crypt32/tests: In more recent Windows versions, updating a data message with no conte...
[wine] / dlls / wined3d / arb_program_shader.c
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 2007-2008 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
32 #include <math.h>
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_caps);
40 WINE_DECLARE_DEBUG_CHANNEL(d3d);
41
42 /* Extract a line. Note that this modifies the source string. */
43 static char *get_line(char **ptr)
44 {
45     char *p, *q;
46
47     p = *ptr;
48     if (!(q = strstr(p, "\n")))
49     {
50         if (!*p) return NULL;
51         *ptr += strlen(p);
52         return p;
53     }
54     *q = '\0';
55     *ptr = q + 1;
56
57     return p;
58 }
59
60 static void shader_arb_dump_program_source(const char *source)
61 {
62     ULONG source_size;
63     char *ptr, *line, *tmp;
64
65     source_size = strlen(source) + 1;
66     tmp = HeapAlloc(GetProcessHeap(), 0, source_size);
67     if (!tmp)
68     {
69         ERR("Failed to allocate %u bytes for shader source.\n", source_size);
70         return;
71     }
72     memcpy(tmp, source, source_size);
73
74     ptr = tmp;
75     while ((line = get_line(&ptr))) FIXME("    %s\n", line);
76     FIXME("\n");
77
78     HeapFree(GetProcessHeap(), 0, tmp);
79 }
80
81 enum arb_helper_value
82 {
83     ARB_ZERO,
84     ARB_ONE,
85     ARB_TWO,
86     ARB_0001,
87     ARB_EPS,
88
89     ARB_VS_REL_OFFSET
90 };
91
92 static const char *arb_get_helper_value(enum wined3d_shader_type shader, enum arb_helper_value value)
93 {
94     if (shader == WINED3D_SHADER_TYPE_GEOMETRY)
95     {
96         ERR("Geometry shaders are unsupported\n");
97         return "bad";
98     }
99
100     if (shader == WINED3D_SHADER_TYPE_PIXEL)
101     {
102         switch (value)
103         {
104             case ARB_ZERO: return "ps_helper_const.x";
105             case ARB_ONE: return "ps_helper_const.y";
106             case ARB_TWO: return "coefmul.x";
107             case ARB_0001: return "helper_const.xxxy";
108             case ARB_EPS: return "ps_helper_const.z";
109             default: break;
110         }
111     }
112     else
113     {
114         switch (value)
115         {
116             case ARB_ZERO: return "helper_const.x";
117             case ARB_ONE: return "helper_const.y";
118             case ARB_TWO: return "helper_const.z";
119             case ARB_EPS: return "helper_const.w";
120             case ARB_0001: return "helper_const.xxxy";
121             case ARB_VS_REL_OFFSET: return "rel_addr_const.y";
122         }
123     }
124     FIXME("Unmanaged %s shader helper constant requested: %u\n",
125           shader == WINED3D_SHADER_TYPE_PIXEL ? "pixel" : "vertex", value);
126     switch (value)
127     {
128         case ARB_ZERO: return "0.0";
129         case ARB_ONE: return "1.0";
130         case ARB_TWO: return "2.0";
131         case ARB_0001: return "{0.0, 0.0, 0.0, 1.0}";
132         case ARB_EPS: return "1e-8";
133         default: return "bad";
134     }
135 }
136
137 static inline BOOL ffp_clip_emul(const struct wined3d_state *state)
138 {
139     return state->lowest_disabled_stage < 7;
140 }
141
142 /* ARB_program_shader private data */
143
144 struct control_frame
145 {
146     struct                          list entry;
147     enum
148     {
149         IF,
150         IFC,
151         LOOP,
152         REP
153     } type;
154     BOOL                            muting;
155     BOOL                            outer_loop;
156     union
157     {
158         unsigned int                loop;
159         unsigned int                ifc;
160     } no;
161     struct wined3d_shader_loop_control loop_control;
162     BOOL                            had_else;
163 };
164
165 struct arb_ps_np2fixup_info
166 {
167     struct ps_np2fixup_info         super;
168     /* For ARB we need a offset value:
169      * With both GLSL and ARB mode the NP2 fixup information (the texture dimensions) are stored in a
170      * consecutive way (GLSL uses a uniform array). Since ARB doesn't know the notion of a "standalone"
171      * array we need an offset to the index inside the program local parameter array. */
172     UINT                            offset;
173 };
174
175 struct arb_ps_compile_args
176 {
177     struct ps_compile_args          super;
178     WORD                            bools;
179     WORD                            clip;  /* only a boolean, use a WORD for alignment */
180     unsigned char                   loop_ctrl[MAX_CONST_I][3];
181 };
182
183 struct stb_const_desc
184 {
185     unsigned char           texunit;
186     UINT                    const_num;
187 };
188
189 struct arb_ps_compiled_shader
190 {
191     struct arb_ps_compile_args      args;
192     struct arb_ps_np2fixup_info     np2fixup_info;
193     struct stb_const_desc           bumpenvmatconst[MAX_TEXTURES];
194     struct stb_const_desc           luminanceconst[MAX_TEXTURES];
195     UINT                            int_consts[MAX_CONST_I];
196     GLuint                          prgId;
197     UINT                            ycorrection;
198     unsigned char                   numbumpenvmatconsts;
199     char                            num_int_consts;
200 };
201
202 struct arb_vs_compile_args
203 {
204     struct vs_compile_args          super;
205     union
206     {
207         struct
208         {
209             WORD                    bools;
210             unsigned char           clip_texcoord;
211             unsigned char           clipplane_mask;
212         }                           boolclip;
213         DWORD                       boolclip_compare;
214     } clip;
215     DWORD                           ps_signature;
216     union
217     {
218         unsigned char               samplers[4];
219         DWORD                       samplers_compare;
220     } vertex;
221     unsigned char                   loop_ctrl[MAX_CONST_I][3];
222 };
223
224 struct arb_vs_compiled_shader
225 {
226     struct arb_vs_compile_args      args;
227     GLuint                          prgId;
228     UINT                            int_consts[MAX_CONST_I];
229     char                            num_int_consts;
230     char                            need_color_unclamp;
231     UINT                            pos_fixup;
232 };
233
234 struct recorded_instruction
235 {
236     struct wined3d_shader_instruction ins;
237     struct list entry;
238 };
239
240 struct shader_arb_ctx_priv
241 {
242     char addr_reg[20];
243     enum
244     {
245         /* plain GL_ARB_vertex_program or GL_ARB_fragment_program */
246         ARB,
247         /* GL_NV_vertex_progam2_option or GL_NV_fragment_program_option */
248         NV2,
249         /* GL_NV_vertex_program3 or GL_NV_fragment_program2 */
250         NV3
251     } target_version;
252
253     const struct arb_vs_compile_args    *cur_vs_args;
254     const struct arb_ps_compile_args    *cur_ps_args;
255     const struct arb_ps_compiled_shader *compiled_fprog;
256     const struct arb_vs_compiled_shader *compiled_vprog;
257     struct arb_ps_np2fixup_info         *cur_np2fixup_info;
258     struct list                         control_frames;
259     struct list                         record;
260     BOOL                                recording;
261     BOOL                                muted;
262     unsigned int                        num_loops, loop_depth, num_ifcs;
263     int                                 aL;
264
265     unsigned int                        vs_clipplanes;
266     BOOL                                footer_written;
267     BOOL                                in_main_func;
268
269     /* For 3.0 vertex shaders */
270     const char                          *vs_output[MAX_REG_OUTPUT];
271     /* For 2.x and earlier vertex shaders */
272     const char                          *texcrd_output[8], *color_output[2], *fog_output;
273
274     /* 3.0 pshader input for compatibility with fixed function */
275     const char                          *ps_input[MAX_REG_INPUT];
276 };
277
278 struct ps_signature
279 {
280     struct wined3d_shader_signature_element *sig;
281     DWORD                               idx;
282     struct wine_rb_entry                entry;
283 };
284
285 struct arb_pshader_private {
286     struct arb_ps_compiled_shader   *gl_shaders;
287     UINT                            num_gl_shaders, shader_array_size;
288     BOOL                            has_signature_idx;
289     DWORD                           input_signature_idx;
290     DWORD                           clipplane_emulation;
291     BOOL                            clamp_consts;
292 };
293
294 struct arb_vshader_private {
295     struct arb_vs_compiled_shader   *gl_shaders;
296     UINT                            num_gl_shaders, shader_array_size;
297     UINT rel_offset;
298 };
299
300 struct shader_arb_priv
301 {
302     GLuint                  current_vprogram_id;
303     GLuint                  current_fprogram_id;
304     const struct arb_ps_compiled_shader *compiled_fprog;
305     const struct arb_vs_compiled_shader *compiled_vprog;
306     GLuint                  depth_blt_vprogram_id;
307     GLuint                  depth_blt_fprogram_id_full[tex_type_count];
308     GLuint                  depth_blt_fprogram_id_masked[tex_type_count];
309     BOOL                    use_arbfp_fixed_func;
310     struct wine_rb_tree     fragment_shaders;
311     BOOL                    last_ps_const_clamped;
312     BOOL                    last_vs_color_unclamp;
313
314     struct wine_rb_tree     signature_tree;
315     DWORD ps_sig_number;
316 };
317
318 /* GL locking for state handlers is done by the caller. */
319 static BOOL need_rel_addr_const(const struct arb_vshader_private *shader_data,
320         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
321 {
322     if (shader_data->rel_offset) return TRUE;
323     if (!reg_maps->usesmova) return FALSE;
324     return !gl_info->supported[NV_VERTEX_PROGRAM2_OPTION];
325 }
326
327 /* Returns TRUE if result.clip from GL_NV_vertex_program2 should be used and FALSE otherwise */
328 static inline BOOL use_nv_clip(const struct wined3d_gl_info *gl_info)
329 {
330     return gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
331             && !(gl_info->quirks & WINED3D_QUIRK_NV_CLIP_BROKEN);
332 }
333
334 static BOOL need_helper_const(const struct arb_vshader_private *shader_data,
335         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
336 {
337     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) return TRUE;
338     if (!gl_info->supported[NV_VERTEX_PROGRAM]) return TRUE; /* Need to init colors. */
339     if (gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT) return TRUE; /* Load the immval offset. */
340     if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) return TRUE; /* Have to init texcoords. */
341     if (!use_nv_clip(gl_info)) return TRUE; /* Init the clip texcoord */
342     if (reg_maps->usesnrm) return TRUE; /* 0.0 */
343     if (reg_maps->usesrcp) return TRUE; /* EPS */
344     return FALSE;
345 }
346
347 static unsigned int reserved_vs_const(const struct arb_vshader_private *shader_data,
348         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
349 {
350     unsigned int ret = 1;
351     /* We use one PARAM for the pos fixup, and in some cases one to load
352      * some immediate values into the shader. */
353     if (need_helper_const(shader_data, reg_maps, gl_info)) ++ret;
354     if (need_rel_addr_const(shader_data, reg_maps, gl_info)) ++ret;
355     return ret;
356 }
357
358 /* Loads floating point constants into the currently set ARB_vertex/fragment_program.
359  * When constant_list == NULL, it will load all the constants.
360  *
361  * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
362  *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
363  */
364 /* GL locking is done by the caller */
365 static unsigned int shader_arb_load_constantsF(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
366         GLuint target_type, unsigned int max_constants, const float *constants, char *dirty_consts)
367 {
368     local_constant* lconst;
369     DWORD i, j;
370     unsigned int ret;
371
372     if (TRACE_ON(d3d_constants))
373     {
374         for(i = 0; i < max_constants; i++) {
375             if(!dirty_consts[i]) continue;
376             TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
377                         constants[i * 4 + 0], constants[i * 4 + 1],
378                         constants[i * 4 + 2], constants[i * 4 + 3]);
379         }
380     }
381
382     i = 0;
383
384     /* In 1.X pixel shaders constants are implicitly clamped in the range [-1;1] */
385     if (target_type == GL_FRAGMENT_PROGRAM_ARB && This->baseShader.reg_maps.shader_version.major == 1)
386     {
387         float lcl_const[4];
388         /* ps 1.x supports only 8 constants, clamp only those. When switching between 1.x and higher
389          * shaders, the first 8 constants are marked dirty for reload
390          */
391         for(; i < min(8, max_constants); i++) {
392             if(!dirty_consts[i]) continue;
393             dirty_consts[i] = 0;
394
395             j = 4 * i;
396             if (constants[j + 0] > 1.0f) lcl_const[0] = 1.0f;
397             else if (constants[j + 0] < -1.0f) lcl_const[0] = -1.0f;
398             else lcl_const[0] = constants[j + 0];
399
400             if (constants[j + 1] > 1.0f) lcl_const[1] = 1.0f;
401             else if (constants[j + 1] < -1.0f) lcl_const[1] = -1.0f;
402             else lcl_const[1] = constants[j + 1];
403
404             if (constants[j + 2] > 1.0f) lcl_const[2] = 1.0f;
405             else if (constants[j + 2] < -1.0f) lcl_const[2] = -1.0f;
406             else lcl_const[2] = constants[j + 2];
407
408             if (constants[j + 3] > 1.0f) lcl_const[3] = 1.0f;
409             else if (constants[j + 3] < -1.0f) lcl_const[3] = -1.0f;
410             else lcl_const[3] = constants[j + 3];
411
412             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, lcl_const));
413         }
414
415         /* If further constants are dirty, reload them without clamping.
416          *
417          * The alternative is not to touch them, but then we cannot reset the dirty constant count
418          * to zero. That's bad for apps that only use PS 1.x shaders, because in that case the code
419          * above would always re-check the first 8 constants since max_constant remains at the init
420          * value
421          */
422     }
423
424     if (gl_info->supported[EXT_GPU_PROGRAM_PARAMETERS])
425     {
426         /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
427          * or just reloading *all* constants at once
428          *
429         GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, i, max_constants, constants + (i * 4)));
430          */
431         for(; i < max_constants; i++) {
432             if(!dirty_consts[i]) continue;
433
434             /* Find the next block of dirty constants */
435             dirty_consts[i] = 0;
436             j = i;
437             for(i++; (i < max_constants) && dirty_consts[i]; i++) {
438                 dirty_consts[i] = 0;
439             }
440
441             GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, constants + (j * 4)));
442         }
443     } else {
444         for(; i < max_constants; i++) {
445             if(dirty_consts[i]) {
446                 dirty_consts[i] = 0;
447                 GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, constants + (i * 4)));
448             }
449         }
450     }
451     checkGLcall("glProgramEnvParameter4fvARB()");
452
453     /* Load immediate constants */
454     if(This->baseShader.load_local_constsF) {
455         if (TRACE_ON(d3d_shader)) {
456             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
457                 GLfloat* values = (GLfloat*)lconst->value;
458                 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
459                         values[0], values[1], values[2], values[3]);
460             }
461         }
462         /* Immediate constants are clamped for 1.X shaders at loading times */
463         ret = 0;
464         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
465             dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
466             ret = max(ret, lconst->idx + 1);
467             GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
468         }
469         checkGLcall("glProgramEnvParameter4fvARB()");
470         return ret; /* The loaded immediate constants need reloading for the next shader */
471     } else {
472         return 0; /* No constants are dirty now */
473     }
474 }
475
476 /**
477  * Loads the texture dimensions for NP2 fixup into the currently set ARB_[vertex/fragment]_programs.
478  */
479 static void shader_arb_load_np2fixup_constants(void *shader_priv,
480         const struct wined3d_gl_info *gl_info, const struct wined3d_state *state)
481 {
482     const struct shader_arb_priv * priv = shader_priv;
483
484     /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
485     if (!use_ps(state)) return;
486
487     if (priv->compiled_fprog && priv->compiled_fprog->np2fixup_info.super.active) {
488         const struct arb_ps_np2fixup_info* const fixup = &priv->compiled_fprog->np2fixup_info;
489         UINT i;
490         WORD active = fixup->super.active;
491         GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
492
493         for (i = 0; active; active >>= 1, ++i)
494         {
495             const IWineD3DBaseTextureImpl *tex = state->textures[i];
496             const unsigned char idx = fixup->super.idx[i];
497             GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
498
499             if (!(active & 1)) continue;
500
501             if (!tex) {
502                 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
503                 continue;
504             }
505
506             if (idx % 2) {
507                 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
508             } else {
509                 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
510             }
511         }
512
513         for (i = 0; i < fixup->super.num_consts; ++i) {
514             GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
515                                                    fixup->offset + i, &np2fixup_constants[i * 4]));
516         }
517     }
518 }
519
520 /* GL locking is done by the caller. */
521 static void shader_arb_ps_local_constants(const struct arb_ps_compiled_shader *gl_shader,
522         const struct wined3d_context *context, const struct wined3d_state *state, float rt_height)
523 {
524     const struct wined3d_gl_info *gl_info = context->gl_info;
525     unsigned char i;
526
527     for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
528     {
529         int texunit = gl_shader->bumpenvmatconst[i].texunit;
530
531         /* The state manager takes care that this function is always called if the bump env matrix changes */
532         const float *data = (const float *)&state->texture_states[texunit][WINED3DTSS_BUMPENVMAT00];
533         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
534                 gl_shader->bumpenvmatconst[i].const_num, data));
535
536         if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
537         {
538             /* WINED3DTSS_BUMPENVLSCALE and WINED3DTSS_BUMPENVLOFFSET are next to each other.
539              * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
540              * don't care about them. The pointers are valid for sure because the stateblock is bigger.
541              * (they're WINED3DTSS_TEXTURETRANSFORMFLAGS and WINED3DTSS_ADDRESSW, so most likely 0 or NaN
542             */
543             const float *scale = (const float *)&state->texture_states[texunit][WINED3DTSS_BUMPENVLSCALE];
544             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
545                     gl_shader->luminanceconst[i].const_num, scale));
546         }
547     }
548     checkGLcall("Load bumpmap consts");
549
550     if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
551     {
552         /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
553         * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
554         * ycorrection.z: 1.0
555         * ycorrection.w: 0.0
556         */
557         float val[4];
558         val[0] = context->render_offscreen ? 0.0f : rt_height;
559         val[1] = context->render_offscreen ? 1.0f : -1.0f;
560         val[2] = 1.0f;
561         val[3] = 0.0f;
562         GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
563         checkGLcall("y correction loading");
564     }
565
566     if (!gl_shader->num_int_consts) return;
567
568     for(i = 0; i < MAX_CONST_I; i++)
569     {
570         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
571         {
572             float val[4];
573             val[0] = (float)state->ps_consts_i[4 * i];
574             val[1] = (float)state->ps_consts_i[4 * i + 1];
575             val[2] = (float)state->ps_consts_i[4 * i + 2];
576             val[3] = -1.0f;
577
578             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
579         }
580     }
581     checkGLcall("Load ps int consts");
582 }
583
584 /* GL locking is done by the caller. */
585 static void shader_arb_vs_local_constants(const struct arb_vs_compiled_shader *gl_shader,
586         const struct wined3d_context *context, const struct wined3d_state *state)
587 {
588     const struct wined3d_gl_info *gl_info = context->gl_info;
589     float position_fixup[4];
590     unsigned char i;
591
592     /* Upload the position fixup */
593     shader_get_position_fixup(context, state, position_fixup);
594     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, position_fixup));
595
596     if (!gl_shader->num_int_consts) return;
597
598     for(i = 0; i < MAX_CONST_I; i++)
599     {
600         if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
601         {
602             float val[4];
603             val[0] = (float)state->vs_consts_i[4 * i];
604             val[1] = (float)state->vs_consts_i[4 * i + 1];
605             val[2] = (float)state->vs_consts_i[4 * i + 2];
606             val[3] = -1.0f;
607
608             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
609         }
610     }
611     checkGLcall("Load vs int consts");
612 }
613
614 /**
615  * Loads the app-supplied constants into the currently set ARB_[vertex/fragment]_programs.
616  *
617  * We only support float constants in ARB at the moment, so don't
618  * worry about the Integers or Booleans
619  */
620 /* GL locking is done by the caller (state handler) */
621 static void shader_arb_load_constants(const struct wined3d_context *context, char usePixelShader, char useVertexShader)
622 {
623     IWineD3DDeviceImpl *device = context->swapchain->device;
624     struct wined3d_stateblock *stateBlock = device->stateBlock;
625     const struct wined3d_gl_info *gl_info = context->gl_info;
626     struct shader_arb_priv *priv = device->shader_priv;
627
628     if (useVertexShader)
629     {
630         IWineD3DBaseShaderImpl *vshader = (IWineD3DBaseShaderImpl *)stateBlock->state.vertex_shader;
631         const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
632
633         /* Load DirectX 9 float constants for vertex shader */
634         device->highest_dirty_vs_const = shader_arb_load_constantsF(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
635                 device->highest_dirty_vs_const, stateBlock->state.vs_consts_f, context->vshader_const_dirty);
636         shader_arb_vs_local_constants(gl_shader, context, &stateBlock->state);
637     }
638
639     if (usePixelShader)
640     {
641         IWineD3DBaseShaderImpl *pshader = (IWineD3DBaseShaderImpl *)stateBlock->state.pixel_shader;
642         const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
643         float rt_height = device->render_targets[0]->currentDesc.Height;
644
645         /* Load DirectX 9 float constants for pixel shader */
646         device->highest_dirty_ps_const = shader_arb_load_constantsF(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
647                 device->highest_dirty_ps_const, stateBlock->state.ps_consts_f, context->pshader_const_dirty);
648         shader_arb_ps_local_constants(gl_shader, context, &stateBlock->state, rt_height);
649     }
650 }
651
652 static void shader_arb_update_float_vertex_constants(IWineD3DDeviceImpl *device, UINT start, UINT count)
653 {
654     struct wined3d_context *context = context_get_current();
655
656     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
657      * context. On a context switch the old context will be fully dirtified */
658     if (!context || context->swapchain->device != device) return;
659
660     memset(context->vshader_const_dirty + start, 1, sizeof(*context->vshader_const_dirty) * count);
661     device->highest_dirty_vs_const = max(device->highest_dirty_vs_const, start + count);
662 }
663
664 static void shader_arb_update_float_pixel_constants(IWineD3DDeviceImpl *device, UINT start, UINT count)
665 {
666     struct wined3d_context *context = context_get_current();
667
668     /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
669      * context. On a context switch the old context will be fully dirtified */
670     if (!context || context->swapchain->device != device) return;
671
672     memset(context->pshader_const_dirty + start, 1, sizeof(*context->pshader_const_dirty) * count);
673     device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, start + count);
674 }
675
676 static DWORD *local_const_mapping(IWineD3DBaseShaderImpl *This)
677 {
678     DWORD *ret;
679     DWORD idx = 0;
680     const local_constant *lconst;
681
682     if(This->baseShader.load_local_constsF || list_empty(&This->baseShader.constantsF)) return NULL;
683
684     ret = HeapAlloc(GetProcessHeap(), 0, sizeof(DWORD) * This->baseShader.limits.constant_float);
685     if(!ret) {
686         ERR("Out of memory\n");
687         return NULL;
688     }
689
690     LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
691         ret[lconst->idx] = idx++;
692     }
693     return ret;
694 }
695
696 /* Generate the variable & register declarations for the ARB_vertex_program output target */
697 static DWORD shader_generate_arb_declarations(IWineD3DBaseShaderImpl *shader,
698         const struct wined3d_shader_reg_maps *reg_maps, struct wined3d_shader_buffer *buffer,
699         const struct wined3d_gl_info *gl_info, DWORD *lconst_map,
700         DWORD *num_clipplanes, struct shader_arb_ctx_priv *ctx)
701 {
702     DWORD i, next_local = 0;
703     char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
704     unsigned max_constantsF;
705     const local_constant *lconst;
706     DWORD map;
707
708     /* In pixel shaders, all private constants are program local, we don't need anything
709      * from program.env. Thus we can advertise the full set of constants in pixel shaders.
710      * If we need a private constant the GL implementation will squeeze it in somewhere
711      *
712      * With vertex shaders we need the posFixup and on some GL implementations 4 helper
713      * immediate values. The posFixup is loaded using program.env for now, so always
714      * subtract one from the number of constants. If the shader uses indirect addressing,
715      * account for the helper const too because we have to declare all availabke d3d constants
716      * and don't know which are actually used.
717      */
718     if (pshader)
719     {
720         max_constantsF = gl_info->limits.arb_ps_native_constants;
721         /* 24 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value. */
722         if (max_constantsF < 24)
723             max_constantsF = gl_info->limits.arb_ps_float_constants;
724     }
725     else
726     {
727         const struct arb_vshader_private *shader_data = shader->baseShader.backend_data;
728         max_constantsF = gl_info->limits.arb_vs_native_constants;
729         /* 96 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value.
730          * Also prevents max_constantsF from becoming less than 0 and
731          * wrapping . */
732         if (max_constantsF < 96)
733             max_constantsF = gl_info->limits.arb_vs_float_constants;
734
735         if (reg_maps->usesrelconstF)
736         {
737             DWORD highest_constf = 0, clip_limit;
738
739             max_constantsF -= reserved_vs_const(shader_data, reg_maps, gl_info);
740             max_constantsF -= count_bits(reg_maps->integer_constants);
741
742             for (i = 0; i < shader->baseShader.limits.constant_float; ++i)
743             {
744                 DWORD idx = i >> 5;
745                 DWORD shift = i & 0x1f;
746                 if(reg_maps->constf[idx] & (1 << shift)) highest_constf = i;
747             }
748
749             if(use_nv_clip(gl_info) && ctx->target_version >= NV2)
750             {
751                 if(ctx->cur_vs_args->super.clip_enabled)
752                     clip_limit = gl_info->limits.clipplanes;
753                 else
754                     clip_limit = 0;
755             }
756             else
757             {
758                 unsigned int mask = ctx->cur_vs_args->clip.boolclip.clipplane_mask;
759                 clip_limit = min(count_bits(mask), 4);
760             }
761             *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
762             max_constantsF -= *num_clipplanes;
763             if(*num_clipplanes < clip_limit)
764             {
765                 WARN("Only %u clipplanes out of %u enabled\n", *num_clipplanes, gl_info->limits.clipplanes);
766             }
767         }
768         else
769         {
770             if (ctx->target_version >= NV2) *num_clipplanes = gl_info->limits.clipplanes;
771             else *num_clipplanes = min(gl_info->limits.clipplanes, 4);
772         }
773     }
774
775     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
776     {
777         if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
778     }
779
780     for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
781     {
782         if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
783     }
784
785     if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
786     {
787         for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
788         {
789             if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
790         }
791     }
792
793     /* Load local constants using the program-local space,
794      * this avoids reloading them each time the shader is used
795      */
796     if (lconst_map)
797     {
798         LIST_FOR_EACH_ENTRY(lconst, &shader->baseShader.constantsF, local_constant, entry)
799         {
800             shader_addline(buffer, "PARAM C%u = program.local[%u];\n", lconst->idx,
801                            lconst_map[lconst->idx]);
802             next_local = max(next_local, lconst_map[lconst->idx] + 1);
803         }
804     }
805
806     /* After subtracting privately used constants from the hardware limit(they are loaded as
807      * local constants), make sure the shader doesn't violate the env constant limit
808      */
809     if(pshader)
810     {
811         max_constantsF = min(max_constantsF, gl_info->limits.arb_ps_float_constants);
812     }
813     else
814     {
815         max_constantsF = min(max_constantsF, gl_info->limits.arb_vs_float_constants);
816     }
817
818     /* Avoid declaring more constants than needed */
819     max_constantsF = min(max_constantsF, shader->baseShader.limits.constant_float);
820
821     /* we use the array-based constants array if the local constants are marked for loading,
822      * because then we use indirect addressing, or when the local constant list is empty,
823      * because then we don't know if we're using indirect addressing or not. If we're hardcoding
824      * local constants do not declare the loaded constants as an array because ARB compilers usually
825      * do not optimize unused constants away
826      */
827     if (reg_maps->usesrelconstF)
828     {
829         /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
830         shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
831                     max_constantsF, max_constantsF - 1);
832     } else {
833         for(i = 0; i < max_constantsF; i++) {
834             DWORD idx, mask;
835             idx = i >> 5;
836             mask = 1 << (i & 0x1f);
837             if (!shader_constant_is_local(shader, i) && (reg_maps->constf[idx] & mask))
838             {
839                 shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
840             }
841         }
842     }
843
844     return next_local;
845 }
846
847 static const char * const shift_tab[] = {
848     "dummy",     /*  0 (none) */
849     "coefmul.x", /*  1 (x2)   */
850     "coefmul.y", /*  2 (x4)   */
851     "coefmul.z", /*  3 (x8)   */
852     "coefmul.w", /*  4 (x16)  */
853     "dummy",     /*  5 (x32)  */
854     "dummy",     /*  6 (x64)  */
855     "dummy",     /*  7 (x128) */
856     "dummy",     /*  8 (d256) */
857     "dummy",     /*  9 (d128) */
858     "dummy",     /* 10 (d64)  */
859     "dummy",     /* 11 (d32)  */
860     "coefdiv.w", /* 12 (d16)  */
861     "coefdiv.z", /* 13 (d8)   */
862     "coefdiv.y", /* 14 (d4)   */
863     "coefdiv.x"  /* 15 (d2)   */
864 };
865
866 static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
867         const struct wined3d_shader_dst_param *dst, char *write_mask)
868 {
869     char *ptr = write_mask;
870
871     if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
872     {
873         *ptr++ = '.';
874         if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
875         if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
876         if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
877         if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
878     }
879
880     *ptr = '\0';
881 }
882
883 static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
884 {
885     /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
886      * but addressed as "rgba". To fix this we need to swap the register's x
887      * and z components. */
888     const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
889     char *ptr = swizzle_str;
890
891     /* swizzle bits fields: wwzzyyxx */
892     DWORD swizzle = param->swizzle;
893     DWORD swizzle_x = swizzle & 0x03;
894     DWORD swizzle_y = (swizzle >> 2) & 0x03;
895     DWORD swizzle_z = (swizzle >> 4) & 0x03;
896     DWORD swizzle_w = (swizzle >> 6) & 0x03;
897
898     /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
899      * generate a swizzle string. Unless we need to our own swizzling. */
900     if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
901     {
902         *ptr++ = '.';
903         if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
904             *ptr++ = swizzle_chars[swizzle_x];
905         } else {
906             *ptr++ = swizzle_chars[swizzle_x];
907             *ptr++ = swizzle_chars[swizzle_y];
908             *ptr++ = swizzle_chars[swizzle_z];
909             *ptr++ = swizzle_chars[swizzle_w];
910         }
911     }
912
913     *ptr = '\0';
914 }
915
916 static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
917 {
918     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
919     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
920
921     if (!strcmp(priv->addr_reg, src)) return;
922
923     strcpy(priv->addr_reg, src);
924     shader_addline(buffer, "ARL A0.x, %s;\n", src);
925 }
926
927 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
928         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
929
930 static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
931         const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
932 {
933     /* oPos, oFog and oPts in D3D */
934     static const char * const rastout_reg_names[] = {"TMP_OUT", "result.fogcoord", "result.pointsize"};
935     struct IWineD3DBaseShaderImpl *shader = ins->ctx->shader;
936     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
937     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
938     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
939
940     *is_color = FALSE;
941
942     switch (reg->type)
943     {
944         case WINED3DSPR_TEMP:
945             sprintf(register_name, "R%u", reg->idx);
946             break;
947
948         case WINED3DSPR_INPUT:
949             if (pshader)
950             {
951                 if (reg_maps->shader_version.major < 3)
952                 {
953                     if (!reg->idx) strcpy(register_name, "fragment.color.primary");
954                     else strcpy(register_name, "fragment.color.secondary");
955                 }
956                 else
957                 {
958                     if(reg->rel_addr)
959                     {
960                         char rel_reg[50];
961                         shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
962
963                         if (!strcmp(rel_reg, "**aL_emul**"))
964                         {
965                             DWORD idx = ctx->aL + reg->idx;
966                             if(idx < MAX_REG_INPUT)
967                             {
968                                 strcpy(register_name, ctx->ps_input[idx]);
969                             }
970                             else
971                             {
972                                 ERR("Pixel shader input register out of bounds: %u\n", idx);
973                                 sprintf(register_name, "out_of_bounds_%u", idx);
974                             }
975                         }
976                         else if (reg_maps->input_registers & 0x0300)
977                         {
978                             /* There are two ways basically:
979                              *
980                              * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
981                              *    That means trouble if the loop also contains a breakc or if the control values
982                              *    aren't local constants.
