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, 2013-2014 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 "config.h"
31 #include "wine/port.h"
32 
33 #include <stdio.h>
34 
35 #include "wined3d_private.h"
36 
37 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
38 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
39 WINE_DECLARE_DEBUG_CHANNEL(d3d);
40 WINE_DECLARE_DEBUG_CHANNEL(d3d_perf);
41 
42 static BOOL shader_is_pshader_version(enum wined3d_shader_type type)
43 {
44     return type == WINED3D_SHADER_TYPE_PIXEL;
45 }
46 
47 static BOOL shader_is_vshader_version(enum wined3d_shader_type type)
48 {
49     return type == WINED3D_SHADER_TYPE_VERTEX;
50 }
51 
52 static const char *get_line(const char **ptr)
53 {
54     const char *p, *q;
55 
56     p = *ptr;
57     if (!(q = strstr(p, "\n")))
58     {
59         if (!*p) return NULL;
60         *ptr += strlen(p);
61         return p;
62     }
63     *ptr = q + 1;
64 
65     return p;
66 }
67 
68 enum arb_helper_value
69 {
70     ARB_ZERO,
71     ARB_ONE,
72     ARB_TWO,
73     ARB_0001,
74     ARB_EPS,
75 
76     ARB_VS_REL_OFFSET
77 };
78 
79 static const char *arb_get_helper_value(enum wined3d_shader_type shader, enum arb_helper_value value)
80 {
81     if (shader != WINED3D_SHADER_TYPE_VERTEX && shader != WINED3D_SHADER_TYPE_PIXEL)
82     {
83         ERR("Unsupported shader type '%s'.\n", debug_shader_type(shader));
84         return "bad";
85     }
86 
87     if (shader == WINED3D_SHADER_TYPE_PIXEL)
88     {
89         switch (value)
90         {
91             case ARB_ZERO: return "ps_helper_const.x";
92             case ARB_ONE: return "ps_helper_const.y";
93             case ARB_TWO: return "coefmul.x";
94             case ARB_0001: return "ps_helper_const.xxxy";
95             case ARB_EPS: return "ps_helper_const.z";
96             default: break;
97         }
98     }
99     else
100     {
101         switch (value)
102         {
103             case ARB_ZERO: return "helper_const.x";
104             case ARB_ONE: return "helper_const.y";
105             case ARB_TWO: return "helper_const.z";
106             case ARB_EPS: return "helper_const.w";
107             case ARB_0001: return "helper_const.xxxy";
108             case ARB_VS_REL_OFFSET: return "rel_addr_const.y";
109         }
110     }
111     FIXME("Unmanaged %s shader helper constant requested: %u.\n",
112           shader == WINED3D_SHADER_TYPE_PIXEL ? "pixel" : "vertex", value);
113     switch (value)
114     {
115         case ARB_ZERO: return "0.0";
116         case ARB_ONE: return "1.0";
117         case ARB_TWO: return "2.0";
118         case ARB_0001: return "{0.0, 0.0, 0.0, 1.0}";
119         case ARB_EPS: return "1e-8";
120         default: return "bad";
121     }
122 }
123 
124 static inline BOOL ffp_clip_emul(const struct wined3d_context *context)
125 {
126     return context->lowest_disabled_stage < 7;
127 }
128 
129 /* ARB_program_shader private data */
130 
131 struct control_frame
132 {
133     struct                          list entry;
134     enum
135     {
136         IF,
137         IFC,
138         LOOP,
139         REP
140     } type;
141     BOOL                            muting;
142     BOOL                            outer_loop;
143     union
144     {
145         unsigned int                loop;
146         unsigned int                ifc;
147     } no;
148     struct wined3d_shader_loop_control loop_control;
149     BOOL                            had_else;
150 };
151 
152 struct arb_ps_np2fixup_info
153 {
154     struct ps_np2fixup_info         super;
155     /* For ARB we need an offset value:
156      * With both GLSL and ARB mode the NP2 fixup information (the texture dimensions) are stored in a
157      * consecutive way (GLSL uses a uniform array). Since ARB doesn't know the notion of a "standalone"
158      * array we need an offset to the index inside the program local parameter array. */
159     UINT                            offset;
160 };
161 
162 struct arb_ps_compile_args
163 {
164     struct ps_compile_args          super;
165     WORD                            bools;
166     WORD                            clip;  /* only a boolean, use a WORD for alignment */
167     unsigned char                   loop_ctrl[WINED3D_MAX_CONSTS_I][3];
168 };
169 
170 struct stb_const_desc
171 {
172     unsigned char           texunit;
173     UINT                    const_num;
174 };
175 
176 struct arb_ps_compiled_shader
177 {
178     struct arb_ps_compile_args      args;
179     struct arb_ps_np2fixup_info     np2fixup_info;
180     struct stb_const_desc           bumpenvmatconst[MAX_TEXTURES];
181     struct stb_const_desc           luminanceconst[MAX_TEXTURES];
182     UINT                            int_consts[WINED3D_MAX_CONSTS_I];
183     GLuint                          prgId;
184     UINT                            ycorrection;
185     unsigned char                   numbumpenvmatconsts;
186     char                            num_int_consts;
187 };
188 
189 struct arb_vs_compile_args
190 {
191     struct vs_compile_args          super;
192     union
193     {
194         struct
195         {
196             WORD                    bools;
197             unsigned char           clip_texcoord;
198             unsigned char           clipplane_mask;
199         }                           boolclip;
200         DWORD                       boolclip_compare;
201     } clip;
202     DWORD                           ps_signature;
203     union
204     {
205         unsigned char               samplers[4];
206         DWORD                       samplers_compare;
207     } vertex;
208     unsigned char                   loop_ctrl[WINED3D_MAX_CONSTS_I][3];
209 };
210 
211 struct arb_vs_compiled_shader
212 {
213     struct arb_vs_compile_args      args;
214     GLuint                          prgId;
215     UINT                            int_consts[WINED3D_MAX_CONSTS_I];
216     char                            num_int_consts;
217     char                            need_color_unclamp;
218     UINT                            pos_fixup;
219 };
220 
221 struct recorded_instruction
222 {
223     struct wined3d_shader_instruction ins;
224     struct list entry;
225 };
226 
227 struct shader_arb_ctx_priv
228 {
229     char addr_reg[20];
230     enum
231     {
232         /* plain GL_ARB_vertex_program or GL_ARB_fragment_program */
233         ARB,
234         /* GL_NV_vertex_program2_option or GL_NV_fragment_program_option */
235         NV2,
236         /* GL_NV_vertex_program3 or GL_NV_fragment_program2 */
237         NV3
238     } target_version;
239 
240     const struct arb_vs_compile_args    *cur_vs_args;
241     const struct arb_ps_compile_args    *cur_ps_args;
242     const struct arb_ps_compiled_shader *compiled_fprog;
243     const struct arb_vs_compiled_shader *compiled_vprog;
244     struct arb_ps_np2fixup_info         *cur_np2fixup_info;
245     struct list                         control_frames;
246     struct list                         record;
247     BOOL                                recording;
248     BOOL                                muted;
249     unsigned int                        num_loops, loop_depth, num_ifcs;
250     int                                 aL;
251     BOOL                                ps_post_process;
252 
253     unsigned int                        vs_clipplanes;
254     BOOL                                footer_written;
255     BOOL                                in_main_func;
256 
257     /* For 3.0 vertex shaders */
258     const char                          *vs_output[MAX_REG_OUTPUT];
259     /* For 2.x and earlier vertex shaders */
260     const char                          *texcrd_output[8], *color_output[2], *fog_output;
261 
262     /* 3.0 pshader input for compatibility with fixed function */
263     const char                          *ps_input[MAX_REG_INPUT];
264 };
265 
266 struct ps_signature
267 {
268     struct wined3d_shader_signature sig;
269     DWORD                               idx;
270     struct wine_rb_entry                entry;
271 };
272 
273 struct arb_pshader_private {
274     struct arb_ps_compiled_shader   *gl_shaders;
275     UINT                            num_gl_shaders, shader_array_size;
276     DWORD                           input_signature_idx;
277     DWORD                           clipplane_emulation;
278     BOOL                            clamp_consts;
279 };
280 
281 struct arb_vshader_private {
282     struct arb_vs_compiled_shader   *gl_shaders;
283     UINT                            num_gl_shaders, shader_array_size;
284     UINT rel_offset;
285 };
286 
287 struct shader_arb_priv
288 {
289     GLuint                  current_vprogram_id;
290     GLuint                  current_fprogram_id;
291     const struct arb_ps_compiled_shader *compiled_fprog;
292     const struct arb_vs_compiled_shader *compiled_vprog;
293     BOOL                    use_arbfp_fixed_func;
294     struct wine_rb_tree     fragment_shaders;
295     BOOL                    last_ps_const_clamped;
296     BOOL                    last_vs_color_unclamp;
297 
298     struct wine_rb_tree     signature_tree;
299     DWORD ps_sig_number;
300 
301     unsigned int highest_dirty_ps_const, highest_dirty_vs_const;
302     char vshader_const_dirty[WINED3D_MAX_VS_CONSTS_F];
303     char pshader_const_dirty[WINED3D_MAX_PS_CONSTS_F];
304     const struct wined3d_context *last_context;
305 
306     const struct wined3d_vertex_pipe_ops *vertex_pipe;
307     const struct fragment_pipeline *fragment_pipe;
308     BOOL ffp_proj_control;
309 };
310 
311 /* Context activation for state handlers is done by the caller. */
312 
313 static BOOL need_rel_addr_const(const struct arb_vshader_private *shader_data,
314         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
315 {
316     if (shader_data->rel_offset) return TRUE;
317     if (!reg_maps->usesmova) return FALSE;
318     return !gl_info->supported[NV_VERTEX_PROGRAM2_OPTION];
319 }
320 
321 /* Returns TRUE if result.clip from GL_NV_vertex_program2 should be used and FALSE otherwise */
322 static inline BOOL use_nv_clip(const struct wined3d_gl_info *gl_info)
323 {
324     return gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
325             && !(gl_info->quirks & WINED3D_QUIRK_NV_CLIP_BROKEN);
326 }
327 
328 static BOOL need_helper_const(const struct arb_vshader_private *shader_data,
329         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
330 {
331     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) return TRUE;
332     if (!gl_info->supported[NV_VERTEX_PROGRAM]) return TRUE; /* Need to init colors. */
333     if (gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT) return TRUE; /* Load the immval offset. */
334     if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) return TRUE; /* Have to init texcoords. */
335     if (!use_nv_clip(gl_info)) return TRUE; /* Init the clip texcoord */
336     if (reg_maps->usesnrm) return TRUE; /* 0.0 */
337     if (reg_maps->usespow) return TRUE; /* EPS, 0.0 and 1.0 */
338     if (reg_maps->fog) return TRUE; /* Clamping fog coord, 0.0 and 1.0 */
339     return FALSE;
340 }
341 
342 static unsigned int reserved_vs_const(const struct arb_vshader_private *shader_data,
343         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
344 {
345     unsigned int ret = 1;
346     /* We use one PARAM for the pos fixup, and in some cases one to load
347      * some immediate values into the shader. */
348     if (need_helper_const(shader_data, reg_maps, gl_info)) ++ret;
349     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) ++ret;
350     return ret;
351 }
352 
353 /* Loads floating point constants into the currently set ARB_vertex/fragment_program.
354  * When constant_list == NULL, it will load all the constants.
355  *
356  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
357  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
358  */
359 /* Context activation is done by the caller. */
360 static unsigned int shader_arb_load_constants_f(const struct wined3d_shader *shader,
361         const struct wined3d_gl_info *gl_info, GLuint target_type, unsigned int max_constants,
362         const struct wined3d_vec4 *constants, char *dirty_consts)
363 {
364     struct wined3d_shader_lconst *lconst;
365     unsigned int ret, i, j;
366 
367     if (TRACE_ON(d3d_constants))
368     {
369         for (i = 0; i < max_constants; ++i)
370         {
371             if (!dirty_consts[i])
372                 continue;
373             TRACE_(d3d_constants)("Loading constant %u: %s.\n", i, debug_vec4(&constants[i]));
374         }
375     }
376 
377     i = 0;
378 
379     /* In 1.X pixel shaders constants are implicitly clamped in the range [-1;1] */
380     if (target_type == GL_FRAGMENT_PROGRAM_ARB && shader->reg_maps.shader_version.major == 1)
381     {
382         float lcl_const[4];
383         /* ps 1.x supports only 8 constants, clamp only those. When switching between 1.x and higher
384          * shaders, the first 8 constants are marked dirty for reload
385          */
386         for (; i < min(8, max_constants); ++i)
387         {
388             if (!dirty_consts[i])
389                 continue;
390             dirty_consts[i] = 0;
391 
392             if (constants[i].x > 1.0f)
393                 lcl_const[0] = 1.0f;
394             else if (constants[i].x < -1.0f)
395                 lcl_const[0] = -1.0f;
396             else
397                 lcl_const[0] = constants[i].x;
398 
399             if (constants[i].y > 1.0f)
400                 lcl_const[1] = 1.0f;
401             else if (constants[i].y < -1.0f)
402                 lcl_const[1] = -1.0f;
403             else
404                 lcl_const[1] = constants[i].y;
405 
406             if (constants[i].z > 1.0f)
407                 lcl_const[2] = 1.0f;
408             else if (constants[i].z < -1.0f)
409                 lcl_const[2] = -1.0f;
410             else
411                 lcl_const[2] = constants[i].z;
412 
413             if (constants[i].w > 1.0f)
414                 lcl_const[3] = 1.0f;
415             else if (constants[i].w < -1.0f)
416                 lcl_const[3] = -1.0f;
417             else
418                 lcl_const[3] = constants[i].w;
419 
420             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, lcl_const));
421         }
422 
423         /* If further constants are dirty, reload them without clamping.
424          *
425          * The alternative is not to touch them, but then we cannot reset the dirty constant count
426          * to zero. That's bad for apps that only use PS 1.x shaders, because in that case the code
427          * above would always re-check the first 8 constants since max_constant remains at the init
428          * value
429          */
430     }
431 
432     if (gl_info->supported[EXT_GPU_PROGRAM_PARAMETERS])
433     {
434         /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
435          * or just reloading *all* constants at once
436          *
437         GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, i, max_constants, constants + (i * 4)));
438          */
439         for (; i < max_constants; ++i)
440         {
441             if (!dirty_consts[i])
442                 continue;
443 
444             /* Find the next block of dirty constants */
445             dirty_consts[i] = 0;
446             j = i;
447             for (++i; (i < max_constants) && dirty_consts[i]; ++i)
448             {
449                 dirty_consts[i] = 0;
450             }
451 
452             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, &constants[j].x));
453         }
454     }
455     else
456     {
457         for (; i < max_constants; ++i)
458         {
459             if (dirty_consts[i])
460             {
461                 dirty_consts[i] = 0;
462                 GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, &constants[i].x));
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 < WINED3D_MAX_CONSTS_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[i].x;
595             val[1] = (float)state->ps_consts_i[i].y;
596             val[2] = (float)state->ps_consts_i[i].z;
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 < WINED3D_MAX_CONSTS_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[i].x;
625             val[1] = (float)state->vs_consts_i[i].y;
626             val[2] = (float)state->vs_consts_i[i].z;
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_constants_f(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]->height;
704 
705         /* Load DirectX 9 float constants for pixel shader */
706         priv->highest_dirty_ps_const = shader_arb_load_constants_f(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 
729     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
730      * context. On a context switch the old context will be fully dirtified */
731     if (!context || context->device != device)
732         return;
733 
734     memset(priv->vshader_const_dirty + start, 1, sizeof(*priv->vshader_const_dirty) * count);
735     priv->highest_dirty_vs_const = max(priv->highest_dirty_vs_const, start + count);
736 }
737 
738 static void shader_arb_update_float_pixel_constants(struct wined3d_device *device, UINT start, UINT count)
739 {
740     struct wined3d_context *context = context_get_current();
741     struct shader_arb_priv *priv = device->shader_priv;
742 
743     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
744      * context. On a context switch the old context will be fully dirtified */
745     if (!context || context->device != device)
746         return;
747 
748     memset(priv->pshader_const_dirty + start, 1, sizeof(*priv->pshader_const_dirty) * count);
749     priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, start + count);
750 }
751 
752 static void shader_arb_append_imm_vec4(struct wined3d_string_buffer *buffer, const float *values)
753 {
754     char str[4][17];
755 
756     wined3d_ftoa(values[0], str[0]);
757     wined3d_ftoa(values[1], str[1]);
758     wined3d_ftoa(values[2], str[2]);
759     wined3d_ftoa(values[3], str[3]);
760     shader_addline(buffer, "{%s, %s, %s, %s}", str[0], str[1], str[2], str[3]);
761 }
762 
763 /* Generate the variable & register declarations for the ARB_vertex_program output target */
764 static void shader_generate_arb_declarations(const struct wined3d_shader *shader,
765         const struct wined3d_shader_reg_maps *reg_maps, struct wined3d_string_buffer *buffer,
766         const struct wined3d_gl_info *gl_info, DWORD *num_clipplanes,
767         const struct shader_arb_ctx_priv *ctx)
768 {
769     DWORD i;
770     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
771     const struct wined3d_shader_lconst *lconst;
772     unsigned max_constantsF;
773     DWORD map;
774 
775     /* In pixel shaders, all private constants are program local, we don't need anything
776      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
777      * If we need a private constant the GL implementation will squeeze it in somewhere
778      *
779      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
780      * immediate values. The posFixup is loaded using program.env for now, so always
781      * subtract one from the number of constants. If the shader uses indirect addressing,
782      * account for the helper const too because we have to declare all available d3d constants
783      * and don't know which are actually used.
784      */
785     if (pshader)
786     {
787         max_constantsF = gl_info->limits.arb_ps_native_constants;
788         /* 24 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value. */
789         if (max_constantsF < 24)
790             max_constantsF = gl_info->limits.arb_ps_float_constants;
791     }
792     else
793     {
794         const struct arb_vshader_private *shader_data = shader->backend_data;
795         max_constantsF = gl_info->limits.arb_vs_native_constants;
796         /* 96 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value.
797          * Also prevents max_constantsF from becoming less than 0 and
798          * wrapping . */
799         if (max_constantsF < 96)
800             max_constantsF = gl_info->limits.arb_vs_float_constants;
801 
802         if (reg_maps->usesrelconstF)
803         {
804             DWORD highest_constf = 0, clip_limit;
805 
806             max_constantsF -= reserved_vs_const(shader_data, reg_maps, gl_info);
807             max_constantsF -= wined3d_popcount(reg_maps->integer_constants);
808             max_constantsF -= gl_info->reserved_arb_constants;
809 
810             for (i = 0; i < shader->limits->constant_float; ++i)
811             {
812                 DWORD idx = i >> 5;
813                 DWORD shift = i & 0x1f;
814                 if (reg_maps->constf[idx] & (1u << shift))
815                     highest_constf = i;
816             }
817 
818             if(use_nv_clip(gl_info) && ctx->target_version >= NV2)
819             {
820                 if(ctx->cur_vs_args->super.clip_enabled)
821                     clip_limit = gl_info->limits.user_clip_distances;
822                 else
823                     clip_limit = 0;
824             }
825             else
826             {
827                 unsigned int mask = ctx->cur_vs_args->clip.boolclip.clipplane_mask;
828                 clip_limit = min(wined3d_popcount(mask), 4);
829             }
830             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
831             max_constantsF -= *num_clipplanes;
832             if(*num_clipplanes < clip_limit)
833             {
834                 WARN("Only %u clip planes out of %u enabled.\n", *num_clipplanes,
835                         gl_info->limits.user_clip_distances);
836             }
837         }
838         else
839         {
840             if (ctx->target_version >= NV2)
841                 *num_clipplanes = gl_info->limits.user_clip_distances;
842             else
843                 *num_clipplanes = min(gl_info->limits.user_clip_distances, 4);
844         }
845     }
846 
847     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
848     {
849         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
850     }
851 
852     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
853     {
854         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
855     }
856 
857     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
858     {
859         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
860         {
861             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
862         }
863     }
864 
865     if (!shader->load_local_constsF)
866     {
867         LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
868         {
869             const float *value;
870             value = (const float *)lconst->value;
871             shader_addline(buffer, "PARAM C%u = ", lconst->idx);
872             shader_arb_append_imm_vec4(buffer, value);
873             shader_addline(buffer, ";\n");
874         }
875     }
876 
877     /* After subtracting privately used constants from the hardware limit(they are loaded as
878      * local constants), make sure the shader doesn't violate the env constant limit
879      */
880     if (pshader)
881     {
882         max_constantsF = min(max_constantsF, gl_info->limits.arb_ps_float_constants);
883     }
884     else
885     {
886         max_constantsF = min(max_constantsF, gl_info->limits.arb_vs_float_constants);
887     }
888 
889     /* Avoid declaring more constants than needed */
890     max_constantsF = min(max_constantsF, shader->limits->constant_float);
891 
892     /* we use the array-based constants array if the local constants are marked for loading,
893      * because then we use indirect addressing, or when the local constant list is empty,
894      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
895      * local constants do not declare the loaded constants as an array because ARB compilers usually
896      * do not optimize unused constants away
897      */
898     if (reg_maps->usesrelconstF)
899     {
900         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
901         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
902                     max_constantsF, max_constantsF - 1);
903     }
904     else
905     {
906         for (i = 0; i < max_constantsF; ++i)
907         {
908             if (!shader_constant_is_local(shader, i) && wined3d_extract_bits(reg_maps->constf, i, 1))
909             {
910                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
911             }
912         }
913     }
914 }
915 
916 static const char * const shift_tab[] = {
917     "dummy",     /*  0 (none) */
918     "coefmul.x", /*  1 (x2)   */
919     "coefmul.y", /*  2 (x4)   */
920     "coefmul.z", /*  3 (x8)   */
921     "coefmul.w", /*  4 (x16)  */
922     "dummy",     /*  5 (x32)  */
923     "dummy",     /*  6 (x64)  */
924     "dummy",     /*  7 (x128) */
925     "dummy",     /*  8 (d256) */
926     "dummy",     /*  9 (d128) */
927     "dummy",     /* 10 (d64)  */
928     "dummy",     /* 11 (d32)  */
929     "coefdiv.w", /* 12 (d16)  */
930     "coefdiv.z", /* 13 (d8)   */
931     "coefdiv.y", /* 14 (d4)   */
932     "coefdiv.x"  /* 15 (d2)   */
933 };
934 
935 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
936         const struct wined3d_shader_dst_param *dst, char *write_mask)
937 {
938     char *ptr = write_mask;
939 
940     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
941     {
942         *ptr++ = '.';
943         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
944         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
945         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
946         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
947     }
948 
949     *ptr = '\0';
950 }
951 
952 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
953 {
954     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
955      * but addressed as "rgba". To fix this we need to swap the register's x
956      * and z components. */
957     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
958     char *ptr = swizzle_str;
959 
960     /* swizzle bits fields: wwzzyyxx */
961     DWORD swizzle = param->swizzle;
962     DWORD swizzle_x = swizzle & 0x03;
963     DWORD swizzle_y = (swizzle >> 2) & 0x03;
964     DWORD swizzle_z = (swizzle >> 4) & 0x03;
965     DWORD swizzle_w = (swizzle >> 6) & 0x03;
966 
967     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
968      * generate a swizzle string. Unless we need to our own swizzling. */
969     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
970     {
971         *ptr++ = '.';
972         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
973             *ptr++ = swizzle_chars[swizzle_x];
974         } else {
975             *ptr++ = swizzle_chars[swizzle_x];
976             *ptr++ = swizzle_chars[swizzle_y];
977             *ptr++ = swizzle_chars[swizzle_z];
978             *ptr++ = swizzle_chars[swizzle_w];
979         }
980     }
981 
982     *ptr = '\0';
983 }
984 
985 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
986 {
987     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
988     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
989 
990     if (!strcmp(priv->addr_reg, src)) return;
991 
992     strcpy(priv->addr_reg, src);
993     shader_addline(buffer, "ARL A0.x, %s;\n", src);
994 }
995 
996 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
997         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
998 
999 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
1000         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
1001 {
1002     /* oPos, oFog and oPts in D3D */
1003     static const char * const rastout_reg_names[] = {"TMP_OUT", "TMP_FOGCOORD", "result.pointsize"};
1004     const struct wined3d_shader *shader = ins->ctx->shader;
1005     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1006     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1007     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1008 
1009     *is_color = FALSE;
1010 
1011     switch (reg->type)
1012     {
1013         case WINED3DSPR_TEMP:
1014             sprintf(register_name, "R%u", reg->idx[0].offset);
1015             break;
1016 
1017         case WINED3DSPR_INPUT:
1018             if (pshader)
1019             {
1020                 if (reg_maps->shader_version.major < 3)
1021                 {
1022                     if (!reg->idx[0].offset)
1023                         strcpy(register_name, "fragment.color.primary");
1024                     else
1025                         strcpy(register_name, "fragment.color.secondary");
1026                 }
1027                 else
1028                 {
1029                     if (reg->idx[0].rel_addr)
1030                     {
1031                         char rel_reg[50];
1032                         shader_arb_get_src_param(ins, reg->idx[0].rel_addr, 0, rel_reg);
1033 
1034                         if (!strcmp(rel_reg, "**aL_emul**"))
1035                         {
1036                             DWORD idx = ctx->aL + reg->idx[0].offset;
1037                             if(idx < MAX_REG_INPUT)
1038                             {
1039                                 strcpy(register_name, ctx->ps_input[idx]);
1040                             }
1041                             else
1042                             {
1043                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
1044                                 sprintf(register_name, "out_of_bounds_%u", idx);
1045                             }
1046                         }
1047                         else if (reg_maps->input_registers & 0x0300)
1048                         {
1049                             /* There are two ways basically:
1050                              *
1051                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
1052                              *    That means trouble if the loop also contains a breakc or if the control values
1053                              *    aren't local constants.
1054                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
1055                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
1056                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
1057                              *    ADAC to load the condition code register and pop it again afterwards
1058                              */
1059                             FIXME("Relative input register addressing with more than 8 registers\n");
1060 
1061                             /* This is better than nothing for now */
1062                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1063                         }
1064                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
1065                         {
1066                             /* This is problematic because we'd have to consult the ctx->ps_input strings
1067                              * for where to find the varying. Some may be "0.0", others can be texcoords or
1068                              * colors. This needs either a pipeline replacement to make the vertex shader feed
1069                              * proper varyings, or loop unrolling
1070                              *
1071                              * For now use the texcoords and hope for the best
1072                              */
1073                             FIXME("Non-vertex shader varying input with indirect addressing\n");
1074                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1075                         }
1076                         else
1077                         {
1078                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
1079                              * pulls GL_NV_fragment_program2 in
1080                              */
1081                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
1082                         }
1083                     }
1084                     else
1085                     {
1086                         if (reg->idx[0].offset < MAX_REG_INPUT)
1087                         {
1088                             strcpy(register_name, ctx->ps_input[reg->idx[0].offset]);
1089                         }
1090                         else
1091                         {
1092                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx[0].offset);
1093                             sprintf(register_name, "out_of_bounds_%u", reg->idx[0].offset);
1094                         }
1095                     }
1096                 }
1097             }
1098             else
1099             {
1100                 if (ctx->cur_vs_args->super.swizzle_map & (1u << reg->idx[0].offset))
1101                     *is_color = TRUE;
1102                 sprintf(register_name, "vertex.attrib[%u]", reg->idx[0].offset);
1103             }
1104             break;
1105 
1106         case WINED3DSPR_CONST:
1107             if (!pshader && reg->idx[0].rel_addr)
1108             {
1109                 const struct arb_vshader_private *shader_data = shader->backend_data;
1110                 UINT rel_offset = ctx->target_version == ARB ? shader_data->rel_offset : 0;
1111                 BOOL aL = FALSE;
1112                 char rel_reg[50];
1113                 if (reg_maps->shader_version.major < 2)
1114                 {
1115                     sprintf(rel_reg, "A0.x");
1116                 }
1117                 else
1118                 {
1119                     shader_arb_get_src_param(ins, reg->idx[0].rel_addr, 0, rel_reg);
1120                     if (ctx->target_version == ARB)
1121                     {
1122                         if (!strcmp(rel_reg, "**aL_emul**"))
1123                         {
1124                             aL = TRUE;
1125                         } else {
1126                             shader_arb_request_a0(ins, rel_reg);
1127                             sprintf(rel_reg, "A0.x");
1128                         }
1129                     }
1130                 }
1131                 if (aL)
1132                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx[0].offset);
1133                 else if (reg->idx[0].offset >= rel_offset)
1134                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx[0].offset - rel_offset);
1135                 else
1136                     sprintf(register_name, "C[%s - %u]", rel_reg, rel_offset - reg->idx[0].offset);
1137             }
1138             else
1139             {
1140                 if (reg_maps->usesrelconstF)
1141                     sprintf(register_name, "C[%u]", reg->idx[0].offset);
1142                 else
1143                     sprintf(register_name, "C%u", reg->idx[0].offset);
1144             }
1145             break;
1146 
1147         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1148             if (pshader)
1149             {
1150                 if (reg_maps->shader_version.major == 1
1151                         && reg_maps->shader_version.minor <= 3)
1152                     /* In ps <= 1.3, Tx is a temporary register as destination
1153                      * to all instructions, and as source to most instructions.
