_LSCALE];
565             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
566                     gl_shader->luminanceconst[i].const_num, scale));
567         }
568     }
569     checkGLcall("Load bumpmap consts");
570 
571     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
572     {
573         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
574         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
575         * ycorrection.z: 1.0
576         * ycorrection.w: 0.0
577         */
578         float val[4];
579         val[0] = context->render_offscreen ? 0.0f : (float) rt_height;
580         val[1] = context->render_offscreen ? 1.0f : -1.0f;
581         val[2] = 1.0f;
582         val[3] = 0.0f;
583         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
584         checkGLcall("y correction loading");
585     }
586 
587     if (!gl_shader->num_int_consts) return;
588 
589     for(i = 0; i < MAX_CONST_I; i++)
590     {
591         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
592         {
593             float val[4];
594             val[0] = (float)state->ps_consts_i[4 * i];
595             val[1] = (float)state->ps_consts_i[4 * i + 1];
596             val[2] = (float)state->ps_consts_i[4 * i + 2];
597             val[3] = -1.0f;
598 
599             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
600         }
601     }
602     checkGLcall("Load ps int consts");
603 }
604 
605 /* Context activation is done by the caller. */
606 static void shader_arb_vs_local_constants(const struct arb_vs_compiled_shader *gl_shader,
607         const struct wined3d_context *context, const struct wined3d_state *state)
608 {
609     const struct wined3d_gl_info *gl_info = context->gl_info;
610     float position_fixup[4];
611     unsigned char i;
612 
613     /* Upload the position fixup */
614     shader_get_position_fixup(context, state, position_fixup);
615     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, position_fixup));
616 
617     if (!gl_shader->num_int_consts) return;
618 
619     for(i = 0; i < MAX_CONST_I; i++)
620     {
621         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
622         {
623             float val[4];
624             val[0] = (float)state->vs_consts_i[4 * i];
625             val[1] = (float)state->vs_consts_i[4 * i + 1];
626             val[2] = (float)state->vs_consts_i[4 * i + 2];
627             val[3] = -1.0f;
628 
629             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
630         }
631     }
632     checkGLcall("Load vs int consts");
633 }
634 
635 static void shader_arb_select(void *shader_priv, struct wined3d_context *context,
636         const struct wined3d_state *state);
637 
638 /**
639  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
640  *
641  * We only support float constants in ARB at the moment, so don't
642  * worry about the Integers or Booleans
643  */
644 /* Context activation is done by the caller (state handler). */
645 static void shader_arb_load_constants_internal(struct shader_arb_priv *priv,
646         struct wined3d_context *context, const struct wined3d_state *state,
647         BOOL usePixelShader, BOOL useVertexShader, BOOL from_shader_select)
648 {
649     const struct wined3d_d3d_info *d3d_info = context->d3d_info;
650     const struct wined3d_gl_info *gl_info = context->gl_info;
651 
652     if (!from_shader_select)
653     {
654         const struct wined3d_shader *vshader = state->shader[WINED3D_SHADER_TYPE_VERTEX];
655         const struct wined3d_shader *pshader = state->shader[WINED3D_SHADER_TYPE_PIXEL];
656 
657         if (vshader
658                 && (vshader->reg_maps.boolean_constants
659                 || (!gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
660                 && (vshader->reg_maps.integer_constants & ~vshader->reg_maps.local_int_consts))))
661         {
662             TRACE("bool/integer vertex shader constants potentially modified, forcing shader reselection.\n");
663             shader_arb_select(priv, context, state);
664         }
665         else if (pshader
666                 && (pshader->reg_maps.boolean_constants
667                 || (!gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION]
668                 && (pshader->reg_maps.integer_constants & ~pshader->reg_maps.local_int_consts))))
669         {
670             TRACE("bool/integer pixel shader constants potentially modified, forcing shader reselection.\n");
671             shader_arb_select(priv, context, state);
672         }
673     }
674 
675     if (context != priv->last_context)
676     {
677         memset(priv->vshader_const_dirty, 1,
678                 sizeof(*priv->vshader_const_dirty) * d3d_info->limits.vs_uniform_count);
679         priv->highest_dirty_vs_const = d3d_info->limits.vs_uniform_count;
680 
681         memset(priv->pshader_const_dirty, 1,
682                 sizeof(*priv->pshader_const_dirty) * d3d_info->limits.ps_uniform_count);
683         priv->highest_dirty_ps_const = d3d_info->limits.ps_uniform_count;
684 
685         priv->last_context = context;
686     }
687 
688     if (useVertexShader)
689     {
690         const struct wined3d_shader *vshader = state->shader[WINED3D_SHADER_TYPE_VERTEX];
691         const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
692 
693         /* Load DirectX 9 float constants for vertex shader */
694         priv->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
695                 priv->highest_dirty_vs_const, state->vs_consts_f, priv->vshader_const_dirty);
696         shader_arb_vs_local_constants(gl_shader, context, state);
697     }
698 
699     if (usePixelShader)
700     {
701         const struct wined3d_shader *pshader = state->shader[WINED3D_SHADER_TYPE_PIXEL];
702         const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
703         UINT rt_height = state->fb->render_targets[0]->resource.height;
704 
705         /* Load DirectX 9 float constants for pixel shader */
706         priv->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
707                 priv->highest_dirty_ps_const, state->ps_consts_f, priv->pshader_const_dirty);
708         shader_arb_ps_local_constants(gl_shader, context, state, rt_height);
709 
710         if (context->constant_update_mask & WINED3D_SHADER_CONST_PS_NP2_FIXUP)
711             shader_arb_load_np2fixup_constants(&gl_shader->np2fixup_info, gl_info, state);
712     }
713 }
714 
715 static void shader_arb_load_constants(void *shader_priv, struct wined3d_context *context,
716         const struct wined3d_state *state)
717 {
718     BOOL vs = use_vs(state);
719     BOOL ps = use_ps(state);
720 
721     shader_arb_load_constants_internal(shader_priv, context, state, ps, vs, FALSE);
722 }
723 
724 static void shader_arb_update_float_vertex_constants(struct wined3d_device *device, UINT start, UINT count)
725 {
726     struct wined3d_context *context = context_get_current();
727     struct shader_arb_priv *priv = device->shader_priv;
728     unsigned int i;
729 
730     for (i = 0; i < device->context_count; ++i)
731     {
732         device->contexts[i]->constant_update_mask |= WINED3D_SHADER_CONST_VS_F;
733     }
734 
735     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
736      * context. On a context switch the old context will be fully dirtified */
737     if (!context || context->swapchain->device != device) return;
738 
739     memset(priv->vshader_const_dirty + start, 1, sizeof(*priv->vshader_const_dirty) * count);
740     priv->highest_dirty_vs_const = max(priv->highest_dirty_vs_const, start + count);
741 }
742 
743 static void shader_arb_update_float_pixel_constants(struct wined3d_device *device, UINT start, UINT count)
744 {
745     struct wined3d_context *context = context_get_current();
746     struct shader_arb_priv *priv = device->shader_priv;
747     unsigned int i;
748 
749     for (i = 0; i < device->context_count; ++i)
750     {
751         device->contexts[i]->constant_update_mask |= WINED3D_SHADER_CONST_PS_F;
752     }
753 
754     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
755      * context. On a context switch the old context will be fully dirtified */
756     if (!context || context->swapchain->device != device) return;
757 
758     memset(priv->pshader_const_dirty + start, 1, sizeof(*priv->pshader_const_dirty) * count);
759     priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, start + count);
760 }
761 
762 static void shader_arb_append_imm_vec4(struct wined3d_shader_buffer *buffer, const float *values)
763 {
764     char str[4][17];
765 
766     wined3d_ftoa(values[0], str[0]);
767     wined3d_ftoa(values[1], str[1]);
768     wined3d_ftoa(values[2], str[2]);
769     wined3d_ftoa(values[3], str[3]);
770     shader_addline(buffer, "{%s, %s, %s, %s}", str[0], str[1], str[2], str[3]);
771 }
772 
773 /* Generate the variable & register declarations for the ARB_vertex_program output target */
774 static void shader_generate_arb_declarations(const struct wined3d_shader *shader,
775         const struct wined3d_shader_reg_maps *reg_maps, struct wined3d_shader_buffer *buffer,
776         const struct wined3d_gl_info *gl_info, DWORD *num_clipplanes,
777         const struct shader_arb_ctx_priv *ctx)
778 {
779     DWORD i;
780     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
781     const struct wined3d_shader_lconst *lconst;
782     unsigned max_constantsF;
783     DWORD map;
784 
785     /* In pixel shaders, all private constants are program local, we don't need anything
786      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
787      * If we need a private constant the GL implementation will squeeze it in somewhere
788      *
789      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
790      * immediate values. The posFixup is loaded using program.env for now, so always
791      * subtract one from the number of constants. If the shader uses indirect addressing,
792      * account for the helper const too because we have to declare all available d3d constants
793      * and don't know which are actually used.
794      */
795     if (pshader)
796     {
797         max_constantsF = gl_info->limits.arb_ps_native_constants;
798         /* 24 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value. */
799         if (max_constantsF < 24)
800             max_constantsF = gl_info->limits.arb_ps_float_constants;
801     }
802     else
803     {
804         const struct arb_vshader_private *shader_data = shader->backend_data;
805         max_constantsF = gl_info->limits.arb_vs_native_constants;
806         /* 96 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value.
807          * Also prevents max_constantsF from becoming less than 0 and
808          * wrapping . */
809         if (max_constantsF < 96)
810             max_constantsF = gl_info->limits.arb_vs_float_constants;
811 
812         if (reg_maps->usesrelconstF)
813         {
814             DWORD highest_constf = 0, clip_limit;
815 
816             max_constantsF -= reserved_vs_const(shader_data, reg_maps, gl_info);
817             max_constantsF -= count_bits(reg_maps->integer_constants);
818             max_constantsF -= gl_info->reserved_arb_constants;
819 
820             for (i = 0; i < shader->limits.constant_float; ++i)
821             {
822                 DWORD idx = i >> 5;
823                 DWORD shift = i & 0x1f;
824                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
825             }
826 
827             if(use_nv_clip(gl_info) && ctx->target_version >= NV2)
828             {
829                 if(ctx->cur_vs_args->super.clip_enabled)
830                     clip_limit = gl_info->limits.clipplanes;
831                 else
832                     clip_limit = 0;
833             }
834             else
835             {
836                 unsigned int mask = ctx->cur_vs_args->clip.boolclip.clipplane_mask;
837                 clip_limit = min(count_bits(mask), 4);
838             }
839             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
840             max_constantsF -= *num_clipplanes;
841             if(*num_clipplanes < clip_limit)
842             {
843                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, gl_info->limits.clipplanes);
844             }
845         }
846         else
847         {
848             if (ctx->target_version >= NV2) *num_clipplanes = gl_info->limits.clipplanes;
849             else *num_clipplanes = min(gl_info->limits.clipplanes, 4);
850         }
851     }
852 
853     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
854     {
855         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
856     }
857 
858     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
859     {
860         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
861     }
862 
863     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
864     {
865         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
866         {
867             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
868         }
869     }
870 
871     if (!shader->load_local_constsF)
872     {
873         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
874         {
875             const float *value;
876             value = (const float *)lconst->value;
877             shader_addline(buffer, "PARAM C%u = ", lconst->idx);
878             shader_arb_append_imm_vec4(buffer, value);
879             shader_addline(buffer, ";\n");
880         }
881     }
882 
883     /* After subtracting privately used constants from the hardware limit(they are loaded as
884      * local constants), make sure the shader doesn't violate the env constant limit
885      */
886     if(pshader)
887     {
888         max_constantsF = min(max_constantsF, gl_info->limits.arb_ps_float_constants);
889     }
890     else
891     {
892         max_constantsF = min(max_constantsF, gl_info->limits.arb_vs_float_constants);
893     }
894 
895     /* Avoid declaring more constants than needed */
896     max_constantsF = min(max_constantsF, shader->limits.constant_float);
897 
898     /* we use the array-based constants array if the local constants are marked for loading,
899      * because then we use indirect addressing, or when the local constant list is empty,
900      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
901      * local constants do not declare the loaded constants as an array because ARB compilers usually
902      * do not optimize unused constants away
903      */
904     if (reg_maps->usesrelconstF)
905     {
906         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
907         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
908                     max_constantsF, max_constantsF - 1);
909     } else {
910         for(i = 0; i < max_constantsF; i++) {
911             DWORD idx, mask;
912             idx = i >> 5;
913             mask = 1 << (i & 0x1f);
914             if (!shader_constant_is_local(shader, i) && (reg_maps->constf[idx] & mask))
915             {
916                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
917             }
918         }
919     }
920 }
921 
922 static const char * const shift_tab[] = {
923     "dummy",     /*  0 (none) */
924     "coefmul.x", /*  1 (x2)   */
925     "coefmul.y", /*  2 (x4)   */
926     "coefmul.z", /*  3 (x8)   */
927     "coefmul.w", /*  4 (x16)  */
928     "dummy",     /*  5 (x32)  */
929     "dummy",     /*  6 (x64)  */
930     "dummy",     /*  7 (x128) */
931     "dummy",     /*  8 (d256) */
932     "dummy",     /*  9 (d128) */
933     "dummy",     /* 10 (d64)  */
934     "dummy",     /* 11 (d32)  */
935     "coefdiv.w", /* 12 (d16)  */
936     "coefdiv.z", /* 13 (d8)   */
937     "coefdiv.y", /* 14 (d4)   */
938     "coefdiv.x"  /* 15 (d2)   */
939 };
940 
941 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
942         const struct wined3d_shader_dst_param *dst, char *write_mask)
943 {
944     char *ptr = write_mask;
945 
946     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
947     {
948         *ptr++ = '.';
949         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
950         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
951         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
952         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
953     }
954 
955     *ptr = '\0';
956 }
957 
958 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
959 {
960     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
961      * but addressed as "rgba". To fix this we need to swap the register's x
962      * and z components. */
963     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
964     char *ptr = swizzle_str;
965 
966     /* swizzle bits fields: wwzzyyxx */
967     DWORD swizzle = param->swizzle;
968     DWORD swizzle_x = swizzle & 0x03;
969     DWORD swizzle_y = (swizzle >> 2) & 0x03;
970     DWORD swizzle_z = (swizzle >> 4) & 0x03;
971     DWORD swizzle_w = (swizzle >> 6) & 0x03;
972 
973     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
974      * generate a swizzle string. Unless we need to our own swizzling. */
975     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
976     {
977         *ptr++ = '.';
978         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
979             *ptr++ = swizzle_chars[swizzle_x];
980         } else {
981             *ptr++ = swizzle_chars[swizzle_x];
982             *ptr++ = swizzle_chars[swizzle_y];
983             *ptr++ = swizzle_chars[swizzle_z];
984             *ptr++ = swizzle_chars[swizzle_w];
985         }
986     }
987 
988     *ptr = '\0';
989 }
990 
991 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
992 {
993     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
994     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
995 
996     if (!strcmp(priv->addr_reg, src)) return;
997 
998     strcpy(priv->addr_reg, src);
999     shader_addline(buffer, "ARL A0.x, %s;\n", src);
1000 }
1001 
1002 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1003         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
1004 
1005 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
1006         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
1007 {
1008     /* oPos, oFog and oPts in D3D */
1009     static const char * const rastout_reg_names[] = {"TMP_OUT", "TMP_FOGCOORD", "result.pointsize"};
1010     const struct wined3d_shader *shader = ins->ctx->shader;
1011     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1012     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1013     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1014 
1015     *is_color = FALSE;
1016 
1017     switch (reg->type)
1018     {
1019         case WINED3DSPR_TEMP:
1020             sprintf(register_name, "R%u", reg->idx[0].offset);
1021             break;
1022 
1023         case WINED3DSPR_INPUT:
1024             if (pshader)
1025             {
1026                 if (reg_maps->shader_version.major < 3)
1027                 {
1028                     if (!reg->idx[0].offset)
1029                         strcpy(register_name, "fragment.color.primary");
1030                     else
1031                         strcpy(register_name, "fragment.color.secondary");
1032                 }
1033                 else
1034                 {
1035                     if (reg->idx[0].rel_addr)
1036                     {
1037                         char rel_reg[50];
1038                         shader_arb_get_src_param(ins, reg->idx[0].rel_addr, 0, rel_reg);
1039 
1040                         if (!strcmp(rel_reg, "**aL_emul**"))
1041                         {
1042                             DWORD idx = ctx->aL + reg->idx[0].offset;
1043                             if(idx < MAX_REG_INPUT)
1044                             {
1045                                 strcpy(register_name, ctx->ps_input[idx]);
1046                             }
1047                             else
1048                             {
1049                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
1050                                 sprintf(register_name, "out_of_bounds_%u", idx);
1051                             }
1052                         }
1053                         else if (reg_maps->input_registers & 0x0300)
1054                         {
1055                             /* There are two ways basically:
1056                              *
1057                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
1058                              *    That means trouble if the loop also contains a breakc or if the control values
1059                              *    aren't local constants.
1060                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
1061                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
1062                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
1063                              *    ADAC to load the condition code register and pop it again afterwards
1064                              */
1065                             FIXME("Relative input register addressing with more than 8 registers\n");
1066 
1067                             /* This is better than nothing for now */
1068                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1069                         }
1070                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
1071                         {
1072                             /* This is problematic because we'd have to consult the ctx->ps_input strings
1073                              * for where to find the varying. Some may be "0.0", others can be texcoords or
1074                              * colors. This needs either a pipeline replacement to make the vertex shader feed
1075                              * proper varyings, or loop unrolling
1076                              *
1077                              * For now use the texcoords and hope for the best
1078                              */
1079                             FIXME("Non-vertex shader varying input with indirect addressing\n");
1080                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1081                         }
1082                         else
1083                         {
1084                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
1085                              * pulls GL_NV_fragment_program2 in
1086                              */
1087                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1088                         }
1089                     }
1090                     else
1091                     {
1092                         if (reg->idx[0].offset < MAX_REG_INPUT)
1093                         {
1094                             strcpy(register_name, ctx->ps_input[reg->idx[0].offset]);
1095                         }
1096                         else
1097                         {
1098                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx[0].offset);
1099                             sprintf(register_name, "out_of_bounds_%u", reg->idx[0].offset);
1100                         }
1101                     }
1102                 }
1103             }
1104             else
1105             {
1106                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx[0].offset))
1107                     *is_color = TRUE;
1108                 sprintf(register_name, "vertex.attrib[%u]", reg->idx[0].offset);
1109             }
1110             break;
1111 
1112         case WINED3DSPR_CONST:
1113             if (!pshader && reg->idx[0].rel_addr)
1114             {
1115                 const struct arb_vshader_private *shader_data = shader->backend_data;
1116                 UINT rel_offset = ctx->target_version == ARB ? shader_data->rel_offset : 0;
1117                 BOOL aL = FALSE;
1118                 char rel_reg[50];
1119                 if (reg_maps->shader_version.major < 2)
1120                 {
1121                     sprintf(rel_reg, "A0.x");
1122                 }
1123                 else
1124                 {
1125                     shader_arb_get_src_param(ins, reg->idx[0].rel_addr, 0, rel_reg);
1126                     if (ctx->target_version == ARB)
1127                     {
1128                         if (!strcmp(rel_reg, "**aL_emul**"))
1129                         {
1130                             aL = TRUE;
1131                         } else {
1132                             shader_arb_request_a0(ins, rel_reg);
1133                             sprintf(rel_reg, "A0.x");
1134                         }
1135                     }
1136                 }
1137                 if (aL)
1138                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx[0].offset);
1139                 else if (reg->idx[0].offset >= rel_offset)
1140                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx[0].offset - rel_offset);
1141                 else
1142                     sprintf(register_name, "C[%s - %u]", rel_reg, rel_offset - reg->idx[0].offset);
1143             }
1144             else
1145             {
1146                 if (reg_maps->usesrelconstF)
1147                     sprintf(register_name, "C[%u]", reg->idx[0].offset);
1148                 else
1149                     sprintf(register_name, "C%u", reg->idx[0].offset);
1150             }
1151             break;
1152 
1153         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1154             if (pshader)
1155             {
1156                 if (reg_maps->shader_version.major == 1
1157                         && reg_maps->shader_version.minor <= 3)
1158                     /* In ps <= 1.3, Tx is a temporary register as destination
1159                      * to all instructions, and as source to most instructions.
