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, 1, 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 != WINED3D_VP_MODE_SHADER)
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 char *tex_dst = dst_str;
1420     struct color_fixup_masks masks;
1421 
1422     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1423     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1424 
1425     switch (resource_type)
1426     {
1427         case WINED3D_SHADER_RESOURCE_TEXTURE_1D:
1428             tex_type = "1D";
1429             break;
1430 
1431         case WINED3D_SHADER_RESOURCE_TEXTURE_2D:
1432             if (pshader && priv->cur_ps_args->super.np2_fixup & (1u << sampler_idx)
1433                     && ins->ctx->gl_info->supported[ARB_TEXTURE_RECTANGLE])
1434                 tex_type = "RECT";
1435             else
1436                 tex_type = "2D";
1437             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1438             {
1439                 if (priv->cur_np2fixup_info->super.active & (1u << sampler_idx))
1440                 {
1441                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1442                     else np2_fixup = TRUE;
1443                 }
1444             }
1445             break;
1446 
1447         case WINED3D_SHADER_RESOURCE_TEXTURE_3D:
1448             tex_type = "3D";
1449             break;
1450 
1451         case WINED3D_SHADER_RESOURCE_TEXTURE_CUBE:
1452             tex_type = "CUBE";
1453             break;
1454 
1455         default:
1456             ERR("Unexpected resource type %#x.\n", resource_type);
1457             tex_type = "";
1458     }
1459 
1460     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1461      * so don't use shader_arb_get_modifier
1462      */
1463     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1464     else mod = "";
1465 
1466     /* Fragment samplers always have indentity mapping */
1467     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1468     {
1469         sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1470     }
1471 
1472     if (pshader)
1473     {
1474         masks = calc_color_correction(priv->cur_ps_args->super.color_fixup[sampler_idx],
1475                 ins->dst[0].write_mask);
1476 
1477         if (masks.source || masks.sign)
1478             tex_dst = "TA";
1479     }
1480 
1481     if (flags & TEX_DERIV)
1482     {
1483         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1484         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1485         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, tex_dst, coord_reg,
1486                        dsx, dsy, sampler_idx, tex_type);
1487     }
1488     else if(flags & TEX_LOD)
1489     {
1490         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1491         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1492         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, tex_dst, coord_reg,
1493                        sampler_idx, tex_type);
1494     }
1495     else if (flags & TEX_BIAS)
1496     {
1497         /* Shouldn't be possible, but let's check for it */
1498         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1499         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1500         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, tex_dst, coord_reg, sampler_idx, tex_type);
1501     }
1502     else if (flags & TEX_PROJ)
1503     {
1504         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, tex_dst, coord_reg, sampler_idx, tex_type);
1505     }
1506     else
1507     {
1508         if (np2_fixup)
1509         {
1510             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1511             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1512                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1513 
1514             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, tex_dst, sampler_idx, tex_type);
1515         }
1516         else
1517             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, tex_dst, coord_reg, sampler_idx, tex_type);
1518     }
1519 
1520     if (pshader)
1521     {
1522         gen_color_correction(buffer, dst_str, tex_dst,
1523                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_ONE),
1524                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_TWO),
1525                 priv->cur_ps_args->super.color_fixup[sampler_idx], masks);
1526     }
1527 }
1528 
1529 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1530         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1531 {
1532     /* Generate a line that does the input modifier computation and return the input register to use */
1533     BOOL is_color = FALSE, insert_line;
1534     char regstr[256];
1535     char swzstr[20];
1536     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
1537     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1538     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
1539     const char *two = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_TWO);
1540 
1541     /* Assume a new line will be added */
1542     insert_line = TRUE;
1543 
1544     /* Get register name */
1545     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1546     shader_arb_get_swizzle(src, is_color, swzstr);
1547 
1548     switch (src->modifiers)
1549     {
1550     case WINED3DSPSM_NONE:
1551         sprintf(outregstr, "%s%s", regstr, swzstr);
1552         insert_line = FALSE;
1553         break;
1554     case WINED3DSPSM_NEG:
1555         sprintf(outregstr, "-%s%s", regstr, swzstr);
1556         insert_line = FALSE;
1557         break;
1558     case WINED3DSPSM_BIAS:
1559         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1560         break;
1561     case WINED3DSPSM_BIASNEG:
1562         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1563         break;
1564     case WINED3DSPSM_SIGN:
1565         shader_addline(buffer, "MAD T%c, %s, %s, -%s;\n", 'A' + tmpreg, regstr, two, one);
1566         break;
1567     case WINED3DSPSM_SIGNNEG:
1568         shader_addline(buffer, "MAD T%c, %s, -%s, %s;\n", 'A' + tmpreg, regstr, two, one);
1569         break;
1570     case WINED3DSPSM_COMP:
1571         shader_addline(buffer, "SUB T%c, %s, %s;\n", 'A' + tmpreg, one, regstr);
1572         break;
1573     case WINED3DSPSM_X2:
1574         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1575         break;
1576     case WINED3DSPSM_X2NEG:
1577         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1578         break;
1579     case WINED3DSPSM_DZ:
1580         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1581         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1582         break;
1583     case WINED3DSPSM_DW:
1584         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1585         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1586         break;
1587     case WINED3DSPSM_ABS:
1588         if(ctx->target_version >= NV2) {
1589             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1590             insert_line = FALSE;
1591         } else {
1592             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1593         }
1594         break;
1595     case WINED3DSPSM_ABSNEG:
1596         if(ctx->target_version >= NV2) {
1597             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1598         } else {
1599             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1600             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1601         }
1602         insert_line = FALSE;
1603         break;
1604     default:
1605         sprintf(outregstr, "%s%s", regstr, swzstr);
1606         insert_line = FALSE;
1607     }
1608 
1609     /* Return modified or original register, with swizzle */
1610     if (insert_line)
1611         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1612 }
1613 
1614 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1615 {
1616     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1617     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
1618     DWORD sampler_code = dst->reg.idx[0].offset;
1619     char dst_name[50];
1620     char src_name[2][50];
1621 
1622     shader_arb_get_dst_param(ins, dst, dst_name);
1623 
1624     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1625      *
1626      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1627      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1628      * temps is done.
1629      */
1630     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1631     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1632     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1633     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1634     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1635 
1636     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1637     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1638 }
1639 
1640 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1641 {
1642     *extra_char = ' ';
1643     switch(mod)
1644     {
1645         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1646         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1647         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1648         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1649         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1650         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1651         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1652         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1653         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1654         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1655         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1656         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1657         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1658     }
1659     FIXME("Unknown modifier %u\n", mod);
1660     return mod;
1661 }
1662 
1663 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1664 {
1665     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1666     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
1667     char dst_name[50];
1668     char src_name[3][50];
1669     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1670             ins->ctx->reg_maps->shader_version.minor);
1671 
1672     shader_arb_get_dst_param(ins, dst, dst_name);
1673     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1674 
1675     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue
1676             && ins->dst->write_mask != WINED3DSP_WRITEMASK_3)
1677     {
1678         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1679     }
1680     else
1681     {
1682         struct wined3d_shader_src_param src0_copy = ins->src[0];
1683         char extra_neg;
1684 
1685         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1686         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1687 
1688         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1689         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1690         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1691         shader_addline(buffer, "CMP%s %s, TA, %s, %s;\n", shader_arb_get_modifier(ins),
1692                 dst_name, src_name[1], src_name[2]);
1693     }
1694 }
1695 
1696 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1697 {
1698     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1699     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
1700     char dst_name[50];
1701     char src_name[3][50];
1702 
1703     shader_arb_get_dst_param(ins, dst, dst_name);
1704 
1705     /* Generate input register names (with modifiers) */
1706     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1707     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1708     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1709 
1710     shader_addline(buffer, "CMP%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1711             dst_name, src_name[0], src_name[2], src_name[1]);
1712 }
1713 
1714 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1715  * dst = dot2(src0, src1) + src2 */
1716 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1717 {
1718     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1719     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
1720     char dst_name[50];
1721     char src_name[3][50];
1722     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1723 
1724     shader_arb_get_dst_param(ins, dst, dst_name);
1725     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1726     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1727 
1728     if(ctx->target_version >= NV3)
1729     {
1730         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1731         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1732         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1733                        dst_name, src_name[0], src_name[1], src_name[2]);
1734     }
1735     else if(ctx->target_version >= NV2)
1736     {
1737         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1738          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1739          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1740          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1741          *
1742          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1743          *
1744          * .xyxy and other swizzles that we could get with this are not valid in
1745          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1746          */
1747         struct wined3d_shader_src_param tmp_param = ins->src[1];
1748         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1749         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1750 
1751         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1752 
1753         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1754                        dst_name, src_name[2], src_name[0], src_name[1]);
1755     }
1756     else
1757     {
1758         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1759         /* 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]
1760         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1761         */
1762         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1763         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1764         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1765         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1766     }
1767 }
1768 
1769 /* Map the opcode 1-to-1 to the GL code */
1770 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1771 {
1772     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
1773     const char *instruction;
1774     char arguments[256], dst_str[50];
1775     unsigned int i;
1776     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1777 
1778     switch (ins->handler_idx)
1779     {
1780         case WINED3DSIH_ABS: instruction = "ABS"; break;
1781         case WINED3DSIH_ADD: instruction = "ADD"; break;
1782         case WINED3DSIH_CRS: instruction = "XPD"; break;
1783         case WINED3DSIH_DP3: instruction = "DP3"; break;
1784         case WINED3DSIH_DP4: instruction = "DP4"; break;
1785         case WINED3DSIH_DST: instruction = "DST"; break;
1786         case WINED3DSIH_FRC: instruction = "FRC"; break;
1787         case WINED3DSIH_LIT: instruction = "LIT"; break;
1788         case WINED3DSIH_LRP: instruction = "LRP"; break;
1789         case WINED3DSIH_MAD: instruction = "MAD"; break;
1790         case WINED3DSIH_MAX: instruction = "MAX"; break;
1791         case WINED3DSIH_MIN: instruction = "MIN"; break;
1792         case WINED3DSIH_MOV: instruction = "MOV"; break;
1793         case WINED3DSIH_MUL: instruction = "MUL"; break;
1794         case WINED3DSIH_SGE: instruction = "SGE"; break;
1795         case WINED3DSIH_SLT: instruction = "SLT"; break;
1796         case WINED3DSIH_SUB: instruction = "SUB"; break;
1797         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1798         case WINED3DSIH_DSX: instruction = "DDX"; break;
1799         default: instruction = "";
1800             FIXME("Unhandled opcode %s.\n", debug_d3dshaderinstructionhandler(ins->handler_idx));
1801             break;
1802     }
1803 
1804     /* Note that shader_arb_add_dst_param() adds spaces. */
1805     arguments[0] = '\0';
1806     shader_arb_get_dst_param(ins, dst, dst_str);
1807     for (i = 0; i < ins->src_count; ++i)
1808     {
1809         char operand[100];
1810         strcat(arguments, ", ");
1811         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1812         strcat(arguments, operand);
1813     }
1814     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1815 }
1816 
1817 static void shader_hw_nop(const struct wined3d_shader_instruction *ins) {}
1818 
1819 static DWORD shader_arb_select_component(DWORD swizzle, DWORD component)
1820 {
1821     return ((swizzle >> 2 * component) & 0x3) * 0x55;
1822 }
1823 
1824 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1825 {
1826     const struct wined3d_shader *shader = ins->ctx->shader;
1827     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1828     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1829     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1830     const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
1831     const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
1832     const char *two = arb_get_helper_value(reg_maps->shader_version.type, ARB_TWO);
1833 
1834     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
1835     char src0_param[256];
1836 
1837     if (ins->handler_idx == WINED3DSIH_MOVA)
1838     {
1839         const struct arb_vshader_private *shader_data = shader->backend_data;
1840         char write_mask[6];
1841         const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1842 
1843         if(ctx->target_version >= NV2) {
1844             shader_hw_map2gl(ins);
1845             return;
1846         }
1847         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1848         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1849 
1850         /* This implements the mova formula used in GLSL. The first two instructions
1851          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1852          * in this case:
1853          * mova A0.x, 0.0
1854          *
1855          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1856          *
1857          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1858          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1859          */
1860         shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
1861         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
1862 
1863         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1864         shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
1865         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1866         if (shader_data->rel_offset)
1867         {
1868             shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
1869         }
1870         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1871 
1872         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1873     }
1874     else if (reg_maps->shader_version.major == 1
1875           && !shader_is_pshader_version(reg_maps->shader_version.type)
1876           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1877     {
1878         const struct arb_vshader_private *shader_data = shader->backend_data;
1879         src0_param[0] = '\0';
1880 
1881         if (shader_data->rel_offset && ctx->target_version == ARB)
1882         {
1883             const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1884             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1885             shader_addline(buffer, "ADD TA.x, %s, %s;\n", src0_param, offset);
1886             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1887         }
1888         else
1889         {
1890             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1891              * with more than one component. Thus replicate the first source argument over all
1892              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1893             struct wined3d_shader_src_param tmp_src = ins->src[0];
1894             tmp_src.swizzle = shader_arb_select_component(tmp_src.swizzle, 0);
1895             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1896             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1897         }
1898     }
1899     else if (ins->dst[0].reg.type == WINED3DSPR_COLOROUT && !ins->dst[0].reg.idx[0].offset && pshader)
1900     {
1901         if (ctx->ps_post_process && shader->u.ps.color0_mov)
1902         {
1903             shader_addline(buffer, "#mov handled in srgb write or fog code\n");
1904             return;
1905         }
1906         shader_hw_map2gl(ins);
1907     }
1908     else
1909     {
1910         shader_hw_map2gl(ins);
1911     }
1912 }
1913 
1914 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1915 {
1916     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1917     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
1918     char reg_dest[40];
1919 
1920     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1921      * but >= 2.0 honors it (undocumented, but tested by the d3d9 testsuite)
1922      */
1923     shader_arb_get_dst_param(ins, dst, reg_dest);
1924 
1925     if (ins->ctx->reg_maps->shader_version.major >= 2)
1926     {
1927         const char *kilsrc = "TA";
1928         BOOL is_color;
1929 
1930         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1931         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1932         {
1933             kilsrc = reg_dest;
1934         }
1935         else
1936         {
1937             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1938              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1939              * masked out components to 0(won't kill)
1940              */
1941             char x = '0', y = '0', z = '0', w = '0';
1942             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1943             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1944             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1945             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1946             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1947         }
1948         shader_addline(buffer, "KIL %s;\n", kilsrc);
1949     }
1950     else
1951     {
1952         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1953          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1954          *
1955          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1956          * or pass in any temporary register(in shader phase 2)
1957          */
1958         if (ins->ctx->reg_maps->shader_version.minor <= 3)
1959             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx[0].offset);
1960         else
1961             shader_arb_get_dst_param(ins, dst, reg_dest);
1962         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1963         shader_addline(buffer, "KIL TA;\n");
1964     }
1965 }
1966 
1967 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1968 {
1969     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1970     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1971     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1972             ins->ctx->reg_maps->shader_version.minor);
1973     struct wined3d_shader_src_param src;
1974 
1975     char reg_dest[40];
1976     char reg_coord[40];
1977     DWORD reg_sampler_code;
1978     WORD myflags = 0;
1979     BOOL swizzle_coord = FALSE;
1980 
1981     /* All versions have a destination register */
1982     shader_arb_get_dst_param(ins, dst, reg_dest);
1983 
1984     /* 1.0-1.4: Use destination register number as texture code.
1985        2.0+: Use provided sampler number as texture code. */
1986     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1987         reg_sampler_code = dst->reg.idx[0].offset;
1988     else
1989         reg_sampler_code = ins->src[1].reg.idx[0].offset;
1990 
1991     /* 1.0-1.3: Use the texcoord varying.
1992        1.4+: Use provided coordinate source register. */
1993     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1994         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1995     else {
1996         /* TEX is the only instruction that can handle DW and DZ natively */
1997         src = ins->src[0];
1998         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1999         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
2000         shader_arb_get_src_param(ins, &src, 0, reg_coord);
2001     }
2002 
2003     /* projection flag:
2004      * 1.1, 1.2, 1.3: Use WINED3D_TSS_TEXTURETRANSFORMFLAGS
2005      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
2006      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
2007      */
2008     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2009     {
2010         DWORD flags = 0;
2011         if (reg_sampler_code < MAX_TEXTURES)
2012             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
2013         if (flags & WINED3D_PSARGS_PROJECTED)
2014         {
2015             myflags |= TEX_PROJ;
2016             if ((flags & ~WINED3D_PSARGS_PROJECTED) == WINED3D_TTFF_COUNT3)
2017                 swizzle_coord = TRUE;
2018         }
2019     }
2020     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
2021     {
2022         enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
2023         if (src_mod == WINED3DSPSM_DZ)
2024         {
2025             swizzle_coord = TRUE;
2026             myflags |= TEX_PROJ;
2027         } else if(src_mod == WINED3DSPSM_DW) {
2028             myflags |= TEX_PROJ;
2029         }
2030     } else {
2031         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
2032         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
2033     }
2034 
2035     if (swizzle_coord)
2036     {
2037         /* TXP cannot handle DZ natively, so move the z coordinate to .w.
2038          * reg_coord is a read-only varying register, so we need a temp reg */
2039         shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
2040         strcpy(reg_coord, "TA");
2041     }
2042 
2043     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
2044 }
2045 
2046 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
2047 {
2048     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2049     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2050     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2051             ins->ctx->reg_maps->shader_version.minor);
2052     char dst_str[50];
2053 
2054     if (shader_version < WINED3D_SHADER_VERSION(1,4))
2055     {
2056         DWORD reg = dst->reg.idx[0].offset;
2057 
2058         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2059         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
2060     } else {
2061         char reg_src[40];
2062 
2063         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
2064         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2065         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
2066     }
2067 }
2068 
2069 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
2070 {
2071      struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2072      DWORD flags = 0;
2073 
2074      DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2075      char dst_str[50];
2076      char src_str[50];
2077 
2078      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
2079      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2080      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2081      /* Move .x first in case src_str is "TA" */
2082      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
2083      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
2084      if (reg1 < MAX_TEXTURES)
2085      {
2086          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2087          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
2088      }
2089      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2090 }
2091 
2092 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
2093 {
2094      struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2095 
2096      DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2097      char dst_str[50];
2098      char src_str[50];
2099 
2100      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2101      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2102      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2103      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2104      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2105      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2106 }
2107 
2108 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2109 {
2110     DWORD reg1 = ins->dst[0].reg.idx[0].offset;
2111     char dst_str[50];
2112     char src_str[50];
2113 
2114     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2115     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2116     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2117     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2118 }
2119 
2120 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2121 {
2122     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2123     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2124     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2125     char reg_coord[40], dst_reg[50], src_reg[50];
2126     DWORD reg_dest_code;
2127 
2128     /* All versions have a destination register. The Tx where the texture coordinates come
2129      * from is the varying incarnation of the texture register
2130      */
2131     reg_dest_code = dst->reg.idx[0].offset;
2132     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2133     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2134     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2135 
2136     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2137      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2138      *
2139      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2140      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2141      *
2142      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2143      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2144      * extension.
2145      */
2146     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2147     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2148     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2149     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2150 
2151     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2152      * so we can't let the GL handle this.
2153      */
2154     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2155             & WINED3D_PSARGS_PROJECTED)
2156     {
2157         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2158         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2159         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2160     } else {
2161         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2162     }
2163 
2164     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2165 
2166     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2167     {
2168         /* No src swizzles are allowed, so this is ok */
2169         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2170                        src_reg, reg_dest_code, reg_dest_code);
2171         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2172     }
2173 }
2174 
2175 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2176 {
2177     DWORD reg = ins->dst[0].reg.idx[0].offset;
2178     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2179     char src0_name[50], dst_name[50];
2180     BOOL is_color;
2181     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2182 
2183     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2184     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2185      * T<reg+1> register. Use this register to store the calculated vector
2186      */
2187     tmp_reg.idx[0].offset = reg + 1;
2188     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2189     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2190 }
2191 
2192 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2193 {
2194     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2195     DWORD flags;
2196     DWORD reg = ins->dst[0].reg.idx[0].offset;
2197     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2198     char dst_str[50];
2199     char src0_name[50];
2200     char dst_reg[50];
2201     BOOL is_color;
2202 
2203     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2204     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2205 
2206     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2207     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2208     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2209     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2210     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2211 }
2212 
2213 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2214 {
2215     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2216     DWORD reg = ins->dst[0].reg.idx[0].offset;
2217     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2218     char src0_name[50], dst_name[50];
2219     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2220     BOOL is_color;
2221 
2222     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2223      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2224      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2225      */
2226     tmp_reg.idx[0].offset = reg + 2 - tex_mx->current_row;
2227     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2228 
2229     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2230     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2231                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2232     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2233 }
2234 
2235 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2236 {
2237     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2238     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2239     DWORD flags;
2240     DWORD reg = ins->dst[0].reg.idx[0].offset;
2241     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2242     char dst_str[50];
2243     char src0_name[50], dst_name[50];
2244     BOOL is_color;
2245 
2246     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2247     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2248     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2249 
2250     /* Sample the texture using the calculated coordinates */
2251     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2252     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2253     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2254     tex_mx->current_row = 0;
2255 }
2256 
2257 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2258 {
2259     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2260     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2261     DWORD flags;
2262     DWORD reg = ins->dst[0].reg.idx[0].offset;
2263     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2264     char dst_str[50];
2265     char src0_name[50];
2266     char dst_reg[50];
2267     BOOL is_color;
2268 
2269     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2270      * components for temporary data storage
2271      */
2272     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2273     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2274     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2275 
2276     /* Construct the eye-ray vector from w coordinates */
2277     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2278     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2279     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2280 
2281     /* Calculate reflection vector
2282      */
2283     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2284     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2285     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2286     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2287     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2288     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2289     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2290 
2291     /* Sample the texture using the calculated coordinates */
2292     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2293     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2294     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2295     tex_mx->current_row = 0;
2296 }
2297 
2298 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2299 {
2300     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2301     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2302     DWORD flags;
2303     DWORD reg = ins->dst[0].reg.idx[0].offset;
2304     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2305     char dst_str[50];
2306     char src0_name[50];
2307     char src1_name[50];
2308     char dst_reg[50];
2309     BOOL is_color;
2310 
2311     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2312     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2313     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2314     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2315     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2316 
2317     /* Calculate reflection vector.
2318      *
2319      *                   dot(N, E)
2320      * dst_reg.xyz = 2 * --------- * N - E
2321      *                   dot(N, N)
2322      *
2323      * Which normalizes the normal vector
2324      */
2325     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2326     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2327     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2328     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2329     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2330     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2331 
2332     /* Sample the texture using the calculated coordinates */
2333     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2334     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2335     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2336     tex_mx->current_row = 0;
2337 }
2338 
2339 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2340 {
2341     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2342     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2343     char dst_name[50];
2344     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2345     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2346 
2347     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2348      * which is essentially an input, is the destination register because it is the first
2349      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2350      * here(writemasks/swizzles are not valid on texdepth)
2351      */
2352     shader_arb_get_dst_param(ins, dst, dst_name);
2353 
2354     /* According to the msdn, the source register(must be r5) is unusable after
2355      * the texdepth instruction, so we're free to modify it
2356      */
2357     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2358 
2359     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2360      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2361      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2362      */
2363     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2364     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2365     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2366     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2367 }
2368 
2369 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2370  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2371  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2372 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2373 {
2374     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2375     DWORD sampler_idx = ins->dst[0].reg.idx[0].offset;
2376     char src0[50];
2377     char dst_str[50];
2378 
2379     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2380     shader_addline(buffer, "MOV TB, 0.0;\n");
2381     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2382 
2383     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2384     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2385 }
2386 
2387 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2388  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2389 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2390 {
2391     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2392     char src0[50];
2393     char dst_str[50];
2394     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2395 
2396     /* Handle output register */
2397     shader_arb_get_dst_param(ins, dst, dst_str);
2398     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2399     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx[0].offset, src0);
2400 }
2401 
2402 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2403  * Perform the 3rd row of a 3x3 matrix multiply */
2404 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2405 {
2406     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2407     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2408     char dst_str[50], dst_name[50];
2409     char src0[50];
2410     BOOL is_color;
2411 
2412     shader_arb_get_dst_param(ins, dst, dst_str);
2413     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2414     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2415     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx[0].offset, src0);
2416     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2417 }
2418 
2419 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2420  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2421  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2422  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2423  */
2424 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2425 {
2426     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2427     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2428     char src0[50], dst_name[50];
2429     BOOL is_color;
2430     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2431     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2432 
2433     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2434     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2435     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx[0].offset, src0);
2436 
2437     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2438      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2439      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2440      */
2441     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2442     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2443     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2444     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2445 }
2446 
2447 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2448     Vertex/Pixel shaders to ARB_vertex_program codes */
2449 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2450 {
2451     int i;
2452     int nComponents = 0;
2453     struct wined3d_shader_dst_param tmp_dst = {{0}};
2454     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2455     struct wined3d_shader_instruction tmp_ins;
2456 
2457     memset(&tmp_ins, 0, sizeof(tmp_ins));
2458 
2459     /* Set constants for the temporary argument */
2460     tmp_ins.ctx = ins->ctx;
2461     tmp_ins.dst_count = 1;
2462     tmp_ins.dst = &tmp_dst;
2463     tmp_ins.src_count = 2;
2464     tmp_ins.src = tmp_src;
2465 
2466     switch(ins->handler_idx)
2467     {
2468         case WINED3DSIH_M4x4:
2469             nComponents = 4;
2470             tmp_ins.handler_idx = WINED3DSIH_DP4;
2471             break;
2472         case WINED3DSIH_M4x3:
2473             nComponents = 3;
2474             tmp_ins.handler_idx = WINED3DSIH_DP4;
2475             break;
2476         case WINED3DSIH_M3x4:
2477             nComponents = 4;
2478             tmp_ins.handler_idx = WINED3DSIH_DP3;
2479             break;
2480         case WINED3DSIH_M3x3:
2481             nComponents = 3;
2482             tmp_ins.handler_idx = WINED3DSIH_DP3;
2483             break;
2484         case WINED3DSIH_M3x2:
2485             nComponents = 2;
2486             tmp_ins.handler_idx = WINED3DSIH_DP3;
2487             break;
2488         default:
2489             FIXME("Unhandled opcode %s.\n", debug_d3dshaderinstructionhandler(ins->handler_idx));
2490             break;
2491     }
2492 
2493     tmp_dst = ins->dst[0];
2494     tmp_src[0] = ins->src[0];
2495     tmp_src[1] = ins->src[1];
2496     for (i = 0; i < nComponents; ++i)
2497     {
2498         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2499         shader_hw_map2gl(&tmp_ins);
2500         ++tmp_src[1].reg.idx[0].offset;
2501     }
2502 }
2503 
2504 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2505 {
2506     *need_abs = FALSE;
2507 
2508     switch(mod)
2509     {
2510         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2511         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2512         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2513         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2514         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2515         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2516         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2517         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2518         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2519         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2520         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2521         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2522         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2523     }
2524     FIXME("Unknown modifier %u\n", mod);
2525     return mod;
2526 }
2527 
2528 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2529 {
2530     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2531     const char *instruction;
2532     struct wined3d_shader_src_param src0_copy = ins->src[0];
2533     BOOL need_abs = FALSE;
2534 
2535     char dst[50];
2536     char src[50];
2537 
2538     switch(ins->handler_idx)
2539     {
2540         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2541         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2542         case WINED3DSIH_EXPP:
2543             if (ins->ctx->reg_maps->shader_version.major < 2)
2544             {
2545                 instruction = "EXP";
2546                 break;
2547             }
2548             /* Drop through. */
2549         case WINED3DSIH_EXP:
2550             instruction = "EX2";
2551             break;
2552         case WINED3DSIH_LOG:
2553         case WINED3DSIH_LOGP:
2554             /* The precision requirements suggest that LOGP matches ARBvp's LOG
2555              * instruction, but notice that the output of those instructions is
2556              * different. */
2557             src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2558             instruction = "LG2";
2559             break;
2560         default: instruction = "";
2561             FIXME("Unhandled opcode %s.\n", debug_d3dshaderinstructionhandler(ins->handler_idx));
2562             break;
2563     }
2564 
2565     /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2566      * .w is used. */
2567     src0_copy.swizzle = shader_arb_select_component(src0_copy.swizzle, 3);
2568 
2569     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2570     shader_arb_get_src_param(ins, &src0_copy, 0, src);
2571 
2572     if(need_abs)
2573     {
2574         shader_addline(buffer, "ABS TA.w, %s;\n", src);
2575         shader_addline(buffer, "%s%s %s, TA.w;\n", instruction, shader_arb_get_modifier(ins), dst);
2576     }
2577     else
2578     {
2579         shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2580     }
2581 
2582 }
2583 
2584 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2585 {
2586     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2587     char dst_name[50];
2588     char src_name[50];
2589     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2590     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2591     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2592 
2593     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2594     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2595 
2596     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2597      * otherwise NRM or RSQ would return NaN */
2598     if(pshader && priv->target_version >= NV3)
2599     {
2600         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2601          *
2602          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2603          */
2604         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2605         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2606         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2607     }
2608     else if(priv->target_version >= NV2)
2609     {
2610         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2611         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2612         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2613                        src_name);
2614     }
2615     else
2616     {
2617         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2618 
2619         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2620         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2621          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2622          */
2623         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2624         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2625 
2626         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2627         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2628         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2629                     src_name);
2630     }
2631 }
2632 
2633 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2634 {
2635     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2636     char dst_name[50];
2637     char src_name[3][50];
2638 
2639     /* ARB_fragment_program has a convenient LRP instruction */
2640     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2641         shader_hw_map2gl(ins);
2642         return;
2643     }
2644 
2645     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2646     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2647     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2648     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2649 
2650     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2651     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2652                    dst_name, src_name[0], src_name[2]);
2653 }
2654 
2655 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2656 {
2657     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2658      * must contain fixed constants. So we need a separate function to filter those constants and
2659      * can't use map2gl
2660      */
2661     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2662     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2663     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2664     char dst_name[50];
2665     char src_name0[50], src_name1[50], src_name2[50];
2666     BOOL is_color;
2667 
2668     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2669     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2670         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2671         /* No modifiers are supported on SCS */
2672         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2673 
2674         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2675         {
2676             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2677             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2678         }
2679     } else if(priv->target_version >= NV2) {
2680         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2681 
2682         /* Sincos writemask must be .x, .y or .xy */
2683         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2684             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2685         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2686             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2687     } else {
2688         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2689          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2690          *
2691          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2692          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2693          *
2694          * The constants we get are:
2695          *
2696          *  +1   +1,     -1     -1     +1      +1      -1       -1
2697          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2698          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2699          *
2700          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2701          *
2702          * (x/2)^2 = x^2 / 4
2703          * (x/2)^3 = x^3 / 8
2704          * (x/2)^4 = x^4 / 16
2705          * (x/2)^5 = x^5 / 32
2706          * etc
2707          *
2708          * To get the final result:
2709          * sin(x) = 2 * sin(x/2) * cos(x/2)
2710          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2711          * (from sin(x+y) and cos(x+y) rules)
2712          *
2713          * As per MSDN, dst.z is undefined after the operation, and so is
2714          * dst.x and dst.y if they're masked out by the writemask. Ie
2715          * sincos dst.y, src1, c0, c1
2716          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2717          * vsa.exe also stops with an error if the dest register is the same register as the source
2718          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2719          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2720          */
2721         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2722         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2723         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2724 
2725         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2726         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2727         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2728         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2729         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2730         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2731 
2732         /* sin(x/2)
2733          *
2734          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2735          * properly merge that with MULs in the code above?
