VirtualBox

source: vbox/trunk/src/VBox/Devices/Graphics/shaderlib/glsl_shader.c@ 66058

Last change on this file since 66058 was 62948, checked in by vboxsync, 8 years ago

VBoxSVGA3D: warnings

  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 218.1 KB
Line 
1/*
2 * GLSL pixel and vertex shader implementation
3 *
4 * Copyright 2006 Jason Green
5 * Copyright 2006-2007 Henri Verbeet
6 * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
7 * Copyright 2009 Henri Verbeet for CodeWeavers
8 *
9 * This library is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
13 *
14 * This library is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with this library; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
22 */
23
24/*
25 * Oracle LGPL Disclaimer: For the avoidance of doubt, except that if any license choice
26 * other than GPL or LGPL is available it will apply instead, Oracle elects to use only
27 * the Lesser General Public License version 2.1 (LGPLv2) at this time for any software where
28 * a choice of LGPL license versions is made available with the language indicating
29 * that LGPLv2 or any later version may be used, or where a choice of which version
30 * of the LGPL is applied is otherwise unspecified.
31 */
32
33/*
34 * D3D shader asm has swizzles on source parameters, and write masks for
35 * destination parameters. GLSL uses swizzles for both. The result of this is
36 * that for example "mov dst.xw, src.zyxw" becomes "dst.xw = src.zw" in GLSL.
37 * Ie, to generate a proper GLSL source swizzle, we need to take the D3D write
38 * mask for the destination parameter into account.
39 */
40
41#include "config.h"
42#include "wine/port.h"
43#include <limits.h>
44#include <stdio.h>
45#include "wined3d_private.h"
46
47WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
48WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
49WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
50WINE_DECLARE_DEBUG_CHANNEL(d3d);
51
52#ifdef VBOX_WITH_VMSVGA
53#define LOG_GROUP LOG_GROUP_DEV_VMSVGA
54#include <VBox/log.h>
55#undef WDLOG
56#define WDLOG(_m) Log(_m)
57#undef CONST
58#define CONST const
59#endif
60
61#define GLINFO_LOCATION (*gl_info)
62
63#define WINED3D_GLSL_SAMPLE_PROJECTED 0x1
64#define WINED3D_GLSL_SAMPLE_RECT 0x2
65#define WINED3D_GLSL_SAMPLE_LOD 0x4
66#define WINED3D_GLSL_SAMPLE_GRAD 0x8
67
68typedef struct {
69 char reg_name[150];
70 char mask_str[6];
71} glsl_dst_param_t;
72
73typedef struct {
74 char reg_name[150];
75 char param_str[200];
76} glsl_src_param_t;
77
78typedef struct {
79 const char *name;
80 DWORD coord_mask;
81} glsl_sample_function_t;
82
83enum heap_node_op
84{
85 HEAP_NODE_TRAVERSE_LEFT,
86 HEAP_NODE_TRAVERSE_RIGHT,
87 HEAP_NODE_POP,
88};
89
90struct constant_entry
91{
92 unsigned int idx;
93 unsigned int version;
94};
95
96struct constant_heap
97{
98 struct constant_entry *entries;
99 unsigned int *positions;
100 unsigned int size;
101};
102
103/* GLSL shader private data */
104struct shader_glsl_priv {
105 struct wined3d_shader_buffer shader_buffer;
106 struct wine_rb_tree program_lookup;
107 struct glsl_shader_prog_link *glsl_program;
108 struct constant_heap vconst_heap;
109 struct constant_heap pconst_heap;
110 unsigned char *stack;
111 GLhandleARB depth_blt_program[tex_type_count];
112 UINT next_constant_version;
113};
114
115/* Struct to maintain data about a linked GLSL program */
116struct glsl_shader_prog_link {
117 struct wine_rb_entry program_lookup_entry;
118 struct list vshader_entry;
119 struct list pshader_entry;
120 GLhandleARB programId;
121 GLint *vuniformF_locations;
122 GLint *puniformF_locations;
123 GLint vuniformI_locations[MAX_CONST_I];
124 GLint puniformI_locations[MAX_CONST_I];
125 GLint posFixup_location;
126 GLint np2Fixup_location;
127 GLint bumpenvmat_location[MAX_TEXTURES];
128 GLint luminancescale_location[MAX_TEXTURES];
129 GLint luminanceoffset_location[MAX_TEXTURES];
130 GLint ycorrection_location;
131 GLenum vertex_color_clamp;
132 IWineD3DVertexShader *vshader;
133 IWineD3DPixelShader *pshader;
134 struct vs_compile_args vs_args;
135 struct ps_compile_args ps_args;
136 UINT constant_version;
137 const struct wined3d_context *context;
138 UINT inp2Fixup_info;
139};
140
141#ifdef VBOX_WITH_VMSVGA
142# define WINEFIXUPINFO_NOINDEX (~0U)
143#else
144#define WINEFIXUPINFO_NOINDEX (~0UL)
145#endif
146#define WINEFIXUPINFO_GET(_p) get_fixup_info((const IWineD3DPixelShaderImpl*)(_p)->pshader, (_p)->inp2Fixup_info)
147#define WINEFIXUPINFO_ISVALID(_p) ((_p)->inp2Fixup_info != WINEFIXUPINFO_NOINDEX)
148#ifdef VBOX_WITH_VMSVGA
149# define WINEFIXUPINFO_INIT(_p) do { (_p)->inp2Fixup_info = WINEFIXUPINFO_NOINDEX; } while (0)
150#else
151#define WINEFIXUPINFO_INIT(_p) ((_p)->inp2Fixup_info == WINEFIXUPINFO_NOINDEX)
152#endif
153
154typedef struct {
155 IWineD3DVertexShader *vshader;
156 IWineD3DPixelShader *pshader;
157 struct ps_compile_args ps_args;
158 struct vs_compile_args vs_args;
159 const struct wined3d_context *context;
160} glsl_program_key_t;
161
162struct shader_glsl_ctx_priv {
163 const struct vs_compile_args *cur_vs_args;
164 const struct ps_compile_args *cur_ps_args;
165 struct ps_np2fixup_info *cur_np2fixup_info;
166};
167
168struct glsl_ps_compiled_shader
169{
170 struct ps_compile_args args;
171 struct ps_np2fixup_info np2fixup;
172 GLhandleARB prgId;
173 const struct wined3d_context *context;
174};
175
176struct glsl_pshader_private
177{
178 struct glsl_ps_compiled_shader *gl_shaders;
179 UINT num_gl_shaders, shader_array_size;
180};
181
182struct glsl_vs_compiled_shader
183{
184 struct vs_compile_args args;
185 GLhandleARB prgId;
186 const struct wined3d_context *context;
187};
188
189struct glsl_vshader_private
190{
191 struct glsl_vs_compiled_shader *gl_shaders;
192 UINT num_gl_shaders, shader_array_size;
193};
194
195static const char *debug_gl_shader_type(GLenum type)
196{
197 switch (type)
198 {
199#define WINED3D_TO_STR(u) case u: return #u
200 WINED3D_TO_STR(GL_VERTEX_SHADER_ARB);
201 WINED3D_TO_STR(GL_GEOMETRY_SHADER_ARB);
202 WINED3D_TO_STR(GL_FRAGMENT_SHADER_ARB);
203#undef WINED3D_TO_STR
204 default:
205 return wine_dbg_sprintf("UNKNOWN(%#x)", type);
206 }
207}
208
209/* Extract a line from the info log.
210 * Note that this modifies the source string. */
211static char *get_info_log_line(char **ptr, int *pcbStr)
212{
213 char *p, *q;
214 const int cbStr = *pcbStr;
215
216 if (!cbStr)
217 {
218 /* zero-length string */
219 return NULL;
220 }
221
222 if ((*ptr)[cbStr-1] != '\0')
223 {
224 ERR("string should be null-rerminated, forcing it!");
225 (*ptr)[cbStr-1] = '\0';
226 }
227 p = *ptr;
228 if (!*p)
229 {
230 *pcbStr = 0;
231 return NULL;
232 }
233
234 if (!(q = strstr(p, "\n")))
235 {
236 /* the string contains a single line! */
237 *ptr += strlen(p);
238 *pcbStr = 0;
239 return p;
240 }
241
242 *q = '\0';
243 *pcbStr = cbStr - (((uintptr_t)q) - ((uintptr_t)p)) - 1;
244 Assert((*pcbStr) >= 0);
245 Assert((*pcbStr) < cbStr);
246 *ptr = q + 1;
247
248 return p;
249}
250
251/** Prints the GLSL info log which will contain error messages if they exist */
252/* GL locking is done by the caller */
253static void print_glsl_info_log(const struct wined3d_gl_info *gl_info, GLhandleARB obj)
254{
255 int infologLength = 0;
256 char *infoLog;
257 unsigned int i;
258 BOOL is_spam;
259
260 static const char * const spam[] =
261 {
262 "Vertex shader was successfully compiled to run on hardware.\n", /* fglrx */
263 "Fragment shader was successfully compiled to run on hardware.\n", /* fglrx, with \n */
264 "Fragment shader was successfully compiled to run on hardware.", /* fglrx, no \n */
265 "Fragment shader(s) linked, vertex shader(s) linked. \n ", /* fglrx, with \n */
266 "Fragment shader(s) linked, vertex shader(s) linked.", /* fglrx, no \n */
267 "Vertex shader(s) linked, no fragment shader(s) defined. \n ", /* fglrx, with \n */
268 "Vertex shader(s) linked, no fragment shader(s) defined.", /* fglrx, no \n */
269 "Fragment shader(s) linked, no vertex shader(s) defined. \n ", /* fglrx, with \n */
270 "Fragment shader(s) linked, no vertex shader(s) defined.", /* fglrx, no \n */
271 };
272
273#ifndef VBOXWINEDBG_SHADERS
274 if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
275#endif
276
277 GL_EXTCALL(glGetObjectParameterivARB(obj,
278 GL_OBJECT_INFO_LOG_LENGTH_ARB,
279 &infologLength));
280
281 /* A size of 1 is just a null-terminated string, so the log should be bigger than
282 * that if there are errors. */
283 if (infologLength > 1)
284 {
285 char *ptr, *line;
286 int cbPtr;
287
288 /* Fglrx doesn't terminate the string properly, but it tells us the proper length.
289 * So use HEAP_ZERO_MEMORY to avoid uninitialized bytes
290 */
291 infoLog = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, infologLength);
292 GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
293 is_spam = FALSE;
294
295 for(i = 0; i < sizeof(spam) / sizeof(spam[0]); i++) {
296 if(strcmp(infoLog, spam[i]) == 0) {
297 is_spam = TRUE;
298 break;
299 }
300 }
301
302 ptr = infoLog;
303 cbPtr = infologLength;
304 if (is_spam)
305 {
306 WDLOG(("Spam received from GLSL shader #%u:\n", obj));
307 while ((line = get_info_log_line(&ptr, &cbPtr))) WDLOG((" %s\n", line));
308 }
309 else
310 {
311 WDLOG(("Error received from GLSL shader #%u:\n", obj));
312 while ((line = get_info_log_line(&ptr, &cbPtr))) WDLOG((" %s\n", line));
313 }
314 HeapFree(GetProcessHeap(), 0, infoLog);
315 }
316}
317
318static void shader_glsl_dump_shader_source(const struct wined3d_gl_info *gl_info, GLhandleARB shader)
319{
320 char *ptr;
321 GLint tmp, source_size;
322 char *source = NULL;
323 int cbPtr;
324
325 GL_EXTCALL(glGetObjectParameterivARB(shader, GL_OBJECT_SHADER_SOURCE_LENGTH_ARB, &tmp));
326
327 source = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, tmp);
328 if (!source)
329 {
330 ERR("Failed to allocate %d bytes for shader source.\n", tmp);
331 return;
332 }
333
334 source_size = tmp;
335
336 WDLOG(("Object %u:\n", shader));
337 GL_EXTCALL(glGetObjectParameterivARB(shader, GL_OBJECT_SUBTYPE_ARB, &tmp));
338 WDLOG((" GL_OBJECT_SUBTYPE_ARB: %s.\n", debug_gl_shader_type(tmp)));
339 GL_EXTCALL(glGetObjectParameterivARB(shader, GL_OBJECT_COMPILE_STATUS_ARB, &tmp));
340 WDLOG((" GL_OBJECT_COMPILE_STATUS_ARB: %d.\n", tmp));
341 WDLOG(("\n"));
342
343 ptr = source;
344 cbPtr = source_size;
345 GL_EXTCALL(glGetShaderSourceARB(shader, source_size, NULL, source));
346#if 0
347 while ((line = get_info_log_line(&ptr, &cbPtr))) WDLOG((" %s\n", line));
348#else
349 WDLOG(("*****shader source***\n"));
350 WDLOG((" %s\n", source));
351 WDLOG(("\n*****END shader source***\n\n"));
352#endif
353 WDLOG(("\n"));
354}
355
356/* GL locking is done by the caller. */
357static void shader_glsl_dump_program_source(const struct wined3d_gl_info *gl_info, GLhandleARB program)
358{
359 GLint i, object_count;
360 GLhandleARB *objects;
361 char *source = NULL;
362
363 WDLOG(("\n***************************dumping program %d******************************\n", program));
364
365 GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_ATTACHED_OBJECTS_ARB, &object_count));
366 objects = HeapAlloc(GetProcessHeap(), 0, object_count * sizeof(*objects));
367 if (!objects)
368 {
369 ERR("Failed to allocate object array memory.\n");
370 return;
371 }
372
373 GL_EXTCALL(glGetAttachedObjectsARB(program, object_count, NULL, objects));
374 for (i = 0; i < object_count; ++i)
375 {
376 shader_glsl_dump_shader_source(gl_info, objects[i]);
377 }
378
379 HeapFree(GetProcessHeap(), 0, source);
380 HeapFree(GetProcessHeap(), 0, objects);
381
382 WDLOG(("\n***************************END dumping program %d******************************\n\n", program));
383}
384
385/* GL locking is done by the caller. */
386static void shader_glsl_validate_compile_link(const struct wined3d_gl_info *gl_info, GLhandleARB program, GLboolean fIsProgram)
387{
388 GLint tmp = -1;
389
390#ifndef VBOXWINEDBG_SHADERS
391 if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
392#endif
393
394 GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_TYPE_ARB, &tmp));
395 if (tmp == GL_PROGRAM_OBJECT_ARB)
396 {
397 if (!fIsProgram)
398 {
399 ERR("this is a program, but shader expected");
400 }
401 GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_LINK_STATUS_ARB, &tmp));
402 if (!tmp)
403 {
404 ERR("Program %p link status invalid.\n", (void *)(uintptr_t)program);
405#ifndef VBOXWINEDBG_SHADERS
406 shader_glsl_dump_program_source(gl_info, program);
407#endif
408 }
409#if defined(VBOX_WITH_VMSVGA) && defined(DEBUG)
410 shader_glsl_dump_program_source(gl_info, program);
411#endif
412 }
413 else if (tmp == GL_SHADER_OBJECT_ARB)
414 {
415 if (fIsProgram)
416 {
417 ERR("this is a shader, but program expected");
418 }
419
420 GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_COMPILE_STATUS_ARB, &tmp));
421 if (!tmp)
422 {
423 ERR("Shader %p compile status invalid.\n", (void *)(uintptr_t)program);
424 shader_glsl_dump_shader_source(gl_info, program);
425 }
426 }
427 else
428 {
429 ERR("unexpected oject type(%d)!", tmp);
430 }
431
432 print_glsl_info_log(gl_info, program);
433}
434
435/**
436 * Loads (pixel shader) samplers
437 */
438/* GL locking is done by the caller */
439static void shader_glsl_load_psamplers(const struct wined3d_gl_info *gl_info,
440 DWORD *tex_unit_map, GLhandleARB programId)
441{
442 GLint name_loc;
443 int i;
444 char sampler_name[20];
445
446 for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
447 snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
448 name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
449 if (name_loc != -1) {
450 DWORD mapped_unit = tex_unit_map[i];
451 if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.fragment_samplers)
452 {
453 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
454 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
455 checkGLcall("glUniform1iARB");
456 } else {
457 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
458 }
459 }
460 }
461}
462
463/* GL locking is done by the caller */
464static void shader_glsl_load_vsamplers(const struct wined3d_gl_info *gl_info,
465 DWORD *tex_unit_map, GLhandleARB programId)
466{
467 GLint name_loc;
468 char sampler_name[20];
469 int i;
470
471 for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
472 snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
473 name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
474 if (name_loc != -1) {
475 DWORD mapped_unit = tex_unit_map[MAX_FRAGMENT_SAMPLERS + i];
476 if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.combined_samplers)
477 {
478 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
479 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
480 checkGLcall("glUniform1iARB");
481 } else {
482 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
483 }
484 }
485 }
486}
487
488/* GL locking is done by the caller */
489static inline void walk_constant_heap(const struct wined3d_gl_info *gl_info, const float *constants,
490 const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
491{
492 int stack_idx = 0;
493 unsigned int heap_idx = 1;
494 unsigned int idx;
495
496 if (heap->entries[heap_idx].version <= version) return;
497
498 idx = heap->entries[heap_idx].idx;
499 if (constant_locations[idx] != -1) GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
500 stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
501
502 while (stack_idx >= 0)
503 {
504 /* Note that we fall through to the next case statement. */
505 switch(stack[stack_idx])
506 {
507 case HEAP_NODE_TRAVERSE_LEFT:
508 {
509 unsigned int left_idx = heap_idx << 1;
510 if (left_idx < heap->size && heap->entries[left_idx].version > version)
511 {
512 heap_idx = left_idx;
513 idx = heap->entries[heap_idx].idx;
514 if (constant_locations[idx] != -1)
515 GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
516
517 stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
518 stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
519 break;
520 }
521 }
522
523 case HEAP_NODE_TRAVERSE_RIGHT:
524 {
525 unsigned int right_idx = (heap_idx << 1) + 1;
526 if (right_idx < heap->size && heap->entries[right_idx].version > version)
527 {
528 heap_idx = right_idx;
529 idx = heap->entries[heap_idx].idx;
530 if (constant_locations[idx] != -1)
531 GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
532
533 stack[stack_idx++] = HEAP_NODE_POP;
534 stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
535 break;
536 }
537 }
538
539 case HEAP_NODE_POP:
540 {
541 heap_idx >>= 1;
542 --stack_idx;
543 break;
544 }
545 }
546 }
547 checkGLcall("walk_constant_heap()");
548}
549
550/* GL locking is done by the caller */
551static inline void apply_clamped_constant(const struct wined3d_gl_info *gl_info, GLint location, const GLfloat *data)
552{
553 GLfloat clamped_constant[4];
554
555 if (location == -1) return;
556
557 clamped_constant[0] = data[0] < -1.0f ? -1.0f : data[0] > 1.0f ? 1.0f : data[0];
558 clamped_constant[1] = data[1] < -1.0f ? -1.0f : data[1] > 1.0f ? 1.0f : data[1];
559 clamped_constant[2] = data[2] < -1.0f ? -1.0f : data[2] > 1.0f ? 1.0f : data[2];
560 clamped_constant[3] = data[3] < -1.0f ? -1.0f : data[3] > 1.0f ? 1.0f : data[3];
561
562 GL_EXTCALL(glUniform4fvARB(location, 1, clamped_constant));
563}
564
565/* GL locking is done by the caller */
566static inline void walk_constant_heap_clamped(const struct wined3d_gl_info *gl_info, const float *constants,
567 const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
568{
569 int stack_idx = 0;
570 unsigned int heap_idx = 1;
571 unsigned int idx;
572
573 if (heap->entries[heap_idx].version <= version) return;
574
575 idx = heap->entries[heap_idx].idx;
576 apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
577 stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
578
579 while (stack_idx >= 0)
580 {
581 /* Note that we fall through to the next case statement. */
582 switch(stack[stack_idx])
583 {
584 case HEAP_NODE_TRAVERSE_LEFT:
585 {
586 unsigned int left_idx = heap_idx << 1;
587 if (left_idx < heap->size && heap->entries[left_idx].version > version)
588 {
589 heap_idx = left_idx;
590 idx = heap->entries[heap_idx].idx;
591 apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
592
593 stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
594 stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
595 break;
596 }
597 }
598
599 case HEAP_NODE_TRAVERSE_RIGHT:
600 {
601 unsigned int right_idx = (heap_idx << 1) + 1;
602 if (right_idx < heap->size && heap->entries[right_idx].version > version)
603 {
604 heap_idx = right_idx;
605 idx = heap->entries[heap_idx].idx;
606 apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
607
608 stack[stack_idx++] = HEAP_NODE_POP;
609 stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
610 break;
611 }
612 }
613
614 case HEAP_NODE_POP:
615 {
616 heap_idx >>= 1;
617 --stack_idx;
618 break;
619 }
620 }
621 }
622 checkGLcall("walk_constant_heap_clamped()");
623}
624
625/* Loads floating point constants (aka uniforms) into the currently set GLSL program. */
626/* GL locking is done by the caller */
627static void shader_glsl_load_constantsF(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