983                              * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
984                              *    source dynamically. The trouble is that we cannot simply read aL.y because it
985                              *    is an ADDRESS register. We could however push it, load .zw with a value and use
986                              *    ADAC to load the condition code register and pop it again afterwards
987                              */
988                             FIXME("Relative input register addressing with more than 8 registers\n");
989
990                             /* This is better than nothing for now */
991                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
992                         }
993                         else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
994                         {
995                             /* This is problematic because we'd have to consult the ctx->ps_input strings
996                              * for where to find the varying. Some may be "0.0", others can be texcoords or
997                              * colors. This needs either a pipeline replacement to make the vertex shader feed
998                              * proper varyings, or loop unrolling
999                              *
1000                              * For now use the texcoords and hope for the best
1001                              */
1002                             FIXME("Non-vertex shader varying input with indirect addressing\n");
1003                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1004                         }
1005                         else
1006                         {
1007                             /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
1008                              * pulls GL_NV_fragment_program2 in
1009                              */
1010                             sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx);
1011                         }
1012                     }
1013                     else
1014                     {
1015                         if(reg->idx < MAX_REG_INPUT)
1016                         {
1017                             strcpy(register_name, ctx->ps_input[reg->idx]);
1018                         }
1019                         else
1020                         {
1021                             ERR("Pixel shader input register out of bounds: %u\n", reg->idx);
1022                             sprintf(register_name, "out_of_bounds_%u", reg->idx);
1023                         }
1024                     }
1025                 }
1026             }
1027             else
1028             {
1029                 if (ctx->cur_vs_args->super.swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1030                 sprintf(register_name, "vertex.attrib[%u]", reg->idx);
1031             }
1032             break;
1033
1034         case WINED3DSPR_CONST:
1035             if (!pshader && reg->rel_addr)
1036             {
1037                 const struct arb_vshader_private *shader_data = shader->baseShader.backend_data;
1038                 UINT rel_offset = shader_data->rel_offset;
1039                 BOOL aL = FALSE;
1040                 char rel_reg[50];
1041                 if (reg_maps->shader_version.major < 2)
1042                 {
1043                     sprintf(rel_reg, "A0.x");
1044                 } else {
1045                     shader_arb_get_src_param(ins, reg->rel_addr, 0, rel_reg);
1046                     if(ctx->target_version == ARB) {
1047                         if (!strcmp(rel_reg, "**aL_emul**"))
1048                         {
1049                             aL = TRUE;
1050                         } else {
1051                             shader_arb_request_a0(ins, rel_reg);
1052                             sprintf(rel_reg, "A0.x");
1053                         }
1054                     }
1055                 }
1056                 if(aL)
1057                     sprintf(register_name, "C[%u]", ctx->aL + reg->idx);
1058                 else if (reg->idx >= rel_offset)
1059                     sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx - rel_offset);
1060                 else
1061                     sprintf(register_name, "C[%s - %u]", rel_reg, rel_offset - reg->idx);
1062             }
1063             else
1064             {
1065                 if (reg_maps->usesrelconstF)
1066                     sprintf(register_name, "C[%u]", reg->idx);
1067                 else
1068                     sprintf(register_name, "C%u", reg->idx);
1069             }
1070             break;
1071
1072         case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1073             if (pshader)
1074             {
1075                 if (reg_maps->shader_version.major == 1
1076                         && reg_maps->shader_version.minor <= 3)
1077                 {
1078                     /* In ps <= 1.3, Tx is a temporary register as destination to all instructions,
1079                      * and as source to most instructions. For some instructions it is the texcoord
1080                      * input. Those instructions know about the special use
1081                      */
1082                     sprintf(register_name, "T%u", reg->idx);
1083                 } else {
1084                     /* in ps 1.4 and 2.x Tx is always a (read-only) varying */
1085                     sprintf(register_name, "fragment.texcoord[%u]", reg->idx);
1086                 }
1087             }
1088             else
1089             {
1090                 if (reg_maps->shader_version.major == 1 || ctx->target_version >= NV2)
1091                 {
1092                     sprintf(register_name, "A%u", reg->idx);
1093                 }
1094                 else
1095                 {
1096                     sprintf(register_name, "A%u_SHADOW", reg->idx);
1097                 }
1098             }
1099             break;
1100
1101         case WINED3DSPR_COLOROUT:
1102             if (ctx->cur_ps_args->super.srgb_correction && !reg->idx)
1103             {
1104                 strcpy(register_name, "TMP_COLOR");
1105             }
1106             else
1107             {
1108                 if(ctx->cur_ps_args->super.srgb_correction) FIXME("sRGB correction on higher render targets\n");
1109                 if (reg_maps->highest_render_target > 0)
1110                 {
1111                     sprintf(register_name, "result.color[%u]", reg->idx);
1112                 }
1113                 else
1114                 {
1115                     strcpy(register_name, "result.color");
1116                 }
1117             }
1118             break;
1119
1120         case WINED3DSPR_RASTOUT:
1121             if(reg->idx == 1) sprintf(register_name, "%s", ctx->fog_output);
1122             else sprintf(register_name, "%s", rastout_reg_names[reg->idx]);
1123             break;
1124
1125         case WINED3DSPR_DEPTHOUT:
1126             strcpy(register_name, "result.depth");
1127             break;
1128
1129         case WINED3DSPR_ATTROUT:
1130         /* case WINED3DSPR_OUTPUT: */
1131             if (pshader) sprintf(register_name, "oD[%u]", reg->idx);
1132             else strcpy(register_name, ctx->color_output[reg->idx]);
1133             break;
1134
1135         case WINED3DSPR_TEXCRDOUT:
1136             if (pshader)
1137             {
1138                 sprintf(register_name, "oT[%u]", reg->idx);
1139             }
1140             else
1141             {
1142                 if (reg_maps->shader_version.major < 3)
1143                 {
1144                     strcpy(register_name, ctx->texcrd_output[reg->idx]);
1145                 }
1146                 else
1147                 {
1148                     strcpy(register_name, ctx->vs_output[reg->idx]);
1149                 }
1150             }
1151             break;
1152
1153         case WINED3DSPR_LOOP:
1154             if(ctx->target_version >= NV2)
1155             {
1156                 /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
1157                 if(pshader) sprintf(register_name, "A0.x");
1158                 else sprintf(register_name, "aL.y");
1159             }
1160             else
1161             {
1162                 /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
1163                  * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
1164                  * thus the result would be something like C[15 + 30], which is not valid in the ARB program
1165                  * grammar. So return a marker for the emulated aL and intercept it in constant and varying
1166                  * indexing
1167                  */
1168                 sprintf(register_name, "**aL_emul**");
1169             }
1170
1171             break;
1172
1173         case WINED3DSPR_CONSTINT:
1174             sprintf(register_name, "I%u", reg->idx);
1175             break;
1176
1177         case WINED3DSPR_MISCTYPE:
1178             if (!reg->idx)
1179             {
1180                 sprintf(register_name, "vpos");
1181             }
1182             else if(reg->idx == 1)
1183             {
1184                 sprintf(register_name, "fragment.facing.x");
1185             }
1186             else
1187             {
1188                 FIXME("Unknown MISCTYPE register index %u\n", reg->idx);
1189             }
1190             break;
1191
1192         default:
1193             FIXME("Unhandled register type %#x[%u]\n", reg->type, reg->idx);
1194             sprintf(register_name, "unrecognized_register[%u]", reg->idx);
1195             break;
1196     }
1197 }
1198
1199 static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
1200         const struct wined3d_shader_dst_param *wined3d_dst, char *str)
1201 {
1202     char register_name[255];
1203     char write_mask[6];
1204     BOOL is_color;
1205
1206     shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
1207     strcpy(str, register_name);
1208
1209     shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
1210     strcat(str, write_mask);
1211 }
1212
1213 static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
1214 {
1215     switch(channel_source)
1216     {
1217         case CHANNEL_SOURCE_ZERO: return "0";
1218         case CHANNEL_SOURCE_ONE: return "1";
1219         case CHANNEL_SOURCE_X: return "x";
1220         case CHANNEL_SOURCE_Y: return "y";
1221         case CHANNEL_SOURCE_Z: return "z";
1222         case CHANNEL_SOURCE_W: return "w";
1223         default:
1224             FIXME("Unhandled channel source %#x\n", channel_source);
1225             return "undefined";
1226     }
1227 }
1228
1229 static void gen_color_correction(struct wined3d_shader_buffer *buffer, const char *reg,
1230         DWORD dst_mask, const char *one, const char *two, struct color_fixup_desc fixup)
1231 {
1232     DWORD mask;
1233
1234     if (is_complex_fixup(fixup))
1235     {
1236         enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1237         FIXME("Complex fixup (%#x) not supported\n", complex_fixup);
1238         return;
1239     }
1240
1241     mask = 0;
1242     if (fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1243     if (fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1244     if (fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1245     if (fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1246     mask &= dst_mask;
1247
1248     if (mask)
1249     {
1250         shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", reg, reg,
1251                 shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
1252                 shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
1253     }
1254
1255     mask = 0;
1256     if (fixup.x_sign_fixup) mask |= WINED3DSP_WRITEMASK_0;
1257     if (fixup.y_sign_fixup) mask |= WINED3DSP_WRITEMASK_1;
1258     if (fixup.z_sign_fixup) mask |= WINED3DSP_WRITEMASK_2;
1259     if (fixup.w_sign_fixup) mask |= WINED3DSP_WRITEMASK_3;
1260     mask &= dst_mask;
1261
1262     if (mask)
1263     {
1264         char reg_mask[6];
1265         char *ptr = reg_mask;
1266
1267         if (mask != WINED3DSP_WRITEMASK_ALL)
1268         {
1269             *ptr++ = '.';
1270             if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1271             if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1272             if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1273             if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1274         }
1275         *ptr = '\0';
1276
1277         shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", reg, reg_mask, reg, two, one);
1278     }
1279 }
1280
1281 static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
1282 {
1283     DWORD mod;
1284     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1285     if (!ins->dst_count) return "";
1286
1287     mod = ins->dst[0].modifiers;
1288
1289     /* Silently ignore PARTIALPRECISION if its not supported */
1290     if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
1291
1292     if(mod & WINED3DSPDM_MSAMPCENTROID)
1293     {
1294         FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
1295         mod &= ~WINED3DSPDM_MSAMPCENTROID;
1296     }
1297
1298     switch(mod)
1299     {
1300         case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
1301             return "H_SAT";
1302
1303         case WINED3DSPDM_SATURATE:
1304             return "_SAT";
1305
1306         case WINED3DSPDM_PARTIALPRECISION:
1307             return "H";
1308
1309         case 0:
1310             return "";
1311
1312         default:
1313             FIXME("Unknown modifiers 0x%08x\n", mod);
1314             return "";
1315     }
1316 }
1317
1318 #define TEX_PROJ        0x1
1319 #define TEX_BIAS        0x2
1320 #define TEX_LOD         0x4
1321 #define TEX_DERIV       0x10
1322
1323 static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
1324         const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
1325 {
1326     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1327     DWORD sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
1328     IWineD3DBaseTextureImpl *texture;
1329     const char *tex_type;
1330     BOOL np2_fixup = FALSE;
1331     struct IWineD3DBaseShaderImpl *shader = ins->ctx->shader;
1332     IWineD3DDeviceImpl *device = shader->baseShader.device;
1333     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1334     const char *mod;
1335     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
1336
1337     /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
1338     if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
1339
1340     switch(sampler_type) {
1341         case WINED3DSTT_1D:
1342             tex_type = "1D";
1343             break;
1344
1345         case WINED3DSTT_2D:
1346             texture = device->stateBlock->state.textures[sampler_idx];
1347             if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
1348             {
1349                 tex_type = "RECT";
1350             } else {
1351                 tex_type = "2D";
1352             }
1353             if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
1354             {
1355                 if (priv->cur_np2fixup_info->super.active & (1 << sampler_idx))
1356                 {
1357                     if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
1358                     else np2_fixup = TRUE;
1359                 }
1360             }
1361             break;
1362
1363         case WINED3DSTT_VOLUME:
1364             tex_type = "3D";
1365             break;
1366
1367         case WINED3DSTT_CUBE:
1368             tex_type = "CUBE";
1369             break;
1370
1371         default:
1372             ERR("Unexpected texture type %d\n", sampler_type);
1373             tex_type = "";
1374     }
1375
1376     /* TEX, TXL, TXD and TXP do not support the "H" modifier,
1377      * so don't use shader_arb_get_modifier
1378      */
1379     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
1380     else mod = "";
1381
1382     /* Fragment samplers always have indentity mapping */
1383     if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
1384     {
1385         sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
1386     }
1387
1388     if (flags & TEX_DERIV)
1389     {
1390         if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
1391         if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
1392         shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1393                        dsx, dsy,sampler_idx, tex_type);
1394     }
1395     else if(flags & TEX_LOD)
1396     {
1397         if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
1398         if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
1399         shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg,
1400                        sampler_idx, tex_type);
1401     }
1402     else if (flags & TEX_BIAS)
1403     {
1404         /* Shouldn't be possible, but let's check for it */
1405         if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
1406         /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
1407         shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1408     }
1409     else if (flags & TEX_PROJ)
1410     {
1411         shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1412     }
1413     else
1414     {
1415         if (np2_fixup)
1416         {
1417             const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
1418             shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
1419                            (idx % 2) ? "zwxy" : "xyzw", coord_reg);
1420
1421             shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, dst_str, sampler_idx, tex_type);
1422         }
1423         else
1424             shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, dst_str, coord_reg, sampler_idx, tex_type);
1425     }
1426
1427     if (pshader)
1428     {
1429         gen_color_correction(buffer, dst_str, ins->dst[0].write_mask,
1430                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_ONE),
1431                 arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_TWO),
1432                 priv->cur_ps_args->super.color_fixup[sampler_idx]);
1433     }
1434 }
1435
1436 static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
1437         const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
1438 {
1439     /* Generate a line that does the input modifier computation and return the input register to use */
1440     BOOL is_color = FALSE;
1441     char regstr[256];
1442     char swzstr[20];
1443     int insert_line;
1444     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1445     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1446     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
1447     const char *two = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_TWO);
1448
1449     /* Assume a new line will be added */
1450     insert_line = 1;
1451
1452     /* Get register name */
1453     shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
1454     shader_arb_get_swizzle(src, is_color, swzstr);
1455
1456     switch (src->modifiers)
1457     {
1458     case WINED3DSPSM_NONE:
1459         sprintf(outregstr, "%s%s", regstr, swzstr);
1460         insert_line = 0;
1461         break;
1462     case WINED3DSPSM_NEG:
1463         sprintf(outregstr, "-%s%s", regstr, swzstr);
1464         insert_line = 0;
1465         break;
1466     case WINED3DSPSM_BIAS:
1467         shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
1468         break;
1469     case WINED3DSPSM_BIASNEG:
1470         shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
1471         break;
1472     case WINED3DSPSM_SIGN:
1473         shader_addline(buffer, "MAD T%c, %s, %s, -%s;\n", 'A' + tmpreg, regstr, two, one);
1474         break;
1475     case WINED3DSPSM_SIGNNEG:
1476         shader_addline(buffer, "MAD T%c, %s, %s, %s;\n", 'A' + tmpreg, regstr, two, one);
1477         break;
1478     case WINED3DSPSM_COMP:
1479         shader_addline(buffer, "SUB T%c, %s, %s;\n", 'A' + tmpreg, one, regstr);
1480         break;
1481     case WINED3DSPSM_X2:
1482         shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
1483         break;
1484     case WINED3DSPSM_X2NEG:
1485         shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
1486         break;
1487     case WINED3DSPSM_DZ:
1488         shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
1489         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1490         break;
1491     case WINED3DSPSM_DW:
1492         shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
1493         shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
1494         break;
1495     case WINED3DSPSM_ABS:
1496         if(ctx->target_version >= NV2) {
1497             sprintf(outregstr, "|%s%s|", regstr, swzstr);
1498             insert_line = 0;
1499         } else {
1500             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1501         }
1502         break;
1503     case WINED3DSPSM_ABSNEG:
1504         if(ctx->target_version >= NV2) {
1505             sprintf(outregstr, "-|%s%s|", regstr, swzstr);
1506         } else {
1507             shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
1508             sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
1509         }
1510         insert_line = 0;
1511         break;
1512     default:
1513         sprintf(outregstr, "%s%s", regstr, swzstr);
1514         insert_line = 0;
1515     }
1516
1517     /* Return modified or original register, with swizzle */
1518     if (insert_line)
1519         sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
1520 }
1521
1522 static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
1523 {
1524     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1525     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1526     char dst_name[50];
1527     char src_name[2][50];
1528     DWORD sampler_code = dst->reg.idx;
1529
1530     shader_arb_get_dst_param(ins, dst, dst_name);
1531
1532     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
1533      *
1534      * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
1535      * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
1536      * temps is done.
1537      */
1538     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1539     shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
1540     shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
1541     shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
1542     shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
1543
1544     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1545     shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
1546 }
1547
1548 static DWORD negate_modifiers(DWORD mod, char *extra_char)
1549 {
1550     *extra_char = ' ';
1551     switch(mod)
1552     {
1553         case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
1554         case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
1555         case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
1556         case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
1557         case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
1558         case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
1559         case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
1560         case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
1561         case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
1562         case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
1563         case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
1564         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
1565         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
1566     }
1567     FIXME("Unknown modifier %u\n", mod);
1568     return mod;
1569 }
1570
1571 static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
1572 {
1573     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1574     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1575     char dst_name[50];
1576     char src_name[3][50];
1577     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1578             ins->ctx->reg_maps->shader_version.minor);
1579     BOOL is_color;
1580
1581     shader_arb_get_dst_param(ins, dst, dst_name);
1582     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1583
1584     /* The coissue flag changes the semantic of the cnd instruction in <= 1.3 shaders */
1585     if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue)
1586     {
1587         shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
1588     } else {
1589         struct wined3d_shader_src_param src0_copy = ins->src[0];
1590         char extra_neg;
1591
1592         /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
1593         src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
1594
1595         shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
1596         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1597         shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
1598         /* No modifiers supported on CMP */
1599         shader_addline(buffer, "CMP %s, TA, %s, %s;\n", dst_name, src_name[1], src_name[2]);
1600
1601         /* _SAT on CMP doesn't make much sense, but it is not a pure NOP */
1602         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1603         {
1604             shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1605             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, dst_name);
1606         }
1607     }
1608 }
1609
1610 static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
1611 {
1612     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1613     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1614     char dst_name[50];
1615     char src_name[3][50];
1616     BOOL is_color;
1617
1618     shader_arb_get_dst_param(ins, dst, dst_name);
1619
1620     /* Generate input register names (with modifiers) */
1621     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1622     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1623     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1624
1625     /* No modifiers are supported on CMP */
1626     shader_addline(buffer, "CMP %s, %s, %s, %s;\n", dst_name,
1627                    src_name[0], src_name[2], src_name[1]);
1628
1629     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
1630     {
1631         shader_arb_get_register_name(ins, &dst->reg, src_name[0], &is_color);
1632         shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name[0]);
1633     }
1634 }
1635
1636 /** Process the WINED3DSIO_DP2ADD instruction in ARB.
1637  * dst = dot2(src0, src1) + src2 */
1638 static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
1639 {
1640     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1641     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1642     char dst_name[50];
1643     char src_name[3][50];
1644     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1645
1646     shader_arb_get_dst_param(ins, dst, dst_name);
1647     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
1648     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
1649
1650     if(ctx->target_version >= NV3)
1651     {
1652         /* GL_NV_fragment_program2 has a 1:1 matching instruction */
1653         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1654         shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1655                        dst_name, src_name[0], src_name[1], src_name[2]);
1656     }
1657     else if(ctx->target_version >= NV2)
1658     {
1659         /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
1660          * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
1661          * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
1662          * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
1663          *
1664          * Make sure that src1.zw = src1.xy, then we get a classic dp2add
1665          *
1666          * .xyxy and other swizzles that we could get with this are not valid in
1667          * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
1668          */
1669         struct wined3d_shader_src_param tmp_param = ins->src[1];
1670         DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
1671         tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
1672
1673         shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
1674
1675         shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
1676                        dst_name, src_name[2], src_name[0], src_name[1]);
1677     }
1678     else
1679     {
1680         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
1681         /* 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]
1682         * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
1683         */
1684         shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
1685         shader_addline(buffer, "MOV TA.z, 0.0;\n");
1686         shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
1687         shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
1688     }
1689 }
1690
1691 /* Map the opcode 1-to-1 to the GL code */
1692 static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
1693 {
1694     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1695     const char *instruction;
1696     char arguments[256], dst_str[50];
1697     unsigned int i;
1698     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1699
1700     switch (ins->handler_idx)
1701     {
1702         case WINED3DSIH_ABS: instruction = "ABS"; break;
1703         case WINED3DSIH_ADD: instruction = "ADD"; break;
1704         case WINED3DSIH_CRS: instruction = "XPD"; break;
1705         case WINED3DSIH_DP3: instruction = "DP3"; break;
1706         case WINED3DSIH_DP4: instruction = "DP4"; break;
1707         case WINED3DSIH_DST: instruction = "DST"; break;
1708         case WINED3DSIH_FRC: instruction = "FRC"; break;
1709         case WINED3DSIH_LIT: instruction = "LIT"; break;
1710         case WINED3DSIH_LRP: instruction = "LRP"; break;
1711         case WINED3DSIH_MAD: instruction = "MAD"; break;
1712         case WINED3DSIH_MAX: instruction = "MAX"; break;
1713         case WINED3DSIH_MIN: instruction = "MIN"; break;
1714         case WINED3DSIH_MOV: instruction = "MOV"; break;
1715         case WINED3DSIH_MUL: instruction = "MUL"; break;
1716         case WINED3DSIH_SGE: instruction = "SGE"; break;
1717         case WINED3DSIH_SLT: instruction = "SLT"; break;
1718         case WINED3DSIH_SUB: instruction = "SUB"; break;
1719         case WINED3DSIH_MOVA:instruction = "ARR"; break;
1720         case WINED3DSIH_DSX: instruction = "DDX"; break;
1721         default: instruction = "";
1722             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
1723             break;
1724     }
1725
1726     /* Note that shader_arb_add_dst_param() adds spaces. */
1727     arguments[0] = '\0';
1728     shader_arb_get_dst_param(ins, dst, dst_str);
1729     for (i = 0; i < ins->src_count; ++i)
1730     {
1731         char operand[100];
1732         strcat(arguments, ", ");
1733         shader_arb_get_src_param(ins, &ins->src[i], i, operand);
1734         strcat(arguments, operand);
1735     }
1736     shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
1737 }
1738
1739 static void shader_hw_nop(const struct wined3d_shader_instruction *ins)
1740 {
1741     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1742     shader_addline(buffer, "NOP;\n");
1743 }
1744
1745 static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
1746 {
1747     struct IWineD3DBaseShaderImpl *shader = ins->ctx->shader;
1748     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1749     BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1750     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
1751     const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
1752     const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
1753     const char *two = arb_get_helper_value(reg_maps->shader_version.type, ARB_TWO);
1754
1755     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1756     char src0_param[256];
1757
1758     if (ins->handler_idx == WINED3DSIH_MOVA)
1759     {
1760         const struct arb_vshader_private *shader_data = shader->baseShader.backend_data;
1761         char write_mask[6];
1762         const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1763
1764         if(ctx->target_version >= NV2) {
1765             shader_hw_map2gl(ins);
1766             return;
1767         }
1768         shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1769         shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
1770
1771         /* This implements the mova formula used in GLSL. The first two instructions
1772          * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
1773          * in this case:
1774          * mova A0.x, 0.0
1775          *
1776          * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
1777          *
1778          * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
1779          * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
1780          */
1781         shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
1782         shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
1783
1784         shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
1785         shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
1786         shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
1787         if (shader_data->rel_offset)
1788         {
1789             shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
1790         }
1791         shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
1792
1793         ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
1794     }
1795     else if (reg_maps->shader_version.major == 1
1796           && !shader_is_pshader_version(reg_maps->shader_version.type)
1797           && ins->dst[0].reg.type == WINED3DSPR_ADDR)
1798     {
1799         const struct arb_vshader_private *shader_data = shader->baseShader.backend_data;
1800         src0_param[0] = '\0';
1801
1802         if (shader_data->rel_offset)
1803         {
1804             const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
1805             shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
1806             shader_addline(buffer, "ADD TA.x, %s, %s;\n", src0_param, offset);
1807             shader_addline(buffer, "ARL A0.x, TA.x;\n");
1808         }
1809         else
1810         {
1811             /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
1812              * with more than one component. Thus replicate the first source argument over all
1813              * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
1814             struct wined3d_shader_src_param tmp_src = ins->src[0];
1815             tmp_src.swizzle = (tmp_src.swizzle & 0x3) * 0x55;
1816             shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
1817             shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
1818         }
1819     }
1820     else if (ins->dst[0].reg.type == WINED3DSPR_COLOROUT && !ins->dst[0].reg.idx && pshader)
1821     {
1822         IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) shader;
1823         if(ctx->cur_ps_args->super.srgb_correction && ps->color0_mov)
1824         {
1825             shader_addline(buffer, "#mov handled in srgb write code\n");
1826             return;
1827         }
1828         shader_hw_map2gl(ins);
1829     }
1830     else
1831     {
1832         shader_hw_map2gl(ins);
1833     }
1834 }
1835
1836 static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
1837 {
1838     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1839     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1840     char reg_dest[40];
1841
1842     /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
1843      * but >= 2.0 honors it(undocumented, but tested by the d3d9 testsuit)
1844      */
1845     shader_arb_get_dst_param(ins, dst, reg_dest);
1846
1847     if (ins->ctx->reg_maps->shader_version.major >= 2)
1848     {
1849         const char *kilsrc = "TA";
1850         BOOL is_color;
1851
1852         shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
1853         if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
1854         {
1855             kilsrc = reg_dest;
1856         }
1857         else
1858         {
1859             /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
1860              * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
1861              * masked out components to 0(won't kill)
1862              */
1863             char x = '0', y = '0', z = '0', w = '0';
1864             if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
1865             if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
1866             if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
1867             if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
1868             shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
1869         }
1870         shader_addline(buffer, "KIL %s;\n", kilsrc);
1871     } else {
1872         /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
1873          * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
1874          *
1875          * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
1876          * or pass in any temporary register(in shader phase 2)
1877          */
1878         if(ins->ctx->reg_maps->shader_version.minor <= 3) {
1879             sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx);
1880         } else {
1881             shader_arb_get_dst_param(ins, dst, reg_dest);
1882         }
1883         shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
1884         shader_addline(buffer, "KIL TA;\n");
1885     }
1886 }
1887
1888 static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
1889 {
1890     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1891     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1892     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1893             ins->ctx->reg_maps->shader_version.minor);
1894     struct wined3d_shader_src_param src;
1895
1896     char reg_dest[40];
1897     char reg_coord[40];
1898     DWORD reg_sampler_code;
1899     WORD myflags = 0;
1900
1901     /* All versions have a destination register */
1902     shader_arb_get_dst_param(ins, dst, reg_dest);
1903
1904     /* 1.0-1.4: Use destination register number as texture code.
1905        2.0+: Use provided sampler number as texure code. */
1906     if (shader_version < WINED3D_SHADER_VERSION(2,0))
1907         reg_sampler_code = dst->reg.idx;
1908     else
1909         reg_sampler_code = ins->src[1].reg.idx;
1910
1911     /* 1.0-1.3: Use the texcoord varying.
1912        1.4+: Use provided coordinate source register. */
1913     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1914         sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
1915     else {
1916         /* TEX is the only instruction that can handle DW and DZ natively */
1917         src = ins->src[0];
1918         if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
1919         if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
1920         shader_arb_get_src_param(ins, &src, 0, reg_coord);
1921     }
1922
1923     /* projection flag:
1924      * 1.1, 1.2, 1.3: Use WINED3DTSS_TEXTURETRANSFORMFLAGS
1925      * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
1926      * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
1927      */
1928     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1929     {
1930         DWORD flags = 0;
1931         if (reg_sampler_code < MAX_TEXTURES)
1932             flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1933         if (flags & WINED3D_PSARGS_PROJECTED)
1934             myflags |= TEX_PROJ;
1935     }
1936     else if (shader_version < WINED3D_SHADER_VERSION(2,0))
1937     {
1938         DWORD src_mod = ins->src[0].modifiers;
1939         if (src_mod == WINED3DSPSM_DZ) {
1940             /* TXP cannot handle DZ natively, so move the z coordinate to .w. reg_coord is a read-only
1941              * varying register, so we need a temp reg
1942              */
1943             shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
1944             strcpy(reg_coord, "TA");
1945             myflags |= TEX_PROJ;
1946         } else if(src_mod == WINED3DSPSM_DW) {
1947             myflags |= TEX_PROJ;
1948         }
1949     } else {
1950         if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
1951         if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
1952     }
1953     shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
1954 }
1955
1956 static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
1957 {
1958     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
1959     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1960     DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
1961             ins->ctx->reg_maps->shader_version.minor);
1962     char dst_str[50];
1963
1964     if (shader_version < WINED3D_SHADER_VERSION(1,4))
1965     {
1966         DWORD reg = dst->reg.idx;
1967
1968         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1969         shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
1970     } else {
1971         char reg_src[40];
1972
1973         shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
1974         shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1975         shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
1976    }
1977 }
1978
1979 static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
1980 {
1981      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
1982      DWORD flags = 0;
1983
1984      DWORD reg1 = ins->dst[0].reg.idx;
1985      char dst_str[50];
1986      char src_str[50];
1987
1988      /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
1989      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
1990      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
1991      /* Move .x first in case src_str is "TA" */
1992      shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
1993      shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
1994      if (reg1 < MAX_TEXTURES)
1995      {
1996          struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
1997          flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
1998      }
1999      shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2000 }
2001
2002 static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
2003 {
2004      struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2005
2006      DWORD reg1 = ins->dst[0].reg.idx;
2007      char dst_str[50];
2008      char src_str[50];
2009
2010      /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
2011      shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2012      shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2013      shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
2014      shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
2015      shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
2016 }
2017
2018 static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
2019 {
2020     DWORD reg1 = ins->dst[0].reg.idx;
2021     char dst_str[50];
2022     char src_str[50];
2023
2024     /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
2025     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2026     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
2027     shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
2028 }
2029
2030 static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
2031 {
2032     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2033     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2034     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2035     char reg_coord[40], dst_reg[50], src_reg[50];
2036     DWORD reg_dest_code;
2037
2038     /* All versions have a destination register. The Tx where the texture coordinates come
2039      * from is the varying incarnation of the texture register
2040      */
2041     reg_dest_code = dst->reg.idx;
2042     shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
2043     shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
2044     sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
2045
2046     /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
2047      * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
2048      *
2049      * GL_NV_fragment_program_option could handle this in one instruction via X2D:
2050      * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
2051      *
2052      * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
2053      * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
2054      * extension.
2055      */
2056     shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
2057     shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
2058     shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
2059     shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
2060
2061     /* with projective textures, texbem only divides the static texture coord, not the displacement,
2062      * so we can't let the GL handle this.