1154                      * For some instructions it is the texcoord input. Those
1155                      * instructions know about the special use. */
1156                     sprintf(register_name, "T%u", reg->idx[0].offset);
1157                 else
1158                     /* In ps 1.4 and 2.x Tx is always a (read-only) varying. */
1159                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx[0].offset);
1160             }
1161             else
1162             {
1163                 if (reg_maps->shader_version.major == 1 || ctx->target_version >= NV2)
1164                     sprintf(register_name, "A%u", reg->idx[0].offset);
1165                 else
1166                     sprintf(register_name, "A%u_SHADOW", reg->idx[0].offset);
1167             }
1168             break;
1169 
1170         case WINED3DSPR_COLOROUT:
1171             if (ctx->ps_post_process && !reg->idx[0].offset)
1172             {
1173                 strcpy(register_name, "TMP_COLOR");
1174             }
1175             else
1176             {
1177                 if (ctx->cur_ps_args->super.srgb_correction)
1178                     FIXME("sRGB correction on higher render targets.\n");
1179                 if (reg_maps->rt_mask > 1)
1180                     sprintf(register_name, "result.color[%u]", reg->idx[0].offset);
1181                 else
1182                     strcpy(register_name, "result.color");
1183             }
1184             break;
1185 
1186         case WINED3DSPR_RASTOUT:
1187             if (reg->idx[0].offset == 1)
1188                 sprintf(register_name, "%s", ctx->fog_output);
1189             else
1190                 sprintf(register_name, "%s", rastout_reg_names[reg->idx[0].offset]);
1191             break;
1192 
1193         case WINED3DSPR_DEPTHOUT:
1194             strcpy(register_name, "result.depth");
1195             break;
1196 
1197         case WINED3DSPR_ATTROUT:
1198         /* case WINED3DSPR_OUTPUT: */
1199             if (pshader)
1200                 sprintf(register_name, "oD[%u]", reg->idx[0].offset);
1201             else
1202                 strcpy(register_name, ctx->color_output[reg->idx[0].offset]);
1203             break;
1204 
1205         case WINED3DSPR_TEXCRDOUT:
1206             if (pshader)
1207                 sprintf(register_name, "oT[%u]", reg->idx[0].offset);
1208             else if (reg_maps->shader_version.major < 3)
1209                 strcpy(register_name, ctx->texcrd_output[reg->idx[0].offset]);
1210             else
1211                 strcpy(register_name, ctx->vs_output[reg->idx[0].offset]);
1212             break;
1213 
1214         case WINED3DSPR_LOOP:
1215             if(ctx->target_version >= NV2)
1216             {
1217                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1218                 if(pshader) sprintf(register_name, "A0.x");
1219                 else sprintf(register_name, "aL.y");
1220             }
1221             else
1222             {
1223                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1224                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1225                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1226                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1227                  * indexing
1228                  */
1229                 sprintf(register_name, "**aL_emul**");
1230             }
1231 
1232             break;
1233 
1234         case WINED3DSPR_CONSTINT:
1235             sprintf(register_name, "I%u", reg->idx[0].offset);
1236             break;
1237 
1238         case WINED3DSPR_MISCTYPE:
1239             if (!reg->idx[0].offset)
1240                 sprintf(register_name, "vpos");
1241             else if (reg->idx[0].offset == 1)
1242                 sprintf(register_name, "fragment.facing.x");
1243             else
1244                 FIXME("Unknown MISCTYPE register index %u.\n", reg->idx[0].offset);
1245             break;
1246 
1247         default:
1248             FIXME("Unhandled register type %#x[%u].\n", reg->type, reg->idx[0].offset);
1249             sprintf(register_name, "unrecognized_register[%u]", reg->idx[0].offset);
1250             break;
1251     }
1252 }
1253 
1254 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1255         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1256 {
1257     char register_name[255];
1258     char write_mask[6];
1259     BOOL is_color;
1260 
1261     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1262     strcpy(str, register_name);
1263 
1264     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1265     strcat(str, write_mask);
1266 }
1267 
1268 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1269 {
1270     switch(channel_source)
1271     {
1272         case CHANNEL_SOURCE_ZERO: return "0";
1273         case CHANNEL_SOURCE_ONE: return "1";
1274         case CHANNEL_SOURCE_X: return "x";
1275         case CHANNEL_SOURCE_Y: return "y";
1276         case CHANNEL_SOURCE_Z: return "z";
1277         case CHANNEL_SOURCE_W: return "w";
1278         default:
1279             FIXME("Unhandled channel source %#x\n", channel_source);
1280             return "undefined";
1281     }
1282 }
1283 
1284 struct color_fixup_masks
1285 {
1286     DWORD source;
1287     DWORD sign;
1288 };
1289 
1290 static struct color_fixup_masks calc_color_correction(struct color_fixup_desc fixup, DWORD dst_mask)
1291 {
1292     struct color_fixup_masks masks = {0, 0};
1293 
1294     if (is_complex_fixup(fixup))
1295     {
1296         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1297         FIXME("Complex fixup (%#x) not supported\n", complex_fixup);
1298         return masks;
1299     }
1300 
1301     if (fixup.x_source != CHANNEL_SOURCE_X)
1302         masks.source |= WINED3DSP_WRITEMASK_0;
1303     if (fixup.y_source != CHANNEL_SOURCE_Y)
1304         masks.source |= WINED3DSP_WRITEMASK_1;
1305     if (fixup.z_source != CHANNEL_SOURCE_Z)
1306         masks.source |= WINED3DSP_WRITEMASK_2;
1307     if (fixup.w_source != CHANNEL_SOURCE_W)
1308         masks.source |= WINED3DSP_WRITEMASK_3;
1309     masks.source &= dst_mask;
1310 
1311     if (fixup.x_sign_fixup)
1312         masks.sign |= WINED3DSP_WRITEMASK_0;
1313     if (fixup.y_sign_fixup)
1314         masks.sign |= WINED3DSP_WRITEMASK_1;
1315     if (fixup.z_sign_fixup)
1316         masks.sign |= WINED3DSP_WRITEMASK_2;
1317     if (fixup.w_sign_fixup)
1318         masks.sign |= WINED3DSP_WRITEMASK_3;
1319     masks.sign &= dst_mask;
1320 
1321     return masks;
1322 }
1323 
1324 static void gen_color_correction(struct wined3d_string_buffer *buffer, const char *dst,
1325         const char *src, const char *one, const char *two,
1326         struct color_fixup_desc fixup, struct color_fixup_masks masks)
1327 {
1328     const char *sign_fixup_src = dst;
1329 
1330     if (masks.source)
1331     {
1332         if (masks.sign)
1333             sign_fixup_src = "TA";
1334 
1335         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", sign_fixup_src, src,
1336                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1337                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1338     }
1339     else if (masks.sign)
1340     {
1341         sign_fixup_src = src;
1342     }
1343 
1344     if (masks.sign)
1345     {
1346         char reg_mask[6];
1347         char *ptr = reg_mask;
1348 
1349         if (masks.sign != WINED3DSP_WRITEMASK_ALL)
1350         {
1351             *ptr++ = '.';
1352             if (masks.sign & WINED3DSP_WRITEMASK_0)
1353                 *ptr++ = 'x';
1354             if (masks.sign & WINED3DSP_WRITEMASK_1)
1355                 *ptr++ = 'y';
1356             if (masks.sign & WINED3DSP_WRITEMASK_2)
1357                 *ptr++ = 'z';
1358             if (masks.sign & WINED3DSP_WRITEMASK_3)
1359                 *ptr++ = 'w';
1360         }
1361         *ptr = '\0';
1362 
1363         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", dst, reg_mask, sign_fixup_src, two, one);
1364     }
1365 }
1366 
1367 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1368 {
1369     DWORD mod;
1370     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1371     if (!ins->dst_count) return "";
1372 
1373     mod = ins->dst[0].modifiers;
1374 
1375     /* Silently ignore PARTIALPRECISION if it's not supported */
1376     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1377 
1378     if(mod & WINED3DSPDM_MSAMPCENTROID)
1379     {
1380         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1381         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1382     }
1383 
1384     switch(mod)
1385     {
1386         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1387             return "H_SAT";
1388 
1389         case WINED3DSPDM_SATURATE:
1390             return "_SAT";
1391 
1392         case WINED3DSPDM_PARTIALPRECISION:
1393             return "H";
1394 
1395         case 0:
1396             return "";
1397 
1398         default:
1399             FIXME("Unknown modifiers 0x%08x\n", mod);
1400             return "";
1401     }
1402 }
1403 
1404 #define TEX_PROJ        0x1
1405 #define TEX_BIAS        0x2
1406 #define TEX_LOD         0x4
1407 #define TEX_DERIV       0x10
1408 
1409 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1410         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1411 {
1412     enum wined3d_shader_resource_type resource_type = ins->ctx->reg_maps->resource_info[sampler_idx].type;
1413     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
1414     const char *tex_type;
1415     BOOL np2_fixup = FALSE;
1416     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1417     const char *mod;
1418     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1419     const struct wined3d_shader *shader;
1420     const struct wined3d_device *device;
1421     const struct wined3d_gl_info *gl_info;
1422     const char *tex_dst = dst_str;
1423     struct color_fixup_masks masks;
1424 
1425     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1426     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1427 
1428     switch (resource_type)
1429     {
1430         case WINED3D_SHADER_RESOURCE_TEXTURE_1D:
1431             tex_type = "1D";
1432             break;
1433 
1434         case WINED3D_SHADER_RESOURCE_TEXTURE_2D:
1435             shader = ins->ctx->shader;
1436             device = shader->device;
1437             gl_info = &device->adapter->gl_info;
1438 
1439             if (pshader && priv->cur_ps_args->super.np2_fixup & (1u << sampler_idx)
1440                     && gl_info->supported[ARB_TEXTURE_RECTANGLE])
1441                 tex_type = "RECT";
1442             else
1443                 tex_type = "2D";
1444             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1445             {
1446                 if (priv->cur_np2fixup_info->super.active & (1u << sampler_idx))
1447                 {
1448                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1449                     else np2_fixup = TRUE;
1450                 }
1451             }
1452             break;
1453 
1454         case WINED3D_SHADER_RESOURCE_TEXTURE_3D:
1455             tex_type = "3D";
1456             break;
1457 
1458         case WINED3D_SHADER_RESOURCE_TEXTURE_CUBE:
1459             tex_type = "CUBE";
1460             break;
1461 
1462         default:
1463             ERR("Unexpected resource type %#x.\n", resource_type);
1464             tex_type = "";
1465     }
1466 
1467     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1468      * so don't use shader_arb_get_modifier
1469      */
1470     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1471     else mod = "";
1472 
1473     /* Fragment samplers always have indentity mapping */
1474     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1475     {
1476         sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1477     }
1478 
1479     if (pshader)
1480     {
1481         masks = calc_color_correction(priv->cur_ps_args->super.color_fixup[sampler_idx],
1482                 ins->dst[0].write_mask);
1483 
1484         if (masks.source || masks.sign)
1485             tex_dst = "TA";
1486     }
1487 
1488     if (flags & TEX_DERIV)
1489     {
1490         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1491         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1492         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, tex_dst, coord_reg,
1493                        dsx, dsy, sampler_idx, tex_type);
1494     }
1495     else if(flags & TEX_LOD)
1496     {
1497         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1498         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1499         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, tex_dst, coord_reg,
1500                        sampler_idx, tex_type);
1501     }
1502     else if (flags & TEX_BIAS)
1503     {
1504         /* Shouldn't be possible, but let's check for it */
1505         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1506         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1507         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, tex_dst, coord_reg, sampler_idx, tex_type);
1508     }
1509     else if (flags & TEX_PROJ)
1510     {
1511         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, tex_dst, coord_reg, sampler_idx, tex_type);
1512     }
1513     else
1514     {
1515         if (np2_fixup)
1516         {
1517             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1518             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1519                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1520 
1521             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, tex_dst, sampler_idx, tex_type);
1522         }
1523         else
1524             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, tex_dst, coord_reg, sampler_idx, tex_type);
1525     }
1526 
1527     if (pshader)
1528     {
1529         gen_color_correction(buffer, dst_str, tex_dst,
1530                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_ONE),
1531                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_TWO),
1532                 priv->cur_ps_args->super.color_fixup[sampler_idx], masks);
1533     }
1534 }
1535 
1536 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1537         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1538 {
1539     /* Generate a line that does the input modifier computation and return the input register to use */
1540     BOOL is_color = FALSE, insert_line;
1541     char regstr[256];
1542     char swzstr[20];
1543     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
1544     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1545     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
1546     const char *two = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_TWO);
1547 
1548     /* Assume a new line will be added */
1549     insert_line = TRUE;
1550 
1551     /* Get register name */
1552     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1553     shader_arb_get_swizzle(src, is_color, swzstr);
1554 
1555     switch (src->modifiers)
1556     {
1557     case WINED3DSPSM_NONE:
1558         sprintf(outregstr, "%s%s", regstr, swzstr);
1559         insert_line = FALSE;
1560         break;
1561     case WINED3DSPSM_NEG:
1562         sprintf(outregstr, "-%s%s", regstr, swzstr);
1563         insert_line = FALSE;
1564         break;
1565     case WINED3DSPSM_BIAS:
1566         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1567         break;
1568     case WINED3DSPSM_BIASNEG:
1569         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1570         break;
1571     case WINED3DSPSM_SIGN:
1572         shader_addline(buffer, "MAD T%c, %s, %s, -%s;\n", 'A' + tmpreg, regstr, two, one);
1573         break;
1574     case WINED3DSPSM_SIGNNEG:
1575         shader_addline(buffer, "MAD T%c, %s, -%s, %s;\n", 'A' + tmpreg, regstr, two, one);
1576         break;
1577     case WINED3DSPSM_COMP:
1578         shader_addline(buffer, "SUB T%c, %s, %s;\n", 'A' + tmpreg, one, regstr);
1579         break;
1580     case WINED3DSPSM_X2:
1581         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1582         break;
1583     case WINED3DSPSM_X2NEG:
1584         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1585         break;
1586     case WINED3DSPSM_DZ:
1587         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1588         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1589         break;
1590     case WINED3DSPSM_DW:
1591         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1592         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1593         break;
1594     case WINED3DSPSM_ABS:
1595         if(ctx->target_version >= NV2) {
1596             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1597             insert_line = FALSE;
1598         } else {
1599             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1600         }
1601         break;
1602     case WINED3DSPSM_ABSNEG:
1603         if(ctx->target_version >= NV2) {
1604             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1605         } else {
1606             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1607             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1608         }
1609         insert_line = FALSE;
1610         break;
1611     default:
1612         sprintf(outregstr, "%s%s", regstr, swzstr);
1613         insert_line = FALSE;
1614     }
1615 
1616     /* Return modified or original register, with swizzle */
1617     if (insert_line)
1618         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1619 }
1620 
1621 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1622 {
1623     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1624     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
1625     DWORD sampler_code = dst->reg.idx[0].offset;
1626     char dst_name[50];
1627     char src_name[2][50];
1628 
1629     shader_arb_get_dst_param(ins, dst, dst_name);
1630 
1631     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1632      *
1633      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1634      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1635      * temps is done.
1636      */
1637     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1638     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1639     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1640     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1641     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1642 
1643     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1644     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1645 }
1646 
1647 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1648 {
1649     *extra_char = ' ';
1650     switch(mod)
1651     {
1652         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1653         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1654         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1655         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1656         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1657         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1658         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1659         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1660         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1661         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1662         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1663         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1664         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1665     }
1666     FIXME("Unknown modifier %u\n", mod);
1667     return mod;
1668 }
1669 
1670 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1671 {
1672     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1673     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
1674     char dst_name[50];
1675     char src_name[3][50];
1676     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1677             ins->ctx->reg_maps->shader_version.minor);
1678 
1679     shader_arb_get_dst_param(ins, dst, dst_name);
1680     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1681 
1682     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue
1683             && ins->dst->write_mask != WINED3DSP_WRITEMASK_3)
1684     {
1685         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1686     }
1687     else
1688     {
1689         struct wined3d_shader_src_param src0_copy = ins->src[0];
1690         char extra_neg;
1691 
1692         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1693         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1694 
1695         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1696         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1697         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1698         shader_addline(buffer, "CMP%s %s, TA, %s, %s;\n", shader_arb_get_modifier(ins),
1699                 dst_name, src_name[1], src_name[2]);
1700     }
1701 }
1702 
1703 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1704 {
1705     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1706     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
1707     char dst_name[50];
1708     char src_name[3][50];
1709 
1710     shader_arb_get_dst_param(ins, dst, dst_name);
1711 
1712     /* Generate input register names (with modifiers) */
1713     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1714     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1715     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1716 
1717     shader_addline(buffer, "CMP%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1718             dst_name, src_name[0], src_name[2], src_name[1]);
1719 }
1720 
1721 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1722  * dst = dot2(src0, src1) + src2 */
1723 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1724 {
1725     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1726     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
1727     char dst_name[50];
1728     char src_name[3][50];
1729     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1730 
1731     shader_arb_get_dst_param(ins, dst, dst_name);
1732     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1733     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1734 
1735     if(ctx->target_version >= NV3)
1736     {
1737         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1738         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1739         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1740                        dst_name, src_name[0], src_name[1], src_name[2]);
1741     }
1742     else if(ctx->target_version >= NV2)
1743     {
1744         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1745          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1746          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1747          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1748          *
1749          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1750          *
1751          * .xyxy and other swizzles that we could get with this are not valid in
1752          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1753          */
1754         struct wined3d_shader_src_param tmp_param = ins->src[1];
1755         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1756         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1757 
1758         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1759 
1760         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1761                        dst_name, src_name[2], src_name[0], src_name[1]);
1762     }
1763     else
1764     {
1765         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1766         /* 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]
1767         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1768         */
1769         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1770         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1771         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1772         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1773     }
1774 }
1775 
1776 /* Map the opcode 1-to-1 to the GL code */
1777 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1778 {
1779     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
1780     const char *instruction;
1781     char arguments[256], dst_str[50];
1782     unsigned int i;
1783     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1784 
1785     switch (ins->handler_idx)
1786     {
1787         case WINED3DSIH_ABS: instruction = "ABS"; break;
1788         case WINED3DSIH_ADD: instruction = "ADD"; break;
1789         case WINED3DSIH_CRS: instruction = "XPD"; break;
1790         case WINED3DSIH_DP3: instruction = "DP3"; break;
1791         case WINED3DSIH_DP4: instruction = "DP4"; break;
1792         case WINED3DSIH_DST: instruction = "DST"; break;
1793         case WINED3DSIH_FRC: instruction = "FRC"; break;
1794         case WINED3DSIH_LIT: instruction = "LIT"; break;
1795         case WINED3DSIH_LRP: instruction = "LRP"; break;
1796         case WINED3DSIH_MAD: instruction = "MAD"; break;
1797         case WINED3DSIH_MAX: instruction = "MAX"; break;
1798         case WINED3DSIH_MIN: instruction = "MIN"; break;
1799         case WINED3DSIH_MOV: instruction = "MOV"; break;
1800         case WINED3DSIH_MUL: instruction = "MUL"; break;
1801         case WINED3DSIH_SGE: instruction = "SGE"; break;
1802         case WINED3DSIH_SLT: instruction = "SLT"; break;
1803         case WINED3DSIH_SUB: instruction = "SUB"; break;
1804         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1805         case WINED3DSIH_DSX: instruction = "DDX"; break;
1806         default: instruction = "";
1807             FIXME("Unhandled opcode %s.\n", debug_d3dshaderinstructionhandler(ins->handler_idx));
1808             break;
1809     }
1810 
1811     /* Note that shader_arb_add_dst_param() adds spaces. */
1812     arguments[0] = '\0';
1813     shader_arb_get_dst_param(ins, dst, dst_str);
1814     for (i = 0; i < ins->src_count; ++i)
1815     {
1816         char operand[100];
1817         strcat(arguments, ", ");
1818         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1819         strcat(arguments, operand);
1820     }
1821     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1822 }
1823 
1824 static void shader_hw_nop(const struct wined3d_shader_instruction *ins) {}
1825 
1826 static DWORD shader_arb_select_component(DWORD swizzle, DWORD component)
1827 {
1828     return ((swizzle >> 2 * component) & 0x3) * 0x55;
1829 }
1830 
1831 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1832 {
1833     const struct wined3d_shader *shader = ins->ctx->shader;
1834     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1835     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1836     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1837     const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
1838     const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
1839     const char *two = arb_get_helper_value(reg_maps->shader_version.type, ARB_TWO);
1840 
1841     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
1842     char src0_param[256];
1843 
1844     if (ins->handler_idx == WINED3DSIH_MOVA)
1845     {
1846         const struct arb_vshader_private *shader_data = shader->backend_data;
1847         char write_mask[6];
1848         const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1849 
1850         if(ctx->target_version >= NV2) {
1851             shader_hw_map2gl(ins);
1852             return;
1853         }
1854         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1855         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1856 
1857         /* This implements the mova formula used in GLSL. The first two instructions
1858          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1859          * in this case:
1860          * mova A0.x, 0.0
1861          *
1862          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1863          *
1864          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1865          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1866          */
1867         shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
1868         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
1869 
1870         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1871         shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
1872         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1873         if (shader_data->rel_offset)
1874         {
1875             shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
1876         }
1877         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1878 
1879         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1880     }
1881     else if (reg_maps->shader_version.major == 1
1882           && !shader_is_pshader_version(reg_maps->shader_version.type)
1883           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1884     {
1885         const struct arb_vshader_private *shader_data = shader->backend_data;
1886         src0_param[0] = '\0';
1887 
1888         if (shader_data->rel_offset && ctx->target_version == ARB)
1889         {
1890             const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1891             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1892             shader_addline(buffer, "ADD TA.x, %s, %s;\n", src0_param, offset);
1893             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1894         }
1895         else
1896         {
1897             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1898              * with more than one component. Thus replicate the first source argument over all
1899              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1900             struct wined3d_shader_src_param tmp_src = ins->src[0];
1901             tmp_src.swizzle = shader_arb_select_component(tmp_src.swizzle, 0);
1902             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1903             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1904         }
1905     }
1906     else if (ins->dst[0].reg.type == WINED3DSPR_COLOROUT && !ins->dst[0].reg.idx[0].offset && pshader)
1907     {
1908         if (ctx->ps_post_process && shader->u.ps.color0_mov)
1909         {
1910             shader_addline(buffer, "#mov handled in srgb write or fog code\n");
1911             return;
1912         }
1913         shader_hw_map2gl(ins);
1914     }
1915     else
1916     {
1917         shader_hw_map2gl(ins);
1918     }
1919 }
1920 
1921 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1922 {
1923     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1924     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
1925     char reg_dest[40];
1926 
1927     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1928      * but >= 2.0 honors it (undocumented, but tested by the d3d9 testsuite)
1929      */
1930     shader_arb_get_dst_param(ins, dst, reg_dest);
1931 
1932     if (ins->ctx->reg_maps->shader_version.major >= 2)
1933     {
1934         const char *kilsrc = "TA";
1935         BOOL is_color;
1936 
1937         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1938         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1939         {
1940             kilsrc = reg_dest;
1941         }
1942         else
1943         {
1944             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1945              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1946              * masked out components to 0(won't kill)
1947              */
1948             char x = '0', y = '0', z = '0', w = '0';
1949             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1950             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1951             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1952             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1953             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1954         }
1955         shader_addline(buffer, "KIL %s;\n", kilsrc);
1956     }
1957     else
1958     {
1959         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1960          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1961          *
1962          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1963          * or pass in any temporary register(in shader phase 2)
1964          */
1965         if (ins->ctx->reg_maps->shader_version.minor <= 3)
1966             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx[0].offset);
1967         else
1968             shader_arb_get_dst_param(ins, dst, reg_dest);
1969         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1970         shader_addline(buffer, "KIL TA;\n");
1971     }
1972 }
1973 
1974 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1975 {
1976     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1977     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1978     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1979             ins->ctx->reg_maps->shader_version.minor);
1980     struct wined3d_shader_src_param src;
1981 
1982     char reg_dest[40];
1983     char reg_coord[40];
1984     DWORD reg_sampler_code;
1985     WORD myflags = 0;
1986     BOOL swizzle_coord = FALSE;
1987 
1988     /* All versions have a destination register */
1989     shader_arb_get_dst_param(ins, dst, reg_dest);
1990 
1991     /* 1.0-1.4: Use destination register number as texture code.
1992        2.0+: Use provided sampler number as texture code. */
1993     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1994         reg_sampler_code = dst->reg.idx[0].offset;
1995     else
1996         reg_sampler_code = ins->src[1].reg.idx[0].offset;
1997 
1998     /* 1.0-1.3: Use the texcoord varying.
1999        1.4+: Use provided coordinate source register. */
2000     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2001         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
2002     else {
2003         /* TEX is the only instruction that can handle DW and DZ natively */
2004         src = ins->src[0];
2005         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
2006         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
2007         shader_arb_get_src_param(ins, &src, 0, reg_coord);
2008     }
2009 
2010     /* projection flag:
2011      * 1.1, 1.2, 1.3: Use WINED3D_TSS_TEXTURETRANSFORMFLAGS
2012      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
2013      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
2014      */
2015     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2016     {
2017         DWORD flags = 0;
2018         if (reg_sampler_code < MAX_TEXTURES)
2019             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
2020         if (flags & WINED3D_PSARGS_PROJECTED)
2021         {
2022             myflags |= TEX_PROJ;
2023             if ((flags & ~WINED3D_PSARGS_PROJECTED) == WINED3D_TTFF_COUNT3)
2024                 swizzle_coord = TRUE;
2025         }
2026     }
2027     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
2028     {
2029         enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
2030         if (src_mod == WINED3DSPSM_DZ)
2031         {
2032             swizzle_coord = TRUE;
2033             myflags |= TEX_PROJ;
2034         } else if(src_mod == WINED3DSPSM_DW) {
2035             myflags |= TEX_PROJ;
2036         }
2037     } else {
2038         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
2039         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
2040     }
2041 
2042     if (swizzle_coord)
2043     {
2044         /* TXP cannot handle DZ natively, so move the z coordinate to .w.
2045          * reg_coord is a read-only varying register, so we need a temp reg */
2046         shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
2047         strcpy(reg_coord, "TA");
2048     }
2049 
2050     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
2051 }
2052 
2053 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
2054 {
2055     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2056     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2057     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2058             ins->ctx->reg_maps->shader_version.minor);
2059     char dst_str[50];
2060 
2061     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2062     {
2063         DWORD reg = dst->reg.idx[0].offset;
2064 
2065         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2066         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
2067     } else {
2068         char reg_src[40];
2069 
2070         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
2071         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2072         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
2073     }
2074 }
2075 
2076 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
2077 {
2078      struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2079      DWORD flags = 0;
2080 
2081      DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2082      char dst_str[50];
2083      char src_str[50];
2084 
2085      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
2086      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2087      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2088      /* Move .x first in case src_str is "TA" */
2089      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
2090      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
2091      if (reg1 < MAX_TEXTURES)
2092      {
2093          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2094          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
2095      }
2096      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2097 }
2098 
2099 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
2100 {
2101      struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2102 
2103      DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2104      char dst_str[50];
2105      char src_str[50];
2106 
2107      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2108      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2109      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2110      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2111      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2112      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2113 }
2114 
2115 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2116 {
2117     DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2118     char dst_str[50];
2119     char src_str[50];
2120 
2121     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2122     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2123     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2124     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2125 }
2126 
2127 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2128 {
2129     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2130     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2131     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2132     char reg_coord[40], dst_reg[50], src_reg[50];
2133     DWORD reg_dest_code;
2134 
2135     /* All versions have a destination register. The Tx where the texture coordinates come
2136      * from is the varying incarnation of the texture register
2137      */
2138     reg_dest_code = dst->reg.idx[0].offset;
2139     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2140     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2141     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2142 
2143     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2144      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2145      *
2146      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2147      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2148      *
2149      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2150      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2151      * extension.
2152      */
2153     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2154     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2155     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2156     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2157 
2158     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2159      * so we can't let the GL handle this.
2160      */
2161     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2162             & WINED3D_PSARGS_PROJECTED)
2163     {
2164         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2165         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2166         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2167     } else {
2168         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2169     }
2170 
2171     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2172 
2173     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2174     {
2175         /* No src swizzles are allowed, so this is ok */
2176         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2177                        src_reg, reg_dest_code, reg_dest_code);
2178         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2179     }
2180 }
2181 
2182 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2183 {
2184     DWORD reg = ins->dst[0].reg.idx[0].offset;
2185     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2186     char src0_name[50], dst_name[50];
2187     BOOL is_color;
2188     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2189 
2190     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2191     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2192      * T<reg+1> register. Use this register to store the calculated vector
2193      */
2194     tmp_reg.idx[0].offset = reg + 1;
2195     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2196     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2197 }
2198 
2199 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2200 {
2201     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2202     DWORD flags;
2203     DWORD reg = ins->dst[0].reg.idx[0].offset;
2204     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2205     char dst_str[50];
2206     char src0_name[50];
2207     char dst_reg[50];
2208     BOOL is_color;
2209 
2210     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2211     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2212 
2213     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2214     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2215     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2216     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2217     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2218 }
2219 
2220 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2221 {
2222     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2223     DWORD reg = ins->dst[0].reg.idx[0].offset;
2224     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2225     char src0_name[50], dst_name[50];
2226     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2227     BOOL is_color;
2228 
2229     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2230      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2231      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2232      */
2233     tmp_reg.idx[0].offset = reg + 2 - tex_mx->current_row;
2234     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2235 
2236     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2237     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2238                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2239     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2240 }
2241 
2242 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2243 {
2244     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2245     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2246     DWORD flags;
2247     DWORD reg = ins->dst[0].reg.idx[0].offset;
2248     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2249     char dst_str[50];
2250     char src0_name[50], dst_name[50];
2251     BOOL is_color;
2252 
2253     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2254     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2255     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2256 
2257     /* Sample the texture using the calculated coordinates */
2258     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2259     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2260     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2261     tex_mx->current_row = 0;
2262 }
2263 
2264 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2265 {
2266     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2267     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2268     DWORD flags;
2269     DWORD reg = ins->dst[0].reg.idx[0].offset;
2270     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2271     char dst_str[50];
2272     char src0_name[50];
2273     char dst_reg[50];
2274     BOOL is_color;
2275 
2276     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2277      * components for temporary data storage
2278      */
2279     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2280     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2281     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2282 
2283     /* Construct the eye-ray vector from w coordinates */
2284     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2285     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2286     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2287 
2288     /* Calculate reflection vector
2289      */
2290     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2291     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2292     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2293     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2294     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2295     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2296     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2297 
2298     /* Sample the texture using the calculated coordinates */
2299     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2300     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2301     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2302     tex_mx->current_row = 0;
2303 }
2304 
2305 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2306 {
2307     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2308     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2309     DWORD flags;
2310     DWORD reg = ins->dst[0].reg.idx[0].offset;
2311     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2312     char dst_str[50];
2313     char src0_name[50];
2314     char src1_name[50];
2315     char dst_reg[50];
2316     BOOL is_color;
2317 
2318     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2319     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2320     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2321     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2322     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2323 
2324     /* Calculate reflection vector.