2736          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2737          * we can merge the sine and cosine MAD rows to calculate them together.
2738          */
2739         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2740         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2741         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2742         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2743 
2744         /* cos(x/2) */
2745         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2746         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2747         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2748 
2749         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2750             /* cos x */
2751             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2752             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2753         }
2754         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2755             /* sin x */
2756             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2757             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2758         }
2759     }
2760 }
2761 
2762 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2763 {
2764     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2765     char dst_name[50];
2766     char src_name[50];
2767     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2768 
2769     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2770     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2771 
2772     /* SGN is only valid in vertex shaders */
2773     if(ctx->target_version >= NV2) {
2774         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2775         return;
2776     }
2777 
2778     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2779      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2780      */
2781     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2782         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2783     } else {
2784         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2785          * Then use TA, and calculate the final result
2786          *
2787          * Not reading from TA? Store the first result in TA to avoid overwriting the
2788          * destination if src reg = dst reg
2789          */
2790         if(strstr(src_name, "TA"))
2791         {
2792             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2793             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2794             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2795         }
2796         else
2797         {
2798             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2799             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2800             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2801         }
2802     }
2803 }
2804 
2805 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2806 {
2807     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2808     char src[50];
2809     char dst[50];
2810     char dst_name[50];
2811     BOOL is_color;
2812 
2813     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2814     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2815     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2816 
2817     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2818     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2819 }
2820 
2821 static void shader_hw_pow(const struct wined3d_shader_instruction *ins)
2822 {
2823     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2824     char src0[50], src1[50], dst[50];
2825     struct wined3d_shader_src_param src0_copy = ins->src[0];
2826     BOOL need_abs = FALSE;
2827     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2828     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2829 
2830     /* POW operates on the absolute value of the input */
2831     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2832 
2833     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2834     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2835     shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2836 
2837     if (need_abs)
2838         shader_addline(buffer, "ABS TA.x, %s;\n", src0);
2839     else
2840         shader_addline(buffer, "MOV TA.x, %s;\n", src0);
2841 
2842     if (priv->target_version >= NV2)
2843     {
2844         shader_addline(buffer, "MOVC TA.y, %s;\n", src1);
2845         shader_addline(buffer, "POW%s %s, TA.x, TA.y;\n", shader_arb_get_modifier(ins), dst);
2846         shader_addline(buffer, "MOV %s (EQ.y), %s;\n", dst, one);
2847     }
2848     else
2849     {
2850         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2851         const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2852 
2853         shader_addline(buffer, "ABS TA.y, %s;\n", src1);
2854         shader_addline(buffer, "SGE TA.y, -TA.y, %s;\n", zero);
2855         /* Possibly add flt_eps to avoid getting float special values */
2856         shader_addline(buffer, "MAD TA.z, TA.y, %s, %s;\n", flt_eps, src1);
2857         shader_addline(buffer, "POW%s TA.x, TA.x, TA.z;\n", shader_arb_get_modifier(ins));
2858         shader_addline(buffer, "MAD TA.x, -TA.x, TA.y, TA.x;\n");
2859         shader_addline(buffer, "MAD %s, TA.y, %s, TA.x;\n", dst, one);
2860     }
2861 }
2862 
2863 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2864 {
2865     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2866     char src_name[50];
2867     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2868 
2869     /* src0 is aL */
2870     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2871 
2872     if(vshader)
2873     {
2874         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2875         struct list *e = list_head(&priv->control_frames);
2876         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2877 
2878         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2879         /* The constant loader makes sure to load -1 into iX.w */
2880         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2881         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2882         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2883     }
2884     else
2885     {
2886         shader_addline(buffer, "LOOP %s;\n", src_name);
2887     }
2888 }
2889 
2890 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2891 {
2892     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2893     char src_name[50];
2894     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2895 
2896     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2897 
2898     /* The constant loader makes sure to load -1 into iX.w */
2899     if(vshader)
2900     {
2901         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2902         struct list *e = list_head(&priv->control_frames);
2903         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2904 
2905         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2906 
2907         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2908         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2909         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2910     }
2911     else
2912     {
2913         shader_addline(buffer, "REP %s;\n", src_name);
2914     }
2915 }
2916 
2917 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2918 {
2919     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2920     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2921 
2922     if(vshader)
2923     {
2924         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2925         struct list *e = list_head(&priv->control_frames);
2926         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2927 
2928         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2929         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2930         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2931 
2932         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2933     }
2934     else
2935     {
2936         shader_addline(buffer, "ENDLOOP;\n");
2937     }
2938 }
2939 
2940 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2941 {
2942     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2943     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2944 
2945     if(vshader)
2946     {
2947         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2948         struct list *e = list_head(&priv->control_frames);
2949         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2950 
2951         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2952         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2953         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2954 
2955         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2956     }
2957     else
2958     {
2959         shader_addline(buffer, "ENDREP;\n");
2960     }
2961 }
2962 
2963 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2964 {
2965     struct control_frame *control_frame;
2966 
2967     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2968     {
2969         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2970     }
2971     ERR("Could not find loop for break\n");
2972     return NULL;
2973 }
2974 
2975 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2976 {
2977     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
2978     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2979     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2980 
2981     if(vshader)
2982     {
2983         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2984     }
2985     else
2986     {
2987         shader_addline(buffer, "BRK;\n");
2988     }
2989 }
2990 
2991 static const char *get_compare(enum wined3d_shader_rel_op op)
2992 {
2993     switch (op)
2994     {
2995         case WINED3D_SHADER_REL_OP_GT: return "GT";
2996         case WINED3D_SHADER_REL_OP_EQ: return "EQ";
2997         case WINED3D_SHADER_REL_OP_GE: return "GE";
2998         case WINED3D_SHADER_REL_OP_LT: return "LT";
2999         case WINED3D_SHADER_REL_OP_NE: return "NE";
3000         case WINED3D_SHADER_REL_OP_LE: return "LE";
3001         default:
3002             FIXME("Unrecognized operator %#x.\n", op);
3003             return "(\?\?)";
3004     }
3005 }
3006 
3007 static enum wined3d_shader_rel_op invert_compare(enum wined3d_shader_rel_op op)
3008 {
3009     switch (op)
3010     {
3011         case WINED3D_SHADER_REL_OP_GT: return WINED3D_SHADER_REL_OP_LE;
3012         case WINED3D_SHADER_REL_OP_EQ: return WINED3D_SHADER_REL_OP_NE;
3013         case WINED3D_SHADER_REL_OP_GE: return WINED3D_SHADER_REL_OP_LT;
3014         case WINED3D_SHADER_REL_OP_LT: return WINED3D_SHADER_REL_OP_GE;
3015         case WINED3D_SHADER_REL_OP_NE: return WINED3D_SHADER_REL_OP_EQ;
3016         case WINED3D_SHADER_REL_OP_LE: return WINED3D_SHADER_REL_OP_GT;
3017         default:
3018             FIXME("Unrecognized operator %#x.\n", op);
3019             return -1;
3020     }
3021 }
3022 
3023 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
3024 {
3025     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
3026     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3027     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
3028     char src_name0[50];
3029     char src_name1[50];
3030     const char *comp = get_compare(ins->flags);
3031 
3032     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3033     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3034 
3035     if(vshader)
3036     {
3037         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
3038          * away the subtraction result
3039          */
3040         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3041         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
3042     }
3043     else
3044     {
3045         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3046         shader_addline(buffer, "BRK (%s.x);\n", comp);
3047     }
3048 }
3049 
3050 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
3051 {
3052     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
3053     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3054     struct list *e = list_head(&priv->control_frames);
3055     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3056     const char *comp;
3057     char src_name0[50];
3058     char src_name1[50];
3059     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3060 
3061     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
3062     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
3063 
3064     if(vshader)
3065     {
3066         /* Invert the flag. We jump to the else label if the condition is NOT true */
3067         comp = get_compare(invert_compare(ins->flags));
3068         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3069         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
3070     }
3071     else
3072     {
3073         comp = get_compare(ins->flags);
3074         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3075         shader_addline(buffer, "IF %s.x;\n", comp);
3076     }
3077 }
3078 
3079 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3080 {
3081     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
3082     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3083     struct list *e = list_head(&priv->control_frames);
3084     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3085     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3086 
3087     if(vshader)
3088     {
3089         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3090         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3091         control_frame->had_else = TRUE;
3092     }
3093     else
3094     {
3095         shader_addline(buffer, "ELSE;\n");
3096     }
3097 }
3098 
3099 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3100 {
3101     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
3102     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3103     struct list *e = list_head(&priv->control_frames);
3104     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3105     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3106 
3107     if(vshader)
3108     {
3109         if(control_frame->had_else)
3110         {
3111             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3112         }
3113         else
3114         {
3115             shader_addline(buffer, "#No else branch. else is endif\n");
3116             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3117         }
3118     }
3119     else
3120     {
3121         shader_addline(buffer, "ENDIF;\n");
3122     }
3123 }
3124 
3125 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3126 {
3127     DWORD sampler_idx = ins->src[1].reg.idx[0].offset;
3128     char reg_dest[40];
3129     char reg_src[3][40];
3130     WORD flags = TEX_DERIV;
3131 
3132     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3133     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3134     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3135     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3136 
3137     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3138     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3139 
3140     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3141 }
3142 
3143 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3144 {
3145     DWORD sampler_idx = ins->src[1].reg.idx[0].offset;
3146     char reg_dest[40];
3147     char reg_coord[40];
3148     WORD flags = TEX_LOD;
3149 
3150     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3151     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3152 
3153     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3154     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3155 
3156     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3157 }
3158 
3159 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3160 {
3161     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
3162     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3163 
3164     priv->in_main_func = FALSE;
3165     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3166      * subroutine, don't generate a label that will make GL complain
3167      */
3168     if(priv->target_version == ARB) return;
3169 
3170     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx[0].offset);
3171 }
3172 
3173 static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
3174         const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
3175         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
3176         struct wined3d_string_buffer *buffer)
3177 {
3178     unsigned int i;
3179 
3180     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3181      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3182      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3183      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3184      */
3185     if (args->super.fog_src == VS_FOG_Z)
3186     {
3187         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3188     }
3189     else
3190     {
3191         if (!reg_maps->fog)
3192         {
3193             /* posFixup.x is always 1.0, so we can safely use it */
3194             shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3195         }
3196         else
3197         {
3198             /* Clamp fogcoord */
3199             const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
3200             const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
3201 
3202             shader_addline(buffer, "MIN TMP_FOGCOORD.x, TMP_FOGCOORD.x, %s;\n", one);
3203             shader_addline(buffer, "MAX result.fogcoord.x, TMP_FOGCOORD.x, %s;\n", zero);
3204         }
3205     }
3206 
3207     /* Clipplanes are always stored without y inversion */
3208     if (use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3209     {
3210         if (args->super.clip_enabled)
3211         {
3212             for (i = 0; i < priv_ctx->vs_clipplanes; i++)
3213             {
3214                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3215             }
3216         }
3217     }
3218     else if (args->clip.boolclip.clip_texcoord)
3219     {
3220         static const char component[4] = {'x', 'y', 'z', 'w'};
3221         unsigned int cur_clip = 0;
3222         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3223 
3224         for (i = 0; i < gl_info->limits.user_clip_distances; ++i)
3225         {
3226             if (args->clip.boolclip.clipplane_mask & (1u << i))
3227             {
3228                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3229                                component[cur_clip++], i);
3230             }
3231         }
3232         switch (cur_clip)
3233         {
3234             case 0:
3235                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3236                 break;
3237             case 1:
3238                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3239                 break;
3240             case 2:
3241                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3242                 break;
3243             case 3:
3244                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3245                 break;
3246         }
3247         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3248                        args->clip.boolclip.clip_texcoord - 1);
3249     }
3250 
3251     /* Write the final position.
3252      *
3253      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3254      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3255      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3256      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3257      */
3258     if (!gl_info->supported[ARB_CLIP_CONTROL])
3259     {
3260         shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3261         shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3262         shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3263 
3264         /* Z coord [0;1]->[-1;1] mapping, see comment in
3265          * get_projection_matrix() in utils.c. */
3266         if (need_helper_const(shader_data, reg_maps, gl_info))
3267         {
3268             const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3269             shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3270         }
3271         else
3272         {
3273             shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3274             shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3275         }
3276     }
3277 
3278     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3279 
3280     priv_ctx->footer_written = TRUE;
3281 }
3282 
3283 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3284 {
3285     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
3286     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3287     const struct wined3d_shader *shader = ins->ctx->shader;
3288     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3289 
3290     if(priv->target_version == ARB) return;
3291 
3292     if(vshader)
3293     {
3294         if (priv->in_main_func) vshader_add_footer(priv, shader->backend_data,
3295                 priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
3296     }
3297 
3298     shader_addline(buffer, "RET;\n");
3299 }
3300 
3301 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3302 {
3303     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
3304     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx[0].offset);
3305 }
3306 
3307 static BOOL shader_arb_compile(const struct wined3d_gl_info *gl_info, GLenum target, const char *src)
3308 {
3309     const char *ptr, *line;
3310     GLint native, pos;
3311 
3312     if (TRACE_ON(d3d_shader))
3313     {
3314         ptr = src;
3315         while ((line = get_line(&ptr))) TRACE_(d3d_shader)("    %.*s", (int)(ptr - line), line);
3316     }
3317 
3318     GL_EXTCALL(glProgramStringARB(target, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(src), src));
3319     checkGLcall("glProgramStringARB()");
3320 
3321     if (FIXME_ON(d3d_shader))
3322     {
3323         gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3324         if (pos != -1)
3325         {
3326             FIXME_(d3d_shader)("Program error at position %d: %s\n\n", pos,
3327                     debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3328             ptr = src;
3329             while ((line = get_line(&ptr))) FIXME_(d3d_shader)("    %.*s", (int)(ptr - line), line);
3330             FIXME_(d3d_shader)("\n");
3331 
3332             return FALSE;
3333         }
3334     }
3335 
3336     if (WARN_ON(d3d_perf))
3337     {
3338         GL_EXTCALL(glGetProgramivARB(target, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3339         checkGLcall("glGetProgramivARB()");
3340         if (!native)
3341             WARN_(d3d_perf)("Program exceeds native resource limits.\n");
3342     }
3343 
3344     return TRUE;
3345 }
3346 
3347 static void arbfp_add_sRGB_correction(struct wined3d_string_buffer *buffer, const char *fragcolor,
3348         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3349 {
3350     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3351 
3352     if(condcode)
3353     {
3354         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3355         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.x;\n", tmp1, fragcolor);
3356         /* Calculate the > 0.0031308 case */
3357         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts0.x;\n", fragcolor, fragcolor);
3358         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts0.x;\n", fragcolor, fragcolor);
3359         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts0.x;\n", fragcolor, fragcolor);
3360         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts0.y;\n", fragcolor, fragcolor);
3361         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts0.z;\n", fragcolor, fragcolor);
3362         /* Calculate the < case */
3363         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts0.w, %s;\n", fragcolor, fragcolor);
3364     }
3365     else
3366     {
3367         /* Calculate the > 0.0031308 case */
3368         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts0.x;\n", tmp1, fragcolor);
3369         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts0.x;\n", tmp1, fragcolor);
3370         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts0.x;\n", tmp1, fragcolor);
3371         shader_addline(buffer, "MUL %s, %s, srgb_consts0.y;\n", tmp1, tmp1);
3372         shader_addline(buffer, "SUB %s, %s, srgb_consts0.z;\n", tmp1, tmp1);
3373         /* Calculate the < case */
3374         shader_addline(buffer, "MUL %s, srgb_consts0.w, %s;\n", tmp2, fragcolor);
3375         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3376         shader_addline(buffer, "SLT %s, srgb_consts1.x, %s;\n", tmp3, fragcolor);
3377         shader_addline(buffer, "SGE %s, srgb_consts1.x, %s;\n", tmp4, fragcolor);
3378         /* Store the components > 0.0031308 in the destination */
3379         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3380         /* Add the components that are < 0.0031308 */
3381         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3382         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3383         * result.color writes(.rgb first, then .a), or handle overwriting already written
3384         * components. The assembler uses a temporary register in this case, which is usually
3385         * not allocated from one of our registers that were used earlier.
3386         */
3387     }
3388     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3389 }
3390 
3391 static const DWORD *find_loop_control_values(const struct wined3d_shader *shader, DWORD idx)
3392 {
3393     const struct wined3d_shader_lconst *constant;
3394 
3395     LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
3396     {
3397         if (constant->idx == idx)
3398         {
3399             return constant->value;
3400         }
3401     }
3402     return NULL;
3403 }
3404 
3405 static void init_ps_input(const struct wined3d_shader *shader,
3406         const struct arb_ps_compile_args *args, struct shader_arb_ctx_priv *priv)
3407 {
3408     static const char * const texcoords[8] =
3409     {
3410         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3411         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3412     };
3413     unsigned int i;
3414     const struct wined3d_shader_signature_element *input;
3415     const char *semantic_name;
3416     DWORD semantic_idx;
3417 
3418     if (args->super.vp_mode == WINED3D_VP_MODE_SHADER)
3419     {
3420         /* That one is easy. The vertex shaders provide v0-v7 in
3421          * fragment.texcoord and v8 and v9 in fragment.color. */
3422         for (i = 0; i < 8; ++i)
3423         {
3424             priv->ps_input[i] = texcoords[i];
3425         }
3426         priv->ps_input[8] = "fragment.color.primary";
3427         priv->ps_input[9] = "fragment.color.secondary";
3428         return;
3429     }
3430 
3431     /* The fragment shader has to collect the varyings on its own. In any case
3432      * properly load color0 and color1. In the case of pre-transformed
3433      * vertices also load texture coordinates. Set other attributes to 0.0.
3434      *
3435      * For fixed-function this behavior is correct, according to the tests.
3436      * For pre-transformed we'd either need a replacement shader that can load
3437      * other attributes like BINORMAL, or load the texture coordinate
3438      * attribute pointers to match the fragment shader signature. */
3439     for (i = 0; i < shader->input_signature.element_count; ++i)
3440     {
3441         input = &shader->input_signature.elements[i];
3442         if (!(semantic_name = input->semantic_name))
3443             continue;
3444         semantic_idx = input->semantic_idx;
3445 
3446         if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
3447         {
3448             if (!semantic_idx)
3449                 priv->ps_input[input->register_idx] = "fragment.color.primary";
3450             else if (semantic_idx == 1)
3451                 priv->ps_input[input->register_idx] = "fragment.color.secondary";
3452             else
3453                 priv->ps_input[input->register_idx] = "0.0";
3454         }
3455         else if (args->super.vp_mode == WINED3D_VP_MODE_FF)
3456         {
3457             priv->ps_input[input->register_idx] = "0.0";
3458         }
3459         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3460         {
3461             if (semantic_idx < 8)
3462                 priv->ps_input[input->register_idx] = texcoords[semantic_idx];
3463             else
3464                 priv->ps_input[input->register_idx] = "0.0";
3465         }
3466         else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
3467         {
3468             if (!semantic_idx)
3469                 priv->ps_input[input->register_idx] = "fragment.fogcoord";
3470             else
3471                 priv->ps_input[input->register_idx] = "0.0";
3472         }
3473         else
3474         {
3475             priv->ps_input[input->register_idx] = "0.0";
3476         }
3477 
3478         TRACE("v%u, semantic %s%u is %s\n", input->register_idx,
3479                 semantic_name, semantic_idx, priv->ps_input[input->register_idx]);
3480     }
3481 }
3482 
3483 static void arbfp_add_linear_fog(struct wined3d_string_buffer *buffer,
3484         const char *fragcolor, const char *tmp)
3485 {
3486     shader_addline(buffer, "SUB %s.x, state.fog.params.z, fragment.fogcoord.x;\n", tmp);
3487     shader_addline(buffer, "MUL_SAT %s.x, %s.x, state.fog.params.w;\n", tmp, tmp);
3488     shader_addline(buffer, "LRP %s.rgb, %s.x, %s, state.fog.color;\n", fragcolor, tmp, fragcolor);
3489 }
3490 
3491 /* Context activation is done by the caller. */
3492 static GLuint shader_arb_generate_pshader(const struct wined3d_shader *shader,
3493         const struct wined3d_gl_info *gl_info, struct wined3d_string_buffer *buffer,
3494         const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
3495 {
3496     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
3497     GLuint retval;
3498     char fragcolor[16];
3499     DWORD next_local = 0;
3500     struct shader_arb_ctx_priv priv_ctx;
3501     BOOL dcl_td = FALSE;
3502     BOOL want_nv_prog = FALSE;
3503     struct arb_pshader_private *shader_priv = shader->backend_data;
3504     DWORD map;
3505     BOOL custom_linear_fog = FALSE;
3506 
3507     char srgbtmp[4][4];
3508     char ftoa_tmp[17];
3509     unsigned int i, found = 0;
3510 
3511     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3512     {
3513         if (!(map & 1)
3514                 || (shader->u.ps.color0_mov && i == shader->u.ps.color0_reg)
3515                 || (reg_maps->shader_version.major < 2 && !i))
3516             continue;
3517 
3518         sprintf(srgbtmp[found], "R%u", i);
3519         ++found;
3520         if (found == 4) break;
3521     }
3522 
3523     switch(found) {
3524         case 0:
3525             sprintf(srgbtmp[0], "TA");
3526             sprintf(srgbtmp[1], "TB");
3527             sprintf(srgbtmp[2], "TC");
3528             sprintf(srgbtmp[3], "TD");
3529             dcl_td = TRUE;
3530             break;
3531         case 1:
3532             sprintf(srgbtmp[1], "TA");
3533             sprintf(srgbtmp[2], "TB");
3534             sprintf(srgbtmp[3], "TC");
3535             break;
3536         case 2:
3537             sprintf(srgbtmp[2], "TA");
3538             sprintf(srgbtmp[3], "TB");
3539             break;
3540         case 3:
3541             sprintf(srgbtmp[3], "TA");
3542             break;
3543         case 4:
3544             break;
3545     }
3546 
3547     /*  Create the hw ARB shader */
3548     memset(&priv_ctx, 0, sizeof(priv_ctx));
3549     priv_ctx.cur_ps_args = args;
3550     priv_ctx.compiled_fprog = compiled;
3551     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3552     init_ps_input(shader, args, &priv_ctx);
3553     list_init(&priv_ctx.control_frames);
3554     priv_ctx.ps_post_process = args->super.srgb_correction;
3555 
3556     /* Avoid enabling NV_fragment_program* if we do not need it.