2063      */
2064     if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
2065             & WINED3D_PSARGS_PROJECTED)
2066     {
2067         shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
2068         shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
2069         shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
2070     } else {
2071         shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
2072     }
2073
2074     shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
2075
2076     if (ins->handler_idx == WINED3DSIH_TEXBEML)
2077     {
2078         /* No src swizzles are allowed, so this is ok */
2079         shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
2080                        src_reg, reg_dest_code, reg_dest_code);
2081         shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
2082     }
2083 }
2084
2085 static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
2086 {
2087     DWORD reg = ins->dst[0].reg.idx;
2088     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2089     char src0_name[50], dst_name[50];
2090     BOOL is_color;
2091     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2092
2093     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2094     /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
2095      * T<reg+1> register. Use this register to store the calculated vector
2096      */
2097     tmp_reg.idx = reg + 1;
2098     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2099     shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2100 }
2101
2102 static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
2103 {
2104     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2105     DWORD flags;
2106     DWORD reg = ins->dst[0].reg.idx;
2107     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2108     char dst_str[50];
2109     char src0_name[50];
2110     char dst_reg[50];
2111     BOOL is_color;
2112
2113     /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
2114     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2115
2116     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2117     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2118     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2119     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2120     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2121 }
2122
2123 static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
2124 {
2125     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2126     DWORD reg = ins->dst[0].reg.idx;
2127     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2128     char src0_name[50], dst_name[50];
2129     struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
2130     BOOL is_color;
2131
2132     /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
2133      * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
2134      * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
2135      */
2136     tmp_reg.idx = reg + 2 - tex_mx->current_row;
2137     shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
2138
2139     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2140     shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
2141                    dst_name, 'x' + tex_mx->current_row, reg, src0_name);
2142     tex_mx->texcoord_w[tex_mx->current_row++] = reg;
2143 }
2144
2145 static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
2146 {
2147     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2148     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2149     DWORD flags;
2150     DWORD reg = ins->dst[0].reg.idx;
2151     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2152     char dst_str[50];
2153     char src0_name[50], dst_name[50];
2154     BOOL is_color;
2155
2156     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2157     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2158     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
2159
2160     /* Sample the texture using the calculated coordinates */
2161     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2162     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2163     shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2164     tex_mx->current_row = 0;
2165 }
2166
2167 static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2168 {
2169     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2170     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2171     DWORD flags;
2172     DWORD reg = ins->dst[0].reg.idx;
2173     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2174     char dst_str[50];
2175     char src0_name[50];
2176     char dst_reg[50];
2177     BOOL is_color;
2178
2179     /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
2180      * components for temporary data storage
2181      */
2182     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2183     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2184     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2185
2186     /* Construct the eye-ray vector from w coordinates */
2187     shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
2188     shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
2189     shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
2190
2191     /* Calculate reflection vector
2192      */
2193     shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
2194     /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
2195     shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
2196     shader_addline(buffer, "RCP TB.w, TB.w;\n");
2197     shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
2198     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2199     shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
2200
2201     /* Sample the texture using the calculated coordinates */
2202     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2203     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2204     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2205     tex_mx->current_row = 0;
2206 }
2207
2208 static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
2209 {
2210     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2211     struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
2212     DWORD flags;
2213     DWORD reg = ins->dst[0].reg.idx;
2214     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2215     char dst_str[50];
2216     char src0_name[50];
2217     char src1_name[50];
2218     char dst_reg[50];
2219     BOOL is_color;
2220
2221     shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
2222     shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
2223     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
2224     /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
2225     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
2226
2227     /* Calculate reflection vector.
2228      *
2229      *                   dot(N, E)
2230      * dst_reg.xyz = 2 * --------- * N - E
2231      *                   dot(N, N)
2232      *
2233      * Which normalizes the normal vector
2234      */
2235     shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
2236     shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
2237     shader_addline(buffer, "RCP TC.w, TC.w;\n");
2238     shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
2239     shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
2240     shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
2241
2242     /* Sample the texture using the calculated coordinates */
2243     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2244     flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
2245     shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3DTTFF_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
2246     tex_mx->current_row = 0;
2247 }
2248
2249 static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
2250 {
2251     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2252     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2253     char dst_name[50];
2254     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2255     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2256
2257     /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
2258      * which is essentially an input, is the destination register because it is the first
2259      * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
2260      * here(writemasks/swizzles are not valid on texdepth)
2261      */
2262     shader_arb_get_dst_param(ins, dst, dst_name);
2263
2264     /* According to the msdn, the source register(must be r5) is unusable after
2265      * the texdepth instruction, so we're free to modify it
2266      */
2267     shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
2268
2269     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2270      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2271      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2272      */
2273     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2274     shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
2275     shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
2276     shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
2277 }
2278
2279 /** Process the WINED3DSIO_TEXDP3TEX instruction in ARB:
2280  * Take a 3-component dot product of the TexCoord[dstreg] and src,
2281  * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2282 static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
2283 {
2284     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2285     DWORD sampler_idx = ins->dst[0].reg.idx;
2286     char src0[50];
2287     char dst_str[50];
2288
2289     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2290     shader_addline(buffer, "MOV TB, 0.0;\n");
2291     shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
2292
2293     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
2294     shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
2295 }
2296
2297 /** Process the WINED3DSIO_TEXDP3 instruction in ARB:
2298  * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2299 static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
2300 {
2301     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2302     char src0[50];
2303     char dst_str[50];
2304     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2305
2306     /* Handle output register */
2307     shader_arb_get_dst_param(ins, dst, dst_str);
2308     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2309     shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx, src0);
2310 }
2311
2312 /** Process the WINED3DSIO_TEXM3X3 instruction in ARB
2313  * Perform the 3rd row of a 3x3 matrix multiply */
2314 static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
2315 {
2316     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2317     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2318     char dst_str[50], dst_name[50];
2319     char src0[50];
2320     BOOL is_color;
2321
2322     shader_arb_get_dst_param(ins, dst, dst_str);
2323     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2324     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2325     shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2326     shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
2327 }
2328
2329 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in ARB:
2330  * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2331  * Calculate tmp0.y = TexCoord[dstreg] . src.xyz;  (tmp0.x has already been calculated)
2332  * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2333  */
2334 static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
2335 {
2336     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2337     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2338     char src0[50], dst_name[50];
2339     BOOL is_color;
2340     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2341     const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2342
2343     shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
2344     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2345     shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx, src0);
2346
2347     /* How to deal with the special case dst_name.g == 0? if r != 0, then
2348      * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
2349      * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
2350      */
2351     shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
2352     shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
2353     shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
2354     shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
2355 }
2356
2357 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2358     Vertex/Pixel shaders to ARB_vertex_program codes */
2359 static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
2360 {
2361     int i;
2362     int nComponents = 0;
2363     struct wined3d_shader_dst_param tmp_dst = {{0}};
2364     struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2365     struct wined3d_shader_instruction tmp_ins;
2366
2367     memset(&tmp_ins, 0, sizeof(tmp_ins));
2368
2369     /* Set constants for the temporary argument */
2370     tmp_ins.ctx = ins->ctx;
2371     tmp_ins.dst_count = 1;
2372     tmp_ins.dst = &tmp_dst;
2373     tmp_ins.src_count = 2;
2374     tmp_ins.src = tmp_src;
2375
2376     switch(ins->handler_idx)
2377     {
2378         case WINED3DSIH_M4x4:
2379             nComponents = 4;
2380             tmp_ins.handler_idx = WINED3DSIH_DP4;
2381             break;
2382         case WINED3DSIH_M4x3:
2383             nComponents = 3;
2384             tmp_ins.handler_idx = WINED3DSIH_DP4;
2385             break;
2386         case WINED3DSIH_M3x4:
2387             nComponents = 4;
2388             tmp_ins.handler_idx = WINED3DSIH_DP3;
2389             break;
2390         case WINED3DSIH_M3x3:
2391             nComponents = 3;
2392             tmp_ins.handler_idx = WINED3DSIH_DP3;
2393             break;
2394         case WINED3DSIH_M3x2:
2395             nComponents = 2;
2396             tmp_ins.handler_idx = WINED3DSIH_DP3;
2397             break;
2398         default:
2399             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2400             break;
2401     }
2402
2403     tmp_dst = ins->dst[0];
2404     tmp_src[0] = ins->src[0];
2405     tmp_src[1] = ins->src[1];
2406     for (i = 0; i < nComponents; i++) {
2407         tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2408         shader_hw_map2gl(&tmp_ins);
2409         ++tmp_src[1].reg.idx;
2410     }
2411 }
2412
2413 static void shader_hw_rcp(const struct wined3d_shader_instruction *ins)
2414 {
2415     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2416     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2417     const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
2418
2419     char dst[50];
2420     char src[50];
2421
2422     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2423     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2424     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2425     {
2426         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2427          * .w is used
2428          */
2429         strcat(src, ".w");
2430     }
2431
2432     /* TODO: If the destination is readable, and not the same as the source, the destination
2433      * can be used instead of TA
2434      */
2435     if (priv->target_version >= NV2)
2436     {
2437         shader_addline(buffer, "MOVC TA.x, %s;\n", src);
2438         shader_addline(buffer, "MOV TA.x (EQ.x), %s;\n", flt_eps);
2439         shader_addline(buffer, "RCP%s %s, TA.x;\n", shader_arb_get_modifier(ins), dst);
2440     }
2441     else
2442     {
2443         const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2444         shader_addline(buffer, "ABS TA.x, %s;\n", src);
2445         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2446         shader_addline(buffer, "MAD TA.x, TA.y, %s, %s;\n", flt_eps, src);
2447         shader_addline(buffer, "RCP%s %s, TA.x;\n", shader_arb_get_modifier(ins), dst);
2448     }
2449 }
2450
2451 static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
2452 {
2453     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2454     const char *instruction;
2455
2456     char dst[50];
2457     char src[50];
2458
2459     switch(ins->handler_idx)
2460     {
2461         case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
2462         case WINED3DSIH_RCP:  instruction = "RCP"; break;
2463         case WINED3DSIH_EXP:  instruction = "EX2"; break;
2464         case WINED3DSIH_EXPP: instruction = "EXP"; break;
2465         default: instruction = "";
2466             FIXME("Unhandled opcode %#x\n", ins->handler_idx);
2467             break;
2468     }
2469
2470     shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
2471     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2472     if (ins->src[0].swizzle == WINED3DSP_NOSWIZZLE)
2473     {
2474         /* Dx sdk says .x is used if no swizzle is given, but our test shows that
2475          * .w is used
2476          */
2477         strcat(src, ".w");
2478     }
2479
2480     shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
2481 }
2482
2483 static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
2484 {
2485     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2486     char dst_name[50];
2487     char src_name[50];
2488     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2489     BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
2490     const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
2491
2492     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2493     shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
2494
2495     /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
2496      * otherwise NRM or RSQ would return NaN */
2497     if(pshader && priv->target_version >= NV3)
2498     {
2499         /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
2500          *
2501          * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
2502          */
2503         shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
2504         shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2505         shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
2506     }
2507     else if(priv->target_version >= NV2)
2508     {
2509         shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
2510         shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
2511         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2512                        src_name);
2513     }
2514     else
2515     {
2516         const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
2517
2518         shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
2519         /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
2520          * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
2521          */
2522         shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
2523         shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
2524
2525         shader_addline(buffer, "RSQ TA.x, TA.x;\n");
2526         /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
2527         shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
2528                     src_name);
2529     }
2530 }
2531
2532 static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
2533 {
2534     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2535     char dst_name[50];
2536     char src_name[3][50];
2537
2538     /* ARB_fragment_program has a convenient LRP instruction */
2539     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2540         shader_hw_map2gl(ins);
2541         return;
2542     }
2543
2544     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2545     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
2546     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
2547     shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
2548
2549     shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
2550     shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
2551                    dst_name, src_name[0], src_name[2]);
2552 }
2553
2554 static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
2555 {
2556     /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
2557      * must contain fixed constants. So we need a separate function to filter those constants and
2558      * can't use map2gl
2559      */
2560     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2561     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2562     const struct wined3d_shader_dst_param *dst = &ins->dst[0];
2563     char dst_name[50];
2564     char src_name0[50], src_name1[50], src_name2[50];
2565     BOOL is_color;
2566
2567     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2568     if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
2569         shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2570         /* No modifiers are supported on SCS */
2571         shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
2572
2573         if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
2574         {
2575             shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
2576             shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
2577         }
2578     } else if(priv->target_version >= NV2) {
2579         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2580
2581         /* Sincos writemask must be .x, .y or .xy */
2582         if(dst->write_mask & WINED3DSP_WRITEMASK_0)
2583             shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2584         if(dst->write_mask & WINED3DSP_WRITEMASK_1)
2585             shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
2586     } else {
2587         /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
2588          * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
2589          *
2590          * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
2591          * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
2592          *
2593          * The constants we get are:
2594          *
2595          *  +1   +1,     -1     -1     +1      +1      -1       -1
2596          *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
2597          *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
2598          *
2599          * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
2600          *
2601          * (x/2)^2 = x^2 / 4
2602          * (x/2)^3 = x^3 / 8
2603          * (x/2)^4 = x^4 / 16
2604          * (x/2)^5 = x^5 / 32
2605          * etc
2606          *
2607          * To get the final result:
2608          * sin(x) = 2 * sin(x/2) * cos(x/2)
2609          * cos(x) = cos(x/2)^2 - sin(x/2)^2
2610          * (from sin(x+y) and cos(x+y) rules)
2611          *
2612          * As per MSDN, dst.z is undefined after the operation, and so is
2613          * dst.x and dst.y if they're masked out by the writemask. Ie
2614          * sincos dst.y, src1, c0, c1
2615          * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
2616          * vsa.exe also stops with an error if the dest register is the same register as the source
2617          * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
2618          * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
2619          */
2620         shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2621         shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
2622         shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
2623
2624         shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
2625         shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
2626         shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
2627         shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
2628         shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
2629         shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
2630
2631         /* sin(x/2)
2632          *
2633          * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
2634          * properly merge that with MULs in the code above?
2635          * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
2636          * we can merge the sine and cosine MAD rows to calculate them together.
2637          */
2638         shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
2639         shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
2640         shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
2641         shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
2642
2643         /* cos(x/2) */
2644         shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
2645         shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
2646         shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
2647
2648         if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
2649             /* cos x */
2650             shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
2651             shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
2652         }
2653         if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
2654             /* sin x */
2655             shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
2656             shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
2657         }
2658     }
2659 }
2660
2661 static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
2662 {
2663     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2664     char dst_name[50];
2665     char src_name[50];
2666     struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
2667
2668     shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
2669     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2670
2671     /* SGN is only valid in vertex shaders */
2672     if(ctx->target_version >= NV2) {
2673         shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
2674         return;
2675     }
2676
2677     /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
2678      * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
2679      */
2680     if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
2681         shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
2682     } else {
2683         /* src contains TA? Write to the dest first. This won't overwrite our destination.
2684          * Then use TA, and calculate the final result
2685          *
2686          * Not reading from TA? Store the first result in TA to avoid overwriting the
2687          * destination if src reg = dst reg
2688          */
2689         if(strstr(src_name, "TA"))
2690         {
2691             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2692             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2693             shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
2694         }
2695         else
2696         {
2697             shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
2698             shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
2699             shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
2700         }
2701     }
2702 }
2703
2704 static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
2705 {
2706     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2707     char src[50];
2708     char dst[50];
2709     char dst_name[50];
2710     BOOL is_color;
2711
2712     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2713     shader_arb_get_src_param(ins, &ins->src[0], 0, src);
2714     shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
2715
2716     shader_addline(buffer, "DDY %s, %s;\n", dst, src);
2717     shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
2718 }
2719
2720 static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
2721 {
2722     *need_abs = FALSE;
2723
2724     switch(mod)
2725     {
2726         case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
2727         case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
2728         case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
2729         case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
2730         case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
2731         case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
2732         case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
2733         case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
2734         case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
2735         case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
2736         case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
2737         case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
2738         case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
2739     }
2740     FIXME("Unknown modifier %u\n", mod);
2741     return mod;
2742 }
2743
2744 static void shader_hw_log_pow(const struct wined3d_shader_instruction *ins)
2745 {
2746     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2747     char src0[50], src1[50], dst[50];
2748     struct wined3d_shader_src_param src0_copy = ins->src[0];
2749     BOOL need_abs = FALSE;
2750     const char *instr;
2751     BOOL arg2 = FALSE;
2752
2753     switch(ins->handler_idx)
2754     {
2755         case WINED3DSIH_LOG:  instr = "LG2"; break;
2756         case WINED3DSIH_LOGP: instr = "LOG"; break;
2757         case WINED3DSIH_POW:  instr = "POW"; arg2 = TRUE; break;
2758         default:
2759             ERR("Unexpected instruction %d\n", ins->handler_idx);
2760             return;
2761     }
2762
2763     /* LOG, LOGP and POW operate on the absolute value of the input */
2764     src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
2765
2766     shader_arb_get_dst_param(ins, &ins->dst[0], dst);
2767     shader_arb_get_src_param(ins, &src0_copy, 0, src0);
2768     if(arg2) shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
2769
2770     if(need_abs)
2771     {
2772         shader_addline(buffer, "ABS TA, %s;\n", src0);
2773         if(arg2)
2774         {
2775             shader_addline(buffer, "%s%s %s, TA, %s;\n", instr, shader_arb_get_modifier(ins), dst, src1);
2776         }
2777         else
2778         {
2779             shader_addline(buffer, "%s%s %s, TA;\n", instr, shader_arb_get_modifier(ins), dst);
2780         }
2781     }
2782     else if(arg2)
2783     {
2784         shader_addline(buffer, "%s%s %s, %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0, src1);
2785     }
2786     else
2787     {
2788         shader_addline(buffer, "%s%s %s, %s;\n", instr, shader_arb_get_modifier(ins), dst, src0);
2789     }
2790 }
2791
2792 static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
2793 {
2794     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2795     char src_name[50];
2796     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2797
2798     /* src0 is aL */
2799     shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
2800
2801     if(vshader)
2802     {
2803         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2804         struct list *e = list_head(&priv->control_frames);
2805         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2806
2807         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2808         /* The constant loader makes sure to load -1 into iX.w */
2809         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2810         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2811         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2812     }
2813     else
2814     {
2815         shader_addline(buffer, "LOOP %s;\n", src_name);
2816     }
2817 }
2818
2819 static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
2820 {
2821     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2822     char src_name[50];
2823     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2824
2825     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
2826
2827     /* The constant loader makes sure to load -1 into iX.w */
2828     if(vshader)
2829     {
2830         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2831         struct list *e = list_head(&priv->control_frames);
2832         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2833
2834         if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
2835
2836         shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
2837         shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
2838         shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
2839     }
2840     else
2841     {
2842         shader_addline(buffer, "REP %s;\n", src_name);
2843     }
2844 }
2845
2846 static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
2847 {
2848     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2849     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2850
2851     if(vshader)
2852     {
2853         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2854         struct list *e = list_head(&priv->control_frames);
2855         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2856
2857         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2858         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2859         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2860
2861         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2862     }
2863     else
2864     {
2865         shader_addline(buffer, "ENDLOOP;\n");
2866     }
2867 }
2868
2869 static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
2870 {
2871     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2872     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2873
2874     if(vshader)
2875     {
2876         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2877         struct list *e = list_head(&priv->control_frames);
2878         struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2879
2880         shader_addline(buffer, "ARAC aL.xy, aL;\n");
2881         shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
2882         shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
2883
2884         if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
2885     }
2886     else
2887     {
2888         shader_addline(buffer, "ENDREP;\n");
2889     }
2890 }
2891
2892 static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
2893 {
2894     struct control_frame *control_frame;
2895
2896     LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
2897     {
2898         if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
2899     }
2900     ERR("Could not find loop for break\n");
2901     return NULL;
2902 }
2903
2904 static void shader_hw_break(const struct wined3d_shader_instruction *ins)
2905 {
2906     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2907     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2908     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2909
2910     if(vshader)
2911     {
2912         shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
2913     }
2914     else
2915     {
2916         shader_addline(buffer, "BRK;\n");
2917     }
2918 }
2919
2920 static const char *get_compare(COMPARISON_TYPE flags)
2921 {
2922     switch (flags)
2923     {
2924         case COMPARISON_GT: return "GT";
2925         case COMPARISON_EQ: return "EQ";
2926         case COMPARISON_GE: return "GE";
2927         case COMPARISON_LT: return "LT";
2928         case COMPARISON_NE: return "NE";
2929         case COMPARISON_LE: return "LE";
2930         default:
2931             FIXME("Unrecognized comparison value: %u\n", flags);
2932             return "(\?\?)";
2933     }
2934 }
2935
2936 static COMPARISON_TYPE invert_compare(COMPARISON_TYPE flags)
2937 {
2938     switch (flags)
2939     {
2940         case COMPARISON_GT: return COMPARISON_LE;
2941         case COMPARISON_EQ: return COMPARISON_NE;
2942         case COMPARISON_GE: return COMPARISON_LT;
2943         case COMPARISON_LT: return COMPARISON_GE;
2944         case COMPARISON_NE: return COMPARISON_EQ;
2945         case COMPARISON_LE: return COMPARISON_GT;
2946         default:
2947             FIXME("Unrecognized comparison value: %u\n", flags);
2948             return -1;
2949     }
2950 }
2951
2952 static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
2953 {
2954     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2955     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2956     const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
2957     char src_name0[50];
2958     char src_name1[50];
2959     const char *comp = get_compare(ins->flags);
2960
2961     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2962     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2963
2964     if(vshader)
2965     {
2966         /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
2967          * away the subtraction result
2968          */
2969         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2970         shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
2971     }
2972     else
2973     {
2974         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2975         shader_addline(buffer, "BRK (%s.x);\n", comp);
2976     }
2977 }
2978
2979 static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
2980 {
2981     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2982     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
2983     struct list *e = list_head(&priv->control_frames);
2984     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
2985     const char *comp;
2986     char src_name0[50];
2987     char src_name1[50];
2988     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
2989
2990     shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
2991     shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
2992
2993     if(vshader)
2994     {
2995         /* Invert the flag. We jump to the else label if the condition is NOT true */
2996         comp = get_compare(invert_compare(ins->flags));
2997         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
2998         shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
2999     }
3000     else
3001     {
3002         comp = get_compare(ins->flags);
3003         shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
3004         shader_addline(buffer, "IF %s.x;\n", comp);
3005     }
3006 }
3007
3008 static void shader_hw_else(const struct wined3d_shader_instruction *ins)
3009 {
3010     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3011     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3012     struct list *e = list_head(&priv->control_frames);
3013     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3014     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3015
3016     if(vshader)
3017     {
3018         shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
3019         shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3020         control_frame->had_else = TRUE;
3021     }
3022     else
3023     {
3024         shader_addline(buffer, "ELSE;\n");
3025     }
3026 }
3027
3028 static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
3029 {
3030     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3031     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3032     struct list *e = list_head(&priv->control_frames);
3033     struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
3034     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3035
3036     if(vshader)
3037     {
3038         if(control_frame->had_else)
3039         {
3040             shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
3041         }
3042         else
3043         {
3044             shader_addline(buffer, "#No else branch. else is endif\n");
3045             shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
3046         }
3047     }
3048     else
3049     {
3050         shader_addline(buffer, "ENDIF;\n");
3051     }
3052 }
3053
3054 static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
3055 {
3056     DWORD sampler_idx = ins->src[1].reg.idx;
3057     char reg_dest[40];
3058     char reg_src[3][40];
3059     WORD flags = TEX_DERIV;
3060
3061     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3062     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
3063     shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
3064     shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
3065
3066     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3067     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3068
3069     shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
3070 }
3071
3072 static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
3073 {
3074     DWORD sampler_idx = ins->src[1].reg.idx;
3075     char reg_dest[40];
3076     char reg_coord[40];
3077     WORD flags = TEX_LOD;
3078
3079     shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
3080     shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
3081
3082     if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
3083     if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
3084
3085     shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
3086 }
3087
3088 static void shader_hw_label(const struct wined3d_shader_instruction *ins)
3089 {
3090     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3091     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3092
3093     priv->in_main_func = FALSE;
3094     /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
3095      * subroutine, don't generate a label that will make GL complain
3096      */
3097     if(priv->target_version == ARB) return;
3098
3099     shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx);
3100 }
3101
3102 static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
3103         const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
3104         const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
3105         struct wined3d_shader_buffer *buffer)
3106 {
3107     unsigned int i;
3108
3109     /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
3110      * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
3111      * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
3112      * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
3113      */
3114     if(args->super.fog_src == VS_FOG_Z) {
3115         shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
3116     } else if (!reg_maps->fog) {
3117         /* posFixup.x is always 1.0, so we can savely use it */
3118         shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
3119     }
3120
3121     /* Write the final position.
3122      *
3123      * OpenGL coordinates specify the center of the pixel while d3d coords specify
3124      * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3125      * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3126      * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
3127      */
3128     shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
3129     shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
3130     shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
3131
3132     if(use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
3133     {
3134         if(args->super.clip_enabled)
3135         {
3136             for(i = 0; i < priv_ctx->vs_clipplanes; i++)
3137             {
3138                 shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
3139             }
3140         }
3141     }
3142     else if(args->clip.boolclip.clip_texcoord)
3143     {
3144         unsigned int cur_clip = 0;
3145         char component[4] = {'x', 'y', 'z', 'w'};
3146         const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
3147
3148         for (i = 0; i < gl_info->limits.clipplanes; ++i)
3149         {
3150             if(args->clip.boolclip.clipplane_mask & (1 << i))
3151             {
3152                 shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
3153                                component[cur_clip++], i);
3154             }
3155         }
3156         switch(cur_clip)
3157         {
3158             case 0:
3159                 shader_addline(buffer, "MOV TA, %s;\n", zero);
3160                 break;
3161             case 1:
3162                 shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
3163                 break;
3164             case 2:
3165                 shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
3166                 break;
3167             case 3:
3168                 shader_addline(buffer, "MOV TA.w, %s;\n", zero);
3169                 break;
3170         }
3171         shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
3172                        args->clip.boolclip.clip_texcoord - 1);
3173     }
3174
3175     /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3176      * and the glsl equivalent
3177      */
3178     if (need_helper_const(shader_data, reg_maps, gl_info))
3179     {
3180         const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
3181         shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
3182     } else {
3183         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
3184         shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
3185     }
3186
3187     shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
3188
3189     priv_ctx->footer_written = TRUE;
3190 }
3191
3192 static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
3193 {
3194     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3195     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
3196     struct IWineD3DBaseShaderImpl *shader = ins->ctx->shader;
3197     BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
3198
3199     if(priv->target_version == ARB) return;
3200
3201     if(vshader)
3202     {
3203         if (priv->in_main_func) vshader_add_footer(priv, shader->baseShader.backend_data,
3204                 priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
3205     }
3206
3207     shader_addline(buffer, "RET;\n");
3208 }
3209
3210 static void shader_hw_call(const struct wined3d_shader_instruction *ins)
3211 {
3212     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3213     shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx);
3214 }
3215
3216 /* GL locking is done by the caller */
3217 static GLuint create_arb_blt_vertex_program(const struct wined3d_gl_info *gl_info)
3218 {
3219     GLuint program_id = 0;
3220     GLint pos;
3221
3222     const char *blt_vprogram =
3223         "!!ARBvp1.0\n"
3224         "PARAM c[1] = { { 1, 0.5 } };\n"
3225         "MOV result.position, vertex.position;\n"
3226         "MOV result.color, c[0].x;\n"
3227         "MOV result.texcoord[0], vertex.texcoord[0];\n"
3228         "END\n";
3229
3230     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3231     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, program_id));
3232     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3233             strlen(blt_vprogram), blt_vprogram));
3234     checkGLcall("glProgramStringARB()");
3235
3236     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3237     if (pos != -1)
3238     {
3239         FIXME("Vertex program error at position %d: %s\n\n", pos,
3240             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3241         shader_arb_dump_program_source(blt_vprogram);
3242     }
3243     else
3244     {
3245         GLint native;
3246
3247         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3248         checkGLcall("glGetProgramivARB()");
3249         if (!native) WARN("Program exceeds native resource limits.\n");
3250     }
3251
3252     return program_id;
3253 }
3254
3255 /* GL locking is done by the caller */
3256 static GLuint create_arb_blt_fragment_program(const struct wined3d_gl_info *gl_info,
3257         enum tex_types tex_type, BOOL masked)
3258 {
3259     GLuint program_id = 0;
3260     const char *fprogram;
3261     GLint pos;
3262
3263     static const char * const blt_fprograms_full[tex_type_count] =
3264     {
3265         /* tex_1d */
3266         NULL,
3267         /* tex_2d */
3268         "!!ARBfp1.0\n"
3269         "TEMP R0;\n"
3270         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3271         "MOV result.depth.z, R0.x;\n"
3272         "END\n",
3273         /* tex_3d */
3274         NULL,
3275         /* tex_cube */
3276         "!!ARBfp1.0\n"
3277         "TEMP R0;\n"
3278         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3279         "MOV result.depth.z, R0.x;\n"
3280         "END\n",
3281         /* tex_rect */
3282         "!!ARBfp1.0\n"
3283         "TEMP R0;\n"
3284         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3285         "MOV result.depth.z, R0.x;\n"
3286         "END\n",
3287     };
3288
3289     static const char * const blt_fprograms_masked[tex_type_count] =
3290     {
3291         /* tex_1d */
3292         NULL,
3293         /* tex_2d */
3294         "!!ARBfp1.0\n"
3295         "PARAM mask = program.local[0];\n"
3296         "TEMP R0;\n"
3297         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3298         "MUL R0.x, R0.x, R0.y;\n"
3299         "KIL -R0.x;\n"
3300         "TEX R0.x, fragment.texcoord[0], texture[0], 2D;\n"
3301         "MOV result.depth.z, R0.x;\n"
3302         "END\n",
3303         /* tex_3d */
3304         NULL,
3305         /* tex_cube */
3306         "!!ARBfp1.0\n"
3307         "PARAM mask = program.local[0];\n"
3308         "TEMP R0;\n"
3309         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3310         "MUL R0.x, R0.x, R0.y;\n"
3311         "KIL -R0.x;\n"
3312         "TEX R0.x, fragment.texcoord[0], texture[0], CUBE;\n"
3313         "MOV result.depth.z, R0.x;\n"
3314         "END\n",
3315         /* tex_rect */
3316         "!!ARBfp1.0\n"
3317         "PARAM mask = program.local[0];\n"
3318         "TEMP R0;\n"
3319         "SLT R0.xy, fragment.position, mask.zwzw;\n"
3320         "MUL R0.x, R0.x, R0.y;\n"
3321         "KIL -R0.x;\n"
3322         "TEX R0.x, fragment.texcoord[0], texture[0], RECT;\n"
3323         "MOV result.depth.z, R0.x;\n"
3324         "END\n",
3325     };
3326
3327     fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3328     if (!fprogram)
3329     {
3330         FIXME("tex_type %#x not supported, falling back to tex_2d\n", tex_type);
3331         tex_type = tex_2d;
3332         fprogram = masked ? blt_fprograms_masked[tex_type] : blt_fprograms_full[tex_type];
3333     }
3334
3335     GL_EXTCALL(glGenProgramsARB(1, &program_id));
3336     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, program_id));
3337     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(fprogram), fprogram));
3338     checkGLcall("glProgramStringARB()");
3339
3340     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
3341     if (pos != -1)
3342     {
3343         FIXME("Fragment program error at position %d: %s\n\n", pos,
3344             debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3345         shader_arb_dump_program_source(fprogram);
3346     }
3347     else
3348     {
3349         GLint native;
3350
3351         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3352         checkGLcall("glGetProgramivARB()");
3353         if (!native) WARN("Program exceeds native resource limits.\n");
3354     }
3355
3356     return program_id;
3357 }
3358
3359 static void arbfp_add_sRGB_correction(struct wined3d_shader_buffer *buffer, const char *fragcolor,
3360         const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
3361 {
3362     /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
3363
3364     if(condcode)
3365     {
3366         /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
3367         shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.y;\n", tmp1, fragcolor);
3368         /* Calculate the > 0.0031308 case */
3369         shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts1.z;\n", fragcolor, fragcolor);
3370         shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts1.z;\n", fragcolor, fragcolor);
3371         shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts1.z;\n", fragcolor, fragcolor);
3372         shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts1.w;\n", fragcolor, fragcolor);
3373         shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts2.x;\n", fragcolor, fragcolor);
3374         /* Calculate the < case */
3375         shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts1.x, %s;\n", fragcolor, fragcolor);
3376     }
3377     else
3378     {
3379         /* Calculate the > 0.0031308 case */
3380         shader_addline(buffer, "POW %s.x, %s.x, srgb_consts1.z;\n", tmp1, fragcolor);
3381         shader_addline(buffer, "POW %s.y, %s.y, srgb_consts1.z;\n", tmp1, fragcolor);
3382         shader_addline(buffer, "POW %s.z, %s.z, srgb_consts1.z;\n", tmp1, fragcolor);
3383         shader_addline(buffer, "MUL %s, %s, srgb_consts1.w;\n", tmp1, tmp1);
3384         shader_addline(buffer, "SUB %s, %s, srgb_consts2.x;\n", tmp1, tmp1);
3385         /* Calculate the < case */
3386         shader_addline(buffer, "MUL %s, srgb_consts1.x, %s;\n", tmp2, fragcolor);
3387         /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
3388         shader_addline(buffer, "SLT %s, srgb_consts1.y, %s;\n", tmp3, fragcolor);
3389         shader_addline(buffer, "SGE %s, srgb_consts1.y, %s;\n", tmp4, fragcolor);
3390         /* Store the components > 0.0031308 in the destination */
3391         shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
3392         /* Add the components that are < 0.0031308 */
3393         shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
3394         /* Move everything into result.color at once. Nvidia hardware cannot handle partial
3395         * result.color writes(.rgb first, then .a), or handle overwriting already written
3396         * components. The assembler uses a temporary register in this case, which is usually
3397         * not allocated from one of our registers that were used earlier.