2325      *
2326      *                   dot(N, E)
2327      * dst_reg.xyz = 2 * --------- * N - E
2328      *                   dot(N, N)
2329      *
2330      * Which normalizes the normal vector
2331      */
2332     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2333     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2334     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2335     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2336     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2337     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2338 
2339     /* Sample the texture using the calculated coordinates */
2340     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2341     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2342     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2343     tex_mx->current_row = 0;
2344 }
2345 
2346 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2347 {
2348     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2349     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2350     char dst_name[50];
2351     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2352     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2353 
2354     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2355      * which is essentially an input, is the destination register because it is the first
2356      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2357      * here(writemasks/swizzles are not valid on texdepth)
2358      */
2359     shader_arb_get_dst_param(ins, dst, dst_name);
2360 
2361     /* According to the msdn, the source register(must be r5) is unusable after
2362      * the texdepth instruction, so we're free to modify it
2363      */
2364     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2365 
2366     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2367      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2368      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2369      */
2370     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2371     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2372     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2373     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2374 }
2375 
2376 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2377  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2378  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2379 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2380 {
2381     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2382     DWORD sampler_idx = ins->dst[0].reg.idx[0].offset;
2383     char src0[50];
2384     char dst_str[50];
2385 
2386     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2387     shader_addline(buffer, "MOV TB, 0.0;\n");
2388     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2389 
2390     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2391     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2392 }
2393 
2394 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2395  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2396 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2397 {
2398     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2399     char src0[50];
2400     char dst_str[50];
2401     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2402 
2403     /* Handle output register */
2404     shader_arb_get_dst_param(ins, dst, dst_str);
2405     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2406     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx[0].offset, src0);
2407 }
2408 
2409 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2410  * Perform the 3rd row of a 3x3 matrix multiply */
2411 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2412 {
2413     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2414     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2415     char dst_str[50], dst_name[50];
2416     char src0[50];
2417     BOOL is_color;
2418 
2419     shader_arb_get_dst_param(ins, dst, dst_str);
2420     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2421     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2422     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx[0].offset, src0);
2423     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2424 }
2425 
2426 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2427  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2428  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2429  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2430  */
2431 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2432 {
2433     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2434     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2435     char src0[50], dst_name[50];
2436     BOOL is_color;
2437     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2438     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2439 
2440     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2441     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2442     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx[0].offset, src0);
2443 
2444     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2445      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2446      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2447      */
2448     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2449     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2450     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2451     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2452 }
2453 
2454 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2455     Vertex/Pixel shaders to ARB_vertex_program codes */
2456 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2457 {
2458     int i;
2459     int nComponents = 0;
2460     struct wined3d_shader_dst_param tmp_dst = {{0}};
2461     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2462     struct wined3d_shader_instruction tmp_ins;
2463 
2464     memset(&tmp_ins, 0, sizeof(tmp_ins));
2465 
2466     /* Set constants for the temporary argument */
2467     tmp_ins.ctx = ins->ctx;
2468     tmp_ins.dst_count = 1;
2469     tmp_ins.dst = &tmp_dst;
2470     tmp_ins.src_count = 2;
2471     tmp_ins.src = tmp_src;
2472 
2473     switch(ins->handler_idx)
2474     {
2475         case WINED3DSIH_M4x4:
2476             nComponents = 4;
2477             tmp_ins.handler_idx = WINED3DSIH_DP4;
2478             break;
2479         case WINED3DSIH_M4x3:
2480             nComponents = 3;
2481             tmp_ins.handler_idx = WINED3DSIH_DP4;
2482             break;
2483         case WINED3DSIH_M3x4:
2484             nComponents = 4;
2485             tmp_ins.handler_idx = WINED3DSIH_DP3;
2486             break;
2487         case WINED3DSIH_M3x3:
2488             nComponents = 3;
2489             tmp_ins.handler_idx = WINED3DSIH_DP3;
2490             break;
2491         case WINED3DSIH_M3x2:
2492             nComponents = 2;
2493             tmp_ins.handler_idx = WINED3DSIH_DP3;
2494             break;
2495         default:
2496             FIXME("Unhandled opcode %s.\n", debug_d3dshaderinstructionhandler(ins->handler_idx));
2497             break;
2498     }
2499 
2500     tmp_dst = ins->dst[0];
2501     tmp_src[0] = ins->src[0];
2502     tmp_src[1] = ins->src[1];
2503     for (i = 0; i < nComponents; ++i)
2504     {
2505         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2506         shader_hw_map2gl(&tmp_ins);
2507         ++tmp_src[1].reg.idx[0].offset;
2508     }
2509 }
2510 
2511 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2512 {
2513     *need_abs = FALSE;
2514 
2515     switch(mod)
2516     {
2517         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2518         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2519         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2520         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2521         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2522         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2523         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2524         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2525         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2526         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2527         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2528         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2529         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2530     }
2531     FIXME("Unknown modifier %u\n", mod);
2532     return mod;
2533 }
2534 
2535 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2536 {
2537     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2538     const char *instruction;
2539     struct wined3d_shader_src_param src0_copy = ins->src[0];
2540     BOOL need_abs = FALSE;
2541 
2542     char dst[50];
2543     char src[50];
2544 
2545     switch(ins->handler_idx)
2546     {
2547         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2548         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2549         case WINED3DSIH_EXPP:
2550             if (ins->ctx->reg_maps->shader_version.major < 2)
2551             {
2552                 instruction = "EXP";
2553                 break;
2554             }
2555             /* Drop through. */
2556         case WINED3DSIH_EXP:
2557             instruction = "EX2";
2558             break;
2559         case WINED3DSIH_LOG:
2560         case WINED3DSIH_LOGP:
2561             /* The precision requirements suggest that LOGP matches ARBvp's LOG
2562              * instruction, but notice that the output of those instructions is
2563              * different. */
2564             src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2565             instruction = "LG2";
2566             break;
2567         default: instruction = "";
2568             FIXME("Unhandled opcode %s.\n", debug_d3dshaderinstructionhandler(ins->handler_idx));
2569             break;
2570     }
2571 
2572     /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2573      * .w is used. */
2574     src0_copy.swizzle = shader_arb_select_component(src0_copy.swizzle, 3);
2575 
2576     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2577     shader_arb_get_src_param(ins, &src0_copy, 0, src);
2578 
2579     if(need_abs)
2580     {
2581         shader_addline(buffer, "ABS TA.w, %s;\n", src);
2582         shader_addline(buffer, "%s%s %s, TA.w;\n", instruction, shader_arb_get_modifier(ins), dst);
2583     }
2584     else
2585     {
2586         shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2587     }
2588 
2589 }
2590 
2591 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2592 {
2593     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2594     char dst_name[50];
2595     char src_name[50];
2596     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2597     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2598     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2599 
2600     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2601     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2602 
2603     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2604      * otherwise NRM or RSQ would return NaN */
2605     if(pshader && priv->target_version >= NV3)
2606     {
2607         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2608          *
2609          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2610          */
2611         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2612         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2613         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2614     }
2615     else if(priv->target_version >= NV2)
2616     {
2617         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2618         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2619         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2620                        src_name);
2621     }
2622     else
2623     {
2624         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2625 
2626         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2627         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2628          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2629          */
2630         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2631         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2632 
2633         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2634         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2635         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2636                     src_name);
2637     }
2638 }
2639 
2640 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2641 {
2642     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2643     char dst_name[50];
2644     char src_name[3][50];
2645 
2646     /* ARB_fragment_program has a convenient LRP instruction */
2647     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2648         shader_hw_map2gl(ins);
2649         return;
2650     }
2651 
2652     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2653     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2654     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2655     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2656 
2657     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2658     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2659                    dst_name, src_name[0], src_name[2]);
2660 }
2661 
2662 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2663 {
2664     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2665      * must contain fixed constants. So we need a separate function to filter those constants and
2666      * can't use map2gl
2667      */
2668     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2669     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2670     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2671     char dst_name[50];
2672     char src_name0[50], src_name1[50], src_name2[50];
2673     BOOL is_color;
2674 
2675     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2676     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2677         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2678         /* No modifiers are supported on SCS */
2679         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2680 
2681         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2682         {
2683             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2684             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2685         }
2686     } else if(priv->target_version >= NV2) {
2687         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2688 
2689         /* Sincos writemask must be .x, .y or .xy */
2690         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2691             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2692         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2693             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2694     } else {
2695         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2696          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2697          *
2698          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2699          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2700          *
2701          * The constants we get are:
2702          *
2703          *  +1   +1,     -1     -1     +1      +1      -1       -1
2704          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2705          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2706          *
2707          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2708          *
2709          * (x/2)^2 = x^2 / 4
2710          * (x/2)^3 = x^3 / 8
2711          * (x/2)^4 = x^4 / 16
2712          * (x/2)^5 = x^5 / 32
2713          * etc
2714          *
2715          * To get the final result:
2716          * sin(x) = 2 * sin(x/2) * cos(x/2)
2717          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2718          * (from sin(x+y) and cos(x+y) rules)
2719          *
2720          * As per MSDN, dst.z is undefined after the operation, and so is
2721          * dst.x and dst.y if they're masked out by the writemask. Ie
2722          * sincos dst.y, src1, c0, c1
2723          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2724          * vsa.exe also stops with an error if the dest register is the same register as the source
2725          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2726          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2727          */
2728         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2729         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2730         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2731 
2732         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2733         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2734         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2735         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2736         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2737         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2738 
2739         /* sin(x/2)
2740          *
2741          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2742          * properly merge that with MULs in the code above?
2743          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2744          * we can merge the sine and cosine MAD rows to calculate them together.
2745          */
2746         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2747         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2748         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2749         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2750 
2751         /* cos(x/2) */
2752         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2753         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2754         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2755 
2756         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2757             /* cos x */
2758             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2759             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2760         }
2761         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2762             /* sin x */
2763             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2764             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2765         }
2766     }
2767 }
2768 
2769 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2770 {
2771     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2772     char dst_name[50];
2773     char src_name[50];
2774     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2775 
2776     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2777     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2778 
2779     /* SGN is only valid in vertex shaders */
2780     if(ctx->target_version >= NV2) {
2781         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2782         return;
2783     }
2784 
2785     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2786      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2787      */
2788     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2789         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2790     } else {
2791         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2792          * Then use TA, and calculate the final result
2793          *
2794          * Not reading from TA? Store the first result in TA to avoid overwriting the
2795          * destination if src reg = dst reg
2796          */
2797         if(strstr(src_name, "TA"))
2798         {
2799             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2800             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2801             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2802         }
2803         else
2804         {
2805             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2806             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2807             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2808         }
2809     }
2810 }
2811 
2812 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2813 {
2814     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2815     char src[50];
2816     char dst[50];
2817     char dst_name[50];
2818     BOOL is_color;
2819 
2820     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2821     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2822     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2823 
2824     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2825     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2826 }
2827 
2828 static void shader_hw_pow(const struct wined3d_shader_instruction *ins)
2829 {
2830     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2831     char src0[50], src1[50], dst[50];
2832     struct wined3d_shader_src_param src0_copy = ins->src[0];
2833     BOOL need_abs = FALSE;
2834     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2835     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2836 
2837     /* POW operates on the absolute value of the input */
2838     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2839 
2840     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2841     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2842     shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2843 
2844     if (need_abs)
2845         shader_addline(buffer, "ABS TA.x, %s;\n", src0);
2846     else
2847         shader_addline(buffer, "MOV TA.x, %s;\n", src0);
2848 
2849     if (priv->target_version >= NV2)
2850     {
2851         shader_addline(buffer, "MOVC TA.y, %s;\n", src1);
2852         shader_addline(buffer, "POW%s %s, TA.x, TA.y;\n", shader_arb_get_modifier(ins), dst);
2853         shader_addline(buffer, "MOV %s (EQ.y), %s;\n", dst, one);
2854     }
2855     else
2856     {
2857         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2858         const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2859 
2860         shader_addline(buffer, "ABS TA.y, %s;\n", src1);
2861         shader_addline(buffer, "SGE TA.y, -TA.y, %s;\n", zero);
2862         /* Possibly add flt_eps to avoid getting float special values */
2863         shader_addline(buffer, "MAD TA.z, TA.y, %s, %s;\n", flt_eps, src1);
2864         shader_addline(buffer, "POW%s TA.x, TA.x, TA.z;\n", shader_arb_get_modifier(ins));
2865         shader_addline(buffer, "MAD TA.x, -TA.x, TA.y, TA.x;\n");
2866         shader_addline(buffer, "MAD %s, TA.y, %s, TA.x;\n", dst, one);
2867     }
2868 }
2869 
2870 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2871 {
2872     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2873     char src_name[50];
2874     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2875 
2876     /* src0 is aL */
2877     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2878 
2879     if(vshader)
2880     {
2881         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2882         struct list *e = list_head(&priv->control_frames);
2883         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2884 
2885         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2886         /* The constant loader makes sure to load -1 into iX.w */
2887         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2888         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2889         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2890     }
2891     else
2892     {
2893         shader_addline(buffer, "LOOP %s;\n", src_name);
2894     }
2895 }
2896 
2897 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2898 {
2899     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2900     char src_name[50];
2901     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2902 
2903     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2904 
2905     /* The constant loader makes sure to load -1 into iX.w */
2906     if(vshader)
2907     {
2908         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2909         struct list *e = list_head(&priv->control_frames);
2910         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2911 
2912         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2913 
2914         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2915         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2916         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2917     }
2918     else
2919     {
2920         shader_addline(buffer, "REP %s;\n", src_name);
2921     }
2922 }
2923 
2924 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2925 {
2926     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2927     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2928 
2929     if(vshader)
2930     {
2931         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2932         struct list *e = list_head(&priv->control_frames);
2933         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2934 
2935         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2936         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2937         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2938 
2939         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2940     }
2941     else
2942     {
2943         shader_addline(buffer, "ENDLOOP;\n");
2944     }
2945 }
2946 
2947 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2948 {
2949     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2950     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2951 
2952     if(vshader)
2953     {
2954         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2955         struct list *e = list_head(&priv->control_frames);
2956         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2957 
2958         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2959         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2960         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2961 
2962         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2963     }
2964     else
2965     {
2966         shader_addline(buffer, "ENDREP;\n");
2967     }
2968 }
2969 
2970 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2971 {
2972     struct control_frame *control_frame;
2973 
2974     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2975     {
2976         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2977     }
2978     ERR("Could not find loop for break\n");
2979     return NULL;
2980 }
2981 
2982 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2983 {
2984     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2985     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2986     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2987 
2988     if(vshader)
2989     {
2990         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2991     }
2992     else
2993     {
2994         shader_addline(buffer, "BRK;\n");
2995     }
2996 }
2997 
2998 static const char *get_compare(enum wined3d_shader_rel_op op)
2999 {
3000     switch (op)
3001     {
3002         case WINED3D_SHADER_REL_OP_GT: return "GT";
3003         case WINED3D_SHADER_REL_OP_EQ: return "EQ";
3004         case WINED3D_SHADER_REL_OP_GE: return "GE";
3005         case WINED3D_SHADER_REL_OP_LT: return "LT";
3006         case WINED3D_SHADER_REL_OP_NE: return "NE";
3007         case WINED3D_SHADER_REL_OP_LE: return "LE";
3008         default:
3009             FIXME("Unrecognized operator %#x.\n", op);
3010             return "(\?\?)";
3011     }
3012 }
3013 
3014 static enum wined3d_shader_rel_op invert_compare(enum wined3d_shader_rel_op op)
3015 {
3016     switch (op)
3017     {
3018         case WINED3D_SHADER_REL_OP_GT: return WINED3D_SHADER_REL_OP_LE;
3019         case WINED3D_SHADER_REL_OP_EQ: return WINED3D_SHADER_REL_OP_NE;
3020         case WINED3D_SHADER_REL_OP_GE: return WINED3D_SHADER_REL_OP_LT;
3021         case WINED3D_SHADER_REL_OP_LT: return WINED3D_SHADER_REL_OP_GE;
3022         case WINED3D_SHADER_REL_OP_NE: return WINED3D_SHADER_REL_OP_EQ;
3023         case WINED3D_SHADER_REL_OP_LE: return WINED3D_SHADER_REL_OP_GT;
3024         default:
3025             FIXME("Unrecognized operator %#x.\n", op);
3026             return -1;
3027     }
3028 }
3029 
3030 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
3031 {
3032     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
3033     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3034     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
3035     char src_name0[50];
3036     char src_name1[50];
3037     const char *comp = get_compare(ins->flags);
3038 
3039     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3040     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3041 
3042     if(vshader)
3043     {
3044         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
3045          * away the subtraction result
3046          */
3047         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3048         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
3049     }
3050     else
3051     {
3052         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3053         shader_addline(buffer, "BRK (%s.x);\n", comp);
3054     }
3055 }
3056 
3057 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
3058 {
3059     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
3060     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3061     struct list *e = list_head(&priv->control_frames);
3062     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3063     const char *comp;
3064     char src_name0[50];
3065     char src_name1[50];
3066     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3067 
3068     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3069     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3070 
3071     if(vshader)
3072     {
3073         /* Invert the flag. We jump to the else label if the condition is NOT true */
3074         comp = get_compare(invert_compare(ins->flags));
3075         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3076         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
3077     }
3078     else
3079     {
3080         comp = get_compare(ins->flags);
3081         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3082         shader_addline(buffer, "IF %s.x;\n", comp);
3083     }
3084 }
3085 
3086 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3087 {
3088     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
3089     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3090     struct list *e = list_head(&priv->control_frames);
3091     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3092     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3093 
3094     if(vshader)
3095     {
3096         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3097         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3098         control_frame->had_else = TRUE;
3099     }
3100     else
3101     {
3102         shader_addline(buffer, "ELSE;\n");
3103     }
3104 }
3105 
3106 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3107 {
3108     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
3109     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3110     struct list *e = list_head(&priv->control_frames);
3111     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3112     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3113 
3114     if(vshader)
3115     {
3116         if(control_frame->had_else)
3117         {
3118             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3119         }
3120         else
3121         {
3122             shader_addline(buffer, "#No else branch. else is endif\n");
3123             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3124         }
3125     }
3126     else
3127     {
3128         shader_addline(buffer, "ENDIF;\n");
3129     }
3130 }
3131 
3132 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3133 {
3134     DWORD sampler_idx = ins->src[1].reg.idx[0].offset;
3135     char reg_dest[40];
3136     char reg_src[3][40];
3137     WORD flags = TEX_DERIV;
3138 
3139     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3140     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3141     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3142     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3143 
3144     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3145     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3146 
3147     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3148 }
3149 
3150 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3151 {
3152     DWORD sampler_idx = ins->src[1].reg.idx[0].offset;
3153     char reg_dest[40];
3154     char reg_coord[40];
3155     WORD flags = TEX_LOD;
3156 
3157     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3158     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3159 
3160     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3161     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3162 
3163     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3164 }
3165 
3166 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3167 {
3168     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
3169     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3170 
3171     priv->in_main_func = FALSE;
3172     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3173      * subroutine, don't generate a label that will make GL complain
3174      */
3175     if(priv->target_version == ARB) return;
3176 
3177     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx[0].offset);
3178 }
3179 
3180 static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
3181         const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
3182         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
3183         struct wined3d_string_buffer *buffer)
3184 {
3185     unsigned int i;
3186 
3187     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3188      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3189      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3190      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3191      */
3192     if (args->super.fog_src == VS_FOG_Z)
3193     {
3194         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3195     }
3196     else
3197     {
3198         if (!reg_maps->fog)
3199         {
3200             /* posFixup.x is always 1.0, so we can safely use it */
3201             shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3202         }
3203         else
3204         {
3205             /* Clamp fogcoord */
3206             const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
3207             const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
3208 
3209             shader_addline(buffer, "MIN TMP_FOGCOORD.x, TMP_FOGCOORD.x, %s;\n", one);
3210             shader_addline(buffer, "MAX result.fogcoord.x, TMP_FOGCOORD.x, %s;\n", zero);
3211         }
3212     }
3213 
3214     /* Clipplanes are always stored without y inversion */
3215     if (use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3216     {
3217         if (args->super.clip_enabled)
3218         {
3219             for (i = 0; i < priv_ctx->vs_clipplanes; i++)
3220             {
3221                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3222             }
3223         }
3224     }
3225     else if (args->clip.boolclip.clip_texcoord)
3226     {
3227         static const char component[4] = {'x', 'y', 'z', 'w'};
3228         unsigned int cur_clip = 0;
3229         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3230 
3231         for (i = 0; i < gl_info->limits.user_clip_distances; ++i)
3232         {
3233             if (args->clip.boolclip.clipplane_mask & (1u << i))
3234             {
3235                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3236                                component[cur_clip++], i);
3237             }
3238         }
3239         switch (cur_clip)
3240         {
3241             case 0:
3242                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3243                 break;
3244             case 1:
3245                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3246                 break;
3247             case 2:
3248                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3249                 break;
3250             case 3:
3251                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3252                 break;
3253         }
3254         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3255                        args->clip.boolclip.clip_texcoord - 1);
3256     }
3257 
3258     /* Write the final position.
3259      *
3260      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3261      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3262      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3263      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3264      */
3265     if (!gl_info->supported[ARB_CLIP_CONTROL])
3266     {
3267         shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3268         shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3269         shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3270 
3271         /* Z coord [0;1]->[-1;1] mapping, see comment in
3272          * get_projection_matrix() in utils.c. */
3273         if (need_helper_const(shader_data, reg_maps, gl_info))
3274         {
3275             const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3276             shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3277         }
3278         else
3279         {
3280             shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3281             shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3282         }
3283     }
3284 
3285     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3286 
3287     priv_ctx->footer_written = TRUE;
3288 }
3289 
3290 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3291 {
3292     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
3293     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3294     const struct wined3d_shader *shader = ins->ctx->shader;
3295     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3296 
3297     if(priv->target_version == ARB) return;
3298 
3299     if(vshader)
3300     {
3301         if (priv->in_main_func) vshader_add_footer(priv, shader->backend_data,
3302                 priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
3303     }
3304 
3305     shader_addline(buffer, "RET;\n");
3306 }
3307 
3308 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3309 {
3310     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
3311     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx[0].offset);
3312 }
3313 
3314 static BOOL shader_arb_compile(const struct wined3d_gl_info *gl_info, GLenum target, const char *src)
3315 {
3316     const char *ptr, *line;
3317     GLint native, pos;
3318 
3319     if (TRACE_ON(d3d_shader))
3320     {
3321         ptr = src;
3322         while ((line = get_line(&ptr))) TRACE_(d3d_shader)("    %.*s", (int)(ptr - line), line);
3323     }
3324 
3325     GL_EXTCALL(glProgramStringARB(target, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(src), src));
3326     checkGLcall("glProgramStringARB()");
3327 
3328     if (FIXME_ON(d3d_shader))
3329     {
3330         gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3331         if (pos != -1)
3332         {
3333             FIXME_(d3d_shader)("Program error at position %d: %s\n\n", pos,
3334                     debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3335             ptr = src;
3336             while ((line = get_line(&ptr))) FIXME_(d3d_shader)("    %.*s", (int)(ptr - line), line);
3337             FIXME_(d3d_shader)("\n");
3338 
3339             return FALSE;
3340         }
3341     }
3342 
3343     if (WARN_ON(d3d_perf))
3344     {
3345         GL_EXTCALL(glGetProgramivARB(target, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3346         checkGLcall("glGetProgramivARB()");
3347         if (!native)
3348             WARN_(d3d_perf)("Program exceeds native resource limits.\n");
3349     }
3350 
3351     return TRUE;
3352 }
3353 
3354 static void arbfp_add_sRGB_correction(struct wined3d_string_buffer *buffer, const char *fragcolor,
3355         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3356 {
3357     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3358 
3359     if(condcode)
3360     {
3361         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3362         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.x;\n", tmp1, fragcolor);
3363         /* Calculate the > 0.0031308 case */
3364         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts0.x;\n", fragcolor, fragcolor);
3365         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts0.x;\n", fragcolor, fragcolor);
3366         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts0.x;\n", fragcolor, fragcolor);
3367         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts0.y;\n", fragcolor, fragcolor);
3368         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts0.z;\n", fragcolor, fragcolor);
3369         /* Calculate the < case */
3370         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts0.w, %s;\n", fragcolor, fragcolor);
3371     }
3372     else
3373     {
3374         /* Calculate the > 0.0031308 case */
3375         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts0.x;\n", tmp1, fragcolor);
3376         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts0.x;\n", tmp1, fragcolor);
3377         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts0.x;\n", tmp1, fragcolor);
3378         shader_addline(buffer, "MUL %s, %s, srgb_consts0.y;\n", tmp1, tmp1);
3379         shader_addline(buffer, "SUB %s, %s, srgb_consts0.z;\n", tmp1, tmp1);
3380         /* Calculate the < case */
3381         shader_addline(buffer, "MUL %s, srgb_consts0.w, %s;\n", tmp2, fragcolor);
3382         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3383         shader_addline(buffer, "SLT %s, srgb_consts1.x, %s;\n", tmp3, fragcolor);
3384         shader_addline(buffer, "SGE %s, srgb_consts1.x, %s;\n", tmp4, fragcolor);
3385         /* Store the components > 0.0031308 in the destination */
3386         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3387         /* Add the components that are < 0.0031308 */
3388         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3389         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3390         * result.color writes(.rgb first, then .a), or handle overwriting already written
3391         * components. The assembler uses a temporary register in this case, which is usually
3392         * not allocated from one of our registers that were used earlier.
3393         */
3394     }
3395     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3396 }
3397 
3398 static const DWORD *find_loop_control_values(const struct wined3d_shader *shader, DWORD idx)
3399 {
3400     const struct wined3d_shader_lconst *constant;
3401 
3402     LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
3403     {
3404         if (constant->idx == idx)
3405         {
3406             return constant->value;
3407         }
3408     }
3409     return NULL;
3410 }
3411 
3412 static void init_ps_input(const struct wined3d_shader *shader,
3413         const struct arb_ps_compile_args *args, struct shader_arb_ctx_priv *priv)
3414 {
3415     static const char * const texcoords[8] =
3416     {
3417         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3418         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3419     };
3420     unsigned int i;
3421     const struct wined3d_shader_signature_element *input;
3422     const char *semantic_name;
3423     DWORD semantic_idx;
3424 
3425     switch(args->super.vp_mode)
3426     {
3427         case pretransformed:
3428         case fixedfunction:
3429             /* The pixelshader has to collect the varyings on its own. In any case properly load
3430              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3431              * other attribs to 0.0.
3432              *
3433              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3434              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3435              * load the texcoord attrib pointers to match the pixel shader signature
3436              */
3437             for (i = 0; i < shader->input_signature.element_count; ++i)
3438             {
3439                 input = &shader->input_signature.elements[i];
3440                 if (!(semantic_name = input->semantic_name))
3441                     continue;
3442                 semantic_idx = input->semantic_idx;
3443 
3444                 if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
3445                 {
3446                     if (!semantic_idx)
3447                         priv->ps_input[input->register_idx] = "fragment.color.primary";
3448                     else if (semantic_idx == 1)
3449                         priv->ps_input[input->register_idx] = "fragment.color.secondary";
3450                     else
3451                         priv->ps_input[input->register_idx] = "0.0";
3452                 }
3453                 else if (args->super.vp_mode == fixedfunction)
3454                 {
3455                     priv->ps_input[input->register_idx] = "0.0";
3456                 }
3457                 else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3458                 {
3459                     if (semantic_idx < 8)
3460                         priv->ps_input[input->register_idx] = texcoords[semantic_idx];
3461                     else
3462                         priv->ps_input[input->register_idx] = "0.0";
3463                 }
3464                 else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
3465                 {
3466                     if (!semantic_idx)
3467                         priv->ps_input[input->register_idx] = "fragment.fogcoord";
3468                     else
3469                         priv->ps_input[input->register_idx] = "0.0";
3470                 }
3471                 else
3472                 {
3473                     priv->ps_input[input->register_idx] = "0.0";
3474                 }
3475 
3476                 TRACE("v%u, semantic %s%u is %s\n", input->register_idx,
3477                         semantic_name, semantic_idx, priv->ps_input[input->register_idx]);
3478             }
3479             break;
3480 
3481         case vertexshader:
3482             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3483              * fragment.color
3484              */
3485             for(i = 0; i < 8; i++)
3486             {
3487                 priv->ps_input[i] = texcoords[i];
3488             }
3489             priv->ps_input[8] = "fragment.color.primary";
3490             priv->ps_input[9] = "fragment.color.secondary";
3491             break;
3492     }
3493 }
3494 
3495 static void arbfp_add_linear_fog(struct wined3d_string_buffer *buffer,
3496         const char *fragcolor, const char *tmp)
3497 {
3498     shader_addline(buffer, "SUB %s.x, state.fog.params.z, fragment.fogcoord.x;\n", tmp);
3499     shader_addline(buffer, "MUL_SAT %s.x, %s.x, state.fog.params.w;\n", tmp, tmp);
3500     shader_addline(buffer, "LRP %s.rgb, %s.x, %s, state.fog.color;\n", fragcolor, tmp, fragcolor);
3501 }
3502 
3503 /* Context activation is done by the caller. */
3504 static GLuint shader_arb_generate_pshader(const struct wined3d_shader *shader,
3505         const struct wined3d_gl_info *gl_info, struct wined3d_string_buffer *buffer,
3506         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3507 {
3508     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3509     GLuint retval;
3510     char fragcolor[16];
3511     DWORD next_local = 0;
3512     struct shader_arb_ctx_priv priv_ctx;
3513     BOOL dcl_td = FALSE;
3514     BOOL want_nv_prog = FALSE;
3515     struct arb_pshader_private *shader_priv = shader->backend_data;
3516     DWORD map;
3517     BOOL custom_linear_fog = FALSE;
3518 
3519     char srgbtmp[4][4];
3520     char ftoa_tmp[17];
3521     unsigned int i, found = 0;
3522 
3523     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3524     {
3525         if (!(map & 1)
3526                 || (shader->u.ps.color0_mov && i == shader->u.ps.color0_reg)
3527                 || (reg_maps->shader_version.major < 2 && !i))
3528             continue;
3529 
3530         sprintf(srgbtmp[found], "R%u", i);
3531         ++found;
3532         if (found == 4) break;
3533     }
3534 
3535     switch(found) {
3536         case 0:
3537             sprintf(srgbtmp[0], "TA");
3538             sprintf(srgbtmp[1], "TB");
3539             sprintf(srgbtmp[2], "TC");
3540             sprintf(srgbtmp[3], "TD");
3541             dcl_td = TRUE;
3542             break;
3543         case 1:
3544             sprintf(srgbtmp[1], "TA");
3545             sprintf(srgbtmp[2], "TB");
3546             sprintf(srgbtmp[3], "TC");
3547             break;
3548         case 2:
3549             sprintf(srgbtmp[2], "TA");
3550             sprintf(srgbtmp[3], "TB");
3551             break;
3552         case 3:
3553             sprintf(srgbtmp[3], "TA");
3554             break;
3555         case 4:
3556             break;
3557     }
3558 
3559     /*  Create the hw ARB shader */
3560     memset(&priv_ctx, 0, sizeof(priv_ctx));
3561     priv_ctx.cur_ps_args = args;
3562     priv_ctx.compiled_fprog = compiled;
3563     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3564     init_ps_input(shader, args, &priv_ctx);
3565     list_init(&priv_ctx.control_frames);
3566     priv_ctx.ps_post_process = args->super.srgb_correction;
3567 
3568     /* Avoid enabling NV_fragment_program* if we do not need it.
3569      *
3570      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3571      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3572      * is faster than what we gain from using higher native instructions. There are some things though
3573      * that cannot be emulated. In that case enable the extensions.
3574      * If the extension is enabled, instruction handlers that support both ways will use it.
3575      *
3576      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3577      * So enable the best we can get.
3578      */
3579     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3580        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3581     {
3582         want_nv_prog = TRUE;
3583     }
3584 
3585     shader_addline(buffer, "!!ARBfp1.0\n");
3586     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3587     {
3588         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3589         priv_ctx.target_version = NV3;
3590     }
3591     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3592     {
3593         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3594         priv_ctx.target_version = NV2;
3595     } else {
3596         if(want_nv_prog)
3597         {
3598             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3599              * limits properly
3600              */
3601             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3602             ERR("Try GLSL\n");
3603         }
3604         priv_ctx.target_version = ARB;
3605     }
3606 
3607     if (reg_maps->rt_mask > 1)
3608     {
3609         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3610     }
3611 
3612     if (reg_maps->shader_version.major < 3)
3613     {
3614         switch (args->super.fog)
3615         {
3616             case WINED3D_FFP_PS_FOG_OFF:
3617                 break;
3618             case WINED3D_FFP_PS_FOG_LINEAR:
3619                 if (gl_info->quirks & WINED3D_QUIRK_BROKEN_ARB_FOG)
3620                 {
3621                     custom_linear_fog = TRUE;
3622                     priv_ctx.ps_post_process = TRUE;
3623                     break;
3624                 }
3625                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3626                 break;
3627             case WINED3D_FFP_PS_FOG_EXP:
3628                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3629                 break;
3630             case WINED3D_FFP_PS_FOG_EXP2:
3631                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3632                 break;
3633         }
3634     }
3635 
3636     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3637      * unused temps away(but occupies them for the whole shader if they're used once). Always
3638      * declaring them avoids tricky bookkeeping work
3639      */
3640     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3641     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3642     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3643     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3644     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3645     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3646     wined3d_ftoa(eps, ftoa_tmp);
3647     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %s, 0.0 };\n", ftoa_tmp);
3648 
3649     if (reg_maps->shader_version.major < 2)
3650     {
3651         strcpy(fragcolor, "R0");
3652     }
3653     else
3654     {
3655         if (priv_ctx.ps_post_process)
3656         {
3657             if (shader->u.ps.color0_mov)
3658             {
3659                 sprintf(fragcolor, "R%u", shader->u.ps.color0_reg);
3660             }
3661             else
3662             {
3663                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3664                 strcpy(fragcolor, "TMP_COLOR");
3665             }
3666         } else {
3667             strcpy(fragcolor, "result.color");
3668         }
3669     }
3670 
3671     if (args->super.srgb_correction)
3672     {
3673         shader_addline(buffer, "PARAM srgb_consts0 = ");
3674         shader_arb_append_imm_vec4(buffer, wined3d_srgb_const0);
3675         shader_addline(buffer, ";\n");
3676         shader_addline(buffer, "PARAM srgb_consts1 = ");
3677         shader_arb_append_imm_vec4(buffer, wined3d_srgb_const1);
3678         shader_addline(buffer, ";\n");
3679     }
3680 
3681     /* Base Declarations */
3682     shader_generate_arb_declarations(shader, reg_maps, buffer, gl_info, NULL, &priv_ctx);
3683 
3684     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3685     {
3686         unsigned char bump_const;
3687 
3688         if (!(map & 1)) continue;
3689 
3690         bump_const = compiled->numbumpenvmatconsts;
3691         compiled->bumpenvmatconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3692         compiled->bumpenvmatconst[bump_const].texunit = i;
3693         compiled->luminanceconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3694         compiled->luminanceconst[bump_const].texunit = i;
3695 
3696         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3697          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3698          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3699          * textures due to conditional NP2 restrictions)
3700          *
3701          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3702          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3703          * their location is shader dependent anyway and they cannot be loaded globally.