1160                      * For some instructions it is the texcoord input. Those
1161                      * instructions know about the special use. */
1162                     sprintf(register_name, "T%u", reg->idx[0].offset);
1163                 else
1164                     /* In ps 1.4 and 2.x Tx is always a (read-only) varying. */
1165                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx[0].offset);
1166             }
1167             else
1168             {
1169                 if (reg_maps->shader_version.major == 1 || ctx->target_version >= NV2)
1170                     sprintf(register_name, "A%u", reg->idx[0].offset);
1171                 else
1172                     sprintf(register_name, "A%u_SHADOW", reg->idx[0].offset);
1173             }
1174             break;
1175 
1176         case WINED3DSPR_COLOROUT:
1177             if (ctx->ps_post_process && !reg->idx[0].offset)
1178             {
1179                 strcpy(register_name, "TMP_COLOR");
1180             }
1181             else
1182             {
1183                 if (ctx->cur_ps_args->super.srgb_correction)
1184                     FIXME("sRGB correction on higher render targets.\n");
1185                 if (reg_maps->rt_mask > 1)
1186                     sprintf(register_name, "result.color[%u]", reg->idx[0].offset);
1187                 else
1188                     strcpy(register_name, "result.color");
1189             }
1190             break;
1191 
1192         case WINED3DSPR_RASTOUT:
1193             if (reg->idx[0].offset == 1)
1194                 sprintf(register_name, "%s", ctx->fog_output);
1195             else
1196                 sprintf(register_name, "%s", rastout_reg_names[reg->idx[0].offset]);
1197             break;
1198 
1199         case WINED3DSPR_DEPTHOUT:
1200             strcpy(register_name, "result.depth");
1201             break;
1202 
1203         case WINED3DSPR_ATTROUT:
1204         /* case WINED3DSPR_OUTPUT: */
1205             if (pshader)
1206                 sprintf(register_name, "oD[%u]", reg->idx[0].offset);
1207             else
1208                 strcpy(register_name, ctx->color_output[reg->idx[0].offset]);
1209             break;
1210 
1211         case WINED3DSPR_TEXCRDOUT:
1212             if (pshader)
1213                 sprintf(register_name, "oT[%u]", reg->idx[0].offset);
1214             else if (reg_maps->shader_version.major < 3)
1215                 strcpy(register_name, ctx->texcrd_output[reg->idx[0].offset]);
1216             else
1217                 strcpy(register_name, ctx->vs_output[reg->idx[0].offset]);
1218             break;
1219 
1220         case WINED3DSPR_LOOP:
1221             if(ctx->target_version >= NV2)
1222             {
1223                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1224                 if(pshader) sprintf(register_name, "A0.x");
1225                 else sprintf(register_name, "aL.y");
1226             }
1227             else
1228             {
1229                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1230                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1231                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1232                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1233                  * indexing
1234                  */
1235                 sprintf(register_name, "**aL_emul**");
1236             }
1237 
1238             break;
1239 
1240         case WINED3DSPR_CONSTINT:
1241             sprintf(register_name, "I%u", reg->idx[0].offset);
1242             break;
1243 
1244         case WINED3DSPR_MISCTYPE:
1245             if (!reg->idx[0].offset)
1246                 sprintf(register_name, "vpos");
1247             else if (reg->idx[0].offset == 1)
1248                 sprintf(register_name, "fragment.facing.x");
1249             else
1250                 FIXME("Unknown MISCTYPE register index %u.\n", reg->idx[0].offset);
1251             break;
1252 
1253         default:
1254             FIXME("Unhandled register type %#x[%u].\n", reg->type, reg->idx[0].offset);
1255             sprintf(register_name, "unrecognized_register[%u]", reg->idx[0].offset);
1256             break;
1257     }
1258 }
1259 
1260 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1261         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1262 {
1263     char register_name[255];
1264     char write_mask[6];
1265     BOOL is_color;
1266 
1267     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1268     strcpy(str, register_name);
1269 
1270     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1271     strcat(str, write_mask);
1272 }
1273 
1274 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1275 {
1276     switch(channel_source)
1277     {
1278         case CHANNEL_SOURCE_ZERO: return "0";
1279         case CHANNEL_SOURCE_ONE: return "1";
1280         case CHANNEL_SOURCE_X: return "x";
1281         case CHANNEL_SOURCE_Y: return "y";
1282         case CHANNEL_SOURCE_Z: return "z";
1283         case CHANNEL_SOURCE_W: return "w";
1284         default:
1285             FIXME("Unhandled channel source %#x\n", channel_source);
1286             return "undefined";
1287     }
1288 }
1289 
1290 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1291         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1292 {
1293     DWORD mask;
1294 
1295     if (is_complex_fixup(fixup))
1296     {
1297         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1298         FIXME("Complex fixup (%#x) not supported\n", complex_fixup);
1299         return;
1300     }
1301 
1302     mask = 0;
1303     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1304     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1305     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1306     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1307     mask &= dst_mask;
1308 
1309     if (mask)
1310     {
1311         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1312                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1313                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1314     }
1315 
1316     mask = 0;
1317     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1318     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1319     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1320     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1321     mask &= dst_mask;
1322 
1323     if (mask)
1324     {
1325         char reg_mask[6];
1326         char *ptr = reg_mask;
1327 
1328         if (mask != WINED3DSP_WRITEMASK_ALL)
1329         {
1330             *ptr++ = '.';
1331             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1332             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1333             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1334             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1335         }
1336         *ptr = '\0';
1337 
1338         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1339     }
1340 }
1341 
1342 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1343 {
1344     DWORD mod;
1345     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1346     if (!ins->dst_count) return "";
1347 
1348     mod = ins->dst[0].modifiers;
1349 
1350     /* Silently ignore PARTIALPRECISION if its not supported */
1351     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1352 
1353     if(mod & WINED3DSPDM_MSAMPCENTROID)
1354     {
1355         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1356         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1357     }
1358 
1359     switch(mod)
1360     {
1361         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1362             return "H_SAT";
1363 
1364         case WINED3DSPDM_SATURATE:
1365             return "_SAT";
1366 
1367         case WINED3DSPDM_PARTIALPRECISION:
1368             return "H";
1369 
1370         case 0:
1371             return "";
1372 
1373         default:
1374             FIXME("Unknown modifiers 0x%08x\n", mod);
1375             return "";
1376     }
1377 }
1378 
1379 #define TEX_PROJ        0x1
1380 #define TEX_BIAS        0x2
1381 #define TEX_LOD         0x4
1382 #define TEX_DERIV       0x10
1383 
1384 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1385         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1386 {
1387     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1388     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1389     const char *tex_type;
1390     BOOL np2_fixup = FALSE;
1391     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1392     const char *mod;
1393     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1394     const struct wined3d_shader *shader;
1395     const struct wined3d_device *device;
1396     const struct wined3d_gl_info *gl_info;
1397 
1398     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1399     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1400 
1401     switch(sampler_type) {
1402         case WINED3DSTT_1D:
1403             tex_type = "1D";
1404             break;
1405 
1406         case WINED3DSTT_2D:
1407             shader = ins->ctx->shader;
1408             device = shader->device;
1409             gl_info = &device->adapter->gl_info;
1410 
1411             if (pshader && priv->cur_ps_args->super.np2_fixup & (1 << sampler_idx)
1412                     && gl_info->supported[ARB_TEXTURE_RECTANGLE])
1413                 tex_type = "RECT";
1414             else
1415                 tex_type = "2D";
1416             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1417             {
1418                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1419                 {
1420                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1421                     else np2_fixup = TRUE;
1422                 }
1423             }
1424             break;
1425 
1426         case WINED3DSTT_VOLUME:
1427             tex_type = "3D";
1428             break;
1429 
1430         case WINED3DSTT_CUBE:
1431             tex_type = "CUBE";
1432             break;
1433 
1434         default:
1435             ERR("Unexpected texture type %d\n", sampler_type);
1436             tex_type = "";
1437     }
1438 
1439     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1440      * so don't use shader_arb_get_modifier
1441      */
1442     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1443     else mod = "";
1444 
1445     /* Fragment samplers always have indentity mapping */
1446     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1447     {
1448         sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1449     }
1450 
1451     if (flags & TEX_DERIV)
1452     {
1453         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1454         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1455         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1456                        dsx, dsy,sampler_idx, tex_type);
1457     }
1458     else if(flags & TEX_LOD)
1459     {
1460         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1461         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1462         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1463                        sampler_idx, tex_type);
1464     }
1465     else if (flags & TEX_BIAS)
1466     {
1467         /* Shouldn't be possible, but let's check for it */
1468         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1469         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1470         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1471     }
1472     else if (flags & TEX_PROJ)
1473     {
1474         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1475     }
1476     else
1477     {
1478         if (np2_fixup)
1479         {
1480             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1481             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1482                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1483 
1484             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1485         }
1486         else
1487             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1488     }
1489 
1490     if (pshader)
1491     {
1492         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1493                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_ONE),
1494                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_TWO),
1495                 priv->cur_ps_args->super.color_fixup[sampler_idx]);
1496     }
1497 }
1498 
1499 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1500         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1501 {
1502     /* Generate a line that does the input modifier computation and return the input register to use */
1503     BOOL is_color = FALSE, insert_line;
1504     char regstr[256];
1505     char swzstr[20];
1506     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1507     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1508     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
1509     const char *two = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_TWO);
1510 
1511     /* Assume a new line will be added */
1512     insert_line = TRUE;
1513 
1514     /* Get register name */
1515     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1516     shader_arb_get_swizzle(src, is_color, swzstr);
1517 
1518     switch (src->modifiers)
1519     {
1520     case WINED3DSPSM_NONE:
1521         sprintf(outregstr, "%s%s", regstr, swzstr);
1522         insert_line = FALSE;
1523         break;
1524     case WINED3DSPSM_NEG:
1525         sprintf(outregstr, "-%s%s", regstr, swzstr);
1526         insert_line = FALSE;
1527         break;
1528     case WINED3DSPSM_BIAS:
1529         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1530         break;
1531     case WINED3DSPSM_BIASNEG:
1532         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1533         break;
1534     case WINED3DSPSM_SIGN:
1535         shader_addline(buffer, "MAD T%c, %s, %s, -%s;\n", 'A' + tmpreg, regstr, two, one);
1536         break;
1537     case WINED3DSPSM_SIGNNEG:
1538         shader_addline(buffer, "MAD T%c, %s, -%s, %s;\n", 'A' + tmpreg, regstr, two, one);
1539         break;
1540     case WINED3DSPSM_COMP:
1541         shader_addline(buffer, "SUB T%c, %s, %s;\n", 'A' + tmpreg, one, regstr);
1542         break;
1543     case WINED3DSPSM_X2:
1544         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1545         break;
1546     case WINED3DSPSM_X2NEG:
1547         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1548         break;
1549     case WINED3DSPSM_DZ:
1550         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1551         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1552         break;
1553     case WINED3DSPSM_DW:
1554         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1555         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1556         break;
1557     case WINED3DSPSM_ABS:
1558         if(ctx->target_version >= NV2) {
1559             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1560             insert_line = FALSE;
1561         } else {
1562             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1563         }
1564         break;
1565     case WINED3DSPSM_ABSNEG:
1566         if(ctx->target_version >= NV2) {
1567             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1568         } else {
1569             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1570             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1571         }
1572         insert_line = FALSE;
1573         break;
1574     default:
1575         sprintf(outregstr, "%s%s", regstr, swzstr);
1576         insert_line = FALSE;
1577     }
1578 
1579     /* Return modified or original register, with swizzle */
1580     if (insert_line)
1581         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1582 }
1583 
1584 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1585 {
1586     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1587     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1588     DWORD sampler_code = dst->reg.idx[0].offset;
1589     char dst_name[50];
1590     char src_name[2][50];
1591 
1592     shader_arb_get_dst_param(ins, dst, dst_name);
1593 
1594     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1595      *
1596      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1597      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1598      * temps is done.
1599      */
1600     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1601     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1602     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1603     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1604     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1605 
1606     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1607     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1608 }
1609 
1610 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1611 {
1612     *extra_char = ' ';
1613     switch(mod)
1614     {
1615         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1616         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1617         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1618         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1619         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1620         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1621         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1622         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1623         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1624         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1625         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1626         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1627         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1628     }
1629     FIXME("Unknown modifier %u\n", mod);
1630     return mod;
1631 }
1632 
1633 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1634 {
1635     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1636     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1637     char dst_name[50];
1638     char src_name[3][50];
1639     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1640             ins->ctx->reg_maps->shader_version.minor);
1641 
1642     shader_arb_get_dst_param(ins, dst, dst_name);
1643     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1644 
1645     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue
1646             && ins->dst->write_mask != WINED3DSP_WRITEMASK_3)
1647     {
1648         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1649     }
1650     else
1651     {
1652         struct wined3d_shader_src_param src0_copy = ins->src[0];
1653         char extra_neg;
1654 
1655         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1656         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1657 
1658         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1659         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1660         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1661         shader_addline(buffer, "CMP%s %s, TA, %s, %s;\n", shader_arb_get_modifier(ins),
1662                 dst_name, src_name[1], src_name[2]);
1663     }
1664 }
1665 
1666 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1667 {
1668     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1669     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1670     char dst_name[50];
1671     char src_name[3][50];
1672 
1673     shader_arb_get_dst_param(ins, dst, dst_name);
1674 
1675     /* Generate input register names (with modifiers) */
1676     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1677     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1678     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1679 
1680     shader_addline(buffer, "CMP%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1681             dst_name, src_name[0], src_name[2], src_name[1]);
1682 }
1683 
1684 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1685  * dst = dot2(src0, src1) + src2 */
1686 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1687 {
1688     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1689     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1690     char dst_name[50];
1691     char src_name[3][50];
1692     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1693 
1694     shader_arb_get_dst_param(ins, dst, dst_name);
1695     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1696     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1697 
1698     if(ctx->target_version >= NV3)
1699     {
1700         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1701         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1702         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1703                        dst_name, src_name[0], src_name[1], src_name[2]);
1704     }
1705     else if(ctx->target_version >= NV2)
1706     {
1707         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1708          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1709          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1710          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1711          *
1712          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1713          *
1714          * .xyxy and other swizzles that we could get with this are not valid in
1715          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1716          */
1717         struct wined3d_shader_src_param tmp_param = ins->src[1];
1718         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1719         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1720 
1721         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1722 
1723         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1724                        dst_name, src_name[2], src_name[0], src_name[1]);
1725     }
1726     else
1727     {
1728         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1729         /* Emulate a DP2 with a DP3 and 0.0. Don't use the dest as temp register, it could be src[1] or src[2]
1730         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1731         */
1732         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1733         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1734         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1735         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1736     }
1737 }
1738 
1739 /* Map the opcode 1-to-1 to the GL code */
1740 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1741 {
1742     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1743     const char *instruction;
1744     char arguments[256], dst_str[50];
1745     unsigned int i;
1746     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1747 
1748     switch (ins->handler_idx)
1749     {
1750         case WINED3DSIH_ABS: instruction = "ABS"; break;
1751         case WINED3DSIH_ADD: instruction = "ADD"; break;
1752         case WINED3DSIH_CRS: instruction = "XPD"; break;
1753         case WINED3DSIH_DP3: instruction = "DP3"; break;
1754         case WINED3DSIH_DP4: instruction = "DP4"; break;
1755         case WINED3DSIH_DST: instruction = "DST"; break;
1756         case WINED3DSIH_FRC: instruction = "FRC"; break;
1757         case WINED3DSIH_LIT: instruction = "LIT"; break;
1758         case WINED3DSIH_LRP: instruction = "LRP"; break;
1759         case WINED3DSIH_MAD: instruction = "MAD"; break;
1760         case WINED3DSIH_MAX: instruction = "MAX"; break;
1761         case WINED3DSIH_MIN: instruction = "MIN"; break;
1762         case WINED3DSIH_MOV: instruction = "MOV"; break;
1763         case WINED3DSIH_MUL: instruction = "MUL"; break;
1764         case WINED3DSIH_SGE: instruction = "SGE"; break;
1765         case WINED3DSIH_SLT: instruction = "SLT"; break;
1766         case WINED3DSIH_SUB: instruction = "SUB"; break;
1767         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1768         case WINED3DSIH_DSX: instruction = "DDX"; break;
1769         default: instruction = "";
1770             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1771             break;
1772     }
1773 
1774     /* Note that shader_arb_add_dst_param() adds spaces. */
1775     arguments[0] = '\0';
1776     shader_arb_get_dst_param(ins, dst, dst_str);
1777     for (i = 0; i < ins->src_count; ++i)
1778     {
1779         char operand[100];
1780         strcat(arguments, ", ");
1781         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1782         strcat(arguments, operand);
1783     }
1784     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1785 }
1786 
1787 static void shader_hw_nop(const struct wined3d_shader_instruction *ins) {}
1788 
1789 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1790 {
1791     const struct wined3d_shader *shader = ins->ctx->shader;
1792     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1793     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1794     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1795     const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
1796     const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
1797     const char *two = arb_get_helper_value(reg_maps->shader_version.type, ARB_TWO);
1798 
1799     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1800     char src0_param[256];
1801 
1802     if (ins->handler_idx == WINED3DSIH_MOVA)
1803     {
1804         const struct arb_vshader_private *shader_data = shader->backend_data;
1805         char write_mask[6];
1806         const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1807 
1808         if(ctx->target_version >= NV2) {
1809             shader_hw_map2gl(ins);
1810             return;
1811         }
1812         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1813         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1814 
1815         /* This implements the mova formula used in GLSL. The first two instructions
1816          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1817          * in this case:
1818          * mova A0.x, 0.0
1819          *
1820          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1821          *
1822          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1823          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1824          */
1825         shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
1826         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
1827 
1828         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1829         shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
1830         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1831         if (shader_data->rel_offset)
1832         {
1833             shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
1834         }
1835         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1836 
1837         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1838     }
1839     else if (reg_maps->shader_version.major == 1
1840           && !shader_is_pshader_version(reg_maps->shader_version.type)
1841           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1842     {
1843         const struct arb_vshader_private *shader_data = shader->backend_data;
1844         src0_param[0] = '\0';
1845 
1846         if (shader_data->rel_offset && ctx->target_version == ARB)
1847         {
1848             const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1849             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1850             shader_addline(buffer, "ADD TA.x, %s, %s;\n", src0_param, offset);
1851             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1852         }
1853         else
1854         {
1855             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1856              * with more than one component. Thus replicate the first source argument over all
1857              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1858             struct wined3d_shader_src_param tmp_src = ins->src[0];
1859             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1860             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1861             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1862         }
1863     }
1864     else if (ins->dst[0].reg.type == WINED3DSPR_COLOROUT && !ins->dst[0].reg.idx[0].offset && pshader)
1865     {
1866         if (ctx->ps_post_process && shader->u.ps.color0_mov)
1867         {
1868             shader_addline(buffer, "#mov handled in srgb write or fog code\n");
1869             return;
1870         }
1871         shader_hw_map2gl(ins);
1872     }
1873     else
1874     {
1875         shader_hw_map2gl(ins);
1876     }
1877 }
1878 
1879 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1880 {
1881     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1882     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1883     char reg_dest[40];
1884 
1885     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1886      * but >= 2.0 honors it (undocumented, but tested by the d3d9 testsuite)
1887      */
1888     shader_arb_get_dst_param(ins, dst, reg_dest);
1889 
1890     if (ins->ctx->reg_maps->shader_version.major >= 2)
1891     {
1892         const char *kilsrc = "TA";
1893         BOOL is_color;
1894 
1895         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1896         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1897         {
1898             kilsrc = reg_dest;
1899         }
1900         else
1901         {
1902             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1903              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1904              * masked out components to 0(won't kill)
1905              */
1906             char x = '0', y = '0', z = '0', w = '0';
1907             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1908             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1909             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1910             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1911             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1912         }
1913         shader_addline(buffer, "KIL %s;\n", kilsrc);
1914     }
1915     else
1916     {
1917         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1918          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1919          *
1920          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1921          * or pass in any temporary register(in shader phase 2)
1922          */
1923         if (ins->ctx->reg_maps->shader_version.minor <= 3)
1924             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx[0].offset);
1925         else
1926             shader_arb_get_dst_param(ins, dst, reg_dest);
1927         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1928         shader_addline(buffer, "KIL TA;\n");
1929     }
1930 }
1931 
1932 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1933 {
1934     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1935     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1936     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1937             ins->ctx->reg_maps->shader_version.minor);
1938     struct wined3d_shader_src_param src;
1939 
1940     char reg_dest[40];
1941     char reg_coord[40];
1942     DWORD reg_sampler_code;
1943     WORD myflags = 0;
1944     BOOL swizzle_coord = FALSE;
1945 
1946     /* All versions have a destination register */
1947     shader_arb_get_dst_param(ins, dst, reg_dest);
1948 
1949     /* 1.0-1.4: Use destination register number as texture code.
1950        2.0+: Use provided sampler number as texure code. */
1951     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1952         reg_sampler_code = dst->reg.idx[0].offset;
1953     else
1954         reg_sampler_code = ins->src[1].reg.idx[0].offset;
1955 
1956     /* 1.0-1.3: Use the texcoord varying.
1957        1.4+: Use provided coordinate source register. */
1958     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1959         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1960     else {
1961         /* TEX is the only instruction that can handle DW and DZ natively */
1962         src = ins->src[0];
1963         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1964         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1965         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1966     }
1967 
1968     /* projection flag:
1969      * 1.1, 1.2, 1.3: Use WINED3D_TSS_TEXTURETRANSFORMFLAGS
1970      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1971      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1972      */
1973     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1974     {
1975         DWORD flags = 0;
1976         if (reg_sampler_code < MAX_TEXTURES)
1977             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1978         if (flags & WINED3D_PSARGS_PROJECTED)
1979         {
1980             myflags |= TEX_PROJ;
1981             if ((flags & ~WINED3D_PSARGS_PROJECTED) == WINED3D_TTFF_COUNT3)
1982                 swizzle_coord = TRUE;
1983         }
1984     }
1985     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1986     {
1987         enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
1988         if (src_mod == WINED3DSPSM_DZ)
1989         {
1990             swizzle_coord = TRUE;
1991             myflags |= TEX_PROJ;
1992         } else if(src_mod == WINED3DSPSM_DW) {
1993             myflags |= TEX_PROJ;
1994         }
1995     } else {
1996         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1997         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1998     }
1999 
2000     if (swizzle_coord)
2001     {
2002         /* TXP cannot handle DZ natively, so move the z coordinate to .w.
2003          * reg_coord is a read-only varying register, so we need a temp reg */
2004         shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
2005         strcpy(reg_coord, "TA");
2006     }
2007 
2008     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
2009 }
2010 
2011 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
2012 {
2013     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2014     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2015     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2016             ins->ctx->reg_maps->shader_version.minor);
2017     char dst_str[50];
2018 
2019     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2020     {
2021         DWORD reg = dst->reg.idx[0].offset;
2022 
2023         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2024         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
2025     } else {
2026         char reg_src[40];
2027 
2028         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
2029         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2030         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
2031     }
2032 }
2033 
2034 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
2035 {
2036      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2037      DWORD flags = 0;
2038 
2039      DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2040      char dst_str[50];
2041      char src_str[50];
2042 
2043      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
2044      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2045      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2046      /* Move .x first in case src_str is "TA" */
2047      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
2048      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
2049      if (reg1 < MAX_TEXTURES)
2050      {
2051          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2052          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
2053      }
2054      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2055 }
2056 
2057 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
2058 {
2059      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2060 
2061      DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2062      char dst_str[50];
2063      char src_str[50];
2064 
2065      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2066      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2067      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2068      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2069      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2070      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2071 }
2072 
2073 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2074 {
2075     DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2076     char dst_str[50];
2077     char src_str[50];
2078 
2079     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2080     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2081     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2082     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2083 }
2084 
2085 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2086 {
2087     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2088     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2089     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2090     char reg_coord[40], dst_reg[50], src_reg[50];
2091     DWORD reg_dest_code;
2092 
2093     /* All versions have a destination register. The Tx where the texture coordinates come
2094      * from is the varying incarnation of the texture register
2095      */
2096     reg_dest_code = dst->reg.idx[0].offset;
2097     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2098     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2099     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2100 
2101     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2102      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2103      *
2104      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2105      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2106      *
2107      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2108      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2109      * extension.
2110      */
2111     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2112     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2113     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2114     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2115 
2116     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2117      * so we can't let the GL handle this.
2118      */
2119     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2120             & WINED3D_PSARGS_PROJECTED)
2121     {
2122         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2123         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2124         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2125     } else {
2126         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2127     }
2128 
2129     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2130 
2131     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2132     {
2133         /* No src swizzles are allowed, so this is ok */
2134         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2135                        src_reg, reg_dest_code, reg_dest_code);
2136         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2137     }
2138 }
2139 
2140 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2141 {
2142     DWORD reg = ins->dst[0].reg.idx[0].offset;
2143     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2144     char src0_name[50], dst_name[50];
2145     BOOL is_color;
2146     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2147 
2148     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2149     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2150      * T<reg+1> register. Use this register to store the calculated vector
2151      */
2152     tmp_reg.idx[0].offset = reg + 1;
2153     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2154     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2155 }
2156 
2157 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2158 {
2159     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2160     DWORD flags;
2161     DWORD reg = ins->dst[0].reg.idx[0].offset;
2162     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2163     char dst_str[50];
2164     char src0_name[50];
2165     char dst_reg[50];
2166     BOOL is_color;
2167 
2168     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2169     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2170 
2171     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2172     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2173     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2174     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2175     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2176 }
2177 
2178 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2179 {
2180     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2181     DWORD reg = ins->dst[0].reg.idx[0].offset;
2182     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2183     char src0_name[50], dst_name[50];
2184     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2185     BOOL is_color;
2186 
2187     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2188      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2189      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2190      */
2191     tmp_reg.idx[0].offset = reg + 2 - tex_mx->current_row;
2192     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2193 
2194     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2195     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2196                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2197     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2198 }
2199 
2200 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2201 {
2202     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2203     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2204     DWORD flags;
2205     DWORD reg = ins->dst[0].reg.idx[0].offset;
2206     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2207     char dst_str[50];
2208     char src0_name[50], dst_name[50];
2209     BOOL is_color;
2210 
2211     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2212     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2213     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2214 
2215     /* Sample the texture using the calculated coordinates */
2216     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2217     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2218     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2219     tex_mx->current_row = 0;
2220 }
2221 
2222 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2223 {
2224     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2225     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2226     DWORD flags;
2227     DWORD reg = ins->dst[0].reg.idx[0].offset;
2228     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2229     char dst_str[50];
2230     char src0_name[50];
2231     char dst_reg[50];
2232     BOOL is_color;
2233 
2234     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2235      * components for temporary data storage
2236      */
2237     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2238     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2239     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2240 
2241     /* Construct the eye-ray vector from w coordinates */
2242     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2243     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2244     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2245 
2246     /* Calculate reflection vector
2247      */
2248     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2249     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2250     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2251     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2252     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2253     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2254     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2255 
2256     /* Sample the texture using the calculated coordinates */
2257     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2258     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2259     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2260     tex_mx->current_row = 0;
2261 }
2262 
2263 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2264 {
2265     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2266     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2267     DWORD flags;
2268     DWORD reg = ins->dst[0].reg.idx[0].offset;
2269     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2270     char dst_str[50];
2271     char src0_name[50];
2272     char src1_name[50];
2273     char dst_reg[50];
2274     BOOL is_color;
2275 
2276     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2277     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2278     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2279     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2280     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2281 
2282     /* Calculate reflection vector.