3398         */
3399     }
3400     /* [0.0;1.0] clamping. Not needed, this is done implicitly */
3401 }
3402
3403 static const DWORD *find_loop_control_values(IWineD3DBaseShaderImpl *This, DWORD idx)
3404 {
3405     const local_constant *constant;
3406
3407     LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
3408     {
3409         if (constant->idx == idx)
3410         {
3411             return constant->value;
3412         }
3413     }
3414     return NULL;
3415 }
3416
3417 static void init_ps_input(const IWineD3DPixelShaderImpl *This, const struct arb_ps_compile_args *args,
3418                           struct shader_arb_ctx_priv *priv)
3419 {
3420     static const char * const texcoords[8] =
3421     {
3422         "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
3423         "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
3424     };
3425     unsigned int i;
3426     const struct wined3d_shader_signature_element *sig = This->baseShader.input_signature;
3427     const char *semantic_name;
3428     DWORD semantic_idx;
3429
3430     switch(args->super.vp_mode)
3431     {
3432         case pretransformed:
3433         case fixedfunction:
3434             /* The pixelshader has to collect the varyings on its own. In any case properly load
3435              * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
3436              * other attribs to 0.0.
3437              *
3438              * For fixedfunction this behavior is correct, according to the tests. For pretransformed
3439              * we'd either need a replacement shader that can load other attribs like BINORMAL, or
3440              * load the texcoord attrib pointers to match the pixel shader signature
3441              */
3442             for(i = 0; i < MAX_REG_INPUT; i++)
3443             {
3444                 semantic_name = sig[i].semantic_name;
3445                 semantic_idx = sig[i].semantic_idx;
3446                 if (!semantic_name) continue;
3447
3448                 if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3449                 {
3450                     if (!semantic_idx) priv->ps_input[i] = "fragment.color.primary";
3451                     else if(semantic_idx == 1) priv->ps_input[i] = "fragment.color.secondary";
3452                     else priv->ps_input[i] = "0.0";
3453                 }
3454                 else if(args->super.vp_mode == fixedfunction)
3455                 {
3456                     priv->ps_input[i] = "0.0";
3457                 }
3458                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3459                 {
3460                     if(semantic_idx < 8) priv->ps_input[i] = texcoords[semantic_idx];
3461                     else priv->ps_input[i] = "0.0";
3462                 }
3463                 else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3464                 {
3465                     if (!semantic_idx) priv->ps_input[i] = "fragment.fogcoord";
3466                     else priv->ps_input[i] = "0.0";
3467                 }
3468                 else
3469                 {
3470                     priv->ps_input[i] = "0.0";
3471                 }
3472
3473                 TRACE("v%u, semantic %s%u is %s\n", i, semantic_name, semantic_idx, priv->ps_input[i]);
3474             }
3475             break;
3476
3477         case vertexshader:
3478             /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
3479              * fragment.color
3480              */
3481             for(i = 0; i < 8; i++)
3482             {
3483                 priv->ps_input[i] = texcoords[i];
3484             }
3485             priv->ps_input[8] = "fragment.color.primary";
3486             priv->ps_input[9] = "fragment.color.secondary";
3487             break;
3488     }
3489 }
3490
3491 /* GL locking is done by the caller */
3492 static GLuint shader_arb_generate_pshader(IWineD3DPixelShaderImpl *This,
3493         const struct wined3d_gl_info *gl_info, struct wined3d_shader_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 = &This->baseShader.reg_maps;
3497     CONST DWORD *function = This->baseShader.function;
3498     const local_constant *lconst;
3499     GLuint retval;
3500     char fragcolor[16];
3501     DWORD *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This), next_local, cur;
3502     struct shader_arb_ctx_priv priv_ctx;
3503     BOOL dcl_td = FALSE;
3504     BOOL want_nv_prog = FALSE;
3505     struct arb_pshader_private *shader_priv = This->baseShader.backend_data;
3506     GLint errPos;
3507     DWORD map;
3508
3509     char srgbtmp[4][4];
3510     unsigned int i, found = 0;
3511
3512     for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
3513     {
3514         if (!(map & 1)
3515                 || (This->color0_mov && i == This->color0_reg)
3516                 || (reg_maps->shader_version.major < 2 && !i))
3517             continue;
3518
3519         sprintf(srgbtmp[found], "R%u", i);
3520         ++found;
3521         if (found == 4) break;
3522     }
3523
3524     switch(found) {
3525         case 0:
3526             sprintf(srgbtmp[0], "TA");
3527             sprintf(srgbtmp[1], "TB");
3528             sprintf(srgbtmp[2], "TC");
3529             sprintf(srgbtmp[3], "TD");
3530             dcl_td = TRUE;
3531             break;
3532         case 1:
3533             sprintf(srgbtmp[1], "TA");
3534             sprintf(srgbtmp[2], "TB");
3535             sprintf(srgbtmp[3], "TC");
3536             break;
3537         case 2:
3538             sprintf(srgbtmp[2], "TA");
3539             sprintf(srgbtmp[3], "TB");
3540             break;
3541         case 3:
3542             sprintf(srgbtmp[3], "TA");
3543             break;
3544         case 4:
3545             break;
3546     }
3547
3548     /*  Create the hw ARB shader */
3549     memset(&priv_ctx, 0, sizeof(priv_ctx));
3550     priv_ctx.cur_ps_args = args;
3551     priv_ctx.compiled_fprog = compiled;
3552     priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
3553     init_ps_input(This, args, &priv_ctx);
3554     list_init(&priv_ctx.control_frames);
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->highest_render_target > 0)
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             case FOG_OFF:
3604                 break;
3605             case FOG_LINEAR:
3606                 shader_addline(buffer, "OPTION ARB_fog_linear;\n");
3607                 break;
3608             case FOG_EXP:
3609                 shader_addline(buffer, "OPTION ARB_fog_exp;\n");
3610                 break;
3611             case FOG_EXP2:
3612                 shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
3613                 break;
3614         }
3615     }
3616
3617     /* For now always declare the temps. At least the Nvidia assembler optimizes completely
3618      * unused temps away(but occupies them for the whole shader if they're used once). Always
3619      * declaring them avoids tricky bookkeeping work
3620      */
3621     shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
3622     shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
3623     shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
3624     if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
3625     shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
3626     shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
3627     shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %1.10f, 0.0 };\n", eps);
3628
3629     if (reg_maps->shader_version.major < 2)
3630     {
3631         strcpy(fragcolor, "R0");
3632     } else {
3633         if(args->super.srgb_correction) {
3634             if(This->color0_mov) {
3635                 sprintf(fragcolor, "R%u", This->color0_reg);
3636             } else {
3637                 shader_addline(buffer, "TEMP TMP_COLOR;\n");
3638                 strcpy(fragcolor, "TMP_COLOR");
3639             }
3640         } else {
3641             strcpy(fragcolor, "result.color");
3642         }
3643     }
3644
3645     if(args->super.srgb_correction) {
3646         shader_addline(buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
3647                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
3648         shader_addline(buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
3649                        srgb_sub_high, 0.0, 0.0, 0.0);
3650     }
3651
3652     /* Base Declarations */
3653     next_local = shader_generate_arb_declarations((IWineD3DBaseShaderImpl *)This,
3654             reg_maps, buffer, gl_info, lconst_map, NULL, &priv_ctx);
3655
3656     for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
3657     {
3658         if (!(map & 1)) continue;
3659
3660         cur = compiled->numbumpenvmatconsts;
3661         compiled->bumpenvmatconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3662         compiled->bumpenvmatconst[cur].texunit = i;
3663         compiled->luminanceconst[cur].const_num = WINED3D_CONST_NUM_UNUSED;
3664         compiled->luminanceconst[cur].texunit = i;
3665
3666         /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
3667          * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
3668          * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
3669          * textures due to conditional NP2 restrictions)
3670          *
3671          * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
3672          * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
3673          * their location is shader dependent anyway and they cannot be loaded globally.
3674          */
3675         compiled->bumpenvmatconst[cur].const_num = next_local++;
3676         shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
3677                        i, compiled->bumpenvmatconst[cur].const_num);
3678         compiled->numbumpenvmatconsts = cur + 1;
3679
3680         if (!(reg_maps->luminanceparams & (1 << i))) continue;
3681
3682         compiled->luminanceconst[cur].const_num = next_local++;
3683         shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
3684                        i, compiled->luminanceconst[cur].const_num);
3685     }
3686
3687     for(i = 0; i < MAX_CONST_I; i++)
3688     {
3689         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
3690         if (reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
3691         {
3692             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
3693
3694             if(control_values)
3695             {
3696                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
3697                                 control_values[0], control_values[1], control_values[2]);
3698             }
3699             else
3700             {
3701                 compiled->int_consts[i] = next_local;
3702                 compiled->num_int_consts++;
3703                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
3704             }
3705         }
3706     }
3707
3708     if(reg_maps->vpos || reg_maps->usesdsy)
3709     {
3710         compiled->ycorrection = next_local;
3711         shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
3712
3713         if(reg_maps->vpos)
3714         {
3715             shader_addline(buffer, "TEMP vpos;\n");
3716             /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
3717              * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
3718              * ycorrection.z: 1.0
3719              * ycorrection.w: 0.0
3720              */
3721             shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
3722             shader_addline(buffer, "FLR vpos.xy, vpos;\n");
3723         }
3724     }
3725     else
3726     {
3727         compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
3728     }
3729
3730     /* Load constants to fixup NP2 texcoords if there are still free constants left:
3731      * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
3732      * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
3733      * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
3734      * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
3735      * shader compilation errors and the subsequent errors when drawing with this shader. */
3736     if (priv_ctx.cur_ps_args->super.np2_fixup) {
3737
3738         struct arb_ps_np2fixup_info* const fixup = priv_ctx.cur_np2fixup_info;
3739         const WORD map = priv_ctx.cur_ps_args->super.np2_fixup;
3740         const UINT max_lconsts = gl_info->limits.arb_ps_local_constants;
3741
3742         fixup->offset = next_local;
3743         fixup->super.active = 0;
3744
3745         cur = 0;
3746         for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3747             if (!(map & (1 << i))) continue;
3748
3749             if (fixup->offset + (cur >> 1) < max_lconsts) {
3750                 fixup->super.active |= (1 << i);
3751                 fixup->super.idx[i] = cur++;
3752             } else {
3753                 FIXME("No free constant found to load NP2 fixup data into shader. "
3754                       "Sampling from this texture will probably look wrong.\n");
3755                 break;
3756             }
3757         }
3758
3759         fixup->super.num_consts = (cur + 1) >> 1;
3760         if (fixup->super.num_consts) {
3761             shader_addline(buffer, "PARAM np2fixup[%u] = { program.env[%u..%u] };\n",
3762                            fixup->super.num_consts, fixup->offset, fixup->super.num_consts + fixup->offset - 1);
3763         }
3764
3765         next_local += fixup->super.num_consts;
3766     }
3767
3768     if (shader_priv->clipplane_emulation != ~0U && args->clip)
3769     {
3770         shader_addline(buffer, "KIL fragment.texcoord[%u];\n", shader_priv->clipplane_emulation);
3771     }
3772
3773     /* Base Shader Body */
3774     shader_generate_main((IWineD3DBaseShaderImpl *)This, buffer, reg_maps, function, &priv_ctx);
3775
3776     if(args->super.srgb_correction) {
3777         arbfp_add_sRGB_correction(buffer, fragcolor, srgbtmp[0], srgbtmp[1], srgbtmp[2], srgbtmp[3],
3778                                   priv_ctx.target_version >= NV2);
3779     }
3780
3781     if(strcmp(fragcolor, "result.color")) {
3782         shader_addline(buffer, "MOV result.color, %s;\n", fragcolor);
3783     }
3784     shader_addline(buffer, "END\n");
3785
3786     /* TODO: change to resource.glObjectHandle or something like that */
3787     GL_EXTCALL(glGenProgramsARB(1, &retval));
3788
3789     TRACE("Creating a hw pixel shader, prg=%d\n", retval);
3790     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, retval));
3791
3792     TRACE("Created hw pixel shader, prg=%d\n", retval);
3793     /* Create the program and check for errors */
3794     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
3795                buffer->bsize, buffer->buffer));
3796     checkGLcall("glProgramStringARB()");
3797
3798     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
3799     if (errPos != -1)
3800     {
3801         FIXME("HW PixelShader Error at position %d: %s\n\n",
3802               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
3803         shader_arb_dump_program_source(buffer->buffer);
3804         retval = 0;
3805     }
3806     else
3807     {
3808         GLint native;
3809
3810         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
3811         checkGLcall("glGetProgramivARB()");
3812         if (!native) WARN("Program exceeds native resource limits.\n");
3813     }
3814
3815     /* Load immediate constants */
3816     if(lconst_map) {
3817         LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
3818             const float *value = (const float *)lconst->value;
3819             GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, lconst_map[lconst->idx], value));
3820             checkGLcall("glProgramLocalParameter4fvARB");
3821         }
3822         HeapFree(GetProcessHeap(), 0, lconst_map);
3823     }
3824
3825     return retval;
3826 }
3827
3828 static int compare_sig(const struct wined3d_shader_signature_element *sig1, const struct wined3d_shader_signature_element *sig2)
3829 {
3830     unsigned int i;
3831     int ret;
3832
3833     for(i = 0; i < MAX_REG_INPUT; i++)
3834     {
3835         if (!sig1[i].semantic_name || !sig2[i].semantic_name)
3836         {
3837             /* Compare pointers, not contents. One string is NULL(element does not exist), the other one is not NULL */
3838             if(sig1[i].semantic_name != sig2[i].semantic_name) return sig1[i].semantic_name < sig2[i].semantic_name ? -1 : 1;
3839             continue;
3840         }
3841
3842         if ((ret = strcmp(sig1[i].semantic_name, sig2[i].semantic_name))) return ret;
3843         if(sig1[i].semantic_idx    != sig2[i].semantic_idx)    return sig1[i].semantic_idx    < sig2[i].semantic_idx    ? -1 : 1;
3844         if(sig1[i].sysval_semantic != sig2[i].sysval_semantic) return sig1[i].sysval_semantic < sig2[i].sysval_semantic ? -1 : 1;
3845         if(sig1[i].component_type  != sig2[i].component_type)  return sig1[i].sysval_semantic < sig2[i].component_type  ? -1 : 1;
3846         if(sig1[i].register_idx    != sig2[i].register_idx)    return sig1[i].register_idx    < sig2[i].register_idx    ? -1 : 1;
3847         if(sig1[i].mask            != sig2->mask)              return sig1[i].mask            < sig2[i].mask            ? -1 : 1;
3848     }
3849     return 0;
3850 }
3851
3852 static struct wined3d_shader_signature_element *clone_sig(const struct wined3d_shader_signature_element *sig)
3853 {
3854     struct wined3d_shader_signature_element *new;
3855     int i;
3856     char *name;
3857
3858     new = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*new) * MAX_REG_INPUT);
3859     for(i = 0; i < MAX_REG_INPUT; i++)
3860     {
3861         if (!sig[i].semantic_name) continue;
3862
3863         new[i] = sig[i];
3864         /* Clone the semantic string */
3865         name = HeapAlloc(GetProcessHeap(), 0, strlen(sig[i].semantic_name) + 1);
3866         strcpy(name, sig[i].semantic_name);
3867         new[i].semantic_name = name;
3868     }
3869     return new;
3870 }
3871
3872 static DWORD find_input_signature(struct shader_arb_priv *priv, const struct wined3d_shader_signature_element *sig)
3873 {
3874     struct wine_rb_entry *entry = wine_rb_get(&priv->signature_tree, sig);
3875     struct ps_signature *found_sig;
3876
3877     if (entry)
3878     {
3879         found_sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
3880         TRACE("Found existing signature %u\n", found_sig->idx);
3881         return found_sig->idx;
3882     }
3883     found_sig = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*sig));
3884     found_sig->sig = clone_sig(sig);
3885     found_sig->idx = priv->ps_sig_number++;
3886     TRACE("New signature stored and assigned number %u\n", found_sig->idx);
3887     if(wine_rb_put(&priv->signature_tree, sig, &found_sig->entry) == -1)
3888     {
3889         ERR("Failed to insert program entry.\n");
3890     }
3891     return found_sig->idx;
3892 }
3893
3894 static void init_output_registers(IWineD3DVertexShaderImpl *shader, DWORD sig_num, struct shader_arb_ctx_priv *priv_ctx,
3895                                   struct arb_vs_compiled_shader *compiled)
3896 {
3897     unsigned int i, j;
3898     static const char * const texcoords[8] =
3899     {
3900         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3901         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]"
3902     };
3903     IWineD3DDeviceImpl *device = shader->baseShader.device;
3904     IWineD3DBaseShaderClass *baseshader = &shader->baseShader;
3905     const struct wined3d_shader_signature_element *sig;
3906     const char *semantic_name;
3907     DWORD semantic_idx, reg_idx;
3908
3909     /* Write generic input varyings 0 to 7 to result.texcoord[], varying 8 to result.color.primary
3910      * and varying 9 to result.color.secondary
3911      */
3912     static const char * const decl_idx_to_string[MAX_REG_INPUT] =
3913     {
3914         "result.texcoord[0]", "result.texcoord[1]", "result.texcoord[2]", "result.texcoord[3]",
3915         "result.texcoord[4]", "result.texcoord[5]", "result.texcoord[6]", "result.texcoord[7]",
3916         "result.color.primary", "result.color.secondary"
3917     };
3918
3919     if(sig_num == ~0)
3920     {
3921         TRACE("Pixel shader uses builtin varyings\n");
3922         /* Map builtins to builtins */
3923         for(i = 0; i < 8; i++)
3924         {
3925             priv_ctx->texcrd_output[i] = texcoords[i];
3926         }
3927         priv_ctx->color_output[0] = "result.color.primary";
3928         priv_ctx->color_output[1] = "result.color.secondary";
3929         priv_ctx->fog_output = "result.fogcoord";
3930
3931         /* Map declared regs to builtins. Use "TA" to /dev/null unread output */
3932         for (i = 0; i < (sizeof(baseshader->output_signature) / sizeof(*baseshader->output_signature)); ++i)
3933         {
3934             semantic_name = baseshader->output_signature[i].semantic_name;
3935             if (!semantic_name) continue;
3936
3937             if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3938             {
3939                 TRACE("o%u is TMP_OUT\n", i);
3940                 if (!baseshader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "TMP_OUT";
3941                 else priv_ctx->vs_output[i] = "TA";
3942             }
3943             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3944             {
3945                 TRACE("o%u is result.pointsize\n", i);
3946                 if (!baseshader->output_signature[i].semantic_idx) priv_ctx->vs_output[i] = "result.pointsize";
3947                 else priv_ctx->vs_output[i] = "TA";
3948             }
3949             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3950             {
3951                 TRACE("o%u is result.color.?, idx %u\n", i, baseshader->output_signature[i].semantic_idx);
3952                 if (!baseshader->output_signature[i].semantic_idx)
3953                     priv_ctx->vs_output[i] = "result.color.primary";
3954                 else if (baseshader->output_signature[i].semantic_idx == 1)
3955                     priv_ctx->vs_output[i] = "result.color.secondary";
3956                 else priv_ctx->vs_output[i] = "TA";
3957             }
3958             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3959             {
3960                 TRACE("o%u is %s\n", i, texcoords[baseshader->output_signature[i].semantic_idx]);
3961                 if (baseshader->output_signature[i].semantic_idx >= 8) priv_ctx->vs_output[i] = "TA";
3962                 else priv_ctx->vs_output[i] = texcoords[baseshader->output_signature[i].semantic_idx];
3963             }
3964             else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3965             {
3966                 TRACE("o%u is result.fogcoord\n", i);
3967                 if (baseshader->output_signature[i].semantic_idx > 0) priv_ctx->vs_output[i] = "TA";
3968                 else priv_ctx->vs_output[i] = "result.fogcoord";
3969             }
3970             else
3971             {
3972                 priv_ctx->vs_output[i] = "TA";
3973             }
3974         }
3975         return;
3976     }
3977
3978     /* Instead of searching for the signature in the signature list, read the one from the current pixel shader.
3979      * Its maybe not the shader where the signature came from, but it is the same signature and faster to find
3980      */
3981     sig = device->stateBlock->state.pixel_shader->baseShader.input_signature;
3982     TRACE("Pixel shader uses declared varyings\n");
3983
3984     /* Map builtin to declared. /dev/null the results by default to the TA temp reg */
3985     for(i = 0; i < 8; i++)
3986     {
3987         priv_ctx->texcrd_output[i] = "TA";
3988     }
3989     priv_ctx->color_output[0] = "TA";
3990     priv_ctx->color_output[1] = "TA";
3991     priv_ctx->fog_output = "TA";
3992
3993     for(i = 0; i < MAX_REG_INPUT; i++)
3994     {
3995         semantic_name = sig[i].semantic_name;
3996         semantic_idx = sig[i].semantic_idx;
3997         reg_idx = sig[i].register_idx;
3998         if (!semantic_name) continue;
3999
4000         /* If a declared input register is not written by builtin arguments, don't write to it.
4001          * GL_NV_vertex_program makes sure the input defaults to 0.0, which is correct with D3D
4002          *
4003          * Don't care about POSITION and PSIZE here - this is a builtin vertex shader, position goes
4004          * to TMP_OUT in any case
4005          */
4006         if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
4007         {
4008             if(semantic_idx < 8) priv_ctx->texcrd_output[semantic_idx] = decl_idx_to_string[reg_idx];
4009         }
4010         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
4011         {
4012             if(semantic_idx < 2) priv_ctx->color_output[semantic_idx] = decl_idx_to_string[reg_idx];
4013         }
4014         else if(shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
4015         {
4016             if (!semantic_idx) priv_ctx->fog_output = decl_idx_to_string[reg_idx];
4017         }
4018         else
4019         {
4020             continue;
4021         }
4022
4023         if (!strcmp(decl_idx_to_string[reg_idx], "result.color.primary")
4024                 || !strcmp(decl_idx_to_string[reg_idx], "result.color.secondary"))
4025         {
4026             compiled->need_color_unclamp = TRUE;
4027         }
4028     }
4029
4030     /* Map declared to declared */
4031     for (i = 0; i < (sizeof(baseshader->output_signature) / sizeof(*baseshader->output_signature)); ++i)
4032     {
4033         /* Write unread output to TA to throw them away */
4034         priv_ctx->vs_output[i] = "TA";
4035         semantic_name = baseshader->output_signature[i].semantic_name;
4036         if (!semantic_name) continue;
4037
4038         if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION)
4039                 && !baseshader->output_signature[i].semantic_idx)
4040         {
4041             priv_ctx->vs_output[i] = "TMP_OUT";
4042             continue;
4043         }
4044         else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE)
4045                 && !baseshader->output_signature[i].semantic_idx)
4046         {
4047             priv_ctx->vs_output[i] = "result.pointsize";
4048             continue;
4049         }
4050
4051         for(j = 0; j < MAX_REG_INPUT; j++)
4052         {
4053             if (!sig[j].semantic_name) continue;
4054
4055             if (!strcmp(sig[j].semantic_name, semantic_name)
4056                     && sig[j].semantic_idx == baseshader->output_signature[i].semantic_idx)
4057             {
4058                 priv_ctx->vs_output[i] = decl_idx_to_string[sig[j].register_idx];
4059
4060                 if (!strcmp(priv_ctx->vs_output[i], "result.color.primary")
4061                         || !strcmp(priv_ctx->vs_output[i], "result.color.secondary"))
4062                 {
4063                     compiled->need_color_unclamp = TRUE;
4064                 }
4065             }
4066         }
4067     }
4068 }
4069
4070 /* GL locking is done by the caller */
4071 static GLuint shader_arb_generate_vshader(IWineD3DVertexShaderImpl *This,
4072         const struct wined3d_gl_info *gl_info, struct wined3d_shader_buffer *buffer,
4073         const struct arb_vs_compile_args *args, struct arb_vs_compiled_shader *compiled)
4074 {
4075     const struct arb_vshader_private *shader_data = This->baseShader.backend_data;
4076     const struct wined3d_shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
4077     CONST DWORD *function = This->baseShader.function;
4078     const local_constant *lconst;
4079     GLuint ret;
4080     DWORD next_local, *lconst_map = local_const_mapping((IWineD3DBaseShaderImpl *) This);
4081     struct shader_arb_ctx_priv priv_ctx;
4082     unsigned int i;
4083     GLint errPos;
4084
4085     memset(&priv_ctx, 0, sizeof(priv_ctx));
4086     priv_ctx.cur_vs_args = args;
4087     list_init(&priv_ctx.control_frames);
4088     init_output_registers(This, args->ps_signature, &priv_ctx, compiled);
4089
4090     /*  Create the hw ARB shader */
4091     shader_addline(buffer, "!!ARBvp1.0\n");
4092
4093     /* Always enable the NV extension if available. Unlike fragment shaders, there is no
4094      * mesurable performance penalty, and we can always make use of it for clipplanes.
4095      */
4096     if (gl_info->supported[NV_VERTEX_PROGRAM3])
4097     {
4098         shader_addline(buffer, "OPTION NV_vertex_program3;\n");
4099         priv_ctx.target_version = NV3;
4100         shader_addline(buffer, "ADDRESS aL;\n");
4101     }
4102     else if (gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4103     {
4104         shader_addline(buffer, "OPTION NV_vertex_program2;\n");
4105         priv_ctx.target_version = NV2;
4106         shader_addline(buffer, "ADDRESS aL;\n");
4107     } else {
4108         priv_ctx.target_version = ARB;
4109     }
4110
4111     shader_addline(buffer, "TEMP TMP_OUT;\n");
4112     if (need_helper_const(shader_data, reg_maps, gl_info))
4113     {
4114         shader_addline(buffer, "PARAM helper_const = { 0.0, 1.0, 2.0, %1.10f};\n", eps);
4115     }
4116     if (need_rel_addr_const(shader_data, reg_maps, gl_info))
4117     {
4118         shader_addline(buffer, "PARAM rel_addr_const = { 0.5, %d.0, 0.0, 0.0 };\n", shader_data->rel_offset);
4119         shader_addline(buffer, "TEMP A0_SHADOW;\n");
4120     }
4121
4122     shader_addline(buffer, "TEMP TA;\n");
4123
4124     /* Base Declarations */
4125     next_local = shader_generate_arb_declarations((IWineD3DBaseShaderImpl *)This,
4126             reg_maps, buffer, gl_info, lconst_map, &priv_ctx.vs_clipplanes, &priv_ctx);
4127
4128     for(i = 0; i < MAX_CONST_I; i++)
4129     {
4130         compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
4131         if(reg_maps->integer_constants & (1 << i) && priv_ctx.target_version >= NV2)
4132         {
4133             const DWORD *control_values = find_loop_control_values((IWineD3DBaseShaderImpl *) This, i);
4134
4135             if(control_values)
4136             {
4137                 shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
4138                                 control_values[0], control_values[1], control_values[2]);
4139             }
4140             else
4141             {
4142                 compiled->int_consts[i] = next_local;
4143                 compiled->num_int_consts++;
4144                 shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
4145             }
4146         }
4147     }
4148
4149     /* We need a constant to fixup the final position */
4150     shader_addline(buffer, "PARAM posFixup = program.local[%u];\n", next_local);
4151     compiled->pos_fixup = next_local++;
4152
4153     /* Initialize output parameters. GL_ARB_vertex_program does not require special initialization values
4154      * for output parameters. D3D in theory does not do that either, but some applications depend on a
4155      * proper initialization of the secondary color, and programs using the fixed function pipeline without
4156      * a replacement shader depend on the texcoord.w being set properly.