3704          */
3705         compiled->bumpenvmatconst[bump_const].const_num = next_local++;
3706         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3707                        i, compiled->bumpenvmatconst[bump_const].const_num);
3708         compiled->numbumpenvmatconsts = bump_const + 1;
3709 
3710         if (!(reg_maps->luminanceparams & (1u << i)))
3711             continue;
3712 
3713         compiled->luminanceconst[bump_const].const_num = next_local++;
3714         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3715                        i, compiled->luminanceconst[bump_const].const_num);
3716     }
3717 
3718     for (i = 0; i < WINED3D_MAX_CONSTS_I; ++i)
3719     {
3720         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3721         if (reg_maps->integer_constants & (1u << i) && priv_ctx.target_version >= NV2)
3722         {
3723             const DWORD *control_values = find_loop_control_values(shader, i);
3724 
3725             if(control_values)
3726             {
3727                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3728                                 control_values[0], control_values[1], control_values[2]);
3729             }
3730             else
3731             {
3732                 compiled->int_consts[i] = next_local;
3733                 compiled->num_int_consts++;
3734                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3735             }
3736         }
3737     }
3738 
3739     if(reg_maps->vpos || reg_maps->usesdsy)
3740     {
3741         compiled->ycorrection = next_local;
3742         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3743 
3744         if(reg_maps->vpos)
3745         {
3746             shader_addline(buffer, "TEMP vpos;\n");
3747             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3748              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3749              * ycorrection.z: 1.0
3750              * ycorrection.w: 0.0
3751              */
3752             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3753             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3754         }
3755     }
3756     else
3757     {
3758         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3759     }
3760 
3761     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3762      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3763      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3764      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3765      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3766      * shader compilation errors and the subsequent errors when drawing with this shader. */
3767     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3768         unsigned char cur_fixup_sampler = 0;
3769 
3770         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3771         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3772         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3773 
3774         fixup->offset = next_local;
3775         fixup->super.active = 0;
3776 
3777         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i)
3778         {
3779             if (!(map & (1u << i)))
3780                 continue;
3781 
3782             if (fixup->offset + (cur_fixup_sampler >> 1) < max_lconsts)
3783             {
3784                 fixup->super.active |= (1u << i);
3785                 fixup->super.idx[i] = cur_fixup_sampler++;
3786             }
3787             else
3788             {
3789                 FIXME("No free constant found to load NP2 fixup data into shader. "
3790                       "Sampling from this texture will probably look wrong.\n");
3791                 break;
3792             }
3793         }
3794 
3795         fixup->super.num_consts = (cur_fixup_sampler + 1) >> 1;
3796         if (fixup->super.num_consts) {
3797             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3798                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3799         }
3800     }
3801 
3802     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3803     {
3804         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3805     }
3806 
3807     /* Base Shader Body */
3808     if (FAILED(shader_generate_code(shader, buffer, reg_maps, &priv_ctx, NULL, NULL)))
3809         return 0;
3810 
3811     if(args->super.srgb_correction) {
3812         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3813                                   priv_ctx.target_version >= NV2);
3814     }
3815 
3816     if (custom_linear_fog)
3817         arbfp_add_linear_fog(buffer, fragcolor, "TA");
3818 
3819     if(strcmp(fragcolor, "result.color")) {
3820         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3821     }
3822     shader_addline(buffer, "END\n");
3823 
3824     /* TODO: change to resource.glObjectHandle or something like that */
3825     GL_EXTCALL(glGenProgramsARB(1, &retval));
3826 
3827     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3828     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3829 
3830     TRACE("Created hw pixel shader, prg=%d\n", retval);
3831     if (!shader_arb_compile(gl_info, GL_FRAGMENT_PROGRAM_ARB, buffer->buffer))
3832         return 0;
3833 
3834     return retval;
3835 }
3836 
3837 static int compare_sig(const struct wined3d_shader_signature *sig1, const struct wined3d_shader_signature *sig2)
3838 {
3839     unsigned int i;
3840     int ret;
3841 
3842     if (sig1->element_count != sig2->element_count)
3843         return sig1->element_count < sig2->element_count ? -1 : 1;
3844 
3845     for (i = 0; i < sig1->element_count; ++i)
3846     {
3847         const struct wined3d_shader_signature_element *e1, *e2;
3848 
3849         e1 = &sig1->elements[i];
3850         e2 = &sig2->elements[i];
3851 
3852         if (!e1->semantic_name || !e2->semantic_name)
3853         {
3854             /* Compare pointers, not contents. One string is NULL (element
3855              * does not exist), the other one is not NULL. */
3856             if (e1->semantic_name != e2->semantic_name)
3857                 return e1->semantic_name < e2->semantic_name ? -1 : 1;
3858             continue;
3859         }
3860 
3861         if ((ret = strcmp(e1->semantic_name, e2->semantic_name)))
3862             return ret;
3863         if (e1->semantic_idx != e2->semantic_idx)
3864             return e1->semantic_idx < e2->semantic_idx ? -1 : 1;
3865         if (e1->sysval_semantic != e2->sysval_semantic)
3866             return e1->sysval_semantic < e2->sysval_semantic ? -1 : 1;
3867         if (e1->component_type != e2->component_type)
3868             return e1->component_type < e2->component_type ? -1 : 1;
3869         if (e1->register_idx != e2->register_idx)
3870             return e1->register_idx < e2->register_idx ? -1 : 1;
3871         if (e1->mask != e2->mask)
3872             return e1->mask < e2->mask ? -1 : 1;
3873     }
3874     return 0;
3875 }
3876 
3877 static void clone_sig(struct wined3d_shader_signature *new, const struct wined3d_shader_signature *sig)
3878 {
3879     unsigned int i;
3880     char *name;
3881 
3882     new->element_count = sig->element_count;
3883     new->elements = heap_calloc(new->element_count, sizeof(*new->elements));
3884     for (i = 0; i < sig->element_count; ++i)
3885     {
3886         new->elements[i] = sig->elements[i];
3887 
3888         if (!new->elements[i].semantic_name)
3889             continue;
3890 
3891         /* Clone the semantic string */
3892         name = heap_alloc(strlen(sig->elements[i].semantic_name) + 1);
3893         strcpy(name, sig->elements[i].semantic_name);
3894         new->elements[i].semantic_name = name;
3895     }
3896 }
3897 
3898 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature *sig)
3899 {
3900     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3901     struct ps_signature *found_sig;
3902 
3903     if (entry)
3904     {
3905         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3906         TRACE("Found existing signature %u\n", found_sig->idx);
3907         return found_sig->idx;
3908     }
3909     found_sig = heap_alloc_zero(sizeof(*found_sig));
3910     clone_sig(&found_sig->sig, sig);
3911     found_sig->idx = priv->ps_sig_number++;
3912     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3913     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3914     {
3915         ERR("Failed to insert program entry.\n");
3916     }
3917     return found_sig->idx;
3918 }
3919 
3920 static void init_output_registers(const struct wined3d_shader *shader,
3921         const struct wined3d_shader_signature *ps_input_sig,
3922         struct shader_arb_ctx_priv *priv_ctx, struct arb_vs_compiled_shader *compiled)
3923 {
3924     unsigned int i, j;
3925     static const char * const texcoords[8] =
3926     {
3927         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3928         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3929     };
3930     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3931      * and varying 9 to result.color.secondary
3932      */
3933     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
3934     {
3935         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3936         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
3937         "result.color.primary", "result.color.secondary"
3938     };
3939 
3940     if (!ps_input_sig)
3941     {
3942         TRACE("Pixel shader uses builtin varyings\n");
3943         /* Map builtins to builtins */
3944         for(i = 0; i < 8; i++)
3945         {
3946             priv_ctx->texcrd_output[i] = texcoords[i];
3947         }
3948         priv_ctx->color_output[0] = "result.color.primary";
3949         priv_ctx->color_output[1] = "result.color.secondary";
3950         priv_ctx->fog_output = "TMP_FOGCOORD";
3951 
3952         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3953         for (i = 0; i < shader->output_signature.element_count; ++i)
3954         {
3955             const struct wined3d_shader_signature_element *output = &shader->output_signature.elements[i];
3956 
3957             if (!output->semantic_name)
3958                 continue;
3959 
3960             if (shader_match_semantic(output->semantic_name, WINED3D_DECL_USAGE_POSITION))
3961             {
3962                 TRACE("o%u is TMP_OUT\n", output->register_idx);
3963                 if (!output->semantic_idx)
3964                     priv_ctx->vs_output[output->register_idx] = "TMP_OUT";
3965                 else
3966                     priv_ctx->vs_output[output->register_idx] = "TA";
3967             }
3968             else if (shader_match_semantic(output->semantic_name, WINED3D_DECL_USAGE_PSIZE))
3969             {
3970                 TRACE("o%u is result.pointsize\n", output->register_idx);
3971                 if (!output->semantic_idx)
3972                     priv_ctx->vs_output[output->register_idx] = "result.pointsize";
3973                 else
3974                     priv_ctx->vs_output[output->register_idx] = "TA";
3975             }
3976             else if (shader_match_semantic(output->semantic_name, WINED3D_DECL_USAGE_COLOR))
3977             {
3978                 TRACE("o%u is result.color.?, idx %u\n", output->register_idx, output->semantic_idx);
3979                 if (!output->semantic_idx)
3980                     priv_ctx->vs_output[output->register_idx] = "result.color.primary";
3981                 else if (output->semantic_idx == 1)
3982                     priv_ctx->vs_output[output->register_idx] = "result.color.secondary";
3983                 else priv_ctx->vs_output[output->register_idx] = "TA";
3984             }
3985             else if (shader_match_semantic(output->semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3986             {
3987                 TRACE("o%u is result.texcoord[%u]\n", output->register_idx, output->semantic_idx);
3988                 if (output->semantic_idx >= 8)
3989                     priv_ctx->vs_output[output->register_idx] = "TA";
3990                 else
3991                     priv_ctx->vs_output[output->register_idx] = texcoords[output->semantic_idx];
3992             }
3993             else if (shader_match_semantic(output->semantic_name, WINED3D_DECL_USAGE_FOG))
3994             {
3995                 TRACE("o%u is result.fogcoord\n", output->register_idx);
3996                 if (output->semantic_idx > 0)
3997                     priv_ctx->vs_output[output->register_idx] = "TA";
3998                 else
3999                     priv_ctx->vs_output[output->register_idx] = "result.fogcoord";
4000             }
4001             else
4002             {
4003                 priv_ctx->vs_output[output->register_idx] = "TA";
4004             }
4005         }
4006         return;
4007     }
4008 
4009     TRACE("Pixel shader uses declared varyings\n");
4010 
4011     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
4012     for(i = 0; i < 8; i++)
4013     {
4014         priv_ctx->texcrd_output[i] = "TA";
4015     }
4016     priv_ctx->color_output[0] = "TA";
4017     priv_ctx->color_output[1] = "TA";
4018     priv_ctx->fog_output = "TA";
4019 
4020     for (i = 0; i < ps_input_sig->element_count; ++i)
4021     {
4022         const struct wined3d_shader_signature_element *input = &ps_input_sig->elements[i];
4023 
4024         if (!input->semantic_name)
4025             continue;
4026 
4027         /* If a declared input register is not written by builtin arguments, don't write to it.
4028          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
4029          *
4030          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4031          * to TMP_OUT in any case
4032          */
4033         if (shader_match_semantic(input->semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4034         {
4035             if (input->semantic_idx < 8)
4036                 priv_ctx->texcrd_output[input->semantic_idx] = decl_idx_to_string[input->register_idx];
4037         }
4038         else if (shader_match_semantic(input->semantic_name, WINED3D_DECL_USAGE_COLOR))
4039         {
4040             if (input->semantic_idx < 2)
4041                 priv_ctx->color_output[input->semantic_idx] = decl_idx_to_string[input->register_idx];
4042         }
4043         else if (shader_match_semantic(input->semantic_name, WINED3D_DECL_USAGE_FOG))
4044         {
4045             if (!input->semantic_idx)
4046                 priv_ctx->fog_output = decl_idx_to_string[input->register_idx];
4047         }
4048         else
4049         {
4050             continue;
4051         }
4052 
4053         if (!strcmp(decl_idx_to_string[input->register_idx], "result.color.primary")
4054                 || !strcmp(decl_idx_to_string[input->register_idx], "result.color.secondary"))
4055         {
4056             compiled->need_color_unclamp = TRUE;
4057         }
4058     }
4059 
4060     /* Map declared to declared */
4061     for (i = 0; i < shader->output_signature.element_count; ++i)
4062     {
4063         const struct wined3d_shader_signature_element *output = &shader->output_signature.elements[i];
4064 
4065         /* Write unread output to TA to throw them away */
4066         priv_ctx->vs_output[output->register_idx] = "TA";
4067 
4068         if (!output->semantic_name)
4069             continue;
4070 
4071         if (shader_match_semantic(output->semantic_name, WINED3D_DECL_USAGE_POSITION) && !output->semantic_idx)
4072         {
4073             priv_ctx->vs_output[output->register_idx] = "TMP_OUT";
4074             continue;
4075         }
4076         else if (shader_match_semantic(output->semantic_name, WINED3D_DECL_USAGE_PSIZE) && !output->semantic_idx)
4077         {
4078             priv_ctx->vs_output[output->register_idx] = "result.pointsize";
4079             continue;
4080         }
4081 
4082         for (j = 0; j < ps_input_sig->element_count; ++j)
4083         {
4084             const struct wined3d_shader_signature_element *input = &ps_input_sig->elements[j];
4085 
4086             if (!input->semantic_name)
4087                 continue;
4088 
4089             if (!strcmp(input->semantic_name, output->semantic_name)
4090                     && input->semantic_idx == output->semantic_idx)
4091             {
4092                 priv_ctx->vs_output[output->register_idx] = decl_idx_to_string[input->register_idx];
4093 
4094                 if (!strcmp(priv_ctx->vs_output[output->register_idx], "result.color.primary")
4095                         || !strcmp(priv_ctx->vs_output[output->register_idx], "result.color.secondary"))
4096                 {
4097                     compiled->need_color_unclamp = TRUE;
4098                 }
4099             }
4100         }
4101     }
4102 }
4103 
4104 /* Context activation is done by the caller. */
4105 static GLuint shader_arb_generate_vshader(const struct wined3d_shader *shader,
4106         const struct wined3d_gl_info *gl_info, struct wined3d_string_buffer *buffer,
4107         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled,
4108         const struct wined3d_shader_signature *ps_input_sig)
4109 {
4110     const struct arb_vshader_private *shader_data = shader->backend_data;
4111     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4112     struct shader_arb_priv *priv = shader->device->shader_priv;
4113     GLuint ret;
4114     DWORD next_local = 0;
4115     struct shader_arb_ctx_priv priv_ctx;
4116     unsigned int i;
4117 
4118     memset(&priv_ctx, 0, sizeof(priv_ctx));
4119     priv_ctx.cur_vs_args = args;
4120     list_init(&priv_ctx.control_frames);
4121     init_output_registers(shader, ps_input_sig, &priv_ctx, compiled);
4122 
4123     /*  Create the hw ARB shader */
4124     shader_addline(buffer, "!!ARBvp1.0\n");
4125 
4126     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4127      * mesurable performance penalty, and we can always make use of it for clipplanes.
4128      */
4129     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4130     {
4131         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4132         priv_ctx.target_version = NV3;
4133         shader_addline(buffer, "ADDRESS aL;\n");
4134     }
4135     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4136     {
4137         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4138         priv_ctx.target_version = NV2;
4139         shader_addline(buffer, "ADDRESS aL;\n");
4140     } else {
4141         priv_ctx.target_version = ARB;
4142     }
4143 
4144     shader_addline(buffer, "TEMP TMP_OUT;\n");
4145     if (reg_maps->fog)
4146         shader_addline(buffer, "TEMP TMP_FOGCOORD;\n");
4147     if (need_helper_const(shader_data, reg_maps, gl_info))
4148     {
4149         char ftoa_tmp[17];
4150         wined3d_ftoa(eps, ftoa_tmp);
4151         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %s};\n", ftoa_tmp);
4152     }
4153     if (need_rel_addr_const(shader_data, reg_maps, gl_info))
4154     {
4155         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", shader_data->rel_offset);
4156         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4157     }
4158 
4159     shader_addline(buffer, "TEMP TA;\n");
4160     shader_addline(buffer, "TEMP TB;\n");
4161 
4162     /* Base Declarations */
4163     shader_generate_arb_declarations(shader, reg_maps, buffer, gl_info,
4164             &priv_ctx.vs_clipplanes, &priv_ctx);
4165 
4166     for (i = 0; i < WINED3D_MAX_CONSTS_I; ++i)
4167     {
4168         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4169         if (reg_maps->integer_constants & (1u << i) && priv_ctx.target_version >= NV2)
4170         {
4171             const DWORD *control_values = find_loop_control_values(shader, i);
4172 
4173             if(control_values)
4174             {
4175                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4176                                 control_values[0], control_values[1], control_values[2]);
4177             }
4178             else
4179             {
4180                 compiled->int_consts[i] = next_local;
4181                 compiled->num_int_consts++;
4182                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4183             }
4184         }
4185     }
4186 
4187     /* We need a constant to fixup the final position */
4188     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4189     compiled->pos_fixup = next_local++;
4190 
4191     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4192      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4193      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4194      * a replacement shader depend on the texcoord.w being set properly.
4195      *
4196      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4197      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4198      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4199      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4200      * this can eat a number of instructions, so skip it unless this cap is set as well
4201      */
4202     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4203     {
4204         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4205         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4206 
4207         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !priv->ffp_proj_control)
4208         {
4209             int i;
4210             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4211             for(i = 0; i < MAX_REG_TEXCRD; i++)
4212             {
4213                 if (reg_maps->u.texcoord_mask[i] && reg_maps->u.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4214                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4215             }
4216         }
4217     }
4218 
4219     /* The shader starts with the main function */
4220     priv_ctx.in_main_func = TRUE;
4221     /* Base Shader Body */
4222     if (FAILED(shader_generate_code(shader, buffer, reg_maps, &priv_ctx, NULL, NULL)))
4223         return -1;
4224 
4225     if (!priv_ctx.footer_written) vshader_add_footer(&priv_ctx,
4226             shader_data, args, reg_maps, gl_info, buffer);
4227 
4228     shader_addline(buffer, "END\n");
4229 
4230     /* TODO: change to resource.glObjectHandle or something like that */
4231     GL_EXTCALL(glGenProgramsARB(1, &ret));
4232 
4233     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4234     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4235 
4236     TRACE("Created hw vertex shader, prg=%d\n", ret);
4237     if (!shader_arb_compile(gl_info, GL_VERTEX_PROGRAM_ARB, buffer->buffer))
4238         return -1;
4239 
4240     return ret;
4241 }
4242 
4243 /* Context activation is done by the caller. */
4244 static struct arb_ps_compiled_shader *find_arb_pshader(struct wined3d_shader *shader,
4245         const struct arb_ps_compile_args *args)
4246 {
4247     struct wined3d_device *device = shader->device;
4248     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4249     const struct wined3d_d3d_info *d3d_info = &device->adapter->d3d_info;
4250     UINT i;
4251     DWORD new_size;
4252     struct arb_ps_compiled_shader *new_array;
4253     struct wined3d_string_buffer buffer;
4254     struct arb_pshader_private *shader_data;
4255     GLuint ret;
4256 
4257     if (!shader->backend_data)
4258     {
4259         struct shader_arb_priv *priv = device->shader_priv;
4260 
4261         shader->backend_data = heap_alloc_zero(sizeof(*shader_data));
4262         shader_data = shader->backend_data;
4263         shader_data->clamp_consts = shader->reg_maps.shader_version.major == 1;
4264 
4265         if (shader->reg_maps.shader_version.major < 3)
4266             shader_data->input_signature_idx = ~0U;
4267         else
4268             shader_data->input_signature_idx = find_input_signature(priv, &shader->input_signature);
4269 
4270         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4271 
4272         if (!d3d_info->vs_clipping)
4273             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->reg_maps,
4274                     d3d_info->limits.ffp_blend_stages - 1);
4275         else
4276             shader_data->clipplane_emulation = ~0U;
4277     }
4278     shader_data = shader->backend_data;
4279 
4280     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4281      * so a linear search is more performant than a hashmap or a binary search
4282      * (cache coherency etc)
4283      */
4284     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4285     {
4286         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4287             return &shader_data->gl_shaders[i];
4288     }
4289 
4290     TRACE("No matching GL shader found, compiling a new shader\n");
4291     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4292         if (shader_data->num_gl_shaders)
4293         {
4294             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4295             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4296                                     new_size * sizeof(*shader_data->gl_shaders));
4297         }
4298         else
4299         {
4300             new_array = heap_alloc_zero(sizeof(*shader_data->gl_shaders));
4301             new_size = 1;
4302         }
4303 
4304         if(!new_array) {
4305             ERR("Out of memory\n");
4306             return 0;
4307         }
4308         shader_data->gl_shaders = new_array;
4309         shader_data->shader_array_size = new_size;
4310     }
4311 
4312     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4313 
4314     pixelshader_update_resource_types(shader, args->super.tex_types);
4315 
4316     if (!string_buffer_init(&buffer))
4317     {
4318         ERR("Failed to initialize shader buffer.\n");
4319         return 0;
4320     }
4321 
4322     ret = shader_arb_generate_pshader(shader, gl_info, &buffer, args,
4323             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4324     string_buffer_free(&buffer);
4325     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4326 
4327     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4328 }
4329 
4330 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4331                                  const DWORD use_map, BOOL skip_int) {
4332     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4333     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4334     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4335     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4336     if(stored->ps_signature != new->ps_signature) return FALSE;
4337     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4338     if(skip_int) return TRUE;
4339 
4340     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4341 }
4342 
4343 static struct arb_vs_compiled_shader *find_arb_vshader(struct wined3d_shader *shader,
4344         const struct wined3d_gl_info *gl_info, DWORD use_map, const struct arb_vs_compile_args *args,
4345         const struct wined3d_shader_signature *ps_input_sig)
4346 {
4347     UINT i;
4348     DWORD new_size;
4349     struct arb_vs_compiled_shader *new_array;
4350     struct wined3d_string_buffer buffer;
4351     struct arb_vshader_private *shader_data;
4352     GLuint ret;
4353 
4354     if (!shader->backend_data)
4355     {
4356         const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4357 
4358         shader->backend_data = heap_alloc_zero(sizeof(*shader_data));
4359         shader_data = shader->backend_data;
4360 
4361         if ((gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
4362                 && reg_maps->min_rel_offset <= reg_maps->max_rel_offset)
4363         {
4364             if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 127)
4365             {
4366                 FIXME("The difference between the minimum and maximum relative offset is > 127.\n");
4367                 FIXME("Which this OpenGL implementation does not support. Try using GLSL.\n");
4368                 FIXME("Min: %u, Max: %u.\n", reg_maps->min_rel_offset, reg_maps->max_rel_offset);
4369             }
4370             else if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 63)
4371                 shader_data->rel_offset = reg_maps->min_rel_offset + 63;
4372             else if (reg_maps->max_rel_offset > 63)
4373                 shader_data->rel_offset = reg_maps->min_rel_offset;
4374         }
4375     }
4376     shader_data = shader->backend_data;
4377 
4378     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4379      * so a linear search is more performant than a hashmap or a binary search
4380      * (cache coherency etc)
4381      */
4382     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4383         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4384                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4385         {
4386             return &shader_data->gl_shaders[i];
4387         }
4388     }
4389 
4390     TRACE("No matching GL shader found, compiling a new shader\n");
4391 
4392     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4393         if (shader_data->num_gl_shaders)
4394         {
4395             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4396             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4397                                     new_size * sizeof(*shader_data->gl_shaders));
4398         }
4399         else
4400         {
4401             new_array = heap_alloc_zero(sizeof(*shader_data->gl_shaders));
4402             new_size = 1;
4403         }
4404 
4405         if(!new_array) {
4406             ERR("Out of memory\n");
4407             return 0;
4408         }
4409         shader_data->gl_shaders = new_array;
4410         shader_data->shader_array_size = new_size;
4411     }
4412 
4413     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4414 
4415     if (!string_buffer_init(&buffer))
4416     {
4417         ERR("Failed to initialize shader buffer.\n");
4418         return 0;
4419     }
4420 
4421     ret = shader_arb_generate_vshader(shader, gl_info, &buffer, args,
4422             &shader_data->gl_shaders[shader_data->num_gl_shaders],
4423             ps_input_sig);
4424     string_buffer_free(&buffer);
4425     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4426 
4427     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4428 }
4429 
4430 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4431         const struct wined3d_context *context, const struct wined3d_shader *shader,
4432         struct arb_ps_compile_args *args)
4433 {
4434     const struct wined3d_gl_info *gl_info = context->gl_info;
4435     const struct wined3d_d3d_info *d3d_info = context->d3d_info;
4436     int i;
4437     WORD int_skip;
4438 
4439     find_ps_compile_args(state, shader, context->stream_info.position_transformed, &args->super, context);
4440 
4441     /* This forces all local boolean constants to 1 to make them stateblock independent */
4442     args->bools = shader->reg_maps.local_bool_consts;
4443 
4444     for (i = 0; i < WINED3D_MAX_CONSTS_B; ++i)
4445     {
4446         if (state->ps_consts_b[i])
4447             args->bools |= ( 1u << i);
4448     }
4449 
4450     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4451      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4452      * duplicate the shader than have a no-op KIL instruction in every shader
4453      */
4454     if (!d3d_info->vs_clipping && use_vs(state)
4455             && state->render_states[WINED3D_RS_CLIPPING]
4456             && state->render_states[WINED3D_RS_CLIPPLANEENABLE])
4457         args->clip = 1;
4458     else
4459         args->clip = 0;
4460 
4461     /* Skip if unused or local, or supported natively */
4462     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4463     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4464     {
4465         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4466         return;
4467     }
4468 
4469     for (i = 0; i < WINED3D_MAX_CONSTS_I; ++i)
4470     {
4471         if (int_skip & (1u << i))
4472         {
4473             args->loop_ctrl[i][0] = 0;
4474             args->loop_ctrl[i][1] = 0;
4475             args->loop_ctrl[i][2] = 0;
4476         }
4477         else
4478         {
4479             args->loop_ctrl[i][0] = state->ps_consts_i[i].x;
4480             args->loop_ctrl[i][1] = state->ps_consts_i[i].y;
4481             args->loop_ctrl[i][2] = state->ps_consts_i[i].z;
4482         }
4483     }
4484 }
4485 
4486 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4487         const struct wined3d_context *context, const struct wined3d_shader *shader,
4488         struct arb_vs_compile_args *args)
4489 {
4490     const struct wined3d_device *device = shader->device;
4491     const struct wined3d_adapter *adapter = device->adapter;
4492     const struct wined3d_gl_info *gl_info = context->gl_info;
4493     const struct wined3d_d3d_info *d3d_info = context->d3d_info;
4494     int i;
4495     WORD int_skip;
4496 
4497     find_vs_compile_args(state, shader, context->stream_info.swizzle_map, &args->super, context);
4498 
4499     args->clip.boolclip_compare = 0;
4500     if (use_ps(state))
4501     {
4502         const struct wined3d_shader *ps = state->shader[WINED3D_SHADER_TYPE_PIXEL];
4503         const struct arb_pshader_private *shader_priv = ps->backend_data;
4504         args->ps_signature = shader_priv->input_signature_idx;
4505 
4506         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4507     }
4508     else
4509     {
4510         args->ps_signature = ~0;
4511         if (!d3d_info->vs_clipping && adapter->fragment_pipe == &arbfp_fragment_pipeline)
4512             args->clip.boolclip.clip_texcoord = ffp_clip_emul(context) ? d3d_info->limits.ffp_blend_stages : 0;
4513         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4514     }
4515 
4516     if (args->clip.boolclip.clip_texcoord)
4517     {
4518         if (state->render_states[WINED3D_RS_CLIPPING])
4519             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3D_RS_CLIPPLANEENABLE];
4520         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4521     }
4522 
4523     /* This forces all local boolean constants to 1 to make them stateblock independent */
4524     args->clip.boolclip.bools = shader->reg_maps.local_bool_consts;
4525     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4526     for (i = 0; i < WINED3D_MAX_CONSTS_B; ++i)
4527     {
4528         if (state->vs_consts_b[i])
4529             args->clip.boolclip.bools |= (1u << i);
4530     }
4531 
4532     args->vertex.samplers[0] = context->tex_unit_map[MAX_FRAGMENT_SAMPLERS + 0];
4533     args->vertex.samplers[1] = context->tex_unit_map[MAX_FRAGMENT_SAMPLERS + 1];
4534     args->vertex.samplers[2] = context->tex_unit_map[MAX_FRAGMENT_SAMPLERS + 2];
4535     args->vertex.samplers[3] = 0;
4536 
4537     /* Skip if unused or local */
4538     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4539     /* This is about flow control, not clipping. */
4540     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4541     {
4542         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4543         return;
4544     }
4545 
4546     for (i = 0; i < WINED3D_MAX_CONSTS_I; ++i)
4547     {
4548         if (int_skip & (1u << i))
4549         {
4550             args->loop_ctrl[i][0] = 0;
4551             args->loop_ctrl[i][1] = 0;
4552             args->loop_ctrl[i][2] = 0;
4553         }
4554         else
4555         {
4556             args->loop_ctrl[i][0] = state->vs_consts_i[i].x;
4557             args->loop_ctrl[i][1] = state->vs_consts_i[i].y;
4558             args->loop_ctrl[i][2] = state->vs_consts_i[i].z;
4559         }
4560     }
4561 }
4562 
4563 /* Context activation is done by the caller. */
4564 static void shader_arb_select(void *shader_priv, struct wined3d_context *context,
4565         const struct wined3d_state *state)
4566 {
4567     struct shader_arb_priv *priv = shader_priv;
4568     const struct wined3d_gl_info *gl_info = context->gl_info;
4569     int i;
4570 
4571     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4572     if (use_ps(state))
4573     {
4574         struct wined3d_shader *ps = state->shader[WINED3D_SHADER_TYPE_PIXEL];
4575         struct arb_ps_compile_args compile_args;
4576         struct arb_ps_compiled_shader *compiled;
4577 
4578         TRACE("Using pixel shader %p.\n", ps);
4579         find_arb_ps_compile_args(state, context, ps, &compile_args);
4580         compiled = find_arb_pshader(ps, &compile_args);
4581         priv->current_fprogram_id = compiled->prgId;
4582         priv->compiled_fprog = compiled;
4583 
4584         /* Bind the fragment program */
4585         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4586         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4587 
4588         if (!priv->use_arbfp_fixed_func)
4589             priv->fragment_pipe->enable_extension(gl_info, FALSE);
4590 
4591         /* Enable OpenGL fragment programs. */
4592         gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
4593         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4594 
4595         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", priv->current_fprogram_id);
4596 
4597         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4598          * a 1.x and newer shader, reload the first 8 constants
4599          */
4600         if (priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->backend_data)->clamp_consts)
4601         {
4602             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->backend_data)->clamp_consts;
4603             priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, 8);
4604             for(i = 0; i < 8; i++)
4605             {
4606                 priv->pshader_const_dirty[i] = 1;
4607             }
4608             /* Also takes care of loading local constants */
4609             shader_arb_load_constants_internal(shader_priv, context, state, TRUE, FALSE, TRUE);
4610         }
4611         else
4612         {
4613             UINT rt_height = state->fb->render_targets[0]->height;
4614             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4615         }
4616 
4617         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4618         if (compiled->np2fixup_info.super.active)
4619             context->constant_update_mask |= WINED3D_SHADER_CONST_PS_NP2_FIXUP;
4620 
4621         if (ps->load_local_constsF)
4622             context->constant_update_mask |= WINED3D_SHADER_CONST_PS_F;
4623     }
4624     else
4625     {
4626         if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4627         {
4628             /* Disable only if we're not using arbfp fixed function fragment
4629              * processing. If this is used, keep GL_FRAGMENT_PROGRAM_ARB
4630              * enabled, and the fixed function pipeline will bind the fixed
4631              * function replacement shader. */
4632             gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4633             checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4634             priv->current_fprogram_id = 0;
4635         }
4636         priv->fragment_pipe->enable_extension(gl_info, TRUE);
4637     }
4638 
4639     if (use_vs(state))
4640     {
4641         struct wined3d_shader *vs = state->shader[WINED3D_SHADER_TYPE_VERTEX];
4642         struct arb_vs_compile_args compile_args;
4643         struct arb_vs_compiled_shader *compiled;
4644         const struct wined3d_shader_signature *ps_input_sig;
4645 
4646         TRACE("Using vertex shader %p\n", vs);