628 const float *constants, const GLint *constant_locations, const struct constant_heap *heap,
629 unsigned char *stack, UINT version)
630{
631 const local_constant *lconst;
632
633 /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
634 if (This->baseShader.reg_maps.shader_version.major == 1
635 && shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type))
636 walk_constant_heap_clamped(gl_info, constants, constant_locations, heap, stack, version);
637 else
638 walk_constant_heap(gl_info, constants, constant_locations, heap, stack, version);
639
640 if (!This->baseShader.load_local_constsF)
641 {
642 TRACE("No need to load local float constants for this shader\n");
643 return;
644 }
645
646 /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
647 LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry)
648 {
649 GLint location = constant_locations[lconst->idx];
650 /* We found this uniform name in the program - go ahead and send the data */
651 if (location != -1) GL_EXTCALL(glUniform4fvARB(location, 1, (const GLfloat *)lconst->value));
652 }
653 checkGLcall("glUniform4fvARB()");
654}
655
656/* Loads integer constants (aka uniforms) into the currently set GLSL program. */
657/* GL locking is done by the caller */
658static void shader_glsl_load_constantsI(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
659 const GLint locations[MAX_CONST_I], const int *constants, WORD constants_set)
660{
661 unsigned int i;
662 struct list* ptr;
663
664 for (i = 0; constants_set; constants_set >>= 1, ++i)
665 {
666 if (!(constants_set & 1)) continue;
667
668 TRACE_(d3d_constants)("Loading constants %u: %i, %i, %i, %i\n",
669 i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
670
671 /* We found this uniform name in the program - go ahead and send the data */
672 GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
673 checkGLcall("glUniform4ivARB");
674 }
675
676 /* Load immediate constants */
677 ptr = list_head(&This->baseShader.constantsI);
678 while (ptr) {
679 const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
680 unsigned int idx = lconst->idx;
681 const GLint *values = (const GLint *)lconst->value;
682
683 TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
684 values[0], values[1], values[2], values[3]);
685
686 /* We found this uniform name in the program - go ahead and send the data */
687 GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
688 checkGLcall("glUniform4ivARB");
689 ptr = list_next(&This->baseShader.constantsI, ptr);
690 }
691}
692
693/* Loads boolean constants (aka uniforms) into the currently set GLSL program. */
694/* GL locking is done by the caller */
695static void shader_glsl_load_constantsB(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
696 GLhandleARB programId, const BOOL *constants, WORD constants_set)
697{
698 GLint tmp_loc;
699 unsigned int i;
700 char tmp_name[8];
701 const char *prefix;
702 struct list* ptr;
703
704 switch (This->baseShader.reg_maps.shader_version.type)
705 {
706 case WINED3D_SHADER_TYPE_VERTEX:
707 prefix = "VB";
708 break;
709
710 case WINED3D_SHADER_TYPE_GEOMETRY:
711 prefix = "GB";
712 break;
713
714 case WINED3D_SHADER_TYPE_PIXEL:
715 prefix = "PB";
716 break;
717
718 default:
719 FIXME("Unknown shader type %#x.\n",
720 This->baseShader.reg_maps.shader_version.type);
721 prefix = "UB";
722 break;
723 }
724
725 /* TODO: Benchmark and see if it would be beneficial to store the
726 * locations of the constants to avoid looking up each time */
727 for (i = 0; constants_set; constants_set >>= 1, ++i)
728 {
729 if (!(constants_set & 1)) continue;
730
731 TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
732
733 /* TODO: Benchmark and see if it would be beneficial to store the
734 * locations of the constants to avoid looking up each time */
735 snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
736 tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
737 if (tmp_loc != -1)
738 {
739 /* We found this uniform name in the program - go ahead and send the data */
740 GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
741 checkGLcall("glUniform1ivARB");
742 }
743 }
744
745 /* Load immediate constants */
746 ptr = list_head(&This->baseShader.constantsB);
747 while (ptr) {
748 const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
749 unsigned int idx = lconst->idx;
750 const GLint *values = (const GLint *)lconst->value;
751
752 TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
753
754 snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
755 tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
756 if (tmp_loc != -1) {
757 /* We found this uniform name in the program - go ahead and send the data */
758 GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
759 checkGLcall("glUniform1ivARB");
760 }
761 ptr = list_next(&This->baseShader.constantsB, ptr);
762 }
763}
764
765static void reset_program_constant_version(struct wine_rb_entry *entry, void *context)
766{
767 WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry)->constant_version = 0;
768}
769
770static const struct ps_np2fixup_info * get_fixup_info(const IWineD3DPixelShaderImpl *shader, UINT inp2fixup_info)
771{
772 struct glsl_pshader_private *shader_data = shader->baseShader.backend_data;
773
774 if (inp2fixup_info == WINEFIXUPINFO_NOINDEX)
775 return NULL;
776
777 if (!shader->baseShader.backend_data)
778 {
779 ERR("no backend data\n");
780 return NULL;
781 }
782 shader_data = shader->baseShader.backend_data;
783
784 if (inp2fixup_info >= shader_data->num_gl_shaders)
785 {
786 ERR("invalid index\n");
787 return NULL;
788 }
789
790 return &shader_data->gl_shaders[inp2fixup_info].np2fixup;
791}
792
793/**
794 * Loads the texture dimensions for NP2 fixup into the currently set GLSL program.
795 */
796/* GL locking is done by the caller (state handler) */
797static void shader_glsl_load_np2fixup_constants(
798 IWineD3DDevice* device,
799 char usePixelShader,
800 char useVertexShader) {
801
802 const IWineD3DDeviceImpl* deviceImpl = (const IWineD3DDeviceImpl*) device;
803 const struct glsl_shader_prog_link* prog = ((struct shader_glsl_priv *)(deviceImpl->shader_priv))->glsl_program;
804
805 if (!prog) {
806 /* No GLSL program set - nothing to do. */
807 return;
808 }
809
810 if (!usePixelShader) {
811 /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
812 return;
813 }
814
815 if (prog->ps_args.np2_fixup && -1 != prog->np2Fixup_location) {
816 const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
817 const IWineD3DStateBlockImpl* stateBlock = (const IWineD3DStateBlockImpl*) deviceImpl->stateBlock;
818 UINT i;
819 UINT fixup = prog->ps_args.np2_fixup;
820 GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
821
822 const struct ps_np2fixup_info *np2Fixup_info = WINEFIXUPINFO_GET(prog);
823
824 for (i = 0; fixup; fixup >>= 1, ++i) {
825 const unsigned char idx = np2Fixup_info->idx[i];
826 const IWineD3DBaseTextureImpl* const tex = (const IWineD3DBaseTextureImpl*) stateBlock->textures[i];
827 GLfloat* tex_dim = &np2fixup_constants[(idx >> 1) * 4];
828
829 if (!tex) {
830 FIXME("Nonexistent texture is flagged for NP2 texcoord fixup\n");
831 continue;
832 }
833
834 if (idx % 2) {
835 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
836 } else {
837 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
838 }
839 }
840
841 GL_EXTCALL(glUniform4fvARB(prog->np2Fixup_location, np2Fixup_info->num_consts, np2fixup_constants));
842 }
843}
844
845/**
846 * Loads the app-supplied constants into the currently set GLSL program.
847 */
848/* GL locking is done by the caller (state handler) */
849static void shader_glsl_load_constants(const struct wined3d_context *context,
850 char usePixelShader, char useVertexShader)
851{
852 const struct wined3d_gl_info *gl_info = context->gl_info;
853 IWineD3DDeviceImpl *device = context_get_device(context);
854 IWineD3DStateBlockImpl* stateBlock = device->stateBlock;
855 struct shader_glsl_priv *priv = device->shader_priv;
856
857 GLhandleARB programId;
858 struct glsl_shader_prog_link *prog = priv->glsl_program;
859 UINT constant_version;
860 int i;
861
862 if (!prog) {
863 /* No GLSL program set - nothing to do. */
864 return;
865 }
866 programId = prog->programId;
867 constant_version = prog->constant_version;
868
869 if (useVertexShader) {
870 IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
871
872 /* Load DirectX 9 float constants/uniforms for vertex shader */
873 shader_glsl_load_constantsF(vshader, gl_info, stateBlock->vertexShaderConstantF,
874 prog->vuniformF_locations, &priv->vconst_heap, priv->stack, constant_version);
875
876 /* Load DirectX 9 integer constants/uniforms for vertex shader */
877 shader_glsl_load_constantsI(vshader, gl_info, prog->vuniformI_locations, stateBlock->vertexShaderConstantI,
878 stateBlock->changed.vertexShaderConstantsI & vshader->baseShader.reg_maps.integer_constants);
879
880 /* Load DirectX 9 boolean constants/uniforms for vertex shader */
881 shader_glsl_load_constantsB(vshader, gl_info, programId, stateBlock->vertexShaderConstantB,
882 stateBlock->changed.vertexShaderConstantsB & vshader->baseShader.reg_maps.boolean_constants);
883
884 /* Upload the position fixup params */
885 GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, &device->posFixup[0]));
886 checkGLcall("glUniform4fvARB");
887 }
888
889 if (usePixelShader) {
890
891 IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
892
893 /* Load DirectX 9 float constants/uniforms for pixel shader */
894 shader_glsl_load_constantsF(pshader, gl_info, stateBlock->pixelShaderConstantF,
895 prog->puniformF_locations, &priv->pconst_heap, priv->stack, constant_version);
896
897 /* Load DirectX 9 integer constants/uniforms for pixel shader */
898 shader_glsl_load_constantsI(pshader, gl_info, prog->puniformI_locations, stateBlock->pixelShaderConstantI,
899 stateBlock->changed.pixelShaderConstantsI & pshader->baseShader.reg_maps.integer_constants);
900
901 /* Load DirectX 9 boolean constants/uniforms for pixel shader */
902 shader_glsl_load_constantsB(pshader, gl_info, programId, stateBlock->pixelShaderConstantB,
903 stateBlock->changed.pixelShaderConstantsB & pshader->baseShader.reg_maps.boolean_constants);
904
905 /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
906 * It can't be 0 for a valid texbem instruction.
907 */
908 for(i = 0; i < MAX_TEXTURES; i++) {
909 const float *data;
910
911 if(prog->bumpenvmat_location[i] == -1) continue;
912
913 data = (const float *)&stateBlock->textureState[i][WINED3DTSS_BUMPENVMAT00];
914 GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
915 checkGLcall("glUniformMatrix2fvARB");
916
917 /* texbeml needs the luminance scale and offset too. If texbeml is used, needsbumpmat
918 * is set too, so we can check that in the needsbumpmat check
919 */
920 if(prog->luminancescale_location[i] != -1) {
921 const GLfloat *scale = (const GLfloat *)&stateBlock->textureState[i][WINED3DTSS_BUMPENVLSCALE];
922 const GLfloat *offset = (const GLfloat *)&stateBlock->textureState[i][WINED3DTSS_BUMPENVLOFFSET];
923
924 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
925 checkGLcall("glUniform1fvARB");
926 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
927 checkGLcall("glUniform1fvARB");
928 }
929 }
930
931 if(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
932 float correction_params[4];
933
934 if (context->render_offscreen)
935 {
936 correction_params[0] = 0.0f;
937 correction_params[1] = 1.0f;
938 } else {
939 /* position is window relative, not viewport relative */
940#ifdef VBOX_WITH_VMSVGA
941 correction_params[0] = device->rtHeight;
942#else
943 correction_params[0] = ((IWineD3DSurfaceImpl *)context->current_rt)->currentDesc.Height;
944#endif
945 correction_params[1] = -1.0f;
946 }
947 GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
948 }
949 }
950
951 if (priv->next_constant_version == UINT_MAX)
952 {
953 TRACE("Max constant version reached, resetting to 0.\n");
954 wine_rb_for_each_entry(&priv->program_lookup, reset_program_constant_version, NULL);
955 priv->next_constant_version = 1;
956 }
957 else
958 {
959 prog->constant_version = priv->next_constant_version++;
960 }
961}
962
963static inline void update_heap_entry(struct constant_heap *heap, unsigned int idx,
964 unsigned int heap_idx, DWORD new_version)
965{
966 struct constant_entry *entries = heap->entries;
967 unsigned int *positions = heap->positions;
968 unsigned int parent_idx;
969
970 while (heap_idx > 1)
971 {
972 parent_idx = heap_idx >> 1;
973
974 if (new_version <= entries[parent_idx].version) break;
975
976 entries[heap_idx] = entries[parent_idx];
977 positions[entries[parent_idx].idx] = heap_idx;
978 heap_idx = parent_idx;
979 }
980
981 entries[heap_idx].version = new_version;
982 entries[heap_idx].idx = idx;
983 positions[idx] = heap_idx;
984}
985
986static void shader_glsl_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
987{
988 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
989 struct shader_glsl_priv *priv = This->shader_priv;
990 struct constant_heap *heap = &priv->vconst_heap;
991 UINT i;
992
993 for (i = start; i < count + start; ++i)
994 {
995 if (!This->stateBlock->changed.vertexShaderConstantsF[i])
996 update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
997 else
998 update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
999 }
1000}
1001
1002static void shader_glsl_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
1003{
1004 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
1005 struct shader_glsl_priv *priv = This->shader_priv;
1006 struct constant_heap *heap = &priv->pconst_heap;
1007 UINT i;
1008
1009 for (i = start; i < count + start; ++i)
1010 {
1011 if (!This->stateBlock->changed.pixelShaderConstantsF[i])
1012 update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
1013 else
1014 update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
1015 }
1016}
1017
1018static unsigned int vec4_varyings(DWORD shader_major, const struct wined3d_gl_info *gl_info)
1019{
1020 unsigned int ret = gl_info->limits.glsl_varyings / 4;
1021 /* 4.0 shaders do not write clip coords because d3d10 does not support user clipplanes */
1022 if(shader_major > 3) return ret;
1023
1024 /* 3.0 shaders may need an extra varying for the clip coord on some cards(mostly dx10 ones) */
1025 if (gl_info->quirks & WINED3D_QUIRK_GLSL_CLIP_VARYING) ret -= 1;
1026 return ret;
1027}
1028
1029/** Generate the variable & register declarations for the GLSL output target */
1030static void shader_generate_glsl_declarations(const struct wined3d_context *context,
1031 struct wined3d_shader_buffer *buffer, IWineD3DBaseShader *iface,
1032 const shader_reg_maps *reg_maps, struct shader_glsl_ctx_priv *ctx_priv)
1033{
1034 IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
1035 IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
1036 const struct ps_compile_args *ps_args = ctx_priv->cur_ps_args;
1037 const struct wined3d_gl_info *gl_info = context->gl_info;
1038 unsigned int i, extra_constants_needed = 0;
1039 const local_constant *lconst;
1040 DWORD map;
1041
1042 /* There are some minor differences between pixel and vertex shaders */
1043 char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
1044 char prefix = pshader ? 'P' : 'V';
1045
1046 /* Prototype the subroutines */
1047 for (i = 0, map = reg_maps->labels; map; map >>= 1, ++i)
1048 {
1049 if (map & 1) shader_addline(buffer, "void subroutine%u();\n", i);
1050 }
1051
1052 /* Declare the constants (aka uniforms) */
1053 if (This->baseShader.limits.constant_float > 0) {
1054 unsigned max_constantsF;
1055 /* Unless the shader uses indirect addressing, always declare the maximum array size and ignore that we need some
1056 * uniforms privately. E.g. if GL supports 256 uniforms, and we need 2 for the pos fixup and immediate values, still
1057 * declare VC[256]. If the shader needs more uniforms than we have it won't work in any case. If it uses less, the
1058 * compiler will figure out which uniforms are really used and strip them out. This allows a shader to use c255 on
1059 * a dx9 card, as long as it doesn't also use all the other constants.
1060 *
1061 * If the shader uses indirect addressing the compiler must assume that all declared uniforms are used. In this case,
1062 * declare only the amount that we're assured to have.
1063 *
1064 * Thus we run into problems in these two cases:
1065 * 1) The shader really uses more uniforms than supported
1066 * 2) The shader uses indirect addressing, less constants than supported, but uses a constant index > #supported consts
1067 */
1068 if (pshader)
1069 {
1070 /* No indirect addressing here. */
1071 max_constantsF = gl_info->limits.glsl_ps_float_constants;
1072 }
1073 else
1074 {
1075 if(This->baseShader.reg_maps.usesrelconstF) {
1076 /* Subtract the other potential uniforms from the max available (bools, ints, and 1 row of projection matrix).
1077 * Subtract another uniform for immediate values, which have to be loaded via uniform by the driver as well.
1078 * The shader code only uses 0.5, 2.0, 1.0, 128 and -128 in vertex shader code, so one vec4 should be enough
1079 * (Unfortunately the Nvidia driver doesn't store 128 and -128 in one float).
1080 *
1081 * Writing gl_ClipVertex requires one uniform for each clipplane as well.
1082 */
1083#ifdef VBOX_WITH_WDDM
1084 if (gl_info->limits.glsl_vs_float_constants == 256)
1085 {
1086 DWORD dwVersion = GetVersion();
1087 DWORD dwMajor = (DWORD)(LOBYTE(LOWORD(dwVersion)));
1088 DWORD dwMinor = (DWORD)(HIBYTE(LOWORD(dwVersion)));
1089 /* tmp workaround Win8 Aero requirement for 256 */
1090 if (dwMajor > 6 || dwMinor > 1)
1091 {
1092 /* tmp work-around to make Internet Explorer in win8 work with GPU supporting only with 256 shader uniform vars
1093 * @todo: make it more robust */
1094 max_constantsF = gl_info->limits.glsl_vs_float_constants - 1;
1095 }
1096 else
1097 max_constantsF = gl_info->limits.glsl_vs_float_constants - 3;
1098 }
1099 else
1100#endif
1101 {
1102 max_constantsF = gl_info->limits.glsl_vs_float_constants - 3;
1103 }
1104
1105 if(ctx_priv->cur_vs_args->clip_enabled)
1106 {
1107 max_constantsF -= gl_info->limits.clipplanes;
1108 }
1109 max_constantsF -= count_bits(This->baseShader.reg_maps.integer_constants);
1110 /* Strictly speaking a bool only uses one scalar, but the nvidia(Linux) compiler doesn't pack them properly,
1111 * so each scalar requires a full vec4. We could work around this by packing the booleans ourselves, but
1112 * for now take this into account when calculating the number of available constants
1113 */
1114 max_constantsF -= count_bits(This->baseShader.reg_maps.boolean_constants);
1115 /* Set by driver quirks in directx.c */
1116 max_constantsF -= gl_info->reserved_glsl_constants;
1117 }
1118 else
1119 {
1120 max_constantsF = gl_info->limits.glsl_vs_float_constants;
1121 }
1122 }
1123 max_constantsF = min(This->baseShader.limits.constant_float, max_constantsF);
1124 shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
1125 }
1126
1127 /* Always declare the full set of constants, the compiler can remove the unused ones because d3d doesn't(yet)
1128 * support indirect int and bool constant addressing. This avoids problems if the app uses e.g. i0 and i9.