4157      *
4158      * GL_NV_vertex_program defines that all output values are initialized to {0.0, 0.0, 0.0, 1.0}. This
4159      * assertion is in effect even when using GL_ARB_vertex_program without any NV specific additions. So
4160      * skip this if NV_vertex_program is supported. Otherwise, initialize the secondary color. For the tex-
4161      * coords, we have a flag in the opengl caps. Many cards do not require the texcoord being set, and
4162      * this can eat a number of instructions, so skip it unless this cap is set as well
4163      */
4164     if (!gl_info->supported[NV_VERTEX_PROGRAM])
4165     {
4166         IWineD3DDeviceImpl *device = This->baseShader.device;
4167         const char *color_init = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_0001);
4168         shader_addline(buffer, "MOV result.color.secondary, %s;\n", color_init);
4169
4170         if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W && !device->frag_pipe->ffp_proj_control)
4171         {
4172             int i;
4173             const char *one = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ONE);
4174             for(i = 0; i < min(8, MAX_REG_TEXCRD); i++)
4175             {
4176                 if (reg_maps->texcoord_mask[i] && reg_maps->texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL)
4177                     shader_addline(buffer, "MOV result.texcoord[%u].w, %s\n", i, one);
4178             }
4179         }
4180     }
4181
4182     /* The shader starts with the main function */
4183     priv_ctx.in_main_func = TRUE;
4184     /* Base Shader Body */
4185     shader_generate_main((IWineD3DBaseShaderImpl *)This, buffer, reg_maps, function, &priv_ctx);
4186
4187     if (!priv_ctx.footer_written) vshader_add_footer(&priv_ctx,
4188             shader_data, args, reg_maps, gl_info, buffer);
4189
4190     shader_addline(buffer, "END\n");
4191
4192     /* TODO: change to resource.glObjectHandle or something like that */
4193     GL_EXTCALL(glGenProgramsARB(1, &ret));
4194
4195     TRACE("Creating a hw vertex shader, prg=%d\n", ret);
4196     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, ret));
4197
4198     TRACE("Created hw vertex shader, prg=%d\n", ret);
4199     /* Create the program and check for errors */
4200     GL_EXTCALL(glProgramStringARB(GL_VERTEX_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
4201                buffer->bsize, buffer->buffer));
4202     checkGLcall("glProgramStringARB()");
4203
4204     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errPos);
4205     if (errPos != -1)
4206     {
4207         FIXME("HW VertexShader Error at position %d: %s\n\n",
4208               errPos, debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
4209         shader_arb_dump_program_source(buffer->buffer);
4210         ret = -1;
4211     }
4212     else
4213     {
4214         GLint native;
4215
4216         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
4217         checkGLcall("glGetProgramivARB()");
4218         if (!native) WARN("Program exceeds native resource limits.\n");
4219
4220         /* Load immediate constants */
4221         if(lconst_map) {
4222             LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
4223                 const float *value = (const float *)lconst->value;
4224                 GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, lconst_map[lconst->idx], value));
4225             }
4226         }
4227     }
4228     HeapFree(GetProcessHeap(), 0, lconst_map);
4229
4230     return ret;
4231 }
4232
4233 /* GL locking is done by the caller */
4234 static struct arb_ps_compiled_shader *find_arb_pshader(IWineD3DPixelShaderImpl *shader, const struct arb_ps_compile_args *args)
4235 {
4236     IWineD3DDeviceImpl *device = shader->baseShader.device;
4237     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4238     UINT i;
4239     DWORD new_size;
4240     struct arb_ps_compiled_shader *new_array;
4241     struct wined3d_shader_buffer buffer;
4242     struct arb_pshader_private *shader_data;
4243     GLuint ret;
4244
4245     if (!shader->baseShader.backend_data)
4246     {
4247         struct shader_arb_priv *priv = device->shader_priv;
4248
4249         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4250         shader_data = shader->baseShader.backend_data;
4251         shader_data->clamp_consts = shader->baseShader.reg_maps.shader_version.major == 1;
4252
4253         if(shader->baseShader.reg_maps.shader_version.major < 3) shader_data->input_signature_idx = ~0;
4254         else shader_data->input_signature_idx = find_input_signature(priv, shader->baseShader.input_signature);
4255
4256         shader_data->has_signature_idx = TRUE;
4257         TRACE("Shader got assigned input signature index %u\n", shader_data->input_signature_idx);
4258
4259         if (!device->vs_clipping)
4260             shader_data->clipplane_emulation = shader_find_free_input_register(&shader->baseShader.reg_maps,
4261                     gl_info->limits.texture_stages - 1);
4262         else
4263             shader_data->clipplane_emulation = ~0U;
4264     }
4265     shader_data = shader->baseShader.backend_data;
4266
4267     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4268      * so a linear search is more performant than a hashmap or a binary search
4269      * (cache coherency etc)
4270      */
4271     for (i = 0; i < shader_data->num_gl_shaders; ++i)
4272     {
4273         if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4274             return &shader_data->gl_shaders[i];
4275     }
4276
4277     TRACE("No matching GL shader found, compiling a new shader\n");
4278     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4279         if (shader_data->num_gl_shaders)
4280         {
4281             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4282             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4283                                     new_size * sizeof(*shader_data->gl_shaders));
4284         } else {
4285             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4286             new_size = 1;
4287         }
4288
4289         if(!new_array) {
4290             ERR("Out of memory\n");
4291             return 0;
4292         }
4293         shader_data->gl_shaders = new_array;
4294         shader_data->shader_array_size = new_size;
4295     }
4296
4297     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4298
4299     pixelshader_update_samplers(&shader->baseShader.reg_maps, device->stateBlock->state.textures);
4300
4301     if (!shader_buffer_init(&buffer))
4302     {
4303         ERR("Failed to initialize shader buffer.\n");
4304         return 0;
4305     }
4306
4307     ret = shader_arb_generate_pshader(shader, gl_info, &buffer, args,
4308             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4309     shader_buffer_free(&buffer);
4310     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4311
4312     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4313 }
4314
4315 static inline BOOL vs_args_equal(const struct arb_vs_compile_args *stored, const struct arb_vs_compile_args *new,
4316                                  const DWORD use_map, BOOL skip_int) {
4317     if((stored->super.swizzle_map & use_map) != new->super.swizzle_map) return FALSE;
4318     if(stored->super.clip_enabled != new->super.clip_enabled) return FALSE;
4319     if(stored->super.fog_src != new->super.fog_src) return FALSE;
4320     if(stored->clip.boolclip_compare != new->clip.boolclip_compare) return FALSE;
4321     if(stored->ps_signature != new->ps_signature) return FALSE;
4322     if(stored->vertex.samplers_compare != new->vertex.samplers_compare) return FALSE;
4323     if(skip_int) return TRUE;
4324
4325     return !memcmp(stored->loop_ctrl, new->loop_ctrl, sizeof(stored->loop_ctrl));
4326 }
4327
4328 static struct arb_vs_compiled_shader *find_arb_vshader(IWineD3DVertexShaderImpl *shader, const struct arb_vs_compile_args *args)
4329 {
4330     IWineD3DDeviceImpl *device = shader->baseShader.device;
4331     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4332     DWORD use_map = device->strided_streams.use_map;
4333     UINT i;
4334     DWORD new_size;
4335     struct arb_vs_compiled_shader *new_array;
4336     struct wined3d_shader_buffer buffer;
4337     struct arb_vshader_private *shader_data;
4338     GLuint ret;
4339
4340     if (!shader->baseShader.backend_data)
4341     {
4342         const struct wined3d_shader_reg_maps *reg_maps = &shader->baseShader.reg_maps;
4343
4344         shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4345         shader_data = shader->baseShader.backend_data;
4346
4347         if ((gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
4348                 && reg_maps->min_rel_offset <= reg_maps->max_rel_offset)
4349         {
4350             if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 127)
4351             {
4352                 FIXME("The difference between the minimum and maximum relative offset is > 127.\n");
4353                 FIXME("Which this OpenGL implementation does not support. Try using GLSL.\n");
4354                 FIXME("Min: %u, Max: %u.\n", reg_maps->min_rel_offset, reg_maps->max_rel_offset);
4355             }
4356             else if (reg_maps->max_rel_offset - reg_maps->min_rel_offset > 63)
4357                 shader_data->rel_offset = reg_maps->min_rel_offset + 63;
4358             else if (reg_maps->max_rel_offset > 63)
4359                 shader_data->rel_offset = reg_maps->min_rel_offset;
4360         }
4361     }
4362     shader_data = shader->baseShader.backend_data;
4363
4364     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4365      * so a linear search is more performant than a hashmap or a binary search
4366      * (cache coherency etc)
4367      */
4368     for(i = 0; i < shader_data->num_gl_shaders; i++) {
4369         if (vs_args_equal(&shader_data->gl_shaders[i].args, args,
4370                 use_map, gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]))
4371         {
4372             return &shader_data->gl_shaders[i];
4373         }
4374     }
4375
4376     TRACE("No matching GL shader found, compiling a new shader\n");
4377
4378     if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4379         if (shader_data->num_gl_shaders)
4380         {
4381             new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4382             new_array = HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, shader_data->gl_shaders,
4383                                     new_size * sizeof(*shader_data->gl_shaders));
4384         } else {
4385             new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4386             new_size = 1;
4387         }
4388
4389         if(!new_array) {
4390             ERR("Out of memory\n");
4391             return 0;
4392         }
4393         shader_data->gl_shaders = new_array;
4394         shader_data->shader_array_size = new_size;
4395     }
4396
4397     shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4398
4399     if (!shader_buffer_init(&buffer))
4400     {
4401         ERR("Failed to initialize shader buffer.\n");
4402         return 0;
4403     }
4404
4405     ret = shader_arb_generate_vshader(shader, gl_info, &buffer, args,
4406             &shader_data->gl_shaders[shader_data->num_gl_shaders]);
4407     shader_buffer_free(&buffer);
4408     shader_data->gl_shaders[shader_data->num_gl_shaders].prgId = ret;
4409
4410     return &shader_data->gl_shaders[shader_data->num_gl_shaders++];
4411 }
4412
4413 static void find_arb_ps_compile_args(const struct wined3d_state *state,
4414         IWineD3DPixelShaderImpl *shader, struct arb_ps_compile_args *args)
4415 {
4416     IWineD3DDeviceImpl *device = shader->baseShader.device;
4417     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4418     int i;
4419     WORD int_skip;
4420
4421     find_ps_compile_args(state, shader, &args->super);
4422
4423     /* This forces all local boolean constants to 1 to make them stateblock independent */
4424     args->bools = shader->baseShader.reg_maps.local_bool_consts;
4425
4426     for(i = 0; i < MAX_CONST_B; i++)
4427     {
4428         if (state->ps_consts_b[i])
4429             args->bools |= ( 1 << i);
4430     }
4431
4432     /* Only enable the clip plane emulation KIL if at least one clipplane is enabled. The KIL instruction
4433      * is quite expensive because it forces the driver to disable early Z discards. It is cheaper to
4434      * duplicate the shader than have a no-op KIL instruction in every shader
4435      */
4436     if (!device->vs_clipping && use_vs(state)
4437             && state->render_states[WINED3DRS_CLIPPING]
4438             && state->render_states[WINED3DRS_CLIPPLANEENABLE])
4439         args->clip = 1;
4440     else
4441         args->clip = 0;
4442
4443     /* Skip if unused or local, or supported natively */
4444     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4445     if (int_skip == 0xffff || gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
4446     {
4447         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4448         return;
4449     }
4450
4451     for(i = 0; i < MAX_CONST_I; i++)
4452     {
4453         if(int_skip & (1 << i))
4454         {
4455             args->loop_ctrl[i][0] = 0;
4456             args->loop_ctrl[i][1] = 0;
4457             args->loop_ctrl[i][2] = 0;
4458         }
4459         else
4460         {
4461             args->loop_ctrl[i][0] = state->ps_consts_i[i * 4];
4462             args->loop_ctrl[i][1] = state->ps_consts_i[i * 4 + 1];
4463             args->loop_ctrl[i][2] = state->ps_consts_i[i * 4 + 2];
4464         }
4465     }
4466 }
4467
4468 static void find_arb_vs_compile_args(const struct wined3d_state *state,
4469         IWineD3DVertexShaderImpl *shader, struct arb_vs_compile_args *args)
4470 {
4471     IWineD3DDeviceImpl *device = shader->baseShader.device;
4472     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4473     int i;
4474     WORD int_skip;
4475
4476     find_vs_compile_args(state, shader, &args->super);
4477
4478     args->clip.boolclip_compare = 0;
4479     if (use_ps(state))
4480     {
4481         IWineD3DPixelShaderImpl *ps = state->pixel_shader;
4482         struct arb_pshader_private *shader_priv = ps->baseShader.backend_data;
4483         args->ps_signature = shader_priv->input_signature_idx;
4484
4485         args->clip.boolclip.clip_texcoord = shader_priv->clipplane_emulation + 1;
4486     }
4487     else
4488     {
4489         args->ps_signature = ~0;
4490         if (!device->vs_clipping)
4491         {
4492             args->clip.boolclip.clip_texcoord = ffp_clip_emul(state) ? gl_info->limits.texture_stages : 0;
4493         }
4494         /* Otherwise: Setting boolclip_compare set clip_texcoord to 0 */
4495     }
4496
4497     if (args->clip.boolclip.clip_texcoord)
4498     {
4499         if (state->render_states[WINED3DRS_CLIPPING])
4500             args->clip.boolclip.clipplane_mask = (unsigned char)state->render_states[WINED3DRS_CLIPPLANEENABLE];
4501         /* clipplane_mask was set to 0 by setting boolclip_compare to 0 */
4502     }
4503
4504     /* This forces all local boolean constants to 1 to make them stateblock independent */
4505     args->clip.boolclip.bools = shader->baseShader.reg_maps.local_bool_consts;
4506     /* TODO: Figure out if it would be better to store bool constants as bitmasks in the stateblock */
4507     for(i = 0; i < MAX_CONST_B; i++)
4508     {
4509         if (state->vs_consts_b[i])
4510             args->clip.boolclip.bools |= ( 1 << i);
4511     }
4512
4513     args->vertex.samplers[0] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 0];
4514     args->vertex.samplers[1] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 1];
4515     args->vertex.samplers[2] = device->texUnitMap[MAX_FRAGMENT_SAMPLERS + 2];
4516     args->vertex.samplers[3] = 0;
4517
4518     /* Skip if unused or local */
4519     int_skip = ~shader->baseShader.reg_maps.integer_constants | shader->baseShader.reg_maps.local_int_consts;
4520     /* This is about flow control, not clipping. */
4521     if (int_skip == 0xffff || gl_info->supported[NV_VERTEX_PROGRAM2_OPTION])
4522     {
4523         memset(&args->loop_ctrl, 0, sizeof(args->loop_ctrl));
4524         return;
4525     }
4526
4527     for(i = 0; i < MAX_CONST_I; i++)
4528     {
4529         if(int_skip & (1 << i))
4530         {
4531             args->loop_ctrl[i][0] = 0;
4532             args->loop_ctrl[i][1] = 0;
4533             args->loop_ctrl[i][2] = 0;
4534         }
4535         else
4536         {
4537             args->loop_ctrl[i][0] = state->vs_consts_i[i * 4];
4538             args->loop_ctrl[i][1] = state->vs_consts_i[i * 4 + 1];
4539             args->loop_ctrl[i][2] = state->vs_consts_i[i * 4 + 2];
4540         }
4541     }
4542 }
4543
4544 /* GL locking is done by the caller */
4545 static void shader_arb_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4546 {
4547     IWineD3DDeviceImpl *This = context->swapchain->device;
4548     struct shader_arb_priv *priv = This->shader_priv;
4549     const struct wined3d_gl_info *gl_info = context->gl_info;
4550     const struct wined3d_state *state = &This->stateBlock->state;
4551     int i;
4552
4553     /* Deal with pixel shaders first so the vertex shader arg function has the input signature ready */
4554     if (usePS)
4555     {
4556         IWineD3DPixelShaderImpl *ps = state->pixel_shader;
4557         struct arb_ps_compile_args compile_args;
4558         struct arb_ps_compiled_shader *compiled;
4559
4560         TRACE("Using pixel shader %p.\n", ps);
4561         find_arb_ps_compile_args(state, ps, &compile_args);
4562         compiled = find_arb_pshader(ps, &compile_args);
4563         priv->current_fprogram_id = compiled->prgId;
4564         priv->compiled_fprog = compiled;
4565
4566         /* Bind the fragment program */
4567         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4568         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id);");
4569
4570         if(!priv->use_arbfp_fixed_func) {
4571             /* Enable OpenGL fragment programs */
4572             glEnable(GL_FRAGMENT_PROGRAM_ARB);
4573             checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB);");
4574         }
4575         TRACE("(%p) : Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB\n", This, priv->current_fprogram_id);
4576
4577         /* Pixel Shader 1.x constants are clamped to [-1;1], Pixel Shader 2.0 constants are not. If switching between
4578          * a 1.x and newer shader, reload the first 8 constants
4579          */
4580         if(priv->last_ps_const_clamped != ((struct arb_pshader_private *)ps->baseShader.backend_data)->clamp_consts)
4581         {
4582             priv->last_ps_const_clamped = ((struct arb_pshader_private *)ps->baseShader.backend_data)->clamp_consts;
4583             This->highest_dirty_ps_const = max(This->highest_dirty_ps_const, 8);
4584             for(i = 0; i < 8; i++)
4585             {
4586                 context->pshader_const_dirty[i] = 1;
4587             }
4588             /* Also takes care of loading local constants */
4589             shader_arb_load_constants(context, TRUE, FALSE);
4590         }
4591         else
4592         {
4593             float rt_height = This->render_targets[0]->currentDesc.Height;
4594             shader_arb_ps_local_constants(compiled, context, state, rt_height);
4595         }
4596
4597         /* Force constant reloading for the NP2 fixup (see comment in shader_glsl_select for more info) */
4598         if (compiled->np2fixup_info.super.active)
4599             shader_arb_load_np2fixup_constants(priv, gl_info, state);
4600     }
4601     else if (gl_info->supported[ARB_FRAGMENT_PROGRAM] && !priv->use_arbfp_fixed_func)
4602     {
4603         /* Disable only if we're not using arbfp fixed function fragment processing. If this is used,
4604         * keep GL_FRAGMENT_PROGRAM_ARB enabled, and the fixed function pipeline will bind the fixed function
4605         * replacement shader
4606         */
4607         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4608         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4609         priv->current_fprogram_id = 0;
4610     }
4611
4612     if (useVS)
4613     {
4614         IWineD3DVertexShaderImpl *vs = state->vertex_shader;
4615         struct arb_vs_compile_args compile_args;
4616         struct arb_vs_compiled_shader *compiled;
4617
4618         TRACE("Using vertex shader %p\n", vs);
4619         find_arb_vs_compile_args(state, vs, &compile_args);
4620         compiled = find_arb_vshader(vs, &compile_args);
4621         priv->current_vprogram_id = compiled->prgId;
4622         priv->compiled_vprog = compiled;
4623
4624         /* Bind the vertex program */
4625         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4626         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id);");
4627
4628         /* Enable OpenGL vertex programs */
4629         glEnable(GL_VERTEX_PROGRAM_ARB);
4630         checkGLcall("glEnable(GL_VERTEX_PROGRAM_ARB);");
4631         TRACE("(%p) : Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB\n", This, priv->current_vprogram_id);
4632         shader_arb_vs_local_constants(compiled, context, state);
4633
4634         if(priv->last_vs_color_unclamp != compiled->need_color_unclamp) {
4635             priv->last_vs_color_unclamp = compiled->need_color_unclamp;
4636
4637             if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4638             {
4639                 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, !compiled->need_color_unclamp));
4640                 checkGLcall("glClampColorARB");
4641             } else {
4642                 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4643             }
4644         }
4645     }
4646     else if (gl_info->supported[ARB_VERTEX_PROGRAM])
4647     {
4648         priv->current_vprogram_id = 0;
4649         glDisable(GL_VERTEX_PROGRAM_ARB);
4650         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4651     }
4652 }
4653
4654 /* GL locking is done by the caller */
4655 static void shader_arb_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4656         enum tex_types tex_type, const SIZE *ds_mask_size)
4657 {
4658     const float mask[] = {0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy};
4659     BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4660     struct shader_arb_priv *priv = shader_priv;
4661     GLuint *blt_fprogram;
4662
4663     if (!priv->depth_blt_vprogram_id) priv->depth_blt_vprogram_id = create_arb_blt_vertex_program(gl_info);
4664     GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->depth_blt_vprogram_id));
4665     glEnable(GL_VERTEX_PROGRAM_ARB);
4666
4667     blt_fprogram = masked ? &priv->depth_blt_fprogram_id_masked[tex_type] : &priv->depth_blt_fprogram_id_full[tex_type];
4668     if (!*blt_fprogram) *blt_fprogram = create_arb_blt_fragment_program(gl_info, tex_type, masked);
4669     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, *blt_fprogram));
4670     if (masked) GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, mask));
4671     glEnable(GL_FRAGMENT_PROGRAM_ARB);
4672 }
4673
4674 /* GL locking is done by the caller */
4675 static void shader_arb_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4676 {
4677     struct shader_arb_priv *priv = shader_priv;
4678
4679     if (priv->current_vprogram_id) {
4680         GL_EXTCALL(glBindProgramARB(GL_VERTEX_PROGRAM_ARB, priv->current_vprogram_id));
4681         checkGLcall("glBindProgramARB(GL_VERTEX_PROGRAM_ARB, vertexShader->prgId);");
4682
4683         TRACE("Bound vertex program %u and enabled GL_VERTEX_PROGRAM_ARB.\n", priv->current_vprogram_id);
4684     }
4685     else
4686     {
4687         glDisable(GL_VERTEX_PROGRAM_ARB);
4688         checkGLcall("glDisable(GL_VERTEX_PROGRAM_ARB)");
4689     }
4690
4691     if (priv->current_fprogram_id) {
4692         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, priv->current_fprogram_id));
4693         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, pixelShader->prgId);");
4694
4695         TRACE("Bound fragment program %u and enabled GL_FRAGMENT_PROGRAM_ARB.\n", priv->current_fprogram_id);
4696     }
4697     else if(!priv->use_arbfp_fixed_func)
4698     {
4699         glDisable(GL_FRAGMENT_PROGRAM_ARB);
4700         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
4701     }
4702 }
4703
4704 static void shader_arb_destroy(IWineD3DBaseShaderImpl *shader)
4705 {
4706     IWineD3DDeviceImpl *device = shader->baseShader.device;
4707     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4708
4709     if (shader_is_pshader_version(shader->baseShader.reg_maps.shader_version.type))
4710     {
4711         struct arb_pshader_private *shader_data = shader->baseShader.backend_data;
4712         UINT i;
4713
4714         if(!shader_data) return; /* This can happen if a shader was never compiled */
4715
4716         if (shader_data->num_gl_shaders)
4717         {
4718             struct wined3d_context *context = context_acquire(device, NULL);
4719
4720             ENTER_GL();
4721             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4722             {
4723                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4724                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4725             }
4726             LEAVE_GL();
4727
4728             context_release(context);
4729         }
4730
4731         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4732         HeapFree(GetProcessHeap(), 0, shader_data);
4733         shader->baseShader.backend_data = NULL;
4734     }
4735     else
4736     {
4737         struct arb_vshader_private *shader_data = shader->baseShader.backend_data;
4738         UINT i;
4739
4740         if(!shader_data) return; /* This can happen if a shader was never compiled */
4741
4742         if (shader_data->num_gl_shaders)
4743         {
4744             struct wined3d_context *context = context_acquire(device, NULL);
4745
4746             ENTER_GL();
4747             for (i = 0; i < shader_data->num_gl_shaders; ++i)
4748             {
4749                 GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId));
4750                 checkGLcall("GL_EXTCALL(glDeleteProgramsARB(1, &shader_data->gl_shaders[i].prgId))");
4751             }
4752             LEAVE_GL();
4753
4754             context_release(context);
4755         }
4756
4757         HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4758         HeapFree(GetProcessHeap(), 0, shader_data);
4759         shader->baseShader.backend_data = NULL;
4760     }
4761 }
4762
4763 static int sig_tree_compare(const void *key, const struct wine_rb_entry *entry)
4764 {
4765     struct ps_signature *e = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4766     return compare_sig(key, e->sig);
4767 }
4768
4769 static const struct wine_rb_functions sig_tree_functions =
4770 {
4771     wined3d_rb_alloc,
4772     wined3d_rb_realloc,
4773     wined3d_rb_free,
4774     sig_tree_compare
4775 };
4776
4777 static HRESULT shader_arb_alloc(IWineD3DDeviceImpl *device)
4778 {
4779     struct shader_arb_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*priv));
4780     if(wine_rb_init(&priv->signature_tree, &sig_tree_functions) == -1)
4781     {
4782         ERR("RB tree init failed\n");
4783         HeapFree(GetProcessHeap(), 0, priv);
4784         return E_OUTOFMEMORY;
4785     }
4786     device->shader_priv = priv;
4787     return WINED3D_OK;
4788 }
4789
4790 static void release_signature(struct wine_rb_entry *entry, void *context)
4791 {
4792     struct ps_signature *sig = WINE_RB_ENTRY_VALUE(entry, struct ps_signature, entry);
4793     int i;
4794     for(i = 0; i < MAX_REG_INPUT; i++)
4795     {
4796         HeapFree(GetProcessHeap(), 0, (char *) sig->sig[i].semantic_name);
4797     }
4798     HeapFree(GetProcessHeap(), 0, sig->sig);
4799     HeapFree(GetProcessHeap(), 0, sig);
4800 }
4801
4802 /* Context activation is done by the caller. */
4803 static void shader_arb_free(IWineD3DDeviceImpl *device)
4804 {
4805     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4806     struct shader_arb_priv *priv = device->shader_priv;
4807     int i;
4808
4809     ENTER_GL();
4810     if(priv->depth_blt_vprogram_id) {
4811         GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_vprogram_id));
4812     }
4813     for (i = 0; i < tex_type_count; ++i)
4814     {
4815         if (priv->depth_blt_fprogram_id_full[i])
4816         {
4817             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_full[i]));
4818         }
4819         if (priv->depth_blt_fprogram_id_masked[i])
4820         {
4821             GL_EXTCALL(glDeleteProgramsARB(1, &priv->depth_blt_fprogram_id_masked[i]));
4822         }
4823     }
4824     LEAVE_GL();
4825
4826     wine_rb_destroy(&priv->signature_tree, release_signature, NULL);
4827     HeapFree(GetProcessHeap(), 0, device->shader_priv);
4828 }
4829
4830 static BOOL shader_arb_dirty_const(void)
4831 {
4832     return TRUE;
4833 }
4834
4835 static void shader_arb_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *pCaps)
4836 {
4837     if (gl_info->supported[ARB_VERTEX_PROGRAM])
4838     {
4839         DWORD vs_consts;
4840
4841         /* 96 is the minimum allowed value of MAX_PROGRAM_ENV_PARAMETERS_ARB
4842          * for vertex programs. If the native limit is less than that it's
4843          * not very useful, and e.g. Mesa swrast returns 0, probably to
4844          * indicate it's a software implementation. */
4845         if (gl_info->limits.arb_vs_native_constants < 96)
4846             vs_consts = gl_info->limits.arb_vs_float_constants;
4847         else
4848             vs_consts = min(gl_info->limits.arb_vs_float_constants, gl_info->limits.arb_vs_native_constants);
4849
4850         if (gl_info->supported[NV_VERTEX_PROGRAM3])
4851         {
4852             pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4853             TRACE_(d3d_caps)("Hardware vertex shader version 3.0 enabled (NV_VERTEX_PROGRAM3)\n");
4854         }
4855         else if (vs_consts >= 256)
4856         {
4857             /* Shader Model 2.0 requires at least 256 vertex shader constants */
4858             pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4859             TRACE_(d3d_caps)("Hardware vertex shader version 2.0 enabled (ARB_PROGRAM)\n");
4860         }
4861         else
4862         {
4863             pCaps->VertexShaderVersion = WINED3DVS_VERSION(1,1);
4864             TRACE_(d3d_caps)("Hardware vertex shader version 1.1 enabled (ARB_PROGRAM)\n");
4865         }
4866         pCaps->MaxVertexShaderConst = vs_consts;
4867     }
4868     else
4869     {
4870         pCaps->VertexShaderVersion = 0;
4871         pCaps->MaxVertexShaderConst = 0;
4872     }
4873
4874     if (gl_info->supported[ARB_FRAGMENT_PROGRAM])
4875     {
4876         DWORD ps_consts;
4877
4878         /* Similar as above for vertex programs, but the minimum for fragment
4879          * programs is 24. */
4880         if (gl_info->limits.arb_ps_native_constants < 24)
4881             ps_consts = gl_info->limits.arb_ps_float_constants;
4882         else
4883             ps_consts = min(gl_info->limits.arb_ps_float_constants, gl_info->limits.arb_ps_native_constants);
4884
4885         if (gl_info->supported[NV_FRAGMENT_PROGRAM2])
4886         {
4887             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(3,0);
4888             TRACE_(d3d_caps)("Hardware pixel shader version 3.0 enabled (NV_FRAGMENT_PROGRAM2)\n");
4889         }
4890         else if (ps_consts >= 32)
4891         {
4892             /* Shader Model 2.0 requires at least 32 pixel shader constants */
4893             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(2,0);
4894             TRACE_(d3d_caps)("Hardware pixel shader version 2.0 enabled (ARB_PROGRAM)\n");
4895         }
4896         else
4897         {
4898             pCaps->PixelShaderVersion    = WINED3DPS_VERSION(1,4);
4899             TRACE_(d3d_caps)("Hardware pixel shader version 1.4 enabled (ARB_PROGRAM)\n");
4900         }
4901         pCaps->PixelShader1xMaxValue = 8.0f;
4902         pCaps->MaxPixelShaderConst = ps_consts;
4903     }
4904     else
4905     {
4906         pCaps->PixelShaderVersion = 0;
4907         pCaps->PixelShader1xMaxValue = 0.0f;
4908         pCaps->MaxPixelShaderConst = 0;
4909     }
4910
4911     pCaps->VSClipping = use_nv_clip(gl_info);
4912 }
4913
4914 static BOOL shader_arb_color_fixup_supported(struct color_fixup_desc fixup)
4915 {
4916     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4917     {
4918         TRACE("Checking support for color_fixup:\n");
4919         dump_color_fixup_desc(fixup);
4920     }
4921
4922     /* We support everything except complex conversions. */
4923     if (!is_complex_fixup(fixup))
4924     {
4925         TRACE("[OK]\n");
4926         return TRUE;
4927     }
4928
4929     TRACE("[FAILED]\n");
4930     return FALSE;
4931 }
4932
4933 static void shader_arb_add_instruction_modifiers(const struct wined3d_shader_instruction *ins) {
4934     DWORD shift;
4935     char write_mask[20], regstr[50];
4936     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
4937     BOOL is_color = FALSE;
4938     const struct wined3d_shader_dst_param *dst;
4939
4940     if (!ins->dst_count) return;
4941
4942     dst = &ins->dst[0];
4943     shift = dst->shift;
4944     if (!shift) return; /* Saturate alone is handled by the instructions */
4945
4946     shader_arb_get_write_mask(ins, dst, write_mask);
4947     shader_arb_get_register_name(ins, &dst->reg, regstr, &is_color);
4948
4949     /* Generate a line that does the output modifier computation
4950      * FIXME: _SAT vs shift? _SAT alone is already handled in the instructions, if this
4951      * maps problems in e.g. _d4_sat modify shader_arb_get_modifier
4952      */
4953     shader_addline(buffer, "MUL%s %s%s, %s, %s;\n", shader_arb_get_modifier(ins),
4954                    regstr, write_mask, regstr, shift_tab[shift]);
4955 }
4956
4957 static const SHADER_HANDLER shader_arb_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4958 {
4959     /* WINED3DSIH_ABS           */ shader_hw_map2gl,
4960     /* WINED3DSIH_ADD           */ shader_hw_map2gl,
4961     /* WINED3DSIH_AND           */ NULL,
4962     /* WINED3DSIH_BEM           */ pshader_hw_bem,
4963     /* WINED3DSIH_BREAK         */ shader_hw_break,
4964     /* WINED3DSIH_BREAKC        */ shader_hw_breakc,
4965     /* WINED3DSIH_BREAKP        */ NULL,
4966     /* WINED3DSIH_CALL          */ shader_hw_call,
4967     /* WINED3DSIH_CALLNZ        */ NULL,
4968     /* WINED3DSIH_CMP           */ pshader_hw_cmp,
4969     /* WINED3DSIH_CND           */ pshader_hw_cnd,
4970     /* WINED3DSIH_CRS           */ shader_hw_map2gl,
4971     /* WINED3DSIH_CUT           */ NULL,
4972     /* WINED3DSIH_DCL           */ NULL,
4973     /* WINED3DSIH_DEF           */ NULL,
4974     /* WINED3DSIH_DEFB          */ NULL,
4975     /* WINED3DSIH_DEFI          */ NULL,
4976     /* WINED3DSIH_DIV           */ NULL,
4977     /* WINED3DSIH_DP2ADD        */ pshader_hw_dp2add,
4978     /* WINED3DSIH_DP3           */ shader_hw_map2gl,
4979     /* WINED3DSIH_DP4           */ shader_hw_map2gl,
4980     /* WINED3DSIH_DST           */ shader_hw_map2gl,
4981     /* WINED3DSIH_DSX           */ shader_hw_map2gl,
4982     /* WINED3DSIH_DSY           */ shader_hw_dsy,
4983     /* WINED3DSIH_ELSE          */ shader_hw_else,
4984     /* WINED3DSIH_EMIT          */ NULL,
4985     /* WINED3DSIH_ENDIF         */ shader_hw_endif,
4986     /* WINED3DSIH_ENDLOOP       */ shader_hw_endloop,
4987     /* WINED3DSIH_ENDREP        */ shader_hw_endrep,
4988     /* WINED3DSIH_EXP           */ shader_hw_scalar_op,
4989     /* WINED3DSIH_EXPP          */ shader_hw_scalar_op,
4990     /* WINED3DSIH_FRC           */ shader_hw_map2gl,
4991     /* WINED3DSIH_FTOI          */ NULL,
4992     /* WINED3DSIH_IADD          */ NULL,
4993     /* WINED3DSIH_IEQ           */ NULL,
4994     /* WINED3DSIH_IF            */ NULL /* Hardcoded into the shader */,
4995     /* WINED3DSIH_IFC           */ shader_hw_ifc,
4996     /* WINED3DSIH_IGE           */ NULL,
4997     /* WINED3DSIH_IMUL          */ NULL,
4998     /* WINED3DSIH_ITOF          */ NULL,
4999     /* WINED3DSIH_LABEL         */ shader_hw_label,
5000     /* WINED3DSIH_LD            */ NULL,
5001     /* WINED3DSIH_LIT           */ shader_hw_map2gl,
5002     /* WINED3DSIH_LOG           */ shader_hw_log_pow,
5003     /* WINED3DSIH_LOGP          */ shader_hw_log_pow,
5004     /* WINED3DSIH_LOOP          */ shader_hw_loop,
5005     /* WINED3DSIH_LRP           */ shader_hw_lrp,
5006     /* WINED3DSIH_LT            */ NULL,
5007     /* WINED3DSIH_M3x2          */ shader_hw_mnxn,
5008     /* WINED3DSIH_M3x3          */ shader_hw_mnxn,
5009     /* WINED3DSIH_M3x4          */ shader_hw_mnxn,
5010     /* WINED3DSIH_M4x3          */ shader_hw_mnxn,
5011     /* WINED3DSIH_M4x4          */ shader_hw_mnxn,
5012     /* WINED3DSIH_MAD           */ shader_hw_map2gl,
5013     /* WINED3DSIH_MAX           */ shader_hw_map2gl,
5014     /* WINED3DSIH_MIN           */ shader_hw_map2gl,
5015     /* WINED3DSIH_MOV           */ shader_hw_mov,
5016     /* WINED3DSIH_MOVA          */ shader_hw_mov,
5017     /* WINED3DSIH_MOVC          */ NULL,
5018     /* WINED3DSIH_MUL           */ shader_hw_map2gl,
5019     /* WINED3DSIH_NOP           */ shader_hw_nop,
5020     /* WINED3DSIH_NRM           */ shader_hw_nrm,
5021     /* WINED3DSIH_PHASE         */ NULL,
5022     /* WINED3DSIH_POW           */ shader_hw_log_pow,
5023     /* WINED3DSIH_RCP           */ shader_hw_rcp,
5024     /* WINED3DSIH_REP           */ shader_hw_rep,
5025     /* WINED3DSIH_RET           */ shader_hw_ret,
5026     /* WINED3DSIH_RSQ           */ shader_hw_scalar_op,
5027     /* WINED3DSIH_SAMPLE        */ NULL,
5028     /* WINED3DSIH_SAMPLE_GRAD   */ NULL,