4647         find_arb_vs_compile_args(state, context, vs, &compile_args);
4648 
4649         /* Instead of searching for the signature in the signature list, read the one from the
4650          * current pixel shader. It's maybe not the shader where the signature came from, but it
4651          * is the same signature and faster to find. */
4652         if (compile_args.ps_signature == ~0U)
4653             ps_input_sig = NULL;
4654         else
4655             ps_input_sig = &state->shader[WINED3D_SHADER_TYPE_PIXEL]->input_signature;
4656 
4657         compiled = find_arb_vshader(vs, context->gl_info, context->stream_info.use_map,
4658                 &compile_args, ps_input_sig);
4659         priv->current_vprogram_id = compiled->prgId;
4660         priv->compiled_vprog = compiled;
4661 
4662         /* Bind the vertex program */
4663         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4664         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4665 
4666         priv->vertex_pipe->vp_enable(gl_info, FALSE);
4667 
4668         /* Enable OpenGL vertex programs */
4669         gl_info->gl_ops.gl.p_glEnable(GL_VERTEX_PROGRAM_ARB);
4670         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4671         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", priv->current_vprogram_id);
4672         shader_arb_vs_local_constants(compiled, context, state);
4673 
4674         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4675             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4676 
4677             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4678             {
4679                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4680                 checkGLcall("glClampColorARB");
4681             } else {
4682                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4683             }
4684         }
4685 
4686         if (vs->load_local_constsF)
4687             context->constant_update_mask |= WINED3D_SHADER_CONST_VS_F;
4688     }
4689     else
4690     {
4691         if (gl_info->supported[ARB_VERTEX_PROGRAM])
4692         {
4693             priv->current_vprogram_id = 0;
4694             gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4695             checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4696         }
4697         priv->vertex_pipe->vp_enable(gl_info, TRUE);
4698     }
4699 }
4700 
4701 static void shader_arb_select_compute(void *shader_priv, struct wined3d_context *context,
4702         const struct wined3d_state *state)
4703 {
4704     ERR("Compute pipeline not supported by the ARB shader backend.\n");
4705 }
4706 
4707 /* Context activation is done by the caller. */
4708 static void shader_arb_disable(void *shader_priv, struct wined3d_context *context)
4709 {
4710     const struct wined3d_gl_info *gl_info = context->gl_info;
4711     struct shader_arb_priv *priv = shader_priv;
4712 
4713     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4714     {
4715         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4716         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4717         priv->current_fprogram_id = 0;
4718     }
4719     priv->fragment_pipe->enable_extension(gl_info, FALSE);
4720 
4721     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4722     {
4723         priv->current_vprogram_id = 0;
4724         gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4725         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4726     }
4727     priv->vertex_pipe->vp_enable(gl_info, FALSE);
4728 
4729     if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT] && priv->last_vs_color_unclamp)
4730     {
4731         GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, GL_FIXED_ONLY_ARB));
4732         checkGLcall("glClampColorARB");
4733         priv->last_vs_color_unclamp = FALSE;
4734     }
4735 
4736     context->shader_update_mask = (1u << WINED3D_SHADER_TYPE_PIXEL)
4737             | (1u << WINED3D_SHADER_TYPE_VERTEX)
4738             | (1u << WINED3D_SHADER_TYPE_GEOMETRY)
4739             | (1u << WINED3D_SHADER_TYPE_HULL)
4740             | (1u << WINED3D_SHADER_TYPE_DOMAIN)
4741             | (1u << WINED3D_SHADER_TYPE_COMPUTE);
4742 }
4743 
4744 static void shader_arb_destroy(struct wined3d_shader *shader)
4745 {
4746     struct wined3d_device *device = shader->device;
4747     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4748 
4749     if (shader_is_pshader_version(shader->reg_maps.shader_version.type))
4750     {
4751         struct arb_pshader_private *shader_data = shader->backend_data;
4752         UINT i;
4753 
4754         if(!shader_data) return; /* This can happen if a shader was never compiled */
4755 
4756         if (shader_data->num_gl_shaders)
4757         {
4758             struct wined3d_context *context = context_acquire(device, NULL, 0);
4759 
4760             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4761             {
4762                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4763                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4764             }
4765 
4766             context_release(context);
4767         }
4768 
4769         heap_free(shader_data->gl_shaders);
4770         heap_free(shader_data);
4771         shader->backend_data = NULL;
4772     }
4773     else
4774     {
4775         struct arb_vshader_private *shader_data = shader->backend_data;
4776         UINT i;
4777 
4778         if(!shader_data) return; /* This can happen if a shader was never compiled */
4779 
4780         if (shader_data->num_gl_shaders)
4781         {
4782             struct wined3d_context *context = context_acquire(device, NULL, 0);
4783 
4784             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4785             {
4786                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4787                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4788             }
4789 
4790             context_release(context);
4791         }
4792 
4793         heap_free(shader_data->gl_shaders);
4794         heap_free(shader_data);
4795         shader->backend_data = NULL;
4796     }
4797 }
4798 
4799 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4800 {
4801     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4802     return compare_sig(key, &e->sig);
4803 }
4804 
4805 static HRESULT shader_arb_alloc(struct wined3d_device *device, const struct wined3d_vertex_pipe_ops *vertex_pipe,
4806         const struct fragment_pipeline *fragment_pipe)
4807 {
4808     const struct wined3d_d3d_info *d3d_info = &device->adapter->d3d_info;
4809     struct fragment_caps fragment_caps;
4810     void *vertex_priv, *fragment_priv;
4811     struct shader_arb_priv *priv;
4812 
4813     if (!(priv = heap_alloc_zero(sizeof(*priv))))
4814         return E_OUTOFMEMORY;
4815 
4816     if (!(vertex_priv = vertex_pipe->vp_alloc(&arb_program_shader_backend, priv)))
4817     {
4818         ERR("Failed to initialize vertex pipe.\n");
4819         heap_free(priv);
4820         return E_FAIL;
4821     }
4822 
4823     if (!(fragment_priv = fragment_pipe->alloc_private(&arb_program_shader_backend, priv)))
4824     {
4825         ERR("Failed to initialize fragment pipe.\n");
4826         vertex_pipe->vp_free(device);
4827         heap_free(priv);
4828         return E_FAIL;
4829     }
4830 
4831     memset(priv->vshader_const_dirty, 1,
4832            sizeof(*priv->vshader_const_dirty) * d3d_info->limits.vs_uniform_count);
4833     memset(priv->pshader_const_dirty, 1,
4834             sizeof(*priv->pshader_const_dirty) * d3d_info->limits.ps_uniform_count);
4835 
4836     wine_rb_init(&priv->signature_tree, sig_tree_compare);
4837 
4838     priv->vertex_pipe = vertex_pipe;
4839     priv->fragment_pipe = fragment_pipe;
4840     fragment_pipe->get_caps(&device->adapter->gl_info, &fragment_caps);
4841     priv->ffp_proj_control = fragment_caps.wined3d_caps & WINED3D_FRAGMENT_CAP_PROJ_CONTROL;
4842 
4843     device->vertex_priv = vertex_priv;
4844     device->fragment_priv = fragment_priv;
4845     device->shader_priv = priv;
4846 
4847     return WINED3D_OK;
4848 }
4849 
4850 static void release_signature(struct wine_rb_entry *entry, void *context)
4851 {
4852     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4853     unsigned int i;
4854 
4855     for (i = 0; i < sig->sig.element_count; ++i)
4856     {
4857         heap_free((char *)sig->sig.elements[i].semantic_name);
4858     }
4859     heap_free(sig->sig.elements);
4860     heap_free(sig);
4861 }
4862 
4863 /* Context activation is done by the caller. */
4864 static void shader_arb_free(struct wined3d_device *device)
4865 {
4866     struct shader_arb_priv *priv = device->shader_priv;
4867 
4868     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4869     priv->fragment_pipe->free_private(device);
4870     priv->vertex_pipe->vp_free(device);
4871     heap_free(device->shader_priv);
4872 }
4873 
4874 static BOOL shader_arb_allocate_context_data(struct wined3d_context *context)
4875 {
4876     return TRUE;
4877 }
4878 
4879 static void shader_arb_free_context_data(struct wined3d_context *context)
4880 {
4881     struct shader_arb_priv *priv;
4882 
4883     priv = context->device->shader_priv;
4884     if (priv->last_context == context)
4885         priv->last_context = NULL;
4886 }
4887 
4888 static void shader_arb_init_context_state(struct wined3d_context *context) {}
4889 
4890 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4891 {
4892     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4893     {
4894         DWORD vs_consts;
4895         UINT vs_version;
4896 
4897         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4898          * for vertex programs. If the native limit is less than that it's
4899          * not very useful, and e.g. Mesa swrast returns 0, probably to
4900          * indicate it's a software implementation. */
4901         if (gl_info->limits.arb_vs_native_constants < 96)
4902             vs_consts = gl_info->limits.arb_vs_float_constants;
4903         else
4904             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
4905 
4906         if (gl_info->supported[NV_VERTEX_PROGRAM3])
4907         {
4908             vs_version = 3;
4909             TRACE("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4910         }
4911         else if (vs_consts >= 256)
4912         {
4913             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4914             vs_version = 2;
4915             TRACE("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4916         }
4917         else
4918         {
4919             vs_version = 1;
4920             TRACE("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4921         }
4922         caps->vs_version = min(wined3d_settings.max_sm_vs, vs_version);
4923         caps->vs_uniform_count = min(WINED3D_MAX_VS_CONSTS_F, vs_consts);
4924     }
4925     else
4926     {
4927         caps->vs_version = 0;
4928         caps->vs_uniform_count = 0;
4929     }
4930 
4931     caps->hs_version = 0;
4932     caps->ds_version = 0;
4933     caps->gs_version = 0;
4934     caps->cs_version = 0;
4935 
4936     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4937     {
4938         DWORD ps_consts;
4939         UINT ps_version;
4940 
4941         /* Similar as above for vertex programs, but the minimum for fragment
4942          * programs is 24. */
4943         if (gl_info->limits.arb_ps_native_constants < 24)
4944             ps_consts = gl_info->limits.arb_ps_float_constants;
4945         else
4946             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
4947 
4948         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
4949         {
4950             ps_version = 3;
4951             TRACE("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4952         }
4953         else if (ps_consts >= 32)
4954         {
4955             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4956             ps_version = 2;
4957             TRACE("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4958         }
4959         else
4960         {
4961             ps_version = 1;
4962             TRACE("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4963         }
4964         caps->ps_version = min(wined3d_settings.max_sm_ps, ps_version);
4965         caps->ps_uniform_count = min(WINED3D_MAX_PS_CONSTS_F, ps_consts);
4966         caps->ps_1x_max_value = 8.0f;
4967     }
4968     else
4969     {
4970         caps->ps_version = 0;
4971         caps->ps_uniform_count = 0;
4972         caps->ps_1x_max_value = 0.0f;
4973     }
4974 
4975     caps->varying_count = 0;
4976     caps->wined3d_caps = WINED3D_SHADER_CAP_SRGB_WRITE;
4977     if (use_nv_clip(gl_info))
4978         caps->wined3d_caps |= WINED3D_SHADER_CAP_VS_CLIPPING;
4979 }
4980 
4981 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
4982 {
4983     /* We support everything except complex conversions. */
4984     return !is_complex_fixup(fixup);
4985 }
4986 
4987 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
4988     DWORD shift;
4989     char write_mask[20], regstr[50];
4990     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
4991     BOOL is_color = FALSE;
4992     const struct wined3d_shader_dst_param *dst;
4993 
4994     if (!ins->dst_count) return;
4995 
4996     dst = &ins->dst[0];
4997     shift = dst->shift;
4998     if (!shift) return; /* Saturate alone is handled by the instructions */
4999 
5000     shader_arb_get_write_mask(ins, dst, write_mask);
5001     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
5002 
5003     /* Generate a line that does the output modifier computation
5004      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
5005      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
5006      */
5007     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
5008                    regstr, write_mask, regstr, shift_tab[shift]);
5009 }
5010 
5011 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
5012 {
5013     /* WINED3DSIH_ABS                              */ shader_hw_map2gl,
5014     /* WINED3DSIH_ADD                              */ shader_hw_map2gl,
5015     /* WINED3DSIH_AND                              */ NULL,
5016     /* WINED3DSIH_ATOMIC_AND                       */ NULL,
5017     /* WINED3DSIH_ATOMIC_CMP_STORE                 */ NULL,
5018     /* WINED3DSIH_ATOMIC_IADD                      */ NULL,
5019     /* WINED3DSIH_ATOMIC_IMAX                      */ NULL,
5020     /* WINED3DSIH_ATOMIC_IMIN                      */ NULL,
5021     /* WINED3DSIH_ATOMIC_OR                        */ NULL,
5022     /* WINED3DSIH_ATOMIC_UMAX                      */ NULL,
5023     /* WINED3DSIH_ATOMIC_UMIN                      */ NULL,
5024     /* WINED3DSIH_ATOMIC_XOR                       */ NULL,
5025     /* WINED3DSIH_BEM                              */ pshader_hw_bem,
5026     /* WINED3DSIH_BFI                              */ NULL,
5027     /* WINED3DSIH_BFREV                            */ NULL,
5028     /* WINED3DSIH_BREAK                            */ shader_hw_break,
5029     /* WINED3DSIH_BREAKC                           */ shader_hw_breakc,
5030     /* WINED3DSIH_BREAKP                           */ NULL,
5031     /* WINED3DSIH_BUFINFO                          */ NULL,
5032     /* WINED3DSIH_CALL                             */ shader_hw_call,
5033     /* WINED3DSIH_CALLNZ                           */ NULL,
5034     /* WINED3DSIH_CASE                             */ NULL,
5035     /* WINED3DSIH_CMP                              */ pshader_hw_cmp,
5036     /* WINED3DSIH_CND                              */ pshader_hw_cnd,
5037     /* WINED3DSIH_CONTINUE                         */ NULL,
5038     /* WINED3DSIH_CONTINUEP                        */ NULL,
5039     /* WINED3DSIH_COUNTBITS                        */ NULL,
5040     /* WINED3DSIH_CRS                              */ shader_hw_map2gl,
5041     /* WINED3DSIH_CUT                              */ NULL,
5042     /* WINED3DSIH_CUT_STREAM                       */ NULL,
5043     /* WINED3DSIH_DCL                              */ shader_hw_nop,
5044     /* WINED3DSIH_DCL_CONSTANT_BUFFER              */ shader_hw_nop,
5045     /* WINED3DSIH_DCL_FUNCTION_BODY                */ NULL,
5046     /* WINED3DSIH_DCL_FUNCTION_TABLE               */ NULL,
5047     /* WINED3DSIH_DCL_GLOBAL_FLAGS                 */ NULL,
5048     /* WINED3DSIH_DCL_GS_INSTANCES                 */ NULL,
5049     /* WINED3DSIH_DCL_HS_FORK_PHASE_INSTANCE_COUNT */ NULL,
5050     /* WINED3DSIH_DCL_HS_JOIN_PHASE_INSTANCE_COUNT */ NULL,
5051     /* WINED3DSIH_DCL_HS_MAX_TESSFACTOR            */ NULL,
5052     /* WINED3DSIH_DCL_IMMEDIATE_CONSTANT_BUFFER    */ NULL,
5053     /* WINED3DSIH_DCL_INDEX_RANGE                  */ NULL,
5054     /* WINED3DSIH_DCL_INDEXABLE_TEMP               */ NULL,
5055     /* WINED3DSIH_DCL_INPUT                        */ NULL,
5056     /* WINED3DSIH_DCL_INPUT_CONTROL_POINT_COUNT    */ NULL,
5057     /* WINED3DSIH_DCL_INPUT_PRIMITIVE              */ shader_hw_nop,
5058     /* WINED3DSIH_DCL_INPUT_PS                     */ NULL,
5059     /* WINED3DSIH_DCL_INPUT_PS_SGV                 */ NULL,
5060     /* WINED3DSIH_DCL_INPUT_PS_SIV                 */ NULL,
5061     /* WINED3DSIH_DCL_INPUT_SGV                    */ NULL,
5062     /* WINED3DSIH_DCL_INPUT_SIV                    */ NULL,
5063     /* WINED3DSIH_DCL_INTERFACE                    */ NULL,
5064     /* WINED3DSIH_DCL_OUTPUT                       */ NULL,
5065     /* WINED3DSIH_DCL_OUTPUT_CONTROL_POINT_COUNT   */ NULL,
5066     /* WINED3DSIH_DCL_OUTPUT_SIV                   */ NULL,
5067     /* WINED3DSIH_DCL_OUTPUT_TOPOLOGY              */ shader_hw_nop,
5068     /* WINED3DSIH_DCL_RESOURCE_RAW                 */ NULL,
5069     /* WINED3DSIH_DCL_RESOURCE_STRUCTURED          */ NULL,
5070     /* WINED3DSIH_DCL_SAMPLER                      */ NULL,
5071     /* WINED3DSIH_DCL_STREAM                       */ NULL,
5072     /* WINED3DSIH_DCL_TEMPS                        */ NULL,
5073     /* WINED3DSIH_DCL_TESSELLATOR_DOMAIN           */ NULL,
5074     /* WINED3DSIH_DCL_TESSELLATOR_OUTPUT_PRIMITIVE */ NULL,
5075     /* WINED3DSIH_DCL_TESSELLATOR_PARTITIONING     */ NULL,
5076     /* WINED3DSIH_DCL_TGSM_RAW                     */ NULL,
5077     /* WINED3DSIH_DCL_TGSM_STRUCTURED              */ NULL,
5078     /* WINED3DSIH_DCL_THREAD_GROUP                 */ NULL,
5079     /* WINED3DSIH_DCL_UAV_RAW                      */ NULL,
5080     /* WINED3DSIH_DCL_UAV_STRUCTURED               */ NULL,
5081     /* WINED3DSIH_DCL_UAV_TYPED                    */ NULL,
5082     /* WINED3DSIH_DCL_VERTICES_OUT                 */ shader_hw_nop,
5083     /* WINED3DSIH_DEF                              */ shader_hw_nop,
5084     /* WINED3DSIH_DEFAULT                          */ NULL,
5085     /* WINED3DSIH_DEFB                             */ shader_hw_nop,
5086     /* WINED3DSIH_DEFI                             */ shader_hw_nop,
5087     /* WINED3DSIH_DIV                              */ NULL,
5088     /* WINED3DSIH_DP2                              */ NULL,
5089     /* WINED3DSIH_DP2ADD                           */ pshader_hw_dp2add,
5090     /* WINED3DSIH_DP3                              */ shader_hw_map2gl,
5091     /* WINED3DSIH_DP4                              */ shader_hw_map2gl,
5092     /* WINED3DSIH_DST                              */ shader_hw_map2gl,
5093     /* WINED3DSIH_DSX                              */ shader_hw_map2gl,
5094     /* WINED3DSIH_DSX_COARSE                       */ NULL,
5095     /* WINED3DSIH_DSX_FINE                         */ NULL,
5096     /* WINED3DSIH_DSY                              */ shader_hw_dsy,
5097     /* WINED3DSIH_DSY_COARSE                       */ NULL,
5098     /* WINED3DSIH_DSY_FINE                         */ NULL,
5099     /* WINED3DSIH_EVAL_SAMPLE_INDEX                */ NULL,
5100     /* WINED3DSIH_ELSE                             */ shader_hw_else,
5101     /* WINED3DSIH_EMIT                             */ NULL,
5102     /* WINED3DSIH_EMIT_STREAM                      */ NULL,
5103     /* WINED3DSIH_ENDIF                            */ shader_hw_endif,
5104     /* WINED3DSIH_ENDLOOP                          */ shader_hw_endloop,
5105     /* WINED3DSIH_ENDREP                           */ shader_hw_endrep,
5106     /* WINED3DSIH_ENDSWITCH                        */ NULL,
5107     /* WINED3DSIH_EQ                               */ NULL,
5108     /* WINED3DSIH_EXP                              */ shader_hw_scalar_op,
5109     /* WINED3DSIH_EXPP                             */ shader_hw_scalar_op,
5110     /* WINED3DSIH_F16TOF32                         */ NULL,
5111     /* WINED3DSIH_F32TOF16                         */ NULL,
5112     /* WINED3DSIH_FCALL                            */ NULL,
5113     /* WINED3DSIH_FIRSTBIT_HI                      */ NULL,
5114     /* WINED3DSIH_FIRSTBIT_LO                      */ NULL,
5115     /* WINED3DSIH_FIRSTBIT_SHI                     */ NULL,
5116     /* WINED3DSIH_FRC                              */ shader_hw_map2gl,
5117     /* WINED3DSIH_FTOI                             */ NULL,
5118     /* WINED3DSIH_FTOU                             */ NULL,
5119     /* WINED3DSIH_GATHER4                          */ NULL,
5120     /* WINED3DSIH_GATHER4_C                        */ NULL,
5121     /* WINED3DSIH_GATHER4_PO                       */ NULL,
5122     /* WINED3DSIH_GATHER4_PO_C                     */ NULL,
5123     /* WINED3DSIH_GE                               */ NULL,
5124     /* WINED3DSIH_HS_CONTROL_POINT_PHASE           */ NULL,
5125     /* WINED3DSIH_HS_DECLS                         */ NULL,
5126     /* WINED3DSIH_HS_FORK_PHASE                    */ NULL,
5127     /* WINED3DSIH_HS_JOIN_PHASE                    */ NULL,
5128     /* WINED3DSIH_IADD                             */ NULL,
5129     /* WINED3DSIH_IBFE                             */ NULL,
5130     /* WINED3DSIH_IEQ                              */ NULL,
5131     /* WINED3DSIH_IF                               */ NULL /* Hardcoded into the shader */,
5132     /* WINED3DSIH_IFC                              */ shader_hw_ifc,
5133     /* WINED3DSIH_IGE                              */ NULL,
5134     /* WINED3DSIH_ILT                              */ NULL,
5135     /* WINED3DSIH_IMAD                             */ NULL,
5136     /* WINED3DSIH_IMAX                             */ NULL,
5137     /* WINED3DSIH_IMIN                             */ NULL,
5138     /* WINED3DSIH_IMM_ATOMIC_ALLOC                 */ NULL,
5139     /* WINED3DSIH_IMM_ATOMIC_AND                   */ NULL,
5140     /* WINED3DSIH_IMM_ATOMIC_CMP_EXCH              */ NULL,
5141     /* WINED3DSIH_IMM_ATOMIC_CONSUME               */ NULL,
5142     /* WINED3DSIH_IMM_ATOMIC_EXCH                  */ NULL,
5143     /* WINED3DSIH_IMM_ATOMIC_IADD                  */ NULL,
5144     /* WINED3DSIH_IMM_ATOMIC_IMAX                  */ NULL,
5145     /* WINED3DSIH_IMM_ATOMIC_IMIN                  */ NULL,
5146     /* WINED3DSIH_IMM_ATOMIC_OR                    */ NULL,
5147     /* WINED3DSIH_IMM_ATOMIC_UMAX                  */ NULL,
5148     /* WINED3DSIH_IMM_ATOMIC_UMIN                  */ NULL,
5149     /* WINED3DSIH_IMM_ATOMIC_XOR                   */ NULL,
5150     /* WINED3DSIH_IMUL                             */ NULL,
5151     /* WINED3DSIH_INE                              */ NULL,
5152     /* WINED3DSIH_INEG                             */ NULL,
5153     /* WINED3DSIH_ISHL                             */ NULL,
5154     /* WINED3DSIH_ISHR                             */ NULL,
5155     /* WINED3DSIH_ITOF                             */ NULL,
5156     /* WINED3DSIH_LABEL                            */ shader_hw_label,
5157     /* WINED3DSIH_LD                               */ NULL,
5158     /* WINED3DSIH_LD2DMS                           */ NULL,
5159     /* WINED3DSIH_LD_RAW                           */ NULL,
5160     /* WINED3DSIH_LD_STRUCTURED                    */ NULL,
5161     /* WINED3DSIH_LD_UAV_TYPED                     */ NULL,
5162     /* WINED3DSIH_LIT                              */ shader_hw_map2gl,
5163     /* WINED3DSIH_LOD                              */ NULL,
5164     /* WINED3DSIH_LOG                              */ shader_hw_scalar_op,
5165     /* WINED3DSIH_LOGP                             */ shader_hw_scalar_op,
5166     /* WINED3DSIH_LOOP                             */ shader_hw_loop,
5167     /* WINED3DSIH_LRP                              */ shader_hw_lrp,
5168     /* WINED3DSIH_LT                               */ NULL,
5169     /* WINED3DSIH_M3x2                             */ shader_hw_mnxn,
5170     /* WINED3DSIH_M3x3                             */ shader_hw_mnxn,
5171     /* WINED3DSIH_M3x4                             */ shader_hw_mnxn,
5172     /* WINED3DSIH_M4x3                             */ shader_hw_mnxn,
5173     /* WINED3DSIH_M4x4                             */ shader_hw_mnxn,
5174     /* WINED3DSIH_MAD                              */ shader_hw_map2gl,
5175     /* WINED3DSIH_MAX                              */ shader_hw_map2gl,
5176     /* WINED3DSIH_MIN                              */ shader_hw_map2gl,
5177     /* WINED3DSIH_MOV                              */ shader_hw_mov,
5178     /* WINED3DSIH_MOVA                             */ shader_hw_mov,
5179     /* WINED3DSIH_MOVC                             */ NULL,
5180     /* WINED3DSIH_MUL                              */ shader_hw_map2gl,
5181     /* WINED3DSIH_NE                               */ NULL,
5182     /* WINED3DSIH_NOP                              */ shader_hw_nop,
5183     /* WINED3DSIH_NOT                              */ NULL,
5184     /* WINED3DSIH_NRM                              */ shader_hw_nrm,
5185     /* WINED3DSIH_OR                               */ NULL,
5186     /* WINED3DSIH_PHASE                            */ shader_hw_nop,
5187     /* WINED3DSIH_POW                              */ shader_hw_pow,
5188     /* WINED3DSIH_RCP                              */ shader_hw_scalar_op,
5189     /* WINED3DSIH_REP                              */ shader_hw_rep,
5190     /* WINED3DSIH_RESINFO                          */ NULL,
5191     /* WINED3DSIH_RET                              */ shader_hw_ret,
5192     /* WINED3DSIH_RETP                             */ NULL,
5193     /* WINED3DSIH_ROUND_NE                         */ NULL,
5194     /* WINED3DSIH_ROUND_NI                         */ NULL,
5195     /* WINED3DSIH_ROUND_PI                         */ NULL,
5196     /* WINED3DSIH_ROUND_Z                          */ NULL,
5197     /* WINED3DSIH_RSQ                              */ shader_hw_scalar_op,
5198     /* WINED3DSIH_SAMPLE                           */ NULL,
5199     /* WINED3DSIH_SAMPLE_B                         */ NULL,
5200     /* WINED3DSIH_SAMPLE_C                         */ NULL,
5201     /* WINED3DSIH_SAMPLE_C_LZ                      */ NULL,
5202     /* WINED3DSIH_SAMPLE_GRAD                      */ NULL,
5203     /* WINED3DSIH_SAMPLE_INFO                      */ NULL,
5204     /* WINED3DSIH_SAMPLE_LOD                       */ NULL,
5205     /* WINED3DSIH_SAMPLE_POS                       */ NULL,
5206     /* WINED3DSIH_SETP                             */ NULL,
5207     /* WINED3DSIH_SGE                              */ shader_hw_map2gl,
5208     /* WINED3DSIH_SGN                              */ shader_hw_sgn,
5209     /* WINED3DSIH_SINCOS                           */ shader_hw_sincos,
5210     /* WINED3DSIH_SLT                              */ shader_hw_map2gl,
5211     /* WINED3DSIH_SQRT                             */ NULL,
5212     /* WINED3DSIH_STORE_RAW                        */ NULL,
5213     /* WINED3DSIH_STORE_STRUCTURED                 */ NULL,
5214     /* WINED3DSIH_STORE_UAV_TYPED                  */ NULL,
5215     /* WINED3DSIH_SUB                              */ shader_hw_map2gl,
5216     /* WINED3DSIH_SWAPC                            */ NULL,
5217     /* WINED3DSIH_SWITCH                           */ NULL,
5218     /* WINED3DSIH_SYNC                             */ NULL,
5219     /* WINED3DSIH_TEX                              */ pshader_hw_tex,
5220     /* WINED3DSIH_TEXBEM                           */ pshader_hw_texbem,
5221     /* WINED3DSIH_TEXBEML                          */ pshader_hw_texbem,
5222     /* WINED3DSIH_TEXCOORD                         */ pshader_hw_texcoord,
5223     /* WINED3DSIH_TEXDEPTH                         */ pshader_hw_texdepth,
5224     /* WINED3DSIH_TEXDP3                           */ pshader_hw_texdp3,
5225     /* WINED3DSIH_TEXDP3TEX                        */ pshader_hw_texdp3tex,
5226     /* WINED3DSIH_TEXKILL                          */ pshader_hw_texkill,
5227     /* WINED3DSIH_TEXLDD                           */ shader_hw_texldd,
5228     /* WINED3DSIH_TEXLDL                           */ shader_hw_texldl,
5229     /* WINED3DSIH_TEXM3x2DEPTH                     */ pshader_hw_texm3x2depth,
5230     /* WINED3DSIH_TEXM3x2PAD                       */ pshader_hw_texm3x2pad,
5231     /* WINED3DSIH_TEXM3x2TEX                       */ pshader_hw_texm3x2tex,
5232     /* WINED3DSIH_TEXM3x3                          */ pshader_hw_texm3x3,
5233     /* WINED3DSIH_TEXM3x3DIFF                      */ NULL,
5234     /* WINED3DSIH_TEXM3x3PAD                       */ pshader_hw_texm3x3pad,
5235     /* WINED3DSIH_TEXM3x3SPEC                      */ pshader_hw_texm3x3spec,
5236     /* WINED3DSIH_TEXM3x3TEX                       */ pshader_hw_texm3x3tex,
5237     /* WINED3DSIH_TEXM3x3VSPEC                     */ pshader_hw_texm3x3vspec,
5238     /* WINED3DSIH_TEXREG2AR                        */ pshader_hw_texreg2ar,
5239     /* WINED3DSIH_TEXREG2GB                        */ pshader_hw_texreg2gb,
5240     /* WINED3DSIH_TEXREG2RGB                       */ pshader_hw_texreg2rgb,
5241     /* WINED3DSIH_UBFE                             */ NULL,
5242     /* WINED3DSIH_UDIV                             */ NULL,
5243     /* WINED3DSIH_UGE                              */ NULL,
5244     /* WINED3DSIH_ULT                              */ NULL,
5245     /* WINED3DSIH_UMAX                             */ NULL,
5246     /* WINED3DSIH_UMIN                             */ NULL,
5247     /* WINED3DSIH_UMUL                             */ NULL,
5248     /* WINED3DSIH_USHR                             */ NULL,
5249     /* WINED3DSIH_UTOF                             */ NULL,
5250     /* WINED3DSIH_XOR                              */ NULL,
5251 };
5252 
5253 static BOOL get_bool_const(const struct wined3d_shader_instruction *ins,
5254         const struct wined3d_shader *shader, DWORD idx)
5255 {
5256     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5257     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5258     const struct wined3d_shader_lconst *constant;
5259     WORD bools = 0;
5260     WORD flag = (1u << idx);
5261     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5262 
5263     if (reg_maps->local_bool_consts & flag)
5264     {
5265         /* What good is an if(bool) with a hardcoded local constant? I don't know, but handle it */
5266         LIST_FOR_EACH_ENTRY(constant, &shader->constantsB, struct wined3d_shader_lconst, entry)
5267         {
5268             if (constant->idx == idx)
5269             {
5270                 return constant->value[0];
5271             }
5272         }
5273         ERR("Local constant not found\n");
5274         return FALSE;
5275     }
5276     else
5277     {
5278         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5279         else bools = priv->cur_ps_args->bools;
5280         return bools & flag;
5281     }
5282 }
5283 
5284 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5285         const struct wined3d_shader *shader, UINT idx, struct wined3d_shader_loop_control *loop_control)
5286 {
5287     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5288     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5289 
5290     /* Integer constants can either be a local constant, or they can be stored in the shader
5291      * type specific compile args. */
5292     if (reg_maps->local_int_consts & (1u << idx))
5293     {
5294         const struct wined3d_shader_lconst *constant;
5295 
5296         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
5297         {
5298             if (constant->idx == idx)
5299             {
5300                 loop_control->count = constant->value[0];
5301                 loop_control->start = constant->value[1];
5302                 /* Step is signed. */
5303                 loop_control->step = (int)constant->value[2];
5304                 return;
5305             }
5306         }
5307         /* If this happens the flag was set incorrectly */
5308         ERR("Local constant not found\n");
5309         loop_control->count = 0;
5310         loop_control->start = 0;
5311         loop_control->step = 0;
5312         return;
5313     }
5314 
5315     switch (reg_maps->shader_version.type)
5316     {
5317         case WINED3D_SHADER_TYPE_VERTEX:
5318             /* Count and aL start value are unsigned */
5319             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5320             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5321             /* Step is signed. */
5322             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5323             break;
5324 
5325         case WINED3D_SHADER_TYPE_PIXEL:
5326             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5327             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5328             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5329             break;
5330 
5331         default:
5332             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5333             break;
5334     }
5335 }
5336 
5337 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5338 {
5339     struct wined3d_shader_src_param *src_param = NULL, *rel_addr;
5340     struct wined3d_shader_dst_param *dst_param;
5341     struct recorded_instruction *rec;
5342     unsigned int i;
5343 
5344     if (!(rec = heap_alloc_zero(sizeof(*rec))))
5345     {
5346         ERR("Out of memory\n");
5347         return;
5348     }
5349 
5350     rec->ins = *ins;
5351     if (!(dst_param = heap_alloc(sizeof(*dst_param))))
5352         goto free;
5353     *dst_param = *ins->dst;
5354     if (ins->dst->reg.idx[0].rel_addr)
5355     {
5356         if (!(rel_addr = heap_alloc(sizeof(*rel_addr))))
5357             goto free;
5358         *rel_addr = *ins->dst->reg.idx[0].rel_addr;
5359         dst_param->reg.idx[0].rel_addr = rel_addr;
5360     }
5361     rec->ins.dst = dst_param;
5362 
5363     if (!(src_param = heap_calloc(ins->src_count, sizeof(*src_param))))
5364         goto free;
5365     for (i = 0; i < ins->src_count; ++i)
5366     {
5367         src_param[i] = ins->src[i];
5368         if (ins->src[i].reg.idx[0].rel_addr)
5369         {
5370             if (!(rel_addr = heap_alloc(sizeof(*rel_addr))))
5371                 goto free;
5372             *rel_addr = *ins->src[i].reg.idx[0].rel_addr;
5373             src_param[i].reg.idx[0].rel_addr = rel_addr;
5374         }
5375     }
5376     rec->ins.src = src_param;
5377     list_add_tail(list, &rec->entry);
5378     return;
5379 
5380 free:
5381     ERR("Out of memory\n");
5382     if (dst_param)
5383     {
5384         heap_free((void *)dst_param->reg.idx[0].rel_addr);
5385         heap_free(dst_param);
5386     }
5387     if (src_param)
5388     {
5389         for (i = 0; i < ins->src_count; ++i)
5390         {
5391             heap_free((void *)src_param[i].reg.idx[0].rel_addr);
5392         }
5393         heap_free(src_param);
5394     }
5395     heap_free(rec);
5396 }
5397 
5398 static void free_recorded_instruction(struct list *list)
5399 {
5400     struct recorded_instruction *rec_ins, *entry2;
5401     unsigned int i;
5402 
5403     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5404     {
5405         list_remove(&rec_ins->entry);
5406         if (rec_ins->ins.dst)
5407         {
5408             heap_free((void *)rec_ins->ins.dst->reg.idx[0].rel_addr);
5409             heap_free((void *)rec_ins->ins.dst);
5410         }
5411         if (rec_ins->ins.src)
5412         {
5413             for (i = 0; i < rec_ins->ins.src_count; ++i)
5414             {
5415                 heap_free((void *)rec_ins->ins.src[i].reg.idx[0].rel_addr);
5416             }
5417             heap_free((void *)rec_ins->ins.src);
5418         }
5419         heap_free(rec_ins);
5420     }
5421 }
5422 
5423 static void pop_control_frame(const struct wined3d_shader_instruction *ins)
5424 {
5425     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5426     struct control_frame *control_frame;
5427 
5428     if (ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5429     {
5430         struct list *e = list_head(&priv->control_frames);
5431         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5432         list_remove(&control_frame->entry);
5433         heap_free(control_frame);
5434         priv->loop_depth--;
5435     }
5436     else if (ins->handler_idx == WINED3DSIH_ENDIF)
5437     {
5438         /* Non-ifc ENDIFs were already handled previously. */
5439         struct list *e = list_head(&priv->control_frames);
5440         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5441         list_remove(&control_frame->entry);
5442         heap_free(control_frame);
5443     }
5444 }
5445 
5446 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5447     SHADER_HANDLER hw_fct;
5448     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5449     const struct wined3d_shader *shader = ins->ctx->shader;
5450     struct control_frame *control_frame;
5451     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
5452     BOOL bool_const;
5453 
5454     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5455     {
5456         control_frame = heap_alloc_zero(sizeof(*control_frame));
5457         list_add_head(&priv->control_frames, &control_frame->entry);
5458 
5459         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5460         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5461 
5462         if(priv->target_version >= NV2)
5463         {
5464             control_frame->no.loop = priv->num_loops++;
5465             priv->loop_depth++;
5466         }
5467         else
5468         {
5469             /* Don't bother recording when we're in a not used if branch */
5470             if(priv->muted)
5471             {
5472                 return;
5473             }
5474 
5475             if(!priv->recording)
5476             {
5477                 list_init(&priv->record);
5478                 priv->recording = TRUE;
5479                 control_frame->outer_loop = TRUE;
5480                 get_loop_control_const(ins, shader, ins->src[0].reg.idx[0].offset, &control_frame->loop_control);
5481                 return; /* Instruction is handled */
5482             }
5483             /* Record this loop in the outer loop's recording */
5484         }
5485     }
5486     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5487     {
5488         if(priv->target_version >= NV2)
5489         {
5490             /* Nothing to do. The control frame is popped after the HW instr handler */
5491         }
5492         else
5493         {
5494             struct list *e = list_head(&priv->control_frames);
5495             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5496             list_remove(&control_frame->entry);
5497 
5498             if(control_frame->outer_loop)
5499             {
5500                 unsigned int iteration;
5501                 int aL = 0;