1129 */
1130 if (This->baseShader.limits.constant_int > 0 && This->baseShader.reg_maps.integer_constants)
1131 shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, This->baseShader.limits.constant_int);
1132
1133 if (This->baseShader.limits.constant_bool > 0 && This->baseShader.reg_maps.boolean_constants)
1134 shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, This->baseShader.limits.constant_bool);
1135
1136 if(!pshader) {
1137 shader_addline(buffer, "uniform vec4 posFixup;\n");
1138 /* Predeclaration; This function is added at link time based on the pixel shader.
1139 * VS 3.0 shaders have an array OUT[] the shader writes to, earlier versions don't have
1140 * that. We know the input to the reorder function at vertex shader compile time, so
1141 * we can deal with that. The reorder function for a 1.x and 2.x vertex shader can just
1142 * read gl_FrontColor. The output depends on the pixel shader. The reorder function for a
1143 * 1.x and 2.x pshader or for fixed function will write gl_FrontColor, and for a 3.0 shader
1144 * it will write to the varying array. Here we depend on the shader optimizer on sorting that
1145 * out. The nvidia driver only does that if the parameter is inout instead of out, hence the
1146 * inout.
1147 */
1148 if (reg_maps->shader_version.major >= 3)
1149 {
1150 shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
1151 } else {
1152 shader_addline(buffer, "void order_ps_input();\n");
1153 }
1154 } else {
1155 for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
1156 {
1157 if (!(map & 1)) continue;
1158
1159 shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
1160
1161 if (reg_maps->luminanceparams & (1 << i))
1162 {
1163 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
1164 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
1165 extra_constants_needed++;
1166 }
1167
1168 extra_constants_needed++;
1169 }
1170
1171 if (ps_args->srgb_correction)
1172 {
1173 shader_addline(buffer, "const vec4 srgb_const0 = vec4(%.8e, %.8e, %.8e, %.8e);\n",
1174 srgb_pow, srgb_mul_high, srgb_sub_high, srgb_mul_low);
1175 shader_addline(buffer, "const vec4 srgb_const1 = vec4(%.8e, 0.0, 0.0, 0.0);\n",
1176 srgb_cmp);
1177 }
1178 if (reg_maps->vpos || reg_maps->usesdsy)
1179 {
1180 if (This->baseShader.limits.constant_float + extra_constants_needed
1181 + 1 < gl_info->limits.glsl_ps_float_constants)
1182 {
1183 shader_addline(buffer, "uniform vec4 ycorrection;\n");
1184 ((IWineD3DPixelShaderImpl *) This)->vpos_uniform = 1;
1185 extra_constants_needed++;
1186 } else {
1187 /* This happens because we do not have proper tracking of the constant registers that are
1188 * actually used, only the max limit of the shader version
1189 */
1190 FIXME("Cannot find a free uniform for vpos correction params\n");
1191 AssertFailed();
1192 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
1193 context->render_offscreen ? 0.0f : ((IWineD3DSurfaceImpl *)device->render_targets[0])->currentDesc.Height,
1194 context->render_offscreen ? 1.0f : -1.0f);
1195 }
1196 shader_addline(buffer, "vec4 vpos;\n");
1197 }
1198 }
1199
1200 /* Declare texture samplers */
1201 for (i = 0; i < This->baseShader.limits.sampler; i++) {
1202 if (reg_maps->sampler_type[i])
1203 {
1204 switch (reg_maps->sampler_type[i])
1205 {
1206 case WINED3DSTT_1D:
1207 shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
1208 break;
1209 case WINED3DSTT_2D:
1210 if(device->stateBlock->textures[i] &&
1211 IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[i]) == GL_TEXTURE_RECTANGLE_ARB) {
1212 shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
1213 } else {
1214 shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
1215 }
1216 break;
1217 case WINED3DSTT_CUBE:
1218 shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
1219 break;
1220 case WINED3DSTT_VOLUME:
1221 shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
1222 break;
1223 default:
1224 shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
1225 FIXME("Unrecognized sampler type: %#x\n", reg_maps->sampler_type[i]);
1226 break;
1227 }
1228 }
1229 }
1230
1231 /* Declare uniforms for NP2 texcoord fixup:
1232 * This is NOT done inside the loop that declares the texture samplers since the NP2 fixup code
1233 * is currently only used for the GeforceFX series and when forcing the ARB_npot extension off.
1234 * Modern cards just skip the code anyway, so put it inside a separate loop. */
1235 if (pshader && ps_args->np2_fixup) {
1236
1237 struct ps_np2fixup_info* const fixup = ctx_priv->cur_np2fixup_info;
1238 UINT cur = 0;
1239
1240 /* NP2/RECT textures in OpenGL use texcoords in the range [0,width]x[0,height]
1241 * while D3D has them in the (normalized) [0,1]x[0,1] range.
1242 * samplerNP2Fixup stores texture dimensions and is updated through
1243 * shader_glsl_load_np2fixup_constants when the sampler changes. */
1244
1245 for (i = 0; i < This->baseShader.limits.sampler; ++i) {
1246 if (reg_maps->sampler_type[i]) {
1247 if (!(ps_args->np2_fixup & (1 << i))) continue;
1248
1249 if (WINED3DSTT_2D != reg_maps->sampler_type[i]) {
1250 FIXME("Non-2D texture is flagged for NP2 texcoord fixup.\n");
1251 continue;
1252 }
1253
1254 fixup->idx[i] = cur++;
1255 }
1256 }
1257
1258 fixup->num_consts = (cur + 1) >> 1;
1259 shader_addline(buffer, "uniform vec4 %csamplerNP2Fixup[%u];\n", prefix, fixup->num_consts);
1260 }
1261
1262 /* Declare address variables */
1263 for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
1264 {
1265 if (map & 1) shader_addline(buffer, "ivec4 A%u;\n", i);
1266 }
1267
1268 /* Declare texture coordinate temporaries and initialize them */
1269 for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
1270 {
1271 if (map & 1) shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
1272 }
1273
1274 /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
1275 * helper function shader that is linked in at link time
1276 */
1277 if (pshader && reg_maps->shader_version.major >= 3)
1278 {
1279 if (use_vs(device->stateBlock))
1280 {
1281 shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1282 } else {
1283 /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
1284 * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
1285 * pixel shader that reads the fixed function color into the packed input registers.
1286 */
1287 shader_addline(buffer, "vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1288 }
1289 }
1290
1291 /* Declare output register temporaries */
1292 if(This->baseShader.limits.packed_output) {
1293 shader_addline(buffer, "vec4 OUT[%u];\n", This->baseShader.limits.packed_output);
1294 }
1295
1296 /* Declare temporary variables */
1297 for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
1298 {
1299 if (map & 1) shader_addline(buffer, "vec4 R%u;\n", i);
1300 }
1301
1302 /* Declare attributes */
1303 if (reg_maps->shader_version.type == WINED3D_SHADER_TYPE_VERTEX)
1304 {
1305 for (i = 0, map = reg_maps->input_registers; map; map >>= 1, ++i)
1306 {
1307 if (map & 1) shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
1308 }
1309 }
1310
1311 /* Declare loop registers aLx */
1312 for (i = 0; i < reg_maps->loop_depth; i++) {
1313 shader_addline(buffer, "int aL%u;\n", i);
1314 shader_addline(buffer, "int tmpInt%u;\n", i);
1315 }
1316
1317 /* Temporary variables for matrix operations */
1318 shader_addline(buffer, "vec4 tmp0;\n");
1319 shader_addline(buffer, "vec4 tmp1;\n");
1320#ifdef VBOX_WITH_VMSVGA
1321 shader_addline(buffer, "bool p0[4];\n");
1322#endif
1323
1324 /* Local constants use a different name so they can be loaded once at shader link time
1325 * They can't be hardcoded into the shader text via LC = {x, y, z, w}; because the
1326 * float -> string conversion can cause precision loss.
1327 */
1328 if(!This->baseShader.load_local_constsF) {
1329 LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
1330 shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
1331 }
1332 }
1333
1334 shader_addline(buffer, "const float FLT_MAX = 1e38;\n");
1335
1336 /* Start the main program */
1337 shader_addline(buffer, "void main() {\n");
1338 if(pshader && reg_maps->vpos) {
1339 /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
1340 * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
1341 * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
1342 * precision troubles when we just substract 0.5.
1343 *
1344 * To deal with that just floor() the position. This will eliminate the fraction on all cards.
1345 *
1346 * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
1347 *
1348 * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
1349 * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
1350 * coordinates specify the pixel centers instead of the pixel corners. This code will behave
1351 * correctly on drivers that returns integer values.
1352 */
1353 shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
1354 }
1355}
1356
1357/*****************************************************************************
1358 * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
1359 *
1360 * For more information, see http://wiki.winehq.org/DirectX-Shaders
1361 ****************************************************************************/
1362
1363/* Prototypes */
1364static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1365 const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src);
1366
1367/** Used for opcode modifiers - They multiply the result by the specified amount */
1368static const char * const shift_glsl_tab[] = {
1369 "", /* 0 (none) */
1370 "2.0 * ", /* 1 (x2) */
1371 "4.0 * ", /* 2 (x4) */
1372 "8.0 * ", /* 3 (x8) */
1373 "16.0 * ", /* 4 (x16) */
1374 "32.0 * ", /* 5 (x32) */
1375 "", /* 6 (x64) */
1376 "", /* 7 (x128) */
1377 "", /* 8 (d256) */
1378 "", /* 9 (d128) */
1379 "", /* 10 (d64) */
1380 "", /* 11 (d32) */
1381 "0.0625 * ", /* 12 (d16) */
1382 "0.125 * ", /* 13 (d8) */
1383 "0.25 * ", /* 14 (d4) */
1384 "0.5 * " /* 15 (d2) */
1385};
1386
1387/* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
1388static void shader_glsl_gen_modifier(DWORD src_modifier, const char *in_reg, const char *in_regswizzle, char *out_str)
1389{
1390 out_str[0] = 0;
1391
1392 switch (src_modifier)
1393 {
1394 case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
1395 case WINED3DSPSM_DW:
1396 case WINED3DSPSM_NONE:
1397 sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1398 break;
1399 case WINED3DSPSM_NEG:
1400 sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
1401 break;
1402 case WINED3DSPSM_NOT:
1403 sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
1404 break;
1405 case WINED3DSPSM_BIAS:
1406 sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1407 break;
1408 case WINED3DSPSM_BIASNEG:
1409 sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1410 break;
1411 case WINED3DSPSM_SIGN:
1412 sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1413 break;
1414 case WINED3DSPSM_SIGNNEG:
1415 sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1416 break;
1417 case WINED3DSPSM_COMP:
1418 sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
1419 break;
1420 case WINED3DSPSM_X2:
1421 sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
1422 break;
1423 case WINED3DSPSM_X2NEG:
1424 sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
1425 break;
1426 case WINED3DSPSM_ABS:
1427 sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
1428 break;
1429 case WINED3DSPSM_ABSNEG:
1430 sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
1431 break;
1432 default:
1433 FIXME("Unhandled modifier %u\n", src_modifier);
1434 sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1435 }
1436}
1437
1438/** Writes the GLSL variable name that corresponds to the register that the
1439 * DX opcode parameter is trying to access */
1440static void shader_glsl_get_register_name(const struct wined3d_shader_register *reg,
1441 char *register_name, BOOL *is_color, const struct wined3d_shader_instruction *ins)
1442{
1443 /* oPos, oFog and oPts in D3D */
1444 static const char * const hwrastout_reg_names[] = { "gl_Position", "gl_FogFragCoord", "gl_PointSize" };
1445
1446 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1447 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
1448 char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
1449
1450 *is_color = FALSE;
1451
1452 switch (reg->type)
1453 {
1454 case WINED3DSPR_TEMP:
1455 sprintf(register_name, "R%u", reg->idx);
1456 break;
1457
1458 case WINED3DSPR_INPUT:
1459 /* vertex shaders */
1460 if (!pshader)
1461 {
1462 struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1463 if (priv->cur_vs_args->swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1464 sprintf(register_name, "attrib%u", reg->idx);
1465 break;
1466 }
1467
1468 /* pixel shaders >= 3.0 */
1469 if (This->baseShader.reg_maps.shader_version.major >= 3)
1470 {
1471 DWORD idx = ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg->idx];
1472 unsigned int in_count = vec4_varyings(This->baseShader.reg_maps.shader_version.major, gl_info);
1473
1474 if (reg->rel_addr)
1475 {
1476 glsl_src_param_t rel_param;
1477
1478 shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1479
1480 /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
1481 * operation there */
1482 if (idx)
1483 {
1484 if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count)
1485 {
1486 sprintf(register_name,
1487 "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
1488 rel_param.param_str, idx, in_count - 1, rel_param.param_str, idx, in_count,
1489 rel_param.param_str, idx);
1490 }
1491 else
1492 {
1493 sprintf(register_name, "IN[%s + %u]", rel_param.param_str, idx);
1494 }
1495 }
1496 else
1497 {
1498 if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count)
1499 {
1500 sprintf(register_name, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
1501 rel_param.param_str, in_count - 1, rel_param.param_str, in_count,
1502 rel_param.param_str);
1503 }
1504 else
1505 {
1506 sprintf(register_name, "IN[%s]", rel_param.param_str);
1507 }
1508 }
1509 }
1510 else
1511 {
1512 if (idx == in_count) sprintf(register_name, "gl_Color");
1513 else if (idx == in_count + 1) sprintf(register_name, "gl_SecondaryColor");
1514 else sprintf(register_name, "IN[%u]", idx);
1515 }
1516 }
1517 else
1518 {
1519 if (reg->idx == 0) strcpy(register_name, "gl_Color");
1520 else strcpy(register_name, "gl_SecondaryColor");
1521 break;
1522 }
1523 break;
1524
1525 case WINED3DSPR_CONST:
1526 {
1527 const char prefix = pshader ? 'P' : 'V';
1528
1529 /* Relative addressing */
1530 if (reg->rel_addr)
1531 {
1532 glsl_src_param_t rel_param;
1533 shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1534 if (reg->idx) sprintf(register_name, "%cC[%s + %u]", prefix, rel_param.param_str, reg->idx);
1535 else sprintf(register_name, "%cC[%s]", prefix, rel_param.param_str);
1536 }
1537 else
1538 {
1539 if (shader_constant_is_local(This, reg->idx))
1540 sprintf(register_name, "%cLC%u", prefix, reg->idx);
1541 else
1542 sprintf(register_name, "%cC[%u]", prefix, reg->idx);
1543 }
1544 }
1545 break;
1546
1547 case WINED3DSPR_CONSTINT:
1548 if (pshader) sprintf(register_name, "PI[%u]", reg->idx);
1549 else sprintf(register_name, "VI[%u]", reg->idx);
1550 break;
1551
1552 case WINED3DSPR_CONSTBOOL:
1553 if (pshader) sprintf(register_name, "PB[%u]", reg->idx);
1554 else sprintf(register_name, "VB[%u]", reg->idx);
1555 break;
1556
1557 case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1558 if (pshader) sprintf(register_name, "T%u", reg->idx);
1559 else sprintf(register_name, "A%u", reg->idx);
1560 break;
1561
1562 case WINED3DSPR_LOOP:
1563 sprintf(register_name, "aL%u", This->baseShader.cur_loop_regno - 1);
1564 break;
1565
1566 case WINED3DSPR_SAMPLER:
1567 if (pshader) sprintf(register_name, "Psampler%u", reg->idx);
1568 else sprintf(register_name, "Vsampler%u", reg->idx);
1569 break;
1570
1571 case WINED3DSPR_COLOROUT:
1572 if (reg->idx >= gl_info->limits.buffers)
1573 WARN("Write to render target %u, only %d supported.\n", reg->idx, gl_info->limits.buffers);
1574
1575 sprintf(register_name, "gl_FragData[%u]", reg->idx);
1576 break;
1577
1578 case WINED3DSPR_RASTOUT:
1579 sprintf(register_name, "%s", hwrastout_reg_names[reg->idx]);
1580 break;
1581
1582 case WINED3DSPR_DEPTHOUT:
1583 sprintf(register_name, "gl_FragDepth");
1584 break;
1585
1586 case WINED3DSPR_ATTROUT:
1587 if (reg->idx == 0) sprintf(register_name, "gl_FrontColor");
1588 else sprintf(register_name, "gl_FrontSecondaryColor");
1589 break;
1590
1591 case WINED3DSPR_TEXCRDOUT:
1592 /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
1593 if (This->baseShader.reg_maps.shader_version.major >= 3) sprintf(register_name, "OUT[%u]", reg->idx);
1594 else sprintf(register_name, "gl_TexCoord[%u]", reg->idx);
1595 break;
1596
1597 case WINED3DSPR_MISCTYPE:
1598 if (reg->idx == 0)
1599 {
1600 /* vPos */
1601 sprintf(register_name, "vpos");
1602 }
1603 else if (reg->idx == 1)
1604 {
1605 /* Note that gl_FrontFacing is a bool, while vFace is
1606 * a float for which the sign determines front/back */
1607 sprintf(register_name, "(gl_FrontFacing ? 1.0 : -1.0)");
1608 }
1609 else
1610 {
1611 FIXME("Unhandled misctype register %d\n", reg->idx);
1612 sprintf(register_name, "unrecognized_register");
1613 }
1614 break;
1615
1616 case WINED3DSPR_IMMCONST:
1617 switch (reg->immconst_type)
1618 {
1619 case WINED3D_IMMCONST_FLOAT:
1620 sprintf(register_name, "%.8e", *(const float *)reg->immconst_data);
1621 break;
1622
1623 case WINED3D_IMMCONST_FLOAT4:
1624 sprintf(register_name, "vec4(%.8e, %.8e, %.8e, %.8e)",
1625 *(const float *)&reg->immconst_data[0], *(const float *)&reg->immconst_data[1],
1626 *(const float *)&reg->immconst_data[2], *(const float *)&reg->immconst_data[3]);
1627 break;
1628
1629 default:
1630 FIXME("Unhandled immconst type %#x\n", reg->immconst_type);
1631 sprintf(register_name, "<unhandled_immconst_type %#x>", reg->immconst_type);
1632 }
1633 break;
1634
1635#ifdef VBOX_WITH_VMSVGA
1636 case WINED3DSPR_PREDICATE:
1637 sprintf(register_name, "p0");
1638 break;
1639#endif
1640
1641 default:
1642 FIXME("Unhandled register name Type(%d)\n", reg->type);
1643 sprintf(register_name, "unrecognized_register");
1644 break;
1645 }
1646}
1647
1648static void shader_glsl_write_mask_to_str(DWORD write_mask, char *str)
1649{
1650 *str++ = '.';
1651 if (write_mask & WINED3DSP_WRITEMASK_0) *str++ = 'x';
1652 if (write_mask & WINED3DSP_WRITEMASK_1) *str++ = 'y';
1653 if (write_mask & WINED3DSP_WRITEMASK_2) *str++ = 'z';
1654 if (write_mask & WINED3DSP_WRITEMASK_3) *str++ = 'w';
1655 *str = '\0';
1656}
1657
1658/* Get the GLSL write mask for the destination register */
1659static DWORD shader_glsl_get_write_mask(const struct wined3d_shader_dst_param *param, char *write_mask)
1660{
1661 DWORD mask = param->write_mask;
1662
1663 if (shader_is_scalar(&param->reg))
1664 {
1665 mask = WINED3DSP_WRITEMASK_0;
1666 *write_mask = '\0';
1667 }
1668 else
1669 {
1670#ifdef VBOX_WITH_VMSVGA
1671 if (param->reg.type == WINED3DSPR_PREDICATE)
1672 {
1673 *write_mask++ = '[';
1674 if (mask & WINED3DSP_WRITEMASK_0) *write_mask++ = '0';
1675 else
1676 if (mask & WINED3DSP_WRITEMASK_1) *write_mask++ = '1';
1677 else
1678 if (mask & WINED3DSP_WRITEMASK_2) *write_mask++ = '2';
1679 else
1680 if (mask & WINED3DSP_WRITEMASK_3) *write_mask++ = '3';
1681 *write_mask++ = ']';
1682 *write_mask = '\0';
1683 }
1684 else
1685#endif
1686 shader_glsl_write_mask_to_str(mask, write_mask);
1687 }
1688
1689 return mask;
1690}
1691
1692static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1693 unsigned int size = 0;
1694
1695 if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1696 if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1697 if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1698 if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1699
1700 return size;
1701}
1702
1703static void shader_glsl_swizzle_to_str(const DWORD swizzle, BOOL fixup, DWORD mask, char *str)
1704{
1705 /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1706 * but addressed as "rgba". To fix this we need to swap the register's x
1707 * and z components. */
1708 const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1709
1710 *str++ = '.';
1711 /* swizzle bits fields: wwzzyyxx */
1712 if (mask & WINED3DSP_WRITEMASK_0) *str++ = swizzle_chars[swizzle & 0x03];
1713 if (mask & WINED3DSP_WRITEMASK_1) *str++ = swizzle_chars[(swizzle >> 2) & 0x03];
1714 if (mask & WINED3DSP_WRITEMASK_2) *str++ = swizzle_chars[(swizzle >> 4) & 0x03];
1715 if (mask & WINED3DSP_WRITEMASK_3) *str++ = swizzle_chars[(swizzle >> 6) & 0x03];
1716 *str = '\0';
1717}
1718
1719static void shader_glsl_get_swizzle(const struct wined3d_shader_src_param *param,
1720 BOOL fixup, DWORD mask, char *swizzle_str)
1721{
1722 if (shader_is_scalar(&param->reg))
1723 *swizzle_str = '\0';
1724 else
1725 shader_glsl_swizzle_to_str(param->swizzle, fixup, mask, swizzle_str);
1726}
1727
1728/* From a given parameter token, generate the corresponding GLSL string.