5029     /* WINED3DSIH_SAMPLE_LOD    */ NULL,
5030     /* WINED3DSIH_SETP          */ NULL,
5031     /* WINED3DSIH_SGE           */ shader_hw_map2gl,
5032     /* WINED3DSIH_SGN           */ shader_hw_sgn,
5033     /* WINED3DSIH_SINCOS        */ shader_hw_sincos,
5034     /* WINED3DSIH_SLT           */ shader_hw_map2gl,
5035     /* WINED3DSIH_SQRT          */ NULL,
5036     /* WINED3DSIH_SUB           */ shader_hw_map2gl,
5037     /* WINED3DSIH_TEX           */ pshader_hw_tex,
5038     /* WINED3DSIH_TEXBEM        */ pshader_hw_texbem,
5039     /* WINED3DSIH_TEXBEML       */ pshader_hw_texbem,
5040     /* WINED3DSIH_TEXCOORD      */ pshader_hw_texcoord,
5041     /* WINED3DSIH_TEXDEPTH      */ pshader_hw_texdepth,
5042     /* WINED3DSIH_TEXDP3        */ pshader_hw_texdp3,
5043     /* WINED3DSIH_TEXDP3TEX     */ pshader_hw_texdp3tex,
5044     /* WINED3DSIH_TEXKILL       */ pshader_hw_texkill,
5045     /* WINED3DSIH_TEXLDD        */ shader_hw_texldd,
5046     /* WINED3DSIH_TEXLDL        */ shader_hw_texldl,
5047     /* WINED3DSIH_TEXM3x2DEPTH  */ pshader_hw_texm3x2depth,
5048     /* WINED3DSIH_TEXM3x2PAD    */ pshader_hw_texm3x2pad,
5049     /* WINED3DSIH_TEXM3x2TEX    */ pshader_hw_texm3x2tex,
5050     /* WINED3DSIH_TEXM3x3       */ pshader_hw_texm3x3,
5051     /* WINED3DSIH_TEXM3x3DIFF   */ NULL,
5052     /* WINED3DSIH_TEXM3x3PAD    */ pshader_hw_texm3x3pad,
5053     /* WINED3DSIH_TEXM3x3SPEC   */ pshader_hw_texm3x3spec,
5054     /* WINED3DSIH_TEXM3x3TEX    */ pshader_hw_texm3x3tex,
5055     /* WINED3DSIH_TEXM3x3VSPEC  */ pshader_hw_texm3x3vspec,
5056     /* WINED3DSIH_TEXREG2AR     */ pshader_hw_texreg2ar,
5057     /* WINED3DSIH_TEXREG2GB     */ pshader_hw_texreg2gb,
5058     /* WINED3DSIH_TEXREG2RGB    */ pshader_hw_texreg2rgb,
5059     /* WINED3DSIH_UTOF          */ NULL,
5060 };
5061
5062 static inline BOOL get_bool_const(const struct wined3d_shader_instruction *ins, IWineD3DBaseShaderImpl *This, DWORD idx)
5063 {
5064     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5065     BOOL vshader = shader_is_vshader_version(reg_maps->shader_version.type);
5066     WORD bools = 0;
5067     WORD flag = (1 << idx);
5068     const local_constant *constant;
5069     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5070
5071     if (reg_maps->local_bool_consts & flag)
5072     {
5073         /* What good is a if(bool) with a hardcoded local constant? I don't know, but handle it */
5074         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsB, local_constant, entry)
5075         {
5076             if (constant->idx == idx)
5077             {
5078                 return constant->value[0];
5079             }
5080         }
5081         ERR("Local constant not found\n");
5082         return FALSE;
5083     }
5084     else
5085     {
5086         if(vshader) bools = priv->cur_vs_args->clip.boolclip.bools;
5087         else bools = priv->cur_ps_args->bools;
5088         return bools & flag;
5089     }
5090 }
5091
5092 static void get_loop_control_const(const struct wined3d_shader_instruction *ins,
5093         IWineD3DBaseShaderImpl *This, UINT idx, struct wined3d_shader_loop_control *loop_control)
5094 {
5095     const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
5096     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5097
5098     /* Integer constants can either be a local constant, or they can be stored in the shader
5099      * type specific compile args. */
5100     if (reg_maps->local_int_consts & (1 << idx))
5101     {
5102         const local_constant *constant;
5103
5104         LIST_FOR_EACH_ENTRY(constant, &This->baseShader.constantsI, local_constant, entry)
5105         {
5106             if (constant->idx == idx)
5107             {
5108                 loop_control->count = constant->value[0];
5109                 loop_control->start = constant->value[1];
5110                 /* Step is signed. */
5111                 loop_control->step = (int)constant->value[2];
5112                 return;
5113             }
5114         }
5115         /* If this happens the flag was set incorrectly */
5116         ERR("Local constant not found\n");
5117         loop_control->count = 0;
5118         loop_control->start = 0;
5119         loop_control->step = 0;
5120         return;
5121     }
5122
5123     switch (reg_maps->shader_version.type)
5124     {
5125         case WINED3D_SHADER_TYPE_VERTEX:
5126             /* Count and aL start value are unsigned */
5127             loop_control->count = priv->cur_vs_args->loop_ctrl[idx][0];
5128             loop_control->start = priv->cur_vs_args->loop_ctrl[idx][1];
5129             /* Step is signed. */
5130             loop_control->step = ((char)priv->cur_vs_args->loop_ctrl[idx][2]);
5131             break;
5132
5133         case WINED3D_SHADER_TYPE_PIXEL:
5134             loop_control->count = priv->cur_ps_args->loop_ctrl[idx][0];
5135             loop_control->start = priv->cur_ps_args->loop_ctrl[idx][1];
5136             loop_control->step = ((char)priv->cur_ps_args->loop_ctrl[idx][2]);
5137             break;
5138
5139         default:
5140             FIXME("Unhandled shader type %#x.\n", reg_maps->shader_version.type);
5141             break;
5142     }
5143 }
5144
5145 static void record_instruction(struct list *list, const struct wined3d_shader_instruction *ins)
5146 {
5147     unsigned int i;
5148     struct wined3d_shader_dst_param *dst_param = NULL;
5149     struct wined3d_shader_src_param *src_param = NULL, *rel_addr = NULL;
5150     struct recorded_instruction *rec = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*rec));
5151     if(!rec)
5152     {
5153         ERR("Out of memory\n");
5154         return;
5155     }
5156
5157     rec->ins = *ins;
5158     dst_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param));
5159     if(!dst_param) goto free;
5160     *dst_param = *ins->dst;
5161     if(ins->dst->reg.rel_addr)
5162     {
5163         rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*dst_param->reg.rel_addr));
5164         if(!rel_addr) goto free;
5165         *rel_addr = *ins->dst->reg.rel_addr;
5166         dst_param->reg.rel_addr = rel_addr;
5167     }
5168     rec->ins.dst = dst_param;
5169
5170     src_param = HeapAlloc(GetProcessHeap(), 0, sizeof(*src_param) * ins->src_count);
5171     if(!src_param) goto free;
5172     for(i = 0; i < ins->src_count; i++)
5173     {
5174         src_param[i] = ins->src[i];
5175         if(ins->src[i].reg.rel_addr)
5176         {
5177             rel_addr = HeapAlloc(GetProcessHeap(), 0, sizeof(*rel_addr));
5178             if(!rel_addr) goto free;
5179             *rel_addr = *ins->src[i].reg.rel_addr;
5180             src_param[i].reg.rel_addr = rel_addr;
5181         }
5182     }
5183     rec->ins.src = src_param;
5184     list_add_tail(list, &rec->entry);
5185     return;
5186
5187 free:
5188     ERR("Out of memory\n");
5189     if(dst_param)
5190     {
5191         HeapFree(GetProcessHeap(), 0, (void *) dst_param->reg.rel_addr);
5192         HeapFree(GetProcessHeap(), 0, dst_param);
5193     }
5194     if(src_param)
5195     {
5196         for(i = 0; i < ins->src_count; i++)
5197         {
5198             HeapFree(GetProcessHeap(), 0, (void *) src_param[i].reg.rel_addr);
5199         }
5200         HeapFree(GetProcessHeap(), 0, src_param);
5201     }
5202     HeapFree(GetProcessHeap(), 0, rec);
5203 }
5204
5205 static void free_recorded_instruction(struct list *list)
5206 {
5207     struct recorded_instruction *rec_ins, *entry2;
5208     unsigned int i;
5209
5210     LIST_FOR_EACH_ENTRY_SAFE(rec_ins, entry2, list, struct recorded_instruction, entry)
5211     {
5212         list_remove(&rec_ins->entry);
5213         if(rec_ins->ins.dst)
5214         {
5215             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst->reg.rel_addr);
5216             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.dst);
5217         }
5218         if(rec_ins->ins.src)
5219         {
5220             for(i = 0; i < rec_ins->ins.src_count; i++)
5221             {
5222                 HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src[i].reg.rel_addr);
5223             }
5224             HeapFree(GetProcessHeap(), 0, (void *) rec_ins->ins.src);
5225         }
5226         HeapFree(GetProcessHeap(), 0, rec_ins);
5227     }
5228 }
5229
5230 static void shader_arb_handle_instruction(const struct wined3d_shader_instruction *ins) {
5231     SHADER_HANDLER hw_fct;
5232     struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
5233     struct IWineD3DBaseShaderImpl *shader = ins->ctx->shader;
5234     struct control_frame *control_frame;
5235     struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
5236     BOOL bool_const;
5237
5238     if(ins->handler_idx == WINED3DSIH_LOOP || ins->handler_idx == WINED3DSIH_REP)
5239     {
5240         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5241         list_add_head(&priv->control_frames, &control_frame->entry);
5242
5243         if(ins->handler_idx == WINED3DSIH_LOOP) control_frame->type = LOOP;
5244         if(ins->handler_idx == WINED3DSIH_REP) control_frame->type = REP;
5245
5246         if(priv->target_version >= NV2)
5247         {
5248             control_frame->no.loop = priv->num_loops++;
5249             priv->loop_depth++;
5250         }
5251         else
5252         {
5253             /* Don't bother recording when we're in a not used if branch */
5254             if(priv->muted)
5255             {
5256                 return;
5257             }
5258
5259             if(!priv->recording)
5260             {
5261                 list_init(&priv->record);
5262                 priv->recording = TRUE;
5263                 control_frame->outer_loop = TRUE;
5264                 get_loop_control_const(ins, shader, ins->src[0].reg.idx, &control_frame->loop_control);
5265                 return; /* Instruction is handled */
5266             }
5267             /* Record this loop in the outer loop's recording */
5268         }
5269     }
5270     else if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5271     {
5272         if(priv->target_version >= NV2)
5273         {
5274             /* Nothing to do. The control frame is popped after the HW instr handler */
5275         }
5276         else
5277         {
5278             struct list *e = list_head(&priv->control_frames);
5279             control_frame = LIST_ENTRY(e, struct control_frame, entry);
5280             list_remove(&control_frame->entry);
5281
5282             if(control_frame->outer_loop)
5283             {
5284                 unsigned int iteration;
5285                 int aL = 0;
5286                 struct list copy;
5287
5288                 /* Turn off recording before playback */
5289                 priv->recording = FALSE;
5290
5291                 /* Move the recorded instructions to a separate list and get them out of the private data
5292                  * structure. If there are nested loops, the shader_arb_handle_instruction below will
5293                  * be recorded again, thus priv->record might be overwritten
5294                  */
5295                 list_init(&copy);
5296                 list_move_tail(&copy, &priv->record);
5297                 list_init(&priv->record);
5298
5299                 if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5300                 {
5301                     shader_addline(buffer, "#unrolling loop: %u iterations, aL=%u, inc %d\n",
5302                                    control_frame->loop_control.count, control_frame->loop_control.start,
5303                                    control_frame->loop_control.step);
5304                     aL = control_frame->loop_control.start;
5305                 }
5306                 else
5307                 {
5308                     shader_addline(buffer, "#unrolling rep: %u iterations\n", control_frame->loop_control.count);
5309                 }
5310
5311                 for (iteration = 0; iteration < control_frame->loop_control.count; ++iteration)
5312                 {
5313                     struct recorded_instruction *rec_ins;
5314                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5315                     {
5316                         priv->aL = aL;
5317                         shader_addline(buffer, "#Iteration %u, aL=%d\n", iteration, aL);
5318                     }
5319                     else
5320                     {
5321                         shader_addline(buffer, "#Iteration %u\n", iteration);
5322                     }
5323
5324                     LIST_FOR_EACH_ENTRY(rec_ins, &copy, struct recorded_instruction, entry)
5325                     {
5326                         shader_arb_handle_instruction(&rec_ins->ins);
5327                     }
5328
5329                     if(ins->handler_idx == WINED3DSIH_ENDLOOP)
5330                     {
5331                         aL += control_frame->loop_control.step;
5332                     }
5333                 }
5334                 shader_addline(buffer, "#end loop/rep\n");
5335
5336                 free_recorded_instruction(&copy);
5337                 HeapFree(GetProcessHeap(), 0, control_frame);
5338                 return; /* Instruction is handled */
5339             }
5340             else
5341             {
5342                 /* This is a nested loop. Proceed to the normal recording function */
5343                 HeapFree(GetProcessHeap(), 0, control_frame);
5344             }
5345         }
5346     }
5347
5348     if(priv->recording)
5349     {
5350         record_instruction(&priv->record, ins);
5351         return;
5352     }
5353
5354     /* boolean if */
5355     if(ins->handler_idx == WINED3DSIH_IF)
5356     {
5357         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5358         list_add_head(&priv->control_frames, &control_frame->entry);
5359         control_frame->type = IF;
5360
5361         bool_const = get_bool_const(ins, shader, ins->src[0].reg.idx);
5362         if(ins->src[0].modifiers == WINED3DSPSM_NOT) bool_const = !bool_const;
5363         if (!priv->muted && !bool_const)
5364         {
5365             shader_addline(buffer, "#if(FALSE){\n");
5366             priv->muted = TRUE;
5367             control_frame->muting = TRUE;
5368         }
5369         else shader_addline(buffer, "#if(TRUE) {\n");
5370
5371         return; /* Instruction is handled */
5372     }
5373     else if(ins->handler_idx == WINED3DSIH_IFC)
5374     {
5375         /* IF(bool) and if_cond(a, b) use the same ELSE and ENDIF tokens */
5376         control_frame = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*control_frame));
5377         control_frame->type = IFC;
5378         control_frame->no.ifc = priv->num_ifcs++;
5379         list_add_head(&priv->control_frames, &control_frame->entry);
5380     }
5381     else if(ins->handler_idx == WINED3DSIH_ELSE)
5382     {
5383         struct list *e = list_head(&priv->control_frames);
5384         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5385
5386         if(control_frame->type == IF)
5387         {
5388             shader_addline(buffer, "#} else {\n");
5389             if(!priv->muted && !control_frame->muting)
5390             {
5391                 priv->muted = TRUE;
5392                 control_frame->muting = TRUE;
5393             }
5394             else if(control_frame->muting) priv->muted = FALSE;
5395             return; /* Instruction is handled. */
5396         }
5397         /* In case of an ifc, generate a HW shader instruction */
5398     }
5399     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5400     {
5401         struct list *e = list_head(&priv->control_frames);
5402         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5403
5404         if(control_frame->type == IF)
5405         {
5406             shader_addline(buffer, "#} endif\n");
5407             if(control_frame->muting) priv->muted = FALSE;
5408             list_remove(&control_frame->entry);
5409             HeapFree(GetProcessHeap(), 0, control_frame);
5410             return; /* Instruction is handled */
5411         }
5412     }
5413
5414     if(priv->muted) return;
5415
5416     /* Select handler */
5417     hw_fct = shader_arb_instruction_handler_table[ins->handler_idx];
5418
5419     /* Unhandled opcode */
5420     if (!hw_fct)
5421     {
5422         FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5423         return;
5424     }
5425     hw_fct(ins);
5426
5427     if(ins->handler_idx == WINED3DSIH_ENDLOOP || ins->handler_idx == WINED3DSIH_ENDREP)
5428     {
5429         struct list *e = list_head(&priv->control_frames);
5430         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5431         list_remove(&control_frame->entry);
5432         HeapFree(GetProcessHeap(), 0, control_frame);
5433         priv->loop_depth--;
5434     }
5435     else if(ins->handler_idx == WINED3DSIH_ENDIF)
5436     {
5437         /* Non-ifc ENDIFs don't reach that place because of the return in the if block above */
5438         struct list *e = list_head(&priv->control_frames);
5439         control_frame = LIST_ENTRY(e, struct control_frame, entry);
5440         list_remove(&control_frame->entry);
5441         HeapFree(GetProcessHeap(), 0, control_frame);
5442     }
5443
5444
5445     shader_arb_add_instruction_modifiers(ins);
5446 }
5447
5448 const shader_backend_t arb_program_shader_backend = {
5449     shader_arb_handle_instruction,
5450     shader_arb_select,
5451     shader_arb_select_depth_blt,
5452     shader_arb_deselect_depth_blt,
5453     shader_arb_update_float_vertex_constants,
5454     shader_arb_update_float_pixel_constants,
5455     shader_arb_load_constants,
5456     shader_arb_load_np2fixup_constants,
5457     shader_arb_destroy,
5458     shader_arb_alloc,
5459     shader_arb_free,
5460     shader_arb_dirty_const,
5461     shader_arb_get_caps,
5462     shader_arb_color_fixup_supported,
5463 };
5464
5465 /* ARB_fragment_program fixed function pipeline replacement definitions */
5466 #define ARB_FFP_CONST_TFACTOR           0
5467 #define ARB_FFP_CONST_SPECULAR_ENABLE   ((ARB_FFP_CONST_TFACTOR) + 1)
5468 #define ARB_FFP_CONST_CONSTANT(i)       ((ARB_FFP_CONST_SPECULAR_ENABLE) + 1 + i)
5469 #define ARB_FFP_CONST_BUMPMAT(i)        ((ARB_FFP_CONST_CONSTANT(7)) + 1 + i)
5470 #define ARB_FFP_CONST_LUMINANCE(i)      ((ARB_FFP_CONST_BUMPMAT(7)) + 1 + i)
5471
5472 struct arbfp_ffp_desc
5473 {
5474     struct ffp_frag_desc parent;
5475     GLuint shader;
5476     unsigned int num_textures_used;
5477 };
5478
5479 /* Context activation is done by the caller. */
5480 static void arbfp_enable(BOOL enable)
5481 {
5482     ENTER_GL();
5483     if(enable) {
5484         glEnable(GL_FRAGMENT_PROGRAM_ARB);
5485         checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
5486     } else {
5487         glDisable(GL_FRAGMENT_PROGRAM_ARB);
5488         checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
5489     }
5490     LEAVE_GL();
5491 }
5492
5493 static HRESULT arbfp_alloc(IWineD3DDeviceImpl *device)
5494 {
5495     struct shader_arb_priv *priv;
5496     /* Share private data between the shader backend and the pipeline replacement, if both
5497      * are the arb implementation. This is needed to figure out whether ARBfp should be disabled
5498      * if no pixel shader is bound or not
5499      */
5500     if (device->shader_backend == &arb_program_shader_backend)
5501     {
5502         device->fragment_priv = device->shader_priv;
5503     }
5504     else
5505     {
5506         device->fragment_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_arb_priv));
5507         if (!device->fragment_priv) return E_OUTOFMEMORY;
5508     }
5509     priv = device->fragment_priv;
5510     if (wine_rb_init(&priv->fragment_shaders, &wined3d_ffp_frag_program_rb_functions) == -1)
5511     {
5512         ERR("Failed to initialize rbtree.\n");
5513         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5514         return E_OUTOFMEMORY;
5515     }
5516     priv->use_arbfp_fixed_func = TRUE;
5517     return WINED3D_OK;
5518 }
5519
5520 /* Context activation is done by the caller. */
5521 static void arbfp_free_ffpshader(struct wine_rb_entry *entry, void *context)
5522 {
5523     const struct wined3d_gl_info *gl_info = context;
5524     struct arbfp_ffp_desc *entry_arb = WINE_RB_ENTRY_VALUE(entry, struct arbfp_ffp_desc, parent.entry);
5525
5526     ENTER_GL();
5527     GL_EXTCALL(glDeleteProgramsARB(1, &entry_arb->shader));
5528     checkGLcall("glDeleteProgramsARB(1, &entry_arb->shader)");
5529     HeapFree(GetProcessHeap(), 0, entry_arb);
5530     LEAVE_GL();
5531 }
5532
5533 /* Context activation is done by the caller. */
5534 static void arbfp_free(IWineD3DDeviceImpl *device)
5535 {
5536     struct shader_arb_priv *priv = device->fragment_priv;
5537
5538     wine_rb_destroy(&priv->fragment_shaders, arbfp_free_ffpshader, &device->adapter->gl_info);
5539     priv->use_arbfp_fixed_func = FALSE;
5540
5541     if (device->shader_backend != &arb_program_shader_backend)
5542     {
5543         HeapFree(GetProcessHeap(), 0, device->fragment_priv);
5544     }
5545 }
5546
5547 static void arbfp_get_caps(const struct wined3d_gl_info *gl_info, struct fragment_caps *caps)
5548 {
5549     caps->PrimitiveMiscCaps = WINED3DPMISCCAPS_TSSARGTEMP;
5550     caps->TextureOpCaps =  WINED3DTEXOPCAPS_DISABLE                     |
5551                            WINED3DTEXOPCAPS_SELECTARG1                  |
5552                            WINED3DTEXOPCAPS_SELECTARG2                  |
5553                            WINED3DTEXOPCAPS_MODULATE4X                  |
5554                            WINED3DTEXOPCAPS_MODULATE2X                  |
5555                            WINED3DTEXOPCAPS_MODULATE                    |
5556                            WINED3DTEXOPCAPS_ADDSIGNED2X                 |
5557                            WINED3DTEXOPCAPS_ADDSIGNED                   |
5558                            WINED3DTEXOPCAPS_ADD                         |
5559                            WINED3DTEXOPCAPS_SUBTRACT                    |
5560                            WINED3DTEXOPCAPS_ADDSMOOTH                   |
5561                            WINED3DTEXOPCAPS_BLENDCURRENTALPHA           |
5562                            WINED3DTEXOPCAPS_BLENDFACTORALPHA            |
5563                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHA           |
5564                            WINED3DTEXOPCAPS_BLENDDIFFUSEALPHA           |
5565                            WINED3DTEXOPCAPS_BLENDTEXTUREALPHAPM         |
5566                            WINED3DTEXOPCAPS_MODULATEALPHA_ADDCOLOR      |
5567                            WINED3DTEXOPCAPS_MODULATECOLOR_ADDALPHA      |
5568                            WINED3DTEXOPCAPS_MODULATEINVCOLOR_ADDALPHA   |
5569                            WINED3DTEXOPCAPS_MODULATEINVALPHA_ADDCOLOR   |
5570                            WINED3DTEXOPCAPS_DOTPRODUCT3                 |
5571                            WINED3DTEXOPCAPS_MULTIPLYADD                 |
5572                            WINED3DTEXOPCAPS_LERP                        |
5573                            WINED3DTEXOPCAPS_BUMPENVMAP                  |
5574                            WINED3DTEXOPCAPS_BUMPENVMAPLUMINANCE;
5575
5576     /* TODO: Implement WINED3DTEXOPCAPS_PREMODULATE */
5577
5578     caps->MaxTextureBlendStages   = 8;
5579     caps->MaxSimultaneousTextures = min(gl_info->limits.fragment_samplers, 8);
5580 }
5581
5582 static void state_texfactor_arbfp(DWORD state_id,
5583         struct wined3d_stateblock *stateblock, struct wined3d_context *context)
5584 {
5585     const struct wined3d_gl_info *gl_info = context->gl_info;
5586     const struct wined3d_state *state = &stateblock->state;
5587     IWineD3DDeviceImpl *device = stateblock->device;
5588     float col[4];
5589
5590     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5591      * application provided constants
5592      */
5593     if (device->shader_backend == &arb_program_shader_backend)
5594     {
5595         if (use_ps(state)) return;
5596
5597         context->pshader_const_dirty[ARB_FFP_CONST_TFACTOR] = 1;
5598         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_TFACTOR + 1);
5599     }
5600
5601     D3DCOLORTOGLFLOAT4(state->render_states[WINED3DRS_TEXTUREFACTOR], col);
5602     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col));
5603     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_TFACTOR, col)");
5604
5605 }
5606
5607 static void state_arb_specularenable(DWORD state_id,
5608         struct wined3d_stateblock *stateblock, struct wined3d_context *context)
5609 {
5610     const struct wined3d_gl_info *gl_info = context->gl_info;
5611     const struct wined3d_state *state = &stateblock->state;
5612     IWineD3DDeviceImpl *device = stateblock->device;
5613     float col[4];
5614
5615     /* Don't load the parameter if we're using an arbfp pixel shader, otherwise we'll overwrite
5616      * application provided constants
5617      */
5618     if (device->shader_backend == &arb_program_shader_backend)
5619     {
5620         if (use_ps(state)) return;
5621
5622         context->pshader_const_dirty[ARB_FFP_CONST_SPECULAR_ENABLE] = 1;
5623         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_SPECULAR_ENABLE + 1);
5624     }
5625
5626     if (state->render_states[WINED3DRS_SPECULARENABLE])
5627     {
5628         /* The specular color has no alpha */
5629         col[0] = 1.0f; col[1] = 1.0f;
5630         col[2] = 1.0f; col[3] = 0.0f;
5631     } else {
5632         col[0] = 0.0f; col[1] = 0.0f;
5633         col[2] = 0.0f; col[3] = 0.0f;
5634     }
5635     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col));
5636     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_SPECULAR_ENABLE, col)");
5637 }
5638
5639 static void set_bumpmat_arbfp(DWORD state_id, struct wined3d_stateblock *stateblock, struct wined3d_context *context)
5640 {
5641     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5642     const struct wined3d_gl_info *gl_info = context->gl_info;
5643     const struct wined3d_state *state = &stateblock->state;
5644     IWineD3DDeviceImpl *device = stateblock->device;
5645     float mat[2][2];
5646
5647     if (use_ps(state))
5648     {
5649         IWineD3DPixelShaderImpl *ps = state->pixel_shader;
5650         if (stage && (ps->baseShader.reg_maps.bumpmat & (1 << stage)))
5651         {
5652             /* The pixel shader has to know the bump env matrix. Do a constants update if it isn't scheduled
5653              * anyway
5654              */
5655             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5656                 stateblock_apply_state(STATE_PIXELSHADERCONSTANT, stateblock, context);
5657         }
5658
5659         if(device->shader_backend == &arb_program_shader_backend) {
5660             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5661             return;
5662         }
5663     } else if(device->shader_backend == &arb_program_shader_backend) {
5664         context->pshader_const_dirty[ARB_FFP_CONST_BUMPMAT(stage)] = 1;
5665         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_BUMPMAT(stage) + 1);
5666     }
5667
5668     mat[0][0] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT00]);
5669     mat[0][1] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT01]);
5670     mat[1][0] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT10]);
5671     mat[1][1] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVMAT11]);
5672
5673     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0]));
5674     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_BUMPMAT(stage), &mat[0][0])");
5675 }
5676
5677 static void tex_bumpenvlum_arbfp(DWORD state_id,
5678         struct wined3d_stateblock *stateblock, struct wined3d_context *context)
5679 {
5680     DWORD stage = (state_id - STATE_TEXTURESTAGE(0, 0)) / (WINED3D_HIGHEST_TEXTURE_STATE + 1);
5681     const struct wined3d_gl_info *gl_info = context->gl_info;
5682     const struct wined3d_state *state = &stateblock->state;
5683     IWineD3DDeviceImpl *device = stateblock->device;
5684     float param[4];
5685
5686     if (use_ps(state))
5687     {
5688         IWineD3DPixelShaderImpl *ps = state->pixel_shader;
5689         if (stage && (ps->baseShader.reg_maps.luminanceparams & (1 << stage)))
5690         {
5691             /* The pixel shader has to know the luminance offset. Do a constants update if it
5692              * isn't scheduled anyway
5693              */
5694             if (!isStateDirty(context, STATE_PIXELSHADERCONSTANT))
5695                 stateblock_apply_state(STATE_PIXELSHADERCONSTANT, stateblock, context);
5696         }
5697
5698         if(device->shader_backend == &arb_program_shader_backend) {
5699             /* Exit now, don't set the bumpmat below, otherwise we may overwrite pixel shader constants */
5700             return;
5701         }
5702     } else if(device->shader_backend == &arb_program_shader_backend) {
5703         context->pshader_const_dirty[ARB_FFP_CONST_LUMINANCE(stage)] = 1;
5704         device->highest_dirty_ps_const = max(device->highest_dirty_ps_const, ARB_FFP_CONST_LUMINANCE(stage) + 1);
5705     }
5706
5707     param[0] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVLSCALE]);
5708     param[1] = *((float *)&state->texture_states[stage][WINED3DTSS_BUMPENVLOFFSET]);
5709     param[2] = 0.0f;
5710     param[3] = 0.0f;
5711
5712     GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param));
5713     checkGLcall("glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, ARB_FFP_CONST_LUMINANCE(stage), param)");
5714 }
5715
5716 static const char *get_argreg(struct wined3d_shader_buffer *buffer, DWORD argnum, unsigned int stage, DWORD arg)
5717 {
5718     const char *ret;
5719
5720     if(arg == ARG_UNUSED) return "unused"; /* This is the marker for unused registers */
5721
5722     switch(arg & WINED3DTA_SELECTMASK) {
5723         case WINED3DTA_DIFFUSE:
5724             ret = "fragment.color.primary"; break;
5725
5726         case WINED3DTA_CURRENT:
5727             if (!stage) ret = "fragment.color.primary";
5728             else ret = "ret";
5729             break;
5730
5731         case WINED3DTA_TEXTURE:
5732             switch(stage) {
5733                 case 0: ret = "tex0"; break;
5734                 case 1: ret = "tex1"; break;
5735                 case 2: ret = "tex2"; break;
5736                 case 3: ret = "tex3"; break;
5737                 case 4: ret = "tex4"; break;
5738                 case 5: ret = "tex5"; break;
5739                 case 6: ret = "tex6"; break;
5740                 case 7: ret = "tex7"; break;
5741                 default: ret = "unknown texture";
5742             }
5743             break;
5744
5745         case WINED3DTA_TFACTOR:
5746             ret = "tfactor"; break;
5747
5748         case WINED3DTA_SPECULAR:
5749             ret = "fragment.color.secondary"; break;
5750
5751         case WINED3DTA_TEMP:
5752             ret = "tempreg"; break;
5753
5754         case WINED3DTA_CONSTANT:
5755             FIXME("Implement perstage constants\n");
5756             switch(stage) {
5757                 case 0: ret = "const0"; break;
5758                 case 1: ret = "const1"; break;
5759                 case 2: ret = "const2"; break;
5760                 case 3: ret = "const3"; break;
5761                 case 4: ret = "const4"; break;
5762                 case 5: ret = "const5"; break;
5763                 case 6: ret = "const6"; break;
5764                 case 7: ret = "const7"; break;
5765                 default: ret = "unknown constant";
5766             }
5767             break;
5768
5769         default:
5770             return "unknown";
5771     }
5772
5773     if(arg & WINED3DTA_COMPLEMENT) {
5774         shader_addline(buffer, "SUB arg%u, const.x, %s;\n", argnum, ret);
5775         if(argnum == 0) ret = "arg0";
5776         if(argnum == 1) ret = "arg1";
5777         if(argnum == 2) ret = "arg2";
5778     }
5779     if(arg & WINED3DTA_ALPHAREPLICATE) {
5780         shader_addline(buffer, "MOV arg%u, %s.w;\n", argnum, ret);
5781         if(argnum == 0) ret = "arg0";
5782         if(argnum == 1) ret = "arg1";
5783         if(argnum == 2) ret = "arg2";
5784     }
5785     return ret;
5786 }
5787
5788 static void gen_ffp_instr(struct wined3d_shader_buffer *buffer, unsigned int stage, BOOL color,
5789         BOOL alpha, DWORD dst, DWORD op, DWORD dw_arg0, DWORD dw_arg1, DWORD dw_arg2)
5790 {
5791     const char *dstmask, *dstreg, *arg0, *arg1, *arg2;
5792     unsigned int mul = 1;
5793     BOOL mul_final_dest = FALSE;
5794
5795     if(color && alpha) dstmask = "";
5796     else if(color) dstmask = ".xyz";
5797     else dstmask = ".w";
5798
5799     if(dst == tempreg) dstreg = "tempreg";
5800     else dstreg = "ret";
5801
5802     arg0 = get_argreg(buffer, 0, stage, dw_arg0);
5803     arg1 = get_argreg(buffer, 1, stage, dw_arg1);
5804     arg2 = get_argreg(buffer, 2, stage, dw_arg2);
5805
5806     switch(op) {
5807         case WINED3DTOP_DISABLE:
5808             if (!stage) shader_addline(buffer, "MOV %s%s, fragment.color.primary;\n", dstreg, dstmask);
5809             break;
5810
5811         case WINED3DTOP_SELECTARG2:
5812             arg1 = arg2;
5813         case WINED3DTOP_SELECTARG1:
5814             shader_addline(buffer, "MOV %s%s, %s;\n", dstreg, dstmask, arg1);
5815             break;
5816
5817         case WINED3DTOP_MODULATE4X:
5818             mul = 2;
5819         case WINED3DTOP_MODULATE2X:
5820             mul *= 2;
5821             if (!strcmp(dstreg, "result.color"))
5822             {
5823                 dstreg = "ret";
5824                 mul_final_dest = TRUE;
5825             }
5826         case WINED3DTOP_MODULATE:
5827             shader_addline(buffer, "MUL %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5828             break;
5829
5830         case WINED3DTOP_ADDSIGNED2X:
5831             mul = 2;
5832             if (!strcmp(dstreg, "result.color"))
5833             {
5834                 dstreg = "ret";
5835                 mul_final_dest = TRUE;
5836             }
5837         case WINED3DTOP_ADDSIGNED:
5838             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5839             arg2 = "arg2";
5840         case WINED3DTOP_ADD:
5841             shader_addline(buffer, "ADD_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5842             break;
5843
5844         case WINED3DTOP_SUBTRACT:
5845             shader_addline(buffer, "SUB_SAT %s%s, %s, %s;\n", dstreg, dstmask, arg1, arg2);
5846             break;
5847
5848         case WINED3DTOP_ADDSMOOTH:
5849             shader_addline(buffer, "SUB arg1, const.x, %s;\n", arg1);
5850             shader_addline(buffer, "MAD_SAT %s%s, arg1, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5851             break;
5852
5853         case WINED3DTOP_BLENDCURRENTALPHA:
5854             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_CURRENT);
5855             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5856             break;
5857         case WINED3DTOP_BLENDFACTORALPHA:
5858             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TFACTOR);
5859             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5860             break;
5861         case WINED3DTOP_BLENDTEXTUREALPHA:
5862             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5863             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5864             break;
5865         case WINED3DTOP_BLENDDIFFUSEALPHA:
5866             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_DIFFUSE);
5867             shader_addline(buffer, "LRP %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5868             break;
5869
5870         case WINED3DTOP_BLENDTEXTUREALPHAPM:
5871             arg0 = get_argreg(buffer, 0, stage, WINED3DTA_TEXTURE);
5872             shader_addline(buffer, "SUB arg0.w, const.x, %s.w;\n", arg0);
5873             shader_addline(buffer, "MAD_SAT %s%s, %s, arg0.w, %s;\n", dstreg, dstmask, arg2, arg1);
5874             break;
5875
5876         /* D3DTOP_PREMODULATE ???? */
5877
5878         case WINED3DTOP_MODULATEINVALPHA_ADDCOLOR:
5879             shader_addline(buffer, "SUB arg0.w, const.x, %s;\n", arg1);
5880             shader_addline(buffer, "MAD_SAT %s%s, arg0.w, %s, %s;\n", dstreg, dstmask, arg2, arg1);
5881             break;
5882         case WINED3DTOP_MODULATEALPHA_ADDCOLOR:
5883             shader_addline(buffer, "MAD_SAT %s%s, %s.w, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg1);
5884             break;
5885         case WINED3DTOP_MODULATEINVCOLOR_ADDALPHA:
5886             shader_addline(buffer, "SUB arg0, const.x, %s;\n", arg1);
5887             shader_addline(buffer, "MAD_SAT %s%s, arg0, %s, %s.w;\n", dstreg, dstmask, arg2, arg1);
5888             break;
5889         case WINED3DTOP_MODULATECOLOR_ADDALPHA:
5890             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s.w;\n", dstreg, dstmask, arg1, arg2, arg1);
5891             break;
5892
5893         case WINED3DTOP_DOTPRODUCT3:
5894             mul = 4;
5895             if (!strcmp(dstreg, "result.color"))
5896             {
5897                 dstreg = "ret";
5898                 mul_final_dest = TRUE;
5899             }
5900             shader_addline(buffer, "SUB arg1, %s, const.w;\n", arg1);
5901             shader_addline(buffer, "SUB arg2, %s, const.w;\n", arg2);
5902             shader_addline(buffer, "DP3_SAT %s%s, arg1, arg2;\n", dstreg, dstmask);
5903             break;
5904
5905         case WINED3DTOP_MULTIPLYADD:
5906             shader_addline(buffer, "MAD_SAT %s%s, %s, %s, %s;\n", dstreg, dstmask, arg1, arg2, arg0);
5907             break;
5908
5909         case WINED3DTOP_LERP:
5910             /* The msdn is not quite right here */
5911             shader_addline(buffer, "LRP %s%s, %s, %s, %s;\n", dstreg, dstmask, arg0, arg1, arg2);
5912             break;
5913
5914         case WINED3DTOP_BUMPENVMAP:
5915         case WINED3DTOP_BUMPENVMAPLUMINANCE:
5916             /* Those are handled in the first pass of the shader(generation pass 1 and 2) already */
5917             break;
5918
5919         default:
5920             FIXME("Unhandled texture op %08x\n", op);
5921     }
5922
5923     if(mul == 2) {
5924         shader_addline(buffer, "MUL_SAT %s%s, %s, const.y;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5925     } else if(mul == 4) {
5926         shader_addline(buffer, "MUL_SAT %s%s, %s, const.z;\n", mul_final_dest ? "result.color" : dstreg, dstmask, dstreg);
5927     }
5928 }
5929
5930 static GLuint gen_arbfp_ffp_shader(const struct ffp_frag_settings *settings, struct wined3d_stateblock *stateblock)
5931 {
5932     const struct wined3d_gl_info *gl_info = &stateblock->device->adapter->gl_info;
5933     unsigned int stage;
5934     struct wined3d_shader_buffer buffer;
5935     BOOL tex_read[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5936     BOOL bump_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5937     BOOL luminance_used[MAX_TEXTURES] = {FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE};
5938     const char *textype;
5939     const char *instr, *sat;
5940     char colorcor_dst[8];
5941     GLuint ret;
5942     DWORD arg0, arg1, arg2;
5943     BOOL tempreg_used = FALSE, tfactor_used = FALSE;
5944     BOOL op_equal;
5945     const char *final_combiner_src = "ret";
5946     GLint pos;
5947
5948     /* Find out which textures are read */
5949     for(stage = 0; stage < MAX_TEXTURES; stage++) {
5950         if(settings->op[stage].cop == WINED3DTOP_DISABLE) break;
5951         arg0 = settings->op[stage].carg0 & WINED3DTA_SELECTMASK;
5952         arg1 = settings->op[stage].carg1 & WINED3DTA_SELECTMASK;
5953         arg2 = settings->op[stage].carg2 & WINED3DTA_SELECTMASK;
5954         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5955         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5956         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5957
5958         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHA) tex_read[stage] = TRUE;
5959         if(settings->op[stage].cop == WINED3DTOP_BLENDTEXTUREALPHAPM) tex_read[stage] = TRUE;
5960         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP) {
5961             bump_used[stage] = TRUE;
5962             tex_read[stage] = TRUE;
5963         }
5964         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
5965             bump_used[stage] = TRUE;
5966             tex_read[stage] = TRUE;
5967             luminance_used[stage] = TRUE;
5968         } else if(settings->op[stage].cop == WINED3DTOP_BLENDFACTORALPHA) {
5969             tfactor_used = TRUE;
5970         }
5971
5972         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5973             tfactor_used = TRUE;
5974         }
5975
5976         if(settings->op[stage].dst == tempreg) tempreg_used = TRUE;
5977         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5978             tempreg_used = TRUE;
5979         }
5980
5981         if(settings->op[stage].aop == WINED3DTOP_DISABLE) continue;
5982         arg0 = settings->op[stage].aarg0 & WINED3DTA_SELECTMASK;
5983         arg1 = settings->op[stage].aarg1 & WINED3DTA_SELECTMASK;
5984         arg2 = settings->op[stage].aarg2 & WINED3DTA_SELECTMASK;
5985         if(arg0 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5986         if(arg1 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5987         if(arg2 == WINED3DTA_TEXTURE) tex_read[stage] = TRUE;
5988
5989         if(arg0 == WINED3DTA_TEMP || arg1 == WINED3DTA_TEMP || arg2 == WINED3DTA_TEMP) {
5990             tempreg_used = TRUE;
5991         }
5992         if(arg0 == WINED3DTA_TFACTOR || arg1 == WINED3DTA_TFACTOR || arg2 == WINED3DTA_TFACTOR) {
5993             tfactor_used = TRUE;
5994         }
5995     }
5996
5997     /* Shader header */
5998     if (!shader_buffer_init(&buffer))
5999     {
6000         ERR("Failed to initialize shader buffer.\n");
6001         return 0;
6002     }
6003
6004     shader_addline(&buffer, "!!ARBfp1.0\n");
6005
6006     switch(settings->fog) {
6007         case FOG_OFF:                                                         break;
6008         case FOG_LINEAR: shader_addline(&buffer, "OPTION ARB_fog_linear;\n"); break;
6009         case FOG_EXP:    shader_addline(&buffer, "OPTION ARB_fog_exp;\n");    break;
6010         case FOG_EXP2:   shader_addline(&buffer, "OPTION ARB_fog_exp2;\n");   break;
6011         default: FIXME("Unexpected fog setting %d\n", settings->fog);
6012     }
6013
6014     shader_addline(&buffer, "PARAM const = {1, 2, 4, 0.5};\n");
6015     shader_addline(&buffer, "TEMP TMP;\n");
6016     shader_addline(&buffer, "TEMP ret;\n");
6017     if(tempreg_used || settings->sRGB_write) shader_addline(&buffer, "TEMP tempreg;\n");
6018     shader_addline(&buffer, "TEMP arg0;\n");
6019     shader_addline(&buffer, "TEMP arg1;\n");
6020     shader_addline(&buffer, "TEMP arg2;\n");
6021     for(stage = 0; stage < MAX_TEXTURES; stage++) {
6022         if(!tex_read[stage]) continue;
6023         shader_addline(&buffer, "TEMP tex%u;\n", stage);
6024         if(!bump_used[stage]) continue;
6025         shader_addline(&buffer, "PARAM bumpmat%u = program.env[%u];\n", stage, ARB_FFP_CONST_BUMPMAT(stage));
6026         if(!luminance_used[stage]) continue;
6027         shader_addline(&buffer, "PARAM luminance%u = program.env[%u];\n", stage, ARB_FFP_CONST_LUMINANCE(stage));
6028     }
6029     if(tfactor_used) {
6030         shader_addline(&buffer, "PARAM tfactor = program.env[%u];\n", ARB_FFP_CONST_TFACTOR);
6031     }
6032         shader_addline(&buffer, "PARAM specular_enable = program.env[%u];\n", ARB_FFP_CONST_SPECULAR_ENABLE);
6033
6034     if(settings->sRGB_write) {
6035         shader_addline(&buffer, "PARAM srgb_consts1 = {%f, %f, %f, %f};\n",
6036                        srgb_mul_low, srgb_cmp, srgb_pow, srgb_mul_high);
6037         shader_addline(&buffer, "PARAM srgb_consts2 = {%f, %f, %f, %f};\n",
6038                        srgb_sub_high, 0.0, 0.0, 0.0);
6039     }
6040
6041     if (ffp_clip_emul(&stateblock->state) && settings->emul_clipplanes)
6042         shader_addline(&buffer, "KIL fragment.texcoord[7];\n");
6043
6044     /* Generate texture sampling instructions) */
6045     for(stage = 0; stage < MAX_TEXTURES && settings->op[stage].cop != WINED3DTOP_DISABLE; stage++) {
6046         if(!tex_read[stage]) continue;
6047
6048         switch(settings->op[stage].tex_type) {
6049             case tex_1d:                    textype = "1D";     break;
6050             case tex_2d:                    textype = "2D";     break;
6051             case tex_3d:                    textype = "3D";     break;
6052             case tex_cube:                  textype = "CUBE";   break;
6053             case tex_rect:                  textype = "RECT";   break;
6054             default: textype = "unexpected_textype";   break;
6055         }
6056
6057         if(settings->op[stage].cop == WINED3DTOP_BUMPENVMAP ||
6058            settings->op[stage].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
6059             sat = "";
6060         } else {
6061             sat = "_SAT";
6062         }
6063
6064         if(settings->op[stage].projected == proj_none) {
6065             instr = "TEX";
6066         } else if(settings->op[stage].projected == proj_count4 ||
6067                   settings->op[stage].projected == proj_count3) {
6068             instr = "TXP";
6069         } else {
6070             FIXME("Unexpected projection mode %d\n", settings->op[stage].projected);
6071             instr = "TXP";
6072         }
6073
6074         if(stage > 0 &&
6075            (settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAP ||
6076             settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE)) {
6077             shader_addline(&buffer, "SWZ arg1, bumpmat%u, x, z, 0, 0;\n", stage - 1);
6078             shader_addline(&buffer, "DP3 ret.x, arg1, tex%u;\n", stage - 1);
6079             shader_addline(&buffer, "SWZ arg1, bumpmat%u, y, w, 0, 0;\n", stage - 1);
6080             shader_addline(&buffer, "DP3 ret.y, arg1, tex%u;\n", stage - 1);
6081
6082             /* with projective textures, texbem only divides the static texture coord, not the displacement,
6083              * so multiply the displacement with the dividing parameter before passing it to TXP
6084              */
6085             if (settings->op[stage].projected != proj_none) {
6086                 if(settings->op[stage].projected == proj_count4) {
6087                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].w;\n", stage);
6088                     shader_addline(&buffer, "MUL ret.xyz, ret, fragment.texcoord[%u].w, fragment.texcoord[%u];\n", stage, stage);
6089                 } else {
6090                     shader_addline(&buffer, "MOV ret.w, fragment.texcoord[%u].z;\n", stage);
6091                     shader_addline(&buffer, "MAD ret.xyz, ret, fragment.texcoord[%u].z, fragment.texcoord[%u];\n", stage, stage);
6092                 }
6093             } else {
6094                 shader_addline(&buffer, "ADD ret, ret, fragment.texcoord[%u];\n", stage);
6095             }
6096
6097             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6098                            instr, sat, stage, stage, textype);
6099             if(settings->op[stage - 1].cop == WINED3DTOP_BUMPENVMAPLUMINANCE) {
6100                 shader_addline(&buffer, "MAD_SAT ret.x, tex%u.z, luminance%u.x, luminance%u.y;\n",
6101                                stage - 1, stage - 1, stage - 1);
6102                 shader_addline(&buffer, "MUL tex%u, tex%u, ret.x;\n", stage, stage);
6103             }
6104         } else if(settings->op[stage].projected == proj_count3) {
6105             shader_addline(&buffer, "MOV ret, fragment.texcoord[%u];\n", stage);
6106             shader_addline(&buffer, "MOV ret.w, ret.z;\n");
6107             shader_addline(&buffer, "%s%s tex%u, ret, texture[%u], %s;\n",
6108                             instr, sat, stage, stage, textype);
6109         } else {
6110             shader_addline(&buffer, "%s%s tex%u, fragment.texcoord[%u], texture[%u], %s;\n",
6111                             instr, sat, stage, stage, stage, textype);
6112         }
6113
6114         sprintf(colorcor_dst, "tex%u", stage);
6115         gen_color_correction(&buffer, colorcor_dst, WINED3DSP_WRITEMASK_ALL, "const.x", "const.y",
6116                 settings->op[stage].color_fixup);
6117     }
6118
6119     /* Generate the main shader */
6120     for (stage = 0; stage < MAX_TEXTURES; ++stage)
6121     {
6122         if (settings->op[stage].cop == WINED3DTOP_DISABLE)
6123         {
6124             if (!stage) final_combiner_src = "fragment.color.primary";
6125             break;
6126         }
6127
6128         if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
6129            settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
6130             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg1;
6131         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG1 &&
6132                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
6133             op_equal = settings->op[stage].carg1 == settings->op[stage].aarg2;
6134         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
6135                   settings->op[stage].aop == WINED3DTOP_SELECTARG1) {
6136             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg1;
6137         } else if(settings->op[stage].cop == WINED3DTOP_SELECTARG2 &&
6138                   settings->op[stage].aop == WINED3DTOP_SELECTARG2) {
6139             op_equal = settings->op[stage].carg2 == settings->op[stage].aarg2;
6140         } else {
6141             op_equal = settings->op[stage].aop   == settings->op[stage].cop &&
6142                        settings->op[stage].carg0 == settings->op[stage].aarg0 &&
6143                        settings->op[stage].carg1 == settings->op[stage].aarg1 &&
6144                        settings->op[stage].carg2 == settings->op[stage].aarg2;
6145         }
6146
6147         if(settings->op[stage].aop == WINED3DTOP_DISABLE) {
6148             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6149                           settings->op[stage].cop, settings->op[stage].carg0,
6150                           settings->op[stage].carg1, settings->op[stage].carg2);
6151             if (!stage)
6152                 shader_addline(&buffer, "MOV ret.w, fragment.color.primary.w;\n");
6153         }
6154         else if (op_equal)
6155         {
6156             gen_ffp_instr(&buffer, stage, TRUE, TRUE, settings->op[stage].dst,
6157                           settings->op[stage].cop, settings->op[stage].carg0,
6158                           settings->op[stage].carg1, settings->op[stage].carg2);
6159         } else {
6160             gen_ffp_instr(&buffer, stage, TRUE, FALSE, settings->op[stage].dst,
6161                           settings->op[stage].cop, settings->op[stage].carg0,
6162                           settings->op[stage].carg1, settings->op[stage].carg2);
6163             gen_ffp_instr(&buffer, stage, FALSE, TRUE, settings->op[stage].dst,
6164                           settings->op[stage].aop, settings->op[stage].aarg0,
6165                           settings->op[stage].aarg1, settings->op[stage].aarg2);
6166         }
6167     }
6168
6169     if(settings->sRGB_write) {
6170         shader_addline(&buffer, "MAD ret, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6171         arbfp_add_sRGB_correction(&buffer, "ret", "arg0", "arg1", "arg2", "tempreg", FALSE);
6172         shader_addline(&buffer, "MOV result.color, ret;\n");
6173     } else {
6174         shader_addline(&buffer, "MAD result.color, fragment.color.secondary, specular_enable, %s;\n", final_combiner_src);
6175     }
6176
6177     /* Footer */
6178     shader_addline(&buffer, "END\n");
6179
6180     /* Generate the shader */
6181     GL_EXTCALL(glGenProgramsARB(1, &ret));
6182     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, ret));
6183     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6184             strlen(buffer.buffer), buffer.buffer));
6185     checkGLcall("glProgramStringARB()");
6186
6187     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6188     if (pos != -1)
6189     {
6190         FIXME("Fragment program error at position %d: %s\n\n", pos,
6191               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6192         shader_arb_dump_program_source(buffer.buffer);
6193     }
6194     else
6195     {
6196         GLint native;
6197
6198         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
6199         checkGLcall("glGetProgramivARB()");
6200         if (!native) WARN("Program exceeds native resource limits.\n");
6201     }
6202
6203     shader_buffer_free(&buffer);
6204     return ret;
6205 }
6206
6207 static void fragment_prog_arbfp(DWORD state_id, struct wined3d_stateblock *stateblock, struct wined3d_context *context)
6208 {
6209     const struct wined3d_gl_info *gl_info = context->gl_info;
6210     const struct wined3d_state *state = &stateblock->state;
6211     IWineD3DDeviceImpl *device = stateblock->device;
6212     struct shader_arb_priv *priv = device->fragment_priv;
6213     BOOL use_vshader = use_vs(state);
6214     BOOL use_pshader = use_ps(state);
6215     struct ffp_frag_settings settings;
6216     const struct arbfp_ffp_desc *desc;
6217     unsigned int i;
6218
6219     TRACE("state_id %#x, stateblock %p, context %p\n", state_id, stateblock, context);
6220
6221     if(isStateDirty(context, STATE_RENDER(WINED3DRS_FOGENABLE))) {
6222         if(!use_pshader && device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
6223             /* Reload fixed function constants since they collide with the pixel shader constants */
6224             for(i = 0; i < MAX_TEXTURES; i++) {
6225                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
6226             }
6227             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
6228             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
6229         } else if(use_pshader && !isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
6230             device->shader_backend->shader_select(context, use_pshader, use_vshader);
6231         }
6232         return;
6233     }
6234
6235     if(!use_pshader) {
6236         /* Find or create a shader implementing the fixed function pipeline settings, then activate it */
6237         gen_ffp_frag_op(stateblock, &settings, FALSE);
6238         desc = (const struct arbfp_ffp_desc *)find_ffp_frag_shader(&priv->fragment_shaders, &settings);
6239         if(!desc) {
6240             struct arbfp_ffp_desc *new_desc = HeapAlloc(GetProcessHeap(), 0, sizeof(*new_desc));
6241             if (!new_desc)
6242             {
6243                 ERR("Out of memory\n");
6244                 return;
6245             }
6246             new_desc->num_textures_used = 0;
6247             for (i = 0; i < gl_info->limits.texture_stages; ++i)
6248             {
6249                 if(settings.op[i].cop == WINED3DTOP_DISABLE) break;
6250                 new_desc->num_textures_used = i;
6251             }
6252
6253             memcpy(&new_desc->parent.settings, &settings, sizeof(settings));
6254             new_desc->shader = gen_arbfp_ffp_shader(&settings, stateblock);
6255             add_ffp_frag_shader(&priv->fragment_shaders, &new_desc->parent);
6256             TRACE("Allocated fixed function replacement shader descriptor %p\n", new_desc);
6257             desc = new_desc;
6258         }
6259
6260         /* Now activate the replacement program. GL_FRAGMENT_PROGRAM_ARB is already active(however, note the
6261          * comment above the shader_select call below). If e.g. GLSL is active, the shader_select call will
6262          * deactivate it.
6263          */
6264         GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader));
6265         checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, desc->shader)");
6266         priv->current_fprogram_id = desc->shader;
6267
6268         if(device->shader_backend == &arb_program_shader_backend && context->last_was_pshader) {
6269             /* Reload fixed function constants since they collide with the pixel shader constants */
6270             for(i = 0; i < MAX_TEXTURES; i++) {
6271                 set_bumpmat_arbfp(STATE_TEXTURESTAGE(i, WINED3DTSS_BUMPENVMAT00), stateblock, context);
6272             }
6273             state_texfactor_arbfp(STATE_RENDER(WINED3DRS_TEXTUREFACTOR), stateblock, context);
6274             state_arb_specularenable(STATE_RENDER(WINED3DRS_SPECULARENABLE), stateblock, context);
6275         }
6276         context->last_was_pshader = FALSE;
6277     } else {
6278         context->last_was_pshader = TRUE;
6279     }
6280
6281     /* Finally, select the shader. If a pixel shader is used, it will be set and enabled by the shader backend.
6282      * If this shader backend is arbfp(most likely), then it will simply overwrite the last fixed function replace-
6283      * ment shader. If the shader backend is not ARB, it currently is important that the opengl implementation
6284      * type overwrites GL_ARB_fragment_program. This is currently the case with GLSL. If we really want to use
6285      * atifs or nvrc pixel shaders with arb fragment programs we'd have to disable GL_FRAGMENT_PROGRAM_ARB here
6286      *
6287      * Don't call shader_select if the vertex shader is dirty, because it will be called later on by the vertex
6288      * shader handler
6289      */
6290     if(!isStateDirty(context, device->StateTable[STATE_VSHADER].representative)) {
6291         device->shader_backend->shader_select(context, use_pshader, use_vshader);
6292
6293         if (!isStateDirty(context, STATE_VERTEXSHADERCONSTANT) && (use_vshader || use_pshader))
6294             stateblock_apply_state(STATE_VERTEXSHADERCONSTANT, stateblock, context);
6295     }
6296     if (use_pshader) stateblock_apply_state(STATE_PIXELSHADERCONSTANT, stateblock, context);
6297 }
6298
6299 /* We can't link the fog states to the fragment state directly since the vertex pipeline links them
6300  * to FOGENABLE. A different linking in different pipeline parts can't be expressed in the combined
6301  * state table, so we need to handle that with a forwarding function. The other invisible side effect
6302  * is that changing the fog start and fog end(which links to FOGENABLE in vertex) results in the
6303  * fragment_prog_arbfp function being called because FOGENABLE is dirty, which calls this function here
6304  */
6305 static void state_arbfp_fog(DWORD state_id, struct wined3d_stateblock *stateblock, struct wined3d_context *context)
6306 {
6307     const struct wined3d_state *state = &stateblock->state;
6308     enum fogsource new_source;
6309
6310     TRACE("state_id %#x, stateblock %p, context %p\n", state_id, stateblock, context);
6311
6312     if(!isStateDirty(context, STATE_PIXELSHADER)) {
6313         fragment_prog_arbfp(state_id, stateblock, context);
6314     }
6315
6316     if (!state->render_states[WINED3DRS_FOGENABLE]) return;
6317
6318     if (state->render_states[WINED3DRS_FOGTABLEMODE] == WINED3DFOG_NONE)
6319     {
6320         if (use_vs(state))
6321         {
6322             new_source = FOGSOURCE_VS;
6323         }
6324         else
6325         {
6326             if (state->render_states[WINED3DRS_FOGVERTEXMODE] == WINED3DFOG_NONE || context->last_was_rhw)
6327                 new_source = FOGSOURCE_COORD;
6328             else
6329                 new_source = FOGSOURCE_FFP;
6330         }
6331     } else {
6332         new_source = FOGSOURCE_FFP;
6333     }
6334     if(new_source != context->fog_source) {
6335         context->fog_source = new_source;
6336         state_fogstartend(STATE_RENDER(WINED3DRS_FOGSTART), stateblock, context);
6337     }
6338 }
6339
6340 static void textransform(DWORD state, struct wined3d_stateblock *stateblock, struct wined3d_context *context)
6341 {
6342     if (!isStateDirty(context, STATE_PIXELSHADER))
6343         fragment_prog_arbfp(state, stateblock, context);
6344 }
6345
6346 static const struct StateEntryTemplate arbfp_fragmentstate_template[] = {
6347     {STATE_RENDER(WINED3DRS_TEXTUREFACTOR),               { STATE_RENDER(WINED3DRS_TEXTUREFACTOR),              state_texfactor_arbfp   }, WINED3D_GL_EXT_NONE             },
6348     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6349     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6350     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6351     {STATE_TEXTURESTAGE(0, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6352     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6353     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6354     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6355     {STATE_TEXTURESTAGE(0, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6356     {STATE_TEXTURESTAGE(0, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6357     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6358     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6359     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6360     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6361     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6362     {STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(0, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6363     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6364     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6365     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6366     {STATE_TEXTURESTAGE(1, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6367     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6368     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6369     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6370     {STATE_TEXTURESTAGE(1, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6371     {STATE_TEXTURESTAGE(1, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6372     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6373     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6374     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6375     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6376     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6377     {STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(1, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6378     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6379     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6380     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6381     {STATE_TEXTURESTAGE(2, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6382     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6383     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6384     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6385     {STATE_TEXTURESTAGE(2, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6386     {STATE_TEXTURESTAGE(2, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6387     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6388     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6389     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6390     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6391     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6392     {STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(2, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6393     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6394     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6395     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6396     {STATE_TEXTURESTAGE(3, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6397     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6398     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6399     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6400     {STATE_TEXTURESTAGE(3, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6401     {STATE_TEXTURESTAGE(3, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6402     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6403     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6404     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6405     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6406     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6407     {STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(3, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6408     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6409     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6410     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6411     {STATE_TEXTURESTAGE(4, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6412     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6413     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6414     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6415     {STATE_TEXTURESTAGE(4, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6416     {STATE_TEXTURESTAGE(4, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6417     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6418     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6419     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6420     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6421     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6422     {STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(4, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6423     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6424     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6425     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6426     {STATE_TEXTURESTAGE(5, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6427     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6428     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6429     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6430     {STATE_TEXTURESTAGE(5, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6431     {STATE_TEXTURESTAGE(5, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6432     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6433     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6434     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6435     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6436     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6437     {STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(5, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6438     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6439     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6440     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6441     {STATE_TEXTURESTAGE(6, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6442     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6443     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6444     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6445     {STATE_TEXTURESTAGE(6, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6446     {STATE_TEXTURESTAGE(6, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6447     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6448     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6449     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6450     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6451     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6452     {STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(6, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6453     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLOROP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6454     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6455     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6456     {STATE_TEXTURESTAGE(7, WINED3DTSS_COLORARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6457     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAOP),           { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6458     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG1),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6459     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG2),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6460     {STATE_TEXTURESTAGE(7, WINED3DTSS_ALPHAARG0),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6461     {STATE_TEXTURESTAGE(7, WINED3DTSS_RESULTARG),         { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6462     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     set_bumpmat_arbfp       }, WINED3D_GL_EXT_NONE             },
6463     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT01),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6464     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT10),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6465     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT11),      { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVMAT00),     NULL                    }, WINED3D_GL_EXT_NONE             },
6466     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),     { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    tex_bumpenvlum_arbfp    }, WINED3D_GL_EXT_NONE             },
6467     {STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLOFFSET),    { STATE_TEXTURESTAGE(7, WINED3DTSS_BUMPENVLSCALE),    NULL                    }, WINED3D_GL_EXT_NONE             },
6468     {STATE_PIXELSHADER,                                   { STATE_PIXELSHADER,                                  fragment_prog_arbfp     }, WINED3D_GL_EXT_NONE             },
6469     {STATE_RENDER(WINED3DRS_FOGENABLE),                   { STATE_RENDER(WINED3DRS_FOGENABLE),                  state_arbfp_fog         }, WINED3D_GL_EXT_NONE             },
6470     {STATE_RENDER(WINED3DRS_FOGTABLEMODE),                { STATE_RENDER(WINED3DRS_FOGENABLE),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6471     {STATE_RENDER(WINED3DRS_FOGVERTEXMODE),               { STATE_RENDER(WINED3DRS_FOGENABLE),                  NULL                    }, WINED3D_GL_EXT_NONE             },
6472     {STATE_RENDER(WINED3DRS_FOGSTART),                    { STATE_RENDER(WINED3DRS_FOGSTART),                   state_fogstartend       }, WINED3D_GL_EXT_NONE             },
6473     {STATE_RENDER(WINED3DRS_FOGEND),                      { STATE_RENDER(WINED3DRS_FOGSTART),                   NULL                    }, WINED3D_GL_EXT_NONE             },
6474     {STATE_RENDER(WINED3DRS_SRGBWRITEENABLE),             { STATE_PIXELSHADER,                                  NULL                    }, WINED3D_GL_EXT_NONE             },
6475     {STATE_RENDER(WINED3DRS_FOGCOLOR),                    { STATE_RENDER(WINED3DRS_FOGCOLOR),                   state_fogcolor          }, WINED3D_GL_EXT_NONE             },
6476     {STATE_RENDER(WINED3DRS_FOGDENSITY),                  { STATE_RENDER(WINED3DRS_FOGDENSITY),                 state_fogdensity        }, WINED3D_GL_EXT_NONE             },
6477     {STATE_TEXTURESTAGE(0,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(0, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6478     {STATE_TEXTURESTAGE(1,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(1, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6479     {STATE_TEXTURESTAGE(2,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(2, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6480     {STATE_TEXTURESTAGE(3,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(3, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6481     {STATE_TEXTURESTAGE(4,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(4, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6482     {STATE_TEXTURESTAGE(5,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(5, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6483     {STATE_TEXTURESTAGE(6,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(6, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6484     {STATE_TEXTURESTAGE(7,WINED3DTSS_TEXTURETRANSFORMFLAGS),{STATE_TEXTURESTAGE(7, WINED3DTSS_TEXTURETRANSFORMFLAGS), textransform      }, WINED3D_GL_EXT_NONE             },
6485     {STATE_RENDER(WINED3DRS_SPECULARENABLE),              { STATE_RENDER(WINED3DRS_SPECULARENABLE),             state_arb_specularenable}, WINED3D_GL_EXT_NONE             },
6486     {0 /* Terminate */,                                   { 0,                                                  0                       }, WINED3D_GL_EXT_NONE             },
6487 };
6488
6489 const struct fragment_pipeline arbfp_fragment_pipeline = {
6490     arbfp_enable,
6491     arbfp_get_caps,
6492     arbfp_alloc,
6493     arbfp_free,
6494     shader_arb_color_fixup_supported,
6495     arbfp_fragmentstate_template,
6496     TRUE /* We can disable projected textures */
6497 };
6498
6499 struct arbfp_blit_priv {
6500     GLenum yuy2_rect_shader, yuy2_2d_shader;
6501     GLenum uyvy_rect_shader, uyvy_2d_shader;
6502     GLenum yv12_rect_shader, yv12_2d_shader;
6503     GLenum p8_rect_shader, p8_2d_shader;
6504     GLuint palette_texture;
6505 };
6506
6507 static HRESULT arbfp_blit_alloc(IWineD3DDeviceImpl *device)
6508 {
6509     device->blit_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct arbfp_blit_priv));
6510     if(!device->blit_priv) {
6511         ERR("Out of memory\n");
6512         return E_OUTOFMEMORY;
6513     }
6514     return WINED3D_OK;
6515 }
6516
6517 /* Context activation is done by the caller. */
6518 static void arbfp_blit_free(IWineD3DDeviceImpl *device)
6519 {
6520     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6521     struct arbfp_blit_priv *priv = device->blit_priv;
6522
6523     ENTER_GL();
6524     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_rect_shader));
6525     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yuy2_2d_shader));
6526     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_rect_shader));
6527     GL_EXTCALL(glDeleteProgramsARB(1, &priv->uyvy_2d_shader));
6528     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_rect_shader));
6529     GL_EXTCALL(glDeleteProgramsARB(1, &priv->yv12_2d_shader));
6530     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_rect_shader));
6531     GL_EXTCALL(glDeleteProgramsARB(1, &priv->p8_2d_shader));
6532     checkGLcall("Delete yuv and p8 programs");
6533
6534     if(priv->palette_texture) glDeleteTextures(1, &priv->palette_texture);
6535     LEAVE_GL();
6536
6537     HeapFree(GetProcessHeap(), 0, device->blit_priv);
6538     device->blit_priv = NULL;
6539 }
6540
6541 static BOOL gen_planar_yuv_read(struct wined3d_shader_buffer *buffer, enum complex_fixup fixup,
6542         GLenum textype, char *luminance)
6543 {
6544     char chroma;
6545     const char *tex, *texinstr;
6546
6547     if (fixup == COMPLEX_FIXUP_UYVY) {
6548         chroma = 'x';
6549         *luminance = 'w';
6550     } else {
6551         chroma = 'w';
6552         *luminance = 'x';
6553     }
6554     switch(textype) {
6555         case GL_TEXTURE_2D:             tex = "2D";     texinstr = "TXP"; break;
6556         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   texinstr = "TEX"; break;
6557         default:
6558             /* This is more tricky than just replacing the texture type - we have to navigate
6559              * properly in the texture to find the correct chroma values
6560              */
6561             FIXME("Implement yuv correction for non-2d, non-rect textures\n");
6562             return FALSE;
6563     }
6564
6565     /* First we have to read the chroma values. This means we need at least two pixels(no filtering),
6566      * or 4 pixels(with filtering). To get the unmodified chromas, we have to rid ourselves of the
6567      * filtering when we sample the texture.