3557      *
3558      * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
3559      * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
3560      * is faster than what we gain from using higher native instructions. There are some things though
3561      * that cannot be emulated. In that case enable the extensions.
3562      * If the extension is enabled, instruction handlers that support both ways will use it.
3563      *
3564      * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
3565      * So enable the best we can get.
3566      */
3567     if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
3568        reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
3569     {
3570         want_nv_prog = TRUE;
3571     }
3572 
3573     shader_addline(buffer, "!!ARBfp1.0\n");
3574     if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
3575     {
3576         shader_addline(buffer, "OPTION NV_fragment_program2;\n");
3577         priv_ctx.target_version = NV3;
3578     }
3579     else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
3580     {
3581         shader_addline(buffer, "OPTION NV_fragment_program;\n");
3582         priv_ctx.target_version = NV2;
3583     } else {
3584         if(want_nv_prog)
3585         {
3586             /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
3587              * limits properly
3588              */
3589             ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
3590             ERR("Try GLSL\n");
3591         }
3592         priv_ctx.target_version = ARB;
3593     }
3594 
3595     if (reg_maps->rt_mask > 1)
3596     {
3597         shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
3598     }
3599 
3600     if (reg_maps->shader_version.major < 3)
3601     {
3602         switch (args->super.fog)
3603         {
3604             case WINED3D_FFP_PS_FOG_OFF:
3605                 break;
3606             case WINED3D_FFP_PS_FOG_LINEAR:
3607                 if (gl_info->quirks & WINED3D_QUIRK_BROKEN_ARB_FOG)
3608                 {
3609                     custom_linear_fog = TRUE;
3610                     priv_ctx.ps_post_process = TRUE;
3611                     break;
3612                 }
3613                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3614                 break;
3615             case WINED3D_FFP_PS_FOG_EXP:
3616                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3617                 break;
3618             case WINED3D_FFP_PS_FOG_EXP2:
3619                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3620                 break;
3621         }
3622     }
3623 
3624     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3625      * unused temps away(but occupies them for the whole shader if they're used once). Always
3626      * declaring them avoids tricky bookkeeping work
3627      */
3628     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3629     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3630     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3631     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3632     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3633     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3634     wined3d_ftoa(eps, ftoa_tmp);
3635     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %s, 0.0 };\n", ftoa_tmp);
3636 
3637     if (reg_maps->shader_version.major < 2)
3638     {
3639         strcpy(fragcolor, "R0");
3640     }
3641     else
3642     {
3643         if (priv_ctx.ps_post_process)
3644         {
3645             if (shader->u.ps.color0_mov)
3646             {
3647                 sprintf(fragcolor, "R%u", shader->u.ps.color0_reg);
3648             }
3649             else
3650             {
3651                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3652                 strcpy(fragcolor, "TMP_COLOR");
3653             }
3654         } else {
3655             strcpy(fragcolor, "result.color");
3656         }
3657     }
3658 
3659     if (args->super.srgb_correction)
3660     {
3661         shader_addline(buffer, "PARAM srgb_consts0 = ");
3662         shader_arb_append_imm_vec4(buffer, wined3d_srgb_const0);
3663         shader_addline(buffer, ";\n");
3664         shader_addline(buffer, "PARAM srgb_consts1 = ");
3665         shader_arb_append_imm_vec4(buffer, wined3d_srgb_const1);
3666         shader_addline(buffer, ";\n");
3667     }
3668 
3669     /* Base Declarations */
3670     shader_generate_arb_declarations(shader, reg_maps, buffer, gl_info, NULL, &priv_ctx);
3671 
3672     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3673     {
3674         unsigned char bump_const;
3675 
3676         if (!(map & 1)) continue;
3677 
3678         bump_const = compiled->numbumpenvmatconsts;
3679         compiled->bumpenvmatconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3680         compiled->bumpenvmatconst[bump_const].texunit = i;
3681         compiled->luminanceconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
3682         compiled->luminanceconst[bump_const].texunit = i;
3683 
3684         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3685          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3686          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3687          * textures due to conditional NP2 restrictions)
3688          *
3689          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3690          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3691          * their location is shader dependent anyway and they cannot be loaded globally.
3692          */
3693         compiled->bumpenvmatconst[bump_const].const_num = next_local++;
3694         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3695                        i, compiled->bumpenvmatconst[bump_const].const_num);
3696         compiled->numbumpenvmatconsts = bump_const + 1;
3697 
3698         if (!(reg_maps->luminanceparams & (1u << i)))
3699             continue;
3700 
3701         compiled->luminanceconst[bump_const].const_num = next_local++;
3702         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3703                        i, compiled->luminanceconst[bump_const].const_num);
3704     }
3705 
3706     for (i = 0; i < WINED3D_MAX_CONSTS_I; ++i)
3707     {
3708         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3709         if (reg_maps->integer_constants & (1u << i) && priv_ctx.target_version >= NV2)
3710         {
3711             const DWORD *control_values = find_loop_control_values(shader, i);
3712 
3713             if(control_values)
3714             {
3715                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3716                                 control_values[0], control_values[1], control_values[2]);
3717             }
3718             else
3719             {
3720                 compiled->int_consts[i] = next_local;
3721                 compiled->num_int_consts++;
3722                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3723             }
3724         }
3725     }
3726 
3727     if(reg_maps->vpos || reg_maps->usesdsy)
3728     {
3729         compiled->ycorrection = next_local;
3730         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3731 
3732         if(reg_maps->vpos)
3733         {
3734             shader_addline(buffer, "TEMP vpos;\n");
3735             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3736              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3737              * ycorrection.z: 1.0
3738              * ycorrection.w: 0.0
3739              */
3740             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3741             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3742         }
3743     }
3744     else
3745     {
3746         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3747     }
3748 
3749     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3750      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3751      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3752      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3753      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3754      * shader compilation errors and the subsequent errors when drawing with this shader. */
3755     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3756         unsigned char cur_fixup_sampler = 0;
3757 
3758         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3759         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3760         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3761 
3762         fixup->offset = next_local;
3763         fixup->super.active = 0;
3764 
3765         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i)
3766         {
3767             if (!(map & (1u << i)))
3768                 continue;
3769 
3770             if (fixup->offset + (cur_fixup_sampler >> 1) < max_lconsts)
3771             {
3772                 fixup->super.active |= (1u << i);
3773                 fixup->super.idx[i] = cur_fixup_sampler++;
3774             }
3775             else
3776             {
3777                 FIXME("No free constant found to load NP2 fixup data into shader. "
3778                       "Sampling from this texture will probably look wrong.\n");
3779                 break;
3780             }
3781         }
3782 
3783         fixup->super.num_consts = (cur_fixup_sampler + 1) >> 1;
3784         if (fixup->super.num_consts) {
3785             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3786                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3787         }
3788     }
3789 
3790     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3791     {
3792         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3793     }
3794 
3795     /* Base Shader Body */
3796     if (FAILED(shader_generate_code(shader, buffer, reg_maps, &priv_ctx, NULL, NULL)))
3797         return 0;
3798 
3799     if(args->super.srgb_correction) {
3800         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3801                                   priv_ctx.target_version >= NV2);
3802     }
3803 
3804     if (custom_linear_fog)
3805         arbfp_add_linear_fog(buffer, fragcolor, "TA");
3806 
3807     if(strcmp(fragcolor, "result.color")) {
3808         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3809     }
3810     shader_addline(buffer, "END\n");
3811 
3812     /* TODO: change to resource.glObjectHandle or something like that */
3813     GL_EXTCALL(glGenProgramsARB(1, &retval));
3814 
3815     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3816     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3817 
3818     TRACE("Created hw pixel shader, prg=%d\n", retval);
3819     if (!shader_arb_compile(gl_info, GL_FRAGMENT_PROGRAM_ARB, buffer->buffer))
3820         return 0;
3821 
3822     return retval;
3823 }
3824 
3825 static int compare_sig(const struct wined3d_shader_signature *sig1, const struct wined3d_shader_signature *sig2)
3826 {
3827     unsigned int i;
3828     int ret;
3829 
3830     if (sig1->element_count != sig2->element_count)
3831         return sig1->element_count < sig2->element_count ? -1 : 1;
3832 
3833     for (i = 0; i < sig1->element_count; ++i)
3834     {
3835         const struct wined3d_shader_signature_element *e1, *e2;
3836 
3837         e1 = &sig1->elements[i];
3838         e2 = &sig2->elements[i];
3839 
3840         if (!e1->semantic_name || !e2->semantic_name)
3841         {
3842             /* Compare pointers, not contents. One string is NULL (element
3843              * does not exist), the other one is not NULL. */
3844             if (e1->semantic_name != e2->semantic_name)
3845                 return e1->semantic_name < e2->semantic_name ? -1 : 1;
3846             continue;
3847         }
3848 
3849         if ((ret = strcmp(e1->semantic_name, e2->semantic_name)))
3850             return ret;
3851         if (e1->semantic_idx != e2->semantic_idx)
3852             return e1->semantic_idx < e2->semantic_idx ? -1 : 1;
3853         if (e1->sysval_semantic != e2->sysval_semantic)
3854             return e1->sysval_semantic < e2->sysval_semantic ? -1 : 1;
3855         if (e1->component_type != e2->component_type)
3856             return e1->component_type < e2->component_type ? -1 : 1;
3857         if (e1->register_idx != e2->register_idx)
3858             return e1->register_idx < e2->register_idx ? -1 : 1;
3859         if (e1->mask != e2->mask)
3860             return e1->mask < e2->mask ? -1 : 1;
3861     }
3862     return 0;
3863 }
3864 
3865 static void clone_sig(struct wined3d_shader_signature *new, const struct wined3d_shader_signature *sig)
3866 {
3867     unsigned int i;
3868     char *name;
3869 
3870     new->element_count = sig->element_count;
3871     new->elements = heap_calloc(new->element_count, sizeof(*new->elements));
3872     for (i = 0; i < sig->element_count; ++i)
3873     {
3874         new->elements[i] = sig->elements[i];
3875 
3876         if (!new->elements[i].semantic_name)
3877             continue;
3878 
3879         /* Clone the semantic string */
3880         name = heap_alloc(strlen(sig->elements[i].semantic_name) + 1);
3881         strcpy(name, sig->elements[i].semantic_name);
3882         new->elements[i].semantic_name = name;
3883     }
3884 }
3885 
3886 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature *sig)
3887 {
3888     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3889     struct ps_signature *found_sig;
3890 
3891     if (entry)
3892     {
3893         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3894         TRACE("Found existing signature %u\n", found_sig->idx);
3895         return found_sig->idx;
3896     }
3897     found_sig = heap_alloc_zero(sizeof(*found_sig));
3898     clone_sig(&found_sig->sig, sig);
3899     found_sig->idx = priv->ps_sig_number++;
3900     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3901     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3902     {
3903         ERR("Failed to insert program entry.\n");
3904     }
3905     return found_sig->idx;
3906 }
3907 
3908 static void init_output_registers(const struct wined3d_shader *shader,
3909         const struct wined3d_shader_signature *ps_input_sig,
3910         struct shader_arb_ctx_priv *priv_ctx, struct arb_vs_compiled_shader *compiled)
3911 {
3912     unsigned int i, j;
3913     static const char * const texcoords[8] =
3914     {
3915         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3916         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3917     };
3918     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3919      * and varying 9 to result.color.secondary
3920      */
3921     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
3922     {
3923         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3924         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
3925         "result.color.primary", "result.color.secondary"
3926     };
3927 
3928     if (!ps_input_sig)
3929     {
3930         TRACE("Pixel shader uses builtin varyings\n");
3931         /* Map builtins to builtins */
3932         for(i = 0; i < 8; i++)
3933         {
3934             priv_ctx->texcrd_output[i] = texcoords[i];
3935         }
3936         priv_ctx->color_output[0] = "result.color.primary";
3937         priv_ctx->color_output[1] = "result.color.secondary";
3938         priv_ctx->fog_output = "TMP_FOGCOORD";
3939 
3940         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3941         for (i = 0; i < shader->output_signature.element_count; ++i)
3942         {
3943             const struct wined3d_shader_signature_element *output = &shader->output_signature.elements[i];
3944 
3945             if (!output->semantic_name)
3946                 continue;
3947 
3948             if (shader_match_semantic(output->semantic_name, WINED3D_DECL_USAGE_POSITION))
3949             {
3950                 TRACE("o%u is TMP_OUT\n", output->register_idx);
3951                 if (!output->semantic_idx)
3952                     priv_ctx->vs_output[output->register_idx] = "TMP_OUT";
3953                 else
3954                     priv_ctx->vs_output[output->register_idx] = "TA";
3955             }
3956             else if (shader_match_semantic(output->semantic_name, WINED3D_DECL_USAGE_PSIZE))
3957             {
3958                 TRACE("o%u is result.pointsize\n", output->register_idx);
3959                 if (!output->semantic_idx)
3960                     priv_ctx->vs_output[output->register_idx] = "result.pointsize";
3961                 else
3962                     priv_ctx->vs_output[output->register_idx] = "TA";
3963             }
3964             else if (shader_match_semantic(output->semantic_name, WINED3D_DECL_USAGE_COLOR))
3965             {
3966                 TRACE("o%u is result.color.?, idx %u\n", output->register_idx, output->semantic_idx);
3967                 if (!output->semantic_idx)
3968                     priv_ctx->vs_output[output->register_idx] = "result.color.primary";
3969                 else if (output->semantic_idx == 1)
3970                     priv_ctx->vs_output[output->register_idx] = "result.color.secondary";
3971                 else priv_ctx->vs_output[output->register_idx] = "TA";
3972             }
3973             else if (shader_match_semantic(output->semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
3974             {
3975                 TRACE("o%u is result.texcoord[%u]\n", output->register_idx, output->semantic_idx);
3976                 if (output->semantic_idx >= 8)
3977                     priv_ctx->vs_output[output->register_idx] = "TA";
3978                 else
3979                     priv_ctx->vs_output[output->register_idx] = texcoords[output->semantic_idx];
3980             }
3981             else if (shader_match_semantic(output->semantic_name, WINED3D_DECL_USAGE_FOG))
3982             {
3983                 TRACE("o%u is result.fogcoord\n", output->register_idx);
3984                 if (output->semantic_idx > 0)
3985                     priv_ctx->vs_output[output->register_idx] = "TA";
3986                 else
3987                     priv_ctx->vs_output[output->register_idx] = "result.fogcoord";
3988             }
3989             else
3990             {
3991                 priv_ctx->vs_output[output->register_idx] = "TA";
3992             }
3993         }
3994         return;
3995     }
3996 
3997     TRACE("Pixel shader uses declared varyings\n");
3998 
3999     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
4000     for(i = 0; i < 8; i++)
4001     {
4002         priv_ctx->texcrd_output[i] = "TA";
4003     }
4004     priv_ctx->color_output[0] = "TA";
4005     priv_ctx->color_output[1] = "TA";
4006     priv_ctx->fog_output = "TA";
4007 
4008     for (i = 0; i < ps_input_sig->element_count; ++i)
4009     {
4010         const struct wined3d_shader_signature_element *input = &ps_input_sig->elements[i];
4011 
4012         if (!input->semantic_name)
4013             continue;
4014 
4015         /* If a declared input register is not written by builtin arguments, don't write to it.
4016          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
4017          *
4018          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4019          * to TMP_OUT in any case
4020          */
4021         if (shader_match_semantic(input->semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
4022         {
4023             if (input->semantic_idx < 8)
4024                 priv_ctx->texcrd_output[input->semantic_idx] = decl_idx_to_string[input->register_idx];
4025         }
4026         else if (shader_match_semantic(input->semantic_name, WINED3D_DECL_USAGE_COLOR))
4027         {
4028             if (input->semantic_idx < 2)
4029                 priv_ctx->color_output[input->semantic_idx] = decl_idx_to_string[input->register_idx];
4030         }
4031         else if (shader_match_semantic(input->semantic_name, WINED3D_DECL_USAGE_FOG))
4032         {
4033             if (!input->semantic_idx)
4034                 priv_ctx->fog_output = decl_idx_to_string[input->register_idx];
4035         }
4036         else
4037         {
4038             continue;
4039         }
4040 
4041         if (!strcmp(decl_idx_to_string[input->register_idx], "result.color.primary")
4042                 || !strcmp(decl_idx_to_string[input->register_idx], "result.color.secondary"))
4043         {
4044             compiled->need_color_unclamp = TRUE;
4045         }
4046     }
4047 
4048     /* Map declared to declared */
4049     for (i = 0; i < shader->output_signature.element_count; ++i)
4050     {
4051         const struct wined3d_shader_signature_element *output = &shader->output_signature.elements[i];
4052 
4053         /* Write unread output to TA to throw them away */
4054         priv_ctx->vs_output[output->register_idx] = "TA";
4055 
4056         if (!output->semantic_name)
4057             continue;
4058 
4059         if (shader_match_semantic(output->semantic_name, WINED3D_DECL_USAGE_POSITION) && !output->semantic_idx)
4060         {
4061             priv_ctx->vs_output[output->register_idx] = "TMP_OUT";
4062             continue;
4063         }
4064         else if (shader_match_semantic(output->semantic_name, WINED3D_DECL_USAGE_PSIZE) && !output->semantic_idx)
4065         {
4066             priv_ctx->vs_output[output->register_idx] = "result.pointsize";
4067             continue;
4068         }
4069 
4070         for (j = 0; j < ps_input_sig->element_count; ++j)
4071         {
4072             const struct wined3d_shader_signature_element *input = &ps_input_sig->elements[j];
4073 
4074             if (!input->semantic_name)
4075                 continue;
4076 
4077             if (!strcmp(input->semantic_name, output->semantic_name)
4078                     && input->semantic_idx == output->semantic_idx)
4079             {
4080                 priv_ctx->vs_output[output->register_idx] = decl_idx_to_string[input->register_idx];
4081 
4082                 if (!strcmp(priv_ctx->vs_output[output->register_idx], "result.color.primary")
4083                         || !strcmp(priv_ctx->vs_output[output->register_idx], "result.color.secondary"))
4084                 {
4085                     compiled->need_color_unclamp = TRUE;
4086                 }
4087             }
4088         }
4089     }
4090 }
4091 
4092 /* Context activation is done by the caller. */
4093 static GLuint shader_arb_generate_vshader(const struct wined3d_shader *shader,
4094         const struct wined3d_gl_info *gl_info, struct wined3d_string_buffer *buffer,
4095         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled,
4096         const struct wined3d_shader_signature *ps_input_sig)
4097 {
4098     const struct arb_vshader_private *shader_data = shader->backend_data;
4099     const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4100     struct shader_arb_priv *priv = shader->device->shader_priv;
4101     GLuint ret;
4102     DWORD next_local = 0;
4103     struct shader_arb_ctx_priv priv_ctx;
4104     unsigned int i;
4105 
4106     memset(&priv_ctx, 0, sizeof(priv_ctx));
4107     priv_ctx.cur_vs_args = args;
4108     list_init(&priv_ctx.control_frames);
4109     init_output_registers(shader, ps_input_sig, &priv_ctx, compiled);
4110 
4111     /*  Create the hw ARB shader */
4112     shader_addline(buffer, "!!ARBvp1.0\n");
4113 
4114     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4115      * mesurable performance penalty, and we can always make use of it for clipplanes.
4116      */
4117     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4118     {
4119         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4120         priv_ctx.target_version = NV3;
4121         shader_addline(buffer, "ADDRESS aL;\n");
4122     }
4123     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4124     {
4125         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4126         priv_ctx.target_version = NV2;
4127         shader_addline(buffer, "ADDRESS aL;\n");
4128     } else {
4129         priv_ctx.target_version = ARB;
4130     }
4131 
4132     shader_addline(buffer, "TEMP TMP_OUT;\n");
4133     if (reg_maps->fog)
4134         shader_addline(buffer, "TEMP TMP_FOGCOORD;\n");
4135     if (need_helper_const(shader_data, reg_maps, gl_info))
4136     {
4137         char ftoa_tmp[17];
4138         wined3d_ftoa(eps, ftoa_tmp);
4139         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %s};\n", ftoa_tmp);
4140     }
4141     if (need_rel_addr_const(shader_data, reg_maps, gl_info))
4142     {
4143         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", shader_data->rel_offset);
4144         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4145     }
4146 
4147     shader_addline(buffer, "TEMP TA;\n");
4148     shader_addline(buffer, "TEMP TB;\n");
4149 
4150     /* Base Declarations */
4151     shader_generate_arb_declarations(shader, reg_maps, buffer, gl_info,
4152             &priv_ctx.vs_clipplanes, &priv_ctx);
4153 
4154     for (i = 0; i < WINED3D_MAX_CONSTS_I; ++i)
4155     {
4156         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4157         if (reg_maps->integer_constants & (1u << i) && priv_ctx.target_version >= NV2)
4158         {
4159             const DWORD *control_values = find_loop_control_values(shader, i);
4160 
4161             if(control_values)
4162             {
4163                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4164                                 control_values[0], control_values[1], control_values[2]);
4165             }
4166             else
4167             {
4168                 compiled->int_consts[i] = next_local;
4169                 compiled->num_int_consts++;
4170                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4171             }
4172         }
4173     }
4174 
4175     /* We need a constant to fixup the final position */
4176     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4177     compiled->pos_fixup = next_local++;
4178 
4179     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4180      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4181      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4182      * a replacement shader depend on the texcoord.w being set properly.