2283      *
2284      *                   dot(N, E)
2285      * dst_reg.xyz = 2 * --------- * N - E
2286      *                   dot(N, N)
2287      *
2288      * Which normalizes the normal vector
2289      */
2290     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2291     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2292     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2293     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2294     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2295     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2296 
2297     /* Sample the texture using the calculated coordinates */
2298     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2299     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2300     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2301     tex_mx->current_row = 0;
2302 }
2303 
2304 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2305 {
2306     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2307     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2308     char dst_name[50];
2309     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2310     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2311 
2312     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2313      * which is essentially an input, is the destination register because it is the first
2314      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2315      * here(writemasks/swizzles are not valid on texdepth)
2316      */
2317     shader_arb_get_dst_param(ins, dst, dst_name);
2318 
2319     /* According to the msdn, the source register(must be r5) is unusable after
2320      * the texdepth instruction, so we're free to modify it
2321      */
2322     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2323 
2324     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2325      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2326      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2327      */
2328     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2329     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2330     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2331     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2332 }
2333 
2334 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2335  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2336  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2337 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2338 {
2339     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2340     DWORD sampler_idx = ins->dst[0].reg.idx[0].offset;
2341     char src0[50];
2342     char dst_str[50];
2343 
2344     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2345     shader_addline(buffer, "MOV TB, 0.0;\n");
2346     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2347 
2348     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2349     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2350 }
2351 
2352 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2353  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2354 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2355 {
2356     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2357     char src0[50];
2358     char dst_str[50];
2359     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2360 
2361     /* Handle output register */
2362     shader_arb_get_dst_param(ins, dst, dst_str);
2363     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2364     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx[0].offset, src0);
2365 }
2366 
2367 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2368  * Perform the 3rd row of a 3x3 matrix multiply */
2369 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2370 {
2371     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2372     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2373     char dst_str[50], dst_name[50];
2374     char src0[50];
2375     BOOL is_color;
2376 
2377     shader_arb_get_dst_param(ins, dst, dst_str);
2378     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2379     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2380     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx[0].offset, src0);
2381     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2382 }
2383 
2384 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2385  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2386  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2387  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2388  */
2389 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2390 {
2391     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2392     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2393     char src0[50], dst_name[50];
2394     BOOL is_color;
2395     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2396     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2397 
2398     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2399     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2400     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx[0].offset, src0);
2401 
2402     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2403      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2404      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2405      */
2406     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2407     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2408     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2409     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2410 }
2411 
2412 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2413     Vertex/Pixel shaders to ARB_vertex_program codes */
2414 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2415 {
2416     int i;
2417     int nComponents = 0;
2418     struct wined3d_shader_dst_param tmp_dst = {{0}};
2419     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2420     struct wined3d_shader_instruction tmp_ins;
2421 
2422     memset(&tmp_ins, 0, sizeof(tmp_ins));
2423 
2424     /* Set constants for the temporary argument */
2425     tmp_ins.ctx = ins->ctx;
2426     tmp_ins.dst_count = 1;
2427     tmp_ins.dst = &tmp_dst;
2428     tmp_ins.src_count = 2;
2429     tmp_ins.src = tmp_src;
2430 
2431     switch(ins->handler_idx)
2432     {
2433         case WINED3DSIH_M4x4:
2434             nComponents = 4;
2435             tmp_ins.handler_idx = WINED3DSIH_DP4;
2436             break;
2437         case WINED3DSIH_M4x3:
2438             nComponents = 3;
2439             tmp_ins.handler_idx = WINED3DSIH_DP4;
2440             break;
2441         case WINED3DSIH_M3x4:
2442             nComponents = 4;
2443             tmp_ins.handler_idx = WINED3DSIH_DP3;
2444             break;
2445         case WINED3DSIH_M3x3:
2446             nComponents = 3;
2447             tmp_ins.handler_idx = WINED3DSIH_DP3;
2448             break;
2449         case WINED3DSIH_M3x2:
2450             nComponents = 2;
2451             tmp_ins.handler_idx = WINED3DSIH_DP3;
2452             break;
2453         default:
2454             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2455             break;
2456     }
2457 
2458     tmp_dst = ins->dst[0];
2459     tmp_src[0] = ins->src[0];
2460     tmp_src[1] = ins->src[1];
2461     for (i = 0; i < nComponents; ++i)
2462     {
2463         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2464         shader_hw_map2gl(&tmp_ins);
2465         ++tmp_src[1].reg.idx[0].offset;
2466     }
2467 }
2468 
2469 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2470 {
2471     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2472 
2473     char dst[50];
2474     char src[50];
2475 
2476     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2477     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2478     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2479     {
2480         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2481          * .w is used
2482          */
2483         strcat(src, ".w");
2484     }
2485 
2486     shader_addline(buffer, "RCP%s %s, %s;\n", shader_arb_get_modifier(ins), dst, src);
2487 }
2488 
2489 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2490 {
2491     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2492     const char *instruction;
2493 
2494     char dst[50];
2495     char src[50];
2496 
2497     switch(ins->handler_idx)
2498     {
2499         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2500         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2501         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2502         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2503         default: instruction = "";
2504             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2505             break;
2506     }
2507 
2508     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2509     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2510     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2511     {
2512         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2513          * .w is used
2514          */
2515         strcat(src, ".w");
2516     }
2517 
2518     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2519 }
2520 
2521 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2522 {
2523     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2524     char dst_name[50];
2525     char src_name[50];
2526     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2527     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2528     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2529 
2530     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2531     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2532 
2533     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2534      * otherwise NRM or RSQ would return NaN */
2535     if(pshader && priv->target_version >= NV3)
2536     {
2537         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2538          *
2539          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2540          */
2541         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2542         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2543         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2544     }
2545     else if(priv->target_version >= NV2)
2546     {
2547         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2548         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2549         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2550                        src_name);
2551     }
2552     else
2553     {
2554         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2555 
2556         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2557         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2558          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2559          */
2560         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2561         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2562 
2563         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2564         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2565         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2566                     src_name);
2567     }
2568 }
2569 
2570 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2571 {
2572     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2573     char dst_name[50];
2574     char src_name[3][50];
2575 
2576     /* ARB_fragment_program has a convenient LRP instruction */
2577     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2578         shader_hw_map2gl(ins);
2579         return;
2580     }
2581 
2582     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2583     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2584     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2585     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2586 
2587     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2588     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2589                    dst_name, src_name[0], src_name[2]);
2590 }
2591 
2592 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2593 {
2594     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2595      * must contain fixed constants. So we need a separate function to filter those constants and
2596      * can't use map2gl
2597      */
2598     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2599     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2600     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2601     char dst_name[50];
2602     char src_name0[50], src_name1[50], src_name2[50];
2603     BOOL is_color;
2604 
2605     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2606     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2607         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2608         /* No modifiers are supported on SCS */
2609         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2610 
2611         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2612         {
2613             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2614             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2615         }
2616     } else if(priv->target_version >= NV2) {
2617         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2618 
2619         /* Sincos writemask must be .x, .y or .xy */
2620         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2621             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2622         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2623             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2624     } else {
2625         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2626          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2627          *
2628          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2629          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2630          *
2631          * The constants we get are:
2632          *
2633          *  +1   +1,     -1     -1     +1      +1      -1       -1
2634          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2635          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2636          *
2637          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2638          *
2639          * (x/2)^2 = x^2 / 4
2640          * (x/2)^3 = x^3 / 8
2641          * (x/2)^4 = x^4 / 16
2642          * (x/2)^5 = x^5 / 32
2643          * etc
2644          *
2645          * To get the final result:
2646          * sin(x) = 2 * sin(x/2) * cos(x/2)
2647          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2648          * (from sin(x+y) and cos(x+y) rules)
2649          *
2650          * As per MSDN, dst.z is undefined after the operation, and so is
2651          * dst.x and dst.y if they're masked out by the writemask. Ie
2652          * sincos dst.y, src1, c0, c1
2653          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2654          * vsa.exe also stops with an error if the dest register is the same register as the source
2655          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2656          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2657          */
2658         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2659         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2660         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2661 
2662         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2663         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2664         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2665         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2666         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2667         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2668 
2669         /* sin(x/2)
2670          *
2671          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2672          * properly merge that with MULs in the code above?
2673          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2674          * we can merge the sine and cosine MAD rows to calculate them together.
2675          */
2676         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2677         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2678         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2679         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2680 
2681         /* cos(x/2) */
2682         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2683         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2684         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2685 
2686         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2687             /* cos x */
2688             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2689             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2690         }
2691         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2692             /* sin x */
2693             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2694             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2695         }
2696     }
2697 }
2698 
2699 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2700 {
2701     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2702     char dst_name[50];
2703     char src_name[50];
2704     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2705 
2706     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2707     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2708 
2709     /* SGN is only valid in vertex shaders */
2710     if(ctx->target_version >= NV2) {
2711         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2712         return;
2713     }
2714 
2715     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2716      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2717      */
2718     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2719         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2720     } else {
2721         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2722          * Then use TA, and calculate the final result
2723          *
2724          * Not reading from TA? Store the first result in TA to avoid overwriting the
2725          * destination if src reg = dst reg
2726          */
2727         if(strstr(src_name, "TA"))
2728         {
2729             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2730             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2731             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2732         }
2733         else
2734         {
2735             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2736             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2737             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2738         }
2739     }
2740 }
2741 
2742 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2743 {
2744     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2745     char src[50];
2746     char dst[50];
2747     char dst_name[50];
2748     BOOL is_color;
2749 
2750     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2751     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2752     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2753 
2754     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2755     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2756 }
2757 
2758 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2759 {
2760     *need_abs = FALSE;
2761 
2762     switch(mod)
2763     {
2764         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2765         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2766         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2767         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2768         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2769         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2770         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2771         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2772         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2773         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2774         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2775         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2776         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2777     }
2778     FIXME("Unknown modifier %u\n", mod);
2779     return mod;
2780 }
2781 
2782 static void shader_hw_log(const struct wined3d_shader_instruction *ins)
2783 {
2784     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2785     char src0[50], dst[50];
2786     struct wined3d_shader_src_param src0_copy = ins->src[0];
2787     BOOL need_abs = FALSE;
2788     const char *instr;
2789 
2790     switch(ins->handler_idx)
2791     {
2792         case WINED3DSIH_LOG:  instr = "LG2"; break;
2793         case WINED3DSIH_LOGP: instr = "LOG"; break;
2794         default:
2795             ERR("Unexpected instruction %d\n", ins->handler_idx);
2796             return;
2797     }
2798 
2799     /* LOG and LOGP operate on the absolute value of the input */
2800     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2801 
2802     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2803     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2804 
2805     if(need_abs)
2806     {
2807         shader_addline(buffer, "ABS TA, %s;\n", src0);
2808         shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2809     }
2810     else
2811     {
2812         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2813     }
2814 }
2815 
2816 static void shader_hw_pow(const struct wined3d_shader_instruction *ins)
2817 {
2818     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2819     char src0[50], src1[50], dst[50];
2820     struct wined3d_shader_src_param src0_copy = ins->src[0];
2821     BOOL need_abs = FALSE;
2822     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2823     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2824 
2825     /* POW operates on the absolute value of the input */
2826     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2827 
2828     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2829     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2830     shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2831 
2832     if (need_abs)
2833         shader_addline(buffer, "ABS TA.x, %s;\n", src0);
2834     else
2835         shader_addline(buffer, "MOV TA.x, %s;\n", src0);
2836 
2837     if (priv->target_version >= NV2)
2838     {
2839         shader_addline(buffer, "MOVC TA.y, %s;\n", src1);
2840         shader_addline(buffer, "POW%s %s, TA.x, TA.y;\n", shader_arb_get_modifier(ins), dst);
2841         shader_addline(buffer, "MOV %s (EQ.y), %s;\n", dst, one);
2842     }
2843     else
2844     {
2845         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2846         const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2847 
2848         shader_addline(buffer, "ABS TA.y, %s;\n", src1);
2849         shader_addline(buffer, "SGE TA.y, -TA.y, %s;\n", zero);
2850         /* Possibly add flt_eps to avoid getting float special values */
2851         shader_addline(buffer, "MAD TA.z, TA.y, %s, %s;\n", flt_eps, src1);
2852         shader_addline(buffer, "POW%s TA.x, TA.x, TA.z;\n", shader_arb_get_modifier(ins));
2853         shader_addline(buffer, "MAD TA.x, -TA.x, TA.y, TA.x;\n");
2854         shader_addline(buffer, "MAD %s, TA.y, %s, TA.x;\n", dst, one);
2855     }
2856 }
2857 
2858 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2859 {
2860     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2861     char src_name[50];
2862     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2863 
2864     /* src0 is aL */
2865     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2866 
2867     if(vshader)
2868     {
2869         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2870         struct list *e = list_head(&priv->control_frames);
2871         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2872 
2873         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2874         /* The constant loader makes sure to load -1 into iX.w */
2875         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2876         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2877         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2878     }
2879     else
2880     {
2881         shader_addline(buffer, "LOOP %s;\n", src_name);
2882     }
2883 }
2884 
2885 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2886 {
2887     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2888     char src_name[50];
2889     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2890 
2891     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2892 
2893     /* The constant loader makes sure to load -1 into iX.w */
2894     if(vshader)
2895     {
2896         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2897         struct list *e = list_head(&priv->control_frames);
2898         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2899 
2900         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2901 
2902         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2903         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2904         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2905     }
2906     else
2907     {
2908         shader_addline(buffer, "REP %s;\n", src_name);
2909     }
2910 }
2911 
2912 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2913 {
2914     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2915     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2916 
2917     if(vshader)
2918     {
2919         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2920         struct list *e = list_head(&priv->control_frames);
2921         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2922 
2923         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2924         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2925         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2926 
2927         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2928     }
2929     else
2930     {
2931         shader_addline(buffer, "ENDLOOP;\n");
2932     }
2933 }
2934 
2935 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2936 {
2937     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2938     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2939 
2940     if(vshader)
2941     {
2942         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2943         struct list *e = list_head(&priv->control_frames);
2944         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2945 
2946         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2947         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2948         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2949 
2950         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2951     }
2952     else
2953     {
2954         shader_addline(buffer, "ENDREP;\n");
2955     }
2956 }
2957 
2958 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2959 {
2960     struct control_frame *control_frame;
2961 
2962     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2963     {
2964         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2965     }
2966     ERR("Could not find loop for break\n");
2967     return NULL;
2968 }
2969 
2970 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2971 {
2972     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2973     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2974     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2975 
2976     if(vshader)
2977     {
2978         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2979     }
2980     else
2981     {
2982         shader_addline(buffer, "BRK;\n");
2983     }
2984 }
2985 
2986 static const char *get_compare(enum wined3d_shader_rel_op op)
2987 {
2988     switch (op)
2989     {
2990         case WINED3D_SHADER_REL_OP_GT: return "GT";
2991         case WINED3D_SHADER_REL_OP_EQ: return "EQ";
2992         case WINED3D_SHADER_REL_OP_GE: return "GE";
2993         case WINED3D_SHADER_REL_OP_LT: return "LT";
2994         case WINED3D_SHADER_REL_OP_NE: return "NE";
2995         case WINED3D_SHADER_REL_OP_LE: return "LE";
2996         default:
2997             FIXME("Unrecognized operator %#x.\n", op);
2998             return "(\?\?)";
2999     }
3000 }
3001 
3002 static enum wined3d_shader_rel_op invert_compare(enum wined3d_shader_rel_op op)
3003 {
3004     switch (op)
3005     {
3006         case WINED3D_SHADER_REL_OP_GT: return WINED3D_SHADER_REL_OP_LE;
3007         case WINED3D_SHADER_REL_OP_EQ: return WINED3D_SHADER_REL_OP_NE;
3008         case WINED3D_SHADER_REL_OP_GE: return WINED3D_SHADER_REL_OP_LT;
3009         case WINED3D_SHADER_REL_OP_LT: return WINED3D_SHADER_REL_OP_GE;
3010         case WINED3D_SHADER_REL_OP_NE: return WINED3D_SHADER_REL_OP_EQ;
3011         case WINED3D_SHADER_REL_OP_LE: return WINED3D_SHADER_REL_OP_GT;
3012         default:
3013             FIXME("Unrecognized operator %#x.\n", op);
3014             return -1;
3015     }
3016 }
3017 
3018 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
3019 {
3020     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3021     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3022     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
3023     char src_name0[50];
3024     char src_name1[50];
3025     const char *comp = get_compare(ins->flags);
3026 
3027     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3028     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3029 
3030     if(vshader)
3031     {
3032         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
3033          * away the subtraction result
3034          */
3035         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3036         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
3037     }
3038     else
3039     {
3040         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3041         shader_addline(buffer, "BRK (%s.x);\n", comp);
3042     }
3043 }
3044 
3045 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
3046 {
3047     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3048     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3049     struct list *e = list_head(&priv->control_frames);
3050     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3051     const char *comp;
3052     char src_name0[50];
3053     char src_name1[50];
3054     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3055 
3056     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3057     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3058 
3059     if(vshader)
3060     {
3061         /* Invert the flag. We jump to the else label if the condition is NOT true */
3062         comp = get_compare(invert_compare(ins->flags));
3063         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3064         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
3065     }
3066     else
3067     {
3068         comp = get_compare(ins->flags);
3069         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3070         shader_addline(buffer, "IF %s.x;\n", comp);
3071     }
3072 }
3073 
3074 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3075 {
3076     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3077     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3078     struct list *e = list_head(&priv->control_frames);
3079     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3080     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3081 
3082     if(vshader)
3083     {
3084         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3085         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3086         control_frame->had_else = TRUE;
3087     }
3088     else
3089     {
3090         shader_addline(buffer, "ELSE;\n");
3091     }
3092 }
3093 
3094 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3095 {
3096     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3097     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3098     struct list *e = list_head(&priv->control_frames);
3099     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3100     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3101 
3102     if(vshader)
3103     {
3104         if(control_frame->had_else)
3105         {
3106             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3107         }
3108         else
3109         {
3110             shader_addline(buffer, "#No else branch. else is endif\n");
3111             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3112         }
3113     }
3114     else
3115     {
3116         shader_addline(buffer, "ENDIF;\n");
3117     }
3118 }
3119 
3120 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3121 {
3122     DWORD sampler_idx = ins->src[1].reg.idx[0].offset;
3123     char reg_dest[40];
3124     char reg_src[3][40];
3125     WORD flags = TEX_DERIV;
3126 
3127     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3128     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3129     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3130     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3131 
3132     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3133     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3134 
3135     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3136 }
3137 
3138 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3139 {
3140     DWORD sampler_idx = ins->src[1].reg.idx[0].offset;
3141     char reg_dest[40];
3142     char reg_coord[40];
3143     WORD flags = TEX_LOD;
3144 
3145     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3146     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3147 
3148     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3149     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3150 
3151     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3152 }
3153 
3154 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3155 {
3156     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3157     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3158 
3159     priv->in_main_func = FALSE;
3160     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3161      * subroutine, don't generate a label that will make GL complain
3162      */
3163     if(priv->target_version == ARB) return;
3164 
3165     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx[0].offset);
3166 }
3167 
3168 static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
3169         const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
3170         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
3171         struct wined3d_shader_buffer *buffer)
3172 {
3173     unsigned int i;
3174 
3175     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3176      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3177      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3178      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3179      */
3180     if (args->super.fog_src == VS_FOG_Z)
3181     {
3182         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3183     }
3184     else
3185     {
3186         if (!reg_maps->fog)
3187         {
3188             /* posFixup.x is always 1.0, so we can safely use it */
3189             shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3190         }
3191         else
3192         {
3193             /* Clamp fogcoord */
3194             const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
3195             const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
3196 
3197             shader_addline(buffer, "MIN TMP_FOGCOORD.x, TMP_FOGCOORD.x, %s;\n", one);
3198             shader_addline(buffer, "MAX result.fogcoord.x, TMP_FOGCOORD.x, %s;\n", zero);
3199         }
3200     }
3201 
3202     /* Clipplanes are always stored without y inversion */
3203     if (use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3204     {
3205         if (args->super.clip_enabled)
3206         {
3207             for (i = 0; i < priv_ctx->vs_clipplanes; i++)
3208             {
3209                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3210             }
3211         }
3212     }
3213     else if (args->clip.boolclip.clip_texcoord)
3214     {
3215         unsigned int cur_clip = 0;
3216         char component[4] = {'x', 'y', 'z', 'w'};
3217         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3218 
3219         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3220         {
3221             if (args->clip.boolclip.clipplane_mask & (1 << i))
3222             {
3223                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3224                                component[cur_clip++], i);
3225             }
3226         }
3227         switch (cur_clip)
3228         {
3229             case 0:
3230                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3231                 break;
3232             case 1:
3233                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3234                 break;
3235             case 2:
3236                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3237                 break;
3238             case 3:
3239                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3240                 break;
3241         }
3242         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3243                        args->clip.boolclip.clip_texcoord - 1);
3244     }
3245 
3246     /* Write the final position.
3247      *
3248      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3249      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3250      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3251      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3252      */
3253     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3254     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3255     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3256 
3257     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3258      * and the glsl equivalent
3259      */
3260     if (need_helper_const(shader_data, reg_maps, gl_info))
3261     {
3262         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3263         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3264     }
3265     else
3266     {
3267         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3268         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3269     }
3270 
3271     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3272 
3273     priv_ctx->footer_written = TRUE;
3274 }
3275 
3276 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3277 {
3278     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3279     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3280     const struct wined3d_shader *shader = ins->ctx->shader;
3281     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3282 
3283     if(priv->target_version == ARB) return;
3284 
3285     if(vshader)
3286     {
3287         if (priv->in_main_func) vshader_add_footer(priv, shader->backend_data,
3288                 priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
3289     }
3290 
3291     shader_addline(buffer, "RET;\n");
3292 }
3293 
3294 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3295 {
3296     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3297     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx[0].offset);
3298 }
3299 
3300 /* Context activation is done by the caller. */
3301 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3302 {
3303     GLuint program_id = 0;
3304     GLint pos;
3305 
3306     const char *blt_vprogram =
3307         "!!ARBvp1.0\n"
3308         "PARAM c[1] = { { 1, 0.5 } };\n"
3309         "MOV result.position, vertex.position;\n"
3310         "MOV result.color, c[0].x;\n"
3311         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3312         "END\n";
3313 
3314     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3315     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3316     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3317             strlen(blt_vprogram), blt_vprogram));
3318     checkGLcall("glProgramStringARB()");
3319 
3320     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3321     if (pos != -1)
3322     {
3323         FIXME("Vertex program error at position %d: %s\n\n", pos,
3324             debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3325         shader_arb_dump_program_source(blt_vprogram);
3326     }
3327     else
3328     {
3329         GLint native;
3330 
3331         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3332         checkGLcall("glGetProgramivARB()");
3333         if (!native) WARN("Program exceeds native resource limits.\n");
3334     }
3335 
3336     return program_id;
3337 }
3338 
3339 /* Context activation is done by the caller. */
3340 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3341         enum tex_types tex_type, BOOL masked)
3342 {
3343     GLuint program_id = 0;
3344     const char *fprogram;
3345     GLint pos;
3346 
3347     static const char * const blt_fprograms_full[tex_type_count] =
3348     {
3349         /* tex_1d */
3350         NULL,
3351         /* tex_2d */
3352         "!!ARBfp1.0\n"
3353         "TEMP R0;\n"
3354         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3355         "MOV result.depth.z, R0.x;\n"
3356         "END\n",
3357         /* tex_3d */
3358         NULL,
3359         /* tex_cube */
3360         "!!ARBfp1.0\n"
3361         "TEMP R0;\n"
3362         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3363         "MOV result.depth.z, R0.x;\n"
3364         "END\n",
3365         /* tex_rect */
3366         "!!ARBfp1.0\n"
3367         "TEMP R0;\n"
3368         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3369         "MOV result.depth.z, R0.x;\n"
3370         "END\n",
3371     };
3372 
3373     static const char * const blt_fprograms_masked[tex_type_count] =
3374     {
3375         /* tex_1d */
3376         NULL,
3377         /* tex_2d */
3378         "!!ARBfp1.0\n"
3379         "PARAM mask = program.local[0];\n"
3380         "TEMP R0;\n"
3381         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3382         "MUL R0.x, R0.x, R0.y;\n"
3383         "KIL -R0.x;\n"
3384         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3385         "MOV result.depth.z, R0.x;\n"
3386         "END\n",
3387         /* tex_3d */
3388         NULL,
3389         /* tex_cube */
3390         "!!ARBfp1.0\n"
3391         "PARAM mask = program.local[0];\n"
3392         "TEMP R0;\n"
3393         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3394         "MUL R0.x, R0.x, R0.y;\n"
3395         "KIL -R0.x;\n"
3396         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3397         "MOV result.depth.z, R0.x;\n"
3398         "END\n",
3399         /* tex_rect */
3400         "!!ARBfp1.0\n"
3401         "PARAM mask = program.local[0];\n"
3402         "TEMP R0;\n"
3403         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3404         "MUL R0.x, R0.x, R0.y;\n"
3405         "KIL -R0.x;\n"
3406         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3407         "MOV result.depth.z, R0.x;\n"
3408         "END\n",
3409     };
3410 
3411     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3412     if (!fprogram)
3413     {
3414         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3415         tex_type = tex_2d;
3416         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3417     }
3418 
3419     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3420     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3421     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3422     checkGLcall("glProgramStringARB()");
3423 
3424     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3425     if (pos != -1)
3426     {
3427         FIXME("Fragment program error at position %d: %s\n\n", pos,
3428             debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3429         shader_arb_dump_program_source(fprogram);
3430     }
3431     else
3432     {
3433         GLint native;
3434 
3435         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3436         checkGLcall("glGetProgramivARB()");
3437         if (!native) WARN("Program exceeds native resource limits.\n");
3438     }
3439 
3440     return program_id;
3441 }
3442 
3443 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3444         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3445 {
3446     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3447 
3448     if(condcode)
3449     {
3450         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3451         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.x;\n", tmp1, fragcolor);
3452         /* Calculate the > 0.0031308 case */
3453         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts0.x;\n", fragcolor, fragcolor);
3454         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts0.x;\n", fragcolor, fragcolor);
3455         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts0.x;\n", fragcolor, fragcolor);
3456         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts0.y;\n", fragcolor, fragcolor);
3457         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts0.z;\n", fragcolor, fragcolor);
3458         /* Calculate the < case */
3459         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts0.w, %s;\n", fragcolor, fragcolor);
3460     }
3461     else
3462     {
3463         /* Calculate the > 0.0031308 case */
3464         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts0.x;\n", tmp1, fragcolor);
3465         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts0.x;\n", tmp1, fragcolor);
3466         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts0.x;\n", tmp1, fragcolor);
3467         shader_addline(buffer, "MUL %s, %s, srgb_consts0.y;\n", tmp1, tmp1);
3468         shader_addline(buffer, "SUB %s, %s, srgb_consts0.z;\n", tmp1, tmp1);
3469         /* Calculate the < case */
3470         shader_addline(buffer, "MUL %s, srgb_consts0.w, %s;\n", tmp2, fragcolor);
3471         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3472         shader_addline(buffer, "SLT %s, srgb_consts1.x, %s;\n", tmp3, fragcolor);
3473         shader_addline(buffer, "SGE %s, srgb_consts1.x, %s;\n", tmp4, fragcolor);
3474         /* Store the components > 0.0031308 in the destination */
3475         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3476         /* Add the components that are < 0.0031308 */
3477         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3478         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3479         * result.color writes(.rgb first, then .a), or handle overwriting already written
3480         * components. The assembler uses a temporary register in this case, which is usually
3481         * not allocated from one of our registers that were used earlier.