6568      *
6569      * These are the rules for reading the chroma:
6570      *
6571      * Even pixel: Cr
6572      * Even pixel: U
6573      * Odd pixel: V
6574      *
6575      * So we have to get the sampling x position in non-normalized coordinates in integers
6576      */
6577     if(textype != GL_TEXTURE_RECTANGLE_ARB) {
6578         shader_addline(buffer, "MUL texcrd.xy, fragment.texcoord[0], size.x;\n");
6579         shader_addline(buffer, "MOV texcrd.w, size.x;\n");
6580     } else {
6581         shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6582     }
6583     /* We must not allow filtering between pixel x and x+1, this would mix U and V
6584      * Vertical filtering is ok. However, bear in mind that the pixel center is at
6585      * 0.5, so add 0.5.
6586      */
6587     shader_addline(buffer, "FLR texcrd.x, texcrd.x;\n");
6588     shader_addline(buffer, "ADD texcrd.x, texcrd.x, coef.y;\n");
6589
6590     /* Divide the x coordinate by 0.5 and get the fraction. This gives 0.25 and 0.75 for the
6591      * even and odd pixels respectively
6592      */
6593     shader_addline(buffer, "MUL texcrd2, texcrd, coef.y;\n");
6594     shader_addline(buffer, "FRC texcrd2, texcrd2;\n");
6595
6596     /* Sample Pixel 1 */
6597     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6598
6599     /* Put the value into either of the chroma values */
6600     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6601     shader_addline(buffer, "MUL chroma.x, luminance.%c, temp.x;\n", chroma);
6602     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6603     shader_addline(buffer, "MUL chroma.y, luminance.%c, temp.x;\n", chroma);
6604
6605     /* Sample pixel 2. If we read an even pixel(SLT above returned 1), sample
6606      * the pixel right to the current one. Otherwise, sample the left pixel.
6607      * Bias and scale the SLT result to -1;1 and add it to the texcrd.x.
6608      */
6609     shader_addline(buffer, "MAD temp.x, temp.x, coef.z, -coef.x;\n");
6610     shader_addline(buffer, "ADD texcrd.x, texcrd, temp.x;\n");
6611     shader_addline(buffer, "%s luminance, texcrd, texture[0], %s;\n", texinstr, tex);
6612
6613     /* Put the value into the other chroma */
6614     shader_addline(buffer, "SGE temp.x, texcrd2.x, coef.y;\n");
6615     shader_addline(buffer, "MAD chroma.y, luminance.%c, temp.x, chroma.y;\n", chroma);
6616     shader_addline(buffer, "SLT temp.x, texcrd2.x, coef.y;\n");
6617     shader_addline(buffer, "MAD chroma.x, luminance.%c, temp.x, chroma.x;\n", chroma);
6618
6619     /* TODO: If filtering is enabled, sample a 2nd pair of pixels left or right of
6620      * the current one and lerp the two U and V values
6621      */
6622
6623     /* This gives the correctly filtered luminance value */
6624     shader_addline(buffer, "TEX luminance, fragment.texcoord[0], texture[0], %s;\n", tex);
6625
6626     return TRUE;
6627 }
6628
6629 static BOOL gen_yv12_read(struct wined3d_shader_buffer *buffer, GLenum textype, char *luminance)
6630 {
6631     const char *tex;
6632
6633     switch(textype) {
6634         case GL_TEXTURE_2D:             tex = "2D";     break;
6635         case GL_TEXTURE_RECTANGLE_ARB:  tex = "RECT";   break;
6636         default:
6637             FIXME("Implement yv12 correction for non-2d, non-rect textures\n");
6638             return FALSE;
6639     }
6640
6641     /* YV12 surfaces contain a WxH sized luminance plane, followed by a (W/2)x(H/2)
6642      * V and a (W/2)x(H/2) U plane, each with 8 bit per pixel. So the effective
6643      * bitdepth is 12 bits per pixel. Since the U and V planes have only half the
6644      * pitch of the luminance plane, the packing into the gl texture is a bit
6645      * unfortunate. If the whole texture is interpreted as luminance data it looks
6646      * approximately like this:
6647      *
6648      *        +----------------------------------+----
6649      *        |                                  |
6650      *        |                                  |
6651      *        |                                  |
6652      *        |                                  |
6653      *        |                                  |   2
6654      *        |            LUMINANCE             |   -
6655      *        |                                  |   3
6656      *        |                                  |
6657      *        |                                  |
6658      *        |                                  |
6659      *        |                                  |
6660      *        +----------------+-----------------+----
6661      *        |                |                 |
6662      *        |  U even rows   |  U odd rows     |
6663      *        |                |                 |   1
6664      *        +----------------+------------------   -
6665      *        |                |                 |   3
6666      *        |  V even rows   |  V odd rows     |
6667      *        |                |                 |
6668      *        +----------------+-----------------+----
6669      *        |                |                 |
6670      *        |     0.5        |       0.5       |
6671      *
6672      * So it appears as if there are 4 chroma images, but in fact the odd rows
6673      * in the chroma images are in the same row as the even ones. So its is
6674      * kinda tricky to read
6675      *
6676      * When reading from rectangle textures, keep in mind that the input y coordinates
6677      * go from 0 to d3d_height, whereas the opengl texture height is 1.5 * d3d_height
6678      */
6679     shader_addline(buffer, "PARAM yv12_coef = {%f, %f, %f, %f};\n",
6680             2.0f / 3.0f, 1.0f / 6.0f, (2.0f / 3.0f) + (1.0f / 6.0f), 1.0f / 3.0f);
6681
6682     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6683     /* the chroma planes have only half the width */
6684     shader_addline(buffer, "MUL texcrd.x, texcrd.x, coef.y;\n");
6685
6686     /* The first value is between 2/3 and 5/6th of the texture's height, so scale+bias
6687      * the coordinate. Also read the right side of the image when reading odd lines
6688      *
6689      * Don't forget to clamp the y values in into the range, otherwise we'll get filtering
6690      * bleeding
6691      */
6692     if(textype == GL_TEXTURE_2D) {
6693
6694         shader_addline(buffer, "RCP chroma.w, size.y;\n");
6695
6696         shader_addline(buffer, "MUL texcrd2.y, texcrd.y, size.y;\n");
6697
6698         shader_addline(buffer, "FLR texcrd2.y, texcrd2.y;\n");
6699         shader_addline(buffer, "MAD texcrd.y, texcrd.y, yv12_coef.y, yv12_coef.x;\n");
6700
6701         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6702         shader_addline(buffer, "ADD texcrd2.x, texcrd2.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6703         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6704         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6705         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6706
6707         /* clamp, keep the half pixel origin in mind */
6708         shader_addline(buffer, "MAD temp.y, coef.y, chroma.w, yv12_coef.x;\n");
6709         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6710         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.z;\n");
6711         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6712     } else {
6713         /* Read from [size - size+size/4] */
6714         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6715         shader_addline(buffer, "MAD texcrd.y, texcrd.y, coef.w, size.y;\n");
6716
6717         /* Read odd lines from the right side(add size * 0.5 to the x coordinate */
6718         shader_addline(buffer, "ADD texcrd2.x, texcrd.y, yv12_coef.y;\n"); /* To avoid 0.5 == 0.5 comparisons */
6719         shader_addline(buffer, "FRC texcrd2.x, texcrd2.x;\n");
6720         shader_addline(buffer, "SGE texcrd2.x, texcrd2.x, coef.y;\n");
6721         shader_addline(buffer, "MUL texcrd2.x, texcrd2.x, size.x;\n");
6722         shader_addline(buffer, "MAD texcrd.x, texcrd2.x, coef.y, texcrd.x;\n");
6723
6724         /* Make sure to read exactly from the pixel center */
6725         shader_addline(buffer, "FLR texcrd.y, texcrd.y;\n");
6726         shader_addline(buffer, "ADD texcrd.y, texcrd.y, coef.y;\n");
6727
6728         /* Clamp */
6729         shader_addline(buffer, "MAD temp.y, size.y, coef.w, size.y;\n");
6730         shader_addline(buffer, "ADD temp.y, temp.y, -coef.y;\n");
6731         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6732         shader_addline(buffer, "ADD temp.y, size.y, -coef.y;\n");
6733         shader_addline(buffer, "MAX texcrd.y, temp.y, texcrd.y;\n");
6734     }
6735     /* Read the texture, put the result into the output register */
6736     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6737     shader_addline(buffer, "MOV chroma.x, temp.w;\n");
6738
6739     /* The other chroma value is 1/6th of the texture lower, from 5/6th to 6/6th
6740      * No need to clamp because we're just reusing the already clamped value from above
6741      */
6742     if(textype == GL_TEXTURE_2D) {
6743         shader_addline(buffer, "ADD texcrd.y, texcrd.y, yv12_coef.y;\n");
6744     } else {
6745         shader_addline(buffer, "MAD texcrd.y, size.y, coef.w, texcrd.y;\n");
6746     }
6747     shader_addline(buffer, "TEX temp, texcrd, texture[0], %s;\n", tex);
6748     shader_addline(buffer, "MOV chroma.y, temp.w;\n");
6749
6750     /* Sample the luminance value. It is in the top 2/3rd of the texture, so scale the y coordinate.
6751      * Clamp the y coordinate to prevent the chroma values from bleeding into the sampled luminance
6752      * values due to filtering
6753      */
6754     shader_addline(buffer, "MOV texcrd, fragment.texcoord[0];\n");
6755     if(textype == GL_TEXTURE_2D) {
6756         /* Multiply the y coordinate by 2/3 and clamp it */
6757         shader_addline(buffer, "MUL texcrd.y, texcrd.y, yv12_coef.x;\n");
6758         shader_addline(buffer, "MAD temp.y, -coef.y, chroma.w, yv12_coef.x;\n");
6759         shader_addline(buffer, "MIN texcrd.y, temp.y, texcrd.y;\n");
6760         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6761     } else {
6762         /* Reading from texture_rectangles is pretty straightforward, just use the unmodified
6763          * texture coordinate. It is still a good idea to clamp it though, since the opengl texture
6764          * is bigger
6765          */
6766         shader_addline(buffer, "ADD temp.x, size.y, -coef.y;\n");
6767         shader_addline(buffer, "MIN texcrd.y, texcrd.y, size.x;\n");
6768         shader_addline(buffer, "TEX luminance, texcrd, texture[0], %s;\n", tex);
6769     }
6770     *luminance = 'a';
6771
6772     return TRUE;
6773 }
6774
6775 static GLuint gen_p8_shader(struct arbfp_blit_priv *priv,
6776         const struct wined3d_gl_info *gl_info, GLenum textype)
6777 {
6778     GLenum shader;
6779     struct wined3d_shader_buffer buffer;
6780     GLint pos;
6781
6782     /* Shader header */
6783     if (!shader_buffer_init(&buffer))
6784     {
6785         ERR("Failed to initialize shader buffer.\n");
6786         return 0;
6787     }
6788
6789     ENTER_GL();
6790     GL_EXTCALL(glGenProgramsARB(1, &shader));
6791     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6792     LEAVE_GL();
6793     if(!shader) {
6794         shader_buffer_free(&buffer);
6795         return 0;
6796     }
6797
6798     shader_addline(&buffer, "!!ARBfp1.0\n");
6799     shader_addline(&buffer, "TEMP index;\n");
6800
6801     /* { 255/256, 0.5/255*255/256, 0, 0 } */
6802     shader_addline(&buffer, "PARAM constants = { 0.996, 0.00195, 0, 0 };\n");
6803
6804     /* The alpha-component contains the palette index */
6805     if(textype == GL_TEXTURE_RECTANGLE_ARB)
6806         shader_addline(&buffer, "TXP index, fragment.texcoord[0], texture[0], RECT;\n");
6807     else
6808         shader_addline(&buffer, "TEX index, fragment.texcoord[0], texture[0], 2D;\n");
6809
6810     /* Scale the index by 255/256 and add a bias of '0.5' in order to sample in the middle */
6811     shader_addline(&buffer, "MAD index.a, index.a, constants.x, constants.y;\n");
6812
6813     /* Use the alpha-component as an index in the palette to get the final color */
6814     shader_addline(&buffer, "TEX result.color, index.a, texture[1], 1D;\n");
6815     shader_addline(&buffer, "END\n");
6816
6817     ENTER_GL();
6818     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6819             strlen(buffer.buffer), buffer.buffer));
6820     checkGLcall("glProgramStringARB()");
6821
6822     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6823     if (pos != -1)
6824     {
6825         FIXME("Fragment program error at position %d: %s\n\n", pos,
6826               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6827         shader_arb_dump_program_source(buffer.buffer);
6828     }
6829
6830     if (textype == GL_TEXTURE_RECTANGLE_ARB)
6831         priv->p8_rect_shader = shader;
6832     else
6833         priv->p8_2d_shader = shader;
6834
6835     shader_buffer_free(&buffer);
6836     LEAVE_GL();
6837
6838     return shader;
6839 }
6840
6841 /* Context activation is done by the caller. */
6842 static void upload_palette(IWineD3DSurfaceImpl *surface)
6843 {
6844     BYTE table[256][4];
6845     IWineD3DDeviceImpl *device = surface->resource.device;
6846     const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
6847     struct arbfp_blit_priv *priv = device->blit_priv;
6848     BOOL colorkey = (surface->CKeyFlags & WINEDDSD_CKSRCBLT) ? TRUE : FALSE;
6849
6850     d3dfmt_p8_init_palette(surface, table, colorkey);
6851
6852     ENTER_GL();
6853     if (!priv->palette_texture)
6854         glGenTextures(1, &priv->palette_texture);
6855
6856     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE1));
6857     glBindTexture(GL_TEXTURE_1D, priv->palette_texture);
6858
6859     glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
6860
6861     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
6862     /* Make sure we have discrete color levels. */
6863     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
6864     glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
6865     /* Upload the palette */
6866     /* TODO: avoid unneeed uploads in the future by adding some SFLAG_PALETTE_DIRTY mechanism */
6867     glTexImage1D(GL_TEXTURE_1D, 0, GL_RGBA, 256, 0, GL_RGBA, GL_UNSIGNED_BYTE, table);
6868
6869     /* Switch back to unit 0 in which the 2D texture will be stored. */
6870     GL_EXTCALL(glActiveTextureARB(GL_TEXTURE0));
6871     LEAVE_GL();
6872 }
6873
6874 /* Context activation is done by the caller. */
6875 static GLuint gen_yuv_shader(struct arbfp_blit_priv *priv, const struct wined3d_gl_info *gl_info,
6876         enum complex_fixup yuv_fixup, GLenum textype)
6877 {
6878     GLenum shader;
6879     struct wined3d_shader_buffer buffer;
6880     char luminance_component;
6881     GLint pos;
6882
6883     /* Shader header */
6884     if (!shader_buffer_init(&buffer))
6885     {
6886         ERR("Failed to initialize shader buffer.\n");
6887         return 0;
6888     }
6889
6890     ENTER_GL();
6891     GL_EXTCALL(glGenProgramsARB(1, &shader));
6892     checkGLcall("GL_EXTCALL(glGenProgramsARB(1, &shader))");
6893     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
6894     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
6895     LEAVE_GL();
6896     if(!shader) {
6897         shader_buffer_free(&buffer);
6898         return 0;
6899     }
6900
6901     /* The YUY2 and UYVY formats contain two pixels packed into a 32 bit macropixel,
6902      * giving effectively 16 bit per pixel. The color consists of a luminance(Y) and
6903      * two chroma(U and V) values. Each macropixel has two luminance values, one for
6904      * each single pixel it contains, and one U and one V value shared between both
6905      * pixels.
6906      *
6907      * The data is loaded into an A8L8 texture. With YUY2, the luminance component
6908      * contains the luminance and alpha the chroma. With UYVY it is vice versa. Thus
6909      * take the format into account when generating the read swizzles
6910      *
6911      * Reading the Y value is straightforward - just sample the texture. The hardware
6912      * takes care of filtering in the horizontal and vertical direction.
6913      *
6914      * Reading the U and V values is harder. We have to avoid filtering horizontally,
6915      * because that would mix the U and V values of one pixel or two adjacent pixels.
6916      * Thus floor the texture coordinate and add 0.5 to get an unfiltered read,
6917      * regardless of the filtering setting. Vertical filtering works automatically
6918      * though - the U and V values of two rows are mixed nicely.
6919      *
6920      * Appart of avoiding filtering issues, the code has to know which value it just
6921      * read, and where it can find the other one. To determine this, it checks if
6922      * it sampled an even or odd pixel, and shifts the 2nd read accordingly.
6923      *
6924      * Handling horizontal filtering of U and V values requires reading a 2nd pair
6925      * of pixels, extracting U and V and mixing them. This is not implemented yet.
6926      *
6927      * An alternative implementation idea is to load the texture as A8R8G8B8 texture,
6928      * with width / 2. This way one read gives all 3 values, finding U and V is easy
6929      * in an unfiltered situation. Finding the luminance on the other hand requires
6930      * finding out if it is an odd or even pixel. The real drawback of this approach
6931      * is filtering. This would have to be emulated completely in the shader, reading
6932      * up two 2 packed pixels in up to 2 rows and interpolating both horizontally and
6933      * vertically. Beyond that it would require adjustments to the texture handling
6934      * code to deal with the width scaling
6935      */
6936     shader_addline(&buffer, "!!ARBfp1.0\n");
6937     shader_addline(&buffer, "TEMP luminance;\n");
6938     shader_addline(&buffer, "TEMP temp;\n");
6939     shader_addline(&buffer, "TEMP chroma;\n");
6940     shader_addline(&buffer, "TEMP texcrd;\n");
6941     shader_addline(&buffer, "TEMP texcrd2;\n");
6942     shader_addline(&buffer, "PARAM coef = {1.0, 0.5, 2.0, 0.25};\n");
6943     shader_addline(&buffer, "PARAM yuv_coef = {1.403, 0.344, 0.714, 1.770};\n");
6944     shader_addline(&buffer, "PARAM size = program.local[0];\n");
6945
6946     switch (yuv_fixup)
6947     {
6948         case COMPLEX_FIXUP_UYVY:
6949         case COMPLEX_FIXUP_YUY2:
6950             if (!gen_planar_yuv_read(&buffer, yuv_fixup, textype, &luminance_component))
6951             {
6952                 shader_buffer_free(&buffer);
6953                 return 0;
6954             }
6955             break;
6956
6957         case COMPLEX_FIXUP_YV12:
6958             if (!gen_yv12_read(&buffer, textype, &luminance_component))
6959             {
6960                 shader_buffer_free(&buffer);
6961                 return 0;
6962             }
6963             break;
6964
6965         default:
6966             FIXME("Unsupported YUV fixup %#x\n", yuv_fixup);
6967             shader_buffer_free(&buffer);
6968             return 0;
6969     }
6970
6971     /* Calculate the final result. Formula is taken from
6972      * http://www.fourcc.org/fccyvrgb.php. Note that the chroma
6973      * ranges from -0.5 to 0.5
6974      */
6975     shader_addline(&buffer, "SUB chroma.xy, chroma, coef.y;\n");
6976
6977     shader_addline(&buffer, "MAD result.color.x, chroma.x, yuv_coef.x, luminance.%c;\n", luminance_component);
6978     shader_addline(&buffer, "MAD temp.x, -chroma.y, yuv_coef.y, luminance.%c;\n", luminance_component);
6979     shader_addline(&buffer, "MAD result.color.y, -chroma.x, yuv_coef.z, temp.x;\n");
6980     shader_addline(&buffer, "MAD result.color.z, chroma.y, yuv_coef.w, luminance.%c;\n", luminance_component);
6981     shader_addline(&buffer, "END\n");
6982
6983     ENTER_GL();
6984     GL_EXTCALL(glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
6985             strlen(buffer.buffer), buffer.buffer));
6986     checkGLcall("glProgramStringARB()");
6987
6988     glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
6989     if (pos != -1)
6990     {
6991         FIXME("Fragment program error at position %d: %s\n\n", pos,
6992               debugstr_a((const char *)glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
6993         shader_arb_dump_program_source(buffer.buffer);
6994     }
6995     else
6996     {
6997         GLint native;
6998
6999         GL_EXTCALL(glGetProgramivARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
7000         checkGLcall("glGetProgramivARB()");
7001         if (!native) WARN("Program exceeds native resource limits.\n");
7002     }
7003
7004     shader_buffer_free(&buffer);
7005     LEAVE_GL();
7006
7007     switch (yuv_fixup)
7008     {
7009         case COMPLEX_FIXUP_YUY2:
7010             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yuy2_rect_shader = shader;
7011             else priv->yuy2_2d_shader = shader;
7012             break;
7013
7014         case COMPLEX_FIXUP_UYVY:
7015             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->uyvy_rect_shader = shader;
7016             else priv->uyvy_2d_shader = shader;
7017             break;
7018
7019         case COMPLEX_FIXUP_YV12:
7020             if (textype == GL_TEXTURE_RECTANGLE_ARB) priv->yv12_rect_shader = shader;
7021             else priv->yv12_2d_shader = shader;
7022             break;
7023         default:
7024             ERR("Unsupported complex fixup: %d\n", yuv_fixup);
7025     }
7026
7027     return shader;
7028 }
7029
7030 /* Context activation is done by the caller. */
7031 static HRESULT arbfp_blit_set(void *blit_priv, const struct wined3d_gl_info *gl_info, IWineD3DSurfaceImpl *surface)
7032 {
7033     GLenum shader;
7034     float size[4] = {(float) surface->pow2Width, (float) surface->pow2Height, 1.0f, 1.0f};
7035     struct arbfp_blit_priv *priv = blit_priv;
7036     enum complex_fixup fixup;
7037     GLenum textype = surface->texture_target;
7038
7039     if (!is_complex_fixup(surface->resource.format->color_fixup))
7040     {
7041         TRACE("Fixup:\n");
7042         dump_color_fixup_desc(surface->resource.format->color_fixup);
7043         /* Don't bother setting up a shader for unconverted formats */
7044         ENTER_GL();
7045         glEnable(textype);
7046         checkGLcall("glEnable(textype)");
7047         LEAVE_GL();
7048         return WINED3D_OK;
7049     }
7050
7051     fixup = get_complex_fixup(surface->resource.format->color_fixup);
7052
7053     switch(fixup)
7054     {
7055         case COMPLEX_FIXUP_YUY2:
7056             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yuy2_rect_shader : priv->yuy2_2d_shader;
7057             break;
7058
7059         case COMPLEX_FIXUP_UYVY:
7060             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->uyvy_rect_shader : priv->uyvy_2d_shader;
7061             break;
7062
7063         case COMPLEX_FIXUP_YV12:
7064             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->yv12_rect_shader : priv->yv12_2d_shader;
7065             break;
7066
7067         case COMPLEX_FIXUP_P8:
7068             shader = textype == GL_TEXTURE_RECTANGLE_ARB ? priv->p8_rect_shader : priv->p8_2d_shader;
7069             if (!shader) shader = gen_p8_shader(priv, gl_info, textype);
7070
7071             upload_palette(surface);
7072             break;
7073
7074         default:
7075             FIXME("Unsupported complex fixup %#x, not setting a shader\n", fixup);
7076             ENTER_GL();
7077             glEnable(textype);
7078             checkGLcall("glEnable(textype)");
7079             LEAVE_GL();
7080             return E_NOTIMPL;
7081     }
7082
7083     if (!shader) shader = gen_yuv_shader(priv, gl_info, fixup, textype);
7084
7085     ENTER_GL();
7086     glEnable(GL_FRAGMENT_PROGRAM_ARB);
7087     checkGLcall("glEnable(GL_FRAGMENT_PROGRAM_ARB)");
7088     GL_EXTCALL(glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader));
7089     checkGLcall("glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader)");
7090     GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, 0, size));
7091     checkGLcall("glProgramLocalParameter4fvARB");
7092     LEAVE_GL();
7093
7094     return WINED3D_OK;
7095 }
7096
7097 /* Context activation is done by the caller. */
7098 static void arbfp_blit_unset(const struct wined3d_gl_info *gl_info)
7099 {
7100     ENTER_GL();
7101     glDisable(GL_FRAGMENT_PROGRAM_ARB);
7102     checkGLcall("glDisable(GL_FRAGMENT_PROGRAM_ARB)");
7103     glDisable(GL_TEXTURE_2D);
7104     checkGLcall("glDisable(GL_TEXTURE_2D)");
7105     if (gl_info->supported[ARB_TEXTURE_CUBE_MAP])
7106     {
7107         glDisable(GL_TEXTURE_CUBE_MAP_ARB);
7108         checkGLcall("glDisable(GL_TEXTURE_CUBE_MAP_ARB)");
7109     }
7110     if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
7111     {
7112         glDisable(GL_TEXTURE_RECTANGLE_ARB);
7113         checkGLcall("glDisable(GL_TEXTURE_RECTANGLE_ARB)");
7114     }
7115     LEAVE_GL();
7116 }
7117
7118 static BOOL arbfp_blit_supported(const struct wined3d_gl_info *gl_info, enum blit_operation blit_op,
7119         const RECT *src_rect, DWORD src_usage, WINED3DPOOL src_pool, const struct wined3d_format *src_format,
7120         const RECT *dst_rect, DWORD dst_usage, WINED3DPOOL dst_pool, const struct wined3d_format *dst_format)
7121 {
7122     enum complex_fixup src_fixup;
7123
7124     if (blit_op != BLIT_OP_BLIT)
7125     {
7126         TRACE("Unsupported blit_op=%d\n", blit_op);
7127         return FALSE;
7128     }
7129
7130     src_fixup = get_complex_fixup(src_format->color_fixup);
7131     if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
7132     {
7133         TRACE("Checking support for fixup:\n");
7134         dump_color_fixup_desc(src_format->color_fixup);
7135     }
7136
7137     if (!is_identity_fixup(dst_format->color_fixup))
7138     {
7139         TRACE("Destination fixups are not supported\n");
7140         return FALSE;
7141     }
7142
7143     if (is_identity_fixup(src_format->color_fixup))
7144     {
7145         TRACE("[OK]\n");
7146         return TRUE;
7147     }
7148
7149      /* We only support YUV conversions. */
7150     if (!is_complex_fixup(src_format->color_fixup))
7151     {
7152         TRACE("[FAILED]\n");
7153         return FALSE;
7154     }
7155
7156     switch(src_fixup)
7157     {
7158         case COMPLEX_FIXUP_YUY2:
7159         case COMPLEX_FIXUP_UYVY:
7160         case COMPLEX_FIXUP_YV12:
7161         case COMPLEX_FIXUP_P8:
7162             TRACE("[OK]\n");
7163             return TRUE;
7164
7165         default:
7166             FIXME("Unsupported YUV fixup %#x\n", src_fixup);
7167             TRACE("[FAILED]\n");
7168             return FALSE;
7169     }
7170 }
7171
7172 HRESULT arbfp_blit_surface(IWineD3DDeviceImpl *device, IWineD3DSurfaceImpl *src_surface, const RECT *src_rect,
7173                            IWineD3DSurfaceImpl *dst_surface, const RECT *dst_rect_in, enum blit_operation blit_op,
7174                            DWORD Filter)
7175 {
7176     IWineD3DSwapChainImpl *dst_swapchain;
7177     struct wined3d_context *context;
7178     RECT dst_rect = *dst_rect_in;
7179
7180     /* Now load the surface */
7181     surface_internal_preload(src_surface, SRGB_RGB);
7182
7183     /* Activate the destination context, set it up for blitting */
7184     context = context_acquire(device, dst_surface);
7185     context_apply_blit_state(context, device);
7186
7187     if (!surface_is_offscreen(dst_surface))
7188         surface_translate_drawable_coords(dst_surface, context->win_handle, &dst_rect);
7189
7190     arbfp_blit_set(device->blit_priv, context->gl_info, src_surface);
7191
7192     ENTER_GL();
7193
7194     /* Draw a textured quad */
7195     draw_textured_quad(src_surface, src_rect, &dst_rect, Filter);
7196
7197     LEAVE_GL();
7198
7199     /* Leave the opengl state valid for blitting */
7200     arbfp_blit_unset(context->gl_info);
7201
7202     dst_swapchain = dst_surface->container.type == WINED3D_CONTAINER_SWAPCHAIN
7203             ? dst_surface->container.u.swapchain : NULL;
7204     if (wined3d_settings.strict_draw_ordering || (dst_swapchain
7205             && (dst_surface == dst_swapchain->front_buffer
7206             || dst_swapchain->num_contexts > 1)))
7207         wglFlush(); /* Flush to ensure ordering across contexts. */
7208
7209     context_release(context);
7210
7211     surface_modify_location(dst_surface, SFLAG_INDRAWABLE, TRUE);
7212     return WINED3D_OK;
7213 }
7214
7215 /* Do not call while under the GL lock. */
7216 static HRESULT arbfp_blit_color_fill(IWineD3DDeviceImpl *device, IWineD3DSurfaceImpl *dst_surface,
7217         const RECT *dst_rect, const WINED3DCOLORVALUE *color)
7218 {
7219     FIXME("Color filling not implemented by arbfp_blit\n");
7220     return WINED3DERR_INVALIDCALL;
7221 }
7222
7223 const struct blit_shader arbfp_blit = {
7224     arbfp_blit_alloc,
7225     arbfp_blit_free,
7226     arbfp_blit_set,
7227     arbfp_blit_unset,
7228     arbfp_blit_supported,
7229     arbfp_blit_color_fill
7230 };