1729 * Also, return the actual register name and swizzle in case the
1730 * caller needs this information as well. */
1731static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1732 const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src)
1733{
1734 BOOL is_color = FALSE;
1735 char swizzle_str[6];
1736
1737 glsl_src->reg_name[0] = '\0';
1738 glsl_src->param_str[0] = '\0';
1739 swizzle_str[0] = '\0';
1740
1741 shader_glsl_get_register_name(&wined3d_src->reg, glsl_src->reg_name, &is_color, ins);
1742 shader_glsl_get_swizzle(wined3d_src, is_color, mask, swizzle_str);
1743 shader_glsl_gen_modifier(wined3d_src->modifiers, glsl_src->reg_name, swizzle_str, glsl_src->param_str);
1744}
1745
1746/* From a given parameter token, generate the corresponding GLSL string.
1747 * Also, return the actual register name and swizzle in case the
1748 * caller needs this information as well. */
1749static DWORD shader_glsl_add_dst_param(const struct wined3d_shader_instruction *ins,
1750 const struct wined3d_shader_dst_param *wined3d_dst, glsl_dst_param_t *glsl_dst)
1751{
1752 BOOL is_color = FALSE;
1753
1754 glsl_dst->mask_str[0] = '\0';
1755 glsl_dst->reg_name[0] = '\0';
1756
1757 shader_glsl_get_register_name(&wined3d_dst->reg, glsl_dst->reg_name, &is_color, ins);
1758 return shader_glsl_get_write_mask(wined3d_dst, glsl_dst->mask_str);
1759}
1760
1761/* Append the destination part of the instruction to the buffer, return the effective write mask */
1762static DWORD shader_glsl_append_dst_ext(struct wined3d_shader_buffer *buffer,
1763 const struct wined3d_shader_instruction *ins, const struct wined3d_shader_dst_param *dst)
1764{
1765 glsl_dst_param_t glsl_dst;
1766 DWORD mask;
1767
1768 mask = shader_glsl_add_dst_param(ins, dst, &glsl_dst);
1769 if (mask) shader_addline(buffer, "%s%s = %s(", glsl_dst.reg_name, glsl_dst.mask_str, shift_glsl_tab[dst->shift]);
1770
1771 return mask;
1772}
1773
1774/* Append the destination part of the instruction to the buffer, return the effective write mask */
1775static DWORD shader_glsl_append_dst(struct wined3d_shader_buffer *buffer, const struct wined3d_shader_instruction *ins)
1776{
1777 return shader_glsl_append_dst_ext(buffer, ins, &ins->dst[0]);
1778}
1779
1780/** Process GLSL instruction modifiers */
1781static void shader_glsl_add_instruction_modifiers(const struct wined3d_shader_instruction *ins)
1782{
1783 glsl_dst_param_t dst_param;
1784 DWORD modifiers;
1785
1786 if (!ins->dst_count) return;
1787
1788 modifiers = ins->dst[0].modifiers;
1789 if (!modifiers) return;
1790
1791 shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
1792
1793 if (modifiers & WINED3DSPDM_SATURATE)
1794 {
1795 /* _SAT means to clamp the value of the register to between 0 and 1 */
1796 shader_addline(ins->ctx->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1797 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1798 }
1799
1800 if (modifiers & WINED3DSPDM_MSAMPCENTROID)
1801 {
1802 FIXME("_centroid modifier not handled\n");
1803 }
1804
1805 if (modifiers & WINED3DSPDM_PARTIALPRECISION)
1806 {
1807 /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1808 }
1809}
1810
1811static inline const char *shader_get_comp_op(DWORD op)
1812{
1813 switch (op) {
1814 case COMPARISON_GT: return ">";
1815 case COMPARISON_EQ: return "==";
1816 case COMPARISON_GE: return ">=";
1817 case COMPARISON_LT: return "<";
1818 case COMPARISON_NE: return "!=";
1819 case COMPARISON_LE: return "<=";
1820 default:
1821 FIXME("Unrecognized comparison value: %u\n", op);
1822 return "(\?\?)";
1823 }
1824}
1825
1826static void shader_glsl_get_sample_function(const struct wined3d_gl_info *gl_info,
1827 DWORD sampler_type, DWORD flags, glsl_sample_function_t *sample_function)
1828{
1829 BOOL projected = flags & WINED3D_GLSL_SAMPLE_PROJECTED;
1830 BOOL texrect = flags & WINED3D_GLSL_SAMPLE_RECT;
1831 BOOL lod = flags & WINED3D_GLSL_SAMPLE_LOD;
1832 BOOL grad = flags & WINED3D_GLSL_SAMPLE_GRAD;
1833
1834 /* Note that there's no such thing as a projected cube texture. */
1835 switch(sampler_type) {
1836 case WINED3DSTT_1D:
1837 if(lod) {
1838 sample_function->name = projected ? "texture1DProjLod" : "texture1DLod";
1839 }
1840 else if (grad)
1841 {
1842 if (gl_info->supported[EXT_GPU_SHADER4])
1843 sample_function->name = projected ? "texture1DProjGrad" : "texture1DGrad";
1844 else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1845 sample_function->name = projected ? "texture1DProjGradARB" : "texture1DGradARB";
1846 else
1847 {
1848 FIXME("Unsupported 1D grad function.\n");
1849 sample_function->name = "unsupported1DGrad";
1850 }
1851 }
1852 else
1853 {
1854 sample_function->name = projected ? "texture1DProj" : "texture1D";
1855 }
1856 sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1857 break;
1858 case WINED3DSTT_2D:
1859 if(texrect) {
1860 if(lod) {
1861 sample_function->name = projected ? "texture2DRectProjLod" : "texture2DRectLod";
1862 }
1863 else if (grad)
1864 {
1865 if (gl_info->supported[EXT_GPU_SHADER4])
1866 sample_function->name = projected ? "texture2DRectProjGrad" : "texture2DRectGrad";
1867 else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1868 sample_function->name = projected ? "texture2DRectProjGradARB" : "texture2DRectGradARB";
1869 else
1870 {
1871 FIXME("Unsupported RECT grad function.\n");
1872 sample_function->name = "unsupported2DRectGrad";
1873 }
1874 }
1875 else
1876 {
1877 sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1878 }
1879 } else {
1880 if(lod) {
1881 sample_function->name = projected ? "texture2DProjLod" : "texture2DLod";
1882 }
1883 else if (grad)
1884 {
1885 if (gl_info->supported[EXT_GPU_SHADER4])
1886 sample_function->name = projected ? "texture2DProjGrad" : "texture2DGrad";
1887 else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1888 sample_function->name = projected ? "texture2DProjGradARB" : "texture2DGradARB";
1889 else
1890 {
1891 FIXME("Unsupported 2D grad function.\n");
1892 sample_function->name = "unsupported2DGrad";
1893 }
1894 }
1895 else
1896 {
1897 sample_function->name = projected ? "texture2DProj" : "texture2D";
1898 }
1899 }
1900 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1901 break;
1902 case WINED3DSTT_CUBE:
1903 if(lod) {
1904 sample_function->name = "textureCubeLod";
1905 }
1906 else if (grad)
1907 {
1908 if (gl_info->supported[EXT_GPU_SHADER4])
1909 sample_function->name = "textureCubeGrad";
1910 else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1911 sample_function->name = "textureCubeGradARB";
1912 else
1913 {
1914 FIXME("Unsupported Cube grad function.\n");
1915 sample_function->name = "unsupportedCubeGrad";
1916 }
1917 }
1918 else
1919 {
1920 sample_function->name = "textureCube";
1921 }
1922 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1923 break;
1924 case WINED3DSTT_VOLUME:
1925 if(lod) {
1926 sample_function->name = projected ? "texture3DProjLod" : "texture3DLod";
1927 }
1928 else if (grad)
1929 {
1930 if (gl_info->supported[EXT_GPU_SHADER4])
1931 sample_function->name = projected ? "texture3DProjGrad" : "texture3DGrad";
1932 else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1933 sample_function->name = projected ? "texture3DProjGradARB" : "texture3DGradARB";
1934 else
1935 {
1936 FIXME("Unsupported 3D grad function.\n");
1937 sample_function->name = "unsupported3DGrad";
1938 }
1939 }
1940 else
1941 {
1942 sample_function->name = projected ? "texture3DProj" : "texture3D";
1943 }
1944 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1945 break;
1946 default:
1947 sample_function->name = "";
1948 sample_function->coord_mask = 0;
1949 FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1950 break;
1951 }
1952}
1953
1954static void shader_glsl_append_fixup_arg(char *arguments, const char *reg_name,
1955 BOOL sign_fixup, enum fixup_channel_source channel_source)
1956{
1957 switch(channel_source)
1958 {
1959 case CHANNEL_SOURCE_ZERO:
1960 strcat(arguments, "0.0");
1961 break;
1962
1963 case CHANNEL_SOURCE_ONE:
1964 strcat(arguments, "1.0");
1965 break;
1966
1967 case CHANNEL_SOURCE_X:
1968 strcat(arguments, reg_name);
1969 strcat(arguments, ".x");
1970 break;
1971
1972 case CHANNEL_SOURCE_Y:
1973 strcat(arguments, reg_name);
1974 strcat(arguments, ".y");
1975 break;
1976
1977 case CHANNEL_SOURCE_Z:
1978 strcat(arguments, reg_name);
1979 strcat(arguments, ".z");
1980 break;
1981
1982 case CHANNEL_SOURCE_W:
1983 strcat(arguments, reg_name);
1984 strcat(arguments, ".w");
1985 break;
1986
1987 default:
1988 FIXME("Unhandled channel source %#x\n", channel_source);
1989 strcat(arguments, "undefined");
1990 break;
1991 }
1992
1993 if (sign_fixup) strcat(arguments, " * 2.0 - 1.0");
1994}
1995
1996static void shader_glsl_color_correction(const struct wined3d_shader_instruction *ins, struct color_fixup_desc fixup)
1997{
1998 struct wined3d_shader_dst_param dst;
1999 unsigned int mask_size, remaining;
2000 glsl_dst_param_t dst_param;
2001 char arguments[256];
2002 DWORD mask;
2003
2004 mask = 0;
2005 if (fixup.x_sign_fixup || fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
2006 if (fixup.y_sign_fixup || fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
2007 if (fixup.z_sign_fixup || fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
2008 if (fixup.w_sign_fixup || fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
2009 mask &= ins->dst[0].write_mask;
2010
2011 if (!mask) return; /* Nothing to do */
2012
2013 if (is_complex_fixup(fixup))
2014 {
2015 enum complex_fixup complex_fixup = get_complex_fixup(fixup);
2016 FIXME("Complex fixup (%#x) not supported\n",complex_fixup); (void)complex_fixup;
2017 return;
2018 }
2019
2020 mask_size = shader_glsl_get_write_mask_size(mask);
2021
2022 dst = ins->dst[0];
2023 dst.write_mask = mask;
2024 shader_glsl_add_dst_param(ins, &dst, &dst_param);
2025
2026 arguments[0] = '\0';
2027 remaining = mask_size;
2028 if (mask & WINED3DSP_WRITEMASK_0)
2029 {
2030 shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.x_sign_fixup, fixup.x_source);
2031 if (--remaining) strcat(arguments, ", ");
2032 }
2033 if (mask & WINED3DSP_WRITEMASK_1)
2034 {
2035 shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.y_sign_fixup, fixup.y_source);
2036 if (--remaining) strcat(arguments, ", ");
2037 }
2038 if (mask & WINED3DSP_WRITEMASK_2)
2039 {
2040 shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.z_sign_fixup, fixup.z_source);
2041 if (--remaining) strcat(arguments, ", ");
2042 }
2043 if (mask & WINED3DSP_WRITEMASK_3)
2044 {
2045 shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.w_sign_fixup, fixup.w_source);
2046 if (--remaining) strcat(arguments, ", ");
2047 }
2048
2049 if (mask_size > 1)
2050 {
2051 shader_addline(ins->ctx->buffer, "%s%s = vec%u(%s);\n",
2052 dst_param.reg_name, dst_param.mask_str, mask_size, arguments);
2053 }
2054 else
2055 {
2056 shader_addline(ins->ctx->buffer, "%s%s = %s;\n", dst_param.reg_name, dst_param.mask_str, arguments);
2057 }
2058}
2059
2060static void PRINTF_ATTR(8, 9) shader_glsl_gen_sample_code(const struct wined3d_shader_instruction *ins,
2061 DWORD sampler, const glsl_sample_function_t *sample_function, DWORD swizzle,
2062 const char *dx, const char *dy,
2063 const char *bias, const char *coord_reg_fmt, ...)