3482         */
3483     }
3484     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3485 }
3486 
3487 static const DWORD *find_loop_control_values(const struct wined3d_shader *shader, DWORD idx)
3488 {
3489     const struct wined3d_shader_lconst *constant;
3490 
3491     LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
3492     {
3493         if (constant->idx == idx)
3494         {
3495             return constant->value;
3496         }
3497     }
3498     return NULL;
3499 }
3500 
3501 static void init_ps_input(const struct wined3d_shader *shader,
3502         const struct arb_ps_compile_args *args, struct shader_arb_ctx_priv *priv)
3503 {
3504     static const char * const texcoords[8] =
3505     {
3506         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3507         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3508     };
3509     unsigned int i;
3510     const struct wined3d_shader_signature_element *sig = shader->input_signature;
3511     const char *semantic_name;
3512     DWORD semantic_idx;
3513 
3514     switch(args->super.vp_mode)
3515     {
3516         case pretransformed:
3517         case fixedfunction:
3518             /* The pixelshader has to collect the varyings on its own. In any case properly load
3519              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3520              * other attribs to 0.0.
3521              *
3522              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3523              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3524              * load the texcoord attrib pointers to match the pixel shader signature
3525              */
3526             for(i = 0; i < MAX_REG_INPUT; i++)
3527             {
3528                 semantic_name = sig[i].semantic_name;
3529                 semantic_idx = sig[i].semantic_idx;
3530                 if (!semantic_name) continue;
3531 
3532                 if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
3533                 {
3534                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3535                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3536                     else priv->ps_input[i] = "0.0";
3537                 }
3538                 else if(args->super.vp_mode == fixedfunction)
3539                 {
3540                     priv->ps_input[i] = "0.0";
3541                 }
3542                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3543                 {
3544                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3545                     else priv->ps_input[i] = "0.0";
3546                 }
3547                 else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
3548                 {
3549                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3550                     else priv->ps_input[i] = "0.0";
3551                 }
3552                 else
3553                 {
3554                     priv->ps_input[i] = "0.0";
3555                 }
3556 
3557                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3558             }
3559             break;
3560 
3561         case vertexshader:
3562             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3563              * fragment.color
3564              */
3565             for(i = 0; i < 8; i++)
3566             {
3567                 priv->ps_input[i] = texcoords[i];
3568             }
3569             priv->ps_input[8] = "fragment.color.primary";
3570             priv->ps_input[9] = "fragment.color.secondary";
3571             break;
3572     }
3573 }
3574 
3575 static void arbfp_add_linear_fog(struct wined3d_shader_buffer *buffer,
3576         const char *fragcolor, const char *tmp)
3577 {
3578     shader_addline(buffer, "SUB %s.x, state.fog.params.z, fragment.fogcoord.x;\n", tmp);
3579     shader_addline(buffer, "MUL_SAT %s.x, %s.x, state.fog.params.w;\n", tmp, tmp);
3580     shader_addline(buffer, "LRP %s.rgb, %s.x, %s, state.fog.color;\n", fragcolor, tmp, fragcolor);
3581 }
3582 
3583 /* Context activation is done by the caller. */
3584 static GLuint shader_arb_generate_pshader(const struct wined3d_shader *shader,
3585         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
3586         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3587 {
3588     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3589     const DWORD *function = shader->function;
3590     GLuint retval;
3591     char fragcolor[16];
3592     DWORD next_local = 0;
3593     struct shader_arb_ctx_priv priv_ctx;
3594     BOOL dcl_td = FALSE;
3595     BOOL want_nv_prog = FALSE;
3596     struct arb_pshader_private *shader_priv = shader->backend_data;
3597     GLint errPos;
3598     DWORD map;
3599     BOOL custom_linear_fog = FALSE;
3600 
3601     char srgbtmp[4][4];
3602     char ftoa_tmp[17];
3603     unsigned int i, found = 0;
3604 
3605     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3606     {
3607         if (!(map & 1)
3608                 || (shader->u.ps.color0_mov && i == shader->u.ps.color0_reg)
3609                 || (reg_maps->shader_version.major < 2 && !i))
3610             continue;
3611 
3612         sprintf(srgbtmp[found], "R%u", i);
3613         ++found;
3614         if (found == 4) break;
3615     }
3616 
3617     switch(found) {
3618         case 0:
3619             sprintf(srgbtmp[0], "TA");
3620             sprintf(srgbtmp[1], "TB");
3621             sprintf(srgbtmp[2], "TC");
3622             sprintf(srgbtmp[3], "TD");
3623             dcl_td = TRUE;
3624             break;
3625         case 1:
3626             sprintf(srgbtmp[1], "TA");
3627             sprintf(srgbtmp[2], "TB");
3628             sprintf(srgbtmp[3], "TC");
3629             break;
3630         case 2:
3631             sprintf(srgbtmp[2], "TA");
3632             sprintf(srgbtmp[3], "TB");
3633             break;
3634         case 3:
3635             sprintf(srgbtmp[3], "TA");
3636             break;
3637         case 4:
3638             break;
3639     }
3640 
3641     /*  Create the hw ARB shader */
3642     memset(&priv_ctx, 0, sizeof(priv_ctx));
3643     priv_ctx.cur_ps_args = args;
3644     priv_ctx.compiled_fprog = compiled;
3645     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3646     init_ps_input(shader, args, &priv_ctx);
3647     list_init(&priv_ctx.control_frames);
3648     priv_ctx.ps_post_process = args->super.srgb_correction;
3649 
3650     /* Avoid enabling NV_fragment_program* if we do not need it.
3651      *
3652      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3653      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3654      * is faster than what we gain from using higher native instructions. There are some things though
3655      * that cannot be emulated. In that case enable the extensions.
3656      * If the extension is enabled, instruction handlers that support both ways will use it.
3657      *
3658      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3659      * So enable the best we can get.
3660      */
3661     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3662        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3663     {
3664         want_nv_prog = TRUE;
3665     }
3666 
3667     shader_addline(buffer, "!!ARBfp1.0\n");
3668     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3669     {
3670         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3671         priv_ctx.target_version = NV3;
3672     }
3673     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3674     {
3675         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3676         priv_ctx.target_version = NV2;
3677     } else {
3678         if(want_nv_prog)
3679         {
3680             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3681              * limits properly
3682              */
3683             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3684             ERR("Try GLSL\n");
3685         }
3686         priv_ctx.target_version = ARB;
3687     }
3688 
3689     if (reg_maps->rt_mask > 1)
3690     {
3691         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3692     }
3693 
3694     if (reg_maps->shader_version.major < 3)
3695     {
3696         switch (args->super.fog)
3697         {
3698             case WINED3D_FFP_PS_FOG_OFF:
3699                 break;
3700             case WINED3D_FFP_PS_FOG_LINEAR:
3701                 if (gl_info->quirks & WINED3D_QUIRK_BROKEN_ARB_FOG)
3702                 {
3703                     custom_linear_fog = TRUE;
3704                     priv_ctx.ps_post_process = TRUE;
3705                     break;
3706                 }
3707                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3708                 break;
3709             case WINED3D_FFP_PS_FOG_EXP:
3710                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3711                 break;
3712             case WINED3D_FFP_PS_FOG_EXP2:
3713                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3714                 break;
3715         }
3716     }
3717 
3718     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3719      * unused temps away(but occupies them for the whole shader if they're used once). Always
3720      * declaring them avoids tricky bookkeeping work
3721      */
3722     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3723     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3724     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3725     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3726     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3727     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3728     wined3d_ftoa(eps, ftoa_tmp);
3729     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %s, 0.0 };\n", ftoa_tmp);
3730 
3731     if (reg_maps->shader_version.major < 2)
3732     {
3733         strcpy(fragcolor, "R0");
3734     }
3735     else
3736     {
3737         if (priv_ctx.ps_post_process)
3738         {
3739             if (shader->u.ps.color0_mov)
3740             {
3741                 sprintf(fragcolor, "R%u", shader->u.ps.color0_reg);
3742             }
3743             else
3744             {
3745                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3746                 strcpy(fragcolor, "TMP_COLOR");
3747             }
3748         } else {
3749             strcpy(fragcolor, "result.color");
3750         }
3751     }
3752 
3753     if (args->super.srgb_correction)
3754     {
3755         shader_addline(buffer, "PARAM srgb_consts0 = ");
3756         shader_arb_append_imm_vec4(buffer, wined3d_srgb_const0);
3757         shader_addline(buffer, ";\n");
3758         shader_addline(buffer, "PARAM srgb_consts1 = ");
3759         shader_arb_append_imm_vec4(buffer, wined3d_srgb_const1);
3760         shader_addline(buffer, ";\n");
3761     }
3762 
3763     /* Base Declarations */
3764     shader_generate_arb_declarations(shader, reg_maps, buffer, gl_info, NULL, &priv_ctx);
3765 
3766     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3767     {
3768         unsigned char bump_const;
3769 
3770         if (!(map & 1)) continue;
3771 
3772         bump_const = compiled->numbumpenvmatconsts;
3773         compiled->bumpenvmatconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3774         compiled->bumpenvmatconst[bump_const].texunit = i;
3775         compiled->luminanceconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3776         compiled->luminanceconst[bump_const].texunit = i;
3777 
3778         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3779          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3780          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3781          * textures due to conditional NP2 restrictions)
3782          *
3783          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3784          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3785          * their location is shader dependent anyway and they cannot be loaded globally.
3786          */
3787         compiled->bumpenvmatconst[bump_const].const_num = next_local++;
3788         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3789                        i, compiled->bumpenvmatconst[bump_const].const_num);
3790         compiled->numbumpenvmatconsts = bump_const + 1;
3791 
3792         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3793 
3794         compiled->luminanceconst[bump_const].const_num = next_local++;
3795         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3796                        i, compiled->luminanceconst[bump_const].const_num);
3797     }
3798 
3799     for(i = 0; i < MAX_CONST_I; i++)
3800     {
3801         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3802         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3803         {
3804             const DWORD *control_values = find_loop_control_values(shader, i);
3805 
3806             if(control_values)
3807             {
3808                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3809                                 control_values[0], control_values[1], control_values[2]);
3810             }
3811             else
3812             {
3813                 compiled->int_consts[i] = next_local;
3814                 compiled->num_int_consts++;
3815                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3816             }
3817         }
3818     }
3819 
3820     if(reg_maps->vpos || reg_maps->usesdsy)
3821     {
3822         compiled->ycorrection = next_local;
3823         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3824 
3825         if(reg_maps->vpos)
3826         {
3827             shader_addline(buffer, "TEMP vpos;\n");
3828             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3829              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3830              * ycorrection.z: 1.0
3831              * ycorrection.w: 0.0
3832              */
3833             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3834             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3835         }
3836     }
3837     else
3838     {
3839         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3840     }
3841 
3842     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3843      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3844      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3845      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3846      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3847      * shader compilation errors and the subsequent errors when drawing with this shader. */
3848     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3849         unsigned char cur_fixup_sampler = 0;
3850 
3851         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3852         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3853         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3854 
3855         fixup->offset = next_local;
3856         fixup->super.active = 0;
3857 
3858         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3859             if (!(map & (1 << i))) continue;
3860 
3861             if (fixup->offset + (cur_fixup_sampler >> 1) < max_lconsts) {
3862                 fixup->super.active |= (1 << i);
3863                 fixup->super.idx[i] = cur_fixup_sampler++;
3864             } else {
3865                 FIXME("No free constant found to load NP2 fixup data into shader. "
3866                       "Sampling from this texture will probably look wrong.\n");
3867                 break;
3868             }
3869         }
3870 
3871         fixup->super.num_consts = (cur_fixup_sampler + 1) >> 1;
3872         if (fixup->super.num_consts) {
3873             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3874                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3875         }
3876     }
3877 
3878     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3879     {
3880         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3881     }
3882 
3883     /* Base Shader Body */
3884     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
3885 
3886     if(args->super.srgb_correction) {
3887         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3888                                   priv_ctx.target_version >= NV2);
3889     }
3890 
3891     if (custom_linear_fog)
3892         arbfp_add_linear_fog(buffer, fragcolor, "TA");
3893 
3894     if(strcmp(fragcolor, "result.color")) {
3895         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3896     }
3897     shader_addline(buffer, "END\n");
3898 
3899     /* TODO: change to resource.glObjectHandle or something like that */
3900     GL_EXTCALL(glGenProgramsARB(1, &retval));
3901 
3902     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3903     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3904 
3905     TRACE("Created hw pixel shader, prg=%d\n", retval);
3906     /* Create the program and check for errors */
3907     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3908                buffer->content_size, buffer->buffer));
3909     checkGLcall("glProgramStringARB()");
3910 
3911     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3912     if (errPos != -1)
3913     {
3914         FIXME("HW PixelShader Error at position %d: %s\n\n",
3915               errPos, debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3916         shader_arb_dump_program_source(buffer->buffer);
3917         retval = 0;
3918     }
3919     else
3920     {
3921         GLint native;
3922 
3923         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3924         checkGLcall("glGetProgramivARB()");
3925         if (!native) WARN("Program exceeds native resource limits.\n");
3926     }
3927 
3928     return retval;
3929 }
3930 
3931 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3932 {
3933     unsigned int i;
3934     int ret;
3935 
3936     for(i = 0; i < MAX_REG_INPUT; i++)
3937     {
3938         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3939         {
3940             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3941             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3942             continue;
3943         }
3944 
3945         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3946         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3947         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3948         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].component_type  < sig2[i].component_type  ? -1 : 1;
3949         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3950         if(sig1[i].mask            != sig2[i].mask)            return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3951     }
3952     return 0;
3953 }
3954 
3955 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3956 {
3957     struct wined3d_shader_signature_element *new;
3958     int i;
3959     char *name;
3960 
3961     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3962     for(i = 0; i < MAX_REG_INPUT; i++)
3963     {
3964         if (!sig[i].semantic_name) continue;
3965 
3966         new[i] = sig[i];
3967         /* Clone the semantic string */
3968         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3969         strcpy(name, sig[i].semantic_name);
3970         new[i].semantic_name = name;
3971     }
3972     return new;
3973 }
3974 
3975 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3976 {
3977     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3978     struct ps_signature *found_sig;
3979 
3980     if (entry)
3981     {
3982         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3983         TRACE("Found existing signature %u\n", found_sig->idx);
3984         return found_sig->idx;
3985     }
3986     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*found_sig));
3987     found_sig->sig = clone_sig(sig);
3988     found_sig->idx = priv->ps_sig_number++;
3989     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3990     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3991     {
3992         ERR("Failed to insert program entry.\n");
3993     }
3994     return found_sig->idx;
3995 }
3996 
3997 static void init_output_registers(const struct wined3d_shader *shader,
3998         const struct wined3d_shader_signature_element *ps_input_sig,
3999         struct shader_arb_ctx_priv *priv_ctx, struct arb_vs_compiled_shader *compiled)
4000 {
4001     unsigned int i, j;
4002     static const char * const texcoords[8] =
4003     {
4004         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
4005         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
4006     };
4007     const char *semantic_name;
4008     DWORD semantic_idx, reg_idx;
4009 
4010     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
4011      * and varying 9 to result.color.secondary
4012      */
4013     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
4014     {
4015         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
4016         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
4017         "result.color.primary", "result.color.secondary"
4018     };
4019 
4020     if (!ps_input_sig)
4021     {
4022         TRACE("Pixel shader uses builtin varyings\n");
4023         /* Map builtins to builtins */
4024         for(i = 0; i < 8; i++)
4025         {
4026             priv_ctx->texcrd_output[i] = texcoords[i];
4027         }
4028         priv_ctx->color_output[0] = "result.color.primary";
4029         priv_ctx->color_output[1] = "result.color.secondary";
4030         priv_ctx->fog_output = "TMP_FOGCOORD";
4031 
4032         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
4033         for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4034         {
4035             semantic_name = shader->output_signature[i].semantic_name;
4036             if (!semantic_name) continue;
4037 
4038             if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION))
4039             {
4040                 TRACE("o%u is TMP_OUT\n", i);
4041                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
4042                 else priv_ctx->vs_output[i] = "TA";
4043             }
4044             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE))
4045             {
4046                 TRACE("o%u is result.pointsize\n", i);
4047                 if (!shader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
4048                 else priv_ctx->vs_output[i] = "TA";
4049             }
4050             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4051             {
4052                 TRACE("o%u is result.color.?, idx %u\n", i, shader->output_signature[i].semantic_idx);
4053                 if (!shader->output_signature[i].semantic_idx)
4054                     priv_ctx->vs_output[i] = "result.color.primary";
4055                 else if (shader->output_signature[i].semantic_idx == 1)
4056                     priv_ctx->vs_output[i] = "result.color.secondary";
4057                 else priv_ctx->vs_output[i] = "TA";
4058             }
4059             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4060             {
4061                 TRACE("o%u is %s\n", i, texcoords[shader->output_signature[i].semantic_idx]);
4062                 if (shader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
4063                 else priv_ctx->vs_output[i] = texcoords[shader->output_signature[i].semantic_idx];
4064             }
4065             else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4066             {
4067                 TRACE("o%u is result.fogcoord\n", i);
4068                 if (shader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
4069                 else priv_ctx->vs_output[i] = "result.fogcoord";
4070             }
4071             else
4072             {
4073                 priv_ctx->vs_output[i] = "TA";
4074             }
4075         }
4076         return;
4077     }
4078 
4079     TRACE("Pixel shader uses declared varyings\n");
4080 
4081     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
4082     for(i = 0; i < 8; i++)
4083     {
4084         priv_ctx->texcrd_output[i] = "TA";
4085     }
4086     priv_ctx->color_output[0] = "TA";
4087     priv_ctx->color_output[1] = "TA";
4088     priv_ctx->fog_output = "TA";
4089 
4090     for(i = 0; i < MAX_REG_INPUT; i++)
4091     {
4092         semantic_name = ps_input_sig[i].semantic_name;
4093         semantic_idx = ps_input_sig[i].semantic_idx;
4094         reg_idx = ps_input_sig[i].register_idx;
4095         if (!semantic_name) continue;
4096 
4097         /* If a declared input register is not written by builtin arguments, don't write to it.
4098          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
4099          *
4100          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4101          * to TMP_OUT in any case
4102          */
4103         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4104         {
4105             if (semantic_idx < 8)
4106                 priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4107         }
4108         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
4109         {
4110             if (semantic_idx < 2)
4111                 priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4112         }
4113         else if(shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
4114         {
4115             if (!semantic_idx)
4116                 priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4117         }
4118         else
4119         {
4120             continue;
4121         }
4122 
4123         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4124                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4125         {
4126             compiled->need_color_unclamp = TRUE;
4127         }
4128     }
4129 
4130     /* Map declared to declared */
4131     for (i = 0; i < (sizeof(shader->output_signature) / sizeof(*shader->output_signature)); ++i)
4132     {
4133         /* Write unread output to TA to throw them away */
4134         priv_ctx->vs_output[i] = "TA";
4135         semantic_name = shader->output_signature[i].semantic_name;
4136         if (!semantic_name) continue;
4137 
4138         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_POSITION)
4139                 && !shader->output_signature[i].semantic_idx)
4140         {
4141             priv_ctx->vs_output[i] = "TMP_OUT";
4142             continue;
4143         }
4144         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_PSIZE)
4145                 && !shader->output_signature[i].semantic_idx)
4146         {
4147             priv_ctx->vs_output[i] = "result.pointsize";
4148             continue;
4149         }
4150 
4151         for(j = 0; j < MAX_REG_INPUT; j++)
4152         {
4153             if (!ps_input_sig[j].semantic_name) continue;
4154 
4155             if (!strcmp(ps_input_sig[j].semantic_name, semantic_name)
4156                     && ps_input_sig[j].semantic_idx == shader->output_signature[i].semantic_idx)
4157             {
4158                 priv_ctx->vs_output[i] = decl_idx_to_string[ps_input_sig[j].register_idx];
4159 
4160                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4161                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4162                 {
4163                     compiled->need_color_unclamp = TRUE;
4164                 }
4165             }
4166         }
4167     }
4168 }
4169 
4170 /* Context activation is done by the caller. */
4171 static GLuint shader_arb_generate_vshader(const struct wined3d_shader *shader,
4172         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
4173         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled,
4174         const struct wined3d_shader_signature_element *ps_input_sig)
4175 {
4176     const struct arb_vshader_private *shader_data = shader->backend_data;
4177     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4178     struct shader_arb_priv *priv = shader->device->shader_priv;
4179     const DWORD *function = shader->function;
4180     GLuint ret;
4181     DWORD next_local = 0;
4182     struct shader_arb_ctx_priv priv_ctx;
4183     unsigned int i;
4184     GLint errPos;
4185 
4186     memset(&priv_ctx, 0, sizeof(priv_ctx));
4187     priv_ctx.cur_vs_args = args;
4188     list_init(&priv_ctx.control_frames);
4189     init_output_registers(shader, ps_input_sig, &priv_ctx, compiled);
4190 
4191     /*  Create the hw ARB shader */
4192     shader_addline(buffer, "!!ARBvp1.0\n");
4193 
4194     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4195      * mesurable performance penalty, and we can always make use of it for clipplanes.
4196      */
4197     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4198     {
4199         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4200         priv_ctx.target_version = NV3;
4201         shader_addline(buffer, "ADDRESS aL;\n");
4202     }
4203     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4204     {
4205         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4206         priv_ctx.target_version = NV2;
4207         shader_addline(buffer, "ADDRESS aL;\n");
4208     } else {
4209         priv_ctx.target_version = ARB;
4210     }
4211 
4212     shader_addline(buffer, "TEMP TMP_OUT;\n");
4213     if (reg_maps->fog)
4214         shader_addline(buffer, "TEMP TMP_FOGCOORD;\n");
4215     if (need_helper_const(shader_data, reg_maps, gl_info))
4216     {
4217         char ftoa_tmp[17];
4218         wined3d_ftoa(eps, ftoa_tmp);
4219         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %s};\n", ftoa_tmp);
4220     }
4221     if (need_rel_addr_const(shader_data, reg_maps, gl_info))
4222     {
4223         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", shader_data->rel_offset);
4224         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4225     }
4226 
4227     shader_addline(buffer, "TEMP TA;\n");
4228     shader_addline(buffer, "TEMP TB;\n");
4229 
4230     /* Base Declarations */
4231     shader_generate_arb_declarations(shader, reg_maps, buffer, gl_info,
4232             &priv_ctx.vs_clipplanes, &priv_ctx);
4233 
4234     for(i = 0; i < MAX_CONST_I; i++)
4235     {
4236         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4237         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4238         {
4239             const DWORD *control_values = find_loop_control_values(shader, i);
4240 
4241             if(control_values)
4242             {
4243                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4244                                 control_values[0], control_values[1], control_values[2]);
4245             }
4246             else
4247             {
4248                 compiled->int_consts[i] = next_local;
4249                 compiled->num_int_consts++;
4250                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4251             }
4252         }
4253     }
4254 
4255     /* We need a constant to fixup the final position */
4256     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4257     compiled->pos_fixup = next_local++;
4258 
4259     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4260      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4261      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4262      * a replacement shader depend on the texcoord.w being set properly.