4183      *
4184      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4185      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4186      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4187      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4188      * this can eat a number of instructions, so skip it unless this cap is set as well
4189      */
4190     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4191     {
4192         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4193         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4194 
4195         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !priv->ffp_proj_control)
4196         {
4197             int i;
4198             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4199             for(i = 0; i < MAX_REG_TEXCRD; i++)
4200             {
4201                 if (reg_maps->u.texcoord_mask[i] && reg_maps->u.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4202                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4203             }
4204         }
4205     }
4206 
4207     /* The shader starts with the main function */
4208     priv_ctx.in_main_func = TRUE;
4209     /* Base Shader Body */
4210     if (FAILED(shader_generate_code(shader, buffer, reg_maps, &priv_ctx, NULL, NULL)))
4211         return -1;
4212 
4213     if (!priv_ctx.footer_written) vshader_add_footer(&priv_ctx,
4214             shader_data, args, reg_maps, gl_info, buffer);
4215 
4216     shader_addline(buffer, "END\n");
4217 
4218     /* TODO: change to resource.glObjectHandle or something like that */
4219     GL_EXTCALL(glGenProgramsARB(1, &ret));
4220 
4221     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4222     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4223 
4224     TRACE("Created hw vertex shader, prg=%d\n", ret);
4225     if (!shader_arb_compile(gl_info, GL_VERTEX_PROGRAM_ARB, buffer->buffer))
4226         return -1;
4227 
4228     return ret;
4229 }
4230 
4231 /* Context activation is done by the caller. */
4232 static struct arb_ps_compiled_shader *find_arb_pshader(struct wined3d_shader *shader,
4233         const struct arb_ps_compile_args *args)
4234 {
4235     struct wined3d_device *device = shader->device;
4236     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4237     const struct wined3d_d3d_info *d3d_info = &device->adapter->d3d_info;
4238     UINT i;
4239     DWORD new_size;
4240     struct arb_ps_compiled_shader *new_array;
4241     struct wined3d_string_buffer buffer;
4242     struct arb_pshader_private *shader_data;
4243     GLuint ret;
4244 
4245     if (!shader->backend_data)
4246     {
4247         struct shader_arb_priv *priv = device->shader_priv;
4248 
4249         shader->backend_data = heap_alloc_zero(sizeof(*shader_data));
4250         shader_data = shader->backend_data;
4251         shader_data->clamp_consts = shader->reg_maps.shader_version.major == 1;
4252 
4253         if (shader->reg_maps.shader_version.major < 3)
4254             shader_data->input_signature_idx = ~0U;
4255         else
4256             shader_data->input_signature_idx = find_input_signature(priv, &shader->input_signature);
4257 
4258         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4259 
4260         if (!d3d_info->vs_clipping)
4261             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->reg_maps,
4262                     d3d_info->limits.ffp_blend_stages - 1);
4263         else
4264             shader_data->clipplane_emulation = ~0U;
4265     }
4266     shader_data = shader->backend_data;
4267 
4268     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4269      * so a linear search is more performant than a hashmap or a binary search
4270      * (cache coherency etc)
4271      */
4272     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4273     {
4274         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4275             return &shader_data->gl_shaders[i];
4276     }
4277 
4278     TRACE("No matching GL shader found, compiling a new shader\n");
4279     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4280         if (shader_data->num_gl_shaders)
4281         {
4282             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4283             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4284                                     new_size * sizeof(*shader_data->gl_shaders));
4285         }
4286         else
4287         {
4288             new_array = heap_alloc_zero(sizeof(*shader_data->gl_shaders));
4289             new_size = 1;
4290         }
4291 
4292         if(!new_array) {
4293             ERR("Out of memory\n");
4294             return 0;
4295         }
4296         shader_data->gl_shaders = new_array;
4297         shader_data->shader_array_size = new_size;
4298     }
4299 
4300     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4301 
4302     pixelshader_update_resource_types(shader, args->super.tex_types);
4303 
4304     if (!string_buffer_init(&buffer))
4305     {
4306         ERR("Failed to initialize shader buffer.\n");
4307         return 0;
4308     }
4309 
4310     ret = shader_arb_generate_pshader(shader, gl_info, &buffer, args,
4311             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4312     string_buffer_free(&buffer);
4313     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4314 
4315     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4316 }
4317 
4318 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4319                                  const DWORD use_map, BOOL skip_int) {
4320     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4321     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4322     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4323     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4324     if(stored->ps_signature != new->ps_signature) return FALSE;
4325     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4326     if(skip_int) return TRUE;
4327 
4328     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4329 }
4330 
4331 static struct arb_vs_compiled_shader *find_arb_vshader(struct wined3d_shader *shader,
4332         const struct wined3d_gl_info *gl_info, DWORD use_map, const struct arb_vs_compile_args *args,
4333         const struct wined3d_shader_signature *ps_input_sig)
4334 {
4335     UINT i;
4336     DWORD new_size;
4337     struct arb_vs_compiled_shader *new_array;
4338     struct wined3d_string_buffer buffer;
4339     struct arb_vshader_private *shader_data;
4340     GLuint ret;
4341 
4342     if (!shader->backend_data)
4343     {
4344         const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
4345 
4346         shader->backend_data = heap_alloc_zero(sizeof(*shader_data));
4347         shader_data = shader->backend_data;
4348 
4349         if ((gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
4350                 && reg_maps->min_rel_offset <= reg_maps->max_rel_offset)
4351         {
4352             if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 127)
4353             {
4354                 FIXME("The difference between the minimum and maximum relative offset is > 127.\n");
4355                 FIXME("Which this OpenGL implementation does not support. Try using GLSL.\n");
4356                 FIXME("Min: %u, Max: %u.\n", reg_maps->min_rel_offset, reg_maps->max_rel_offset);
4357             }
4358             else if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 63)
4359                 shader_data->rel_offset = reg_maps->min_rel_offset + 63;
4360             else if (reg_maps->max_rel_offset > 63)
4361                 shader_data->rel_offset = reg_maps->min_rel_offset;
4362         }
4363     }
4364     shader_data = shader->backend_data;
4365 
4366     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4367      * so a linear search is more performant than a hashmap or a binary search
4368      * (cache coherency etc)
4369      */
4370     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4371         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4372                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4373         {
4374             return &shader_data->gl_shaders[i];
4375         }
4376     }
4377 
4378     TRACE("No matching GL shader found, compiling a new shader\n");
4379 
4380     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4381         if (shader_data->num_gl_shaders)
4382         {
4383             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4384             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4385                                     new_size * sizeof(*shader_data->gl_shaders));
4386         }
4387         else
4388         {
4389             new_array = heap_alloc_zero(sizeof(*shader_data->gl_shaders));
4390             new_size = 1;
4391         }
4392 
4393         if(!new_array) {
4394             ERR("Out of memory\n");
4395             return 0;
4396         }
4397         shader_data->gl_shaders = new_array;
4398         shader_data->shader_array_size = new_size;
4399     }
4400 
4401     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4402 
4403     if (!string_buffer_init(&buffer))
4404     {
4405         ERR("Failed to initialize shader buffer.\n");
4406         return 0;
4407     }
4408 
4409     ret = shader_arb_generate_vshader(shader, gl_info, &buffer, args,
4410             &shader_data->gl_shaders[shader_data->num_gl_shaders],
4411             ps_input_sig);
4412     string_buffer_free(&buffer);
4413     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4414 
4415     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4416 }
4417 
4418 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4419         const struct wined3d_context *context, const struct wined3d_shader *shader,
4420         struct arb_ps_compile_args *args)
4421 {
4422     const struct wined3d_gl_info *gl_info = context->gl_info;
4423     const struct wined3d_d3d_info *d3d_info = context->d3d_info;
4424     int i;
4425     WORD int_skip;
4426 
4427     find_ps_compile_args(state, shader, context->stream_info.position_transformed, &args->super, context);
4428 
4429     /* This forces all local boolean constants to 1 to make them stateblock independent */
4430     args->bools = shader->reg_maps.local_bool_consts;
4431 
4432     for (i = 0; i < WINED3D_MAX_CONSTS_B; ++i)
4433     {
4434         if (state->ps_consts_b[i])
4435             args->bools |= ( 1u << i);
4436     }
4437 
4438     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4439      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4440      * duplicate the shader than have a no-op KIL instruction in every shader
4441      */
4442     if (!d3d_info->vs_clipping && use_vs(state)
4443             && state->render_states[WINED3D_RS_CLIPPING]
4444             && state->render_states[WINED3D_RS_CLIPPLANEENABLE])
4445         args->clip = 1;
4446     else
4447         args->clip = 0;
4448 
4449     /* Skip if unused or local, or supported natively */
4450     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4451     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4452     {
4453         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4454         return;
4455     }
4456 
4457     for (i = 0; i < WINED3D_MAX_CONSTS_I; ++i)
4458     {
4459         if (int_skip & (1u << i))
4460         {
4461             args->loop_ctrl[i][0] = 0;
4462             args->loop_ctrl[i][1] = 0;
4463             args->loop_ctrl[i][2] = 0;
4464         }
4465         else
4466         {
4467             args->loop_ctrl[i][0] = state->ps_consts_i[i].x;
4468             args->loop_ctrl[i][1] = state->ps_consts_i[i].y;
4469             args->loop_ctrl[i][2] = state->ps_consts_i[i].z;
4470         }
4471     }
4472 }
4473 
4474 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4475         const struct wined3d_context *context, const struct wined3d_shader *shader,
4476         struct arb_vs_compile_args *args)
4477 {
4478     const struct wined3d_device *device = shader->device;
4479     const struct wined3d_adapter *adapter = device->adapter;
4480     const struct wined3d_gl_info *gl_info = context->gl_info;
4481     const struct wined3d_d3d_info *d3d_info = context->d3d_info;
4482     int i;
4483     WORD int_skip;
4484 
4485     find_vs_compile_args(state, shader, context->stream_info.swizzle_map, &args->super, context);
4486 
4487     args->clip.boolclip_compare = 0;
4488     if (use_ps(state))
4489     {
4490         const struct wined3d_shader *ps = state->shader[WINED3D_SHADER_TYPE_PIXEL];
4491         const struct arb_pshader_private *shader_priv = ps->backend_data;
4492         args->ps_signature = shader_priv->input_signature_idx;
4493 
4494         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4495     }
4496     else
4497     {
4498         args->ps_signature = ~0;
4499         if (!d3d_info->vs_clipping && adapter->fragment_pipe == &arbfp_fragment_pipeline)
4500             args->clip.boolclip.clip_texcoord = ffp_clip_emul(context) ? d3d_info->limits.ffp_blend_stages : 0;
4501         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4502     }
4503 
4504     if (args->clip.boolclip.clip_texcoord)
4505     {
4506         if (state->render_states[WINED3D_RS_CLIPPING])
4507             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3D_RS_CLIPPLANEENABLE];
4508         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4509     }
4510 
4511     /* This forces all local boolean constants to 1 to make them stateblock independent */
4512     args->clip.boolclip.bools = shader->reg_maps.local_bool_consts;
4513     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4514     for (i = 0; i < WINED3D_MAX_CONSTS_B; ++i)
4515     {
4516         if (state->vs_consts_b[i])
4517             args->clip.boolclip.bools |= (1u << i);
4518     }
4519 
4520     args->vertex.samplers[0] = context->tex_unit_map[MAX_FRAGMENT_SAMPLERS + 0];
4521     args->vertex.samplers[1] = context->tex_unit_map[MAX_FRAGMENT_SAMPLERS + 1];
4522     args->vertex.samplers[2] = context->tex_unit_map[MAX_FRAGMENT_SAMPLERS + 2];
4523     args->vertex.samplers[3] = 0;
4524 
4525     /* Skip if unused or local */
4526     int_skip = ~shader->reg_maps.integer_constants | shader->reg_maps.local_int_consts;
4527     /* This is about flow control, not clipping. */
4528     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4529     {
4530         memset(args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4531         return;
4532     }
4533 
4534     for (i = 0; i < WINED3D_MAX_CONSTS_I; ++i)
4535     {
4536         if (int_skip & (1u << i))
4537         {
4538             args->loop_ctrl[i][0] = 0;
4539             args->loop_ctrl[i][1] = 0;
4540             args->loop_ctrl[i][2] = 0;
4541         }
4542         else
4543         {
4544             args->loop_ctrl[i][0] = state->vs_consts_i[i].x;
4545             args->loop_ctrl[i][1] = state->vs_consts_i[i].y;
4546             args->loop_ctrl[i][2] = state->vs_consts_i[i].z;
4547         }
4548     }
4549 }
4550 
4551 /* Context activation is done by the caller. */
4552 static void shader_arb_select(void *shader_priv, struct wined3d_context *context,
4553         const struct wined3d_state *state)
4554 {
4555     struct shader_arb_priv *priv = shader_priv;
4556     const struct wined3d_gl_info *gl_info = context->gl_info;
4557     int i;
4558 
4559     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4560     if (use_ps(state))
4561     {
4562         struct wined3d_shader *ps = state->shader[WINED3D_SHADER_TYPE_PIXEL];
4563         struct arb_ps_compile_args compile_args;
4564         struct arb_ps_compiled_shader *compiled;
4565 
4566         TRACE("Using pixel shader %p.\n", ps);
4567         find_arb_ps_compile_args(state, context, ps, &compile_args);
4568         compiled = find_arb_pshader(ps, &compile_args);
4569         priv->current_fprogram_id = compiled->prgId;
4570         priv->compiled_fprog = compiled;
4571 
4572         /* Bind the fragment program */
4573         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4574         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4575 
4576         if (!priv->use_arbfp_fixed_func)
4577             priv->fragment_pipe->enable_extension(gl_info, FALSE);
4578 
4579         /* Enable OpenGL fragment programs. */
4580         gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
4581         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4582 
4583         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", priv->current_fprogram_id);
4584 
4585         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4586          * a 1.x and newer shader, reload the first 8 constants
4587          */
4588         if (priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->backend_data)->clamp_consts)
4589         {
4590             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->backend_data)->clamp_consts;
4591             priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, 8);
4592             for(i = 0; i < 8; i++)
4593             {
4594                 priv->pshader_const_dirty[i] = 1;
4595             }
4596             /* Also takes care of loading local constants */
4597             shader_arb_load_constants_internal(shader_priv, context, state, TRUE, FALSE, TRUE);
4598         }
4599         else
4600         {
4601             UINT rt_height = state->fb->render_targets[0]->height;
4602             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4603         }
4604 
4605         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4606         if (compiled->np2fixup_info.super.active)
4607             context->constant_update_mask |= WINED3D_SHADER_CONST_PS_NP2_FIXUP;
4608 
4609         if (ps->load_local_constsF)
4610             context->constant_update_mask |= WINED3D_SHADER_CONST_PS_F;
4611     }
4612     else
4613     {
4614         if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4615         {
4616             /* Disable only if we're not using arbfp fixed function fragment
4617              * processing. If this is used, keep GL_FRAGMENT_PROGRAM_ARB
4618              * enabled, and the fixed function pipeline will bind the fixed
4619              * function replacement shader. */
4620             gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4621             checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4622             priv->current_fprogram_id = 0;
4623         }
4624         priv->fragment_pipe->enable_extension(gl_info, TRUE);
4625     }
4626 
4627     if (use_vs(state))
4628     {
4629         struct wined3d_shader *vs = state->shader[WINED3D_SHADER_TYPE_VERTEX];
4630         struct arb_vs_compile_args compile_args;
4631         struct arb_vs_compiled_shader *compiled;
4632         const struct wined3d_shader_signature *ps_input_sig;
4633 
4634         TRACE("Using vertex shader %p\n", vs);
4635         find_arb_vs_compile_args(state, context, vs, &compile_args);
4636 
4637         /* Instead of searching for the signature in the signature list, read the one from the
4638          * current pixel shader. It's maybe not the shader where the signature came from, but it
4639          * is the same signature and faster to find. */
4640         if (compile_args.ps_signature == ~0U)
4641             ps_input_sig = NULL;
4642         else
4643             ps_input_sig = &state->shader[WINED3D_SHADER_TYPE_PIXEL]->input_signature;
4644 
4645         compiled = find_arb_vshader(vs, context->gl_info, context->stream_info.use_map,
4646                 &compile_args, ps_input_sig);
4647         priv->current_vprogram_id = compiled->prgId;
4648         priv->compiled_vprog = compiled;
4649 
4650         /* Bind the vertex program */
4651         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4652         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4653 
4654         priv->vertex_pipe->vp_enable(gl_info, FALSE);
4655 
4656         /* Enable OpenGL vertex programs */
4657         gl_info->gl_ops.gl.p_glEnable(GL_VERTEX_PROGRAM_ARB);
4658         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4659         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", priv->current_vprogram_id);
4660         shader_arb_vs_local_constants(compiled, context, state);
4661 
4662         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4663             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4664 
4665             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4666             {
4667                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4668                 checkGLcall("glClampColorARB");
4669             } else {
4670                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4671             }
4672         }
4673 
4674         if (vs->load_local_constsF)
4675             context->constant_update_mask |= WINED3D_SHADER_CONST_VS_F;
4676     }
4677     else
4678     {
4679         if (gl_info->supported[ARB_VERTEX_PROGRAM])
4680         {
4681             priv->current_vprogram_id = 0;
4682             gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4683             checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4684         }
4685         priv->vertex_pipe->vp_enable(gl_info, TRUE);
4686     }
4687 }
4688 
4689 static void shader_arb_select_compute(void *shader_priv, struct wined3d_context *context,
4690         const struct wined3d_state *state)
4691 {
4692     ERR("Compute pipeline not supported by the ARB shader backend.\n");
4693 }
4694 
4695 /* Context activation is done by the caller. */
4696 static void shader_arb_disable(void *shader_priv, struct wined3d_context *context)
4697 {
4698     const struct wined3d_gl_info *gl_info = context->gl_info;
4699     struct shader_arb_priv *priv = shader_priv;
4700 
4701     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4702     {
4703         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
4704         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4705         priv->current_fprogram_id = 0;
4706     }
4707     priv->fragment_pipe->enable_extension(gl_info, FALSE);
4708 
4709     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4710     {
4711         priv->current_vprogram_id = 0;
4712         gl_info->gl_ops.gl.p_glDisable(GL_VERTEX_PROGRAM_ARB);
4713         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4714     }
4715     priv->vertex_pipe->vp_enable(gl_info, FALSE);
4716 
4717     if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT] && priv->last_vs_color_unclamp)
4718     {
4719         GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, GL_FIXED_ONLY_ARB));
4720         checkGLcall("glClampColorARB");
4721         priv->last_vs_color_unclamp = FALSE;
4722     }
4723 
4724     context->shader_update_mask = (1u << WINED3D_SHADER_TYPE_PIXEL)
4725             | (1u << WINED3D_SHADER_TYPE_VERTEX)
4726             | (1u << WINED3D_SHADER_TYPE_GEOMETRY)
4727             | (1u << WINED3D_SHADER_TYPE_HULL)
4728             | (1u << WINED3D_SHADER_TYPE_DOMAIN)
4729             | (1u << WINED3D_SHADER_TYPE_COMPUTE);
4730 }
4731 
4732 static void shader_arb_destroy(struct wined3d_shader *shader)
4733 {
4734     struct wined3d_device *device = shader->device;
4735     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4736 
4737     if (shader_is_pshader_version(shader->reg_maps.shader_version.type))
4738     {
4739         struct arb_pshader_private *shader_data = shader->backend_data;
4740         UINT i;
4741 
4742         if(!shader_data) return; /* This can happen if a shader was never compiled */
4743 
4744         if (shader_data->num_gl_shaders)
4745         {
4746             struct wined3d_context *context = context_acquire(device, NULL, 0);
4747 
4748             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4749             {
4750                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4751                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4752             }
4753 
4754             context_release(context);
4755         }
4756 
4757         heap_free(shader_data->gl_shaders);
4758         heap_free(shader_data);
4759         shader->backend_data = NULL;
4760     }
4761     else
4762     {
4763         struct arb_vshader_private *shader_data = shader->backend_data;
4764         UINT i;
4765 
4766         if(!shader_data) return; /* This can happen if a shader was never compiled */
4767 
4768         if (shader_data->num_gl_shaders)
4769         {
4770             struct wined3d_context *context = context_acquire(device, NULL, 0);
4771 
4772             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4773             {
4774                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4775                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4776             }
4777 
4778             context_release(context);
4779         }
4780 
4781         heap_free(shader_data->gl_shaders);
4782         heap_free(shader_data);
4783         shader->backend_data = NULL;
4784     }
4785 }
4786 
4787 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4788 {
4789     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4790     return compare_sig(key, &e->sig);
4791 }
4792 
4793 static HRESULT shader_arb_alloc(struct wined3d_device *device, const struct wined3d_vertex_pipe_ops *vertex_pipe,
4794         const struct fragment_pipeline *fragment_pipe)
4795 {
4796     const struct wined3d_d3d_info *d3d_info = &device->adapter->d3d_info;
4797     struct fragment_caps fragment_caps;
4798     void *vertex_priv, *fragment_priv;
4799     struct shader_arb_priv *priv;
4800 
4801     if (!(priv = heap_alloc_zero(sizeof(*priv))))
4802         return E_OUTOFMEMORY;
4803 
4804     if (!(vertex_priv = vertex_pipe->vp_alloc(&arb_program_shader_backend, priv)))
4805     {
4806         ERR("Failed to initialize vertex pipe.\n");
4807         heap_free(priv);
4808         return E_FAIL;
4809     }
4810 
4811     if (!(fragment_priv = fragment_pipe->alloc_private(&arb_program_shader_backend, priv)))
4812     {
4813         ERR("Failed to initialize fragment pipe.\n");
4814         vertex_pipe->vp_free(device);
4815         heap_free(priv);
4816         return E_FAIL;
4817     }
4818 
4819     memset(priv->vshader_const_dirty, 1,
4820            sizeof(*priv->vshader_const_dirty) * d3d_info->limits.vs_uniform_count);
4821     memset(priv->pshader_const_dirty, 1,
4822             sizeof(*priv->pshader_const_dirty) * d3d_info->limits.ps_uniform_count);
4823 
4824     wine_rb_init(&priv->signature_tree, sig_tree_compare);
4825 
4826     priv->vertex_pipe = vertex_pipe;
4827     priv->fragment_pipe = fragment_pipe;
4828     fragment_pipe->get_caps(&device->adapter->gl_info, &fragment_caps);
4829     priv->ffp_proj_control = fragment_caps.wined3d_caps & WINED3D_FRAGMENT_CAP_PROJ_CONTROL;
4830 
4831     device->vertex_priv = vertex_priv;
4832     device->fragment_priv = fragment_priv;
4833     device->shader_priv = priv;
4834 
4835     return WINED3D_OK;
4836 }
4837 
4838 static void release_signature(struct wine_rb_entry *entry, void *context)
4839 {
4840     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4841     unsigned int i;
4842 
4843     for (i = 0; i < sig->sig.element_count; ++i)
4844     {
4845         heap_free((char *)sig->sig.elements[i].semantic_name);
4846     }
4847     heap_free(sig->sig.elements);
4848     heap_free(sig);
4849 }
4850 
4851 /* Context activation is done by the caller. */
4852 static void shader_arb_free(struct wined3d_device *device)
4853 {
4854     struct shader_arb_priv *priv = device->shader_priv;
4855 
4856     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4857     priv->fragment_pipe->free_private(device);
4858     priv->vertex_pipe->vp_free(device);
4859     heap_free(device->shader_priv);
4860 }
4861 
4862 static BOOL shader_arb_allocate_context_data(struct wined3d_context *context)
4863 {
4864     return TRUE;
4865 }
4866 
4867 static void shader_arb_free_context_data(struct wined3d_context *context)
4868 {
4869     struct shader_arb_priv *priv;
4870 
4871     priv = context->device->shader_priv;
4872     if (priv->last_context == context)
4873         priv->last_context = NULL;
4874 }
4875 
4876 static void shader_arb_init_context_state(struct wined3d_context *context) {}
4877 
4878 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *caps)
4879 {
4880     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4881     {
4882         DWORD vs_consts;
4883         UINT vs_version;
4884 
4885         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4886          * for vertex programs. If the native limit is less than that it's
4887          * not very useful, and e.g. Mesa swrast returns 0, probably to
4888          * indicate it's a software implementation. */
4889         if (gl_info->limits.arb_vs_native_constants < 96)
4890             vs_consts = gl_info->limits.arb_vs_float_constants;
4891         else
4892             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
4893 
4894         if (gl_info->supported[NV_VERTEX_PROGRAM3])
4895         {
4896             vs_version = 3;
4897             TRACE("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4898         }
4899         else if (vs_consts >= 256)
4900         {
4901             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4902             vs_version = 2;
4903             TRACE("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4904         }
4905         else
4906         {
4907             vs_version = 1;
4908             TRACE("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4909         }
4910         caps->vs_version = min(wined3d_settings.max_sm_vs, vs_version);
4911         caps->vs_uniform_count = min(WINED3D_MAX_VS_CONSTS_F, vs_consts);
4912     }
4913     else
4914     {
4915         caps->vs_version = 0;
4916         caps->vs_uniform_count = 0;
4917     }
4918 
4919     caps->hs_version = 0;
4920     caps->ds_version = 0;
4921     caps->gs_version = 0;
4922     caps->cs_version = 0;
4923 
4924     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4925     {
4926         DWORD ps_consts;
4927         UINT ps_version;
4928 
4929         /* Similar as above for vertex programs, but the minimum for fragment
4930          * programs is 24. */
4931         if (gl_info->limits.arb_ps_native_constants < 24)
4932             ps_consts = gl_info->limits.arb_ps_float_constants;
4933         else
4934             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
4935 
4936         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
4937         {
4938             ps_version = 3;
4939             TRACE("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4940         }
4941         else if (ps_consts >= 32)
4942         {
4943             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4944             ps_version = 2;
4945             TRACE("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4946         }
4947         else
4948         {
4949             ps_version = 1;
4950             TRACE("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4951         }
4952         caps->ps_version = min(wined3d_settings.max_sm_ps, ps_version);
4953         caps->ps_uniform_count = min(WINED3D_MAX_PS_CONSTS_F, ps_consts);
4954         caps->ps_1x_max_value = 8.0f;
4955     }
4956     else
4957     {
4958         caps->ps_version = 0;
4959         caps->ps_uniform_count = 0;
4960         caps->ps_1x_max_value = 0.0f;
4961     }
4962 
4963     caps->varying_count = 0;
4964     caps->wined3d_caps = WINED3D_SHADER_CAP_SRGB_WRITE;
4965     if (use_nv_clip(gl_info))
4966         caps->wined3d_caps |= WINED3D_SHADER_CAP_VS_CLIPPING;
4967 }
4968 
4969 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
4970 {
4971     /* We support everything except complex conversions. */
4972     return !is_complex_fixup(fixup);
4973 }
4974 
4975 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
4976     DWORD shift;
4977     char write_mask[20], regstr[50];
4978     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
4979     BOOL is_color = FALSE;
4980     const struct wined3d_shader_dst_param *dst;
4981 
4982     if (!ins->dst_count) return;
4983 
4984     dst = &ins->dst[0];
4985     shift = dst->shift;
4986     if (!shift) return; /* Saturate alone is handled by the instructions */
4987 
4988     shader_arb_get_write_mask(ins, dst, write_mask);
4989     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
4990 
4991     /* Generate a line that does the output modifier computation
4992      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
4993      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
4994      */
4995     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
4996                    regstr, write_mask, regstr, shift_tab[shift]);
4997 }
4998 
4999 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
5000 {
5001     /* WINED3DSIH_ABS                              */ shader_hw_map2gl,
5002     /* WINED3DSIH_ADD                              */ shader_hw_map2gl,
5003     /* WINED3DSIH_AND                              */ NULL,
5004     /* WINED3DSIH_ATOMIC_AND                       */ NULL,
5005     /* WINED3DSIH_ATOMIC_CMP_STORE                 */ NULL,
5006     /* WINED3DSIH_ATOMIC_IADD                      */ NULL,
5007     /* WINED3DSIH_ATOMIC_IMAX                      */ NULL,
5008     /* WINED3DSIH_ATOMIC_IMIN                      */ NULL,
5009     /* WINED3DSIH_ATOMIC_OR                        */ NULL,
5010     /* WINED3DSIH_ATOMIC_UMAX                      */ NULL,
5011     /* WINED3DSIH_ATOMIC_UMIN                      */ NULL,
5012     /* WINED3DSIH_ATOMIC_XOR                       */ NULL,
5013     /* WINED3DSIH_BEM                              */ pshader_hw_bem,
5014     /* WINED3DSIH_BFI                              */ NULL,
5015     /* WINED3DSIH_BFREV                            */ NULL,
5016     /* WINED3DSIH_BREAK                            */ shader_hw_break,
5017     /* WINED3DSIH_BREAKC                           */ shader_hw_breakc,
5018     /* WINED3DSIH_BREAKP                           */ NULL,
5019     /* WINED3DSIH_BUFINFO                          */ NULL,
5020     /* WINED3DSIH_CALL                             */ shader_hw_call,
5021     /* WINED3DSIH_CALLNZ                           */ NULL,
5022     /* WINED3DSIH_CASE                             */ NULL,
5023     /* WINED3DSIH_CMP                              */ pshader_hw_cmp,
5024     /* WINED3DSIH_CND                              */ pshader_hw_cnd,
5025     /* WINED3DSIH_CONTINUE                         */ NULL,
5026     /* WINED3DSIH_CONTINUEP                        */ NULL,
5027     /* WINED3DSIH_COUNTBITS                        */ NULL,
5028     /* WINED3DSIH_CRS                              */ shader_hw_map2gl,
5029     /* WINED3DSIH_CUT                              */ NULL,
5030     /* WINED3DSIH_CUT_STREAM                       */ NULL,
5031     /* WINED3DSIH_DCL                              */ shader_hw_nop,
5032     /* WINED3DSIH_DCL_CONSTANT_BUFFER              */ shader_hw_nop,
5033     /* WINED3DSIH_DCL_FUNCTION_BODY                */ NULL,
5034     /* WINED3DSIH_DCL_FUNCTION_TABLE               */ NULL,
5035     /* WINED3DSIH_DCL_GLOBAL_FLAGS                 */ NULL,
5036     /* WINED3DSIH_DCL_GS_INSTANCES                 */ NULL,
5037     /* WINED3DSIH_DCL_HS_FORK_PHASE_INSTANCE_COUNT */ NULL,
5038     /* WINED3DSIH_DCL_HS_JOIN_PHASE_INSTANCE_COUNT */ NULL,
5039     /* WINED3DSIH_DCL_HS_MAX_TESSFACTOR            */ NULL,
5040     /* WINED3DSIH_DCL_IMMEDIATE_CONSTANT_BUFFER    */ NULL,
5041     /* WINED3DSIH_DCL_INDEX_RANGE                  */ NULL,
5042     /* WINED3DSIH_DCL_INDEXABLE_TEMP               */ NULL,
5043     /* WINED3DSIH_DCL_INPUT                        */ NULL,
5044     /* WINED3DSIH_DCL_INPUT_CONTROL_POINT_COUNT    */ NULL,
5045     /* WINED3DSIH_DCL_INPUT_PRIMITIVE              */ shader_hw_nop,
5046     /* WINED3DSIH_DCL_INPUT_PS                     */ NULL,
5047     /* WINED3DSIH_DCL_INPUT_PS_SGV                 */ NULL,
5048     /* WINED3DSIH_DCL_INPUT_PS_SIV                 */ NULL,
5049     /* WINED3DSIH_DCL_INPUT_SGV                    */ NULL,
5050     /* WINED3DSIH_DCL_INPUT_SIV                    */ NULL,