2064{
2065 const char *sampler_base;
2066 char dst_swizzle[6];
2067 struct color_fixup_desc fixup;
2068 BOOL np2_fixup = FALSE;
2069 BOOL tmirror_tmp_reg = FALSE;
2070 va_list args;
2071
2072 shader_glsl_swizzle_to_str(swizzle, FALSE, ins->dst[0].write_mask, dst_swizzle);
2073
2074 if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
2075 {
2076 const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2077 fixup = priv->cur_ps_args->color_fixup[sampler];
2078 sampler_base = "Psampler";
2079
2080 if (priv->cur_ps_args->np2_fixup & (1 << sampler)) {
2081 if(bias) {
2082 FIXME("Biased sampling from NP2 textures is unsupported\n");
2083 } else {
2084 np2_fixup = TRUE;
2085 }
2086 }
2087
2088 if (priv->cur_ps_args->t_mirror & (1 << sampler))
2089 {
2090 if (ins->ctx->reg_maps->sampler_type[sampler]==WINED3DSTT_2D)
2091 {
2092 if (sample_function->coord_mask & WINED3DSP_WRITEMASK_1)
2093 {
2094 glsl_src_param_t coord_param;
2095 shader_glsl_add_src_param(ins, &ins->src[0], sample_function->coord_mask, &coord_param);
2096
2097 if (ins->src[0].reg.type != WINED3DSPR_INPUT)
2098 {
2099 shader_addline(ins->ctx->buffer, "%s.y=1.0-%s.y;\n",
2100 coord_param.reg_name, coord_param.reg_name);
2101 }
2102 else
2103 {
2104 tmirror_tmp_reg = TRUE;
2105 shader_addline(ins->ctx->buffer, "tmp0.xy=vec2(%s.x, 1.0-%s.y).xy;\n",
2106 coord_param.reg_name, coord_param.reg_name);
2107 }
2108 }
2109 else
2110 {
2111 DebugBreak();
2112 FIXME("Unexpected coord_mask with t_mirror\n");
2113 }
2114 }
2115 }
2116 } else {
2117 sampler_base = "Vsampler";
2118 fixup = COLOR_FIXUP_IDENTITY; /* FIXME: Vshader color fixup */
2119 }
2120
2121 shader_glsl_append_dst(ins->ctx->buffer, ins);
2122
2123 shader_addline(ins->ctx->buffer, "%s(%s%u, ", sample_function->name, sampler_base, sampler);
2124
2125 if (tmirror_tmp_reg)
2126 {
2127 shader_addline(ins->ctx->buffer, "%s", "tmp0.xy");
2128 }
2129 else
2130 {
2131 va_start(args, coord_reg_fmt);
2132 shader_vaddline(ins->ctx->buffer, coord_reg_fmt, args);
2133 va_end(args);
2134 }
2135
2136 if(bias) {
2137 shader_addline(ins->ctx->buffer, ", %s)%s);\n", bias, dst_swizzle);
2138 } else {
2139 if (np2_fixup) {
2140 const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2141 const unsigned char idx = priv->cur_np2fixup_info->idx[sampler];
2142
2143 shader_addline(ins->ctx->buffer, " * PsamplerNP2Fixup[%u].%s)%s);\n", idx >> 1,
2144 (idx % 2) ? "zw" : "xy", dst_swizzle);
2145 } else if(dx && dy) {
2146 shader_addline(ins->ctx->buffer, ", %s, %s)%s);\n", dx, dy, dst_swizzle);
2147 } else {
2148 shader_addline(ins->ctx->buffer, ")%s);\n", dst_swizzle);
2149 }
2150 }
2151
2152 if(!is_identity_fixup(fixup)) {
2153 shader_glsl_color_correction(ins, fixup);
2154 }
2155}
2156
2157/*****************************************************************************
2158 * Begin processing individual instruction opcodes
2159 ****************************************************************************/
2160
2161/* Generate GLSL arithmetic functions (dst = src1 + src2) */
2162static void shader_glsl_arith(const struct wined3d_shader_instruction *ins)
2163{
2164 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2165 glsl_src_param_t src0_param;
2166 glsl_src_param_t src1_param;
2167 DWORD write_mask;
2168 char op;
2169
2170 /* Determine the GLSL operator to use based on the opcode */
2171 switch (ins->handler_idx)
2172 {
2173 case WINED3DSIH_MUL: op = '*'; break;
2174 case WINED3DSIH_ADD: op = '+'; break;
2175 case WINED3DSIH_SUB: op = '-'; break;
2176 default:
2177 op = ' ';
2178 FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2179 break;
2180 }
2181
2182 write_mask = shader_glsl_append_dst(buffer, ins);
2183 shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2184 shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2185 shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
2186}
2187
2188#ifdef VBOX_WITH_VMSVGA
2189static void shader_glsl_mov_impl(const struct wined3d_shader_instruction *ins, int p0_idx);
2190
2191/* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
2192static void shader_glsl_mov(const struct wined3d_shader_instruction *ins)
2193{
2194 if (ins->predicate)
2195 {
2196 int i;
2197 DWORD dst_mask = ins->dst[0].write_mask;
2198 struct wined3d_shader_dst_param *dst = (struct wined3d_shader_dst_param *)&ins->dst[0];
2199
2200 for (i = 0; i < 4; i++)
2201 {
2202 if (dst_mask & RT_BIT(i))
2203 {
2204 dst->write_mask = RT_BIT(i);
2205
2206 shader_glsl_mov_impl(ins, i);
2207 }
2208 }
2209 dst->write_mask = dst_mask;
2210 }
2211 else
2212 shader_glsl_mov_impl(ins, 0);
2213}
2214
2215/* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
2216static void shader_glsl_mov_impl(const struct wined3d_shader_instruction *ins, int p0_idx)
2217
2218#else
2219/* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
2220static void shader_glsl_mov(const struct wined3d_shader_instruction *ins)
2221#endif
2222{
2223 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
2224 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2225 glsl_src_param_t src0_param;
2226 DWORD write_mask;
2227
2228#ifdef VBOX_WITH_VMSVGA
2229 if (ins->predicate)
2230 {
2231 shader_addline(buffer, "if (p0[%d]) {\n", p0_idx);
2232 }
2233#endif
2234
2235 write_mask = shader_glsl_append_dst(buffer, ins);
2236 shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2237
2238 /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
2239 * shader versions WINED3DSIO_MOVA is used for this. */
2240 if (ins->ctx->reg_maps->shader_version.major == 1
2241 && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
2242 && ins->dst[0].reg.type == WINED3DSPR_ADDR)
2243 {
2244 /* This is a simple floor() */
2245 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2246 if (mask_size > 1) {
2247 shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
2248 } else {
2249 shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
2250 }
2251 }
2252 else if(ins->handler_idx == WINED3DSIH_MOVA)
2253 {
2254 /* We need to *round* to the nearest int here. */
2255 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2256
2257 if (gl_info->supported[EXT_GPU_SHADER4])
2258 {
2259 if (mask_size > 1)
2260 shader_addline(buffer, "ivec%d(round(%s)));\n", mask_size, src0_param.param_str);
2261 else
2262 shader_addline(buffer, "int(round(%s)));\n", src0_param.param_str);
2263 }
2264 else
2265 {
2266 if (mask_size > 1)
2267 shader_addline(buffer, "ivec%d(floor(abs(%s) + vec%d(0.5)) * sign(%s)));\n",
2268 mask_size, src0_param.param_str, mask_size, src0_param.param_str);
2269 else
2270 shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n",
2271 src0_param.param_str, src0_param.param_str);
2272 }
2273 }
2274 else
2275 {
2276 shader_addline(buffer, "%s);\n", src0_param.param_str);
2277 }
2278#ifdef VBOX_WITH_VMSVGA
2279 if (ins->predicate)
2280 {
2281 shader_addline(buffer, "}\n");
2282 }
2283#endif
2284}
2285
2286/* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
2287static void shader_glsl_dot(const struct wined3d_shader_instruction *ins)
2288{
2289 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2290 glsl_src_param_t src0_param;
2291 glsl_src_param_t src1_param;
2292 DWORD dst_write_mask, src_write_mask;
2293 unsigned int dst_size = 0;
2294
2295 dst_write_mask = shader_glsl_append_dst(buffer, ins);
2296 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2297
2298 /* dp3 works on vec3, dp4 on vec4 */
2299 if (ins->handler_idx == WINED3DSIH_DP4)
2300 {
2301 src_write_mask = WINED3DSP_WRITEMASK_ALL;
2302 } else {
2303 src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2304 }
2305
2306 shader_glsl_add_src_param(ins, &ins->src[0], src_write_mask, &src0_param);
2307 shader_glsl_add_src_param(ins, &ins->src[1], src_write_mask, &src1_param);
2308
2309 if (dst_size > 1) {
2310 shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
2311 } else {
2312 shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
2313 }
2314}
2315
2316/* Note that this instruction has some restrictions. The destination write mask
2317 * can't contain the w component, and the source swizzles have to be .xyzw */
2318static void shader_glsl_cross(const struct wined3d_shader_instruction *ins)
2319{
2320 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2321 glsl_src_param_t src0_param;
2322 glsl_src_param_t src1_param;
2323 char dst_mask[6];
2324
2325 shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2326 shader_glsl_append_dst(ins->ctx->buffer, ins);
2327 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2328 shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
2329 shader_addline(ins->ctx->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
2330}
2331
2332/* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
2333 * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
2334 * GLSL uses the value as-is. */
2335static void shader_glsl_pow(const struct wined3d_shader_instruction *ins)
2336{
2337 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2338 glsl_src_param_t src0_param;
2339 glsl_src_param_t src1_param;
2340 DWORD dst_write_mask;
2341 unsigned int dst_size;
2342
2343 dst_write_mask = shader_glsl_append_dst(buffer, ins);
2344 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2345
2346 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2347 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2348
2349 if (dst_size > 1) {
2350 shader_addline(buffer, "vec%d(pow(abs(%s), %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
2351 } else {
2352 shader_addline(buffer, "pow(abs(%s), %s));\n", src0_param.param_str, src1_param.param_str);
2353 }
2354}
2355
2356/* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
2357 * Src0 is a scalar. Note that D3D uses the absolute of src0, while
2358 * GLSL uses the value as-is. */
2359static void shader_glsl_log(const struct wined3d_shader_instruction *ins)
2360{
2361 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2362 glsl_src_param_t src0_param;
2363 DWORD dst_write_mask;
2364 unsigned int dst_size;
2365
2366 dst_write_mask = shader_glsl_append_dst(buffer, ins);
2367 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2368
2369 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2370
2371 if (dst_size > 1)
2372 {
2373 shader_addline(buffer, "vec%d(%s == 0.0 ? -FLT_MAX : log2(abs(%s))));\n",
2374 dst_size, src0_param.param_str, src0_param.param_str);
2375 }
2376 else
2377 {
2378 shader_addline(buffer, "%s == 0.0 ? -FLT_MAX : log2(abs(%s)));\n",
2379 src0_param.param_str, src0_param.param_str);
2380 }
2381}
2382
2383/* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
2384static void shader_glsl_map2gl(const struct wined3d_shader_instruction *ins)
2385{
2386 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2387 glsl_src_param_t src_param;
2388 const char *instruction;
2389 DWORD write_mask;
2390 unsigned i;
2391
2392 /* Determine the GLSL function to use based on the opcode */
2393 /* TODO: Possibly make this a table for faster lookups */
2394 switch (ins->handler_idx)
2395 {
2396 case WINED3DSIH_MIN: instruction = "min"; break;
2397 case WINED3DSIH_MAX: instruction = "max"; break;
2398 case WINED3DSIH_ABS: instruction = "abs"; break;
2399 case WINED3DSIH_FRC: instruction = "fract"; break;
2400 case WINED3DSIH_EXP: instruction = "exp2"; break;
2401 case WINED3DSIH_DSX: instruction = "dFdx"; break;
2402 case WINED3DSIH_DSY: instruction = "ycorrection.y * dFdy"; break;
2403 default: instruction = "";
2404 FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2405 break;
2406 }
2407
2408 write_mask = shader_glsl_append_dst(buffer, ins);
2409
2410 shader_addline(buffer, "%s(", instruction);
2411
2412 if (ins->src_count)
2413 {
2414 shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2415 shader_addline(buffer, "%s", src_param.param_str);
2416 for (i = 1; i < ins->src_count; ++i)
2417 {
2418 shader_glsl_add_src_param(ins, &ins->src[i], write_mask, &src_param);
2419 shader_addline(buffer, ", %s", src_param.param_str);
2420 }
2421 }
2422
2423 shader_addline(buffer, "));\n");
2424}
2425
2426static void shader_glsl_nrm(const struct wined3d_shader_instruction *ins)
2427{
2428 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2429 glsl_src_param_t src_param;
2430 unsigned int mask_size;
2431 DWORD write_mask;
2432 char dst_mask[6];
2433
2434 write_mask = shader_glsl_get_write_mask(ins->dst, dst_mask);
2435 mask_size = shader_glsl_get_write_mask_size(write_mask);
2436 shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2437
2438 shader_addline(buffer, "tmp0.x = length(%s);\n", src_param.param_str);
2439 shader_glsl_append_dst(buffer, ins);
2440 if (mask_size > 1)
2441 {
2442 shader_addline(buffer, "tmp0.x == 0.0 ? vec%u(0.0) : (%s / tmp0.x));\n",
2443 mask_size, src_param.param_str);
2444 }
2445 else
2446 {
2447 shader_addline(buffer, "tmp0.x == 0.0 ? 0.0 : (%s / tmp0.x));\n",
2448 src_param.param_str);
2449 }
2450}
2451
2452/** Process the WINED3DSIO_EXPP instruction in GLSL:
2453 * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
2454 * dst.x = 2^(floor(src))
2455 * dst.y = src - floor(src)
2456 * dst.z = 2^src (partial precision is allowed, but optional)
2457 * dst.w = 1.0;
2458 * For 2.0 shaders, just do this (honoring writemask and swizzle):
2459 * dst = 2^src; (partial precision is allowed, but optional)
2460 */
2461static void shader_glsl_expp(const struct wined3d_shader_instruction *ins)
2462{
2463 glsl_src_param_t src_param;
2464
2465 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src_param);
2466
2467 if (ins->ctx->reg_maps->shader_version.major < 2)
2468 {
2469 char dst_mask[6];
2470
2471 shader_addline(ins->ctx->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
2472 shader_addline(ins->ctx->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
2473 shader_addline(ins->ctx->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
2474 shader_addline(ins->ctx->buffer, "tmp0.w = 1.0;\n");
2475
2476 shader_glsl_append_dst(ins->ctx->buffer, ins);
2477 shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2478 shader_addline(ins->ctx->buffer, "tmp0%s);\n", dst_mask);
2479 } else {
2480 DWORD write_mask;
2481 unsigned int mask_size;
2482
2483 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2484 mask_size = shader_glsl_get_write_mask_size(write_mask);
2485
2486 if (mask_size > 1) {
2487 shader_addline(ins->ctx->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
2488 } else {
2489 shader_addline(ins->ctx->buffer, "exp2(%s));\n", src_param.param_str);
2490 }
2491 }
2492}
2493
2494/** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
2495static void shader_glsl_rcp(const struct wined3d_shader_instruction *ins)
2496{
2497 glsl_src_param_t src_param;
2498 DWORD write_mask;
2499 unsigned int mask_size;
2500
2501 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2502 mask_size = shader_glsl_get_write_mask_size(write_mask);
2503 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2504
2505 if (mask_size > 1)
2506 {
2507 shader_addline(ins->ctx->buffer, "vec%d(%s == 0.0 ? FLT_MAX : 1.0 / %s));\n",
2508 mask_size, src_param.param_str, src_param.param_str);
2509 }
2510 else
2511 {
2512 shader_addline(ins->ctx->buffer, "%s == 0.0 ? FLT_MAX : 1.0 / %s);\n",
2513 src_param.param_str, src_param.param_str);
2514 }
2515}
2516
2517static void shader_glsl_rsq(const struct wined3d_shader_instruction *ins)
2518{
2519 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2520 glsl_src_param_t src_param;
2521 DWORD write_mask;
2522 unsigned int mask_size;
2523
2524 write_mask = shader_glsl_append_dst(buffer, ins);
2525 mask_size = shader_glsl_get_write_mask_size(write_mask);
2526
2527 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2528
2529 if (mask_size > 1)
2530 {
2531 shader_addline(buffer, "vec%d(%s == 0.0 ? FLT_MAX : inversesqrt(abs(%s))));\n",
2532 mask_size, src_param.param_str, src_param.param_str);
2533 }
2534 else
2535 {
2536 shader_addline(buffer, "%s == 0.0 ? FLT_MAX : inversesqrt(abs(%s)));\n",
2537 src_param.param_str, src_param.param_str);
2538 }
2539}
2540
2541#ifdef VBOX_WITH_VMSVGA
2542static void shader_glsl_setp(const struct wined3d_shader_instruction *ins)
2543{
2544 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2545 glsl_src_param_t src_param1, src_param2;
2546 DWORD write_mask;
2547
2548 int i;
2549 DWORD dst_mask = ins->dst[0].write_mask;
2550 struct wined3d_shader_dst_param dst = ins->dst[0];
2551
2552 /* Cycle through all source0 channels */
2553 for (i=0; i<4; i++) {
2554 if (dst_mask & RT_BIT(i))
2555 {
2556 write_mask = WINED3DSP_WRITEMASK_0 << i;
2557 dst.write_mask = dst_mask & write_mask;
2558
2559 write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2560 Assert(write_mask);
2561
2562 shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param1);
2563 shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src_param2);
2564
2565 shader_addline(buffer, "%s %s %s);\n",
2566 src_param1.param_str, shader_get_comp_op(ins->flags), src_param2.param_str);
2567 }
2568 }
2569}
2570#endif
2571
2572/** Process signed comparison opcodes in GLSL. */
2573static void shader_glsl_compare(const struct wined3d_shader_instruction *ins)
2574{
2575 glsl_src_param_t src0_param;
2576 glsl_src_param_t src1_param;
2577 DWORD write_mask;
2578 unsigned int mask_size;
2579
2580 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2581 mask_size = shader_glsl_get_write_mask_size(write_mask);
2582 shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2583 shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2584
2585 if (mask_size > 1) {
2586 const char *compare;
2587
2588 switch(ins->handler_idx)
2589 {
2590 case WINED3DSIH_SLT: compare = "lessThan"; break;
2591 case WINED3DSIH_SGE: compare = "greaterThanEqual"; break;
2592 default: compare = "";
2593 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2594 }
2595
2596 shader_addline(ins->ctx->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
2597 src0_param.param_str, src1_param.param_str);
2598 } else {
2599 switch(ins->handler_idx)
2600 {
2601 case WINED3DSIH_SLT:
2602 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
2603 * to return 0.0 but step returns 1.0 because step is not < x
2604 * An alternative is a bvec compare padded with an unused second component.
2605 * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
2606 * issue. Playing with not() is not possible either because not() does not accept
2607 * a scalar.
2608 */
2609 shader_addline(ins->ctx->buffer, "(%s < %s) ? 1.0 : 0.0);\n",
2610 src0_param.param_str, src1_param.param_str);
2611 break;
2612 case WINED3DSIH_SGE:
2613 /* Here we can use the step() function and safe a conditional */
2614 shader_addline(ins->ctx->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
2615 break;
2616 default:
2617 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2618 }
2619
2620 }
2621}
2622
2623/** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
2624static void shader_glsl_cmp(const struct wined3d_shader_instruction *ins)
2625{
2626 glsl_src_param_t src0_param;
2627 glsl_src_param_t src1_param;
2628 glsl_src_param_t src2_param;
2629 DWORD write_mask, cmp_channel = 0;
2630 unsigned int i, j;
2631 char mask_char[6];
2632 BOOL temp_destination = FALSE;
2633
2634 if (shader_is_scalar(&ins->src[0].reg))
2635 {
2636 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2637
2638 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2639 shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2640 shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2641
2642 shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2643 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2644 } else {
2645 DWORD dst_mask = ins->dst[0].write_mask;
2646 struct wined3d_shader_dst_param dst = ins->dst[0];
2647
2648 /* Cycle through all source0 channels */
2649 for (i=0; i<4; i++) {
2650 write_mask = 0;
2651 /* Find the destination channels which use the current source0 channel */
2652 for (j=0; j<4; j++) {
2653 if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2654 {
2655 write_mask |= WINED3DSP_WRITEMASK_0 << j;
2656 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2657 }
2658 }
2659 dst.write_mask = dst_mask & write_mask;
2660
2661 /* Splitting the cmp instruction up in multiple lines imposes a problem:
2662 * The first lines may overwrite source parameters of the following lines.
2663 * Deal with that by using a temporary destination register if needed
2664 */
2665 if ((ins->src[0].reg.idx == ins->dst[0].reg.idx
2666 && ins->src[0].reg.type == ins->dst[0].reg.type)
2667 || (ins->src[1].reg.idx == ins->dst[0].reg.idx
2668 && ins->src[1].reg.type == ins->dst[0].reg.type)
2669 || (ins->src[2].reg.idx == ins->dst[0].reg.idx
2670 && ins->src[2].reg.type == ins->dst[0].reg.type))
2671 {
2672 write_mask = shader_glsl_get_write_mask(&dst, mask_char);
2673 if (!write_mask) continue;
2674 shader_addline(ins->ctx->buffer, "tmp0%s = (", mask_char);
2675 temp_destination = TRUE;
2676 } else {
2677 write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2678 if (!write_mask) continue;
2679 }
2680
2681 shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2682 shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2683 shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2684
2685 shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2686 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2687 }
2688
2689 if(temp_destination) {
2690 shader_glsl_get_write_mask(&ins->dst[0], mask_char);
2691 shader_glsl_append_dst(ins->ctx->buffer, ins);
2692 shader_addline(ins->ctx->buffer, "tmp0%s);\n", mask_char);
2693 }
2694 }
2695
2696}
2697
2698/** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
2699/* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
2700 * the compare is done per component of src0. */
2701static void shader_glsl_cnd(const struct wined3d_shader_instruction *ins)
2702{
2703 struct wined3d_shader_dst_param dst;
2704 glsl_src_param_t src0_param;
2705 glsl_src_param_t src1_param;
2706 glsl_src_param_t src2_param;
2707 DWORD write_mask, cmp_channel = 0;
2708 unsigned int i, j;
2709 DWORD dst_mask;
2710 DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2711 ins->ctx->reg_maps->shader_version.minor);
2712
2713 if (shader_version < WINED3D_SHADER_VERSION(1, 4))
2714 {
2715 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2716 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2717 shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2718 shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2719
2720 /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
2721 if (ins->coissue)
2722 {
2723 shader_addline(ins->ctx->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
2724 } else {
2725 shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2726 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2727 }
2728 return;
2729 }
2730 /* Cycle through all source0 channels */
2731 dst_mask = ins->dst[0].write_mask;
2732 dst = ins->dst[0];
2733 for (i=0; i<4; i++) {
2734 write_mask = 0;
2735 /* Find the destination channels which use the current source0 channel */
2736 for (j=0; j<4; j++) {
2737 if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2738 {
2739 write_mask |= WINED3DSP_WRITEMASK_0 << j;
2740 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2741 }
2742 }
2743
2744 dst.write_mask = dst_mask & write_mask;
2745 write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2746 if (!write_mask) continue;
2747
2748 shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2749 shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2750 shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2751
2752 shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2753 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2754 }
2755}
2756
2757/** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
2758static void shader_glsl_mad(const struct wined3d_shader_instruction *ins)
2759{
2760 glsl_src_param_t src0_param;
2761 glsl_src_param_t src1_param;
2762 glsl_src_param_t src2_param;
2763 DWORD write_mask;
2764
2765 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2766 shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2767 shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2768 shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2769 shader_addline(ins->ctx->buffer, "(%s * %s) + %s);\n",
2770 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2771}
2772
2773/* Handles transforming all WINED3DSIO_M?x? opcodes for
2774 Vertex shaders to GLSL codes */
2775static void shader_glsl_mnxn(const struct wined3d_shader_instruction *ins)
2776{
2777 int i;
2778 int nComponents = 0;
2779 struct wined3d_shader_dst_param tmp_dst = {{0}};
2780 struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2781 struct wined3d_shader_instruction tmp_ins;
2782
2783 memset(&tmp_ins, 0, sizeof(tmp_ins));
2784
2785 /* Set constants for the temporary argument */
2786 tmp_ins.ctx = ins->ctx;
2787 tmp_ins.dst_count = 1;
2788 tmp_ins.dst = &tmp_dst;
2789 tmp_ins.src_count = 2;
2790 tmp_ins.src = tmp_src;
2791
2792 switch(ins->handler_idx)
2793 {
2794 case WINED3DSIH_M4x4:
2795 nComponents = 4;
2796 tmp_ins.handler_idx = WINED3DSIH_DP4;
2797 break;
2798 case WINED3DSIH_M4x3:
2799 nComponents = 3;
2800 tmp_ins.handler_idx = WINED3DSIH_DP4;
2801 break;
2802 case WINED3DSIH_M3x4:
2803 nComponents = 4;
2804 tmp_ins.handler_idx = WINED3DSIH_DP3;
2805 break;
2806 case WINED3DSIH_M3x3:
2807 nComponents = 3;
2808 tmp_ins.handler_idx = WINED3DSIH_DP3;
2809 break;
2810 case WINED3DSIH_M3x2:
2811 nComponents = 2;
2812 tmp_ins.handler_idx = WINED3DSIH_DP3;
2813 break;
2814 default:
2815 break;
2816 }
2817
2818 tmp_dst = ins->dst[0];
2819 tmp_src[0] = ins->src[0];
2820 tmp_src[1] = ins->src[1];
2821 for (i = 0; i < nComponents; ++i)
2822 {
2823 tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2824 shader_glsl_dot(&tmp_ins);
2825 ++tmp_src[1].reg.idx;
2826 }
2827}
2828
2829/**
2830 The LRP instruction performs a component-wise linear interpolation
2831 between the second and third operands using the first operand as the
2832 blend factor. Equation: (dst = src2 + src0 * (src1 - src2))
2833 This is equivalent to mix(src2, src1, src0);
2834*/
2835static void shader_glsl_lrp(const struct wined3d_shader_instruction *ins)
2836{
2837 glsl_src_param_t src0_param;
2838 glsl_src_param_t src1_param;
2839 glsl_src_param_t src2_param;
2840 DWORD write_mask;
2841
2842 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2843
2844 shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2845 shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2846 shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2847
2848 shader_addline(ins->ctx->buffer, "mix(%s, %s, %s));\n",
2849 src2_param.param_str, src1_param.param_str, src0_param.param_str);
2850}
2851
2852/** Process the WINED3DSIO_LIT instruction in GLSL:
2853 * dst.x = dst.w = 1.0
2854 * dst.y = (src0.x > 0) ? src0.x
2855 * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
2856 * where src.w is clamped at +- 128
2857 */
2858static void shader_glsl_lit(const struct wined3d_shader_instruction *ins)
2859{
2860 glsl_src_param_t src0_param;
2861 glsl_src_param_t src1_param;
2862 glsl_src_param_t src3_param;
2863 char dst_mask[6];
2864
2865 shader_glsl_append_dst(ins->ctx->buffer, ins);
2866 shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2867
2868 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2869 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src1_param);
2870 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src3_param);
2871
2872 /* The sdk specifies the instruction like this
2873 * dst.x = 1.0;
2874 * if(src.x > 0.0) dst.y = src.x
2875 * else dst.y = 0.0.