4263      *
4264      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4265      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4266      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4267      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4268      * this can eat a number of instructions, so skip it unless this cap is set as well
4269      */
4270     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4271     {
4272         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4273         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4274 
4275         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !priv->ffp_proj_control)
4276         {
4277             int i;
4278             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4279             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4280             {
4281                 if (reg_maps->texcoord_mask[i] && reg_maps->texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4282                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4283             }
4284         }
4285     }
4286 
4287     /* The shader starts with the main function */
4288     priv_ctx.in_main_func = TRUE;
4289     /* Base Shader Body */
4290     shader_generate_main(shader, buffer, reg_maps, function, &priv_ctx);
4291 
4292     if (!priv_ctx.footer_written) vshader_add_footer(&priv_ctx,
4293             shader_data, args, reg_maps, gl_info, buffer);
4294 
4295     shader_addline(buffer, "END\n");
4296 
4297     /* TODO: change to resource.glObjectHandle or something like that */
4298     GL_EXTCALL(glGenProgramsARB(1, &ret));
4299 
4300     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4301     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4302 
4303     TRACE("Created hw vertex shader, prg=%d\n", ret);
4304     /* Create the program and check for errors */
4305     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4306                buffer->content_size, buffer->buffer));
4307     checkGLcall("glProgramStringARB()");
4308 
4309     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4310     if (errPos != -1)
4311     {
4312         FIXME("HW VertexShader Error at position %d: %s\n\n",
4313               errPos, debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4314         shader_arb_dump_program_source(buffer->buffer);
4315         ret = -1;
4316     }
4317     else
4318     {
4319         GLint native;
4320 
4321         GL_EXTCALL(glGetProgramivARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4322         checkGLcall("glGetProgramivARB()");
4323         if (!native) WARN("Program exceeds native resource limits.\n");
4324     }
4325 
4326     return ret;
4327 }
4328 
4329 /* Context activation is done by the caller. */
4330 static struct arb_ps_compiled_shader *find_arb_pshader(struct wined3d_shader *shader,
4331         const struct arb_ps_compile_args *args)
4332 {
4333     struct wined3d_device *device = shader->device;
4334     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4335     const struct wined3d_d3d_info *d3d_info = &device->adapter->d3d_info;
4336     UINT i;
4337     DWORD new_size;
4338     struct arb_ps_compiled_shader *new_array;
4339     struct wined3d_shader_buffer buffer;
4340     struct arb_pshader_private *shader_data;
4341     GLuint ret;
4342 
4343     if (!shader->backend_data)
4344     {
4345         struct shader_arb_priv *priv = device->shader_priv;
4346 
4347         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4348         shader_data = shader->backend_data;
4349         shader_data->clamp_consts = shader->reg_maps.shader_version.major == 1;
4350 
4351         if (shader->reg_maps.shader_version.major < 3)
4352             shader_data->input_signature_idx = ~0U;
4353         else
4354             shader_data->input_signature_idx = find_input_signature(priv, shader->input_signature);
4355 
4356         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4357 
4358         if (!d3d_info->vs_clipping)
4359             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->reg_maps,
4360                     d3d_info->limits.ffp_blend_stages - 1);
4361         else
4362             shader_data->clipplane_emulation = ~0U;
4363     }
4364     shader_data = shader->backend_data;
4365 
4366     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4367      * so a linear search is more performant than a hashmap or a binary search
4368      * (cache coherency etc)
4369      */
4370     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4371     {
4372         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4373             return &shader_data->gl_shaders[i];
4374     }
4375 
4376     TRACE("No matching GL shader found, compiling a new shader\n");
4377     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4378         if (shader_data->num_gl_shaders)
4379         {
4380             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4381             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4382                                     new_size * sizeof(*shader_data->gl_shaders));
4383         } else {
4384             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4385             new_size = 1;
4386         }
4387 
4388         if(!new_array) {
4389             ERR("Out of memory\n");
4390             return 0;
4391         }
4392         shader_data->gl_shaders = new_array;
4393         shader_data->shader_array_size = new_size;
4394     }
4395 
4396     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4397 
4398     pixelshader_update_samplers(shader, args->super.tex_types);
4399 
4400     if (!shader_buffer_init(&buffer))
4401     {
4402         ERR("Failed to initialize shader buffer.\n");
4403         return 0;
4404     }
4405 
4406     ret = shader_arb_generate_pshader(shader, gl_info, &buffer, args,
4407             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4408     shader_buffer_free(&buffer);
4409     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4410 
4411     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4412 }
4413 
4414 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4415                                  const DWORD use_map, BOOL skip_int) {
4416     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4417     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4418     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4419     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4420     if(stored->ps_signature != new->ps_signature) return FALSE;
4421     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4422     if(skip_int) return TRUE;
4423 
4424     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4425 }
4426 
4427 static struct arb_vs_compiled_shader *find_arb_vshader(struct wined3d_shader *shader,
4428         const struct wined3d_gl_info *gl_info, DWORD use_map, const struct arb_vs_compile_args *args,
4429         const struct wined3d_shader_signature_element *ps_input_sig)
4430 {
4431     UINT i;
4432     DWORD new_size;
4433     struct arb_vs_compiled_shader *new_array;
4434     struct wined3d_shader_buffer buffer;
4435     struct arb_vshader_private *shader_data;
4436     GLuint ret;
4437 
4438     if (!shader->backend_data)
4439     {
4440         const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4441 
4442         shader->backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4443         shader_data = shader->backend_data;
4444 
4445         if ((gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
4446                 && reg_maps->min_rel_offset <= reg_maps->max_rel_offset)
4447         {
4448             if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 127)
4449             {
4450                 FIXME("The difference between the minimum and maximum relative offset is > 127.\n");
4451                 FIXME("Which this OpenGL implementation does not support. Try using GLSL.\n");
4452                 FIXME("Min: %u, Max: %u.\n", reg_maps->min_rel_offset, reg_maps->max_rel_offset);
4453             }
4454             else if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 63)
4455                 shader_data->rel_offset = reg_maps->min_rel_offset + 63;
4456             else if (reg_maps->max_rel_offset > 63)
4457                 shader_data->rel_offset = reg_maps->min_rel_offset;
4458         }
4459     }
4460     shader_data = shader->backend_data;
4461 
4462     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4463      * so a linear search is more performant than a hashmap or a binary search
4464      * (cache coherency etc)
4465      */
4466     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4467         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4468                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4469         {
4470             return &shader_data->gl_shaders[i];
4471         }
4472     }
4473 
4474     TRACE("No matching GL shader found, compiling a new shader\n");
4475 
4476     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4477         if (shader_data->num_gl_shaders)
4478         {
4479             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4480             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4481                                     new_size * sizeof(*shader_data->gl_shaders));
4482         } else {
4483             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4484             new_size = 1;
4485         }
4486 
4487         if(!new_array) {
4488             ERR("Out of memory\n");
4489             return 0;
4490         }
4491         shader_data->gl_shaders = new_array;
4492         shader_data->shader_array_size = new_size;
4493     }
4494 
4495     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4496 
4497     if (!shader_buffer_init(&buffer))
4498     {
4499         ERR("Failed to initialize shader buffer.\n");
4500         return 0;
4501     }
4502 
4503     ret = shader_arb_generate_vshader(shader, gl_info, &buffer, args,
4504             &shader_data->gl_shaders[shader_data->num_gl_shaders],
4505             ps_input_sig);
4506     shader_buffer_free(&buffer);
4507     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4508 
4509     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4510 }
4511 
4512 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4513         const struct wined3d_context *context, const struct wined3d_shader *shader,
4514         struct arb_ps_compile_args *args)
4515 {
4516     const struct wined3d_gl_info *gl_info = context->gl_info;
4517     const struct wined3d_d3d_info *d3d_info = context->d3d_info;
4518     int i;
4519     WORD int_skip;
4520 
4521     find_ps_compile_args(state, shader, context->stream_info.position_transformed, &args->super, gl_info);
4522 
4523     /* This forces all local boolean constants to 1 to make them stateblock independent */
4524     args->bools = shader->reg_maps.local_bool_consts;
4525 
4526     for(i = 0; i < MAX_CONST_B; i++)
4527     {
4528         if (state->ps_consts_b[i])
4529             args->bools |= ( 1 << i);
4530     }
4531 
4532     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4533      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4534      * duplicate the shader than have a no-op KIL instruction in every shader
4535      */
4536     if (!d3d_info->vs_clipping && use_vs(state)
4537             && state->render_states[WINED3D_RS_CLIPPING]
4538             && state->render_states[WINED3D_RS_CLIPPLANEENABLE])
4539         args->clip = 1;
4540     else
4541         args->clip = 0;
4542 
4543     /* Skip if unused or local, or supported natively */
4544     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4545     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4546     {
4547         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4548         return;
4549     }
4550 
4551     for(i = 0; i < MAX_CONST_I; i++)
4552     {
4553         if(int_skip & (1 << i))
4554         {
4555             args->loop_ctrl[i][0] = 0;
4556             args->loop_ctrl[i][1] = 0;
4557             args->loop_ctrl[i][2] = 0;
4558         }
4559         else
4560         {
4561             args->loop_ctrl[i][0] = state->ps_consts_i[i * 4];
4562             args->loop_ctrl[i][1] = state->ps_consts_i[i * 4 + 1];
4563             args->loop_ctrl[i][2] = state->ps_consts_i[i * 4 + 2];
4564         }
4565     }
4566 }
4567 
4568 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4569         const struct wined3d_context *context, const struct wined3d_shader *shader,
4570         struct arb_vs_compile_args *args)
4571 {
4572     const struct wined3d_device *device = shader->device;
4573     const struct wined3d_adapter *adapter = device->adapter;
4574     const struct wined3d_gl_info *gl_info = context->gl_info;
4575     const struct wined3d_d3d_info *d3d_info = context->d3d_info;
4576     int i;
4577     WORD int_skip;
4578 
4579     find_vs_compile_args(state, shader, context->stream_info.swizzle_map, &args->super);
4580 
4581     args->clip.boolclip_compare = 0;
4582     if (use_ps(state))
4583     {
4584         const struct wined3d_shader *ps = state->shader[WINED3D_SHADER_TYPE_PIXEL];
4585         const struct arb_pshader_private *shader_priv = ps->backend_data;
4586         args->ps_signature = shader_priv->input_signature_idx;
4587 
4588         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4589     }
4590     else
4591     {
4592         args->ps_signature = ~0;
4593         if (!d3d_info->vs_clipping && adapter->fragment_pipe == &arbfp_fragment_pipeline)
4594             args->clip.boolclip.clip_texcoord = ffp_clip_emul(context) ? d3d_info->limits.ffp_blend_stages : 0;
4595         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4596     }
4597 
4598     if (args->clip.boolclip.clip_texcoord)
4599     {
4600         if (state->render_states[WINED3D_RS_CLIPPING])
4601             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3D_RS_CLIPPLANEENABLE];
4602         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4603     }
4604 
4605     /* This forces all local boolean constants to 1 to make them stateblock independent */
4606     args->clip.boolclip.bools = shader->reg_maps.local_bool_consts;
4607     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4608     for(i = 0; i < MAX_CONST_B; i++)
4609     {
4610         if (state->vs_consts_b[i])
4611             args->clip.boolclip.bools |= ( 1 << i);
4612     }
4613 
4614     args->vertex.samplers[0] = context->tex_unit_map[MAX_FRAGMENT_SAMPLERS + 0];
4615     args->vertex.samplers[1] = context->tex_unit_map[MAX_FRAGMENT_SAMPLERS + 1];
4616     args->vertex.samplers[2] = context->tex_unit_map[MAX_FRAGMENT_SAMPLERS + 2];
4617     args->vertex.samplers[3] = 0;
4618 
4619     /* Skip if unused or local */
4620     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4621     /* This is about flow control, not clipping. */
4622     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4623     {
4624         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4625         return;
4626     }
4627 
4628     for(i = 0; i < MAX_CONST_I; i++)
4629     {
4630         if(int_skip & (1 << i))
4631         {
4632             args->loop_ctrl[i][0] = 0;
4633             args->loop_ctrl[i][1] = 0;
4634             args->loop_ctrl[i][2] = 0;
4635         }
4636         else
4637         {
4638             args->loop_ctrl[i][0] = state->vs_consts_i[i * 4];
4639             args->loop_ctrl[i][1] = state->vs_consts_i[i * 4 + 1];
4640             args->loop_ctrl[i][2] = state->vs_consts_i[i * 4 + 2];
4641         }
4642     }
4643 }
4644 
4645 /* Context activation is done by the caller. */
4646 static void shader_arb_select(void *shader_priv, struct wined3d_context *context,
4647         const struct wined3d_state *state)
4648 {
4649     struct shader_arb_priv *priv = shader_priv;
4650     const struct wined3d_gl_info *gl_info = context->gl_info;
4651     int i;
4652 
4653     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4654     if (use_ps(state))
4655     {
4656         struct wined3d_shader *ps = state->shader[WINED3D_SHADER_TYPE_PIXEL];
4657         struct arb_ps_compile_args compile_args;
4658         struct arb_ps_compiled_shader *compiled;
4659 
4660         TRACE("Using pixel shader %p.\n", ps);
4661         find_arb_ps_compile_args(state, context, ps, &compile_args);
4662         compiled = find_arb_pshader(ps, &compile_args);
4663         priv->current_fprogram_id = compiled->prgId;
4664         priv->compiled_fprog = compiled;
4665 
4666         /* Bind the fragment program */
4667         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4668         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4669 
4670         if (!priv->use_arbfp_fixed_func)
4671             priv->fragment_pipe->enable_extension(gl_info, FALSE);
4672 
4673         /* Enable OpenGL fragment programs. */
4674         gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
4675         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4676 
4677         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", priv->current_fprogram_id);
4678 
4679         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4680          * a 1.x and newer shader, reload the first 8 constants
4681          */
4682         if (priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->backend_data)->clamp_consts)
4683         {
4684             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->backend_data)->clamp_consts;
4685             priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, 8);
4686             for(i = 0; i < 8; i++)
4687             {
4688                 priv->pshader_const_dirty[i] = 1;
4689             }
4690             /* Also takes care of loading local constants */
4691             shader_arb_load_constants_internal(shader_priv, context, state, TRUE, FALSE, TRUE);
4692         }
4693         else
4694         {
4695             UINT rt_height = state->fb->render_targets[0]->resource.height;
4696             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4697         }
4698 
4699         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4700         if (compiled->np2fixup_info.super.active)
4701             context->constant_update_mask |= WINED3D_SHADER_CONST_PS_NP2_FIXUP;
4702 
4703         if (ps->load_local_constsF)
4704             context->constant_update_mask |= WINED3D_SHADER_CONST_PS_F;
4705     }
4706     else
4707     {
4708         if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4709         {
4710             /* Disable only if we're not using arbfp fixed function fragment
4711              * processing. If this is used, keep GL_FRAGMENT_PROGRAM_ARB
4712              * enabled, and the fixed function pipeline will bind the fixed
4713              * function replacement shader. */
4714             gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4715             checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4716             priv->current_fprogram_id = 0;
4717         }
4718         priv->fragment_pipe->enable_extension(gl_info, TRUE);
4719     }
4720 
4721     if (use_vs(state))
4722     {
4723         struct wined3d_shader *vs = state->shader[WINED3D_SHADER_TYPE_VERTEX];
4724         struct arb_vs_compile_args compile_args;
4725         struct arb_vs_compiled_shader *compiled;
4726         const struct wined3d_shader_signature_element *ps_input_sig;
4727 
4728         TRACE("Using vertex shader %p\n", vs);
4729         find_arb_vs_compile_args(state, context, vs, &compile_args);
4730 
4731         /* Instead of searching for the signature in the signature list, read the one from the
4732          * current pixel shader. It's maybe not the shader where the signature came from, but it
4733          * is the same signature and faster to find. */
4734         if (compile_args.ps_signature == ~0U)
4735             ps_input_sig = NULL;
4736         else
4737             ps_input_sig = state->shader[WINED3D_SHADER_TYPE_PIXEL]->input_signature;
4738 
4739         compiled = find_arb_vshader(vs, context->gl_info, context->stream_info.use_map,
4740                 &compile_args, ps_input_sig);
4741         priv->current_vprogram_id = compiled->prgId;
4742         priv->compiled_vprog = compiled;
4743 
4744         /* Bind the vertex program */
4745         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4746         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4747 
4748         priv->vertex_pipe->vp_enable(gl_info, FALSE);
4749 
4750         /* Enable OpenGL vertex programs */
4751         gl_info->gl_ops.gl.p_glEnable(GL_VERTEX_PROGRAM_ARB);
4752         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4753         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", priv->current_vprogram_id);
4754         shader_arb_vs_local_constants(compiled, context, state);
4755 
4756         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4757             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4758 
4759             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4760             {
4761                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4762                 checkGLcall("glClampColorARB");
4763             } else {
4764                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4765             }
4766         }
4767 
4768         if (vs->load_local_constsF)
4769             context->constant_update_mask |= WINED3D_SHADER_CONST_VS_F;
4770     }
4771     else
4772     {
4773         if (gl_info->supported[ARB_VERTEX_PROGRAM])
4774         {
4775             priv->current_vprogram_id = 0;
4776             gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4777             checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4778         }
4779         priv->vertex_pipe->vp_enable(gl_info, TRUE);
4780     }
4781 }
4782 
4783 
4784 /* Context activation is done by the caller. */
4785 static void shader_arb_disable(void *shader_priv, struct wined3d_context *context)
4786 {
4787     const struct wined3d_gl_info *gl_info = context->gl_info;
4788     struct shader_arb_priv *priv = shader_priv;
4789 
4790     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4791     {
4792         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4793         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4794         priv->current_fprogram_id = 0;
4795     }
4796     priv->fragment_pipe->enable_extension(gl_info, FALSE);
4797 
4798     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4799     {
4800         priv->current_vprogram_id = 0;
4801         gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4802         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4803     }
4804     priv->vertex_pipe->vp_enable(gl_info, FALSE);
4805 
4806     if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT] && priv->last_vs_color_unclamp)
4807     {
4808         GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, GL_FIXED_ONLY_ARB));
4809         checkGLcall("glClampColorARB");
4810         priv->last_vs_color_unclamp = FALSE;
4811     }
4812 
4813     context->shader_update_mask = (1 << WINED3D_SHADER_TYPE_PIXEL)
4814             | (1 << WINED3D_SHADER_TYPE_VERTEX)
4815             | (1 << WINED3D_SHADER_TYPE_GEOMETRY);
4816 }
4817 
4818 /* Context activation is done by the caller. */
4819 static void shader_arb_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4820         enum tex_types tex_type, const SIZE *ds_mask_size)
4821 {
4822     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4823     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4824     struct shader_arb_priv *priv = shader_priv;
4825     GLuint *blt_fprogram;
4826 
4827     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4828     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4829     gl_info->gl_ops.gl.p_glEnable(GL_VERTEX_PROGRAM_ARB);
4830 
4831     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4832     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4833     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4834     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4835     gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
4836 }
4837 
4838 /* Context activation is done by the caller. */
4839 static void shader_arb_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4840 {
4841     struct shader_arb_priv *priv = shader_priv;
4842 
4843     if (priv->current_vprogram_id) {
4844         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4845         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4846 
4847         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB.\n", priv->current_vprogram_id);
4848     }
4849     else
4850     {
4851         gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4852         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4853     }
4854 
4855     if (priv->current_fprogram_id) {
4856         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4857         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4858 
4859         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB.\n", priv->current_fprogram_id);
4860     }
4861     else if(!priv->use_arbfp_fixed_func)
4862     {
4863         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4864         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4865     }
4866 }
4867 
4868 static void shader_arb_destroy(struct wined3d_shader *shader)
4869 {
4870     struct wined3d_device *device = shader->device;
4871     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4872 
4873     if (shader_is_pshader_version(shader->reg_maps.shader_version.type))
4874     {
4875         struct arb_pshader_private *shader_data = shader->backend_data;
4876         UINT i;
4877 
4878         if(!shader_data) return; /* This can happen if a shader was never compiled */
4879 
4880         if (shader_data->num_gl_shaders)
4881         {
4882             struct wined3d_context *context = context_acquire(device, NULL);
4883 
4884             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4885             {
4886                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4887                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4888             }
4889 
4890             context_release(context);
4891         }
4892 
4893         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4894         HeapFree(GetProcessHeap(), 0, shader_data);
4895         shader->backend_data = NULL;
4896     }
4897     else
4898     {
4899         struct arb_vshader_private *shader_data = shader->backend_data;
4900         UINT i;
4901 
4902         if(!shader_data) return; /* This can happen if a shader was never compiled */
4903 
4904         if (shader_data->num_gl_shaders)
4905         {
4906             struct wined3d_context *context = context_acquire(device, NULL);
4907 
4908             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4909             {
4910                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4911                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4912             }
4913 
4914             context_release(context);
4915         }
4916 
4917         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4918         HeapFree(GetProcessHeap(), 0, shader_data);
4919         shader->backend_data = NULL;
4920     }
4921 }
4922 
4923 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4924 {
4925     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4926     return compare_sig(key, e->sig);
4927 }
4928 
4929 static const struct wine_rb_functions sig_tree_functions =
4930 {
4931     wined3d_rb_alloc,
4932     wined3d_rb_realloc,
4933     wined3d_rb_free,
4934     sig_tree_compare
4935 };
4936 
4937 static HRESULT shader_arb_alloc(struct wined3d_device *device, const struct wined3d_vertex_pipe_ops *vertex_pipe,
4938         const struct fragment_pipeline *fragment_pipe)
4939 {
4940     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4941     struct fragment_caps fragment_caps;
4942     void *vertex_priv, *fragment_priv;
4943     const struct wined3d_d3d_info *d3d_info = &device->adapter->d3d_info;
4944 
4945     if (!(vertex_priv = vertex_pipe->vp_alloc(&arb_program_shader_backend, priv)))
4946     {
4947         ERR("Failed to initialize vertex pipe.\n");
4948         HeapFree(GetProcessHeap(), 0, priv);
4949         return E_FAIL;
4950     }
4951 
4952     if (!(fragment_priv = fragment_pipe->alloc_private(&arb_program_shader_backend, priv)))
4953     {
4954         ERR("Failed to initialize fragment pipe.\n");
4955         vertex_pipe->vp_free(device);
4956         HeapFree(GetProcessHeap(), 0, priv);
4957         return E_FAIL;
4958     }
4959 
4960     priv->vshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4961             sizeof(*priv->vshader_const_dirty) * d3d_info->limits.vs_uniform_count);
4962     if (!priv->vshader_const_dirty)
4963         goto fail;
4964     memset(priv->vshader_const_dirty, 1,
4965            sizeof(*priv->vshader_const_dirty) * d3d_info->limits.vs_uniform_count);
4966 
4967     priv->pshader_const_dirty = HeapAlloc(GetProcessHeap(), 0,
4968             sizeof(*priv->pshader_const_dirty) * d3d_info->limits.ps_uniform_count);
4969     if (!priv->pshader_const_dirty)
4970         goto fail;
4971     memset(priv->pshader_const_dirty, 1,
4972             sizeof(*priv->pshader_const_dirty) * d3d_info->limits.ps_uniform_count);
4973 
4974     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4975     {
4976         ERR("RB tree init failed\n");
4977         goto fail;
4978     }
4979 
4980     priv->vertex_pipe = vertex_pipe;
4981     priv->fragment_pipe = fragment_pipe;
4982     fragment_pipe->get_caps(&device->adapter->gl_info, &fragment_caps);
4983     priv->ffp_proj_control = fragment_caps.wined3d_caps & WINED3D_FRAGMENT_CAP_PROJ_CONTROL;
4984 
4985     device->vertex_priv = vertex_priv;
4986     device->fragment_priv = fragment_priv;
4987     device->shader_priv = priv;
4988 
4989     return WINED3D_OK;
4990 
4991 fail:
4992     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
4993     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
4994     fragment_pipe->free_private(device);
4995     vertex_pipe->vp_free(device);
4996     HeapFree(GetProcessHeap(), 0, priv);
4997     return E_OUTOFMEMORY;
4998 }
4999 
5000 static void release_signature(struct wine_rb_entry *entry, void *context)
5001 {
5002     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
5003     int i;
5004     for(i = 0; i < MAX_REG_INPUT; i++)
5005     {
5006         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
5007     }
5008     HeapFree(GetProcessHeap(), 0, sig->sig);
5009     HeapFree(GetProcessHeap(), 0, sig);
5010 }
5011 
5012 /* Context activation is done by the caller. */
5013 static void shader_arb_free(struct wined3d_device *device)
5014 {
5015     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
5016     struct shader_arb_priv *priv = device->shader_priv;
5017     int i;
5018 
5019     if (priv->depth_blt_vprogram_id)
5020         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
5021 
5022     for (i = 0; i < tex_type_count; ++i)
5023     {
5024         if (priv->depth_blt_fprogram_id_full[i])
5025         {
5026             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_full[i]));
5027         }
5028         if (priv->depth_blt_fprogram_id_masked[i])