5051     /* WINED3DSIH_DCL_INTERFACE                    */ NULL,
5052     /* WINED3DSIH_DCL_OUTPUT                       */ NULL,
5053     /* WINED3DSIH_DCL_OUTPUT_CONTROL_POINT_COUNT   */ NULL,
5054     /* WINED3DSIH_DCL_OUTPUT_SIV                   */ NULL,
5055     /* WINED3DSIH_DCL_OUTPUT_TOPOLOGY              */ shader_hw_nop,
5056     /* WINED3DSIH_DCL_RESOURCE_RAW                 */ NULL,
5057     /* WINED3DSIH_DCL_RESOURCE_STRUCTURED          */ NULL,
5058     /* WINED3DSIH_DCL_SAMPLER                      */ NULL,
5059     /* WINED3DSIH_DCL_STREAM                       */ NULL,
5060     /* WINED3DSIH_DCL_TEMPS                        */ NULL,
5061     /* WINED3DSIH_DCL_TESSELLATOR_DOMAIN           */ NULL,
5062     /* WINED3DSIH_DCL_TESSELLATOR_OUTPUT_PRIMITIVE */ NULL,
5063     /* WINED3DSIH_DCL_TESSELLATOR_PARTITIONING     */ NULL,
5064     /* WINED3DSIH_DCL_TGSM_RAW                     */ NULL,
5065     /* WINED3DSIH_DCL_TGSM_STRUCTURED              */ NULL,
5066     /* WINED3DSIH_DCL_THREAD_GROUP                 */ NULL,
5067     /* WINED3DSIH_DCL_UAV_RAW                      */ NULL,
5068     /* WINED3DSIH_DCL_UAV_STRUCTURED               */ NULL,
5069     /* WINED3DSIH_DCL_UAV_TYPED                    */ NULL,
5070     /* WINED3DSIH_DCL_VERTICES_OUT                 */ shader_hw_nop,
5071     /* WINED3DSIH_DEF                              */ shader_hw_nop,
5072     /* WINED3DSIH_DEFAULT                          */ NULL,
5073     /* WINED3DSIH_DEFB                             */ shader_hw_nop,
5074     /* WINED3DSIH_DEFI                             */ shader_hw_nop,
5075     /* WINED3DSIH_DIV                              */ NULL,
5076     /* WINED3DSIH_DP2                              */ NULL,
5077     /* WINED3DSIH_DP2ADD                           */ pshader_hw_dp2add,
5078     /* WINED3DSIH_DP3                              */ shader_hw_map2gl,
5079     /* WINED3DSIH_DP4                              */ shader_hw_map2gl,
5080     /* WINED3DSIH_DST                              */ shader_hw_map2gl,
5081     /* WINED3DSIH_DSX                              */ shader_hw_map2gl,
5082     /* WINED3DSIH_DSX_COARSE                       */ NULL,
5083     /* WINED3DSIH_DSX_FINE                         */ NULL,
5084     /* WINED3DSIH_DSY                              */ shader_hw_dsy,
5085     /* WINED3DSIH_DSY_COARSE                       */ NULL,
5086     /* WINED3DSIH_DSY_FINE                         */ NULL,
5087     /* WINED3DSIH_EVAL_SAMPLE_INDEX                */ NULL,
5088     /* WINED3DSIH_ELSE                             */ shader_hw_else,
5089     /* WINED3DSIH_EMIT                             */ NULL,
5090     /* WINED3DSIH_EMIT_STREAM                      */ NULL,
5091     /* WINED3DSIH_ENDIF                            */ shader_hw_endif,
5092     /* WINED3DSIH_ENDLOOP                          */ shader_hw_endloop,
5093     /* WINED3DSIH_ENDREP                           */ shader_hw_endrep,
5094     /* WINED3DSIH_ENDSWITCH                        */ NULL,
5095     /* WINED3DSIH_EQ                               */ NULL,
5096     /* WINED3DSIH_EXP                              */ shader_hw_scalar_op,
5097     /* WINED3DSIH_EXPP                             */ shader_hw_scalar_op,
5098     /* WINED3DSIH_F16TOF32                         */ NULL,
5099     /* WINED3DSIH_F32TOF16                         */ NULL,
5100     /* WINED3DSIH_FCALL                            */ NULL,
5101     /* WINED3DSIH_FIRSTBIT_HI                      */ NULL,
5102     /* WINED3DSIH_FIRSTBIT_LO                      */ NULL,
5103     /* WINED3DSIH_FIRSTBIT_SHI                     */ NULL,
5104     /* WINED3DSIH_FRC                              */ shader_hw_map2gl,
5105     /* WINED3DSIH_FTOI                             */ NULL,
5106     /* WINED3DSIH_FTOU                             */ NULL,
5107     /* WINED3DSIH_GATHER4                          */ NULL,
5108     /* WINED3DSIH_GATHER4_C                        */ NULL,
5109     /* WINED3DSIH_GATHER4_PO                       */ NULL,
5110     /* WINED3DSIH_GATHER4_PO_C                     */ NULL,
5111     /* WINED3DSIH_GE                               */ NULL,
5112     /* WINED3DSIH_HS_CONTROL_POINT_PHASE           */ NULL,
5113     /* WINED3DSIH_HS_DECLS                         */ NULL,
5114     /* WINED3DSIH_HS_FORK_PHASE                    */ NULL,
5115     /* WINED3DSIH_HS_JOIN_PHASE                    */ NULL,
5116     /* WINED3DSIH_IADD                             */ NULL,
5117     /* WINED3DSIH_IBFE                             */ NULL,
5118     /* WINED3DSIH_IEQ                              */ NULL,
5119     /* WINED3DSIH_IF                               */ NULL /* Hardcoded into the shader */,
5120     /* WINED3DSIH_IFC                              */ shader_hw_ifc,
5121     /* WINED3DSIH_IGE                              */ NULL,
5122     /* WINED3DSIH_ILT                              */ NULL,
5123     /* WINED3DSIH_IMAD                             */ NULL,
5124     /* WINED3DSIH_IMAX                             */ NULL,
5125     /* WINED3DSIH_IMIN                             */ NULL,
5126     /* WINED3DSIH_IMM_ATOMIC_ALLOC                 */ NULL,
5127     /* WINED3DSIH_IMM_ATOMIC_AND                   */ NULL,
5128     /* WINED3DSIH_IMM_ATOMIC_CMP_EXCH              */ NULL,
5129     /* WINED3DSIH_IMM_ATOMIC_CONSUME               */ NULL,
5130     /* WINED3DSIH_IMM_ATOMIC_EXCH                  */ NULL,
5131     /* WINED3DSIH_IMM_ATOMIC_IADD                  */ NULL,
5132     /* WINED3DSIH_IMM_ATOMIC_IMAX                  */ NULL,
5133     /* WINED3DSIH_IMM_ATOMIC_IMIN                  */ NULL,
5134     /* WINED3DSIH_IMM_ATOMIC_OR                    */ NULL,
5135     /* WINED3DSIH_IMM_ATOMIC_UMAX                  */ NULL,
5136     /* WINED3DSIH_IMM_ATOMIC_UMIN                  */ NULL,
5137     /* WINED3DSIH_IMM_ATOMIC_XOR                   */ NULL,
5138     /* WINED3DSIH_IMUL                             */ NULL,
5139     /* WINED3DSIH_INE                              */ NULL,
5140     /* WINED3DSIH_INEG                             */ NULL,
5141     /* WINED3DSIH_ISHL                             */ NULL,
5142     /* WINED3DSIH_ISHR                             */ NULL,
5143     /* WINED3DSIH_ITOF                             */ NULL,
5144     /* WINED3DSIH_LABEL                            */ shader_hw_label,
5145     /* WINED3DSIH_LD                               */ NULL,
5146     /* WINED3DSIH_LD2DMS                           */ NULL,
5147     /* WINED3DSIH_LD_RAW                           */ NULL,
5148     /* WINED3DSIH_LD_STRUCTURED                    */ NULL,
5149     /* WINED3DSIH_LD_UAV_TYPED                     */ NULL,
5150     /* WINED3DSIH_LIT                              */ shader_hw_map2gl,
5151     /* WINED3DSIH_LOD                              */ NULL,
5152     /* WINED3DSIH_LOG                              */ shader_hw_scalar_op,
5153     /* WINED3DSIH_LOGP                             */ shader_hw_scalar_op,
5154     /* WINED3DSIH_LOOP                             */ shader_hw_loop,
5155     /* WINED3DSIH_LRP                              */ shader_hw_lrp,
5156     /* WINED3DSIH_LT                               */ NULL,
5157     /* WINED3DSIH_M3x2                             */ shader_hw_mnxn,
5158     /* WINED3DSIH_M3x3                             */ shader_hw_mnxn,
5159     /* WINED3DSIH_M3x4                             */ shader_hw_mnxn,
5160     /* WINED3DSIH_M4x3                             */ shader_hw_mnxn,
5161     /* WINED3DSIH_M4x4                             */ shader_hw_mnxn,
5162     /* WINED3DSIH_MAD                              */ shader_hw_map2gl,
5163     /* WINED3DSIH_MAX                              */ shader_hw_map2gl,
5164     /* WINED3DSIH_MIN                              */ shader_hw_map2gl,
5165     /* WINED3DSIH_MOV                              */ shader_hw_mov,
5166     /* WINED3DSIH_MOVA                             */ shader_hw_mov,
5167     /* WINED3DSIH_MOVC                             */ NULL,
5168     /* WINED3DSIH_MUL                              */ shader_hw_map2gl,
5169     /* WINED3DSIH_NE                               */ NULL,
5170     /* WINED3DSIH_NOP                              */ shader_hw_nop,
5171     /* WINED3DSIH_NOT                              */ NULL,
5172     /* WINED3DSIH_NRM                              */ shader_hw_nrm,
5173     /* WINED3DSIH_OR                               */ NULL,
5174     /* WINED3DSIH_PHASE                            */ shader_hw_nop,
5175     /* WINED3DSIH_POW                              */ shader_hw_pow,
5176     /* WINED3DSIH_RCP                              */ shader_hw_scalar_op,
5177     /* WINED3DSIH_REP                              */ shader_hw_rep,
5178     /* WINED3DSIH_RESINFO                          */ NULL,
5179     /* WINED3DSIH_RET                              */ shader_hw_ret,
5180     /* WINED3DSIH_RETP                             */ NULL,
5181     /* WINED3DSIH_ROUND_NE                         */ NULL,
5182     /* WINED3DSIH_ROUND_NI                         */ NULL,
5183     /* WINED3DSIH_ROUND_PI                         */ NULL,
5184     /* WINED3DSIH_ROUND_Z                          */ NULL,
5185     /* WINED3DSIH_RSQ                              */ shader_hw_scalar_op,
5186     /* WINED3DSIH_SAMPLE                           */ NULL,
5187     /* WINED3DSIH_SAMPLE_B                         */ NULL,
5188     /* WINED3DSIH_SAMPLE_C                         */ NULL,
5189     /* WINED3DSIH_SAMPLE_C_LZ                      */ NULL,
5190     /* WINED3DSIH_SAMPLE_GRAD                      */ NULL,
5191     /* WINED3DSIH_SAMPLE_INFO                      */ NULL,
5192     /* WINED3DSIH_SAMPLE_LOD                       */ NULL,
5193     /* WINED3DSIH_SAMPLE_POS                       */ NULL,
5194     /* WINED3DSIH_SETP                             */ NULL,
5195     /* WINED3DSIH_SGE                              */ shader_hw_map2gl,
5196     /* WINED3DSIH_SGN                              */ shader_hw_sgn,
5197     /* WINED3DSIH_SINCOS                           */ shader_hw_sincos,
5198     /* WINED3DSIH_SLT                              */ shader_hw_map2gl,
5199     /* WINED3DSIH_SQRT                             */ NULL,
5200     /* WINED3DSIH_STORE_RAW                        */ NULL,
5201     /* WINED3DSIH_STORE_STRUCTURED                 */ NULL,
5202     /* WINED3DSIH_STORE_UAV_TYPED                  */ NULL,
5203     /* WINED3DSIH_SUB                              */ shader_hw_map2gl,
5204     /* WINED3DSIH_SWAPC                            */ NULL,
5205     /* WINED3DSIH_SWITCH                           */ NULL,
5206     /* WINED3DSIH_SYNC                             */ NULL,
5207     /* WINED3DSIH_TEX                              */ pshader_hw_tex,
5208     /* WINED3DSIH_TEXBEM                           */ pshader_hw_texbem,
5209     /* WINED3DSIH_TEXBEML                          */ pshader_hw_texbem,
5210     /* WINED3DSIH_TEXCOORD                         */ pshader_hw_texcoord,
5211     /* WINED3DSIH_TEXDEPTH                         */ pshader_hw_texdepth,
5212     /* WINED3DSIH_TEXDP3                           */ pshader_hw_texdp3,
5213     /* WINED3DSIH_TEXDP3TEX                        */ pshader_hw_texdp3tex,
5214     /* WINED3DSIH_TEXKILL                          */ pshader_hw_texkill,
5215     /* WINED3DSIH_TEXLDD                           */ shader_hw_texldd,
5216     /* WINED3DSIH_TEXLDL                           */ shader_hw_texldl,
5217     /* WINED3DSIH_TEXM3x2DEPTH                     */ pshader_hw_texm3x2depth,
5218     /* WINED3DSIH_TEXM3x2PAD                       */ pshader_hw_texm3x2pad,
5219     /* WINED3DSIH_TEXM3x2TEX                       */ pshader_hw_texm3x2tex,
5220     /* WINED3DSIH_TEXM3x3                          */ pshader_hw_texm3x3,
5221     /* WINED3DSIH_TEXM3x3DIFF                      */ NULL,
5222     /* WINED3DSIH_TEXM3x3PAD                       */ pshader_hw_texm3x3pad,
5223     /* WINED3DSIH_TEXM3x3SPEC                      */ pshader_hw_texm3x3spec,
5224     /* WINED3DSIH_TEXM3x3TEX                       */ pshader_hw_texm3x3tex,
5225     /* WINED3DSIH_TEXM3x3VSPEC                     */ pshader_hw_texm3x3vspec,
5226     /* WINED3DSIH_TEXREG2AR                        */ pshader_hw_texreg2ar,
5227     /* WINED3DSIH_TEXREG2GB                        */ pshader_hw_texreg2gb,
5228     /* WINED3DSIH_TEXREG2RGB                       */ pshader_hw_texreg2rgb,
5229     /* WINED3DSIH_UBFE                             */ NULL,
5230     /* WINED3DSIH_UDIV                             */ NULL,
5231     /* WINED3DSIH_UGE                              */ NULL,
5232     /* WINED3DSIH_ULT                              */ NULL,
5233     /* WINED3DSIH_UMAX                             */ NULL,
5234     /* WINED3DSIH_UMIN                             */ NULL,
5235     /* WINED3DSIH_UMUL                             */ NULL,
5236     /* WINED3DSIH_USHR                             */ NULL,
5237     /* WINED3DSIH_UTOF                             */ NULL,
5238     /* WINED3DSIH_XOR                              */ NULL,
5239 };
5240 
5241 static BOOL get_bool_const(const struct wined3d_shader_instruction *ins,
5242         const struct wined3d_shader *shader, DWORD idx)
5243 {
5244     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5245     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5246     const struct wined3d_shader_lconst *constant;
5247     WORD bools = 0;
5248     WORD flag = (1u << idx);
5249     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5250 
5251     if (reg_maps->local_bool_consts & flag)
5252     {
5253         /* What good is an if(bool) with a hardcoded local constant? I don't know, but handle it */
5254         LIST_FOR_EACH_ENTRY(constant, &shader->constantsB, struct wined3d_shader_lconst, entry)
5255         {
5256             if (constant->idx == idx)
5257             {
5258                 return constant->value[0];
5259             }
5260         }
5261         ERR("Local constant not found\n");
5262         return FALSE;
5263     }
5264     else
5265     {
5266         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5267         else bools = priv->cur_ps_args->bools;
5268         return bools & flag;
5269     }
5270 }
5271 
5272 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5273         const struct wined3d_shader *shader, UINT idx, struct wined3d_shader_loop_control *loop_control)
5274 {
5275     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5276     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5277 
5278     /* Integer constants can either be a local constant, or they can be stored in the shader
5279      * type specific compile args. */
5280     if (reg_maps->local_int_consts & (1u << idx))
5281     {
5282         const struct wined3d_shader_lconst *constant;
5283 
5284         LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
5285         {
5286             if (constant->idx == idx)
5287             {
5288                 loop_control->count = constant->value[0];
5289                 loop_control->start = constant->value[1];
5290                 /* Step is signed. */
5291                 loop_control->step = (int)constant->value[2];
5292                 return;
5293             }
5294         }
5295         /* If this happens the flag was set incorrectly */
5296         ERR("Local constant not found\n");
5297         loop_control->count = 0;
5298         loop_control->start = 0;
5299         loop_control->step = 0;
5300         return;
5301     }
5302 
5303     switch (reg_maps->shader_version.type)
5304     {
5305         case WINED3D_SHADER_TYPE_VERTEX:
5306             /* Count and aL start value are unsigned */
5307             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5308             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5309             /* Step is signed. */
5310             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5311             break;
5312 
5313         case WINED3D_SHADER_TYPE_PIXEL:
5314             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5315             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5316             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5317             break;
5318 
5319         default:
5320             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5321             break;
5322     }
5323 }
5324 
5325 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5326 {
5327     struct wined3d_shader_src_param *src_param = NULL, *rel_addr;
5328     struct wined3d_shader_dst_param *dst_param;
5329     struct recorded_instruction *rec;
5330     unsigned int i;
5331 
5332     if (!(rec = heap_alloc_zero(sizeof(*rec))))
5333     {
5334         ERR("Out of memory\n");
5335         return;
5336     }
5337 
5338     rec->ins = *ins;
5339     if (!(dst_param = heap_alloc(sizeof(*dst_param))))
5340         goto free;
5341     *dst_param = *ins->dst;
5342     if (ins->dst->reg.idx[0].rel_addr)
5343     {
5344         if (!(rel_addr = heap_alloc(sizeof(*rel_addr))))
5345             goto free;
5346         *rel_addr = *ins->dst->reg.idx[0].rel_addr;
5347         dst_param->reg.idx[0].rel_addr = rel_addr;
5348     }
5349     rec->ins.dst = dst_param;
5350 
5351     if (!(src_param = heap_calloc(ins->src_count, sizeof(*src_param))))
5352         goto free;
5353     for (i = 0; i < ins->src_count; ++i)
5354     {
5355         src_param[i] = ins->src[i];
5356         if (ins->src[i].reg.idx[0].rel_addr)
5357         {
5358             if (!(rel_addr = heap_alloc(sizeof(*rel_addr))))
5359                 goto free;
5360             *rel_addr = *ins->src[i].reg.idx[0].rel_addr;
5361             src_param[i].reg.idx[0].rel_addr = rel_addr;
5362         }
5363     }
5364     rec->ins.src = src_param;
5365     list_add_tail(list, &rec->entry);
5366     return;
5367 
5368 free:
5369     ERR("Out of memory\n");
5370     if (dst_param)
5371     {
5372         heap_free((void *)dst_param->reg.idx[0].rel_addr);
5373         heap_free(dst_param);
5374     }
5375     if (src_param)
5376     {
5377         for (i = 0; i < ins->src_count; ++i)
5378         {
5379             heap_free((void *)src_param[i].reg.idx[0].rel_addr);
5380         }
5381         heap_free(src_param);
5382     }
5383     heap_free(rec);
5384 }
5385 
5386 static void free_recorded_instruction(struct list *list)
5387 {
5388     struct recorded_instruction *rec_ins, *entry2;
5389     unsigned int i;
5390 
5391     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5392     {
5393         list_remove(&rec_ins->entry);
5394         if (rec_ins->ins.dst)
5395         {
5396             heap_free((void *)rec_ins->ins.dst->reg.idx[0].rel_addr);
5397             heap_free((void *)rec_ins->ins.dst);
5398         }
5399         if (rec_ins->ins.src)
5400         {
5401             for (i = 0; i < rec_ins->ins.src_count; ++i)
5402             {
5403                 heap_free((void *)rec_ins->ins.src[i].reg.idx[0].rel_addr);
5404             }
5405             heap_free((void *)rec_ins->ins.src);
5406         }
5407         heap_free(rec_ins);
5408     }
5409 }
5410 
5411 static void pop_control_frame(const struct wined3d_shader_instruction *ins)
5412 {
5413     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5414     struct control_frame *control_frame;
5415 
5416     if (ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5417     {
5418         struct list *e = list_head(&priv->control_frames);
5419         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5420         list_remove(&control_frame->entry);
5421         heap_free(control_frame);
5422         priv->loop_depth--;
5423     }
5424     else if (ins->handler_idx == WINED3DSIH_ENDIF)
5425     {
5426         /* Non-ifc ENDIFs were already handled previously. */
5427         struct list *e = list_head(&priv->control_frames);
5428         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5429         list_remove(&control_frame->entry);
5430         heap_free(control_frame);
5431     }
5432 }
5433 
5434 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5435     SHADER_HANDLER hw_fct;
5436     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5437     const struct wined3d_shader *shader = ins->ctx->shader;
5438     struct control_frame *control_frame;
5439     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
5440     BOOL bool_const;
5441 
5442     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5443     {
5444         control_frame = heap_alloc_zero(sizeof(*control_frame));
5445         list_add_head(&priv->control_frames, &control_frame->entry);
5446 
5447         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5448         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5449 
5450         if(priv->target_version >= NV2)
5451         {
5452             control_frame->no.loop = priv->num_loops++;
5453             priv->loop_depth++;
5454         }
5455         else
5456         {
5457             /* Don't bother recording when we're in a not used if branch */
5458             if(priv->muted)
5459             {
5460                 return;
5461             }
5462 
5463             if(!priv->recording)
5464             {
5465                 list_init(&priv->record);
5466                 priv->recording = TRUE;
5467                 control_frame->outer_loop = TRUE;
5468                 get_loop_control_const(ins, shader, ins->src[0].reg.idx[0].offset, &control_frame->loop_control);
5469                 return; /* Instruction is handled */
5470             }
5471             /* Record this loop in the outer loop's recording */
5472         }
5473     }
5474     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5475     {
5476         if(priv->target_version >= NV2)
5477         {
5478             /* Nothing to do. The control frame is popped after the HW instr handler */
5479         }
5480         else
5481         {
5482             struct list *e = list_head(&priv->control_frames);
5483             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5484             list_remove(&control_frame->entry);
5485 
5486             if(control_frame->outer_loop)
5487             {
5488                 unsigned int iteration;
5489                 int aL = 0;
5490                 struct list copy;
5491 
5492                 /* Turn off recording before playback */
5493                 priv->recording = FALSE;
5494 
5495                 /* Move the recorded instructions to a separate list and get them out of the private data
5496                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5497                  * be recorded again, thus priv->record might be overwritten
5498                  */
5499                 list_init(&copy);
5500                 list_move_tail(&copy, &priv->record);
5501                 list_init(&priv->record);
5502 
5503                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5504                 {
5505                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5506                                    control_frame->loop_control.count, control_frame->loop_control.start,
5507                                    control_frame->loop_control.step);
5508                     aL = control_frame->loop_control.start;
5509                 }
5510                 else
5511                 {
5512                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5513                 }
5514 
5515                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5516                 {
5517                     struct recorded_instruction *rec_ins;
5518                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5519                     {
5520                         priv->aL = aL;
5521                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5522                     }
5523                     else
5524                     {
5525                         shader_addline(buffer, "#Iteration %u\n", iteration);
5526                     }
5527 
5528                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5529                     {
5530                         shader_arb_handle_instruction(&rec_ins->ins);
5531                     }
5532 
5533                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5534                     {
5535                         aL += control_frame->loop_control.step;
5536                     }
5537                 }
5538                 shader_addline(buffer, "#end loop/rep\n");
5539 
5540                 free_recorded_instruction(&copy);
5541                 heap_free(control_frame);
5542                 return; /* Instruction is handled */
5543             }
5544             else
5545             {
5546                 /* This is a nested loop. Proceed to the normal recording function */
5547                 heap_free(control_frame);
5548             }
5549         }
5550     }
5551 
5552     if(priv->recording)
5553     {
5554         record_instruction(&priv->record, ins);
5555         return;
5556     }
5557 
5558     /* boolean if */
5559     if(ins->handler_idx == WINED3DSIH_IF)
5560     {
5561         control_frame = heap_alloc_zero(sizeof(*control_frame));
5562         list_add_head(&priv->control_frames, &control_frame->entry);
5563         control_frame->type = IF;
5564 
5565         bool_const = get_bool_const(ins, shader, ins->src[0].reg.idx[0].offset);
5566         if (ins->src[0].modifiers == WINED3DSPSM_NOT)
5567             bool_const = !bool_const;
5568         if (!priv->muted && !bool_const)
5569         {
5570             shader_addline(buffer, "#if(FALSE){\n");
5571             priv->muted = TRUE;
5572             control_frame->muting = TRUE;
5573         }
5574         else shader_addline(buffer, "#if(TRUE) {\n");
5575 
5576         return; /* Instruction is handled */
5577     }
5578     else if(ins->handler_idx == WINED3DSIH_IFC)
5579     {
5580         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5581         control_frame = heap_alloc_zero(sizeof(*control_frame));
5582         control_frame->type = IFC;
5583         control_frame->no.ifc = priv->num_ifcs++;
5584         list_add_head(&priv->control_frames, &control_frame->entry);
5585     }
5586     else if(ins->handler_idx == WINED3DSIH_ELSE)
5587     {
5588         struct list *e = list_head(&priv->control_frames);
5589         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5590 
5591         if(control_frame->type == IF)
5592         {
5593             shader_addline(buffer, "#} else {\n");
5594             if(!priv->muted && !control_frame->muting)
5595             {
5596                 priv->muted = TRUE;
5597                 control_frame->muting = TRUE;
5598             }
5599             else if(control_frame->muting) priv->muted = FALSE;
5600             return; /* Instruction is handled. */
5601         }
5602         /* In case of an ifc, generate a HW shader instruction */
5603         if (control_frame->type != IFC)
5604             ERR("Control frame does not match.\n");
5605     }
5606     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5607     {
5608         struct list *e = list_head(&priv->control_frames);
5609         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5610 
5611         if(control_frame->type == IF)
5612         {
5613             shader_addline(buffer, "#} endif\n");
5614             if(control_frame->muting) priv->muted = FALSE;
5615             list_remove(&control_frame->entry);
5616             heap_free(control_frame);
5617             return; /* Instruction is handled */
5618         }
5619         /* In case of an ifc, generate a HW shader instruction */
5620         if (control_frame->type != IFC)
5621             ERR("Control frame does not match.\n");
5622     }
5623 
5624     if(priv->muted)
5625     {
5626         pop_control_frame(ins);
5627         return;
5628     }
5629 
5630     /* Select handler */
5631     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5632 
5633     /* Unhandled opcode */
5634     if (!hw_fct)
5635     {
5636         FIXME("Backend can't handle opcode %s.\n", debug_d3dshaderinstructionhandler(ins->handler_idx));
5637         return;
5638     }
5639     hw_fct(ins);
5640 
5641     pop_control_frame(ins);
5642 
5643     shader_arb_add_instruction_modifiers(ins);
5644 }
5645 
5646 static BOOL shader_arb_has_ffp_proj_control(void *shader_priv)
5647 {
5648     struct shader_arb_priv *priv = shader_priv;
5649 
5650     return priv->ffp_proj_control;
5651 }
5652 
5653 static void shader_arb_precompile(void *shader_priv, struct wined3d_shader *shader) {}
5654 
5655 const struct wined3d_shader_backend_ops arb_program_shader_backend =
5656 {
5657     shader_arb_handle_instruction,
5658     shader_arb_precompile,
5659     shader_arb_select,
5660     shader_arb_select_compute,
5661     shader_arb_disable,
5662     shader_arb_update_float_vertex_constants,
5663     shader_arb_update_float_pixel_constants,
5664     shader_arb_load_constants,
5665     shader_arb_destroy,
5666     shader_arb_alloc,
5667     shader_arb_free,
5668     shader_arb_allocate_context_data,
5669     shader_arb_free_context_data,
5670     shader_arb_init_context_state,
5671     shader_arb_get_caps,
5672     shader_arb_color_fixup_supported,
5673     shader_arb_has_ffp_proj_control,
5674 };
5675 
5676 /* ARB_fragment_program fixed function pipeline replacement definitions */
5677 #define ARB_FFP_CONST_TFACTOR           0
5678 #define ARB_FFP_CONST_COLOR_KEY_LOW     ((ARB_FFP_CONST_TFACTOR) + 1)
5679 #define ARB_FFP_CONST_COLOR_KEY_HIGH    ((ARB_FFP_CONST_COLOR_KEY_LOW) + 1)
5680 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_COLOR_KEY_HIGH) + 1)
5681 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5682 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5683 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5684 
5685 struct arbfp_ffp_desc
5686 {
5687     struct ffp_frag_desc parent;
5688     GLuint shader;
5689 };
5690 
5691 /* Context activation is done by the caller. */
5692 static void arbfp_enable(const struct wined3d_gl_info *gl_info, BOOL enable)
5693 {
5694     if (enable)
5695     {
5696         gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
5697         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5698     }
5699     else
5700     {
5701         gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
5702         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5703     }
5704 }
5705 
5706 static void *arbfp_alloc(const struct wined3d_shader_backend_ops *shader_backend, void *shader_priv)
5707 {
5708     struct shader_arb_priv *priv;
5709 
5710     /* Share private data between the shader backend and the pipeline
5711      * replacement, if both are the arb implementation. This is needed to
5712      * figure out whether ARBfp should be disabled if no pixel shader is bound
5713      * or not. */
5714     if (shader_backend == &arb_program_shader_backend)
5715         priv = shader_priv;
5716     else if (!(priv = heap_alloc_zero(sizeof(*priv))))
5717         return NULL;
5718 
5719     wine_rb_init(&priv->fragment_shaders, wined3d_ffp_frag_program_key_compare);
5720     priv->use_arbfp_fixed_func = TRUE;
5721 
5722     return priv;
5723 }
5724 
5725 /* Context activation is done by the caller. */
5726 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5727 {
5728     const struct wined3d_gl_info *gl_info = context;
5729     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5730 
5731     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5732     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5733     heap_free(entry_arb);
5734 }
5735 
5736 /* Context activation is done by the caller. */
5737 static void arbfp_free(struct wined3d_device *device)
5738 {
5739     struct shader_arb_priv *priv = device->fragment_priv;
5740 
5741     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5742     priv->use_arbfp_fixed_func = FALSE;
5743 
5744     if (device->shader_backend != &arb_program_shader_backend)
5745         heap_free(device->fragment_priv);
5746 }
5747 
5748 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5749 {
5750     caps->wined3d_caps = WINED3D_FRAGMENT_CAP_PROJ_CONTROL
5751             | WINED3D_FRAGMENT_CAP_SRGB_WRITE
5752             | WINED3D_FRAGMENT_CAP_COLOR_KEY;
5753     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5754     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5755                            WINED3DTEXOPCAPS_SELECTARG1                  |
5756                            WINED3DTEXOPCAPS_SELECTARG2                  |
5757                            WINED3DTEXOPCAPS_MODULATE4X                  |
5758                            WINED3DTEXOPCAPS_MODULATE2X                  |
5759                            WINED3DTEXOPCAPS_MODULATE                    |
5760                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5761                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5762                            WINED3DTEXOPCAPS_ADD                         |
5763                            WINED3DTEXOPCAPS_SUBTRACT                    |
5764                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5765                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5766                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5767                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5768                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5769                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5770                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5771                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5772                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5773                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5774                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5775                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5776                            WINED3DTEXOPCAPS_LERP                        |
5777                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5778                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5779 
5780     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5781 
5782     caps->MaxTextureBlendStages   = MAX_TEXTURES;
5783     caps->MaxSimultaneousTextures = min(gl_info->limits.samplers[WINED3D_SHADER_TYPE_PIXEL], MAX_TEXTURES);
5784 }
5785 
5786 static DWORD arbfp_get_emul_mask(const struct wined3d_gl_info *gl_info)
5787 {
5788     return GL_EXT_EMUL_ARB_MULTITEXTURE | GL_EXT_EMUL_EXT_FOG_COORD;
5789 }
5790 
5791 static void state_texfactor_arbfp(struct wined3d_context *context,
5792         const struct wined3d_state *state, DWORD state_id)
5793 {
5794     const struct wined3d_gl_info *gl_info = context->gl_info;
5795     struct wined3d_device *device = context->device;
5796     struct wined3d_color color;
5797 
5798     if (device->shader_backend == &arb_program_shader_backend)
5799     {
5800         struct shader_arb_priv *priv;
5801 
5802         /* Don't load the parameter if we're using an arbfp pixel shader,
5803          * otherwise we'll overwrite application provided constants. */
5804         if (use_ps(state))
5805             return;
5806 
5807         priv = device->shader_priv;
5808         priv->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5809         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5810     }
5811 
5812     wined3d_color_from_d3dcolor(&color, state->render_states[WINED3D_RS_TEXTUREFACTOR]);
5813     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, &color.r));
5814     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, &color.r)");
5815 }
5816 
5817 static void state_tss_constant_arbfp(struct wined3d_context *context,
5818         const struct wined3d_state *state, DWORD state_id)
5819 {
5820     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5821     const struct wined3d_gl_info *gl_info = context->gl_info;
5822     struct wined3d_device *device = context->device;
5823     struct wined3d_color color;
5824 
5825     if (device->shader_backend == &arb_program_shader_backend)
5826     {
5827         struct shader_arb_priv *priv;
5828 
5829         /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5830          * application provided constants.