2876 * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
2877 * else dst.z = 0.0;
2878 * dst.w = 1.0;
2879 *
2880 * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
2881 * dst.x = 1.0 ... No further explanation needed
2882 * dst.y = max(src.y, 0.0); ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
2883 * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0; ... 0 ^ power is 0, and otherwise we use y anyway
2884 * dst.w = 1.0. ... Nothing fancy.
2885 *
2886 * So we still have one conditional in there. So do this:
2887 * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
2888 *
2889 * step(0.0, x) will return 1 if src.x > 0.0, and 0 otherwise. So if y is 0 we get pow(0.0 * 1.0, power),
2890 * which sets dst.z to 0. If y > 0, but x = 0.0, we get pow(y * 0.0, power), which results in 0 too.
2891 * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to
2892 */
2893 shader_addline(ins->ctx->buffer,
2894 "vec4(1.0, max(%s, 0.0), pow(max(0.0, %s) * step(0.0, %s), clamp(%s, -128.0, 128.0)), 1.0)%s);\n",
2895 src0_param.param_str, src1_param.param_str, src0_param.param_str, src3_param.param_str, dst_mask);
2896}
2897
2898/** Process the WINED3DSIO_DST instruction in GLSL:
2899 * dst.x = 1.0
2900 * dst.y = src0.x * src0.y
2901 * dst.z = src0.z
2902 * dst.w = src1.w
2903 */
2904static void shader_glsl_dst(const struct wined3d_shader_instruction *ins)
2905{
2906 glsl_src_param_t src0y_param;
2907 glsl_src_param_t src0z_param;
2908 glsl_src_param_t src1y_param;
2909 glsl_src_param_t src1w_param;
2910 char dst_mask[6];
2911
2912 shader_glsl_append_dst(ins->ctx->buffer, ins);
2913 shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2914
2915 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src0y_param);
2916 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &src0z_param);
2917 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_1, &src1y_param);
2918 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_3, &src1w_param);
2919
2920 shader_addline(ins->ctx->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
2921 src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
2922}
2923
2924/** Process the WINED3DSIO_SINCOS instruction in GLSL:
2925 * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
2926 * can handle it. But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
2927 *
2928 * dst.x = cos(src0.?)
2929 * dst.y = sin(src0.?)
2930 * dst.z = dst.z
2931 * dst.w = dst.w
2932 */
2933static void shader_glsl_sincos(const struct wined3d_shader_instruction *ins)
2934{
2935 glsl_src_param_t src0_param;
2936 DWORD write_mask;
2937
2938 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2939 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2940
2941 switch (write_mask) {
2942 case WINED3DSP_WRITEMASK_0:
2943 shader_addline(ins->ctx->buffer, "cos(%s));\n", src0_param.param_str);
2944 break;
2945
2946 case WINED3DSP_WRITEMASK_1:
2947 shader_addline(ins->ctx->buffer, "sin(%s));\n", src0_param.param_str);
2948 break;
2949
2950 case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
2951 shader_addline(ins->ctx->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
2952 break;
2953
2954 default:
2955 ERR("Write mask should be .x, .y or .xy\n");
2956 break;
2957 }
2958}
2959
2960/* sgn in vs_2_0 has 2 extra parameters(registers for temporary storage) which we don't use
2961 * here. But those extra parameters require a dedicated function for sgn, since map2gl would
2962 * generate invalid code
2963 */
2964static void shader_glsl_sgn(const struct wined3d_shader_instruction *ins)
2965{
2966 glsl_src_param_t src0_param;
2967 DWORD write_mask;
2968
2969 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2970 shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2971
2972 shader_addline(ins->ctx->buffer, "sign(%s));\n", src0_param.param_str);
2973}
2974
2975/** Process the WINED3DSIO_LOOP instruction in GLSL:
2976 * Start a for() loop where src1.y is the initial value of aL,
2977 * increment aL by src1.z for a total of src1.x iterations.
2978 * Need to use a temporary variable for this operation.
2979 */
2980/* FIXME: I don't think nested loops will work correctly this way. */
2981static void shader_glsl_loop(const struct wined3d_shader_instruction *ins)
2982{
2983 glsl_src_param_t src1_param;
2984 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2985 const DWORD *control_values = NULL;
2986 const local_constant *constant;
2987
2988 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2989
2990 /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2991 * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2992 * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2993 * addressing.
2994 */
2995 if (ins->src[1].reg.type == WINED3DSPR_CONSTINT)
2996 {
2997 LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
2998 if (constant->idx == ins->src[1].reg.idx)
2999 {
3000 control_values = constant->value;
3001 break;
3002 }
3003 }
3004 }
3005
3006 if (control_values)
3007 {
3008 struct wined3d_shader_loop_control loop_control;
3009 loop_control.count = control_values[0];
3010 loop_control.start = control_values[1];
3011 loop_control.step = (int)control_values[2];
3012
3013 if (loop_control.step > 0)
3014 {
3015 shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u < (%u * %d + %u); aL%u += %d) {\n",
3016 shader->baseShader.cur_loop_depth, loop_control.start,
3017 shader->baseShader.cur_loop_depth, loop_control.count, loop_control.step, loop_control.start,
3018 shader->baseShader.cur_loop_depth, loop_control.step);
3019 }
3020 else if (loop_control.step < 0)
3021 {
3022 shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u > (%u * %d + %u); aL%u += %d) {\n",
3023 shader->baseShader.cur_loop_depth, loop_control.start,
3024 shader->baseShader.cur_loop_depth, loop_control.count, loop_control.step, loop_control.start,
3025 shader->baseShader.cur_loop_depth, loop_control.step);
3026 }
3027 else
3028 {
3029 shader_addline(ins->ctx->buffer, "for (aL%u = %u, tmpInt%u = 0; tmpInt%u < %u; tmpInt%u++) {\n",
3030 shader->baseShader.cur_loop_depth, loop_control.start, shader->baseShader.cur_loop_depth,
3031 shader->baseShader.cur_loop_depth, loop_control.count,
3032 shader->baseShader.cur_loop_depth);
3033 }
3034 } else {
3035 shader_addline(ins->ctx->buffer,
3036 "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
3037 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno,
3038 src1_param.reg_name, shader->baseShader.cur_loop_depth, src1_param.reg_name,
3039 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno, src1_param.reg_name);
3040 }
3041
3042 shader->baseShader.cur_loop_depth++;
3043 shader->baseShader.cur_loop_regno++;
3044}
3045
3046static void shader_glsl_end(const struct wined3d_shader_instruction *ins)
3047{
3048 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3049
3050 shader_addline(ins->ctx->buffer, "}\n");
3051
3052 if (ins->handler_idx == WINED3DSIH_ENDLOOP)
3053 {
3054 shader->baseShader.cur_loop_depth--;
3055 shader->baseShader.cur_loop_regno--;
3056 }
3057
3058 if (ins->handler_idx == WINED3DSIH_ENDREP)
3059 {
3060 shader->baseShader.cur_loop_depth--;
3061 }
3062}
3063
3064static void shader_glsl_rep(const struct wined3d_shader_instruction *ins)
3065{
3066 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3067 glsl_src_param_t src0_param;
3068 const DWORD *control_values = NULL;
3069 const local_constant *constant;
3070
3071 /* Try to hardcode local values to help the GLSL compiler to unroll and optimize the loop */
3072 if (ins->src[0].reg.type == WINED3DSPR_CONSTINT)
3073 {
3074 LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry)
3075 {
3076 if (constant->idx == ins->src[0].reg.idx)
3077 {
3078 control_values = constant->value;
3079 break;
3080 }
3081 }
3082 }
3083
3084 if(control_values) {
3085 shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %d; tmpInt%d++) {\n",
3086 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
3087 control_values[0], shader->baseShader.cur_loop_depth);
3088 } else {
3089 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
3090 shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
3091 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
3092 src0_param.param_str, shader->baseShader.cur_loop_depth);
3093 }
3094 shader->baseShader.cur_loop_depth++;
3095}
3096
3097static void shader_glsl_if(const struct wined3d_shader_instruction *ins)
3098{
3099 glsl_src_param_t src0_param;
3100
3101 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
3102 shader_addline(ins->ctx->buffer, "if (%s) {\n", src0_param.param_str);
3103}
3104
3105static void shader_glsl_ifc(const struct wined3d_shader_instruction *ins)
3106{
3107 glsl_src_param_t src0_param;
3108 glsl_src_param_t src1_param;
3109
3110 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
3111 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
3112
3113 shader_addline(ins->ctx->buffer, "if (%s %s %s) {\n",
3114 src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
3115}
3116
3117static void shader_glsl_else(const struct wined3d_shader_instruction *ins)
3118{
3119 shader_addline(ins->ctx->buffer, "} else {\n");
3120}
3121
3122static void shader_glsl_break(const struct wined3d_shader_instruction *ins)
3123{
3124 shader_addline(ins->ctx->buffer, "break;\n");
3125}
3126
3127/* FIXME: According to MSDN the compare is done per component. */
3128static void shader_glsl_breakc(const struct wined3d_shader_instruction *ins)
3129{
3130 glsl_src_param_t src0_param;
3131 glsl_src_param_t src1_param;
3132
3133 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
3134 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
3135
3136 shader_addline(ins->ctx->buffer, "if (%s %s %s) break;\n",
3137 src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
3138}
3139
3140static void shader_glsl_label(const struct wined3d_shader_instruction *ins)
3141{
3142 shader_addline(ins->ctx->buffer, "}\n");
3143 shader_addline(ins->ctx->buffer, "void subroutine%u () {\n", ins->src[0].reg.idx);
3144}
3145
3146static void shader_glsl_call(const struct wined3d_shader_instruction *ins)
3147{
3148 shader_addline(ins->ctx->buffer, "subroutine%u();\n", ins->src[0].reg.idx);
3149}
3150
3151static void shader_glsl_callnz(const struct wined3d_shader_instruction *ins)
3152{
3153 glsl_src_param_t src1_param;
3154
3155 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
3156 shader_addline(ins->ctx->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, ins->src[0].reg.idx);
3157}
3158
3159static void shader_glsl_ret(const struct wined3d_shader_instruction *ins)
3160{
3161 /* No-op. The closing } is written when a new function is started, and at the end of the shader. This
3162 * function only suppresses the unhandled instruction warning
3163 */
3164}
3165
3166/*********************************************
3167 * Pixel Shader Specific Code begins here
3168 ********************************************/
3169static void shader_glsl_tex(const struct wined3d_shader_instruction *ins)
3170{
3171 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3172 IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
3173 DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
3174 ins->ctx->reg_maps->shader_version.minor);
3175 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3176 glsl_sample_function_t sample_function;
3177 DWORD sample_flags = 0;
3178 WINED3DSAMPLER_TEXTURE_TYPE sampler_type;
3179 DWORD sampler_idx;
3180 DWORD mask = 0, swizzle;
3181
3182 /* 1.0-1.4: Use destination register as sampler source.
3183 * 2.0+: Use provided sampler source. */
3184 if (shader_version < WINED3D_SHADER_VERSION(2,0)) sampler_idx = ins->dst[0].reg.idx;
3185 else sampler_idx = ins->src[1].reg.idx;
3186 sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3187
3188 if (shader_version < WINED3D_SHADER_VERSION(1,4))
3189 {
3190 const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
3191 DWORD flags = (priv->cur_ps_args->tex_transform >> (sampler_idx * WINED3D_PSARGS_TEXTRANSFORM_SHIFT))
3192 & WINED3D_PSARGS_TEXTRANSFORM_MASK;
3193
3194 /* Projected cube textures don't make a lot of sense, the resulting coordinates stay the same. */
3195 if (flags & WINED3D_PSARGS_PROJECTED && sampler_type != WINED3DSTT_CUBE) {
3196 sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3197 switch (flags & ~WINED3D_PSARGS_PROJECTED) {
3198 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3199 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
3200 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
3201 case WINED3DTTFF_COUNT4:
3202 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
3203 }
3204 }
3205 }
3206 else if (shader_version < WINED3D_SHADER_VERSION(2,0))
3207 {
3208 DWORD src_mod = ins->src[0].modifiers;
3209
3210 if (src_mod == WINED3DSPSM_DZ) {
3211 sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3212 mask = WINED3DSP_WRITEMASK_2;
3213 } else if (src_mod == WINED3DSPSM_DW) {
3214 sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3215 mask = WINED3DSP_WRITEMASK_3;
3216 }
3217 } else {
3218 if (ins->flags & WINED3DSI_TEXLD_PROJECT)
3219 {
3220 /* ps 2.0 texldp instruction always divides by the fourth component. */
3221 sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3222 mask = WINED3DSP_WRITEMASK_3;
3223 }
3224 }
3225
3226 if(deviceImpl->stateBlock->textures[sampler_idx] &&
3227 IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
3228 sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3229 }
3230
3231 shader_glsl_get_sample_function(gl_info, sampler_type, sample_flags, &sample_function);
3232 mask |= sample_function.coord_mask;
3233
3234 if (shader_version < WINED3D_SHADER_VERSION(2,0)) swizzle = WINED3DSP_NOSWIZZLE;
3235 else swizzle = ins->src[1].swizzle;
3236
3237 /* 1.0-1.3: Use destination register as coordinate source.
3238 1.4+: Use provided coordinate source register. */
3239 if (shader_version < WINED3D_SHADER_VERSION(1,4))
3240 {
3241 char coord_mask[6];
3242 shader_glsl_write_mask_to_str(mask, coord_mask);
3243 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3244 "T%u%s", sampler_idx, coord_mask);
3245 } else {
3246 glsl_src_param_t coord_param;
3247 shader_glsl_add_src_param(ins, &ins->src[0], mask, &coord_param);
3248 if (ins->flags & WINED3DSI_TEXLD_BIAS)
3249 {
3250 glsl_src_param_t bias;
3251 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &bias);
3252 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, bias.param_str,
3253 "%s", coord_param.param_str);
3254 } else {
3255 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3256 "%s", coord_param.param_str);
3257 }
3258 }
3259}
3260
3261static void shader_glsl_texldd(const struct wined3d_shader_instruction *ins)
3262{
3263 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3264 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
3265 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3266 glsl_sample_function_t sample_function;
3267 glsl_src_param_t coord_param, dx_param, dy_param;
3268 DWORD sample_flags = WINED3D_GLSL_SAMPLE_GRAD;
3269 DWORD sampler_type;
3270 DWORD sampler_idx;
3271 DWORD swizzle = ins->src[1].swizzle;
3272
3273 if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4])
3274 {
3275 FIXME("texldd used, but not supported by hardware. Falling back to regular tex\n");
3276 shader_glsl_tex(ins);
3277 return;
3278 }
3279
3280 sampler_idx = ins->src[1].reg.idx;
3281 sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3282 if(deviceImpl->stateBlock->textures[sampler_idx] &&
3283 IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
3284 sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3285 }
3286
3287 shader_glsl_get_sample_function(gl_info, sampler_type, sample_flags, &sample_function);
3288 shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3289 shader_glsl_add_src_param(ins, &ins->src[2], sample_function.coord_mask, &dx_param);
3290 shader_glsl_add_src_param(ins, &ins->src[3], sample_function.coord_mask, &dy_param);
3291
3292 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, dx_param.param_str, dy_param.param_str, NULL,
3293 "%s", coord_param.param_str);
3294}
3295
3296static void shader_glsl_texldl(const struct wined3d_shader_instruction *ins)
3297{
3298 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3299 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
3300 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3301 glsl_sample_function_t sample_function;
3302 glsl_src_param_t coord_param, lod_param;
3303 DWORD sample_flags = WINED3D_GLSL_SAMPLE_LOD;
3304 DWORD sampler_type;
3305 DWORD sampler_idx;
3306 DWORD swizzle = ins->src[1].swizzle;
3307
3308 sampler_idx = ins->src[1].reg.idx;
3309 sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3310 if(deviceImpl->stateBlock->textures[sampler_idx] &&
3311 IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
3312 sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3313 }
3314 shader_glsl_get_sample_function(gl_info, sampler_type, sample_flags, &sample_function);
3315 shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3316
3317 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &lod_param);
3318
3319 if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4]
3320 && shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
3321 {
3322 /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
3323 * However, they seem to work just fine in fragment shaders as well. */
3324 WARN("Using %s in fragment shader.\n", sample_function.name);
3325 }
3326 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, lod_param.param_str,
3327 "%s", coord_param.param_str);
3328}
3329
3330static void shader_glsl_texcoord(const struct wined3d_shader_instruction *ins)
3331{
3332 /* FIXME: Make this work for more than just 2D textures */
3333 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3334 DWORD write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3335
3336 if (!(ins->ctx->reg_maps->shader_version.major == 1 && ins->ctx->reg_maps->shader_version.minor == 4))
3337 {
3338 char dst_mask[6];
3339
3340 shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3341 shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n",
3342 ins->dst[0].reg.idx, dst_mask);
3343 } else {
3344 DWORD reg = ins->src[0].reg.idx;
3345 DWORD src_mod = ins->src[0].modifiers;
3346 char dst_swizzle[6];
3347
3348 shader_glsl_get_swizzle(&ins->src[0], FALSE, write_mask, dst_swizzle);
3349
3350 if (src_mod == WINED3DSPSM_DZ) {
3351 glsl_src_param_t div_param;
3352 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3353 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &div_param);
3354
3355 if (mask_size > 1) {
3356 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3357 } else {
3358 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3359 }
3360 } else if (src_mod == WINED3DSPSM_DW) {
3361 glsl_src_param_t div_param;
3362 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3363 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &div_param);
3364
3365 if (mask_size > 1) {
3366 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3367 } else {
3368 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3369 }
3370 } else {
3371 shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
3372 }
3373 }
3374}
3375
3376/** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
3377 * Take a 3-component dot product of the TexCoord[dstreg] and src,
3378 * then perform a 1D texture lookup from stage dstregnum, place into dst. */
3379static void shader_glsl_texdp3tex(const struct wined3d_shader_instruction *ins)
3380{
3381 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3382 glsl_src_param_t src0_param;
3383 glsl_sample_function_t sample_function;
3384 DWORD sampler_idx = ins->dst[0].reg.idx;
3385 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3386 WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3387 UINT mask_size;
3388
3389 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3390
3391 /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
3392 * scalar, and projected sampling would require 4.