5502                 struct list copy;
5503 
5504                 /* Turn off recording before playback */
5505                 priv->recording = FALSE;
5506 
5507                 /* Move the recorded instructions to a separate list and get them out of the private data
5508                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5509                  * be recorded again, thus priv->record might be overwritten
5510                  */
5511                 list_init(&copy);
5512                 list_move_tail(&copy, &priv->record);
5513                 list_init(&priv->record);
5514 
5515                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5516                 {
5517                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5518                                    control_frame->loop_control.count, control_frame->loop_control.start,
5519                                    control_frame->loop_control.step);
5520                     aL = control_frame->loop_control.start;
5521                 }
5522                 else
5523                 {
5524                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5525                 }
5526 
5527                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5528                 {
5529                     struct recorded_instruction *rec_ins;
5530                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5531                     {
5532                         priv->aL = aL;
5533                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5534                     }
5535                     else
5536                     {
5537                         shader_addline(buffer, "#Iteration %u\n", iteration);
5538                     }
5539 
5540                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5541                     {
5542                         shader_arb_handle_instruction(&rec_ins->ins);
5543                     }
5544 
5545                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5546                     {
5547                         aL += control_frame->loop_control.step;
5548                     }
5549                 }
5550                 shader_addline(buffer, "#end loop/rep\n");
5551 
5552                 free_recorded_instruction(&copy);
5553                 heap_free(control_frame);
5554                 return; /* Instruction is handled */
5555             }
5556             else
5557             {
5558                 /* This is a nested loop. Proceed to the normal recording function */
5559                 heap_free(control_frame);
5560             }
5561         }
5562     }
5563 
5564     if(priv->recording)
5565     {
5566         record_instruction(&priv->record, ins);
5567         return;
5568     }
5569 
5570     /* boolean if */
5571     if(ins->handler_idx == WINED3DSIH_IF)
5572     {
5573         control_frame = heap_alloc_zero(sizeof(*control_frame));
5574         list_add_head(&priv->control_frames, &control_frame->entry);
5575         control_frame->type = IF;
5576 
5577         bool_const = get_bool_const(ins, shader, ins->src[0].reg.idx[0].offset);
5578         if (ins->src[0].modifiers == WINED3DSPSM_NOT)
5579             bool_const = !bool_const;
5580         if (!priv->muted && !bool_const)
5581         {
5582             shader_addline(buffer, "#if(FALSE){\n");
5583             priv->muted = TRUE;
5584             control_frame->muting = TRUE;
5585         }
5586         else shader_addline(buffer, "#if(TRUE) {\n");
5587 
5588         return; /* Instruction is handled */
5589     }
5590     else if(ins->handler_idx == WINED3DSIH_IFC)
5591     {
5592         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5593         control_frame = heap_alloc_zero(sizeof(*control_frame));
5594         control_frame->type = IFC;
5595         control_frame->no.ifc = priv->num_ifcs++;
5596         list_add_head(&priv->control_frames, &control_frame->entry);
5597     }
5598     else if(ins->handler_idx == WINED3DSIH_ELSE)
5599     {
5600         struct list *e = list_head(&priv->control_frames);
5601         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5602 
5603         if(control_frame->type == IF)
5604         {
5605             shader_addline(buffer, "#} else {\n");
5606             if(!priv->muted && !control_frame->muting)
5607             {
5608                 priv->muted = TRUE;
5609                 control_frame->muting = TRUE;
5610             }
5611             else if(control_frame->muting) priv->muted = FALSE;
5612             return; /* Instruction is handled. */
5613         }
5614         /* In case of an ifc, generate a HW shader instruction */
5615         if (control_frame->type != IFC)
5616             ERR("Control frame does not match.\n");
5617     }
5618     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5619     {
5620         struct list *e = list_head(&priv->control_frames);
5621         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5622 
5623         if(control_frame->type == IF)
5624         {
5625             shader_addline(buffer, "#} endif\n");
5626             if(control_frame->muting) priv->muted = FALSE;
5627             list_remove(&control_frame->entry);
5628             heap_free(control_frame);
5629             return; /* Instruction is handled */
5630         }
5631         /* In case of an ifc, generate a HW shader instruction */
5632         if (control_frame->type != IFC)
5633             ERR("Control frame does not match.\n");
5634     }
5635 
5636     if(priv->muted)
5637     {
5638         pop_control_frame(ins);
5639         return;
5640     }
5641 
5642     /* Select handler */
5643     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5644 
5645     /* Unhandled opcode */
5646     if (!hw_fct)
5647     {
5648         FIXME("Backend can't handle opcode %s.\n", debug_d3dshaderinstructionhandler(ins->handler_idx));
5649         return;
5650     }
5651     hw_fct(ins);
5652 
5653     pop_control_frame(ins);
5654 
5655     shader_arb_add_instruction_modifiers(ins);
5656 }
5657 
5658 static BOOL shader_arb_has_ffp_proj_control(void *shader_priv)
5659 {
5660     struct shader_arb_priv *priv = shader_priv;
5661 
5662     return priv->ffp_proj_control;
5663 }
5664 
5665 static void shader_arb_precompile(void *shader_priv, struct wined3d_shader *shader) {}
5666 
5667 const struct wined3d_shader_backend_ops arb_program_shader_backend =
5668 {
5669     shader_arb_handle_instruction,
5670     shader_arb_precompile,
5671     shader_arb_select,
5672     shader_arb_select_compute,
5673     shader_arb_disable,
5674     shader_arb_update_float_vertex_constants,
5675     shader_arb_update_float_pixel_constants,
5676     shader_arb_load_constants,
5677     shader_arb_destroy,
5678     shader_arb_alloc,
5679     shader_arb_free,
5680     shader_arb_allocate_context_data,
5681     shader_arb_free_context_data,
5682     shader_arb_init_context_state,
5683     shader_arb_get_caps,
5684     shader_arb_color_fixup_supported,
5685     shader_arb_has_ffp_proj_control,
5686 };
5687 
5688 /* ARB_fragment_program fixed function pipeline replacement definitions */
5689 #define ARB_FFP_CONST_TFACTOR           0
5690 #define ARB_FFP_CONST_COLOR_KEY_LOW     ((ARB_FFP_CONST_TFACTOR) + 1)
5691 #define ARB_FFP_CONST_COLOR_KEY_HIGH    ((ARB_FFP_CONST_COLOR_KEY_LOW) + 1)
5692 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_COLOR_KEY_HIGH) + 1)
5693 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5694 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5695 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5696 
5697 struct arbfp_ffp_desc
5698 {
5699     struct ffp_frag_desc parent;
5700     GLuint shader;
5701 };
5702 
5703 /* Context activation is done by the caller. */
5704 static void arbfp_enable(const struct wined3d_gl_info *gl_info, BOOL enable)
5705 {
5706     if (enable)
5707     {
5708         gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
5709         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5710     }
5711     else
5712     {
5713         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
5714         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5715     }
5716 }
5717 
5718 static void *arbfp_alloc(const struct wined3d_shader_backend_ops *shader_backend, void *shader_priv)
5719 {
5720     struct shader_arb_priv *priv;
5721 
5722     /* Share private data between the shader backend and the pipeline
5723      * replacement, if both are the arb implementation. This is needed to
5724      * figure out whether ARBfp should be disabled if no pixel shader is bound
5725      * or not. */
5726     if (shader_backend == &arb_program_shader_backend)
5727         priv = shader_priv;
5728     else if (!(priv = heap_alloc_zero(sizeof(*priv))))
5729         return NULL;
5730 
5731     wine_rb_init(&priv->fragment_shaders, wined3d_ffp_frag_program_key_compare);
5732     priv->use_arbfp_fixed_func = TRUE;
5733 
5734     return priv;
5735 }
5736 
5737 /* Context activation is done by the caller. */
5738 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5739 {
5740     const struct wined3d_gl_info *gl_info = context;
5741     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5742 
5743     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5744     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5745     heap_free(entry_arb);
5746 }
5747 
5748 /* Context activation is done by the caller. */
5749 static void arbfp_free(struct wined3d_device *device)
5750 {
5751     struct shader_arb_priv *priv = device->fragment_priv;
5752 
5753     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5754     priv->use_arbfp_fixed_func = FALSE;
5755 
5756     if (device->shader_backend != &arb_program_shader_backend)
5757         heap_free(device->fragment_priv);
5758 }
5759 
5760 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5761 {
5762     caps->wined3d_caps = WINED3D_FRAGMENT_CAP_PROJ_CONTROL
5763             | WINED3D_FRAGMENT_CAP_SRGB_WRITE
5764             | WINED3D_FRAGMENT_CAP_COLOR_KEY;
5765     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5766     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5767                            WINED3DTEXOPCAPS_SELECTARG1                  |
5768                            WINED3DTEXOPCAPS_SELECTARG2                  |
5769                            WINED3DTEXOPCAPS_MODULATE4X                  |
5770                            WINED3DTEXOPCAPS_MODULATE2X                  |
5771                            WINED3DTEXOPCAPS_MODULATE                    |
5772                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5773                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5774                            WINED3DTEXOPCAPS_ADD                         |
5775                            WINED3DTEXOPCAPS_SUBTRACT                    |
5776                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5777                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5778                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5779                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5780                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5781                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5782                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5783                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5784                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5785                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5786                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5787                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5788                            WINED3DTEXOPCAPS_LERP                        |
5789                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5790                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5791 
5792     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5793 
5794     caps->MaxTextureBlendStages   = MAX_TEXTURES;
5795     caps->MaxSimultaneousTextures = min(gl_info->limits.samplers[WINED3D_SHADER_TYPE_PIXEL], MAX_TEXTURES);
5796 }
5797 
5798 static DWORD arbfp_get_emul_mask(const struct wined3d_gl_info *gl_info)
5799 {
5800     return GL_EXT_EMUL_ARB_MULTITEXTURE | GL_EXT_EMUL_EXT_FOG_COORD;
5801 }
5802 
5803 static void state_texfactor_arbfp(struct wined3d_context *context,
5804         const struct wined3d_state *state, DWORD state_id)
5805 {
5806     const struct wined3d_gl_info *gl_info = context->gl_info;
5807     struct wined3d_device *device = context->device;
5808     struct wined3d_color color;
5809 
5810     if (device->shader_backend == &arb_program_shader_backend)
5811     {
5812         struct shader_arb_priv *priv;
5813 
5814         /* Don't load the parameter if we're using an arbfp pixel shader,
5815          * otherwise we'll overwrite application provided constants. */
5816         if (use_ps(state))
5817             return;
5818 
5819         priv = device->shader_priv;
5820         priv->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5821         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5822     }
5823 
5824     wined3d_color_from_d3dcolor(&color, state->render_states[WINED3D_RS_TEXTUREFACTOR]);
5825     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, &color.r));
5826     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, &color.r)");
5827 }
5828 
5829 static void state_tss_constant_arbfp(struct wined3d_context *context,
5830         const struct wined3d_state *state, DWORD state_id)
5831 {
5832     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5833     const struct wined3d_gl_info *gl_info = context->gl_info;
5834     struct wined3d_device *device = context->device;
5835     struct wined3d_color color;
5836 
5837     if (device->shader_backend == &arb_program_shader_backend)
5838     {
5839         struct shader_arb_priv *priv;
5840 
5841         /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5842          * application provided constants.
5843          */
5844         if (use_ps(state))
5845             return;
5846 
5847         priv = device->shader_priv;
5848         priv->pshader_const_dirty[ARB_FFP_CONST_CONSTANT(stage)] = 1;
5849         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_CONSTANT(stage) + 1);
5850     }
5851 
5852     wined3d_color_from_d3dcolor(&color, state->texture_states[stage][WINED3D_TSS_CONSTANT]);
5853     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_CONSTANT(stage), &color.r));
5854     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_CONSTANT(stage), &color.r)");
5855 }
5856 
5857 static void state_arb_specularenable(struct wined3d_context *context,
5858         const struct wined3d_state *state, DWORD state_id)
5859 {
5860     const struct wined3d_gl_info *gl_info = context->gl_info;
5861     struct wined3d_device *device = context->device;
5862     float col[4];
5863 
5864     if (device->shader_backend == &arb_program_shader_backend)
5865     {
5866         struct shader_arb_priv *priv;
5867 
5868         /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5869          * application provided constants.
5870          */
5871         if (use_ps(state))
5872             return;
5873 
5874         priv = device->shader_priv;
5875         priv->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5876         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5877     }
5878 
5879     if (state->render_states[WINED3D_RS_SPECULARENABLE])
5880     {
5881         /* The specular color has no alpha */
5882         col[0] = 1.0f; col[1] = 1.0f;
5883         col[2] = 1.0f; col[3] = 0.0f;
5884     } else {
5885         col[0] = 0.0f; col[1] = 0.0f;
5886         col[2] = 0.0f; col[3] = 0.0f;
5887     }
5888     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5889     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5890 }
5891 
5892 static void set_bumpmat_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5893 {
5894     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5895     const struct wined3d_gl_info *gl_info = context->gl_info;
5896     struct wined3d_device *device = context->device;
5897     float mat[2][2];
5898 
5899     context->constant_update_mask |= WINED3D_SHADER_CONST_PS_BUMP_ENV;
5900 
5901     if (device->shader_backend == &arb_program_shader_backend)
5902     {
5903         struct shader_arb_priv *priv = device->shader_priv;
5904 
5905         /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants. */
5906         if (use_ps(state))
5907             return;
5908 
5909         priv->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5910         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5911     }
5912 
5913     mat[0][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT00]);
5914     mat[0][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT01]);
5915     mat[1][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT10]);
5916     mat[1][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT11]);
5917 
5918     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5919     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5920 }
5921 
5922 static void tex_bumpenvlum_arbfp(struct wined3d_context *context,
5923         const struct wined3d_state *state, DWORD state_id)
5924 {
5925     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5926     const struct wined3d_gl_info *gl_info = context->gl_info;
5927     struct wined3d_device *device = context->device;
5928     float param[4];
5929 
5930     context->constant_update_mask |= WINED3D_SHADER_CONST_PS_BUMP_ENV;
5931 
5932     if (device->shader_backend == &arb_program_shader_backend)
5933     {
5934         struct shader_arb_priv *priv = device->shader_priv;
5935 
5936         /* Exit now, don't set the luminance below, otherwise we may overwrite pixel shader constants. */
5937         if (use_ps(state))
5938             return;
5939 
5940         priv->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5941         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5942     }
5943 
5944     param[0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LSCALE]);
5945     param[1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LOFFSET]);
5946     param[2] = 0.0f;
5947     param[3] = 0.0f;
5948 
5949     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5950     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5951 }
5952 
5953 static void alpha_test_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5954 {
5955     const struct wined3d_gl_info *gl_info = context->gl_info;
5956     int glParm;
5957     float ref;
5958 
5959     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
5960 
5961     if (state->render_states[WINED3D_RS_ALPHATESTENABLE])
5962     {
5963         gl_info->gl_ops.gl.p_glEnable(GL_ALPHA_TEST);
5964         checkGLcall("glEnable GL_ALPHA_TEST");
5965     }
5966     else
5967     {
5968         gl_info->gl_ops.gl.p_glDisable(GL_ALPHA_TEST);
5969         checkGLcall("glDisable GL_ALPHA_TEST");
5970         return;
5971     }
5972 
5973     ref = ((float)state->render_states[WINED3D_RS_ALPHAREF]) / 255.0f;
5974     glParm = wined3d_gl_compare_func(state->render_states[WINED3D_RS_ALPHAFUNC]);
5975 
5976     if (glParm)
5977     {
5978         gl_info->gl_ops.gl.p_glAlphaFunc(glParm, ref);
5979         checkGLcall("glAlphaFunc");
5980     }
5981 }
5982 
5983 static void color_key_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5984 {
5985     const struct wined3d_texture *texture = state->textures[0];
5986     const struct wined3d_gl_info *gl_info = context->gl_info;
5987     struct wined3d_device *device = context->device;
5988     struct wined3d_color float_key[2];
5989 
5990     if (!texture)
5991         return;
5992 
5993     if (device->shader_backend == &arb_program_shader_backend)
5994     {
5995         struct shader_arb_priv *priv;
5996 
5997         /* Don't load the parameter if we're using an arbfp pixel shader,
5998          * otherwise we'll overwrite application provided constants. */
5999         if (use_ps(state))
6000             return;
6001 
6002         priv = device->shader_priv;
6003         priv->pshader_const_dirty[ARB_FFP_CONST_COLOR_KEY_LOW] = 1;
6004         priv->pshader_const_dirty[ARB_FFP_CONST_COLOR_KEY_HIGH] = 1;
6005         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_COLOR_KEY_HIGH + 1);
6006     }
6007 
6008     wined3d_format_get_float_color_key(texture->resource.format, &texture->async.src_blt_color_key, float_key);
6009 
6010     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_COLOR_KEY_LOW, &float_key[0].r));
6011     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_COLOR_KEY_LOW, &float_key[0].r)");
6012     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_COLOR_KEY_HIGH, &float_key[1].r));
6013     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_COLOR_KEY_HIGH, &float_key[1].r)");
6014 }
6015 
6016 static const char *get_argreg(struct wined3d_string_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
6017 {
6018     const char *ret;
6019 
6020     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
6021 
6022     switch(arg & WINED3DTA_SELECTMASK) {
6023         case WINED3DTA_DIFFUSE:
6024             ret = "fragment.color.primary"; break;
6025 
6026         case WINED3DTA_CURRENT:
6027             ret = "ret";
6028             break;
6029 
6030         case WINED3DTA_TEXTURE:
6031             switch(stage) {
6032                 case 0: ret = "tex0"; break;
6033                 case 1: ret = "tex1"; break;
6034                 case 2: ret = "tex2"; break;
6035                 case 3: ret = "tex3"; break;
6036                 case 4: ret = "tex4"; break;
6037                 case 5: ret = "tex5"; break;
6038                 case 6: ret = "tex6"; break;
6039                 case 7: ret = "tex7"; break;
6040                 default: ret = "unknown texture";
6041             }
6042             break;
6043 
6044         case WINED3DTA_TFACTOR:
6045             ret = "tfactor"; break;
6046 
6047         case WINED3DTA_SPECULAR:
6048             ret = "fragment.color.secondary"; break;
6049 
6050         case WINED3DTA_TEMP:
6051             ret = "tempreg"; break;
6052 
6053         case WINED3DTA_CONSTANT:
6054             switch(stage) {
6055                 case 0: ret = "const0"; break;
6056                 case 1: ret = "const1"; break;
6057                 case 2: ret = "const2"; break;
6058                 case 3: ret = "const3"; break;
6059                 case 4: ret = "const4"; break;
6060                 case 5: ret = "const5"; break;
6061                 case 6: ret = "const6"; break;
6062                 case 7: ret = "const7"; break;
6063                 default: ret = "unknown constant";
6064             }
6065             break;
6066 
6067         default:
6068             return "unknown";
6069     }
6070 
6071     if(arg & WINED3DTA_COMPLEMENT) {
6072         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
6073         if(argnum == 0) ret = "arg0";
6074         if(argnum == 1) ret = "arg1";
6075         if(argnum == 2) ret = "arg2";
6076     }
6077     if(arg & WINED3DTA_ALPHAREPLICATE) {
6078         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
6079         if(argnum == 0) ret = "arg0";
6080         if(argnum == 1) ret = "arg1";
6081         if(argnum == 2) ret = "arg2";
6082     }
6083     return ret;
6084 }
6085 
6086 static void gen_ffp_instr(struct wined3d_string_buffer *buffer, unsigned int stage, BOOL color,
6087         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
6088 {
6089     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
6090     unsigned int mul = 1;
6091 
6092     if(color && alpha) dstmask = "";
6093     else if(color) dstmask = ".xyz";
6094     else dstmask = ".w";
6095 
6096     if(dst == tempreg) dstreg = "tempreg";
6097     else dstreg = "ret";
6098 
6099     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
6100     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
6101     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
6102 
6103     switch (op)
6104     {
6105         case WINED3D_TOP_DISABLE:
6106             break;
6107 
6108         case WINED3D_TOP_SELECT_ARG2:
6109             arg1 = arg2;
6110             /* FALLTHROUGH */
6111         case WINED3D_TOP_SELECT_ARG1:
6112             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
6113             break;
6114 
6115         case WINED3D_TOP_MODULATE_4X:
6116             mul = 2;
6117             /* FALLTHROUGH */
6118         case WINED3D_TOP_MODULATE_2X:
6119             mul *= 2;
6120             /* FALLTHROUGH */
6121         case WINED3D_TOP_MODULATE:
6122             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6123             break;
6124 
6125         case WINED3D_TOP_ADD_SIGNED_2X:
6126             mul = 2;
6127             /* FALLTHROUGH */
6128         case WINED3D_TOP_ADD_SIGNED:
6129             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6130             arg2 = "arg2";
6131             /* FALLTHROUGH */
6132         case WINED3D_TOP_ADD:
6133             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6134             break;
6135 
6136         case WINED3D_TOP_SUBTRACT:
6137             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6138             break;
6139 
6140         case WINED3D_TOP_ADD_SMOOTH:
6141             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
6142             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6143             break;
6144 
6145         case WINED3D_TOP_BLEND_CURRENT_ALPHA:
6146             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
6147             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6148             break;
6149         case WINED3D_TOP_BLEND_FACTOR_ALPHA:
6150             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
6151             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6152             break;
6153         case WINED3D_TOP_BLEND_TEXTURE_ALPHA:
6154             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
6155             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6156             break;
6157         case WINED3D_TOP_BLEND_DIFFUSE_ALPHA:
6158             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
6159             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6160             break;
6161 
6162         case WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM:
6163             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
6164             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
6165             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
6166             break;
6167 
6168         /* D3DTOP_PREMODULATE ???? */
6169 
6170         case WINED3D_TOP_MODULATE_INVALPHA_ADD_COLOR:
6171             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
6172             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6173             break;
6174         case WINED3D_TOP_MODULATE_ALPHA_ADD_COLOR:
6175             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
6176             break;
6177         case WINED3D_TOP_MODULATE_INVCOLOR_ADD_ALPHA:
6178             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
6179             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
6180             break;
6181         case WINED3D_TOP_MODULATE_COLOR_ADD_ALPHA:
6182             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
6183             break;
6184 
6185         case WINED3D_TOP_DOTPRODUCT3:
6186             mul = 4;
6187             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
6188             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6189             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
6190             break;
6191 
6192         case WINED3D_TOP_MULTIPLY_ADD:
6193             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
6194             break;
6195 
6196         case WINED3D_TOP_LERP:
6197             /* The msdn is not quite right here */
6198             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6199             break;
6200 
6201         case WINED3D_TOP_BUMPENVMAP:
6202         case WINED3D_TOP_BUMPENVMAP_LUMINANCE:
6203             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
6204             break;
6205 
6206         default:
6207             FIXME("Unhandled texture op %08x\n", op);
6208     }
6209 
6210     if (mul == 2)
6211         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", dstreg, dstmask, dstreg);
6212     else if (mul == 4)
6213         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", dstreg, dstmask, dstreg);
6214 }
6215 
6216 static const char *arbfp_texture_target(enum wined3d_gl_resource_type type)
6217 {
6218     switch(type)
6219     {
6220         case WINED3D_GL_RES_TYPE_TEX_1D:
6221             return "1D";
6222         case WINED3D_GL_RES_TYPE_TEX_2D:
6223             return "2D";
6224         case WINED3D_GL_RES_TYPE_TEX_3D:
6225             return "3D";
6226         case WINED3D_GL_RES_TYPE_TEX_CUBE:
6227             return "CUBE";
6228         case WINED3D_GL_RES_TYPE_TEX_RECT:
6229             return "RECT";
6230         default:
6231             return "unexpected_resource_type";
6232     }
6233 }
6234 
6235 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, const struct wined3d_gl_info *gl_info)
6236 {
6237     BYTE tex_read = 0, bump_used = 0, luminance_used = 0, constant_used = 0;
6238     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
6239     unsigned int stage, lowest_disabled_stage;
6240     struct wined3d_string_buffer buffer;
6241     struct color_fixup_masks masks;
6242     BOOL custom_linear_fog = FALSE;
6243     const char *textype, *instr;
6244     DWORD arg0, arg1, arg2;
6245     char colorcor_dst[8];
6246     BOOL op_equal;
6247     GLuint ret;
6248 
6249     if (!string_buffer_init(&buffer))
6250     {
6251         ERR("Failed to initialize shader buffer.\n");
6252         return 0;
6253     }
6254 
6255     shader_addline(&buffer, "!!ARBfp1.0\n");
6256 
6257     if (settings->color_key_enabled)
6258     {
6259         shader_addline(&buffer, "PARAM color_key_low = program.env[%u];\n", ARB_FFP_CONST_COLOR_KEY_LOW);
6260         shader_addline(&buffer, "PARAM color_key_high = program.env[%u];\n", ARB_FFP_CONST_COLOR_KEY_HIGH);
6261         tex_read |= 1;
6262     }
6263 
6264     /* Find out which textures are read */
6265     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6266     {
6267         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6268             break;
6269 
6270         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
6271         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
6272         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
6273 
6274         if (arg0 == WINED3DTA_TEXTURE || arg1 == WINED3DTA_TEXTURE || arg2 == WINED3DTA_TEXTURE)
6275             tex_read |= 1u << stage;
6276         if (settings->op[stage].dst == tempreg)
6277             tempreg_used = TRUE;
6278         if (arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP)
6279             tempreg_used = TRUE;
6280         if (arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR)
6281             tfactor_used = TRUE;
6282         if (arg0 == WINED3DTA_CONSTANT || arg1 == WINED3DTA_CONSTANT || arg2 == WINED3DTA_CONSTANT)
6283             constant_used |= 1u << stage;
6284 
6285         switch (settings->op[stage].cop)
6286         {
6287             case WINED3D_TOP_BUMPENVMAP_LUMINANCE:
6288                 luminance_used |= 1u << stage;
6289                 /* fall through */
6290             case WINED3D_TOP_BUMPENVMAP:
6291                 bump_used |= 1u << stage;
6292                 /* fall through */
6293             case WINED3D_TOP_BLEND_TEXTURE_ALPHA:
6294             case WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM:
6295                 tex_read |= 1u << stage;
6296                 break;
6297 
6298             case WINED3D_TOP_BLEND_FACTOR_ALPHA:
6299                 tfactor_used = TRUE;
6300                 break;
6301 
6302             default:
6303                 break;
6304         }
6305 
6306         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6307             continue;
6308 
6309         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
6310         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
6311         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
6312 
6313         if (arg0 == WINED3DTA_TEXTURE || arg1 == WINED3DTA_TEXTURE || arg2 == WINED3DTA_TEXTURE)
6314             tex_read |= 1u << stage;
6315         if (arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP)
6316             tempreg_used = TRUE;
6317         if (arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR)
6318             tfactor_used = TRUE;
6319         if (arg0 == WINED3DTA_CONSTANT || arg1 == WINED3DTA_CONSTANT || arg2 == WINED3DTA_CONSTANT)
6320             constant_used |= 1u << stage;
6321     }
6322     lowest_disabled_stage = stage;
6323 
6324     switch (settings->fog)
6325     {
6326         case WINED3D_FFP_PS_FOG_OFF:                                                         break;
6327         case WINED3D_FFP_PS_FOG_LINEAR:
6328             if (gl_info->quirks & WINED3D_QUIRK_BROKEN_ARB_FOG)
6329             {
6330                 custom_linear_fog = TRUE;
6331                 break;
6332             }
6333             shader_addline(&buffer, "OPTION ARB_fog_linear;\n");
6334             break;
6335 
6336         case WINED3D_FFP_PS_FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
6337         case WINED3D_FFP_PS_FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
6338         default: FIXME("Unexpected fog setting %d\n", settings->fog);
6339     }
6340 
6341     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
6342     shader_addline(&buffer, "TEMP TMP;\n");
6343     shader_addline(&buffer, "TEMP ret;\n");
6344     if (tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
6345     shader_addline(&buffer, "TEMP arg0;\n");
6346     shader_addline(&buffer, "TEMP arg1;\n");
6347     shader_addline(&buffer, "TEMP arg2;\n");
6348     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6349     {
6350         if (constant_used & (1u << stage))
6351             shader_addline(&buffer, "PARAM const%u = program.env[%u];\n", stage, ARB_FFP_CONST_CONSTANT(stage));
6352 
6353         if (!(tex_read & (1u << stage)))
6354             continue;
6355 
6356         shader_addline(&buffer, "TEMP tex%u;\n", stage);
6357 
6358         if (!(bump_used & (1u << stage)))
6359             continue;
6360         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
6361 
6362         if (!(luminance_used & (1u << stage)))
6363             continue;
6364         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
6365     }
6366     if (tfactor_used)
6367         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
6368     shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6369 
6370     if (settings->sRGB_write)
6371     {
6372         shader_addline(&buffer, "PARAM srgb_consts0 = ");
6373         shader_arb_append_imm_vec4(&buffer, wined3d_srgb_const0);
6374         shader_addline(&buffer, ";\n");
6375         shader_addline(&buffer, "PARAM srgb_consts1 = ");
6376         shader_arb_append_imm_vec4(&buffer, wined3d_srgb_const1);
6377         shader_addline(&buffer, ";\n");
6378     }
6379 
6380     if (lowest_disabled_stage < 7 && settings->emul_clipplanes)
6381         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6382 
6383     if (tempreg_used || settings->sRGB_write)
6384         shader_addline(&buffer, "MOV tempreg, 0.0;\n");
6385 
6386     /* Generate texture sampling instructions */
6387     for (stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3D_TOP_DISABLE; ++stage)
6388     {
6389         if (!(tex_read & (1u << stage)))
6390             continue;
6391 
6392         textype = arbfp_texture_target(settings->op[stage].tex_type);
6393 
6394         if(settings->op[stage].projected == proj_none) {
6395             instr = "TEX";
6396         } else if(settings->op[stage].projected == proj_count4 ||
6397                   settings->op[stage].projected == proj_count3) {
6398             instr = "TXP";
6399         } else {
6400             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6401             instr = "TXP";
6402         }
6403 
6404         if (stage > 0
6405                 && (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP
6406                 || settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE))
6407         {
6408             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6409             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6410             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6411             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6412 
6413             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6414              * so multiply the displacement with the dividing parameter before passing it to TXP
6415              */
6416             if (settings->op[stage].projected != proj_none) {
6417                 if(settings->op[stage].projected == proj_count4) {
6418                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6419                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6420                 } else {
6421                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6422                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6423                 }
6424             } else {
6425                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6426             }
6427 
6428             shader_addline(&buffer, "%s tex%u, ret, texture[%u], %s;\n",
6429                     instr, stage, stage, textype);
6430             if (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6431             {
6432                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6433                                stage - 1, stage - 1, stage - 1);
6434                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6435             }
6436         } else if(settings->op[stage].projected == proj_count3) {
6437             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6438             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6439             shader_addline(&buffer, "%s tex%u, ret, texture[%u], %s;\n",
6440                             instr, stage, stage, textype);
6441         } else {
6442             shader_addline(&buffer, "%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6443                             instr, stage, stage, stage, textype);
6444         }
6445 
6446         sprintf(colorcor_dst, "tex%u", stage);
6447         masks = calc_color_correction(settings->op[stage].color_fixup, WINED3DSP_WRITEMASK_ALL);
6448         gen_color_correction(&buffer, colorcor_dst, colorcor_dst, "const.x", "const.y",
6449                 settings->op[stage].color_fixup, masks);
6450     }
6451 
6452     if (settings->color_key_enabled)
6453     {
6454         shader_addline(&buffer, "SLT TMP, tex0, color_key_low;\n"); /* below low key */
6455         shader_addline(&buffer, "SGE ret, tex0, color_key_high;\n"); /* above high key */
6456         shader_addline(&buffer, "ADD TMP, TMP, ret;\n"); /* or */
6457         shader_addline(&buffer, "DP4 TMP.b, TMP, TMP;\n"); /* on any channel */