5831          */
5832         if (use_ps(state))
5833             return;
5834 
5835         priv = device->shader_priv;
5836         priv->pshader_const_dirty[ARB_FFP_CONST_CONSTANT(stage)] = 1;
5837         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_CONSTANT(stage) + 1);
5838     }
5839 
5840     wined3d_color_from_d3dcolor(&color, state->texture_states[stage][WINED3D_TSS_CONSTANT]);
5841     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_CONSTANT(stage), &color.r));
5842     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_CONSTANT(stage), &color.r)");
5843 }
5844 
5845 static void state_arb_specularenable(struct wined3d_context *context,
5846         const struct wined3d_state *state, DWORD state_id)
5847 {
5848     const struct wined3d_gl_info *gl_info = context->gl_info;
5849     struct wined3d_device *device = context->device;
5850     float col[4];
5851 
5852     if (device->shader_backend == &arb_program_shader_backend)
5853     {
5854         struct shader_arb_priv *priv;
5855 
5856         /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5857          * application provided constants.
5858          */
5859         if (use_ps(state))
5860             return;
5861 
5862         priv = device->shader_priv;
5863         priv->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5864         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5865     }
5866 
5867     if (state->render_states[WINED3D_RS_SPECULARENABLE])
5868     {
5869         /* The specular color has no alpha */
5870         col[0] = 1.0f; col[1] = 1.0f;
5871         col[2] = 1.0f; col[3] = 0.0f;
5872     } else {
5873         col[0] = 0.0f; col[1] = 0.0f;
5874         col[2] = 0.0f; col[3] = 0.0f;
5875     }
5876     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5877     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5878 }
5879 
5880 static void set_bumpmat_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5881 {
5882     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5883     const struct wined3d_gl_info *gl_info = context->gl_info;
5884     struct wined3d_device *device = context->device;
5885     float mat[2][2];
5886 
5887     context->constant_update_mask |= WINED3D_SHADER_CONST_PS_BUMP_ENV;
5888 
5889     if (device->shader_backend == &arb_program_shader_backend)
5890     {
5891         struct shader_arb_priv *priv = device->shader_priv;
5892 
5893         /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants. */
5894         if (use_ps(state))
5895             return;
5896 
5897         priv->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5898         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5899     }
5900 
5901     mat[0][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT00]);
5902     mat[0][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT01]);
5903     mat[1][0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT10]);
5904     mat[1][1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_MAT11]);
5905 
5906     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5907     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5908 }
5909 
5910 static void tex_bumpenvlum_arbfp(struct wined3d_context *context,
5911         const struct wined3d_state *state, DWORD state_id)
5912 {
5913     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5914     const struct wined3d_gl_info *gl_info = context->gl_info;
5915     struct wined3d_device *device = context->device;
5916     float param[4];
5917 
5918     context->constant_update_mask |= WINED3D_SHADER_CONST_PS_BUMP_ENV;
5919 
5920     if (device->shader_backend == &arb_program_shader_backend)
5921     {
5922         struct shader_arb_priv *priv = device->shader_priv;
5923 
5924         /* Exit now, don't set the luminance below, otherwise we may overwrite pixel shader constants. */
5925         if (use_ps(state))
5926             return;
5927 
5928         priv->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5929         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5930     }
5931 
5932     param[0] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LSCALE]);
5933     param[1] = *((float *)&state->texture_states[stage][WINED3D_TSS_BUMPENV_LOFFSET]);
5934     param[2] = 0.0f;
5935     param[3] = 0.0f;
5936 
5937     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5938     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5939 }
5940 
5941 static void alpha_test_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5942 {
5943     const struct wined3d_gl_info *gl_info = context->gl_info;
5944     int glParm;
5945     float ref;
5946 
5947     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
5948 
5949     if (state->render_states[WINED3D_RS_ALPHATESTENABLE])
5950     {
5951         gl_info->gl_ops.gl.p_glEnable(GL_ALPHA_TEST);
5952         checkGLcall("glEnable GL_ALPHA_TEST");
5953     }
5954     else
5955     {
5956         gl_info->gl_ops.gl.p_glDisable(GL_ALPHA_TEST);
5957         checkGLcall("glDisable GL_ALPHA_TEST");
5958         return;
5959     }
5960 
5961     ref = ((float)state->render_states[WINED3D_RS_ALPHAREF]) / 255.0f;
5962     glParm = wined3d_gl_compare_func(state->render_states[WINED3D_RS_ALPHAFUNC]);
5963 
5964     if (glParm)
5965     {
5966         gl_info->gl_ops.gl.p_glAlphaFunc(glParm, ref);
5967         checkGLcall("glAlphaFunc");
5968     }
5969 }
5970 
5971 static void color_key_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
5972 {
5973     const struct wined3d_texture *texture = state->textures[0];
5974     const struct wined3d_gl_info *gl_info = context->gl_info;
5975     struct wined3d_device *device = context->device;
5976     struct wined3d_color float_key[2];
5977 
5978     if (!texture)
5979         return;
5980 
5981     if (device->shader_backend == &arb_program_shader_backend)
5982     {
5983         struct shader_arb_priv *priv;
5984 
5985         /* Don't load the parameter if we're using an arbfp pixel shader,
5986          * otherwise we'll overwrite application provided constants. */
5987         if (use_ps(state))
5988             return;
5989 
5990         priv = device->shader_priv;
5991         priv->pshader_const_dirty[ARB_FFP_CONST_COLOR_KEY_LOW] = 1;
5992         priv->pshader_const_dirty[ARB_FFP_CONST_COLOR_KEY_HIGH] = 1;
5993         priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, ARB_FFP_CONST_COLOR_KEY_HIGH + 1);
5994     }
5995 
5996     wined3d_format_get_float_color_key(texture->resource.format, &texture->async.src_blt_color_key, float_key);
5997 
5998     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_COLOR_KEY_LOW, &float_key[0].r));
5999     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_COLOR_KEY_LOW, &float_key[0].r)");
6000     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_COLOR_KEY_HIGH, &float_key[1].r));
6001     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_COLOR_KEY_HIGH, &float_key[1].r)");
6002 }
6003 
6004 static const char *get_argreg(struct wined3d_string_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
6005 {
6006     const char *ret;
6007 
6008     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
6009 
6010     switch(arg & WINED3DTA_SELECTMASK) {
6011         case WINED3DTA_DIFFUSE:
6012             ret = "fragment.color.primary"; break;
6013 
6014         case WINED3DTA_CURRENT:
6015             ret = "ret";
6016             break;
6017 
6018         case WINED3DTA_TEXTURE:
6019             switch(stage) {
6020                 case 0: ret = "tex0"; break;
6021                 case 1: ret = "tex1"; break;
6022                 case 2: ret = "tex2"; break;
6023                 case 3: ret = "tex3"; break;
6024                 case 4: ret = "tex4"; break;
6025                 case 5: ret = "tex5"; break;
6026                 case 6: ret = "tex6"; break;
6027                 case 7: ret = "tex7"; break;
6028                 default: ret = "unknown texture";
6029             }
6030             break;
6031 
6032         case WINED3DTA_TFACTOR:
6033             ret = "tfactor"; break;
6034 
6035         case WINED3DTA_SPECULAR:
6036             ret = "fragment.color.secondary"; break;
6037 
6038         case WINED3DTA_TEMP:
6039             ret = "tempreg"; break;
6040 
6041         case WINED3DTA_CONSTANT:
6042             switch(stage) {
6043                 case 0: ret = "const0"; break;
6044                 case 1: ret = "const1"; break;
6045                 case 2: ret = "const2"; break;
6046                 case 3: ret = "const3"; break;
6047                 case 4: ret = "const4"; break;
6048                 case 5: ret = "const5"; break;
6049                 case 6: ret = "const6"; break;
6050                 case 7: ret = "const7"; break;
6051                 default: ret = "unknown constant";
6052             }
6053             break;
6054 
6055         default:
6056             return "unknown";
6057     }
6058 
6059     if(arg & WINED3DTA_COMPLEMENT) {
6060         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
6061         if(argnum == 0) ret = "arg0";
6062         if(argnum == 1) ret = "arg1";
6063         if(argnum == 2) ret = "arg2";
6064     }
6065     if(arg & WINED3DTA_ALPHAREPLICATE) {
6066         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
6067         if(argnum == 0) ret = "arg0";
6068         if(argnum == 1) ret = "arg1";
6069         if(argnum == 2) ret = "arg2";
6070     }
6071     return ret;
6072 }
6073 
6074 static void gen_ffp_instr(struct wined3d_string_buffer *buffer, unsigned int stage, BOOL color,
6075         BOOL alpha, BOOL tmp_dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
6076 {
6077     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
6078     unsigned int mul = 1;
6079 
6080     if (color && alpha)
6081         dstmask = "";
6082     else if (color)
6083         dstmask = ".xyz";
6084     else
6085         dstmask = ".w";
6086 
6087     dstreg = tmp_dst ? "tempreg" : "ret";
6088 
6089     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
6090     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
6091     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
6092 
6093     switch (op)
6094     {
6095         case WINED3D_TOP_DISABLE:
6096             break;
6097 
6098         case WINED3D_TOP_SELECT_ARG2:
6099             arg1 = arg2;
6100             /* FALLTHROUGH */
6101         case WINED3D_TOP_SELECT_ARG1:
6102             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
6103             break;
6104 
6105         case WINED3D_TOP_MODULATE_4X:
6106             mul = 2;
6107             /* FALLTHROUGH */
6108         case WINED3D_TOP_MODULATE_2X:
6109             mul *= 2;
6110             /* FALLTHROUGH */
6111         case WINED3D_TOP_MODULATE:
6112             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6113             break;
6114 
6115         case WINED3D_TOP_ADD_SIGNED_2X:
6116             mul = 2;
6117             /* FALLTHROUGH */
6118         case WINED3D_TOP_ADD_SIGNED:
6119             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6120             arg2 = "arg2";
6121             /* FALLTHROUGH */
6122         case WINED3D_TOP_ADD:
6123             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6124             break;
6125 
6126         case WINED3D_TOP_SUBTRACT:
6127             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
6128             break;
6129 
6130         case WINED3D_TOP_ADD_SMOOTH:
6131             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
6132             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6133             break;
6134 
6135         case WINED3D_TOP_BLEND_CURRENT_ALPHA:
6136             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
6137             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6138             break;
6139         case WINED3D_TOP_BLEND_FACTOR_ALPHA:
6140             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
6141             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6142             break;
6143         case WINED3D_TOP_BLEND_TEXTURE_ALPHA:
6144             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
6145             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6146             break;
6147         case WINED3D_TOP_BLEND_DIFFUSE_ALPHA:
6148             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
6149             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6150             break;
6151 
6152         case WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM:
6153             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
6154             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
6155             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
6156             break;
6157 
6158         /* D3DTOP_PREMODULATE ???? */
6159 
6160         case WINED3D_TOP_MODULATE_INVALPHA_ADD_COLOR:
6161             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
6162             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
6163             break;
6164         case WINED3D_TOP_MODULATE_ALPHA_ADD_COLOR:
6165             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
6166             break;
6167         case WINED3D_TOP_MODULATE_INVCOLOR_ADD_ALPHA:
6168             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
6169             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
6170             break;
6171         case WINED3D_TOP_MODULATE_COLOR_ADD_ALPHA:
6172             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
6173             break;
6174 
6175         case WINED3D_TOP_DOTPRODUCT3:
6176             mul = 4;
6177             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
6178             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
6179             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
6180             break;
6181 
6182         case WINED3D_TOP_MULTIPLY_ADD:
6183             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
6184             break;
6185 
6186         case WINED3D_TOP_LERP:
6187             /* The msdn is not quite right here */
6188             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
6189             break;
6190 
6191         case WINED3D_TOP_BUMPENVMAP:
6192         case WINED3D_TOP_BUMPENVMAP_LUMINANCE:
6193             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
6194             break;
6195 
6196         default:
6197             FIXME("Unhandled texture op %08x\n", op);
6198     }
6199 
6200     if (mul == 2)
6201         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", dstreg, dstmask, dstreg);
6202     else if (mul == 4)
6203         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", dstreg, dstmask, dstreg);
6204 }
6205 
6206 static const char *arbfp_texture_target(enum wined3d_gl_resource_type type)
6207 {
6208     switch(type)
6209     {
6210         case WINED3D_GL_RES_TYPE_TEX_1D:
6211             return "1D";
6212         case WINED3D_GL_RES_TYPE_TEX_2D:
6213             return "2D";
6214         case WINED3D_GL_RES_TYPE_TEX_3D:
6215             return "3D";
6216         case WINED3D_GL_RES_TYPE_TEX_CUBE:
6217             return "CUBE";
6218         case WINED3D_GL_RES_TYPE_TEX_RECT:
6219             return "RECT";
6220         default:
6221             return "unexpected_resource_type";
6222     }
6223 }
6224 
6225 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, const struct wined3d_gl_info *gl_info)
6226 {
6227     BYTE tex_read = 0, bump_used = 0, luminance_used = 0, constant_used = 0;
6228     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
6229     unsigned int stage, lowest_disabled_stage;
6230     struct wined3d_string_buffer buffer;
6231     struct color_fixup_masks masks;
6232     BOOL custom_linear_fog = FALSE;
6233     const char *textype, *instr;
6234     DWORD arg0, arg1, arg2;
6235     char colorcor_dst[8];
6236     BOOL op_equal;
6237     GLuint ret;
6238 
6239     if (!string_buffer_init(&buffer))
6240     {
6241         ERR("Failed to initialize shader buffer.\n");
6242         return 0;
6243     }
6244 
6245     shader_addline(&buffer, "!!ARBfp1.0\n");
6246 
6247     if (settings->color_key_enabled)
6248     {
6249         shader_addline(&buffer, "PARAM color_key_low = program.env[%u];\n", ARB_FFP_CONST_COLOR_KEY_LOW);
6250         shader_addline(&buffer, "PARAM color_key_high = program.env[%u];\n", ARB_FFP_CONST_COLOR_KEY_HIGH);
6251         tex_read |= 1;
6252     }
6253 
6254     /* Find out which textures are read */
6255     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6256     {
6257         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6258             break;
6259 
6260         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
6261         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
6262         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
6263 
6264         if (arg0 == WINED3DTA_TEXTURE || arg1 == WINED3DTA_TEXTURE || arg2 == WINED3DTA_TEXTURE)
6265             tex_read |= 1u << stage;
6266         if (settings->op[stage].tmp_dst)
6267             tempreg_used = TRUE;
6268         if (arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP)
6269             tempreg_used = TRUE;
6270         if (arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR)
6271             tfactor_used = TRUE;
6272         if (arg0 == WINED3DTA_CONSTANT || arg1 == WINED3DTA_CONSTANT || arg2 == WINED3DTA_CONSTANT)
6273             constant_used |= 1u << stage;
6274 
6275         switch (settings->op[stage].cop)
6276         {
6277             case WINED3D_TOP_BUMPENVMAP_LUMINANCE:
6278                 luminance_used |= 1u << stage;
6279                 /* fall through */
6280             case WINED3D_TOP_BUMPENVMAP:
6281                 bump_used |= 1u << stage;
6282                 /* fall through */
6283             case WINED3D_TOP_BLEND_TEXTURE_ALPHA:
6284             case WINED3D_TOP_BLEND_TEXTURE_ALPHA_PM:
6285                 tex_read |= 1u << stage;
6286                 break;
6287 
6288             case WINED3D_TOP_BLEND_FACTOR_ALPHA:
6289                 tfactor_used = TRUE;
6290                 break;
6291 
6292             default:
6293                 break;
6294         }
6295 
6296         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6297             continue;
6298 
6299         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
6300         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
6301         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
6302 
6303         if (arg0 == WINED3DTA_TEXTURE || arg1 == WINED3DTA_TEXTURE || arg2 == WINED3DTA_TEXTURE)
6304             tex_read |= 1u << stage;
6305         if (arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP)
6306             tempreg_used = TRUE;
6307         if (arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR)
6308             tfactor_used = TRUE;
6309         if (arg0 == WINED3DTA_CONSTANT || arg1 == WINED3DTA_CONSTANT || arg2 == WINED3DTA_CONSTANT)
6310             constant_used |= 1u << stage;
6311     }
6312     lowest_disabled_stage = stage;
6313 
6314     switch (settings->fog)
6315     {
6316         case WINED3D_FFP_PS_FOG_OFF:                                                         break;
6317         case WINED3D_FFP_PS_FOG_LINEAR:
6318             if (gl_info->quirks & WINED3D_QUIRK_BROKEN_ARB_FOG)
6319             {
6320                 custom_linear_fog = TRUE;
6321                 break;
6322             }
6323             shader_addline(&buffer, "OPTION ARB_fog_linear;\n");
6324             break;
6325 
6326         case WINED3D_FFP_PS_FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
6327         case WINED3D_FFP_PS_FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
6328         default: FIXME("Unexpected fog setting %d\n", settings->fog);
6329     }
6330 
6331     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
6332     shader_addline(&buffer, "TEMP TMP;\n");
6333     shader_addline(&buffer, "TEMP ret;\n");
6334     if (tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
6335     shader_addline(&buffer, "TEMP arg0;\n");
6336     shader_addline(&buffer, "TEMP arg1;\n");
6337     shader_addline(&buffer, "TEMP arg2;\n");
6338     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6339     {
6340         if (constant_used & (1u << stage))
6341             shader_addline(&buffer, "PARAM const%u = program.env[%u];\n", stage, ARB_FFP_CONST_CONSTANT(stage));
6342 
6343         if (!(tex_read & (1u << stage)))
6344             continue;
6345 
6346         shader_addline(&buffer, "TEMP tex%u;\n", stage);
6347 
6348         if (!(bump_used & (1u << stage)))
6349             continue;
6350         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
6351 
6352         if (!(luminance_used & (1u << stage)))
6353             continue;
6354         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
6355     }
6356     if (tfactor_used)
6357         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
6358     shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6359 
6360     if (settings->sRGB_write)
6361     {
6362         shader_addline(&buffer, "PARAM srgb_consts0 = ");
6363         shader_arb_append_imm_vec4(&buffer, wined3d_srgb_const0);
6364         shader_addline(&buffer, ";\n");
6365         shader_addline(&buffer, "PARAM srgb_consts1 = ");
6366         shader_arb_append_imm_vec4(&buffer, wined3d_srgb_const1);
6367         shader_addline(&buffer, ";\n");
6368     }
6369 
6370     if (lowest_disabled_stage < 7 && settings->emul_clipplanes)
6371         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6372 
6373     if (tempreg_used || settings->sRGB_write)
6374         shader_addline(&buffer, "MOV tempreg, 0.0;\n");
6375 
6376     /* Generate texture sampling instructions */
6377     for (stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3D_TOP_DISABLE; ++stage)
6378     {
6379         if (!(tex_read & (1u << stage)))
6380             continue;
6381 
6382         textype = arbfp_texture_target(settings->op[stage].tex_type);
6383 
6384         if (settings->op[stage].projected == WINED3D_PROJECTION_NONE)
6385         {
6386             instr = "TEX";
6387         }
6388         else if (settings->op[stage].projected == WINED3D_PROJECTION_COUNT4
6389                 || settings->op[stage].projected == WINED3D_PROJECTION_COUNT3)
6390         {
6391             instr = "TXP";
6392         }
6393         else
6394         {
6395             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6396             instr = "TXP";
6397         }
6398 
6399         if (stage > 0
6400                 && (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP
6401                 || settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE))
6402         {
6403             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6404             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6405             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6406             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6407 
6408             /* With projective textures, texbem only divides the static
6409              * texture coordinate, not the displacement, so multiply the
6410              * displacement with the dividing parameter before passing it to
6411              * TXP. */
6412             if (settings->op[stage].projected != WINED3D_PROJECTION_NONE)
6413             {
6414                 if (settings->op[stage].projected == WINED3D_PROJECTION_COUNT4)
6415                 {
6416                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6417                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n",
6418                             stage, stage);
6419                 }
6420                 else
6421                 {
6422                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6423                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n",
6424                             stage, stage);
6425                 }
6426             }
6427             else
6428             {
6429                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6430             }
6431 
6432             shader_addline(&buffer, "%s tex%u, ret, texture[%u], %s;\n",
6433                     instr, stage, stage, textype);
6434             if (settings->op[stage - 1].cop == WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6435             {
6436                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6437                                stage - 1, stage - 1, stage - 1);
6438                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6439             }
6440         }
6441         else if (settings->op[stage].projected == WINED3D_PROJECTION_COUNT3)
6442         {
6443             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6444             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6445             shader_addline(&buffer, "%s tex%u, ret, texture[%u], %s;\n",
6446                             instr, stage, stage, textype);
6447         }
6448         else
6449         {
6450             shader_addline(&buffer, "%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6451                             instr, stage, stage, stage, textype);
6452         }
6453 
6454         sprintf(colorcor_dst, "tex%u", stage);
6455         masks = calc_color_correction(settings->op[stage].color_fixup, WINED3DSP_WRITEMASK_ALL);
6456         gen_color_correction(&buffer, colorcor_dst, colorcor_dst, "const.x", "const.y",
6457                 settings->op[stage].color_fixup, masks);
6458     }
6459 
6460     if (settings->color_key_enabled)
6461     {
6462         shader_addline(&buffer, "SLT TMP, tex0, color_key_low;\n"); /* below low key */
6463         shader_addline(&buffer, "SGE ret, tex0, color_key_high;\n"); /* above high key */
6464         shader_addline(&buffer, "ADD TMP, TMP, ret;\n"); /* or */
6465         shader_addline(&buffer, "DP4 TMP.b, TMP, TMP;\n"); /* on any channel */
6466         shader_addline(&buffer, "SGE TMP, -TMP.b, 0.0;\n"); /* logical not */
6467         shader_addline(&buffer, "KIL -TMP;\n"); /* discard if true */
6468     }
6469 
6470     shader_addline(&buffer, "MOV ret, fragment.color.primary;\n");
6471 
6472     /* Generate the main shader */
6473     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6474     {
6475         if (settings->op[stage].cop == WINED3D_TOP_DISABLE)
6476             break;
6477 
6478         if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6479                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6480             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6481         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG1
6482                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6483             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6484         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6485                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG1)
6486             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6487         else if (settings->op[stage].cop == WINED3D_TOP_SELECT_ARG2
6488                 && settings->op[stage].aop == WINED3D_TOP_SELECT_ARG2)
6489             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6490         else
6491             op_equal = settings->op[stage].aop   == settings->op[stage].cop
6492                     && settings->op[stage].carg0 == settings->op[stage].aarg0
6493                     && settings->op[stage].carg1 == settings->op[stage].aarg1
6494                     && settings->op[stage].carg2 == settings->op[stage].aarg2;
6495 
6496         if (settings->op[stage].aop == WINED3D_TOP_DISABLE)
6497         {
6498             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].tmp_dst,
6499                           settings->op[stage].cop, settings->op[stage].carg0,
6500                           settings->op[stage].carg1, settings->op[stage].carg2);
6501         }
6502         else if (op_equal)
6503         {
6504             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].tmp_dst,
6505                           settings->op[stage].cop, settings->op[stage].carg0,
6506                           settings->op[stage].carg1, settings->op[stage].carg2);
6507         }
6508         else if (settings->op[stage].cop != WINED3D_TOP_BUMPENVMAP
6509                 && settings->op[stage].cop != WINED3D_TOP_BUMPENVMAP_LUMINANCE)
6510         {
6511             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].tmp_dst,
6512                           settings->op[stage].cop, settings->op[stage].carg0,
6513                           settings->op[stage].carg1, settings->op[stage].carg2);
6514             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].tmp_dst,
6515                           settings->op[stage].aop, settings->op[stage].aarg0,
6516                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6517         }
6518     }
6519 
6520     if (settings->sRGB_write || custom_linear_fog)
6521     {
6522         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, ret;\n");
6523         if (settings->sRGB_write)
6524             arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6525         if (custom_linear_fog)
6526             arbfp_add_linear_fog(&buffer, "ret", "arg0");
6527         shader_addline(&buffer, "MOV result.color, ret;\n");
6528     }
6529     else
6530     {
6531         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, ret;\n");
6532     }
6533 
6534     /* Footer */
6535     shader_addline(&buffer, "END\n");
6536 
6537     /* Generate the shader */
6538     GL_EXTCALL(glGenProgramsARB(1, &ret));
6539     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6540     shader_arb_compile(gl_info, GL_FRAGMENT_PROGRAM_ARB, buffer.buffer);
6541 
6542     string_buffer_free(&buffer);
6543     return ret;
6544 }
6545 
6546 static void fragment_prog_arbfp(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6547 {
6548     const struct wined3d_gl_info *gl_info = context->gl_info;
6549     const struct wined3d_device *device = context->device;
6550     struct shader_arb_priv *priv = device->fragment_priv;
6551     BOOL use_pshader = use_ps(state);
6552     struct ffp_frag_settings settings;
6553     const struct arbfp_ffp_desc *desc;
6554     unsigned int i;
6555 
6556     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6557 
6558     if (isStateDirty(context, STATE_RENDER(WINED3D_RS_FOGENABLE)))
6559     {
6560         if (!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6561         {
6562             /* Reload fixed function constants since they collide with the
6563              * pixel shader constants. */
6564             for (i = 0; i < MAX_TEXTURES; ++i)
6565             {
6566                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6567                 state_tss_constant_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_CONSTANT));
6568             }
6569             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6570             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6571             color_key_arbfp(context, state, STATE_COLOR_KEY);
6572         }
6573         else if (use_pshader)
6574         {
6575             context->shader_update_mask |= 1u << WINED3D_SHADER_TYPE_PIXEL;
6576         }
6577         return;
6578     }
6579 
6580     if (!use_pshader)
6581     {
6582         /* Find or create a shader implementing the fixed function pipeline
6583          * settings, then activate it. */
6584         gen_ffp_frag_op(context, state, &settings, FALSE);
6585         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6586         if (!desc)
6587         {
6588             struct arbfp_ffp_desc *new_desc;
6589 
6590             if (!(new_desc = heap_alloc(sizeof(*new_desc))))
6591             {
6592                 ERR("Out of memory\n");
6593                 return;
6594             }
6595 
6596             new_desc->parent.settings = settings;
6597             new_desc->shader = gen_arbfp_ffp_shader(&settings, gl_info);
6598             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6599             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6600             desc = new_desc;
6601         }
6602 
6603         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active (however, note the
6604          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6605          * deactivate it.