3393 *
3394 * It is a dependent read - not valid with conditional NP2 textures
3395 */
3396 shader_glsl_get_sample_function(gl_info, sampler_type, 0, &sample_function);
3397 mask_size = shader_glsl_get_write_mask_size(sample_function.coord_mask);
3398
3399 switch(mask_size)
3400 {
3401 case 1:
3402 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3403 "dot(gl_TexCoord[%u].xyz, %s)", sampler_idx, src0_param.param_str);
3404 break;
3405
3406 case 2:
3407 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3408 "vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0)", sampler_idx, src0_param.param_str);
3409 break;
3410
3411 case 3:
3412 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3413 "vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0)", sampler_idx, src0_param.param_str);
3414 break;
3415
3416 default:
3417 FIXME("Unexpected mask size %u\n", mask_size);
3418 break;
3419 }
3420}
3421
3422/** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
3423 * Take a 3-component dot product of the TexCoord[dstreg] and src. */
3424static void shader_glsl_texdp3(const struct wined3d_shader_instruction *ins)
3425{
3426 glsl_src_param_t src0_param;
3427 DWORD dstreg = ins->dst[0].reg.idx;
3428 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3429 DWORD dst_mask;
3430 unsigned int mask_size;
3431
3432 dst_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3433 mask_size = shader_glsl_get_write_mask_size(dst_mask);
3434 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3435
3436 if (mask_size > 1) {
3437 shader_addline(ins->ctx->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
3438 } else {
3439 shader_addline(ins->ctx->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
3440 }
3441}
3442
3443/** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
3444 * Calculate the depth as dst.x / dst.y */
3445static void shader_glsl_texdepth(const struct wined3d_shader_instruction *ins)
3446{
3447 glsl_dst_param_t dst_param;
3448
3449 shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3450
3451 /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
3452 * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
3453 * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
3454 * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
3455 * >= 1.0 or < 0.0
3456 */
3457 shader_addline(ins->ctx->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n",
3458 dst_param.reg_name, dst_param.reg_name);
3459}
3460
3461/** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
3462 * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
3463 * Calculate tmp0.y = TexCoord[dstreg] . src.xyz; (tmp0.x has already been calculated)
3464 * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
3465 */
3466static void shader_glsl_texm3x2depth(const struct wined3d_shader_instruction *ins)
3467{
3468 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3469 DWORD dstreg = ins->dst[0].reg.idx;
3470 glsl_src_param_t src0_param;
3471
3472 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3473
3474 shader_addline(ins->ctx->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
3475 shader_addline(ins->ctx->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
3476}
3477
3478/** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
3479 * Calculate the 1st of a 2-row matrix multiplication. */
3480static void shader_glsl_texm3x2pad(const struct wined3d_shader_instruction *ins)
3481{
3482 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3483 DWORD reg = ins->dst[0].reg.idx;
3484 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3485 glsl_src_param_t src0_param;
3486
3487 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3488 shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3489}
3490
3491/** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
3492 * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
3493static void shader_glsl_texm3x3pad(const struct wined3d_shader_instruction *ins)
3494{
3495 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3496 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3497 DWORD reg = ins->dst[0].reg.idx;
3498 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3499 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3500 glsl_src_param_t src0_param;
3501
3502 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3503 shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + current_state->current_row, reg, src0_param.param_str);
3504 current_state->texcoord_w[current_state->current_row++] = reg;
3505}
3506
3507static void shader_glsl_texm3x2tex(const struct wined3d_shader_instruction *ins)
3508{
3509 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3510 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3511 DWORD reg = ins->dst[0].reg.idx;
3512 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3513 glsl_src_param_t src0_param;
3514 WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[reg];
3515 glsl_sample_function_t sample_function;
3516
3517 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3518 shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3519
3520 shader_glsl_get_sample_function(gl_info, sampler_type, 0, &sample_function);
3521
3522 /* Sample the texture using the calculated coordinates */
3523 shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xy");
3524}
3525
3526/** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
3527 * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
3528static void shader_glsl_texm3x3tex(const struct wined3d_shader_instruction *ins)
3529{
3530 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3531 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3532 SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
3533 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3534 glsl_src_param_t src0_param;
3535 DWORD reg = ins->dst[0].reg.idx;
3536 WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[reg];
3537 glsl_sample_function_t sample_function;
3538
3539 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3540 shader_addline(ins->ctx->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3541
3542 /* Dependent read, not valid with conditional NP2 */
3543 shader_glsl_get_sample_function(gl_info, sampler_type, 0, &sample_function);
3544
3545 /* Sample the texture using the calculated coordinates */
3546 shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3547
3548 current_state->current_row = 0;
3549}
3550
3551/** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
3552 * Perform the 3rd row of a 3x3 matrix multiply */
3553static void shader_glsl_texm3x3(const struct wined3d_shader_instruction *ins)
3554{
3555 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3556 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3557 SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
3558 glsl_src_param_t src0_param;
3559 char dst_mask[6];
3560 DWORD reg = ins->dst[0].reg.idx;
3561
3562 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3563
3564 shader_glsl_append_dst(ins->ctx->buffer, ins);
3565 shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3566 shader_addline(ins->ctx->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
3567
3568 current_state->current_row = 0;
3569}
3570
3571/* Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
3572 * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3573static void shader_glsl_texm3x3spec(const struct wined3d_shader_instruction *ins)
3574{
3575 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3576 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3577 DWORD reg = ins->dst[0].reg.idx;
3578 glsl_src_param_t src0_param;
3579 glsl_src_param_t src1_param;
3580 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3581 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3582 WINED3DSAMPLER_TEXTURE_TYPE stype = ins->ctx->reg_maps->sampler_type[reg];
3583 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3584 glsl_sample_function_t sample_function;
3585
3586 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3587 shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
3588
3589 /* Perform the last matrix multiply operation */
3590 shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3591 /* Reflection calculation */
3592 shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
3593
3594 /* Dependent read, not valid with conditional NP2 */
3595 shader_glsl_get_sample_function(gl_info, stype, 0, &sample_function);
3596
3597 /* Sample the texture */
3598 shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3599
3600 current_state->current_row = 0;
3601}
3602
3603/* Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
3604 * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3605static void shader_glsl_texm3x3vspec(const struct wined3d_shader_instruction *ins)
3606{
3607 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3608 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3609 DWORD reg = ins->dst[0].reg.idx;
3610 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3611 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3612 glsl_src_param_t src0_param;
3613 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3614 WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[reg];
3615 glsl_sample_function_t sample_function;
3616
3617 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3618
3619 /* Perform the last matrix multiply operation */
3620 shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
3621
3622 /* Construct the eye-ray vector from w coordinates */
3623 shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
3624 current_state->texcoord_w[0], current_state->texcoord_w[1], reg);
3625 shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
3626
3627 /* Dependent read, not valid with conditional NP2 */
3628 shader_glsl_get_sample_function(gl_info, sampler_type, 0, &sample_function);
3629
3630 /* Sample the texture using the calculated coordinates */
3631 shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3632
3633 current_state->current_row = 0;
3634}
3635
3636/** Process the WINED3DSIO_TEXBEM instruction in GLSL.
3637 * Apply a fake bump map transform.
3638 * texbem is pshader <= 1.3 only, this saves a few version checks
3639 */
3640static void shader_glsl_texbem(const struct wined3d_shader_instruction *ins)
3641{
3642 /*IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3643 IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device; - unused */
3644 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3645 const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
3646 glsl_sample_function_t sample_function;
3647 glsl_src_param_t coord_param;
3648 WINED3DSAMPLER_TEXTURE_TYPE sampler_type;
3649 DWORD sampler_idx;
3650 DWORD mask;
3651 DWORD flags;
3652 char coord_mask[6];
3653
3654 sampler_idx = ins->dst[0].reg.idx;
3655 flags = (priv->cur_ps_args->tex_transform >> (sampler_idx * WINED3D_PSARGS_TEXTRANSFORM_SHIFT))
3656 & WINED3D_PSARGS_TEXTRANSFORM_MASK;
3657
3658 sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3659 /* Dependent read, not valid with conditional NP2 */
3660 shader_glsl_get_sample_function(gl_info, sampler_type, 0, &sample_function);
3661 mask = sample_function.coord_mask;
3662
3663 shader_glsl_write_mask_to_str(mask, coord_mask);
3664
3665 /* with projective textures, texbem only divides the static texture coord, not the displacement,
3666 * so we can't let the GL handle this.
3667 */
3668 if (flags & WINED3D_PSARGS_PROJECTED) {
3669 DWORD div_mask=0;
3670 char coord_div_mask[3];
3671 switch (flags & ~WINED3D_PSARGS_PROJECTED) {
3672 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3673 case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
3674 case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
3675 case WINED3DTTFF_COUNT4:
3676 case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
3677 }
3678 shader_glsl_write_mask_to_str(div_mask, coord_div_mask);
3679 shader_addline(ins->ctx->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
3680 }
3681
3682 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &coord_param);
3683
3684 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3685 "T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s", sampler_idx, coord_mask, sampler_idx,
3686 coord_param.param_str, coord_mask);
3687
3688 if (ins->handler_idx == WINED3DSIH_TEXBEML)
3689 {
3690 glsl_src_param_t luminance_param;
3691 glsl_dst_param_t dst_param;
3692
3693 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &luminance_param);
3694 shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3695
3696 shader_addline(ins->ctx->buffer, "%s%s *= (%s * luminancescale%d + luminanceoffset%d);\n",
3697 dst_param.reg_name, dst_param.mask_str,
3698 luminance_param.param_str, sampler_idx, sampler_idx);
3699 }
3700}
3701
3702static void shader_glsl_bem(const struct wined3d_shader_instruction *ins)
3703{
3704 glsl_src_param_t src0_param, src1_param;
3705 DWORD sampler_idx = ins->dst[0].reg.idx;
3706
3707 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3708 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3709
3710 shader_glsl_append_dst(ins->ctx->buffer, ins);
3711 shader_addline(ins->ctx->buffer, "%s + bumpenvmat%d * %s);\n",
3712 src0_param.param_str, sampler_idx, src1_param.param_str);
3713}
3714
3715/** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
3716 * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
3717static void shader_glsl_texreg2ar(const struct wined3d_shader_instruction *ins)
3718{
3719 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3720 glsl_src_param_t src0_param;
3721 DWORD sampler_idx = ins->dst[0].reg.idx;
3722 WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3723 glsl_sample_function_t sample_function;
3724
3725 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3726
3727 shader_glsl_get_sample_function(gl_info, sampler_type, 0, &sample_function);
3728 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3729 "%s.wx", src0_param.reg_name);
3730}
3731
3732/** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
3733 * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
3734static void shader_glsl_texreg2gb(const struct wined3d_shader_instruction *ins)
3735{
3736 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3737 glsl_src_param_t src0_param;
3738 DWORD sampler_idx = ins->dst[0].reg.idx;
3739 WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3740 glsl_sample_function_t sample_function;
3741
3742 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3743
3744 shader_glsl_get_sample_function(gl_info, sampler_type, 0, &sample_function);
3745 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3746 "%s.yz", src0_param.reg_name);
3747}
3748
3749/** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
3750 * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
3751static void shader_glsl_texreg2rgb(const struct wined3d_shader_instruction *ins)
3752{
3753 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3754 glsl_src_param_t src0_param;
3755 DWORD sampler_idx = ins->dst[0].reg.idx;
3756 WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3757 glsl_sample_function_t sample_function;
3758
3759 /* Dependent read, not valid with conditional NP2 */
3760 shader_glsl_get_sample_function(gl_info, sampler_type, 0, &sample_function);
3761 shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &src0_param);
3762
3763 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3764 "%s", src0_param.param_str);
3765}
3766
3767/** Process the WINED3DSIO_TEXKILL instruction in GLSL.
3768 * If any of the first 3 components are < 0, discard this pixel */
3769static void shader_glsl_texkill(const struct wined3d_shader_instruction *ins)
3770{
3771 glsl_dst_param_t dst_param;
3772
3773 /* The argument is a destination parameter, and no writemasks are allowed */
3774 shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3775 if (ins->ctx->reg_maps->shader_version.major >= 2)
3776 {
3777 /* 2.0 shaders compare all 4 components in texkill */
3778 shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
3779 } else {
3780 /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
3781 * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
3782 * 4 components are defined, only the first 3 are used
3783 */
3784 shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
3785 }
3786}
3787
3788/** Process the WINED3DSIO_DP2ADD instruction in GLSL.
3789 * dst = dot2(src0, src1) + src2 */
3790static void shader_glsl_dp2add(const struct wined3d_shader_instruction *ins)
3791{
3792 glsl_src_param_t src0_param;
3793 glsl_src_param_t src1_param;
3794 glsl_src_param_t src2_param;
3795 DWORD write_mask;
3796 unsigned int mask_size;
3797
3798 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3799 mask_size = shader_glsl_get_write_mask_size(write_mask);
3800
3801 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3802 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3803 shader_glsl_add_src_param(ins, &ins->src[2], WINED3DSP_WRITEMASK_0, &src2_param);
3804
3805 if (mask_size > 1) {
3806 shader_addline(ins->ctx->buffer, "vec%d(dot(%s, %s) + %s));\n",
3807 mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
3808 } else {
3809 shader_addline(ins->ctx->buffer, "dot(%s, %s) + %s);\n",
3810 src0_param.param_str, src1_param.param_str, src2_param.param_str);
3811 }
3812}
3813
3814static void shader_glsl_input_pack(IWineD3DPixelShader *iface, struct wined3d_shader_buffer *buffer,
3815 const struct wined3d_shader_signature_element *input_signature, const struct shader_reg_maps *reg_maps,
3816 enum vertexprocessing_mode vertexprocessing)
3817{
3818 unsigned int i;
3819 IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
3820 WORD map = reg_maps->input_registers;
3821
3822 for (i = 0; map; map >>= 1, ++i)
3823 {
3824 const char *semantic_name;
3825 UINT semantic_idx;
3826 char reg_mask[6];
3827
3828 /* Unused */
3829 if (!(map & 1)) continue;
3830
3831 semantic_name = input_signature[i].semantic_name;
3832 semantic_idx = input_signature[i].semantic_idx;
3833 shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3834
3835 if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3836 {
3837 if (semantic_idx < 8 && vertexprocessing == pretransformed)
3838 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
3839 This->input_reg_map[i], reg_mask, semantic_idx, reg_mask);
3840 else
3841 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3842 This->input_reg_map[i], reg_mask, reg_mask);
3843 }
3844 else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3845 {
3846 if (semantic_idx == 0)
3847 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
3848 This->input_reg_map[i], reg_mask, reg_mask);
3849 else if (semantic_idx == 1)
3850 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
3851 This->input_reg_map[i], reg_mask, reg_mask);
3852 else
3853 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3854 This->input_reg_map[i], reg_mask, reg_mask);
3855 }
3856 else
3857 {
3858 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3859 This->input_reg_map[i], reg_mask, reg_mask);
3860 }
3861 }
3862}
3863
3864/*********************************************
3865 * Vertex Shader Specific Code begins here
3866 ********************************************/
3867
3868static void add_glsl_program_entry(struct shader_glsl_priv *priv, struct glsl_shader_prog_link *entry) {
3869 glsl_program_key_t key;
3870
3871 key.vshader = entry->vshader;
3872 key.pshader = entry->pshader;
3873 key.vs_args = entry->vs_args;
3874 key.ps_args = entry->ps_args;
3875 key.context = entry->context;
3876
3877 if (wine_rb_put(&priv->program_lookup, &key, &entry->program_lookup_entry) == -1)
3878 {
3879 ERR("Failed to insert program entry.\n");
3880 }
3881}
3882
3883static struct glsl_shader_prog_link *get_glsl_program_entry(struct shader_glsl_priv *priv,
3884 IWineD3DVertexShader *vshader, IWineD3DPixelShader *pshader, struct vs_compile_args *vs_args,
3885 struct ps_compile_args *ps_args, const struct wined3d_context *context) {
3886 struct wine_rb_entry *entry;
3887 glsl_program_key_t key;
3888
3889 key.vshader = vshader;
3890 key.pshader = pshader;
3891 key.vs_args = *vs_args;
3892 key.ps_args = *ps_args;
3893 key.context = context;
3894
3895 entry = wine_rb_get(&priv->program_lookup, &key);
3896 return entry ? WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry) : NULL;
3897}
3898
3899/* GL locking is done by the caller */
3900static void delete_glsl_program_entry(struct shader_glsl_priv *priv, const struct wined3d_gl_info *gl_info,
3901 struct glsl_shader_prog_link *entry)
3902{
3903 glsl_program_key_t key;
3904
3905 key.vshader = entry->vshader;
3906 key.pshader = entry->pshader;
3907 key.vs_args = entry->vs_args;
3908 key.ps_args = entry->ps_args;
3909 key.context = entry->context;
3910 wine_rb_remove(&priv->program_lookup, &key);
3911
3912 if (context_get_current() == entry->context)
3913 {
3914 TRACE("deleting program %p\n", (void *)(uintptr_t)entry->programId);
3915 GL_EXTCALL(glDeleteObjectARB(entry->programId));
3916 checkGLcall("glDeleteObjectARB");
3917 }
3918 else
3919 {
3920 WARN("Attempting to delete program %p created in ctx %p from ctx %p\n", (void *)(uintptr_t)entry->programId, entry->context, context_get_current());
3921 }
3922
3923 if (entry->vshader) list_remove(&entry->vshader_entry);
3924 if (entry->pshader) list_remove(&entry->pshader_entry);
3925 HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
3926 HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
3927 HeapFree(GetProcessHeap(), 0, entry);
3928}
3929
3930static void handle_ps3_input(struct wined3d_shader_buffer *buffer, const struct wined3d_gl_info *gl_info, const DWORD *map,
3931 const struct wined3d_shader_signature_element *input_signature, const struct shader_reg_maps *reg_maps_in,
3932 const struct wined3d_shader_signature_element *output_signature, const struct shader_reg_maps *reg_maps_out)
3933{
3934 unsigned int i, j;
3935 const char *semantic_name_in, *semantic_name_out;
3936 UINT semantic_idx_in, semantic_idx_out;
3937 DWORD *set;
3938 DWORD in_idx;
3939 unsigned int in_count = vec4_varyings(3, gl_info);
3940 char reg_mask[6], reg_mask_out[6];
3941 char destination[50];
3942 WORD input_map, output_map;
3943
3944 set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
3945
3946 if (!output_signature)
3947 {
3948 /* Save gl_FrontColor & gl_FrontSecondaryColor before overwriting them. */
3949 shader_addline(buffer, "vec4 front_color = gl_FrontColor;\n");
3950 shader_addline(buffer, "vec4 front_secondary_color = gl_FrontSecondaryColor;\n");
3951 }
3952
3953 input_map = reg_maps_in->input_registers;
3954 for (i = 0; input_map; input_map >>= 1, ++i)
3955 {
3956 if (!(input_map & 1)) continue;
3957
3958 in_idx = map[i];
3959 if (in_idx >= (in_count + 2)) {
3960 FIXME("More input varyings declared than supported, expect issues\n");
3961 continue;
3962 }
3963 else if (map[i] == ~0U)
3964 {
3965 /* Declared, but not read register */
3966 continue;
3967 }
3968
3969 if (in_idx == in_count) {
3970 sprintf(destination, "gl_FrontColor");
3971 } else if (in_idx == in_count + 1) {
3972 sprintf(destination, "gl_FrontSecondaryColor");
3973 } else {
3974 sprintf(destination, "IN[%u]", in_idx);
3975 }
3976
3977 semantic_name_in = input_signature[i].semantic_name;
3978 semantic_idx_in = input_signature[i].semantic_idx;
3979 set[map[i]] = input_signature[i].mask;
3980 shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3981
3982 if (!output_signature)
3983 {
3984 if (shader_match_semantic(semantic_name_in, WINED3DDECLUSAGE_COLOR))
3985 {
3986 if (semantic_idx_in == 0)
3987 shader_addline(buffer, "%s%s = front_color%s;\n",
3988 destination, reg_mask, reg_mask);
3989 else if (semantic_idx_in == 1)
3990 shader_addline(buffer, "%s%s = front_secondary_color%s;\n",
3991 destination, reg_mask, reg_mask);
3992 else
3993 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3994 destination, reg_mask, reg_mask);
3995 }
3996 else if (shader_match_semantic(semantic_name_in, WINED3DDECLUSAGE_TEXCOORD))
3997 {
3998 if (semantic_idx_in < 8)
3999 {
4000 shader_addline(buffer, "%s%s = gl_TexCoord[%u]%s;\n",
4001 destination, reg_mask, semantic_idx_in, reg_mask);
4002 }
4003 else
4004 {
4005 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
4006 destination, reg_mask, reg_mask);
4007 }
4008 }
4009 else if (shader_match_semantic(semantic_name_in, WINED3DDECLUSAGE_FOG))
4010 {
4011 shader_addline(buffer, "%s%s = vec4(gl_FogFragCoord, 0.0, 0.0, 0.0)%s;\n",
4012 destination, reg_mask, reg_mask);
4013 }
4014 else
4015 {
4016 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
4017 destination, reg_mask, reg_mask);
4018 }
4019 } else {
4020 BOOL found = FALSE;
4021
4022 output_map = reg_maps_out->output_registers;
4023 for (j = 0; output_map; output_map >>= 1, ++j)
4024 {
4025 if (!(output_map & 1)) continue;
4026
4027 semantic_name_out = output_signature[j].semantic_name;
4028 semantic_idx_out = output_signature[j].semantic_idx;
4029 shader_glsl_write_mask_to_str(output_signature[j].mask, reg_mask_out);
4030
4031 if (semantic_idx_in == semantic_idx_out
4032 && !strcmp(semantic_name_in, semantic_name_out))
4033 {
4034 shader_addline(buffer, "%s%s = OUT[%u]%s;\n",
4035 destination, reg_mask, j, reg_mask);
4036 found = TRUE;
4037 }
4038 }
4039 if(!found) {
4040 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
4041 destination, reg_mask, reg_mask);
4042 }
4043 }
4044 }
4045
4046 /* This is solely to make the compiler / linker happy and avoid warning about undefined
4047 * varyings. It shouldn't result in any real code executed on the GPU, since all read
4048 * input varyings are assigned above, if the optimizer works properly.