5029         {
5030             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_masked[i]));
5031         }
5032     }
5033 
5034     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
5035     HeapFree(GetProcessHeap(), 0, priv->pshader_const_dirty);
5036     HeapFree(GetProcessHeap(), 0, priv->vshader_const_dirty);
5037     priv->fragment_pipe->free_private(device);
5038     priv->vertex_pipe->vp_free(device);
5039     HeapFree(GetProcessHeap(), 0, device->shader_priv);
5040 }
5041 
5042 static BOOL shader_arb_allocate_context_data(struct wined3d_context *context)
5043 {
5044     return TRUE;
5045 }
5046 
5047 static void shader_arb_free_context_data(struct wined3d_context *context)
5048 {
5049     struct shader_arb_priv *priv = context->swapchain->device->shader_priv;
5050 
5051     if (priv->last_context == context)
5052         priv->last_context = NULL;
5053 }
5054 
5055 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
5056 {
5057     if (gl_info->supported[ARB_VERTEX_PROGRAM])
5058     {
5059         DWORD vs_consts;
5060         UINT vs_version;
5061 
5062         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
5063          * for vertex programs. If the native limit is less than that it's
5064          * not very useful, and e.g. Mesa swrast returns 0, probably to
5065          * indicate it's a software implementation. */
5066         if (gl_info->limits.arb_vs_native_constants < 96)
5067             vs_consts = gl_info->limits.arb_vs_float_constants;
5068         else
5069             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
5070 
5071         if (gl_info->supported[NV_VERTEX_PROGRAM3])
5072         {
5073             vs_version = 3;
5074             TRACE("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
5075         }
5076         else if (vs_consts >= 256)
5077         {
5078             /* Shader Model 2.0 requires at least 256 vertex shader constants */
5079             vs_version = 2;
5080             TRACE("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
5081         }
5082         else
5083         {
5084             vs_version = 1;
5085             TRACE("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
5086         }
5087         caps->vs_version = min(wined3d_settings.max_sm_vs, vs_version);
5088         caps->vs_uniform_count = vs_consts;
5089     }
5090     else
5091     {
5092         caps->vs_version = 0;
5093         caps->vs_uniform_count = 0;
5094     }
5095 
5096     caps->gs_version = 0;
5097 
5098     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
5099     {
5100         DWORD ps_consts;
5101         UINT ps_version;
5102 
5103         /* Similar as above for vertex programs, but the minimum for fragment
5104          * programs is 24. */
5105         if (gl_info->limits.arb_ps_native_constants < 24)
5106             ps_consts = gl_info->limits.arb_ps_float_constants;
5107         else
5108             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
5109 
5110         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
5111         {
5112             ps_version = 3;
5113             TRACE("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
5114         }
5115         else if (ps_consts >= 32)
5116         {
5117             /* Shader Model 2.0 requires at least 32 pixel shader constants */
5118             ps_version = 2;
5119             TRACE("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
5120         }
5121         else
5122         {
5123             ps_version = 1;
5124             TRACE("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
5125         }
5126         caps->ps_version = min(wined3d_settings.max_sm_ps, ps_version);
5127         caps->ps_uniform_count = ps_consts;
5128         caps->ps_1x_max_value = 8.0f;
5129     }
5130     else
5131     {
5132         caps->ps_version = 0;
5133         caps->ps_uniform_count = 0;
5134         caps->ps_1x_max_value = 0.0f;
5135     }
5136 
5137     caps->wined3d_caps = WINED3D_SHADER_CAP_SRGB_WRITE;
5138     if (use_nv_clip(gl_info))
5139         caps->wined3d_caps |= WINED3D_SHADER_CAP_VS_CLIPPING;
5140 }
5141 
5142 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
5143 {
5144     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
5145     {
5146         TRACE("Checking support for color_fixup:\n");
5147         dump_color_fixup_desc(fixup);
5148     }
5149 
5150     /* We support everything except complex conversions. */
5151     if (!is_complex_fixup(fixup))
5152     {
5153         TRACE("[OK]\n");
5154         return TRUE;
5155     }
5156 
5157     TRACE("[FAILED]\n");
5158     return FALSE;
5159 }
5160 
5161 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
5162     DWORD shift;
5163     char write_mask[20], regstr[50];
5164     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5165     BOOL is_color = FALSE;
5166     const struct wined3d_shader_dst_param *dst;
5167 
5168     if (!ins->dst_count) return;
5169 
5170     dst = &ins->dst[0];
5171     shift = dst->shift;
5172     if (!shift) return; /* Saturate alone is handled by the instructions */
5173 
5174     shader_arb_get_write_mask(ins, dst, write_mask);
5175     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
5176 
5177     /* Generate a line that does the output modifier computation
5178      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
5179      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
5180      */
5181     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
5182                    regstr, write_mask, regstr, shift_tab[shift]);
5183 }
5184 
5185 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
5186 {
5187     /* WINED3DSIH_ABS                   */ shader_hw_map2gl,
5188     /* WINED3DSIH_ADD                   */ shader_hw_map2gl,
5189     /* WINED3DSIH_AND                   */ NULL,
5190     /* WINED3DSIH_BEM                   */ pshader_hw_bem,
5191     /* WINED3DSIH_BREAK                 */ shader_hw_break,
5192     /* WINED3DSIH_BREAKC                */ shader_hw_breakc,
5193     /* WINED3DSIH_BREAKP                */ NULL,
5194     /* WINED3DSIH_CALL                  */ shader_hw_call,
5195     /* WINED3DSIH_CALLNZ                */ NULL,
5196     /* WINED3DSIH_CMP                   */ pshader_hw_cmp,
5197     /* WINED3DSIH_CND                   */ pshader_hw_cnd,
5198     /* WINED3DSIH_CRS                   */ shader_hw_map2gl,
5199     /* WINED3DSIH_CUT                   */ NULL,
5200     /* WINED3DSIH_DCL                   */ shader_hw_nop,
5201     /* WINED3DSIH_DCL_CONSTANT_BUFFER   */ shader_hw_nop,
5202     /* WINED3DSIH_DCL_INPUT_PRIMITIVE   */ shader_hw_nop,
5203     /* WINED3DSIH_DCL_OUTPUT_TOPOLOGY   */ shader_hw_nop,
5204     /* WINED3DSIH_DCL_VERTICES_OUT      */ shader_hw_nop,
5205     /* WINED3DSIH_DEF                   */ shader_hw_nop,
5206     /* WINED3DSIH_DEFB                  */ shader_hw_nop,
5207     /* WINED3DSIH_DEFI                  */ shader_hw_nop,
5208     /* WINED3DSIH_DIV                   */ NULL,
5209     /* WINED3DSIH_DP2ADD                */ pshader_hw_dp2add,
5210     /* WINED3DSIH_DP3                   */ shader_hw_map2gl,
5211     /* WINED3DSIH_DP4                   */ shader_hw_map2gl,
5212     /* WINED3DSIH_DST                   */ shader_hw_map2gl,
5213     /* WINED3DSIH_DSX                   */ shader_hw_map2gl,
5214     /* WINED3DSIH_DSY                   */ shader_hw_dsy,
5215     /* WINED3DSIH_ELSE                  */ shader_hw_else,
5216     /* WINED3DSIH_EMIT                  */ NULL,
5217     /* WINED3DSIH_ENDIF                 */ shader_hw_endif,
5218     /* WINED3DSIH_ENDLOOP               */ shader_hw_endloop,
5219     /* WINED3DSIH_ENDREP                */ shader_hw_endrep,
5220     /* WINED3DSIH_EQ                    */ NULL,
5221     /* WINED3DSIH_EXP                   */ shader_hw_scalar_op,
5222     /* WINED3DSIH_EXPP                  */ shader_hw_scalar_op,
5223     /* WINED3DSIH_FRC                   */ shader_hw_map2gl,
5224     /* WINED3DSIH_FTOI                  */ NULL,
5225     /* WINED3DSIH_GE                    */ NULL,
5226     /* WINED3DSIH_IADD                  */ NULL,
5227     /* WINED3DSIH_IEQ                   */ NULL,
5228     /* WINED3DSIH_IF                    */ NULL /* Hardcoded into the shader */,
5229     /* WINED3DSIH_IFC                   */ shader_hw_ifc,
5230     /* WINED3DSIH_IGE                   */ NULL,
5231     /* WINED3DSIH_IMUL                  */ NULL,
5232     /* WINED3DSIH_ITOF                  */ NULL,
5233     /* WINED3DSIH_LABEL                 */ shader_hw_label,
5234     /* WINED3DSIH_LD                    */ NULL,
5235     /* WINED3DSIH_LIT                   */ shader_hw_map2gl,
5236     /* WINED3DSIH_LOG                   */ shader_hw_log,
5237     /* WINED3DSIH_LOGP                  */ shader_hw_log,
5238     /* WINED3DSIH_LOOP                  */ shader_hw_loop,
5239     /* WINED3DSIH_LRP                   */ shader_hw_lrp,
5240     /* WINED3DSIH_LT                    */ NULL,
5241     /* WINED3DSIH_M3x2                  */ shader_hw_mnxn,
5242     /* WINED3DSIH_M3x3                  */ shader_hw_mnxn,
5243     /* WINED3DSIH_M3x4                  */ shader_hw_mnxn,
5244     /* WINED3DSIH_M4x3                  */ shader_hw_mnxn,
5245     /* WINED3DSIH_M4x4                  */ shader_hw_mnxn,
5246     /* WINED3DSIH_MAD                   */ shader_hw_map2gl,
5247     /* WINED3DSIH_MAX                   */ shader_hw_map2gl,
5248     /* WINED3DSIH_MIN                   */ shader_hw_map2gl,
5249     /* WINED3DSIH_MOV                   */ shader_hw_mov,
5250     /* WINED3DSIH_MOVA                  */ shader_hw_mov,
5251     /* WINED3DSIH_MOVC                  */ NULL,
5252     /* WINED3DSIH_MUL                   */ shader_hw_map2gl,
5253     /* WINED3DSIH_NOP                   */ shader_hw_nop,
5254     /* WINED3DSIH_NRM                   */ shader_hw_nrm,
5255     /* WINED3DSIH_PHASE                 */ shader_hw_nop,
5256     /* WINED3DSIH_POW                   */ shader_hw_pow,
5257     /* WINED3DSIH_RCP                   */ shader_hw_rcp,
5258     /* WINED3DSIH_REP                   */ shader_hw_rep,
5259     /* WINED3DSIH_RET                   */ shader_hw_ret,
5260     /* WINED3DSIH_ROUND_NI              */ NULL,
5261     /* WINED3DSIH_RSQ                   */ shader_hw_scalar_op,
5262     /* WINED3DSIH_SAMPLE                */ NULL,
5263     /* WINED3DSIH_SAMPLE_GRAD           */ NULL,
5264     /* WINED3DSIH_SAMPLE_LOD            */ NULL,
5265     /* WINED3DSIH_SETP                  */ NULL,
5266     /* WINED3DSIH_SGE                   */ shader_hw_map2gl,
5267     /* WINED3DSIH_SGN                   */ shader_hw_sgn,
5268     /* WINED3DSIH_SINCOS                */ shader_hw_sincos,
5269     /* WINED3DSIH_SLT                   */ shader_hw_map2gl,
5270     /* WINED3DSIH_SQRT                  */ NULL,
5271     /* WINED3DSIH_SUB                   */ shader_hw_map2gl,
5272     /* WINED3DSIH_TEX                   */ pshader_hw_tex,
5273     /* WINED3DSIH_TEXBEM                */ pshader_hw_texbem,
5274     /* WINED3DSIH_TEXBEML               */ pshader_hw_texbem,
5275     /* WINED3DSIH_TEXCOORD              */ pshader_hw_texcoord,
5276     /* WINED3DSIH_TEXDEPTH              */ pshader_hw_texdepth,
5277     /* WINED3DSIH_TEXDP3                */ pshader_hw_texdp3,
5278     /* WINED3DSIH_TEXDP3TEX             */ pshader_hw_texdp3tex,
5279     /* WINED3DSIH_TEXKILL               */ pshader_hw_texkill,
5280     /* WINED3DSIH_TEXLDD                */ shader_hw_texldd,
5281     /* WINED3DSIH_TEXLDL                */ shader_hw_texldl,
5282     /* WINED3DSIH_TEXM3x2DEPTH          */ pshader_hw_texm3x2depth,
5283     /* WINED3DSIH_TEXM3x2PAD            */ pshader_hw_texm3x2pad,
5284     /* WINED3DSIH_TEXM3x2TEX            */ pshader_hw_texm3x2tex,
5285     /* WINED3DSIH_TEXM3x3               */ pshader_hw_texm3x3,
5286     /* WINED3DSIH_TEXM3x3DIFF           */ NULL,
5287     /* WINED3DSIH_TEXM3x3PAD            */ pshader_hw_texm3x3pad,
5288     /* WINED3DSIH_TEXM3x3SPEC           */ pshader_hw_texm3x3spec,
5289     /* WINED3DSIH_TEXM3x3TEX            */ pshader_hw_texm3x3tex,
5290     /* WINED3DSIH_TEXM3x3VSPEC          */ pshader_hw_texm3x3vspec,
5291     /* WINED3DSIH_TEXREG2AR             */ pshader_hw_texreg2ar,
5292     /* WINED3DSIH_TEXREG2GB             */ pshader_hw_texreg2gb,
5293     /* WINED3DSIH_TEXREG2RGB            */ pshader_hw_texreg2rgb,
5294     /* WINED3DSIH_UDIV                  */ NULL,
5295     /* WINED3DSIH_USHR                  */ NULL,
5296     /* WINED3DSIH_UTOF                  */ NULL,
5297     /* WINED3DSIH_XOR                   */ NULL,
5298 };
5299 
5300 static BOOL get_bool_const(const struct wined3d_shader_instruction *ins,
5301         const struct wined3d_shader *shader, DWORD idx)
5302 {
5303     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5304     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5305     const struct wined3d_shader_lconst *constant;
5306     WORD bools = 0;
5307     WORD flag = (1 << idx);
5308     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5309 
5310     if (reg_maps->local_bool_consts & flag)
5311     {
5312         /* What good is an if(bool) with a hardcoded local constant? I don't know, but handle it */
5313         LIST_FOR_EACH_ENTRY(constant, &shader->constantsB, struct wined3d_shader_lconst, entry)
5314         {
5315             if (constant->idx == idx)
5316             {
5317                 return constant->value[0];
5318             }
5319         }
5320         ERR("Local constant not found\n");
5321         return FALSE;
5322     }
5323     else
5324     {
5325         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5326         else bools = priv->cur_ps_args->bools;
5327         return bools & flag;
5328     }
5329 }
5330 
5331 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5332         const struct wined3d_shader *shader, UINT idx, struct wined3d_shader_loop_control *loop_control)
5333 {
5334     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5335     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5336 
5337     /* Integer constants can either be a local constant, or they can be stored in the shader
5338      * type specific compile args. */
5339     if (reg_maps->local_int_consts & (1 << idx))
5340     {
5341         const struct wined3d_shader_lconst *constant;
5342 
5343         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
5344         {
5345             if (constant->idx == idx)
5346             {
5347                 loop_control->count = constant->value[0];
5348                 loop_control->start = constant->value[1];
5349                 /* Step is signed. */
5350                 loop_control->step = (int)constant->value[2];
5351                 return;
5352             }
5353         }
5354         /* If this happens the flag was set incorrectly */
5355         ERR("Local constant not found\n");
5356         loop_control->count = 0;
5357         loop_control->start = 0;
5358         loop_control->step = 0;
5359         return;
5360     }
5361 
5362     switch (reg_maps->shader_version.type)
5363     {
5364         case WINED3D_SHADER_TYPE_VERTEX:
5365             /* Count and aL start value are unsigned */
5366             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5367             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5368             /* Step is signed. */
5369             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5370             break;
5371 
5372         case WINED3D_SHADER_TYPE_PIXEL:
5373             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5374             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5375             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5376             break;
5377 
5378         default:
5379             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5380             break;
5381     }
5382 }
5383 
5384 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5385 {
5386     unsigned int i;
5387     struct wined3d_shader_dst_param *dst_param = NULL;
5388     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5389     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5390     if(!rec)
5391     {
5392         ERR("Out of memory\n");
5393         return;
5394     }
5395 
5396     rec->ins = *ins;
5397     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5398     if(!dst_param) goto free;
5399     *dst_param = *ins->dst;
5400     if (ins->dst->reg.idx[0].rel_addr)
5401     {
5402         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5403         if (!rel_addr)
5404             goto free;
5405         *rel_addr = *ins->dst->reg.idx[0].rel_addr;
5406         dst_param->reg.idx[0].rel_addr = rel_addr;
5407     }
5408     rec->ins.dst = dst_param;
5409 
5410     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5411     if (!src_param)
5412         goto free;
5413     for (i = 0; i < ins->src_count; ++i)
5414     {
5415         src_param[i] = ins->src[i];
5416         if (ins->src[i].reg.idx[0].rel_addr)
5417         {
5418             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5419             if (!rel_addr)
5420                 goto free;
5421             *rel_addr = *ins->src[i].reg.idx[0].rel_addr;
5422             src_param[i].reg.idx[0].rel_addr = rel_addr;
5423         }
5424     }
5425     rec->ins.src = src_param;
5426     list_add_tail(list, &rec->entry);
5427     return;
5428 
5429 free:
5430     ERR("Out of memory\n");
5431     if(dst_param)
5432     {
5433         HeapFree(GetProcessHeap(), 0, (void *)dst_param->reg.idx[0].rel_addr);
5434         HeapFree(GetProcessHeap(), 0, dst_param);
5435     }
5436     if(src_param)
5437     {
5438         for(i = 0; i < ins->src_count; i++)
5439         {
5440             HeapFree(GetProcessHeap(), 0, (void *)src_param[i].reg.idx[0].rel_addr);
5441         }
5442         HeapFree(GetProcessHeap(), 0, src_param);
5443     }
5444     HeapFree(GetProcessHeap(), 0, rec);
5445 }
5446 
5447 static void free_recorded_instruction(struct list *list)
5448 {
5449     struct recorded_instruction *rec_ins, *entry2;
5450     unsigned int i;
5451 
5452     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5453     {
5454         list_remove(&rec_ins->entry);
5455         if (rec_ins->ins.dst)
5456         {
5457             HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.dst->reg.idx[0].rel_addr);
5458             HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.dst);
5459         }
5460         if (rec_ins->ins.src)
5461         {
5462             for (i = 0; i < rec_ins->ins.src_count; ++i)
5463             {
5464                 HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.src[i].reg.idx[0].rel_addr);
5465             }
5466             HeapFree(GetProcessHeap(), 0, (void *)rec_ins->ins.src);
5467         }
5468         HeapFree(GetProcessHeap(), 0, rec_ins);
5469     }
5470 }
5471 
5472 static void pop_control_frame(const struct wined3d_shader_instruction *ins)
5473 {
5474     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5475     struct control_frame *control_frame;
5476 
5477     if (ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5478     {
5479         struct list *e = list_head(&priv->control_frames);
5480         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5481         list_remove(&control_frame->entry);
5482         HeapFree(GetProcessHeap(), 0, control_frame);
5483         priv->loop_depth--;
5484     }
5485     else if (ins->handler_idx == WINED3DSIH_ENDIF)
5486     {
5487         /* Non-ifc ENDIFs were already handled previously. */
5488         struct list *e = list_head(&priv->control_frames);
5489         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5490         list_remove(&control_frame->entry);
5491         HeapFree(GetProcessHeap(), 0, control_frame);
5492     }
5493 }
5494 
5495 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5496     SHADER_HANDLER hw_fct;
5497     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5498     const struct wined3d_shader *shader = ins->ctx->shader;
5499     struct control_frame *control_frame;
5500     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5501     BOOL bool_const;
5502 
5503     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5504     {
5505         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5506         list_add_head(&priv->control_frames, &control_frame->entry);
5507 
5508         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5509         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5510 
5511         if(priv->target_version >= NV2)
5512         {
5513             control_frame->no.loop = priv->num_loops++;
5514             priv->loop_depth++;
5515         }
5516         else
5517         {
5518             /* Don't bother recording when we're in a not used if branch */
5519             if(priv->muted)
5520             {
5521                 return;
5522             }
5523 
5524             if(!priv->recording)
5525             {
5526                 list_init(&priv->record);
5527                 priv->recording = TRUE;
5528                 control_frame->outer_loop = TRUE;
5529                 get_loop_control_const(ins, shader, ins->src[0].reg.idx[0].offset, &control_frame->loop_control);
5530                 return; /* Instruction is handled */
5531             }
5532             /* Record this loop in the outer loop's recording */
5533         }
5534     }
5535     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5536     {
5537         if(priv->target_version >= NV2)
5538         {
5539             /* Nothing to do. The control frame is popped after the HW instr handler */
5540         }
5541         else
5542         {
5543             struct list *e = list_head(&priv->control_frames);
5544             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5545             list_remove(&control_frame->entry);
5546 
5547             if(control_frame->outer_loop)
5548             {
5549                 unsigned int iteration;
5550                 int aL = 0;
5551                 struct list copy;
5552 
5553                 /* Turn off recording before playback */
5554                 priv->recording = FALSE;
5555 
5556                 /* Move the recorded instructions to a separate list and get them out of the private data
5557                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5558                  * be recorded again, thus priv->record might be overwritten
5559                  */
5560                 list_init(&copy);
5561                 list_move_tail(&copy, &priv->record);
5562                 list_init(&priv->record);
5563 
5564                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5565                 {
5566                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5567                                    control_frame->loop_control.count, control_frame->loop_control.start,
5568                                    control_frame->loop_control.step);
5569                     aL = control_frame->loop_control.start;
5570                 }
5571                 else
5572                 {
5573                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5574                 }
5575 
5576                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5577                 {
5578                     struct recorded_instruction *rec_ins;
5579                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5580                     {
5581                         priv->aL = aL;
5582                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5583                     }
5584                     else
5585                     {
5586                         shader_addline(buffer, "#Iteration %u\n", iteration);
5587                     }
5588 
5589                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5590                     {
5591                         shader_arb_handle_instruction(&rec_ins->ins);
5592                     }
5593 
5594                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5595                     {
5596                         aL += control_frame->loop_control.step;
5597                     }
5598                 }
5599                 shader_addline(buffer, "#end loop/rep\n");
5600 
5601                 free_recorded_instruction(&copy);
5602                 HeapFree(GetProcessHeap(), 0, control_frame);
5603                 return; /* Instruction is handled */
5604             }
5605             else
5606             {
5607                 /* This is a nested loop. Proceed to the normal recording function */
5608                 HeapFree(GetProcessHeap(), 0, control_frame);
5609             }
5610         }
5611     }
5612 
5613     if(priv->recording)
5614     {
5615         record_instruction(&priv->record, ins);
5616         return;
5617     }
5618 
5619     /* boolean if */
5620     if(ins->handler_idx == WINED3DSIH_IF)
5621     {
5622         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5623         list_add_head(&priv->control_frames, &control_frame->entry);
5624         control_frame->type = IF;
5625 
5626         bool_const = get_bool_const(ins, shader, ins->src[0].reg.idx[0].offset);
5627         if (ins->src[0].modifiers == WINED3DSPSM_NOT)
5628             bool_const = !bool_const;
5629         if (!priv->muted && !bool_const)
5630         {
5631             shader_addline(buffer, "#if(FALSE){\n");
5632             priv->muted = TRUE;
5633             control_frame->muting = TRUE;
5634         }
5635         else shader_addline(buffer, "#if(TRUE) {\n");
5636 
5637         return; /* Instruction is handled */
5638     }
5639     else if(ins->handler_idx == WINED3DSIH_IFC)
5640     {
5641         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5642         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5643         control_frame->type = IFC;
5644         control_frame->no.ifc = priv->num_ifcs++;
5645         list_add_head(&priv->control_frames, &control_frame->entry);
5646     }
5647     else if(ins->handler_idx == WINED3DSIH_ELSE)
5648     {
5649         struct list *e = list_head(&priv->control_frames);
5650         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5651 
5652         if(control_frame->type == IF)
5653         {
5654             shader_addline(buffer, "#} else {\n");
5655             if(!priv->muted && !control_frame->muting)
5656             {
5657                 priv->muted = TRUE;
5658                 control_frame->muting = TRUE;
5659             }
5660             else if(control_frame->muting) priv->muted = FALSE;
5661             return; /* Instruction is handled. */
5662         }
5663         /* In case of an ifc, generate a HW shader instruction */
5664         if (control_frame->type != IFC)
5665             ERR("Control frame does not match.\n");
5666     }
5667     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5668     {
5669         struct list *e = list_head(&priv->control_frames);
5670         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5671 
5672         if(control_frame->type == IF)
5673         {
5674             shader_addline(buffer, "#} endif\n");
5675             if(control_frame->muting) priv->muted = FALSE;
5676             list_remove(&control_frame->entry);
5677             HeapFree(GetProcessHeap(), 0, control_frame);
5678             return; /* Instruction is handled */
5679         }
5680         /* In case of an ifc, generate a HW shader instruction */
5681         if (control_frame->type != IFC)
5682             ERR("Control frame does not match.\n");
5683     }
5684 
5685     if(priv->muted)
5686     {
5687         pop_control_frame(ins);
5688         return;
5689     }
5690 
5691     /* Select handler */
5692     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5693 
5694     /* Unhandled opcode */
5695     if (!hw_fct)
5696     {
5697         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5698         return;
5699     }
5700     hw_fct(ins);
5701 
5702     pop_control_frame(ins);
5703 
5704     shader_arb_add_instruction_modifiers(ins);
5705 }
5706 
5707 static BOOL shader_arb_has_ffp_proj_control(void *shader_priv)
5708 {
5709     struct shader_arb_priv *priv = shader_priv;
5710 
5711     return priv->ffp_proj_control;
5712 }
5713 
5714 const struct wined3d_shader_backend_ops arb_program_shader_backend =
5715 {
5716     shader_arb_handle_instruction,
5717     shader_arb_select,
5718     shader_arb_disable,
5719     shader_arb_select_depth_blt,
5720     shader_arb_deselect_depth_blt,
5721     shader_arb_update_float_vertex_constants,
5722     shader_arb_update_float_pixel_constants,
5723     shader_arb_load_constants,
5724     shader_arb_destroy,
5725     shader_arb_alloc,
5726     shader_arb_free,
5727     shader_arb_allocate_context_data,
5728     shader_arb_free_context_data,
5729     shader_arb_get_caps,
5730     shader_arb_color_fixup_supported,
5731     shader_arb_has_ffp_proj_control,
5732 };
5733 
5734 /* ARB_fragment_program fixed function pipeline replacement definitions */
5735 #define ARB_FFP_CONST_TFACTOR           0
5736 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5737 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5738 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5739 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5740 
5741 struct arbfp_ffp_desc
5742 {
5743     struct ffp_frag_desc parent;
5744     GLuint shader;
5745 };
5746 
5747 /* Context activation is done by the caller. */
5748 static void arbfp_enable(const struct wined3d_gl_info *gl_info, BOOL enable)
5749 {
5750     if (enable)
5751     {
5752         gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
5753         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5754     }
5755     else
5756     {
5757         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
5758         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5759     }
5760 }
5761 
5762 static void *arbfp_alloc(const struct wined3d_shader_backend_ops *shader_backend, void *shader_priv)
5763 {
5764     struct shader_arb_priv *priv;
5765 
5766     /* Share private data between the shader backend and the pipeline
5767      * replacement, if both are the arb implementation. This is needed to
5768      * figure out whether ARBfp should be disabled if no pixel shader is bound
5769      * or not. */
5770     if (shader_backend == &arb_program_shader_backend)
5771         priv = shader_priv;
5772     else if (!(priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv))))
5773         return NULL;
5774 
5775     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5776     {
5777         ERR("Failed to initialize rbtree.\n");
5778         if (priv != shader_priv)
5779             HeapFree(GetProcessHeap(), 0, priv);
5780         return NULL;
5781     }
5782     priv->use_arbfp_fixed_func = TRUE;
5783 
5784     return priv;
5785 }
5786 
5787 /* Context activation is done by the caller. */
5788 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5789 {
5790     const struct wined3d_gl_info *gl_info = context;
5791     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5792 
5793     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5794     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5795     HeapFree(GetProcessHeap(), 0, entry_arb);
5796 }
5797 
5798 /* Context activation is done by the caller. */
5799 static void arbfp_free(struct wined3d_device *device)
5800 {