6606          */
6607         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6608         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6609         priv->current_fprogram_id = desc->shader;
6610 
6611         if (device->shader_backend == &arb_program_shader_backend && context->last_was_pshader)
6612         {
6613             /* Reload fixed function constants since they collide with the
6614              * pixel shader constants. */
6615             for (i = 0; i < MAX_TEXTURES; ++i)
6616             {
6617                 set_bumpmat_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_BUMPENV_MAT00));
6618                 state_tss_constant_arbfp(context, state, STATE_TEXTURESTAGE(i, WINED3D_TSS_CONSTANT));
6619             }
6620             state_texfactor_arbfp(context, state, STATE_RENDER(WINED3D_RS_TEXTUREFACTOR));
6621             state_arb_specularenable(context, state, STATE_RENDER(WINED3D_RS_SPECULARENABLE));
6622             color_key_arbfp(context, state, STATE_COLOR_KEY);
6623         }
6624         context->last_was_pshader = FALSE;
6625     }
6626     else if (!context->last_was_pshader)
6627     {
6628         if (device->shader_backend == &arb_program_shader_backend)
6629             context->constant_update_mask |= WINED3D_SHADER_CONST_PS_F;
6630         context->last_was_pshader = TRUE;
6631     }
6632 
6633     context->shader_update_mask |= 1u << WINED3D_SHADER_TYPE_PIXEL;
6634 }
6635 
6636 /* We can't link the fog states to the fragment state directly since the
6637  * vertex pipeline links them to FOGENABLE. A different linking in different
6638  * pipeline parts can't be expressed in the combined state table, so we need
6639  * to handle that with a forwarding function. The other invisible side effect
6640  * is that changing the fog start and fog end (which links to FOGENABLE in
6641  * vertex) results in the fragment_prog_arbfp function being called because
6642  * FOGENABLE is dirty, which calls this function here. */
6643 static void state_arbfp_fog(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6644 {
6645     enum fogsource new_source;
6646     DWORD fogstart = state->render_states[WINED3D_RS_FOGSTART];
6647     DWORD fogend = state->render_states[WINED3D_RS_FOGEND];
6648 
6649     TRACE("context %p, state %p, state_id %#x.\n", context, state, state_id);
6650 
6651     if (!isStateDirty(context, STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL)))
6652         fragment_prog_arbfp(context, state, state_id);
6653 
6654     if (!state->render_states[WINED3D_RS_FOGENABLE])
6655         return;
6656 
6657     if (state->render_states[WINED3D_RS_FOGTABLEMODE] == WINED3D_FOG_NONE)
6658     {
6659         if (use_vs(state))
6660         {
6661             new_source = FOGSOURCE_VS;
6662         }
6663         else
6664         {
6665             if (state->render_states[WINED3D_RS_FOGVERTEXMODE] == WINED3D_FOG_NONE || context->last_was_rhw)
6666                 new_source = FOGSOURCE_COORD;
6667             else
6668                 new_source = FOGSOURCE_FFP;
6669         }
6670     }
6671     else
6672     {
6673         new_source = FOGSOURCE_FFP;
6674     }
6675 
6676     if (new_source != context->fog_source || fogstart == fogend)
6677     {
6678         context->fog_source = new_source;
6679         state_fogstartend(context, state, STATE_RENDER(WINED3D_RS_FOGSTART));
6680     }
6681 }
6682 
6683 static void textransform(struct wined3d_context *context, const struct wined3d_state *state, DWORD state_id)
6684 {
6685     if (!isStateDirty(context, STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL)))
6686         fragment_prog_arbfp(context, state, state_id);
6687 }
6688 
6689 static const struct StateEntryTemplate arbfp_fragmentstate_template[] =
6690 {
6691     {STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),              { STATE_RENDER(WINED3D_RS_TEXTUREFACTOR),             state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6692     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6693     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6694     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6695     {STATE_TEXTURESTAGE(0, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6696     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6697     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6698     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6699     {STATE_TEXTURESTAGE(0, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6700     {STATE_TEXTURESTAGE(0, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6701     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6702     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6703     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6704     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6705     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6706     {STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(0, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6707     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6708     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6709     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6710     {STATE_TEXTURESTAGE(1, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6711     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6712     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6713     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6714     {STATE_TEXTURESTAGE(1, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6715     {STATE_TEXTURESTAGE(1, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6716     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6717     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6718     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6719     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6720     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6721     {STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(1, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6722     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6723     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6724     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6725     {STATE_TEXTURESTAGE(2, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6726     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6727     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6728     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6729     {STATE_TEXTURESTAGE(2, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6730     {STATE_TEXTURESTAGE(2, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6731     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6732     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6733     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6734     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6735     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6736     {STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(2, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6737     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6738     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6739     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6740     {STATE_TEXTURESTAGE(3, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6741     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6742     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6743     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6744     {STATE_TEXTURESTAGE(3, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6745     {STATE_TEXTURESTAGE(3, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6746     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6747     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6748     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6749     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6750     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6751     {STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(3, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6752     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6753     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6754     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6755     {STATE_TEXTURESTAGE(4, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6756     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6757     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6758     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6759     {STATE_TEXTURESTAGE(4, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6760     {STATE_TEXTURESTAGE(4, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6761     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6762     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6763     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6764     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6765     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6766     {STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(4, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6767     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6768     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6769     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6770     {STATE_TEXTURESTAGE(5, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6771     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6772     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6773     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6774     {STATE_TEXTURESTAGE(5, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6775     {STATE_TEXTURESTAGE(5, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6776     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6777     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6778     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6779     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6780     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6781     {STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(5, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6782     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6783     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6784     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6785     {STATE_TEXTURESTAGE(6, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6786     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6787     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6788     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6789     {STATE_TEXTURESTAGE(6, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6790     {STATE_TEXTURESTAGE(6, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6791     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6792     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6793     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6794     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6795     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6796     {STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(6, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6797     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6798     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6799     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6800     {STATE_TEXTURESTAGE(7, WINED3D_TSS_COLOR_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6801     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_OP),         { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6802     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG1),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6803     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG2),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6804     {STATE_TEXTURESTAGE(7, WINED3D_TSS_ALPHA_ARG0),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6805     {STATE_TEXTURESTAGE(7, WINED3D_TSS_RESULT_ARG),       { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6806     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6807     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT01),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6808     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT10),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6809     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT11),    { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_MAT00),   NULL                    }, WINED3D_GL_EXT_NONE             },
6810     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),   { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6811     {STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LOFFSET),  { STATE_TEXTURESTAGE(7, WINED3D_TSS_BUMPENV_LSCALE),  NULL                    }, WINED3D_GL_EXT_NONE             },
6812     {STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),             { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6813     {STATE_RENDER(WINED3D_RS_ALPHAFUNC),                  { STATE_RENDER(WINED3D_RS_ALPHATESTENABLE),           NULL                    }, WINED3D_GL_EXT_NONE             },
6814     {STATE_RENDER(WINED3D_RS_ALPHAREF),                   { STATE_RENDER(WINED3D_RS_ALPHATESTENABLE),           NULL                    }, WINED3D_GL_EXT_NONE             },
6815     {STATE_RENDER(WINED3D_RS_ALPHATESTENABLE),            { STATE_RENDER(WINED3D_RS_ALPHATESTENABLE),           alpha_test_arbfp        }, WINED3D_GL_EXT_NONE             },
6816     {STATE_RENDER(WINED3D_RS_COLORKEYENABLE),             { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6817     {STATE_COLOR_KEY,                                     { STATE_COLOR_KEY,                                    color_key_arbfp         }, WINED3D_GL_EXT_NONE             },
6818     {STATE_RENDER(WINED3D_RS_FOGENABLE),                  { STATE_RENDER(WINED3D_RS_FOGENABLE),                 state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6819     {STATE_RENDER(WINED3D_RS_FOGTABLEMODE),               { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6820     {STATE_RENDER(WINED3D_RS_FOGVERTEXMODE),              { STATE_RENDER(WINED3D_RS_FOGENABLE),                 NULL                    }, WINED3D_GL_EXT_NONE             },
6821     {STATE_RENDER(WINED3D_RS_FOGSTART),                   { STATE_RENDER(WINED3D_RS_FOGSTART),                  state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6822     {STATE_RENDER(WINED3D_RS_FOGEND),                     { STATE_RENDER(WINED3D_RS_FOGSTART),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6823     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),           state_srgbwrite         }, ARB_FRAMEBUFFER_SRGB            },
6824     {STATE_RENDER(WINED3D_RS_SRGBWRITEENABLE),            { STATE_SHADER(WINED3D_SHADER_TYPE_PIXEL),            NULL                    }, WINED3D_GL_EXT_NONE             },
6825     {STATE_RENDER(WINED3D_RS_FOGCOLOR),                   { STATE_RENDER(WINED3D_RS_FOGCOLOR),                  state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6826     {STATE_RENDER(WINED3D_RS_FOGDENSITY),                 { STATE_RENDER(WINED3D_RS_FOGDENSITY),                state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6827     {STATE_TEXTURESTAGE(0,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(0,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform}, WINED3D_GL_EXT_NONE             },
6828     {STATE_TEXTURESTAGE(1,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(1,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform}, WINED3D_GL_EXT_NONE             },
6829     {STATE_TEXTURESTAGE(2,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(2,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform}, WINED3D_GL_EXT_NONE             },
6830     {STATE_TEXTURESTAGE(3,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(3,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform}, WINED3D_GL_EXT_NONE             },
6831     {STATE_TEXTURESTAGE(4,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(4,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform}, WINED3D_GL_EXT_NONE             },
6832     {STATE_TEXTURESTAGE(5,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(5,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform}, WINED3D_GL_EXT_NONE             },
6833     {STATE_TEXTURESTAGE(6,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(6,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform}, WINED3D_GL_EXT_NONE             },
6834     {STATE_TEXTURESTAGE(7,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), {STATE_TEXTURESTAGE(7,WINED3D_TSS_TEXTURE_TRANSFORM_FLAGS), textransform}, WINED3D_GL_EXT_NONE             },
6835     {STATE_TEXTURESTAGE(0, WINED3D_TSS_CONSTANT),         { STATE_TEXTURESTAGE(0, WINED3D_TSS_CONSTANT),        state_tss_constant_arbfp}, WINED3D_GL_EXT_NONE             },
6836     {STATE_TEXTURESTAGE(1, WINED3D_TSS_CONSTANT),         { STATE_TEXTURESTAGE(1, WINED3D_TSS_CONSTANT),        state_tss_constant_arbfp}, WINED3D_GL_EXT_NONE             },
6837     {STATE_TEXTURESTAGE(2, WINED3D_TSS_CONSTANT),         { STATE_TEXTURESTAGE(2, WINED3D_TSS_CONSTANT),        state_tss_constant_arbfp}, WINED3D_GL_EXT_NONE             },
6838     {STATE_TEXTURESTAGE(3, WINED3D_TSS_CONSTANT),         { STATE_TEXTURESTAGE(3, WINED3D_TSS_CONSTANT),        state_tss_constant_arbfp}, WINED3D_GL_EXT_NONE             },
6839     {STATE_TEXTURESTAGE(4, WINED3D_TSS_CONSTANT),         { STATE_TEXTURESTAGE(4, WINED3D_TSS_CONSTANT),        state_tss_constant_arbfp}, WINED3D_GL_EXT_NONE             },
6840     {STATE_TEXTURESTAGE(5, WINED3D_TSS_CONSTANT),         { STATE_TEXTURESTAGE(5, WINED3D_TSS_CONSTANT),        state_tss_constant_arbfp}, WINED3D_GL_EXT_NONE             },
6841     {STATE_TEXTURESTAGE(6, WINED3D_TSS_CONSTANT),         { STATE_TEXTURESTAGE(6, WINED3D_TSS_CONSTANT),        state_tss_constant_arbfp}, WINED3D_GL_EXT_NONE             },
6842     {STATE_TEXTURESTAGE(7, WINED3D_TSS_CONSTANT),         { STATE_TEXTURESTAGE(7, WINED3D_TSS_CONSTANT),        state_tss_constant_arbfp}, WINED3D_GL_EXT_NONE             },
6843     {STATE_RENDER(WINED3D_RS_SPECULARENABLE),             { STATE_RENDER(WINED3D_RS_SPECULARENABLE),            state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6844     {STATE_RENDER(WINED3D_RS_SHADEMODE),                  { STATE_RENDER(WINED3D_RS_SHADEMODE),                 state_shademode         }, WINED3D_GL_EXT_NONE             },
6845     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6846 };
6847 
6848 static BOOL arbfp_alloc_context_data(struct wined3d_context *context)
6849 {
6850     return TRUE;
6851 }
6852 
6853 static void arbfp_free_context_data(struct wined3d_context *context)
6854 {
6855 }
6856 
6857 const struct fragment_pipeline arbfp_fragment_pipeline = {
6858     arbfp_enable,
6859     arbfp_get_caps,
6860     arbfp_get_emul_mask,
6861     arbfp_alloc,
6862     arbfp_free,
6863     arbfp_alloc_context_data,
6864     arbfp_free_context_data,
6865     shader_arb_color_fixup_supported,
6866     arbfp_fragmentstate_template,
6867 };
6868 
6869 struct arbfp_blit_type
6870 {
6871     enum complex_fixup fixup : 4;
6872     enum wined3d_gl_resource_type res_type : 3;
6873     DWORD use_color_key : 1;
6874     DWORD padding : 24;
6875 };
6876 
6877 struct arbfp_blit_desc
6878 {
6879     GLuint shader;
6880     struct arbfp_blit_type type;
6881     struct wine_rb_entry entry;
6882 };
6883 
6884 #define ARBFP_BLIT_PARAM_SIZE 0
6885 #define ARBFP_BLIT_PARAM_COLOR_KEY_LOW 1
6886 #define ARBFP_BLIT_PARAM_COLOR_KEY_HIGH 2
6887 
6888 struct wined3d_arbfp_blitter
6889 {
6890     struct wined3d_blitter blitter;
6891     struct wine_rb_tree shaders;
6892     GLuint palette_texture;
6893 };
6894 
6895 static int arbfp_blit_type_compare(const void *key, const struct wine_rb_entry *entry)
6896 {
6897     const struct arbfp_blit_type *ka = key;
6898     const struct arbfp_blit_type *kb = &WINE_RB_ENTRY_VALUE(entry, const struct arbfp_blit_desc, entry)->type;
6899 
6900     return memcmp(ka, kb, sizeof(*ka));
6901 }
6902 
6903 /* Context activation is done by the caller. */
6904 static void arbfp_free_blit_shader(struct wine_rb_entry *entry, void *ctx)
6905 {
6906     struct arbfp_blit_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_blit_desc, entry);
6907     const struct wined3d_gl_info *gl_info;
6908     struct wined3d_context *context;
6909 
6910     context = ctx;
6911     gl_info = context->gl_info;
6912 
6913     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
6914     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
6915     heap_free(entry_arb);
6916 }
6917 
6918 /* Context activation is done by the caller. */
6919 static void arbfp_blitter_destroy(struct wined3d_blitter *blitter, struct wined3d_context *context)
6920 {
6921     const struct wined3d_gl_info *gl_info = context->gl_info;
6922     struct wined3d_arbfp_blitter *arbfp_blitter;
6923     struct wined3d_blitter *next;
6924 
6925     if ((next = blitter->next))
6926         next->ops->blitter_destroy(next, context);
6927 
6928     arbfp_blitter = CONTAINING_RECORD(blitter, struct wined3d_arbfp_blitter, blitter);
6929 
6930     wine_rb_destroy(&arbfp_blitter->shaders, arbfp_free_blit_shader, context);
6931     checkGLcall("Delete blit programs");
6932 
6933     if (arbfp_blitter->palette_texture)
6934         gl_info->gl_ops.gl.p_glDeleteTextures(1, &arbfp_blitter->palette_texture);
6935 
6936     heap_free(arbfp_blitter);
6937 }
6938 
6939 static void gen_packed_yuv_read(struct wined3d_string_buffer *buffer,
6940         const struct arbfp_blit_type *type, char *luminance)
6941 {
6942     char chroma;
6943     const char *tex, *texinstr = "TXP";
6944 
6945     if (type->fixup == COMPLEX_FIXUP_UYVY)
6946     {
6947         chroma = 'x';
6948         *luminance = 'w';
6949     }
6950     else
6951     {
6952         chroma = 'w';
6953         *luminance = 'x';
6954     }
6955 
6956     tex = arbfp_texture_target(type->res_type);
6957     if (type->res_type == WINED3D_GL_RES_TYPE_TEX_RECT)
6958         texinstr = "TEX";
6959 
6960     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6961      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6962      * filtering when we sample the texture.
6963      *
6964      * These are the rules for reading the chroma:
6965      *
6966      * Even pixel: Cr
6967      * Even pixel: U
6968      * Odd pixel: V
6969      *
6970      * So we have to get the sampling x position in non-normalized coordinates in integers
6971      */
6972     if (type->res_type != WINED3D_GL_RES_TYPE_TEX_RECT)
6973     {
6974         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6975         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6976     }
6977     else
6978     {
6979         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6980     }
6981     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6982      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6983      * 0.5, so add 0.5.
6984      */
6985     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6986     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6987 
6988     /* Multiply the x coordinate by 0.5 and get the fraction. This gives 0.25
6989      * and 0.75 for the even and odd pixels respectively. */
6990     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6991     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6992 
6993     /* Sample Pixel 1. */
6994     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6995 
6996     /* Put the value into either of the chroma values */
6997     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6998     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6999     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
7000     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
7001 
7002     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
7003      * the pixel right to the current one. Otherwise, sample the left pixel.
7004      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
7005      */
7006     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
7007     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
7008     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
7009 
7010     /* Put the value into the other chroma */
7011     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
7012     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
7013     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
7014     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
7015 
7016     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
7017      * the current one and lerp the two U and V values
7018      */
7019 
7020     /* This gives the correctly filtered luminance value */
7021     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
7022 }
7023 
7024 static void gen_yv12_read(struct wined3d_string_buffer *buffer,
7025         const struct arbfp_blit_type *type, char *luminance)
7026 {
7027     const char *tex;
7028     static const float yv12_coef[]
7029             = {2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f};
7030 
7031     tex = arbfp_texture_target(type->res_type);
7032 
7033     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
7034      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
7035      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
7036      * pitch of the luminance plane, the packing into the gl texture is a bit
7037      * unfortunate. If the whole texture is interpreted as luminance data it looks
7038      * approximately like this:
7039      *
7040      *        +----------------------------------+----
7041      *        |                                  |
7042      *        |                                  |
7043      *        |                                  |
7044      *        |                                  |
7045      *        |                                  |   2
7046      *        |            LUMINANCE             |   -
7047      *        |                                  |   3
7048      *        |                                  |
7049      *        |                                  |
7050      *        |                                  |
7051      *        |                                  |
7052      *        +----------------+-----------------+----
7053      *        |                |                 |
7054      *        |  V even rows   |  V odd rows     |
7055      *        |                |                 |   1
7056      *        +----------------+------------------   -
7057      *        |                |                 |   3
7058      *        |  U even rows   |  U odd rows     |
7059      *        |                |                 |
7060      *        +----------------+-----------------+----
7061      *        |                |                 |
7062      *        |     0.5        |       0.5       |
7063      *
7064      * So it appears as if there are 4 chroma images, but in fact the odd rows
7065      * in the chroma images are in the same row as the even ones. So it is
7066      * kinda tricky to read
7067      *
7068      * When reading from rectangle textures, keep in mind that the input y coordinates
7069      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
7070      */
7071     shader_addline(buffer, "PARAM yv12_coef = ");
7072     shader_arb_append_imm_vec4(buffer, yv12_coef);
7073     shader_addline(buffer, ";\n");
7074 
7075     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
7076     /* the chroma planes have only half the width */
7077     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
7078 
7079     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
7080      * the coordinate. Also read the right side of the image when reading odd lines
7081      *
7082      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
7083      * bleeding
7084      */
7085     if (type->res_type == WINED3D_GL_RES_TYPE_TEX_2D)
7086     {
7087         shader_addline(buffer, "RCP chroma.w, size.y;\n");
7088 
7089         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
7090 
7091         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
7092         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
7093 
7094         /* Read odd lines from the right side (add size * 0.5 to the x coordinate). */
7095         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
7096         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
7097         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
7098         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
7099 
7100         /* clamp, keep the half pixel origin in mind */
7101         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
7102         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
7103         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
7104         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
7105     }
7106     else
7107     {
7108         /* The y coordinate for V is in the range [size, size + size / 4). */
7109         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
7110         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
7111 
7112         /* Read odd lines from the right side (add size * 0.5 to the x coordinate). */
7113         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
7114         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
7115         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
7116         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
7117         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
7118 
7119         /* Make sure to read exactly from the pixel center */
7120         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
7121         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
7122 
7123         /* Clamp */
7124         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
7125         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
7126         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
7127         shader_addline(buffer, "ADD temp.y, size.y, coef.y;\n");
7128         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
7129     }
7130     /* Read the texture, put the result into the output register */
7131     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
7132     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
7133 
7134     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
7135      * No need to clamp because we're just reusing the already clamped value from above
7136      */
7137     if (type->res_type == WINED3D_GL_RES_TYPE_TEX_2D)
7138         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
7139     else
7140         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
7141     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
7142     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
7143 
7144     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
7145      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
7146      * values due to filtering
7147      */
7148     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
7149     if (type->res_type == WINED3D_GL_RES_TYPE_TEX_2D)
7150     {
7151         /* Multiply the y coordinate by 2/3 and clamp it */
7152         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
7153         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
7154         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
7155         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
7156     }
7157     else
7158     {
7159         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
7160          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
7161          * is bigger
7162          */
7163         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
7164         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
7165         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
7166     }
7167     *luminance = 'a';
7168 }
7169 
7170 static void gen_nv12_read(struct wined3d_string_buffer *buffer,
7171         const struct arbfp_blit_type *type, char *luminance)
7172 {
7173     const char *tex;
7174     static const float nv12_coef[]
7175             = {2.0f / 3.0f, 1.0f / 3.0f, 1.0f, 1.0f};
7176 
7177     tex = arbfp_texture_target(type->res_type);
7178 
7179     /* NV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
7180      * sized plane where each component is an UV pair. So the effective
7181      * bitdepth is 12 bits per pixel If the whole texture is interpreted as luminance
7182      * data it looks approximately like this:
7183      *
7184      *        +----------------------------------+----
7185      *        |                                  |
7186      *        |                                  |
7187      *        |                                  |
7188      *        |                                  |
7189      *        |                                  |   2
7190      *        |            LUMINANCE             |   -
7191      *        |                                  |   3
7192      *        |                                  |
7193      *        |                                  |
7194      *        |                                  |
7195      *        |                                  |
7196      *        +----------------------------------+----
7197      *        |UVUVUVUVUVUVUVUVUVUVUVUVUVUVUVUVUV|
7198      *        |UVUVUVUVUVUVUVUVUVUVUVUVUVUVUVUVUV|
7199      *        |                                  |   1
7200      *        |                                  |   -
7201      *        |                                  |   3
7202      *        |                                  |
7203      *        |                                  |
7204      *        +----------------------------------+----
7205      *
7206      * When reading from rectangle textures, keep in mind that the input y coordinates
7207      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height. */
7208 
7209     shader_addline(buffer, "PARAM nv12_coef = ");
7210     shader_arb_append_imm_vec4(buffer, nv12_coef);
7211     shader_addline(buffer, ";\n");
7212 
7213     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
7214     /* We only have half the number of chroma pixels. */
7215     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
7216 
7217     if (type->res_type == WINED3D_GL_RES_TYPE_TEX_2D)
7218     {
7219         shader_addline(buffer, "RCP chroma.w, size.x;\n");
7220         shader_addline(buffer, "RCP chroma.z, size.y;\n");
7221 
7222         shader_addline(buffer, "MAD texcrd.y, texcrd.y, nv12_coef.y, nv12_coef.x;\n");
7223 
7224         /* We must not allow filtering horizontally, this would mix U and V.