6458         shader_addline(&buffer, "SGE TMP, -TMP.b, 0.0;\n"); /* logical not */
6459         shader_addline(&buffer, "KIL -TMP;\n"); /* discard if true */
6460     }
6461 
6462     shader_addline(&buffer, "MOV ret, fragment.color.primary;\n");
6463 
6464     /* Generate the main shader */
6465     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6466     {
6467         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6468             break;
6469 
6470         if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6471                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6472             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6473         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6474                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6475             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6476         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6477                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6478             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6479         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6480                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6481             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6482         else
6483             op_equal = settings->op[stage].aop   == settings->op[stage].cop
6484                     && settings->op[stage].carg0 == settings->op[stage].aarg0
6485                     && settings->op[stage].carg1 == settings->op[stage].aarg1
6486                     && settings->op[stage].carg2 == settings->op[stage].aarg2;
6487 
6488         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6489         {
6490             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6491                           settings->op[stage].cop, settings->op[stage].carg0,
6492                           settings->op[stage].carg1, settings->op[stage].carg2);
6493         }
6494         else if (op_equal)
6495         {
6496             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6497                           settings->op[stage].cop, settings->op[stage].carg0,
6498                           settings->op[stage].carg1, settings->op[stage].carg2);
6499         }
6500         else if (settings->op[stage].cop != WINED3D_TOP_BUMPENVMAP
6501                 && settings->op[stage].cop != WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6502         {
6503             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6504                           settings->op[stage].cop, settings->op[stage].carg0,
6505                           settings->op[stage].carg1, settings->op[stage].carg2);
6506             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6507                           settings->op[stage].aop, settings->op[stage].aarg0,
6508                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6509         }
6510     }
6511 
6512     if (settings->sRGB_write || custom_linear_fog)
6513     {
6514         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, ret;\n");
6515         if (settings->sRGB_write)
6516             arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6517         if (custom_linear_fog)
6518             arbfp_add_linear_fog(&buffer, "ret", "arg0");
6519         shader_addline(&buffer, "MOV result.color, ret;\n");
6520     }
6521     else
6522     {
6523         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, ret;\n");
6524     }
6525 
6526     /* Footer */
6527     shader_addline(&buffer, "END\n");
6528 
6529     /* Generate the shader */
6530     GL_EXTCALL(glGenProgramsARB(1, &ret));
6531     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6532     shader_arb_compile(gl_info, GL_FRAGMENT_PROGRAM_ARB, buffer.buffer);
6533 
6534     string_buffer_free(&buffer);
6535     return ret;
6536 }
6537 
6538 static void fragment_prog_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6539 {
6540     const struct wined3d_gl_info *gl_info = context->gl_info;
6541     const struct wined3d_device *device = context->device;
6542     struct shader_arb_priv *priv = device->fragment_priv;
6543     BOOL use_pshader = use_ps(state);
6544     struct ffp_frag_settings settings;
6545     const struct arbfp_ffp_desc *desc;
6546     unsigned int i;
6547 
6548     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6549 
6550     if (isStateDirty(context, STATE_RENDER(WINED3D_RS_FOGENABLE)))
6551     {
6552         if (!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6553         {
6554             /* Reload fixed function constants since they collide with the
6555              * pixel shader constants. */
6556             for (i = 0; i < MAX_TEXTURES; ++i)
6557             {
6558                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6559                 state_tss_constant_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_CONSTANT));
6560             }
6561             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6562             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6563             color_key_arbfp(context, state, STATE_COLOR_KEY);
6564         }
6565         else if (use_pshader)
6566         {
6567             context->shader_update_mask |= 1u << WINED3D_SHADER_TYPE_PIXEL;
6568         }
6569         return;
6570     }
6571 
6572     if (!use_pshader)
6573     {
6574         /* Find or create a shader implementing the fixed function pipeline
6575          * settings, then activate it. */
6576         gen_ffp_frag_op(context, state, &settings, FALSE);
6577         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6578         if (!desc)
6579         {
6580             struct arbfp_ffp_desc *new_desc;
6581 
6582             if (!(new_desc = heap_alloc(sizeof(*new_desc))))
6583             {
6584                 ERR("Out of memory\n");
6585                 return;
6586             }
6587 
6588             new_desc->parent.settings = settings;
6589             new_desc->shader = gen_arbfp_ffp_shader(&settings, gl_info);
6590             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6591             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6592             desc = new_desc;
6593         }
6594 
6595         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active (however, note the
6596          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6597          * deactivate it.
6598          */
6599         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6600         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6601         priv->current_fprogram_id = desc->shader;
6602 
6603         if (device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6604         {
6605             /* Reload fixed function constants since they collide with the
6606              * pixel shader constants. */
6607             for (i = 0; i < MAX_TEXTURES; ++i)
6608             {
6609                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6610                 state_tss_constant_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_CONSTANT));
6611             }
6612             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6613             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6614             color_key_arbfp(context, state, STATE_COLOR_KEY);
6615         }
6616         context->last_was_pshader = FALSE;
6617     }
6618     else if (!context->last_was_pshader)
6619     {
6620         if (device->shader_backend == &arb_program_shader_backend)
6621             context->constant_update_mask |= WINED3D_SHADER_CONST_PS_F;
6622         context->last_was_pshader = TRUE;
6623     }
6624 
6625     context->shader_update_mask |= 1u << WINED3D_SHADER_TYPE_PIXEL;
6626 }
6627 
6628 /* We can't link the fog states to the fragment state directly since the
6629  * vertex pipeline links them to FOGENABLE. A different linking in different
6630  * pipeline parts can't be expressed in the combined state table, so we need
6631  * to handle that with a forwarding function. The other invisible side effect
6632  * is that changing the fog start and fog end (which links to FOGENABLE in
6633  * vertex) results in the fragment_prog_arbfp function being called because
6634  * FOGENABLE is dirty, which calls this function here. */
6635 static void state_arbfp_fog(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6636 {
6637     enum fogsource new_source;
6638     DWORD fogstart = state->render_states[WINED3D_RS_FOGSTART];
6639     DWORD fogend = state->render_states[WINED3D_RS_FOGEND];
6640 
6641     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6642 
6643     if (!isStateDirty(context, STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL)))
6644         fragment_prog_arbfp(context, state, state_id);
6645 
6646     if (!state->render_states[WINED3D_RS_FOGENABLE])
6647         return;
6648 
6649     if (state->render_states[WINED3D_RS_FOGTABLEMODE] == WINED3D_FOG_NONE)
6650     {
6651         if (use_vs(state))
6652         {
6653             new_source = FOGSOURCE_VS;
6654         }
6655         else
6656         {
6657             if (state->render_states[WINED3D_RS_FOGVERTEXMODE] == WINED3D_FOG_NONE || context->last_was_rhw)
6658                 new_source = FOGSOURCE_COORD;
6659             else
6660                 new_source = FOGSOURCE_FFP;
6661         }
6662     }
6663     else
6664     {
6665         new_source = FOGSOURCE_FFP;
6666     }
6667 
6668     if (new_source != context->fog_source || fogstart == fogend)
6669     {
6670         context->fog_source = new_source;
6671         state_fogstartend(context, state, STATE_RENDER(WINED3D_RS_FOGSTART));
6672     }
6673 }
6674 
6675 static void textransform(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6676 {
6677     if (!isStateDirty(context, STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL)))
6678         fragment_prog_arbfp(context, state, state_id);
6679 }
6680 
6681 static const struct StateEntryTemplate arbfp_fragmentstate_template[] =
6682 {
6683     {STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),              { STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),             state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6684     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6685     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6686     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6687     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6688     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6689     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6690     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6691     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6692     {STATE_TEXTURESTAGE(0, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6693     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6694     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6695     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6696     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6697     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6698     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6699     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6700     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6701     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6702     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6703     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6704     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6705     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6706     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6707     {STATE_TEXTURESTAGE(1, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6708     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6709     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6710     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6711     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6712     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6713     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6714     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6715     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6716     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6717     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6718     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6719     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6720     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6721     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6722     {STATE_TEXTURESTAGE(2, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6723     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6724     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6725     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6726     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6727     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6728     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6729     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6730     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6731     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6732     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6733     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6734     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6735     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6736     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6737     {STATE_TEXTURESTAGE(3, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6738     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6739     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6740     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6741     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6742     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6743     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6744     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6745     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6746     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6747     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6748     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6749     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6750     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6751     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6752     {STATE_TEXTURESTAGE(4, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6753     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6754     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6755     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6756     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6757     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6758     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6759     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6760     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6761     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6762     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6763     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6764     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6765     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6766     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6767     {STATE_TEXTURESTAGE(5, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6768     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6769     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6770     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6771     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6772     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6773     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6774     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6775     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6776     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6777     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6778     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6779     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6780     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6781     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6782     {STATE_TEXTURESTAGE(6, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6783     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6784     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6785     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6786     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6787     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6788     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6789     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6790     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6791     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6792     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6793     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6794     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6795     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6796     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6797     {STATE_TEXTURESTAGE(7, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6798     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6799     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6800     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6801     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6802     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6803     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6804     {STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),             { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6805     {STATE_RENDER(WINED3D_RS_ALPHAFUNC),                  { STATE_RENDER(WINED3D_RS_ALPHATESTENABLE),           NULL                    }, WINED3D_GL_EXT_NONE             },
6806     {STATE_RENDER(WINED3D_RS_ALPHAREF),                   { STATE_RENDER(WINED3D_RS_ALPHATESTENABLE),           NULL                    }, WINED3D_GL_EXT_NONE             },
6807     {STATE_RENDER(WINED3D_RS_ALPHATESTENABLE),            { STATE_RENDER(WINED3D_RS_ALPHATESTENABLE),           alpha_test_arbfp        }, WINED3D_GL_EXT_NONE             },
6808     {STATE_RENDER(WINED3D_RS_COLORKEYENABLE),             { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6809     {STATE_COLOR_KEY,                                     { STATE_COLOR_KEY,                                    color_key_arbfp         }, WINED3D_GL_EXT_NONE             },
6810     {STATE_RENDER(WINED3D_RS_FOGENABLE),                  { STATE_RENDER(WINED3D_RS_FOGENABLE),                 state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6811     {STATE_RENDER(WINED3D_RS_FOGTABLEMODE),               { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6812     {STATE_RENDER(WINED3D_RS_FOGVERTEXMODE),              { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6813     {STATE_RENDER(WINED3D_RS_FOGSTART),                   { STATE_RENDER(WINED3D_RS_FOGSTART),                  state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6814     {STATE_RENDER(WINED3D_RS_FOGEND),                     { STATE_RENDER(WINED3D_RS_FOGSTART),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6815     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),           state_srgbwrite         }, ARB_FRAMEBUFFER_SRGB            },
6816     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6817     {STATE_RENDER(WINED3D_RS_FOGCOLOR),                   { STATE_RENDER(WINED3D_RS_FOGCOLOR),                  state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6818     {STATE_RENDER(WINED3D_RS_FOGDENSITY),                 { STATE_RENDER(WINED3D_RS_FOGDENSITY),                state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6819     {STATE_TEXTURESTAGE(0,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(0,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform}, WINED3D_GL_EXT_NONE             },
6820     {STATE_TEXTURESTAGE(1,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(1,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform}, WINED3D_GL_EXT_NONE             },
6821     {STATE_TEXTURESTAGE(2,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(2,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform}, WINED3D_GL_EXT_NONE             },
6822     {STATE_TEXTURESTAGE(3,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(3,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform}, WINED3D_GL_EXT_NONE             },
6823     {STATE_TEXTURESTAGE(4,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(4,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform}, WINED3D_GL_EXT_NONE             },
6824     {STATE_TEXTURESTAGE(5,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(5,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform}, WINED3D_GL_EXT_NONE             },
6825     {STATE_TEXTURESTAGE(6,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(6,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform}, WINED3D_GL_EXT_NONE             },
6826     {STATE_TEXTURESTAGE(7,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(7,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform}, WINED3D_GL_EXT_NONE             },
6827     {STATE_TEXTURESTAGE(0, WINED3D_TSS_CONSTANT),         { STATE_TEXTURESTAGE(0, WINED3D_TSS_CONSTANT),        state_tss_constant_arbfp}, WINED3D_GL_EXT_NONE             },
6828     {STATE_TEXTURESTAGE(1, WINED3D_TSS_CONSTANT),         { STATE_TEXTURESTAGE(1, WINED3D_TSS_CONSTANT),        state_tss_constant_arbfp}, WINED3D_GL_EXT_NONE             },
6829     {STATE_TEXTURESTAGE(2, WINED3D_TSS_CONSTANT),         { STATE_TEXTURESTAGE(2, WINED3D_TSS_CONSTANT),        state_tss_constant_arbfp}, WINED3D_GL_EXT_NONE             },
6830     {STATE_TEXTURESTAGE(3, WINED3D_TSS_CONSTANT),         { STATE_TEXTURESTAGE(3, WINED3D_TSS_CONSTANT),        state_tss_constant_arbfp}, WINED3D_GL_EXT_NONE             },
6831     {STATE_TEXTURESTAGE(4, WINED3D_TSS_CONSTANT),         { STATE_TEXTURESTAGE(4, WINED3D_TSS_CONSTANT),        state_tss_constant_arbfp}, WINED3D_GL_EXT_NONE             },
6832     {STATE_TEXTURESTAGE(5, WINED3D_TSS_CONSTANT),         { STATE_TEXTURESTAGE(5, WINED3D_TSS_CONSTANT),        state_tss_constant_arbfp}, WINED3D_GL_EXT_NONE             },
6833     {STATE_TEXTURESTAGE(6, WINED3D_TSS_CONSTANT),         { STATE_TEXTURESTAGE(6, WINED3D_TSS_CONSTANT),        state_tss_constant_arbfp}, WINED3D_GL_EXT_NONE             },
6834     {STATE_TEXTURESTAGE(7, WINED3D_TSS_CONSTANT),         { STATE_TEXTURESTAGE(7, WINED3D_TSS_CONSTANT),        state_tss_constant_arbfp}, WINED3D_GL_EXT_NONE             },
6835     {STATE_RENDER(WINED3D_RS_SPECULARENABLE),             { STATE_RENDER(WINED3D_RS_SPECULARENABLE),            state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6836     {STATE_RENDER(WINED3D_RS_SHADEMODE),                  { STATE_RENDER(WINED3D_RS_SHADEMODE),                 state_shademode         }, WINED3D_GL_EXT_NONE             },
6837     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6838 };
6839 
6840 static BOOL arbfp_alloc_context_data(struct wined3d_context *context)
6841 {
6842     return TRUE;
6843 }
6844 
6845 static void arbfp_free_context_data(struct wined3d_context *context)
6846 {
6847 }
6848 
6849 const struct fragment_pipeline arbfp_fragment_pipeline = {
6850     arbfp_enable,
6851     arbfp_get_caps,
6852     arbfp_get_emul_mask,
6853     arbfp_alloc,
6854     arbfp_free,
6855     arbfp_alloc_context_data,
6856     arbfp_free_context_data,
6857     shader_arb_color_fixup_supported,
6858     arbfp_fragmentstate_template,
6859 };
6860 
6861 struct arbfp_blit_type
6862 {
6863     enum complex_fixup fixup : 4;
6864     enum wined3d_gl_resource_type res_type : 3;
6865     DWORD use_color_key : 1;
6866     DWORD padding : 24;
6867 };
6868 
6869 struct arbfp_blit_desc
6870 {
6871     GLuint shader;
6872     struct arbfp_blit_type type;
6873     struct wine_rb_entry entry;
6874 };
6875 
6876 #define ARBFP_BLIT_PARAM_SIZE 0
6877 #define ARBFP_BLIT_PARAM_COLOR_KEY_LOW 1
6878 #define ARBFP_BLIT_PARAM_COLOR_KEY_HIGH 2
6879 
6880 struct wined3d_arbfp_blitter
6881 {
6882     struct wined3d_blitter blitter;
6883     struct wine_rb_tree shaders;
6884     GLuint palette_texture;
6885 };
6886 
6887 static int arbfp_blit_type_compare(const void *key, const struct wine_rb_entry *entry)
6888 {
6889     const struct arbfp_blit_type *ka = key;
6890     const struct arbfp_blit_type *kb = &WINE_RB_ENTRY_VALUE(entry, const struct arbfp_blit_desc, entry)->type;
6891 
6892     return memcmp(ka, kb, sizeof(*ka));
6893 }
6894 
6895 /* Context activation is done by the caller. */
6896 static void arbfp_free_blit_shader(struct wine_rb_entry *entry, void *ctx)
6897 {
6898     struct arbfp_blit_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_blit_desc, entry);
6899     const struct wined3d_gl_info *gl_info;
6900     struct wined3d_context *context;
6901 
6902     context = ctx;
6903     gl_info = context->gl_info;
6904 
6905     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
6906     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
6907     heap_free(entry_arb);
6908 }
6909 
6910 /* Context activation is done by the caller. */
6911 static void arbfp_blitter_destroy(struct wined3d_blitter *blitter, struct wined3d_context *context)
6912 {
6913     const struct wined3d_gl_info *gl_info = context->gl_info;
6914     struct wined3d_arbfp_blitter *arbfp_blitter;
6915     struct wined3d_blitter *next;
6916 
6917     if ((next = blitter->next))
6918         next->ops->blitter_destroy(next, context);
6919 
6920     arbfp_blitter = CONTAINING_RECORD(blitter, struct wined3d_arbfp_blitter, blitter);
6921 
6922     wine_rb_destroy(&arbfp_blitter->shaders, arbfp_free_blit_shader, context);
6923     checkGLcall("Delete blit programs");
6924 
6925     if (arbfp_blitter->palette_texture)
6926         gl_info->gl_ops.gl.p_glDeleteTextures(1, &arbfp_blitter->palette_texture);
6927 
6928     heap_free(arbfp_blitter);
6929 }
6930 
6931 static BOOL gen_planar_yuv_read(struct wined3d_string_buffer *buffer, const struct arbfp_blit_type *type,
6932         char *luminance)
6933 {
6934     char chroma;
6935     const char *tex, *texinstr = "TXP";
6936 
6937     if (type->fixup == COMPLEX_FIXUP_UYVY)
6938     {
6939         chroma = 'x';
6940         *luminance = 'w';
6941     }
6942     else
6943     {
6944         chroma = 'w';
6945         *luminance = 'x';
6946     }
6947 
6948     tex = arbfp_texture_target(type->res_type);
6949     if (type->res_type == WINED3D_GL_RES_TYPE_TEX_RECT)
6950         texinstr = "TEX";
6951 
6952     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6953      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6954      * filtering when we sample the texture.
6955      *
6956      * These are the rules for reading the chroma:
6957      *
6958      * Even pixel: Cr
6959      * Even pixel: U
6960      * Odd pixel: V
6961      *
6962      * So we have to get the sampling x position in non-normalized coordinates in integers
6963      */
6964     if (type->res_type != WINED3D_GL_RES_TYPE_TEX_RECT)
6965     {
6966         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6967         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6968     }
6969     else
6970     {
6971         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6972     }
6973     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6974      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6975      * 0.5, so add 0.5.
6976      */
6977     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6978     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6979 
6980     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6981      * even and odd pixels respectively
6982      */
6983     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6984     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6985 
6986     /* Sample Pixel 1 */
6987     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6988 
6989     /* Put the value into either of the chroma values */
6990     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6991     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6992     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6993     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6994 
6995     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6996      * the pixel right to the current one. Otherwise, sample the left pixel.
6997      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6998      */
6999     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
7000     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
7001     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
7002 
7003     /* Put the value into the other chroma */
7004     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
7005     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
7006     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
7007     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
7008 
7009     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
7010      * the current one and lerp the two U and V values
7011      */
7012 
7013     /* This gives the correctly filtered luminance value */
7014     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
7015 
7016     return TRUE;
7017 }
7018 
7019 static BOOL gen_yv12_read(struct wined3d_string_buffer *buffer, const struct arbfp_blit_type *type,
7020         char *luminance)
7021 {
7022     const char *tex;
7023     static const float yv12_coef[]
7024             = {2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f};
7025 
7026     tex = arbfp_texture_target(type->res_type);
7027 
7028     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
7029      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
7030      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
7031      * pitch of the luminance plane, the packing into the gl texture is a bit
7032      * unfortunate. If the whole texture is interpreted as luminance data it looks
7033      * approximately like this:
7034      *
7035      *        +----------------------------------+----
7036      *        |                                  |
7037      *        |                                  |
7038      *        |                                  |
7039      *        |                                  |
7040      *        |                                  |   2
7041      *        |            LUMINANCE             |   -
7042      *        |                                  |   3
7043      *        |                                  |
7044      *        |                                  |
7045      *        |                                  |
7046      *        |                                  |
7047      *        +----------------+-----------------+----
7048      *        |                |                 |
7049      *        |  V even rows   |  V odd rows     |
7050      *        |                |                 |   1
7051      *        +----------------+------------------   -
7052      *        |                |                 |   3
7053      *        |  U even rows   |  U odd rows     |
7054      *        |                |                 |
7055      *        +----------------+-----------------+----
7056      *        |                |                 |
7057      *        |     0.5        |       0.5       |
7058      *
7059      * So it appears as if there are 4 chroma images, but in fact the odd rows
7060      * in the chroma images are in the same row as the even ones. So it is
7061      * kinda tricky to read
7062      *
7063      * When reading from rectangle textures, keep in mind that the input y coordinates
7064      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
7065      */
7066     shader_addline(buffer, "PARAM yv12_coef = ");
7067     shader_arb_append_imm_vec4(buffer, yv12_coef);
7068     shader_addline(buffer, ";\n");
7069 
7070     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
7071     /* the chroma planes have only half the width */
7072     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
7073 
7074     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
7075      * the coordinate. Also read the right side of the image when reading odd lines
7076      *
7077      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
7078      * bleeding
7079      */
7080     if (type->res_type == WINED3D_GL_RES_TYPE_TEX_2D)
7081     {
7082 
7083         shader_addline(buffer, "RCP chroma.w, size.y;\n");
7084 
7085         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
7086 
7087         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
7088         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
7089 
7090         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
7091         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
7092         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
7093         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
7094         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
7095 
7096         /* clamp, keep the half pixel origin in mind */
7097         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
7098         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
7099         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
7100         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
7101     }
7102     else
7103     {
7104         /* Read from [size - size+size/4] */
7105         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
7106         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
7107 
7108         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
7109         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
7110         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
7111         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
7112         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
7113         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
7114 
7115         /* Make sure to read exactly from the pixel center */
7116         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
7117         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
7118 
7119         /* Clamp */
7120         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
7121         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
7122         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
7123         shader_addline(buffer, "ADD temp.y, size.y, coef.y;\n");
7124         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
7125     }
7126     /* Read the texture, put the result into the output register */
7127     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
7128     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
7129 
7130     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
7131      * No need to clamp because we're just reusing the already clamped value from above
7132      */
7133     if (type->res_type == WINED3D_GL_RES_TYPE_TEX_2D)
7134         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
7135     else
7136         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
7137     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
7138     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
7139 
7140     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
7141      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
7142      * values due to filtering
7143      */
7144     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
7145     if (type->res_type == WINED3D_GL_RES_TYPE_TEX_2D)
7146     {
7147         /* Multiply the y coordinate by 2/3 and clamp it */
7148         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
7149         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
7150         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
7151         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
7152     }
7153     else
7154     {
7155         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
7156          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
7157          * is bigger
7158          */
7159         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
7160         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
7161         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
7162     }
7163     *luminance = 'a';
7164 
7165     return TRUE;
7166 }
7167 
7168 static BOOL gen_nv12_read(struct wined3d_string_buffer *buffer, const struct arbfp_blit_type *type,
7169         char *luminance)
7170 {
7171     const char *tex;
7172     static const float nv12_coef[]
7173             = {2.0f / 3.0f, 1.0f / 3.0f, 1.0f, 1.0f};
7174 
7175     tex = arbfp_texture_target(type->res_type);
7176 
7177     /* NV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
7178      * sized plane where each component is an UV pair. So the effective
7179      * bitdepth is 12 bits per pixel If the whole texture is interpreted as luminance
7180      * data it looks approximately like this:
7181      *
7182      *        +----------------------------------+----
7183      *        |                                  |
7184      *        |                                  |
7185      *        |                                  |
7186      *        |                                  |
7187      *        |                                  |   2
7188      *        |            LUMINANCE             |   -
7189      *        |                                  |   3
7190      *        |                                  |
7191      *        |                                  |
7192      *        |                                  |
7193      *        |                                  |
7194      *        +----------------------------------+----
7195      *        |UVUVUVUVUVUVUVUVUVUVUVUVUVUVUVUVUV|
7196      *        |UVUVUVUVUVUVUVUVUVUVUVUVUVUVUVUVUV|
7197      *        |                                  |   1
7198      *        |                                  |   -
7199      *        |                                  |   3
7200      *        |                                  |
7201      *        |                                  |
7202      *        +----------------------------------+----
7203      *
7204      * When reading from rectangle textures, keep in mind that the input y coordinates
7205      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height. */
7206 
7207     shader_addline(buffer, "PARAM nv12_coef = ");
7208     shader_arb_append_imm_vec4(buffer, nv12_coef);
7209     shader_addline(buffer, ";\n");
7210 
7211     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
7212     /* We only have half the number of chroma pixels. */
7213     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
7214 
7215     if (type->res_type == WINED3D_GL_RES_TYPE_TEX_2D)
7216     {
7217         shader_addline(buffer, "RCP chroma.w, size.x;\n");
7218         shader_addline(buffer, "RCP chroma.z, size.y;\n");
7219 
7220         shader_addline(buffer, "MAD texcrd.y, texcrd.y, nv12_coef.y, nv12_coef.x;\n");
7221 
7222         /* We must not allow filtering horizontally, this would mix U and V.