5801     struct shader_arb_priv *priv = device->fragment_priv;
5802 
5803     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5804     priv->use_arbfp_fixed_func = FALSE;
5805 
5806     if (device->shader_backend != &arb_program_shader_backend)
5807     {
5808         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5809     }
5810 }
5811 
5812 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5813 {
5814     caps->wined3d_caps = WINED3D_FRAGMENT_CAP_PROJ_CONTROL
5815             | WINED3D_FRAGMENT_CAP_SRGB_WRITE;
5816     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5817     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5818                            WINED3DTEXOPCAPS_SELECTARG1                  |
5819                            WINED3DTEXOPCAPS_SELECTARG2                  |
5820                            WINED3DTEXOPCAPS_MODULATE4X                  |
5821                            WINED3DTEXOPCAPS_MODULATE2X                  |
5822                            WINED3DTEXOPCAPS_MODULATE                    |
5823                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5824                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5825                            WINED3DTEXOPCAPS_ADD                         |
5826                            WINED3DTEXOPCAPS_SUBTRACT                    |
5827                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5828                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5829                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5830                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5831                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5832                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5833                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5834                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5835                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5836                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5837                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5838                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5839                            WINED3DTEXOPCAPS_LERP                        |
5840                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5841                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5842 
5843     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5844 
5845     caps->MaxTextureBlendStages   = 8;
5846     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5847 }
5848 
5849 static void state_texfactor_arbfp(struct wined3d_context *context,
5850         const struct wined3d_state *state, DWORD state_id)
5851 {
5852     struct wined3d_device *device = context->swapchain->device;
5853     const struct wined3d_gl_info *gl_info = context->gl_info;
5854     float col[4];
5855 
5856     /* Don't load the parameter if we're using an arbfp pixel shader,
5857      * otherwise we'll overwrite application provided constants. */
5858     if (device->shader_backend == &arb_program_shader_backend)
5859     {
5860         struct shader_arb_priv *priv;
5861 
5862         if (use_ps(state)) return;
5863 
5864         priv = device->shader_priv;
5865         priv->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5866         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5867     }
5868 
5869     D3DCOLORTOGLFLOAT4(state->render_states[WINED3D_RS_TEXTUREFACTOR], col);
5870     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5871     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5872 }
5873 
5874 static void state_arb_specularenable(struct wined3d_context *context,
5875         const struct wined3d_state *state, DWORD state_id)
5876 {
5877     struct wined3d_device *device = context->swapchain->device;
5878     const struct wined3d_gl_info *gl_info = context->gl_info;
5879     float col[4];
5880 
5881     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5882      * application provided constants
5883      */
5884     if (device->shader_backend == &arb_program_shader_backend)
5885     {
5886         struct shader_arb_priv *priv;
5887 
5888         if (use_ps(state)) return;
5889 
5890         priv = device->shader_priv;
5891         priv->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5892         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5893     }
5894 
5895     if (state->render_states[WINED3D_RS_SPECULARENABLE])
5896     {
5897         /* The specular color has no alpha */
5898         col[0] = 1.0f; col[1] = 1.0f;
5899         col[2] = 1.0f; col[3] = 0.0f;
5900     } else {
5901         col[0] = 0.0f; col[1] = 0.0f;
5902         col[2] = 0.0f; col[3] = 0.0f;
5903     }
5904     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5905     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5906 }
5907 
5908 static void set_bumpmat_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5909 {
5910     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5911     struct wined3d_device *device = context->swapchain->device;
5912     const struct wined3d_gl_info *gl_info = context->gl_info;
5913     float mat[2][2];
5914 
5915     context->constant_update_mask |= WINED3D_SHADER_CONST_PS_BUMP_ENV;
5916 
5917     if (device->shader_backend == &arb_program_shader_backend)
5918     {
5919         struct shader_arb_priv *priv = device->shader_priv;
5920 
5921         /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants. */
5922         if (use_ps(state))
5923             return;
5924 
5925         priv->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5926         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5927     }
5928 
5929     mat[0][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT00]);
5930     mat[0][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT01]);
5931     mat[1][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT10]);
5932     mat[1][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT11]);
5933 
5934     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5935     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5936 }
5937 
5938 static void tex_bumpenvlum_arbfp(struct wined3d_context *context,
5939         const struct wined3d_state *state, DWORD state_id)
5940 {
5941     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5942     struct wined3d_device *device = context->swapchain->device;
5943     const struct wined3d_gl_info *gl_info = context->gl_info;
5944     float param[4];
5945 
5946     context->constant_update_mask |= WINED3D_SHADER_CONST_PS_BUMP_ENV;
5947 
5948     if (device->shader_backend == &arb_program_shader_backend)
5949     {
5950         struct shader_arb_priv *priv = device->shader_priv;
5951 
5952         /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants. */
5953         if (use_ps(state))
5954             return;
5955 
5956         priv->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5957         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5958     }
5959 
5960     param[0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LSCALE]);
5961     param[1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LOFFSET]);
5962     param[2] = 0.0f;
5963     param[3] = 0.0f;
5964 
5965     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5966     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5967 }
5968 
5969 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5970 {
5971     const char *ret;
5972 
5973     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5974 
5975     switch(arg & WINED3DTA_SELECTMASK) {
5976         case WINED3DTA_DIFFUSE:
5977             ret = "fragment.color.primary"; break;
5978 
5979         case WINED3DTA_CURRENT:
5980             if (!stage) ret = "fragment.color.primary";
5981             else ret = "ret";
5982             break;
5983 
5984         case WINED3DTA_TEXTURE:
5985             switch(stage) {
5986                 case 0: ret = "tex0"; break;
5987                 case 1: ret = "tex1"; break;
5988                 case 2: ret = "tex2"; break;
5989                 case 3: ret = "tex3"; break;
5990                 case 4: ret = "tex4"; break;
5991                 case 5: ret = "tex5"; break;
5992                 case 6: ret = "tex6"; break;
5993                 case 7: ret = "tex7"; break;
5994                 default: ret = "unknown texture";
5995             }
5996             break;
5997 
5998         case WINED3DTA_TFACTOR:
5999             ret = "tfactor"; break;
6000 
6001         case WINED3DTA_SPECULAR:
6002             ret = "fragment.color.secondary"; break;
6003 
6004         case WINED3DTA_TEMP:
6005             ret = "tempreg"; break;
6006 
6007         case WINED3DTA_CONSTANT:
6008             FIXME("Implement perstage constants\n");
6009             switch(stage) {
6010                 case 0: ret = "const0"; break;
6011                 case 1: ret = "const1"; break;
6012                 case 2: ret = "const2"; break;
6013                 case 3: ret = "const3"; break;
6014                 case 4: ret = "const4"; break;
6015                 case 5: ret = "const5"; break;
6016                 case 6: ret = "const6"; break;
6017                 case 7: ret = "const7"; break;
6018                 default: ret = "unknown constant";
6019             }
6020             break;
6021 
6022         default:
6023             return "unknown";
6024     }
6025 
6026     if(arg & WINED3DTA_COMPLEMENT) {
6027         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
6028         if(argnum == 0) ret = "arg0";
6029         if(argnum == 1) ret = "arg1";
6030         if(argnum == 2) ret = "arg2";
6031     }
6032     if(arg & WINED3DTA_ALPHAREPLICATE) {
6033         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
6034         if(argnum == 0) ret = "arg0";
6035         if(argnum == 1) ret = "arg1";
6036         if(argnum == 2) ret = "arg2";
6037     }
6038     return ret;
6039 }
6040 
6041 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
6042         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
6043 {
6044     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
6045     unsigned int mul = 1;
6046 
6047     if(color && alpha) dstmask = "";
6048     else if(color) dstmask = ".xyz";
6049     else dstmask = ".w";
6050 
6051     if(dst == tempreg) dstreg = "tempreg";
6052     else dstreg = "ret";
6053 
6054     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
6055     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
6056     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
6057 
6058     switch (op)
6059     {
6060         case WINED3D_TOP_DISABLE:
6061             if (!stage)
6062                 shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
6063             break;
6064 
6065         case WINED3D_TOP_SELECT_ARG2:
6066             arg1 = arg2;
6067             /* FALLTHROUGH */
6068         case WINED3D_TOP_SELECT_ARG1:
6069             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
6070             break;
6071 
6072         case WINED3D_TOP_MODULATE_4X:
6073             mul = 2;
6074             /* FALLTHROUGH */
6075         case WINED3D_TOP_MODULATE_2X:
6076             mul *= 2;
6077             /* FALLTHROUGH */
6078         case WINED3D_TOP_MODULATE:
6079             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6080             break;
6081 
6082         case WINED3D_TOP_ADD_SIGNED_2X:
6083             mul = 2;
6084             /* FALLTHROUGH */
6085         case WINED3D_TOP_ADD_SIGNED:
6086             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6087             arg2 = "arg2";
6088             /* FALLTHROUGH */
6089         case WINED3D_TOP_ADD:
6090             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6091             break;
6092 
6093         case WINED3D_TOP_SUBTRACT:
6094             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6095             break;
6096 
6097         case WINED3D_TOP_ADD_SMOOTH:
6098             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
6099             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6100             break;
6101 
6102         case WINED3D_TOP_BLEND_CURRENT_ALPHA:
6103             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
6104             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6105             break;
6106         case WINED3D_TOP_BLEND_FACTOR_ALPHA:
6107             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
6108             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6109             break;
6110         case WINED3D_TOP_BLEND_TEXTURE_ALPHA:
6111             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
6112             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6113             break;
6114         case WINED3D_TOP_BLEND_DIFFUSE_ALPHA:
6115             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
6116             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6117             break;
6118 
6119         case WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM:
6120             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
6121             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
6122             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
6123             break;
6124 
6125         /* D3DTOP_PREMODULATE ???? */
6126 
6127         case WINED3D_TOP_MODULATE_INVALPHA_ADD_COLOR:
6128             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
6129             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6130             break;
6131         case WINED3D_TOP_MODULATE_ALPHA_ADD_COLOR:
6132             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
6133             break;
6134         case WINED3D_TOP_MODULATE_INVCOLOR_ADD_ALPHA:
6135             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
6136             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
6137             break;
6138         case WINED3D_TOP_MODULATE_COLOR_ADD_ALPHA:
6139             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
6140             break;
6141 
6142         case WINED3D_TOP_DOTPRODUCT3:
6143             mul = 4;
6144             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
6145             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6146             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
6147             break;
6148 
6149         case WINED3D_TOP_MULTIPLY_ADD:
6150             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
6151             break;
6152 
6153         case WINED3D_TOP_LERP:
6154             /* The msdn is not quite right here */
6155             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6156             break;
6157 
6158         case WINED3D_TOP_BUMPENVMAP:
6159         case WINED3D_TOP_BUMPENVMAP_LUMINANCE:
6160             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
6161             break;
6162 
6163         default:
6164             FIXME("Unhandled texture op %08x\n", op);
6165     }
6166 
6167     if (mul == 2)
6168         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", dstreg, dstmask, dstreg);
6169     else if (mul == 4)
6170         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", dstreg, dstmask, dstreg);
6171 }
6172 
6173 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, const struct wined3d_gl_info *gl_info)
6174 {
6175     unsigned int stage;
6176     struct wined3d_shader_buffer buffer;
6177     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6178     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6179     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
6180     UINT lowest_disabled_stage;
6181     const char *textype;
6182     const char *instr;
6183     char colorcor_dst[8];
6184     GLuint ret;
6185     DWORD arg0, arg1, arg2;
6186     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
6187     BOOL op_equal;
6188     const char *final_combiner_src = "ret";
6189     GLint pos;
6190     BOOL custom_linear_fog = FALSE;
6191 
6192     /* Find out which textures are read */
6193     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6194     {
6195         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6196             break;
6197         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
6198         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
6199         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
6200         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6201         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6202         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6203 
6204         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA)
6205             tex_read[stage] = TRUE;
6206         if (settings->op[stage].cop == WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM)
6207             tex_read[stage] = TRUE;
6208         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP)
6209         {
6210             bump_used[stage] = TRUE;
6211             tex_read[stage] = TRUE;
6212         }
6213         if (settings->op[stage].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6214         {
6215             bump_used[stage] = TRUE;
6216             tex_read[stage] = TRUE;
6217             luminance_used[stage] = TRUE;
6218         }
6219         else if (settings->op[stage].cop == WINED3D_TOP_BLEND_FACTOR_ALPHA)
6220         {
6221             tfactor_used = TRUE;
6222         }
6223 
6224         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6225             tfactor_used = TRUE;
6226         }
6227 
6228         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
6229         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6230             tempreg_used = TRUE;
6231         }
6232 
6233         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6234             continue;
6235         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
6236         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
6237         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
6238         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6239         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6240         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
6241 
6242         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
6243             tempreg_used = TRUE;
6244         }
6245         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
6246             tfactor_used = TRUE;
6247         }
6248     }
6249     lowest_disabled_stage = stage;
6250 
6251     /* Shader header */
6252     if (!shader_buffer_init(&buffer))
6253     {
6254         ERR("Failed to initialize shader buffer.\n");
6255         return 0;
6256     }
6257 
6258     shader_addline(&buffer, "!!ARBfp1.0\n");
6259 
6260     switch (settings->fog)
6261     {
6262         case WINED3D_FFP_PS_FOG_OFF:                                                         break;
6263         case WINED3D_FFP_PS_FOG_LINEAR:
6264             if (gl_info->quirks & WINED3D_QUIRK_BROKEN_ARB_FOG)
6265             {
6266                 custom_linear_fog = TRUE;
6267                 break;
6268             }
6269             shader_addline(&buffer, "OPTION ARB_fog_linear;\n");
6270             break;
6271 
6272         case WINED3D_FFP_PS_FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
6273         case WINED3D_FFP_PS_FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
6274         default: FIXME("Unexpected fog setting %d\n", settings->fog);
6275     }
6276 
6277     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
6278     shader_addline(&buffer, "TEMP TMP;\n");
6279     shader_addline(&buffer, "TEMP ret;\n");
6280     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
6281     shader_addline(&buffer, "TEMP arg0;\n");
6282     shader_addline(&buffer, "TEMP arg1;\n");
6283     shader_addline(&buffer, "TEMP arg2;\n");
6284     for(stage = 0; stage < MAX_TEXTURES; stage++) {
6285         if(!tex_read[stage]) continue;
6286         shader_addline(&buffer, "TEMP tex%u;\n", stage);
6287         if(!bump_used[stage]) continue;
6288         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
6289         if(!luminance_used[stage]) continue;
6290         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
6291     }
6292     if(tfactor_used) {
6293         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
6294     }
6295     shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6296 
6297     if (settings->sRGB_write)
6298     {
6299         shader_addline(&buffer, "PARAM srgb_consts0 = ");
6300         shader_arb_append_imm_vec4(&buffer, wined3d_srgb_const0);
6301         shader_addline(&buffer, ";\n");
6302         shader_addline(&buffer, "PARAM srgb_consts1 = ");
6303         shader_arb_append_imm_vec4(&buffer, wined3d_srgb_const1);
6304         shader_addline(&buffer, ";\n");
6305     }
6306 
6307     if (lowest_disabled_stage < 7 && settings->emul_clipplanes)
6308         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6309 
6310     /* Generate texture sampling instructions) */
6311     for (stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3D_TOP_DISABLE; ++stage)
6312     {
6313         if (!tex_read[stage])
6314             continue;
6315 
6316         switch(settings->op[stage].tex_type) {
6317             case tex_1d:                    textype = "1D";     break;
6318             case tex_2d:                    textype = "2D";     break;
6319             case tex_3d:                    textype = "3D";     break;
6320             case tex_cube:                  textype = "CUBE";   break;
6321             case tex_rect:                  textype = "RECT";   break;
6322             default: textype = "unexpected_textype";   break;
6323         }
6324 
6325         if(settings->op[stage].projected == proj_none) {
6326             instr = "TEX";
6327         } else if(settings->op[stage].projected == proj_count4 ||
6328                   settings->op[stage].projected == proj_count3) {
6329             instr = "TXP";
6330         } else {
6331             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6332             instr = "TXP";
6333         }
6334 
6335         if (stage > 0
6336                 && (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP
6337                 || settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE))
6338         {
6339             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6340             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6341             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6342             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6343 
6344             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6345              * so multiply the displacement with the dividing parameter before passing it to TXP
6346              */
6347             if (settings->op[stage].projected != proj_none) {
6348                 if(settings->op[stage].projected == proj_count4) {
6349                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6350                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6351                 } else {
6352                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6353                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6354                 }
6355             } else {
6356                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6357             }
6358 
6359             shader_addline(&buffer, "%s tex%u, ret, texture[%u], %s;\n",
6360                     instr, stage, stage, textype);
6361             if (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6362             {
6363                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6364                                stage - 1, stage - 1, stage - 1);
6365                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6366             }
6367         } else if(settings->op[stage].projected == proj_count3) {
6368             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6369             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6370             shader_addline(&buffer, "%s tex%u, ret, texture[%u], %s;\n",
6371                             instr, stage, stage, textype);
6372         } else {
6373             shader_addline(&buffer, "%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6374                             instr, stage, stage, stage, textype);
6375         }
6376 
6377         sprintf(colorcor_dst, "tex%u", stage);
6378         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6379                 settings->op[stage].color_fixup);
6380     }
6381 
6382     /* Generate the main shader */
6383     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6384     {
6385         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6386         {
6387             if (!stage)
6388                 final_combiner_src = "fragment.color.primary";
6389             break;
6390         }
6391 
6392         if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6393                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6394             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6395         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6396                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6397             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6398         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6399                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6400             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6401         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6402                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6403             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6404         else
6405             op_equal = settings->op[stage].aop   == settings->op[stage].cop
6406                     && settings->op[stage].carg0 == settings->op[stage].aarg0
6407                     && settings->op[stage].carg1 == settings->op[stage].aarg1
6408                     && settings->op[stage].carg2 == settings->op[stage].aarg2;
6409 
6410         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6411         {
6412             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6413                           settings->op[stage].cop, settings->op[stage].carg0,
6414                           settings->op[stage].carg1, settings->op[stage].carg2);
6415             if (!stage)
6416                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6417         }
6418         else if (op_equal)
6419         {
6420             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6421                           settings->op[stage].cop, settings->op[stage].carg0,
6422                           settings->op[stage].carg1, settings->op[stage].carg2);
6423         } else {
6424             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6425                           settings->op[stage].cop, settings->op[stage].carg0,
6426                           settings->op[stage].carg1, settings->op[stage].carg2);
6427             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6428                           settings->op[stage].aop, settings->op[stage].aarg0,
6429                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6430         }
6431     }
6432 
6433     if (settings->sRGB_write || custom_linear_fog)
6434     {
6435         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6436         if (settings->sRGB_write)
6437             arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6438         if (custom_linear_fog)
6439             arbfp_add_linear_fog(&buffer, "ret", "arg0");
6440         shader_addline(&buffer, "MOV result.color, ret;\n");
6441     }
6442     else
6443     {
6444         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n",
6445                 final_combiner_src);
6446     }
6447 
6448     /* Footer */
6449     shader_addline(&buffer, "END\n");
6450 
6451     /* Generate the shader */
6452     GL_EXTCALL(glGenProgramsARB(1, &ret));
6453     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6454     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6455             strlen(buffer.buffer), buffer.buffer));
6456     checkGLcall("glProgramStringARB()");
6457 
6458     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6459     if (pos != -1)
6460     {
6461         FIXME("Fragment program error at position %d: %s\n\n", pos,
6462               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6463         shader_arb_dump_program_source(buffer.buffer);
6464     }
6465     else
6466     {
6467         GLint native;
6468 
6469         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6470         checkGLcall("glGetProgramivARB()");
6471         if (!native) WARN("Program exceeds native resource limits.\n");
6472     }
6473 
6474     shader_buffer_free(&buffer);
6475     return ret;
6476 }
6477 
6478 static void fragment_prog_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6479 {
6480     const struct wined3d_device *device = context->swapchain->device;
6481     const struct wined3d_gl_info *gl_info = context->gl_info;
6482     struct shader_arb_priv *priv = device->fragment_priv;
6483     BOOL use_pshader = use_ps(state);
6484     struct ffp_frag_settings settings;
6485     const struct arbfp_ffp_desc *desc;
6486     unsigned int i;
6487 
6488     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6489 
6490     if (isStateDirty(context, STATE_RENDER(WINED3D_RS_FOGENABLE)))
6491     {
6492         if (!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6493         {
6494             /* Reload fixed function constants since they collide with the
6495              * pixel shader constants. */
6496             for (i = 0; i < MAX_TEXTURES; ++i)
6497             {
6498                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6499             }
6500             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6501             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6502         }
6503         else if (use_pshader)
6504         {
6505             context->shader_update_mask |= 1 << WINED3D_SHADER_TYPE_PIXEL;
6506         }
6507         return;
6508     }
6509 
6510     if (!use_pshader)
6511     {
6512         /* Find or create a shader implementing the fixed function pipeline
6513          * settings, then activate it. */
6514         gen_ffp_frag_op(context, state, &settings, FALSE);
6515         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6516         if(!desc) {
6517             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
6518             if (!new_desc)
6519             {
6520                 ERR("Out of memory\n");
6521                 return;
6522             }
6523 
6524             new_desc->parent.settings = settings;
6525             new_desc->shader = gen_arbfp_ffp_shader(&settings, gl_info);
6526             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6527             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6528             desc = new_desc;
6529         }
6530 
6531         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active (however, note the
6532          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6533          * deactivate it.