7225          * Vertical filtering is ok. However, bear in mind that the pixel center is at
7226          * 0.5, so add 0.5. */
7227 
7228         /* Convert to non-normalized coordinates so we can find the
7229          * individual pixel. */
7230         shader_addline(buffer, "MUL texcrd.x, texcrd.x, size.x;\n");
7231         shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
7232         /* Multiply by 2 since chroma components are stored in UV pixel pairs,
7233          * add 0.5 to hit the center of the pixel. */
7234         shader_addline(buffer, "MAD texcrd.x, texcrd.x, coef.z, coef.y;\n");
7235 
7236         /* Convert back to normalized coordinates. */
7237         shader_addline(buffer, "MUL texcrd.x, texcrd.x, chroma.w;\n");
7238 
7239         /* Clamp, keep the half pixel origin in mind. */
7240         shader_addline(buffer, "MAD temp.y, coef.y, chroma.z, nv12_coef.x;\n");
7241         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
7242         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.z, nv12_coef.z;\n");
7243         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
7244     }
7245     else
7246     {
7247         /* The y coordinate for chroma is in the range [size, size + size / 2). */
7248         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.y, size.y;\n");
7249 
7250         shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
7251         /* Multiply by 2 since chroma components are stored in UV pixel pairs,
7252          * add 0.5 to hit the center of the pixel. */
7253         shader_addline(buffer, "MAD texcrd.x, texcrd.x, coef.z, coef.y;\n");
7254 
7255         /* Clamp */
7256         shader_addline(buffer, "MAD temp.y, size.y, coef.y, size.y;\n");
7257         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
7258         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
7259         shader_addline(buffer, "ADD temp.y, size.y, coef.y;\n");
7260         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
7261     }
7262     /* Read the texture, put the result into the output register. */
7263     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
7264     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
7265 
7266     if (type->res_type == WINED3D_GL_RES_TYPE_TEX_2D)
7267     {
7268         /* Add 1/size.x */
7269         shader_addline(buffer, "ADD texcrd.x, texcrd.x, chroma.w;\n");
7270     }
7271     else
7272     {
7273         /* Add 1 */
7274         shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.x;\n");
7275     }
7276     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
7277     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
7278 
7279     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
7280      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
7281      * values due to filtering. */
7282     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
7283     if (type->res_type == WINED3D_GL_RES_TYPE_TEX_2D)
7284     {
7285         /* Multiply the y coordinate by 2/3 and clamp it */
7286         shader_addline(buffer, "MUL texcrd.y, texcrd.y, nv12_coef.x;\n");
7287         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, nv12_coef.x;\n");
7288         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
7289         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
7290     }
7291     else
7292     {
7293         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
7294          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
7295          * is bigger
7296          */
7297         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
7298         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
7299         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
7300     }
7301     *luminance = 'a';
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             gen_packed_yuv_read(&buffer, type, &luminance_component);
7468             break;
7469 
7470         case COMPLEX_FIXUP_YV12:
7471             gen_yv12_read(&buffer, type, &luminance_component);
7472             break;
7473 
7474         case COMPLEX_FIXUP_NV12:
7475             gen_nv12_read(&buffer, type, &luminance_component);
7476             break;
7477 
7478         default:
7479             FIXME("Unsupported YUV fixup %#x\n", type->fixup);
7480             string_buffer_free(&buffer);
7481             return 0;
7482     }
7483 
7484     /* Calculate the final result. Formula is taken from
7485      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
7486      * ranges from -0.5 to 0.5
7487      */
7488     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
7489 
7490     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
7491     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
7492     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
7493     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
7494     shader_addline(&buffer, "END\n");
7495 
7496     shader_arb_compile(gl_info, GL_FRAGMENT_PROGRAM_ARB, buffer.buffer);
7497 
7498     string_buffer_free(&buffer);
7499 
7500     return shader;
7501 }
7502 
7503 /* Context activation is done by the caller. */
7504 static GLuint arbfp_gen_plain_shader(const struct wined3d_gl_info *gl_info, const struct arbfp_blit_type *type)
7505 {
7506     GLuint shader;
7507     struct wined3d_string_buffer buffer;
7508     const char *tex_target = arbfp_texture_target(type->res_type);
7509 
7510     /* Shader header */
7511     if (!string_buffer_init(&buffer))
7512     {
7513         ERR("Failed to initialize shader buffer.\n");
7514         return 0;
7515     }
7516 
7517     GL_EXTCALL(glGenProgramsARB(1, &shader));
7518     if (!shader)
7519     {
7520         string_buffer_free(&buffer);
7521         return 0;
7522     }
7523     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7524 
7525     shader_addline(&buffer, "!!ARBfp1.0\n");
7526 
7527     if (type->use_color_key)
7528     {
7529         shader_addline(&buffer, "TEMP color;\n");
7530         shader_addline(&buffer, "TEMP less, greater;\n");
7531         shader_addline(&buffer, "PARAM color_key_low = program.local[%u];\n", ARBFP_BLIT_PARAM_COLOR_KEY_LOW);
7532         shader_addline(&buffer, "PARAM color_key_high = program.local[%u];\n", ARBFP_BLIT_PARAM_COLOR_KEY_HIGH);
7533         shader_addline(&buffer, "TEX color, fragment.texcoord[0], texture[0], %s;\n", tex_target);
7534         shader_addline(&buffer, "SLT less, color, color_key_low;\n"); /* below low key */
7535         shader_addline(&buffer, "SGE greater, color, color_key_high;\n"); /* above high key */
7536         shader_addline(&buffer, "ADD less, less, greater;\n"); /* or */
7537         shader_addline(&buffer, "DP4 less.b, less, less;\n"); /* on any channel */
7538         shader_addline(&buffer, "SGE less, -less.b, 0.0;\n"); /* logical not */
7539         shader_addline(&buffer, "KIL -less;\n"); /* discard if true */
7540         shader_addline(&buffer, "MOV result.color, color;\n");
7541     }
7542     else
7543     {
7544         shader_addline(&buffer, "TEX result.color, fragment.texcoord[0], texture[0], %s;\n", tex_target);
7545     }
7546 
7547     shader_addline(&buffer, "END\n");
7548 
7549     shader_arb_compile(gl_info, GL_FRAGMENT_PROGRAM_ARB, buffer.buffer);
7550 
7551     string_buffer_free(&buffer);
7552 
7553     return shader;
7554 }
7555 
7556 /* Context activation is done by the caller. */
7557 static HRESULT arbfp_blit_set(struct wined3d_arbfp_blitter *blitter, struct wined3d_context *context,
7558         const struct wined3d_texture *texture, unsigned int sub_resource_idx,
7559         const struct wined3d_color_key *color_key)
7560 {
7561     enum complex_fixup fixup;
7562     const struct wined3d_gl_info *gl_info = context->gl_info;
7563     struct wine_rb_entry *entry;
7564     struct arbfp_blit_type type;
7565     struct arbfp_blit_desc *desc;
7566     struct wined3d_color float_color_key[2];
7567     struct wined3d_vec4 size;
7568     unsigned int level;
7569     GLuint shader;
7570 
7571     level = sub_resource_idx % texture->level_count;
7572     size.x = wined3d_texture_get_level_pow2_width(texture, level);
7573     size.y = wined3d_texture_get_level_pow2_height(texture, level);
7574     size.z = 1.0f;
7575     size.w = 1.0f;
7576 
7577     if (is_complex_fixup(texture->resource.format->color_fixup))
7578         fixup = get_complex_fixup(texture->resource.format->color_fixup);
7579     else
7580         fixup = COMPLEX_FIXUP_NONE;
7581 
7582     switch (texture->target)
7583     {
7584         case GL_TEXTURE_1D:
7585             type.res_type = WINED3D_GL_RES_TYPE_TEX_1D;
7586             break;
7587 
7588         case GL_TEXTURE_2D:
7589             type.res_type = WINED3D_GL_RES_TYPE_TEX_2D;
7590             break;
7591 
7592         case GL_TEXTURE_3D:
7593             type.res_type = WINED3D_GL_RES_TYPE_TEX_3D;
7594             break;
7595 
7596         case GL_TEXTURE_CUBE_MAP_ARB:
7597             type.res_type = WINED3D_GL_RES_TYPE_TEX_CUBE;
7598             break;
7599 
7600         case GL_TEXTURE_RECTANGLE_ARB:
7601             type.res_type = WINED3D_GL_RES_TYPE_TEX_RECT;
7602             break;
7603 
7604         default:
7605             ERR("Unexpected GL texture type %#x.\n", texture->target);
7606             type.res_type = WINED3D_GL_RES_TYPE_TEX_2D;
7607     }
7608     type.fixup = fixup;
7609     type.use_color_key = !!color_key;
7610     type.padding = 0;
7611 
7612     if ((entry = wine_rb_get(&blitter->shaders, &type)))
7613     {
7614         desc = WINE_RB_ENTRY_VALUE(entry, struct arbfp_blit_desc, entry);
7615         shader = desc->shader;
7616     }
7617     else
7618     {
7619         switch (fixup)
7620         {
7621             case COMPLEX_FIXUP_NONE:
7622                 if (!is_identity_fixup(texture->resource.format->color_fixup))
7623                     FIXME("Implement support for sign or swizzle fixups.\n");
7624                 shader = arbfp_gen_plain_shader(gl_info, &type);
7625                 break;
7626 
7627             case COMPLEX_FIXUP_P8:
7628                 shader = gen_p8_shader(gl_info, &type);
7629                 break;
7630 
7631             case COMPLEX_FIXUP_YUY2:
7632             case COMPLEX_FIXUP_UYVY:
7633             case COMPLEX_FIXUP_YV12:
7634             case COMPLEX_FIXUP_NV12:
7635                 shader = gen_yuv_shader(gl_info, &type);
7636                 break;
7637         }
7638 
7639         if (!shader)
7640         {
7641             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7642             return E_NOTIMPL;
7643         }
7644 
7645         if (!(desc = heap_alloc(sizeof(*desc))))
7646             goto err_out;
7647 
7648         desc->type = type;
7649         desc->shader = shader;
7650         if (wine_rb_put(&blitter->shaders, &desc->type, &desc->entry) == -1)
7651         {
7652 err_out:
7653             ERR("Out of memory\n");
7654             GL_EXTCALL(glDeleteProgramsARB(1, &shader));
7655             checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader))");
7656             GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, 0));
7657             checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, 0)");
7658             heap_free(desc);
7659             return E_OUTOFMEMORY;
7660         }
7661     }
7662 
7663     if (fixup == COMPLEX_FIXUP_P8)
7664         upload_palette(blitter, texture, context);
7665 
7666     gl_info->gl_ops.gl.p_glEnable(GL_FRAGMENT_PROGRAM_ARB);
7667     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7668     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7669     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7670     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARBFP_BLIT_PARAM_SIZE, &size.x));
7671     checkGLcall("glProgramLocalParameter4fvARB");
7672     if (type.use_color_key)
7673     {
7674         wined3d_format_get_float_color_key(texture->resource.format, color_key, float_color_key);
7675         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
7676                 ARBFP_BLIT_PARAM_COLOR_KEY_LOW, &float_color_key[0].r));
7677         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
7678                 ARBFP_BLIT_PARAM_COLOR_KEY_HIGH, &float_color_key[1].r));
7679         checkGLcall("glProgramLocalParameter4fvARB");
7680     }
7681 
7682     return WINED3D_OK;
7683 }
7684 
7685 /* Context activation is done by the caller. */
7686 static void arbfp_blit_unset(const struct wined3d_gl_info *gl_info)
7687 {
7688     gl_info->gl_ops.gl.p_glDisable(GL_FRAGMENT_PROGRAM_ARB);
7689     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7690 }
7691 
7692 static BOOL arbfp_blit_supported(enum wined3d_blit_op blit_op, const struct wined3d_context *context,
7693         const struct wined3d_resource *src_resource, DWORD src_location,
7694         const struct wined3d_resource *dst_resource, DWORD dst_location)
7695 {
7696     const struct wined3d_format *src_format = src_resource->format;
7697     const struct wined3d_format *dst_format = dst_resource->format;
7698     enum complex_fixup src_fixup;
7699     BOOL decompress;
7700 
7701     if (src_resource->type != WINED3D_RTYPE_TEXTURE_2D)
7702         return FALSE;
7703 
7704     if (blit_op == WINED3D_BLIT_OP_RAW_BLIT && dst_format->id == src_format->id)
7705     {
7706         if (dst_format->flags[WINED3D_GL_RES_TYPE_TEX_2D] & (WINED3DFMT_FLAG_DEPTH | WINED3DFMT_FLAG_STENCIL))
7707             blit_op = WINED3D_BLIT_OP_DEPTH_BLIT;
7708         else
7709             blit_op = WINED3D_BLIT_OP_COLOR_BLIT;
7710     }
7711 
7712     switch (blit_op)
7713     {
7714         case WINED3D_BLIT_OP_COLOR_BLIT_CKEY:
7715             if (!context->d3d_info->shader_color_key)
7716             {
7717                 /* The conversion modifies the alpha channel so the color key might no longer match. */
7718                 TRACE("Color keying not supported with converted textures.\n");
7719                 return FALSE;
7720             }
7721         case WINED3D_BLIT_OP_COLOR_BLIT_ALPHATEST:
7722         case WINED3D_BLIT_OP_COLOR_BLIT:
7723             break;
7724 
7725         default:
7726             TRACE("Unsupported blit_op=%d\n", blit_op);
7727             return FALSE;
7728     }
7729 
7730     decompress = src_format && (src_format->flags[WINED3D_GL_RES_TYPE_TEX_2D] & WINED3DFMT_FLAG_COMPRESSED)
7731             && !(dst_format->flags[WINED3D_GL_RES_TYPE_TEX_2D] & WINED3DFMT_FLAG_COMPRESSED);
7732     if (!decompress && !(src_resource->access & dst_resource->access & WINED3D_RESOURCE_ACCESS_GPU))
7733         return FALSE;
7734 
7735     src_fixup = get_complex_fixup(src_format->color_fixup);
7736     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7737     {
7738         TRACE("Checking support for fixup:\n");
7739         dump_color_fixup_desc(src_format->color_fixup);
7740     }
7741 
7742     if (!is_identity_fixup(dst_format->color_fixup)
7743             && (dst_format->id != src_format->id || dst_location != WINED3D_LOCATION_DRAWABLE))
7744     {
7745         TRACE("Destination fixups are not supported\n");
7746         return FALSE;
7747     }
7748 
7749     if (is_identity_fixup(src_format->color_fixup))
7750     {
7751         TRACE("[OK]\n");
7752         return TRUE;
7753     }
7754 
7755      /* We only support YUV conversions. */
7756     if (!is_complex_fixup(src_format->color_fixup))
7757     {
7758         if (wined3d_settings.offscreen_rendering_mode == ORM_BACKBUFFER)
7759         {
7760             WARN("Claiming fixup support because of ORM_BACKBUFFER.\n");
7761             return TRUE;
7762         }
7763 
7764         TRACE("[FAILED]\n");
7765         return FALSE;
7766     }
7767 
7768     switch(src_fixup)
7769     {
7770         case COMPLEX_FIXUP_YUY2:
7771         case COMPLEX_FIXUP_UYVY:
7772         case COMPLEX_FIXUP_YV12:
7773         case COMPLEX_FIXUP_NV12:
7774         case COMPLEX_FIXUP_P8:
7775             TRACE("[OK]\n");
7776             return TRUE;
7777 
7778         default:
7779             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7780             TRACE("[FAILED]\n");
7781             return FALSE;
7782     }
7783 }
7784 
7785 static DWORD arbfp_blitter_blit(struct wined3d_blitter *blitter, enum wined3d_blit_op op,
7786         struct wined3d_context *context, struct wined3d_texture *src_texture, unsigned int src_sub_resource_idx,
7787         DWORD src_location, const RECT *src_rect, struct wined3d_texture *dst_texture,
7788         unsigned int dst_sub_resource_idx, DWORD dst_location, const RECT *dst_rect,
7789         const struct wined3d_color_key *color_key, enum wined3d_texture_filter_type filter)
7790 {
7791     struct wined3d_device *device = dst_texture->resource.device;
7792     struct wined3d_texture *staging_texture = NULL;
7793     struct wined3d_arbfp_blitter *arbfp_blitter;
7794     struct wined3d_color_key alpha_test_key;
7795     struct wined3d_blitter *next;
7796     unsigned int src_level;
7797     RECT s, d;
7798 
7799     TRACE("blitter %p, op %#x, context %p, src_texture %p, src_sub_resource_idx %u, src_location %s, src_rect %s, "
7800             "dst_texture %p, dst_sub_resource_idx %u, dst_location %s, dst_rect %s, colour_key %p, filter %s.\n",
7801             blitter, op, context, src_texture, src_sub_resource_idx, wined3d_debug_location(src_location),
7802             wine_dbgstr_rect(src_rect), dst_texture, dst_sub_resource_idx, wined3d_debug_location(dst_location),
7803             wine_dbgstr_rect(dst_rect), color_key, debug_d3dtexturefiltertype(filter));
7804 
7805     if (!arbfp_blit_supported(op, context, &src_texture->resource, src_location,
7806             &dst_texture->resource, dst_location))
7807     {
7808         if (!(next = blitter->next))
7809         {
7810             ERR("No blitter to handle blit op %#x.\n", op);
7811             return dst_location;
7812         }
7813 
7814         TRACE("Forwarding to blitter %p.\n", next);
7815         return next->ops->blitter_blit(next, op, context, src_texture, src_sub_resource_idx, src_location,
7816                 src_rect, dst_texture, dst_sub_resource_idx, dst_location, dst_rect, color_key, filter);
7817     }
7818 
7819     arbfp_blitter = CONTAINING_RECORD(blitter, struct wined3d_arbfp_blitter, blitter);
7820 
7821     if (!(src_texture->resource.access & WINED3D_RESOURCE_ACCESS_GPU))
7822     {
7823         struct wined3d_resource_desc desc;
7824         struct wined3d_box upload_box;
7825         HRESULT hr;
7826 
7827         TRACE("Source texture is not GPU accessible, creating a staging texture.\n");
7828 
7829         src_level = src_sub_resource_idx % src_texture->level_count;
7830         desc.resource_type = WINED3D_RTYPE_TEXTURE_2D;
7831         desc.format = src_texture->resource.format->id;
7832         desc.multisample_type = src_texture->resource.multisample_type;
7833         desc.multisample_quality = src_texture->resource.multisample_quality;
7834         desc.usage = WINED3DUSAGE_PRIVATE;
7835         desc.access = WINED3D_RESOURCE_ACCESS_GPU;
7836         desc.width = wined3d_texture_get_level_width(src_texture, src_level);
7837         desc.height = wined3d_texture_get_level_height(src_texture, src_level);
7838         desc.depth = 1;
7839         desc.size = 0;
7840 
7841         if (FAILED(hr = wined3d_texture_create(device, &desc, 1, 1, 0,
7842                 NULL, NULL, &wined3d_null_parent_ops, &staging_texture)))
7843         {
7844             ERR("Failed to create staging texture, hr %#x.\n", hr);
7845             return dst_location;
7846         }
7847 
7848         wined3d_box_set(&upload_box, 0, 0, desc.width, desc.height, 0, desc.depth);
7849         wined3d_texture_upload_from_texture(staging_texture, 0, 0, 0, 0,
7850                 src_texture, src_sub_resource_idx, &upload_box);
7851 
7852         src_texture = staging_texture;
7853         src_sub_resource_idx = 0;
7854     }
7855     else if (wined3d_settings.offscreen_rendering_mode != ORM_FBO
7856             && (src_texture->sub_resources[src_sub_resource_idx].locations
7857             & (WINED3D_LOCATION_TEXTURE_RGB | WINED3D_LOCATION_DRAWABLE)) == WINED3D_LOCATION_DRAWABLE
7858             && !wined3d_resource_is_offscreen(&src_texture->resource))
7859     {
7860 
7861         /* Without FBO blits transferring from the drawable to the texture is
7862          * expensive, because we have to flip the data in sysmem. Since we can
7863          * flip in the blitter, we don't actually need that flip anyway. So we
7864          * use the surface's texture as scratch texture, and flip the source
7865          * rectangle instead. */
7866         texture2d_load_fb_texture(src_texture, src_sub_resource_idx, FALSE, context);
7867 
7868         s = *src_rect;
7869         src_level = src_sub_resource_idx % src_texture->level_count;
7870         s.top = wined3d_texture_get_level_height(src_texture, src_level) - s.top;
7871         s.bottom = wined3d_texture_get_level_height(src_texture, src_level) - s.bottom;
7872         src_rect = &s;
7873     }
7874     else
7875     {
7876         wined3d_texture_load(src_texture, context, FALSE);
7877     }
7878 
7879     context_apply_ffp_blit_state(context, device);
7880 
7881     if (dst_location == WINED3D_LOCATION_DRAWABLE)
7882     {
7883         d = *dst_rect;
7884         wined3d_texture_translate_drawable_coords(dst_texture, context->win_handle, &d);
7885         dst_rect = &d;
7886     }
7887 
7888     if (wined3d_settings.offscreen_rendering_mode == ORM_FBO)
7889     {
7890         GLenum buffer;
7891 
7892         if (dst_location == WINED3D_LOCATION_DRAWABLE)
7893         {
7894             TRACE("Destination texture %p is onscreen.\n", dst_texture);
7895             buffer = wined3d_texture_get_gl_buffer(dst_texture);
7896         }
7897         else
7898         {
7899             TRACE("Destination texture %p is offscreen.\n", dst_texture);
7900             buffer = GL_COLOR_ATTACHMENT0;
7901         }
7902         context_apply_fbo_state_blit(context, GL_DRAW_FRAMEBUFFER,
7903                 &dst_texture->resource, dst_sub_resource_idx, NULL, 0, dst_location);
7904         context_set_draw_buffer(context, buffer);
7905         context_check_fbo_status(context, GL_DRAW_FRAMEBUFFER);
7906         context_invalidate_state(context, STATE_FRAMEBUFFER);
7907     }
7908 
7909     if (op == WINED3D_BLIT_OP_COLOR_BLIT_ALPHATEST)
7910     {
7911         const struct wined3d_format *fmt = src_texture->resource.format;
7912         alpha_test_key.color_space_low_value = 0;
7913         alpha_test_key.color_space_high_value = ~(((1u << fmt->alpha_size) - 1) << fmt->alpha_offset);
7914         color_key = &alpha_test_key;
7915     }
7916 
7917     arbfp_blit_set(arbfp_blitter, context, src_texture, src_sub_resource_idx, color_key);
7918 
7919     /* Draw a textured quad */
7920     context_draw_textured_quad(context, src_texture, src_sub_resource_idx, src_rect, dst_rect, filter);
7921 
7922     /* Leave the opengl state valid for blitting */
7923     arbfp_blit_unset(context->gl_info);
7924 
7925     if (dst_texture->swapchain && (dst_texture->swapchain->front_buffer == dst_texture))
7926         context->gl_info->gl_ops.gl.p_glFlush();
7927 
7928     if (staging_texture)
7929         wined3d_texture_decref(staging_texture);
7930 
7931     return dst_location;
7932 }
7933 
7934 static void arbfp_blitter_clear(struct wined3d_blitter *blitter, struct wined3d_device *device,
7935         unsigned int rt_count, const struct wined3d_fb_state *fb, unsigned int rect_count, const RECT *clear_rects,
7936         const RECT *draw_rect, DWORD flags, const struct wined3d_color *colour, float depth, DWORD stencil)
7937 {
7938     struct wined3d_blitter *next;
7939 
7940     if ((next = blitter->next))
7941         next->ops->blitter_clear(next, device, rt_count, fb, rect_count,
7942                 clear_rects, draw_rect, flags, colour, depth, stencil);
7943 }
7944 
7945 static const struct wined3d_blitter_ops arbfp_blitter_ops =
7946 {
7947     arbfp_blitter_destroy,
7948     arbfp_blitter_clear,
7949     arbfp_blitter_blit,
7950 };
7951 
7952 void wined3d_arbfp_blitter_create(struct wined3d_blitter **next, const struct wined3d_device *device)
7953 {
7954     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
7955     struct wined3d_arbfp_blitter *blitter;
7956 
7957     if (device->shader_backend != &arb_program_shader_backend
7958             && device->shader_backend != &glsl_shader_backend)
7959         return;
7960 
7961     if (!gl_info->supported[ARB_FRAGMENT_PROGRAM])
7962         return;
7963 
7964     if (!gl_info->supported[WINED3D_GL_LEGACY_CONTEXT])
7965         return;
7966 
7967     if (!(blitter = heap_alloc(sizeof(*blitter))))
7968     {
7969         ERR("Failed to allocate blitter.\n");
7970         return;
7971     }
7972 
7973     TRACE("Created blitter %p.\n", blitter);
7974 
7975     blitter->blitter.ops = &arbfp_blitter_ops;
7976     blitter->blitter.next = *next;
7977     wine_rb_init(&blitter->shaders, arbfp_blit_type_compare);
7978     blitter->palette_texture = 0;
7979     *next = &blitter->blitter;
7980 }
7981