7223          * Vertical filtering is ok. However, bear in mind that the pixel center is at
7224          * 0.5, so add 0.5. */
7225 
7226         /* Convert to non-normalized coordinates so we can find the
7227          * individual pixel. */
7228         shader_addline(buffer, "MUL texcrd.x, texcrd.x, size.x;\n");
7229         shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
7230         /* Multiply by 2 since chroma components are stored in UV pixel pairs,
7231          * add 0.5 to hit the center of the pixel. */
7232         shader_addline(buffer, "MAD texcrd.x, texcrd.x, coef.z, coef.y;\n");
7233 
7234         /* Convert back to normalized coordinates. */
7235         shader_addline(buffer, "MUL texcrd.x, texcrd.x, chroma.w;\n");
7236 
7237         /* Clamp, keep the half pixel origin in mind. */
7238         shader_addline(buffer, "MAD temp.y, coef.y, chroma.z, nv12_coef.x;\n");
7239         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
7240         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.z, nv12_coef.z;\n");
7241         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
7242     }
7243     else
7244     {
7245         /* Read from [size - size+size/2] */
7246         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.y, size.y;\n");
7247 
7248         shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
7249         /* Multiply by 2 since chroma components are stored in UV pixel pairs,
7250          * add 0.5 to hit the center of the pixel. */
7251         shader_addline(buffer, "MAD texcrd.x, texcrd.x, coef.z, coef.y;\n");
7252 
7253         /* Clamp */
7254         shader_addline(buffer, "MAD temp.y, size.y, coef.y, size.y;\n");
7255         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
7256         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
7257         shader_addline(buffer, "ADD temp.y, size.y, coef.y;\n");
7258         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
7259     }
7260     /* Read the texture, put the result into the output register. */
7261     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
7262     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
7263 
7264     if (type->res_type == WINED3D_GL_RES_TYPE_TEX_2D)
7265     {
7266         /* Add 1/size.x */
7267         shader_addline(buffer, "ADD texcrd.x, texcrd.x, chroma.w;\n");
7268     }
7269     else
7270     {
7271         /* Add 1 */
7272         shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.x;\n");
7273     }
7274     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
7275     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
7276 
7277     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
7278      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
7279      * values due to filtering. */
7280     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
7281     if (type->res_type == WINED3D_GL_RES_TYPE_TEX_2D)
7282     {
7283         /* Multiply the y coordinate by 2/3 and clamp it */
7284         shader_addline(buffer, "MUL texcrd.y, texcrd.y, nv12_coef.x;\n");
7285         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, nv12_coef.x;\n");
7286         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
7287         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
7288     }
7289     else
7290     {
7291         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
7292          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
7293          * is bigger
7294          */
7295         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
7296         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
7297         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
7298     }
7299     *luminance = 'a';
7300 
7301     return TRUE;
7302 }
7303 
7304 /* Context activation is done by the caller. */
7305 static GLuint gen_p8_shader(const struct wined3d_gl_info *gl_info, const struct arbfp_blit_type *type)
7306 {
7307     GLuint shader;
7308     struct wined3d_string_buffer buffer;
7309     const char *tex_target = arbfp_texture_target(type->res_type);
7310 
7311     /* This should not happen because we only use this conversion for
7312      * present blits which don't use color keying. */
7313     if (type->use_color_key)
7314         FIXME("Implement P8 color keying.\n");
7315 
7316     /* Shader header */
7317     if (!string_buffer_init(&buffer))
7318     {
7319         ERR("Failed to initialize shader buffer.\n");
7320         return 0;
7321     }
7322 
7323     GL_EXTCALL(glGenProgramsARB(1, &shader));
7324     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7325     if (!shader)
7326     {
7327         string_buffer_free(&buffer);
7328         return 0;
7329     }
7330 
7331     shader_addline(&buffer, "!!ARBfp1.0\n");
7332     shader_addline(&buffer, "TEMP index;\n");
7333 
7334     /* { 255/256, 0.5/255*255/256, 0, 0 } */
7335     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
7336 
7337     /* The alpha-component contains the palette index */
7338     shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], %s;\n", tex_target);
7339 
7340     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
7341     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
7342 
7343     /* Use the alpha-component as an index in the palette to get the final color */
7344     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
7345     shader_addline(&buffer, "END\n");
7346 
7347     shader_arb_compile(gl_info, GL_FRAGMENT_PROGRAM_ARB, buffer.buffer);
7348 
7349     string_buffer_free(&buffer);
7350 
7351     return shader;
7352 }
7353 
7354 /* Context activation is done by the caller. */
7355 static void upload_palette(struct wined3d_arbfp_blitter *blitter,
7356         const struct wined3d_texture *texture, struct wined3d_context *context)
7357 {
7358     const struct wined3d_palette *palette = texture->swapchain ? texture->swapchain->palette : NULL;
7359     const struct wined3d_gl_info *gl_info = context->gl_info;
7360 
7361     if (!blitter->palette_texture)
7362         gl_info->gl_ops.gl.p_glGenTextures(1, &blitter->palette_texture);
7363 
7364     GL_EXTCALL(glActiveTexture(GL_TEXTURE1));
7365     gl_info->gl_ops.gl.p_glBindTexture(GL_TEXTURE_1D, blitter->palette_texture);
7366 
7367     gl_info->gl_ops.gl.p_glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
7368 
7369     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
7370     /* Make sure we have discrete color levels. */
7371     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
7372     gl_info->gl_ops.gl.p_glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
7373     /* TODO: avoid unneeded uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
7374     if (palette)
7375     {
7376         gl_info->gl_ops.gl.p_glTexImage1D(GL_TEXTURE_1D, 0, GL_RGB, 256, 0, GL_BGRA,
7377                 GL_UNSIGNED_INT_8_8_8_8_REV, palette->colors);
7378     }
7379     else
7380     {
7381         static const DWORD black;
7382         FIXME("P8 surface loaded without a palette.\n");
7383         gl_info->gl_ops.gl.p_glTexImage1D(GL_TEXTURE_1D, 0, GL_RGB, 1, 0, GL_BGRA,
7384                 GL_UNSIGNED_INT_8_8_8_8_REV, &black);
7385     }
7386 
7387     /* Switch back to unit 0 in which the 2D texture will be stored. */
7388     context_active_texture(context, gl_info, 0);
7389 }
7390 
7391 /* Context activation is done by the caller. */
7392 static GLuint gen_yuv_shader(const struct wined3d_gl_info *gl_info, const struct arbfp_blit_type *type)
7393 {
7394     GLuint shader;
7395     struct wined3d_string_buffer buffer;
7396     char luminance_component;
7397 
7398     if (type->use_color_key)
7399         FIXME("Implement YUV color keying.\n");
7400 
7401     /* Shader header */
7402     if (!string_buffer_init(&buffer))
7403     {
7404         ERR("Failed to initialize shader buffer.\n");
7405         return 0;
7406     }
7407 
7408     GL_EXTCALL(glGenProgramsARB(1, &shader));
7409     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
7410     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7411     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7412     if (!shader)
7413     {
7414         string_buffer_free(&buffer);
7415         return 0;
7416     }
7417 
7418     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
7419      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
7420      * two chroma(U and V) values. Each macropixel has two luminance values, one for
7421      * each single pixel it contains, and one U and one V value shared between both
7422      * pixels.
7423      *
7424      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
7425      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
7426      * take the format into account when generating the read swizzles
7427      *
7428      * Reading the Y value is straightforward - just sample the texture. The hardware
7429      * takes care of filtering in the horizontal and vertical direction.
7430      *
7431      * Reading the U and V values is harder. We have to avoid filtering horizontally,
7432      * because that would mix the U and V values of one pixel or two adjacent pixels.
7433      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
7434      * regardless of the filtering setting. Vertical filtering works automatically
7435      * though - the U and V values of two rows are mixed nicely.
7436      *
7437      * Apart of avoiding filtering issues, the code has to know which value it just
7438      * read, and where it can find the other one. To determine this, it checks if
7439      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
7440      *
7441      * Handling horizontal filtering of U and V values requires reading a 2nd pair
7442      * of pixels, extracting U and V and mixing them. This is not implemented yet.
7443      *
7444      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
7445      * with width / 2. This way one read gives all 3 values, finding U and V is easy
7446      * in an unfiltered situation. Finding the luminance on the other hand requires
7447      * finding out if it is an odd or even pixel. The real drawback of this approach
7448      * is filtering. This would have to be emulated completely in the shader, reading
7449      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
7450      * vertically. Beyond that it would require adjustments to the texture handling
7451      * code to deal with the width scaling
7452      */
7453     shader_addline(&buffer, "!!ARBfp1.0\n");
7454     shader_addline(&buffer, "TEMP luminance;\n");
7455     shader_addline(&buffer, "TEMP temp;\n");
7456     shader_addline(&buffer, "TEMP chroma;\n");
7457     shader_addline(&buffer, "TEMP texcrd;\n");
7458     shader_addline(&buffer, "TEMP texcrd2;\n");
7459     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
7460     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
7461     shader_addline(&buffer, "PARAM size = program.local[%u];\n", ARBFP_BLIT_PARAM_SIZE);
7462 
7463     switch (type->fixup)
7464     {
7465         case COMPLEX_FIXUP_UYVY:
7466         case COMPLEX_FIXUP_YUY2:
7467             if (!gen_planar_yuv_read(&buffer, type, &luminance_component))
7468             {
7469                 string_buffer_free(&buffer);
7470                 return 0;
7471             }
7472             break;
7473 
7474         case COMPLEX_FIXUP_YV12:
7475             if (!gen_yv12_read(&buffer, type, &luminance_component))
7476             {
7477                 string_buffer_free(&buffer);
7478                 return 0;
7479             }
7480             break;
7481 
7482         case COMPLEX_FIXUP_NV12:
7483             if (!gen_nv12_read(&buffer, type, &luminance_component))
7484             {
7485                 string_buffer_free(&buffer);
7486                 return 0;
7487             }
7488             break;
7489 
7490         default:
7491             FIXME("Unsupported YUV fixup %#x\n", type->fixup);
7492             string_buffer_free(&buffer);
7493             return 0;
7494     }
7495 
7496     /* Calculate the final result. Formula is taken from
7497      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
7498      * ranges from -0.5 to 0.5
7499      */
7500     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
7501 
7502     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
7503     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
7504     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
7505     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
7506     shader_addline(&buffer, "END\n");
7507 
7508     shader_arb_compile(gl_info, GL_FRAGMENT_PROGRAM_ARB, buffer.buffer);
7509 
7510     string_buffer_free(&buffer);
7511 
7512     return shader;
7513 }
7514 
7515 /* Context activation is done by the caller. */
7516 static GLuint arbfp_gen_plain_shader(const struct wined3d_gl_info *gl_info, const struct arbfp_blit_type *type)
7517 {
7518     GLuint shader;
7519     struct wined3d_string_buffer buffer;
7520     const char *tex_target = arbfp_texture_target(type->res_type);
7521 
7522     /* Shader header */
7523     if (!string_buffer_init(&buffer))
7524     {
7525         ERR("Failed to initialize shader buffer.\n");
7526         return 0;
7527     }
7528 
7529     GL_EXTCALL(glGenProgramsARB(1, &shader));
7530     if (!shader)
7531     {
7532         string_buffer_free(&buffer);
7533         return 0;
7534     }
7535     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7536 
7537     shader_addline(&buffer, "!!ARBfp1.0\n");
7538 
7539     if (type->use_color_key)
7540     {
7541         shader_addline(&buffer, "TEMP color;\n");
7542         shader_addline(&buffer, "TEMP less, greater;\n");
7543         shader_addline(&buffer, "PARAM color_key_low = program.local[%u];\n", ARBFP_BLIT_PARAM_COLOR_KEY_LOW);
7544         shader_addline(&buffer, "PARAM color_key_high = program.local[%u];\n", ARBFP_BLIT_PARAM_COLOR_KEY_HIGH);
7545         shader_addline(&buffer, "TEX color, fragment.texcoord[0], texture[0], %s;\n", tex_target);
7546         shader_addline(&buffer, "SLT less, color, color_key_low;\n"); /* below low key */
7547         shader_addline(&buffer, "SGE greater, color, color_key_high;\n"); /* above high key */
7548         shader_addline(&buffer, "ADD less, less, greater;\n"); /* or */
7549         shader_addline(&buffer, "DP4 less.b, less, less;\n"); /* on any channel */
7550         shader_addline(&buffer, "SGE less, -less.b, 0.0;\n"); /* logical not */
7551         shader_addline(&buffer, "KIL -less;\n"); /* discard if true */
7552         shader_addline(&buffer, "MOV result.color, color;\n");
7553     }
7554     else
7555     {
7556         shader_addline(&buffer, "TEX result.color, fragment.texcoord[0], texture[0], %s;\n", tex_target);
7557     }
7558 
7559     shader_addline(&buffer, "END\n");
7560 
7561     shader_arb_compile(gl_info, GL_FRAGMENT_PROGRAM_ARB, buffer.buffer);
7562 
7563     string_buffer_free(&buffer);
7564 
7565     return shader;
7566 }
7567 
7568 /* Context activation is done by the caller. */
7569 static HRESULT arbfp_blit_set(struct wined3d_arbfp_blitter *blitter, struct wined3d_context *context,
7570         const struct wined3d_texture *texture, unsigned int sub_resource_idx,
7571         const struct wined3d_color_key *color_key)
7572 {
7573     enum complex_fixup fixup;
7574     const struct wined3d_gl_info *gl_info = context->gl_info;
7575     struct wine_rb_entry *entry;
7576     struct arbfp_blit_type type;
7577     struct arbfp_blit_desc *desc;
7578     struct wined3d_color float_color_key[2];
7579     struct wined3d_vec4 size;
7580     unsigned int level;
7581     GLuint shader;
7582 
7583     level = sub_resource_idx % texture->level_count;
7584     size.x = wined3d_texture_get_level_pow2_width(texture, level);
7585     size.y = wined3d_texture_get_level_pow2_height(texture, level);
7586     size.z = 1.0f;
7587     size.w = 1.0f;
7588 
7589     if (is_complex_fixup(texture->resource.format->color_fixup))
7590         fixup = get_complex_fixup(texture->resource.format->color_fixup);
7591     else
7592         fixup = COMPLEX_FIXUP_NONE;
7593 
7594     switch (texture->target)
7595     {
7596         case GL_TEXTURE_1D:
7597             type.res_type = WINED3D_GL_RES_TYPE_TEX_1D;
7598             break;
7599 
7600         case GL_TEXTURE_2D:
7601             type.res_type = WINED3D_GL_RES_TYPE_TEX_2D;
7602             break;
7603 
7604         case GL_TEXTURE_3D:
7605             type.res_type = WINED3D_GL_RES_TYPE_TEX_3D;
7606             break;
7607 
7608         case GL_TEXTURE_CUBE_MAP_ARB:
7609             type.res_type = WINED3D_GL_RES_TYPE_TEX_CUBE;
7610             break;
7611 
7612         case GL_TEXTURE_RECTANGLE_ARB:
7613             type.res_type = WINED3D_GL_RES_TYPE_TEX_RECT;
7614             break;
7615 
7616         default:
7617             ERR("Unexpected GL texture type %#x.\n", texture->target);
7618             type.res_type = WINED3D_GL_RES_TYPE_TEX_2D;
7619     }
7620     type.fixup = fixup;
7621     type.use_color_key = !!color_key;
7622     type.padding = 0;
7623 
7624     if ((entry = wine_rb_get(&blitter->shaders, &type)))
7625     {
7626         desc = WINE_RB_ENTRY_VALUE(entry, struct arbfp_blit_desc, entry);
7627         shader = desc->shader;
7628     }
7629     else
7630     {
7631         switch (fixup)
7632         {
7633             case COMPLEX_FIXUP_NONE:
7634                 if (!is_identity_fixup(texture->resource.format->color_fixup))
7635                     FIXME("Implement support for sign or swizzle fixups.\n");
7636                 shader = arbfp_gen_plain_shader(gl_info, &type);
7637                 break;
7638 
7639             case COMPLEX_FIXUP_P8:
7640                 shader = gen_p8_shader(gl_info, &type);
7641                 break;
7642 
7643             case COMPLEX_FIXUP_YUY2:
7644             case COMPLEX_FIXUP_UYVY:
7645             case COMPLEX_FIXUP_YV12:
7646             case COMPLEX_FIXUP_NV12:
7647                 shader = gen_yuv_shader(gl_info, &type);
7648                 break;
7649         }
7650 
7651         if (!shader)
7652         {
7653             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7654             return E_NOTIMPL;
7655         }
7656 
7657         if (!(desc = heap_alloc(sizeof(*desc))))
7658             goto err_out;
7659 
7660         desc->type = type;
7661         desc->shader = shader;
7662         if (wine_rb_put(&blitter->shaders, &desc->type, &desc->entry) == -1)
7663         {
7664 err_out:
7665             ERR("Out of memory\n");
7666             GL_EXTCALL(glDeleteProgramsARB(1, &shader));
7667             checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader))");
7668             GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, 0));
7669             checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, 0)");
7670             heap_free(desc);
7671             return E_OUTOFMEMORY;
7672         }
7673     }
7674 
7675     if (fixup == COMPLEX_FIXUP_P8)
7676         upload_palette(blitter, texture, context);
7677 
7678     gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
7679     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7680     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7681     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7682     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARBFP_BLIT_PARAM_SIZE, &size.x));
7683     checkGLcall("glProgramLocalParameter4fvARB");
7684     if (type.use_color_key)
7685     {
7686         wined3d_format_get_float_color_key(texture->resource.format, color_key, float_color_key);
7687         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
7688                 ARBFP_BLIT_PARAM_COLOR_KEY_LOW, &float_color_key[0].r));
7689         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
7690                 ARBFP_BLIT_PARAM_COLOR_KEY_HIGH, &float_color_key[1].r));
7691         checkGLcall("glProgramLocalParameter4fvARB");
7692     }
7693 
7694     return WINED3D_OK;
7695 }
7696 
7697 /* Context activation is done by the caller. */
7698 static void arbfp_blit_unset(const struct wined3d_gl_info *gl_info)
7699 {
7700     gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
7701     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7702 }
7703 
7704 static BOOL arbfp_blit_supported(enum wined3d_blit_op blit_op, const struct wined3d_context *context,
7705         const struct wined3d_resource *src_resource, DWORD src_location,
7706         const struct wined3d_resource *dst_resource, DWORD dst_location)
7707 {
7708     const struct wined3d_format *src_format = src_resource->format;
7709     const struct wined3d_format *dst_format = dst_resource->format;
7710     enum complex_fixup src_fixup;
7711     BOOL decompress;
7712 
7713     if (!context->gl_info->supported[ARB_FRAGMENT_PROGRAM])
7714         return FALSE;
7715 
7716     if (blit_op == WINED3D_BLIT_OP_RAW_BLIT && dst_format->id == src_format->id)
7717     {
7718         if (dst_format->flags[WINED3D_GL_RES_TYPE_TEX_2D] & (WINED3DFMT_FLAG_DEPTH | WINED3DFMT_FLAG_STENCIL))
7719             blit_op = WINED3D_BLIT_OP_DEPTH_BLIT;
7720         else
7721             blit_op = WINED3D_BLIT_OP_COLOR_BLIT;
7722     }
7723 
7724     switch (blit_op)
7725     {
7726         case WINED3D_BLIT_OP_COLOR_BLIT_CKEY:
7727             if (!context->d3d_info->shader_color_key)
7728             {
7729                 /* The conversion modifies the alpha channel so the color key might no longer match. */
7730                 TRACE("Color keying not supported with converted textures.\n");
7731                 return FALSE;
7732             }
7733         case WINED3D_BLIT_OP_COLOR_BLIT_ALPHATEST:
7734         case WINED3D_BLIT_OP_COLOR_BLIT:
7735             break;
7736 
7737         default:
7738             TRACE("Unsupported blit_op=%d\n", blit_op);
7739             return FALSE;
7740     }
7741 
7742     decompress = src_format && (src_format->flags[WINED3D_GL_RES_TYPE_TEX_2D] & WINED3DFMT_FLAG_COMPRESSED)
7743             && !(dst_format->flags[WINED3D_GL_RES_TYPE_TEX_2D] & WINED3DFMT_FLAG_COMPRESSED);
7744     if (!decompress && !(src_resource->access & dst_resource->access & WINED3D_RESOURCE_ACCESS_GPU))
7745         return FALSE;
7746 
7747     src_fixup = get_complex_fixup(src_format->color_fixup);
7748     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7749     {
7750         TRACE("Checking support for fixup:\n");
7751         dump_color_fixup_desc(src_format->color_fixup);
7752     }
7753 
7754     if (!is_identity_fixup(dst_format->color_fixup)
7755             && (dst_format->id != src_format->id || dst_location != WINED3D_LOCATION_DRAWABLE))
7756     {
7757         TRACE("Destination fixups are not supported\n");
7758         return FALSE;
7759     }
7760 
7761     if (is_identity_fixup(src_format->color_fixup))
7762     {
7763         TRACE("[OK]\n");
7764         return TRUE;
7765     }
7766 
7767      /* We only support YUV conversions. */
7768     if (!is_complex_fixup(src_format->color_fixup))
7769     {
7770         if (wined3d_settings.offscreen_rendering_mode == ORM_BACKBUFFER)
7771         {
7772             WARN("Claiming fixup support because of ORM_BACKBUFFER.\n");
7773             return TRUE;
7774         }
7775 
7776         TRACE("[FAILED]\n");
7777         return FALSE;
7778     }
7779 
7780     switch(src_fixup)
7781     {
7782         case COMPLEX_FIXUP_YUY2:
7783         case COMPLEX_FIXUP_UYVY:
7784         case COMPLEX_FIXUP_YV12:
7785         case COMPLEX_FIXUP_NV12:
7786         case COMPLEX_FIXUP_P8:
7787             TRACE("[OK]\n");
7788             return TRUE;
7789 
7790         default:
7791             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7792             TRACE("[FAILED]\n");
7793             return FALSE;
7794     }
7795 }
7796 
7797 static DWORD arbfp_blitter_blit(struct wined3d_blitter *blitter, enum wined3d_blit_op op,
7798         struct wined3d_context *context, struct wined3d_surface *src_surface, DWORD src_location,
7799         const RECT *src_rect, struct wined3d_surface *dst_surface, DWORD dst_location, const RECT *dst_rect,
7800         const struct wined3d_color_key *color_key, enum wined3d_texture_filter_type filter)
7801 {
7802     unsigned int src_sub_resource_idx = surface_get_sub_resource_idx(src_surface);
7803     struct wined3d_texture *src_texture = src_surface->container;
7804     struct wined3d_texture *dst_texture = dst_surface->container;
7805     struct wined3d_device *device = dst_texture->resource.device;
7806     struct wined3d_arbfp_blitter *arbfp_blitter;
7807     struct wined3d_color_key alpha_test_key;
7808     struct wined3d_blitter *next;
7809     RECT s, d;
7810 
7811     if (!arbfp_blit_supported(op, context, &src_texture->resource, src_location,
7812             &dst_texture->resource, dst_location))
7813     {
7814         if ((next = blitter->next))
7815             return next->ops->blitter_blit(next, op, context, src_surface, src_location,
7816                     src_rect, dst_surface, dst_location, dst_rect, color_key, filter);
7817     }
7818 
7819     arbfp_blitter = CONTAINING_RECORD(blitter, struct wined3d_arbfp_blitter, blitter);
7820 
7821     /* Now load the surface */
7822     if (wined3d_settings.offscreen_rendering_mode != ORM_FBO
7823             && (surface_get_sub_resource(src_surface)->locations
7824             & (WINED3D_LOCATION_TEXTURE_RGB | WINED3D_LOCATION_DRAWABLE))
7825             == WINED3D_LOCATION_DRAWABLE
7826             && !wined3d_resource_is_offscreen(&src_texture->resource))
7827     {
7828         unsigned int src_level = src_sub_resource_idx % src_texture->level_count;
7829 
7830         /* Without FBO blits transferring from the drawable to the texture is
7831          * expensive, because we have to flip the data in sysmem. Since we can
7832          * flip in the blitter, we don't actually need that flip anyway. So we
7833          * use the surface's texture as scratch texture, and flip the source
7834          * rectangle instead. */
7835         surface_load_fb_texture(src_surface, FALSE, context);
7836 
7837         s = *src_rect;
7838         s.top = wined3d_texture_get_level_height(src_texture, src_level) - s.top;
7839         s.bottom = wined3d_texture_get_level_height(src_texture, src_level) - s.bottom;
7840         src_rect = &s;
7841     }
7842     else
7843         wined3d_texture_load(src_texture, context, FALSE);
7844 
7845     context_apply_blit_state(context, device);
7846 
7847     if (dst_location == WINED3D_LOCATION_DRAWABLE)
7848     {
7849         d = *dst_rect;
7850         surface_translate_drawable_coords(dst_surface, context->win_handle, &d);
7851         dst_rect = &d;
7852     }
7853 
7854     if (wined3d_settings.offscreen_rendering_mode == ORM_FBO)
7855     {
7856         GLenum buffer;
7857 
7858         if (dst_location == WINED3D_LOCATION_DRAWABLE)
7859         {
7860             TRACE("Destination surface %p is onscreen.\n", dst_surface);
7861             buffer = wined3d_texture_get_gl_buffer(dst_texture);
7862         }
7863         else
7864         {
7865             TRACE("Destination surface %p is offscreen.\n", dst_surface);
7866             buffer = GL_COLOR_ATTACHMENT0;
7867         }
7868         context_apply_fbo_state_blit(context, GL_DRAW_FRAMEBUFFER, dst_surface, NULL, dst_location);
7869         context_set_draw_buffer(context, buffer);
7870         context_check_fbo_status(context, GL_DRAW_FRAMEBUFFER);
7871         context_invalidate_state(context, STATE_FRAMEBUFFER);
7872     }
7873 
7874     if (op == WINED3D_BLIT_OP_COLOR_BLIT_ALPHATEST)
7875     {
7876         const struct wined3d_format *fmt = src_texture->resource.format;
7877         alpha_test_key.color_space_low_value = 0;
7878         alpha_test_key.color_space_high_value = ~(((1u << fmt->alpha_size) - 1) << fmt->alpha_offset);
7879         color_key = &alpha_test_key;
7880     }
7881 
7882     arbfp_blit_set(arbfp_blitter, context, src_texture, src_sub_resource_idx, color_key);
7883 
7884     /* Draw a textured quad */
7885     draw_textured_quad(src_texture, src_sub_resource_idx, context, src_rect, dst_rect, filter);
7886 
7887     /* Leave the opengl state valid for blitting */
7888     arbfp_blit_unset(context->gl_info);
7889 
7890     if (wined3d_settings.strict_draw_ordering
7891             || (dst_texture->swapchain && (dst_texture->swapchain->front_buffer == dst_texture)))
7892         context->gl_info->gl_ops.gl.p_glFlush(); /* Flush to ensure ordering across contexts. */
7893 
7894     return dst_location;
7895 }
7896 
7897 static void arbfp_blitter_clear(struct wined3d_blitter *blitter, struct wined3d_device *device,
7898         unsigned int rt_count, const struct wined3d_fb_state *fb, unsigned int rect_count, const RECT *clear_rects,
7899         const RECT *draw_rect, DWORD flags, const struct wined3d_color *colour, float depth, DWORD stencil)
7900 {
7901     struct wined3d_blitter *next;
7902 
7903     if ((next = blitter->next))
7904         next->ops->blitter_clear(next, device, rt_count, fb, rect_count,
7905                 clear_rects, draw_rect, flags, colour, depth, stencil);
7906 }
7907 
7908 static const struct wined3d_blitter_ops arbfp_blitter_ops =
7909 {
7910     arbfp_blitter_destroy,
7911     arbfp_blitter_clear,
7912     arbfp_blitter_blit,
7913 };
7914 
7915 void wined3d_arbfp_blitter_create(struct wined3d_blitter **next, const struct wined3d_device *device)
7916 {
7917     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
7918     struct wined3d_arbfp_blitter *blitter;
7919 
7920     if (device->shader_backend != &arb_program_shader_backend
7921             && device->shader_backend != &glsl_shader_backend)
7922         return;
7923 
7924     if (!gl_info->supported[ARB_FRAGMENT_PROGRAM])
7925         return;
7926 
7927     if (!gl_info->supported[WINED3D_GL_LEGACY_CONTEXT])
7928         return;
7929 
7930     if (!(blitter = heap_alloc(sizeof(*blitter))))
7931     {
7932         ERR("Failed to allocate blitter.\n");
7933         return;
7934     }
7935 
7936     TRACE("Created blitter %p.\n", blitter);
7937 
7938     blitter->blitter.ops = &arbfp_blitter_ops;
7939     blitter->blitter.next = *next;
7940     wine_rb_init(&blitter->shaders, arbfp_blit_type_compare);
7941     blitter->palette_texture = 0;
7942     *next = &blitter->blitter;
7943 }
7944