6534          */
6535         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6536         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6537         priv->current_fprogram_id = desc->shader;
6538 
6539         if (device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6540         {
6541             /* Reload fixed function constants since they collide with the
6542              * pixel shader constants. */
6543             for (i = 0; i < MAX_TEXTURES; ++i)
6544             {
6545                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6546             }
6547             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6548             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6549         }
6550         context->last_was_pshader = FALSE;
6551     } else {
6552         context->last_was_pshader = TRUE;
6553     }
6554 
6555     context->shader_update_mask |= 1 << WINED3D_SHADER_TYPE_PIXEL;
6556 }
6557 
6558 /* We can't link the fog states to the fragment state directly since the
6559  * vertex pipeline links them to FOGENABLE. A different linking in different
6560  * pipeline parts can't be expressed in the combined state table, so we need
6561  * to handle that with a forwarding function. The other invisible side effect
6562  * is that changing the fog start and fog end (which links to FOGENABLE in
6563  * vertex) results in the fragment_prog_arbfp function being called because
6564  * FOGENABLE is dirty, which calls this function here. */
6565 static void state_arbfp_fog(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6566 {
6567     enum fogsource new_source;
6568     DWORD fogstart = state->render_states[WINED3D_RS_FOGSTART];
6569     DWORD fogend = state->render_states[WINED3D_RS_FOGEND];
6570 
6571     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6572 
6573     if (!isStateDirty(context, STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL)))
6574         fragment_prog_arbfp(context, state, state_id);
6575 
6576     if (!state->render_states[WINED3D_RS_FOGENABLE])
6577         return;
6578 
6579     if (state->render_states[WINED3D_RS_FOGTABLEMODE] == WINED3D_FOG_NONE)
6580     {
6581         if (use_vs(state))
6582         {
6583             new_source = FOGSOURCE_VS;
6584         }
6585         else
6586         {
6587             if (state->render_states[WINED3D_RS_FOGVERTEXMODE] == WINED3D_FOG_NONE || context->last_was_rhw)
6588                 new_source = FOGSOURCE_COORD;
6589             else
6590                 new_source = FOGSOURCE_FFP;
6591         }
6592     }
6593     else
6594     {
6595         new_source = FOGSOURCE_FFP;
6596     }
6597 
6598     if (new_source != context->fog_source || fogstart == fogend)
6599     {
6600         context->fog_source = new_source;
6601         state_fogstartend(context, state, STATE_RENDER(WINED3D_RS_FOGSTART));
6602     }
6603 }
6604 
6605 static void textransform(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6606 {
6607     if (!isStateDirty(context, STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL)))
6608         fragment_prog_arbfp(context, state, state_id);
6609 }
6610 
6611 static const struct StateEntryTemplate arbfp_fragmentstate_template[] =
6612 {
6613     {STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),              { STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),             state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6614     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6615     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6616     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6617     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6618     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6619     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6620     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6621     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6622     {STATE_TEXTURESTAGE(0, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6623     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6624     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6625     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6626     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6627     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6628     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6629     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6630     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6631     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6632     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6633     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6634     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6635     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6636     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6637     {STATE_TEXTURESTAGE(1, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6638     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6639     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6640     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6641     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6642     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6643     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6644     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6645     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6646     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6647     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6648     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6649     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6650     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6651     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6652     {STATE_TEXTURESTAGE(2, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6653     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6654     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6655     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6656     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6657     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6658     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6659     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6660     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6661     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6662     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6663     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6664     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6665     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6666     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6667     {STATE_TEXTURESTAGE(3, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6668     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6669     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6670     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6671     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6672     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6673     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6674     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6675     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6676     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6677     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6678     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6679     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6680     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6681     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6682     {STATE_TEXTURESTAGE(4, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6683     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6684     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6685     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6686     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6687     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6688     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6689     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6690     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6691     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6692     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6693     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6694     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6695     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6696     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6697     {STATE_TEXTURESTAGE(5, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6698     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6699     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6700     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6701     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6702     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6703     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6704     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6705     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6706     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6707     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6708     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6709     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6710     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6711     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6712     {STATE_TEXTURESTAGE(6, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6713     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6714     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6715     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6716     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6717     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6718     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6719     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6720     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6721     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6722     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6723     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6724     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6725     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6726     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6727     {STATE_TEXTURESTAGE(7, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6728     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6729     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6730     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6731     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6732     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6733     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6734     {STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),             { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6735     {STATE_RENDER(WINED3D_RS_FOGENABLE),                  { STATE_RENDER(WINED3D_RS_FOGENABLE),                 state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6736     {STATE_RENDER(WINED3D_RS_FOGTABLEMODE),               { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6737     {STATE_RENDER(WINED3D_RS_FOGVERTEXMODE),              { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6738     {STATE_RENDER(WINED3D_RS_FOGSTART),                   { STATE_RENDER(WINED3D_RS_FOGSTART),                  state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6739     {STATE_RENDER(WINED3D_RS_FOGEND),                     { STATE_RENDER(WINED3D_RS_FOGSTART),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6740     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),           state_srgbwrite         }, ARB_FRAMEBUFFER_SRGB            },
6741     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6742     {STATE_RENDER(WINED3D_RS_FOGCOLOR),                   { STATE_RENDER(WINED3D_RS_FOGCOLOR),                  state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6743     {STATE_RENDER(WINED3D_RS_FOGDENSITY),                 { STATE_RENDER(WINED3D_RS_FOGDENSITY),                state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6744     {STATE_TEXTURESTAGE(0,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(0, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6745     {STATE_TEXTURESTAGE(1,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(1, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6746     {STATE_TEXTURESTAGE(2,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(2, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6747     {STATE_TEXTURESTAGE(3,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(3, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6748     {STATE_TEXTURESTAGE(4,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(4, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6749     {STATE_TEXTURESTAGE(5,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(5, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6750     {STATE_TEXTURESTAGE(6,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(6, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6751     {STATE_TEXTURESTAGE(7,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(7, WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6752     {STATE_RENDER(WINED3D_RS_SPECULARENABLE),             { STATE_RENDER(WINED3D_RS_SPECULARENABLE),            state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6753     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6754 };
6755 
6756 const struct fragment_pipeline arbfp_fragment_pipeline = {
6757     arbfp_enable,
6758     arbfp_get_caps,
6759     arbfp_alloc,
6760     arbfp_free,
6761     shader_arb_color_fixup_supported,
6762     arbfp_fragmentstate_template,
6763 };
6764 
6765 struct arbfp_blit_priv {
6766     GLenum yuy2_rect_shader, yuy2_2d_shader;
6767     GLenum uyvy_rect_shader, uyvy_2d_shader;
6768     GLenum yv12_rect_shader, yv12_2d_shader;
6769     GLenum p8_rect_shader, p8_2d_shader;
6770     GLuint palette_texture;
6771 };
6772 
6773 static HRESULT arbfp_blit_alloc(struct wined3d_device *device)
6774 {
6775     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6776     if(!device->blit_priv) {
6777         ERR("Out of memory\n");
6778         return E_OUTOFMEMORY;
6779     }
6780     return WINED3D_OK;
6781 }
6782 
6783 /* Context activation is done by the caller. */
6784 static void arbfp_blit_free(struct wined3d_device *device)
6785 {
6786     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6787     struct arbfp_blit_priv *priv = device->blit_priv;
6788 
6789     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6790     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6791     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6792     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6793     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6794     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6795     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6796     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6797     checkGLcall("Delete yuv and p8 programs");
6798 
6799     if (priv->palette_texture)
6800         gl_info->gl_ops.gl.p_glDeleteTextures(1, &priv->palette_texture);
6801 
6802     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6803     device->blit_priv = NULL;
6804 }
6805 
6806 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6807         GLenum textype, char *luminance)
6808 {
6809     char chroma;
6810     const char *tex, *texinstr;
6811 
6812     if (fixup == COMPLEX_FIXUP_UYVY) {
6813         chroma = 'x';
6814         *luminance = 'w';
6815     } else {
6816         chroma = 'w';
6817         *luminance = 'x';
6818     }
6819     switch(textype) {
6820         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6821         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6822         default:
6823             /* This is more tricky than just replacing the texture type - we have to navigate
6824              * properly in the texture to find the correct chroma values
6825              */
6826             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6827             return FALSE;
6828     }
6829 
6830     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6831      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6832      * filtering when we sample the texture.
6833      *
6834      * These are the rules for reading the chroma:
6835      *
6836      * Even pixel: Cr
6837      * Even pixel: U
6838      * Odd pixel: V
6839      *
6840      * So we have to get the sampling x position in non-normalized coordinates in integers
6841      */
6842     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6843         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6844         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6845     } else {
6846         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6847     }
6848     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6849      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6850      * 0.5, so add 0.5.
6851      */
6852     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6853     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6854 
6855     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6856      * even and odd pixels respectively
6857      */
6858     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6859     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6860 
6861     /* Sample Pixel 1 */
6862     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6863 
6864     /* Put the value into either of the chroma values */
6865     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6866     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6867     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6868     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6869 
6870     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6871      * the pixel right to the current one. Otherwise, sample the left pixel.
6872      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6873      */
6874     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6875     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6876     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6877 
6878     /* Put the value into the other chroma */
6879     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6880     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6881     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6882     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6883 
6884     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6885      * the current one and lerp the two U and V values
6886      */
6887 
6888     /* This gives the correctly filtered luminance value */
6889     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6890 
6891     return TRUE;
6892 }
6893 
6894 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6895 {
6896     const char *tex;
6897     static const float yv12_coef[]
6898             = {2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f};
6899 
6900     switch(textype) {
6901         case GL_TEXTURE_2D:             tex = "2D";     break;
6902         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6903         default:
6904             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6905             return FALSE;
6906     }
6907 
6908     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6909      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6910      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6911      * pitch of the luminance plane, the packing into the gl texture is a bit
6912      * unfortunate. If the whole texture is interpreted as luminance data it looks
6913      * approximately like this:
6914      *
6915      *        +----------------------------------+----
6916      *        |                                  |
6917      *        |                                  |
6918      *        |                                  |
6919      *        |                                  |
6920      *        |                                  |   2
6921      *        |            LUMINANCE             |   -
6922      *        |                                  |   3
6923      *        |                                  |
6924      *        |                                  |
6925      *        |                                  |
6926      *        |                                  |
6927      *        +----------------+-----------------+----
6928      *        |                |                 |
6929      *        |  U even rows   |  U odd rows     |
6930      *        |                |                 |   1
6931      *        +----------------+------------------   -
6932      *        |                |                 |   3
6933      *        |  V even rows   |  V odd rows     |
6934      *        |                |                 |
6935      *        +----------------+-----------------+----
6936      *        |                |                 |
6937      *        |     0.5        |       0.5       |
6938      *
6939      * So it appears as if there are 4 chroma images, but in fact the odd rows
6940      * in the chroma images are in the same row as the even ones. So its is
6941      * kinda tricky to read
6942      *
6943      * When reading from rectangle textures, keep in mind that the input y coordinates
6944      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6945      */
6946     shader_addline(buffer, "PARAM yv12_coef = ");
6947     shader_arb_append_imm_vec4(buffer, yv12_coef);
6948     shader_addline(buffer, ";\n");
6949 
6950     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6951     /* the chroma planes have only half the width */
6952     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6953 
6954     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6955      * the coordinate. Also read the right side of the image when reading odd lines
6956      *
6957      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6958      * bleeding
6959      */
6960     if(textype == GL_TEXTURE_2D) {
6961 
6962         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6963 
6964         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6965 
6966         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6967         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6968 
6969         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6970         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6971         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6972         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6973         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6974 
6975         /* clamp, keep the half pixel origin in mind */
6976         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6977         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6978         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6979         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6980     } else {
6981         /* Read from [size - size+size/4] */
6982         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6983         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6984 
6985         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6986         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6987         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6988         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6989         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6990         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6991 
6992         /* Make sure to read exactly from the pixel center */
6993         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6994         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6995 
6996         /* Clamp */
6997         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6998         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6999         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
7000         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
7001         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
7002     }
7003     /* Read the texture, put the result into the output register */
7004     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
7005     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
7006 
7007     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
7008      * No need to clamp because we're just reusing the already clamped value from above
7009      */
7010     if(textype == GL_TEXTURE_2D) {
7011         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
7012     } else {
7013         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
7014     }
7015     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
7016     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
7017 
7018     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
7019      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
7020      * values due to filtering
7021      */
7022     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
7023     if(textype == GL_TEXTURE_2D) {
7024         /* Multiply the y coordinate by 2/3 and clamp it */
7025         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
7026         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
7027         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
7028         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
7029     } else {
7030         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
7031          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
7032          * is bigger
7033          */
7034         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
7035         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
7036         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
7037     }
7038     *luminance = 'a';
7039 
7040     return TRUE;
7041 }
7042 
7043 static GLuint gen_p8_shader(struct arbfp_blit_priv *priv,
7044         const struct wined3d_gl_info *gl_info, GLenum textype)
7045 {
7046     GLenum shader;
7047     struct wined3d_shader_buffer buffer;
7048     GLint pos;
7049 
7050     /* Shader header */
7051     if (!shader_buffer_init(&buffer))
7052     {
7053         ERR("Failed to initialize shader buffer.\n");
7054         return 0;
7055     }
7056 
7057     GL_EXTCALL(glGenProgramsARB(1, &shader));
7058     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7059     if (!shader)
7060     {
7061         shader_buffer_free(&buffer);
7062         return 0;
7063     }
7064 
7065     shader_addline(&buffer, "!!ARBfp1.0\n");
7066     shader_addline(&buffer, "TEMP index;\n");
7067 
7068     /* { 255/256, 0.5/255*255/256, 0, 0 } */
7069     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
7070 
7071     /* The alpha-component contains the palette index */
7072     if(textype == GL_TEXTURE_RECTANGLE_ARB)
7073         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
7074     else
7075         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
7076 
7077     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
7078     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
7079 
7080     /* Use the alpha-component as an index in the palette to get the final color */
7081     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
7082     shader_addline(&buffer, "END\n");
7083 
7084     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7085             strlen(buffer.buffer), buffer.buffer));
7086     checkGLcall("glProgramStringARB()");
7087 
7088     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7089     if (pos != -1)
7090     {
7091         FIXME("Fragment program error at position %d: %s\n\n", pos,
7092               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7093         shader_arb_dump_program_source(buffer.buffer);
7094     }
7095 
7096     if (textype == GL_TEXTURE_RECTANGLE_ARB)
7097         priv->p8_rect_shader = shader;
7098     else
7099         priv->p8_2d_shader = shader;
7100 
7101     shader_buffer_free(&buffer);
7102 
7103     return shader;
7104 }
7105 
7106 /* Context activation is done by the caller. */
7107 static void upload_palette(const struct wined3d_surface *surface, struct wined3d_context *context)
7108 {
7109     BYTE table[256][4];
7110     struct wined3d_device *device = surface->resource.device;
7111     const struct wined3d_gl_info *gl_info = context->gl_info;
7112     struct arbfp_blit_priv *priv = device->blit_priv;
7113     BOOL colorkey = !!(surface->container->color_key_flags & WINEDDSD_CKSRCBLT);
7114 
7115     d3dfmt_p8_init_palette(surface, table, colorkey);
7116 
7117     if (!priv->palette_texture)
7118         gl_info->gl_ops.gl.p_glGenTextures(1, &priv->palette_texture);
7119 
7120     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
7121     gl_info->gl_ops.gl.p_glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
7122 
7123     gl_info->gl_ops.gl.p_glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
7124 
7125     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
7126     /* Make sure we have discrete color levels. */
7127     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
7128     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
7129     /* Upload the palette */
7130     /* TODO: avoid unneeded uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
7131     gl_info->gl_ops.gl.p_glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
7132 
7133     /* Switch back to unit 0 in which the 2D texture will be stored. */
7134     context_active_texture(context, gl_info, 0);
7135 }
7136 
7137 /* Context activation is done by the caller. */
7138 static GLuint gen_yuv_shader(struct arbfp_blit_priv *priv, const struct wined3d_gl_info *gl_info,
7139         enum complex_fixup yuv_fixup, GLenum textype)
7140 {
7141     GLenum shader;
7142     struct wined3d_shader_buffer buffer;
7143     char luminance_component;
7144     GLint pos;
7145 
7146     /* Shader header */
7147     if (!shader_buffer_init(&buffer))
7148     {
7149         ERR("Failed to initialize shader buffer.\n");
7150         return 0;
7151     }
7152 
7153     GL_EXTCALL(glGenProgramsARB(1, &shader));
7154     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
7155     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7156     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7157     if (!shader)
7158     {
7159         shader_buffer_free(&buffer);
7160         return 0;
7161     }
7162 
7163     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
7164      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
7165      * two chroma(U and V) values. Each macropixel has two luminance values, one for
7166      * each single pixel it contains, and one U and one V value shared between both
7167      * pixels.
7168      *
7169      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
7170      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
7171      * take the format into account when generating the read swizzles
7172      *
7173      * Reading the Y value is straightforward - just sample the texture. The hardware
7174      * takes care of filtering in the horizontal and vertical direction.
7175      *
7176      * Reading the U and V values is harder. We have to avoid filtering horizontally,
7177      * because that would mix the U and V values of one pixel or two adjacent pixels.
7178      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
7179      * regardless of the filtering setting. Vertical filtering works automatically
7180      * though - the U and V values of two rows are mixed nicely.
7181      *
7182      * Apart of avoiding filtering issues, the code has to know which value it just
7183      * read, and where it can find the other one. To determine this, it checks if
7184      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
7185      *
7186      * Handling horizontal filtering of U and V values requires reading a 2nd pair
7187      * of pixels, extracting U and V and mixing them. This is not implemented yet.
7188      *
7189      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
7190      * with width / 2. This way one read gives all 3 values, finding U and V is easy
7191      * in an unfiltered situation. Finding the luminance on the other hand requires
7192      * finding out if it is an odd or even pixel. The real drawback of this approach
7193      * is filtering. This would have to be emulated completely in the shader, reading
7194      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
7195      * vertically. Beyond that it would require adjustments to the texture handling
7196      * code to deal with the width scaling
7197      */
7198     shader_addline(&buffer, "!!ARBfp1.0\n");
7199     shader_addline(&buffer, "TEMP luminance;\n");
7200     shader_addline(&buffer, "TEMP temp;\n");
7201     shader_addline(&buffer, "TEMP chroma;\n");
7202     shader_addline(&buffer, "TEMP texcrd;\n");
7203     shader_addline(&buffer, "TEMP texcrd2;\n");
7204     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
7205     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
7206     shader_addline(&buffer, "PARAM size = program.local[0];\n");
7207 
7208     switch (yuv_fixup)
7209     {
7210         case COMPLEX_FIXUP_UYVY:
7211         case COMPLEX_FIXUP_YUY2:
7212             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
7213             {
7214                 shader_buffer_free(&buffer);
7215                 return 0;
7216             }
7217             break;
7218 
7219         case COMPLEX_FIXUP_YV12:
7220             if (!gen_yv12_read(&buffer, textype, &luminance_component))
7221             {
7222                 shader_buffer_free(&buffer);
7223                 return 0;
7224             }
7225             break;
7226 
7227         default:
7228             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
7229             shader_buffer_free(&buffer);
7230             return 0;
7231     }
7232 
7233     /* Calculate the final result. Formula is taken from
7234      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
7235      * ranges from -0.5 to 0.5
7236      */
7237     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
7238 
7239     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
7240     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
7241     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
7242     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
7243     shader_addline(&buffer, "END\n");
7244 
7245     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
7246             strlen(buffer.buffer), buffer.buffer));
7247     checkGLcall("glProgramStringARB()");
7248 
7249     gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
7250     if (pos != -1)
7251     {
7252         FIXME("Fragment program error at position %d: %s\n\n", pos,
7253               debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
7254         shader_arb_dump_program_source(buffer.buffer);
7255     }
7256     else
7257     {
7258         GLint native;
7259 
7260         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
7261         checkGLcall("glGetProgramivARB()");
7262         if (!native) WARN("Program exceeds native resource limits.\n");
7263     }
7264 
7265     shader_buffer_free(&buffer);
7266 
7267     switch (yuv_fixup)
7268     {
7269         case COMPLEX_FIXUP_YUY2:
7270             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
7271             else priv->yuy2_2d_shader = shader;
7272             break;
7273 
7274         case COMPLEX_FIXUP_UYVY:
7275             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
7276             else priv->uyvy_2d_shader = shader;
7277             break;
7278 
7279         case COMPLEX_FIXUP_YV12:
7280             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
7281             else priv->yv12_2d_shader = shader;
7282             break;
7283         default:
7284             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
7285     }
7286 
7287     return shader;
7288 }
7289 
7290 /* Context activation is done by the caller. */
7291 static HRESULT arbfp_blit_set(void *blit_priv, struct wined3d_context *context, const struct wined3d_surface *surface)
7292 {
7293     GLenum shader;
7294     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
7295     struct arbfp_blit_priv *priv = blit_priv;
7296     enum complex_fixup fixup;
7297     const struct wined3d_gl_info *gl_info = context->gl_info;
7298     GLenum textype = surface->container->target;
7299 
7300     if (surface->flags & SFLAG_CONVERTED)
7301     {
7302         gl_info->gl_ops.gl.p_glEnable(textype);
7303         checkGLcall("glEnable(textype)");
7304         return WINED3D_OK;
7305     }
7306 
7307     if (!is_complex_fixup(surface->resource.format->color_fixup))
7308     {
7309         TRACE("Fixup:\n");
7310         dump_color_fixup_desc(surface->resource.format->color_fixup);
7311         /* Don't bother setting up a shader for unconverted formats */
7312         gl_info->gl_ops.gl.p_glEnable(textype);
7313         checkGLcall("glEnable(textype)");
7314         return WINED3D_OK;
7315     }
7316 
7317     fixup = get_complex_fixup(surface->resource.format->color_fixup);
7318 
7319     switch(fixup)
7320     {
7321         case COMPLEX_FIXUP_YUY2:
7322             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7323             break;
7324 
7325         case COMPLEX_FIXUP_UYVY:
7326             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7327             break;
7328 
7329         case COMPLEX_FIXUP_YV12:
7330             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7331             break;
7332 
7333         case COMPLEX_FIXUP_P8:
7334             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7335             if (!shader) shader = gen_p8_shader(priv, gl_info, textype);
7336 
7337             upload_palette(surface, context);
7338             break;
7339 
7340         default:
7341             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7342             gl_info->gl_ops.gl.p_glEnable(textype);
7343             checkGLcall("glEnable(textype)");
7344             return E_NOTIMPL;
7345     }
7346 
7347     if (!shader) shader = gen_yuv_shader(priv, gl_info, fixup, textype);
7348 
7349     gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
7350     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7351     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7352     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7353     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7354     checkGLcall("glProgramLocalParameter4fvARB");
7355 
7356     return WINED3D_OK;
7357 }
7358 
7359 /* Context activation is done by the caller. */
7360 static void arbfp_blit_unset(const struct wined3d_gl_info *gl_info)
7361 {
7362     gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
7363     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7364     gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_2D);
7365     checkGLcall("glDisable(GL_TEXTURE_2D)");
7366     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7367     {
7368         gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7369         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7370     }
7371     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7372     {
7373         gl_info->gl_ops.gl.p_glDisable(GL_TEXTURE_RECTANGLE_ARB);
7374         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7375     }
7376 }
7377 
7378 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum wined3d_blit_op blit_op,
7379         const RECT *src_rect, DWORD src_usage, enum wined3d_pool src_pool, const struct wined3d_format *src_format,
7380         const RECT *dst_rect, DWORD dst_usage, enum wined3d_pool dst_pool, const struct wined3d_format *dst_format)
7381 {
7382     enum complex_fixup src_fixup;
7383 
7384     if (!gl_info->supported[ARB_FRAGMENT_PROGRAM])
7385         return FALSE;
7386 
7387     if (blit_op != WINED3D_BLIT_OP_COLOR_BLIT)
7388     {
7389         TRACE("Unsupported blit_op=%d\n", blit_op);
7390         return FALSE;
7391     }
7392 
7393     if (src_pool == WINED3D_POOL_SYSTEM_MEM || dst_pool == WINED3D_POOL_SYSTEM_MEM)
7394         return FALSE;
7395 
7396     src_fixup = get_complex_fixup(src_format->color_fixup);
7397     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7398     {
7399         TRACE("Checking support for fixup:\n");
7400         dump_color_fixup_desc(src_format->color_fixup);
7401     }
7402 
7403     if (!is_identity_fixup(dst_format->color_fixup))
7404     {
7405         TRACE("Destination fixups are not supported\n");
7406         return FALSE;
7407     }
7408 
7409     if (is_identity_fixup(src_format->color_fixup))
7410     {
7411         TRACE("[OK]\n");
7412         return TRUE;
7413     }
7414 
7415      /* We only support YUV conversions. */
7416     if (!is_complex_fixup(src_format->color_fixup))
7417     {
7418         TRACE("[FAILED]\n");
7419         return FALSE;
7420     }
7421 
7422     switch(src_fixup)
7423     {
7424         case COMPLEX_FIXUP_YUY2:
7425         case COMPLEX_FIXUP_UYVY:
7426         case COMPLEX_FIXUP_YV12:
7427         case COMPLEX_FIXUP_P8:
7428             TRACE("[OK]\n");
7429             return TRUE;
7430 
7431         default:
7432             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7433             TRACE("[FAILED]\n");
7434             return FALSE;
7435     }
7436 }
7437 
7438 HRESULT arbfp_blit_surface(struct wined3d_device *device, DWORD filter,
7439         struct wined3d_surface *src_surface, const RECT *src_rect_in,
7440         struct wined3d_surface *dst_surface, const RECT *dst_rect_in)
7441 {
7442     struct wined3d_context *context;
7443     RECT src_rect = *src_rect_in;
7444     RECT dst_rect = *dst_rect_in;
7445 
7446     /* Activate the destination context, set it up for blitting */
7447     context = context_acquire(device, dst_surface);
7448 
7449     /* Now load the surface */
7450     if (wined3d_settings.offscreen_rendering_mode != ORM_FBO
7451             && (src_surface->locations & (WINED3D_LOCATION_TEXTURE_RGB | WINED3D_LOCATION_DRAWABLE))
7452             == WINED3D_LOCATION_DRAWABLE
7453             && !surface_is_offscreen(src_surface))
7454     {
7455         /* Without FBO blits transferring from the drawable to the texture is
7456          * expensive, because we have to flip the data in sysmem. Since we can
7457          * flip in the blitter, we don't actually need that flip anyway. So we
7458          * use the surface's texture as scratch texture, and flip the source
7459          * rectangle instead. */
7460         surface_load_fb_texture(src_surface, FALSE);
7461 
7462         src_rect.top = src_surface->resource.height - src_rect.top;
7463         src_rect.bottom = src_surface->resource.height - src_rect.bottom;
7464     }
7465     else
7466         wined3d_texture_load(src_surface->container, context, FALSE);
7467 
7468     context_apply_blit_state(context, device);
7469 
7470     if (!surface_is_offscreen(dst_surface))
7471         surface_translate_drawable_coords(dst_surface, context->win_handle, &dst_rect);
7472 
7473     arbfp_blit_set(device->blit_priv, context, src_surface);
7474 
7475     /* Draw a textured quad */
7476     draw_textured_quad(src_surface, context, &src_rect, &dst_rect, filter);
7477 
7478     /* Leave the opengl state valid for blitting */
7479     arbfp_blit_unset(context->gl_info);
7480 
7481     if (wined3d_settings.strict_draw_ordering
7482             || (dst_surface->swapchain && (dst_surface->swapchain->front_buffer == dst_surface)))
7483         context->gl_info->gl_ops.gl.p_glFlush(); /* Flush to ensure ordering across contexts. */
7484 
7485     context_release(context);
7486 
7487     surface_validate_location(dst_surface, dst_surface->draw_binding);
7488     surface_invalidate_location(dst_surface, ~dst_surface->draw_binding);
7489 
7490     return WINED3D_OK;
7491 }
7492 
7493 static HRESULT arbfp_blit_color_fill(struct wined3d_device *device, struct wined3d_surface *dst_surface,
7494         const RECT *dst_rect, const struct wined3d_color *color)
7495 {
7496     FIXME("Color filling not implemented by arbfp_blit\n");
7497     return WINED3DERR_INVALIDCALL;
7498 }
7499 
7500 static HRESULT arbfp_blit_depth_fill(struct wined3d_device *device,
7501         struct wined3d_surface *surface, const RECT *rect, float depth)
7502 {
7503     FIXME("Depth filling not implemented by arbfp_blit.\n");
7504     return WINED3DERR_INVALIDCALL;
7505 }
7506 
7507 const struct blit_shader arbfp_blit = {
7508     arbfp_blit_alloc,
7509     arbfp_blit_free,
7510     arbfp_blit_set,
7511     arbfp_blit_unset,
7512     arbfp_blit_supported,
7513     arbfp_blit_color_fill,
7514     arbfp_blit_depth_fill,
7515 };
7516