4049 */
4050 for(i = 0; i < in_count + 2; i++) {
4051 if (set[i] && set[i] != WINED3DSP_WRITEMASK_ALL)
4052 {
4053 unsigned int size = 0;
4054 memset(reg_mask, 0, sizeof(reg_mask));
4055 if(!(set[i] & WINED3DSP_WRITEMASK_0)) {
4056 reg_mask[size] = 'x';
4057 size++;
4058 }
4059 if(!(set[i] & WINED3DSP_WRITEMASK_1)) {
4060 reg_mask[size] = 'y';
4061 size++;
4062 }
4063 if(!(set[i] & WINED3DSP_WRITEMASK_2)) {
4064 reg_mask[size] = 'z';
4065 size++;
4066 }
4067 if(!(set[i] & WINED3DSP_WRITEMASK_3)) {
4068 reg_mask[size] = 'w';
4069 size++;
4070 }
4071
4072 if (i == in_count) {
4073 sprintf(destination, "gl_FrontColor");
4074 } else if (i == in_count + 1) {
4075 sprintf(destination, "gl_FrontSecondaryColor");
4076 } else {
4077 sprintf(destination, "IN[%u]", i);
4078 }
4079
4080 if (size == 1) {
4081 shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
4082 } else {
4083 shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
4084 }
4085 }
4086 }
4087
4088 HeapFree(GetProcessHeap(), 0, set);
4089}
4090
4091static void generate_texcoord_assignment(struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *vs, IWineD3DPixelShaderImpl *ps)
4092{
4093 DWORD map;
4094 unsigned int i;
4095 char reg_mask[6];
4096
4097 if (!ps)
4098 return;
4099
4100 for (i = 0, map = ps->baseShader.reg_maps.texcoord; map && i < min(8, MAX_REG_TEXCRD); map >>= 1, ++i)
4101 {
4102 if (!(map & 1))
4103 continue;
4104
4105 /* so far we assume that if texcoord_mask has any write flags, they are assigned appropriately with pixel shader */
4106 if ((vs->baseShader.reg_maps.texcoord_mask[i]) & WINED3DSP_WRITEMASK_ALL)
4107 continue;
4108
4109 shader_glsl_write_mask_to_str(WINED3DSP_WRITEMASK_ALL, reg_mask);
4110 shader_addline(buffer, "gl_TexCoord[%u]%s = gl_MultiTexCoord%u%s;\n", i, reg_mask, i, reg_mask);
4111 }
4112}
4113
4114/* GL locking is done by the caller */
4115static GLhandleARB generate_param_reorder_function(struct wined3d_shader_buffer *buffer,
4116 IWineD3DVertexShader *a_vertexshader, IWineD3DPixelShader *pixelshader, const struct wined3d_gl_info *gl_info)
4117{
4118 GLhandleARB ret = 0;
4119 IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) a_vertexshader;
4120 IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pixelshader;
4121 IWineD3DDeviceImpl *device;
4122 DWORD vs_major = vs->baseShader.reg_maps.shader_version.major;
4123 DWORD ps_major = ps ? ps->baseShader.reg_maps.shader_version.major : 0;
4124 unsigned int i;
4125 const char *semantic_name;
4126 UINT semantic_idx;
4127 char reg_mask[6];
4128 const struct wined3d_shader_signature_element *output_signature;
4129
4130 shader_buffer_clear(buffer);
4131
4132 shader_addline(buffer, "#version 120\n");
4133
4134 if(vs_major < 3 && ps_major < 3) {
4135 /* That one is easy: The vertex shader writes to the builtin varyings, the pixel shader reads from them.
4136 * Take care about the texcoord .w fixup though if we're using the fixed function fragment pipeline
4137 */
4138 device = (IWineD3DDeviceImpl *) vs->baseShader.device;
4139 if ((gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W)
4140 && ps_major == 0 && vs_major > 0 && !device->frag_pipe->ffp_proj_control)
4141 {
4142 shader_addline(buffer, "void order_ps_input() {\n");
4143 for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
4144 if(vs->baseShader.reg_maps.texcoord_mask[i] != 0 &&
4145 vs->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
4146 shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", i);
4147 }
4148 }
4149 shader_addline(buffer, "}\n");
4150 } else {
4151 shader_addline(buffer, "void order_ps_input() {\n");
4152 generate_texcoord_assignment(buffer, vs, ps);
4153 shader_addline(buffer, "}\n");
4154 }
4155 } else if(ps_major < 3 && vs_major >= 3) {
4156 WORD map = vs->baseShader.reg_maps.output_registers;
4157
4158 /* The vertex shader writes to its own varyings, the pixel shader needs them in the builtin ones */
4159 output_signature = vs->baseShader.output_signature;
4160
4161 shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
4162 for (i = 0; map; map >>= 1, ++i)
4163 {
4164 DWORD write_mask;
4165
4166 if (!(map & 1)) continue;
4167
4168 semantic_name = output_signature[i].semantic_name;
4169 semantic_idx = output_signature[i].semantic_idx;
4170 write_mask = output_signature[i].mask;
4171 shader_glsl_write_mask_to_str(write_mask, reg_mask);
4172
4173 if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
4174 {
4175 if (semantic_idx == 0)
4176 shader_addline(buffer, "gl_FrontColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
4177 else if (semantic_idx == 1)
4178 shader_addline(buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
4179 }
4180 else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
4181 {
4182 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
4183 }
4184 else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
4185 {
4186 if (semantic_idx < 8)
4187 {
4188 if (!(gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) || ps_major > 0)
4189 write_mask |= WINED3DSP_WRITEMASK_3;
4190
4191 shader_addline(buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
4192 semantic_idx, reg_mask, i, reg_mask);
4193 if (!(write_mask & WINED3DSP_WRITEMASK_3))
4194 shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", semantic_idx);
4195 }
4196 }
4197 else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
4198 {
4199 shader_addline(buffer, "gl_PointSize = OUT[%u].x;\n", i);
4200 }
4201 else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
4202 {
4203 shader_addline(buffer, "gl_FogFragCoord = OUT[%u].%c;\n", i, reg_mask[1]);
4204 }
4205 }
4206 shader_addline(buffer, "}\n");
4207
4208 } else if(ps_major >= 3 && vs_major >= 3) {
4209 WORD map = vs->baseShader.reg_maps.output_registers;
4210
4211 output_signature = vs->baseShader.output_signature;
4212
4213 /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
4214 shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(3, gl_info));
4215 shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
4216
4217 /* First, sort out position and point size. Those are not passed to the pixel shader */
4218 for (i = 0; map; map >>= 1, ++i)
4219 {
4220 if (!(map & 1)) continue;
4221
4222 semantic_name = output_signature[i].semantic_name;
4223 shader_glsl_write_mask_to_str(output_signature[i].mask, reg_mask);
4224
4225 if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
4226 {
4227 shader_addline(buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
4228 }
4229 else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
4230 {
4231 shader_addline(buffer, "gl_PointSize = OUT[%u].x;\n", i);
4232 }
4233 }
4234
4235 /* Then, fix the pixel shader input */
4236 handle_ps3_input(buffer, gl_info, ps->input_reg_map, ps->baseShader.input_signature,
4237 &ps->baseShader.reg_maps, output_signature, &vs->baseShader.reg_maps);
4238
4239 shader_addline(buffer, "}\n");
4240 } else if(ps_major >= 3 && vs_major < 3) {
4241 shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(3, gl_info));
4242 shader_addline(buffer, "void order_ps_input() {\n");
4243 /* The vertex shader wrote to the builtin varyings. There is no need to figure out position and
4244 * point size, but we depend on the optimizers kindness to find out that the pixel shader doesn't
4245 * read gl_TexCoord and gl_ColorX, otherwise we'll run out of varyings
4246 */
4247 handle_ps3_input(buffer, gl_info, ps->input_reg_map, ps->baseShader.input_signature,
4248 &ps->baseShader.reg_maps, NULL, NULL);
4249 shader_addline(buffer, "}\n");
4250 } else {
4251 ERR("Unexpected vertex and pixel shader version condition: vs: %d, ps: %d\n", vs_major, ps_major);
4252 }
4253
4254 ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4255 checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
4256 GL_EXTCALL(glShaderSourceARB(ret, 1, (const char**)&buffer->buffer, NULL));
4257 checkGLcall("glShaderSourceARB(ret, 1, &buffer->buffer, NULL)");
4258 GL_EXTCALL(glCompileShaderARB(ret));
4259 checkGLcall("glCompileShaderARB(ret)");
4260 shader_glsl_validate_compile_link(gl_info, ret, FALSE);
4261 return ret;
4262}
4263
4264#ifdef VBOX_WITH_VMSVGA
4265static GLhandleARB generate_passthrough_vshader(const struct wined3d_gl_info *gl_info)
4266{
4267 GLhandleARB ret = 0;
4268 static const char *passthrough_vshader[] =
4269 {
4270 "#version 120\n"
4271 "vec4 R0;\n"
4272 "void main(void)\n"
4273 "{\n"
4274 " R0 = gl_Vertex;\n"
4275 " R0.w = 1.0;\n"
4276 " R0.z = 0.0;\n"
4277 " gl_Position = gl_ModelViewProjectionMatrix * R0;\n"
4278 "}\n"
4279 };
4280
4281 ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4282 checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
4283 GL_EXTCALL(glShaderSourceARB(ret, 1, passthrough_vshader, NULL));
4284 checkGLcall("glShaderSourceARB(ret, 1, passthrough_vshader, NULL)");
4285 GL_EXTCALL(glCompileShaderARB(ret));
4286 checkGLcall("glCompileShaderARB(ret)");
4287 shader_glsl_validate_compile_link(gl_info, ret, FALSE);
4288
4289 return ret;
4290}
4291
4292#endif
4293
4294/* GL locking is done by the caller */
4295static void hardcode_local_constants(IWineD3DBaseShaderImpl *shader, const struct wined3d_gl_info *gl_info,
4296 GLhandleARB programId, char prefix)
4297{
4298 const local_constant *lconst;
4299 GLint tmp_loc;
4300 const float *value;
4301 char glsl_name[8];
4302
4303 LIST_FOR_EACH_ENTRY(lconst, &shader->baseShader.constantsF, local_constant, entry) {
4304 value = (const float *)lconst->value;
4305 snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
4306 tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4307 GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
4308 }
4309 checkGLcall("Hardcoding local constants");
4310}
4311
4312/* GL locking is done by the caller */
4313#ifdef VBOX_WITH_VMSVGA
4314static GLhandleARB shader_glsl_generate_pshader(const struct wined3d_context *context,
4315#else
4316static GLuint shader_glsl_generate_pshader(const struct wined3d_context *context,
4317#endif
4318 struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *This,
4319 const struct ps_compile_args *args, struct ps_np2fixup_info *np2fixup_info)
4320{
4321 const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
4322 const struct wined3d_gl_info *gl_info = context->gl_info;
4323 CONST DWORD *function = This->baseShader.function;
4324 struct shader_glsl_ctx_priv priv_ctx;
4325
4326 /* Create the hw GLSL shader object and assign it as the shader->prgId */
4327 GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
4328
4329 memset(&priv_ctx, 0, sizeof(priv_ctx));
4330 priv_ctx.cur_ps_args = args;
4331 priv_ctx.cur_np2fixup_info = np2fixup_info;
4332
4333 shader_addline(buffer, "#version 120\n");
4334
4335 if (gl_info->supported[ARB_SHADER_TEXTURE_LOD] && reg_maps->usestexldd)
4336 {
4337 shader_addline(buffer, "#extension GL_ARB_shader_texture_lod : enable\n");
4338 }
4339 if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
4340 {
4341 /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
4342 * drivers write a warning if we don't do so
4343 */
4344 shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
4345 }
4346 if (gl_info->supported[EXT_GPU_SHADER4])
4347 {
4348 shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
4349 }
4350
4351 /* Base Declarations */
4352 shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShader *)This, reg_maps, &priv_ctx);
4353
4354 /* Pack 3.0 inputs */
4355 if (reg_maps->shader_version.major >= 3 && args->vp_mode != vertexshader)
4356 {
4357 shader_glsl_input_pack((IWineD3DPixelShader *) This, buffer,
4358 This->baseShader.input_signature, reg_maps, args->vp_mode);
4359 }
4360
4361 /* Base Shader Body */
4362 shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
4363
4364 /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
4365 if (reg_maps->shader_version.major < 2)
4366 {
4367 /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
4368 shader_addline(buffer, "gl_FragData[0] = R0;\n");
4369 }
4370
4371 if (args->srgb_correction)
4372 {
4373 shader_addline(buffer, "tmp0.xyz = pow(gl_FragData[0].xyz, vec3(srgb_const0.x));\n");
4374 shader_addline(buffer, "tmp0.xyz = tmp0.xyz * vec3(srgb_const0.y) - vec3(srgb_const0.z);\n");
4375 shader_addline(buffer, "tmp1.xyz = gl_FragData[0].xyz * vec3(srgb_const0.w);\n");
4376 shader_addline(buffer, "bvec3 srgb_compare = lessThan(gl_FragData[0].xyz, vec3(srgb_const1.x));\n");
4377 shader_addline(buffer, "gl_FragData[0].xyz = mix(tmp0.xyz, tmp1.xyz, vec3(srgb_compare));\n");
4378 shader_addline(buffer, "gl_FragData[0] = clamp(gl_FragData[0], 0.0, 1.0);\n");
4379 }
4380 /* Pixel shader < 3.0 do not replace the fog stage.
4381 * This implements linear fog computation and blending.
4382 * TODO: non linear fog
4383 * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
4384 * -1/(e-s) and e/(e-s) respectively.
4385 */
4386 if (reg_maps->shader_version.major < 3)
4387 {
4388 switch(args->fog) {
4389 case FOG_OFF: break;
4390 case FOG_LINEAR:
4391 shader_addline(buffer, "float fogstart = -1.0 / (gl_Fog.end - gl_Fog.start);\n");
4392 shader_addline(buffer, "float fogend = gl_Fog.end * -fogstart;\n");
4393 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * fogstart + fogend, 0.0, 1.0);\n");
4394 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4395 break;
4396 case FOG_EXP:
4397 /* Fog = e^(-gl_Fog.density * gl_FogFragCoord) */
4398 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_FogFragCoord);\n");
4399 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4400 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4401 break;
4402 case FOG_EXP2:
4403 /* Fog = e^(-(gl_Fog.density * gl_FogFragCoord)^2) */
4404 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_Fog.density * gl_FogFragCoord * gl_FogFragCoord);\n");
4405 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4406 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4407 break;
4408 }
4409 }
4410
4411 shader_addline(buffer, "}\n");
4412
4413 TRACE("Compiling shader object %p\n", (void *)(uintptr_t)shader_obj);
4414 GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
4415 GL_EXTCALL(glCompileShaderARB(shader_obj));
4416 shader_glsl_validate_compile_link(gl_info, shader_obj, FALSE);
4417
4418 /* Store the shader object */
4419 return shader_obj;
4420}
4421
4422/* GL locking is done by the caller */
4423#ifdef VBOX_WITH_VMSVGA
4424static GLhandleARB shader_glsl_generate_vshader(const struct wined3d_context *context,
4425#else
4426static GLuint shader_glsl_generate_vshader(const struct wined3d_context *context,
4427#endif
4428 struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *This,
4429 const struct vs_compile_args *args)
4430{
4431 const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
4432 const struct wined3d_gl_info *gl_info = context->gl_info;
4433 CONST DWORD *function = This->baseShader.function;
4434 struct shader_glsl_ctx_priv priv_ctx;
4435
4436 /* Create the hw GLSL shader program and assign it as the shader->prgId */
4437 GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4438
4439 shader_addline(buffer, "#version 120\n");
4440
4441 if (gl_info->supported[EXT_GPU_SHADER4])
4442 {
4443 shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
4444 }
4445
4446 memset(&priv_ctx, 0, sizeof(priv_ctx));
4447 priv_ctx.cur_vs_args = args;
4448
4449 /* Base Declarations */
4450 shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShader *)This, reg_maps, &priv_ctx);
4451
4452 /* Base Shader Body */
4453 shader_generate_main((IWineD3DBaseShader*)This, buffer, reg_maps, function, &priv_ctx);
4454
4455 /* Unpack 3.0 outputs */
4456 if (reg_maps->shader_version.major >= 3) shader_addline(buffer, "order_ps_input(OUT);\n");
4457 else shader_addline(buffer, "order_ps_input();\n");
4458
4459 /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
4460 * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
4461 * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
4462 * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
4463 */
4464 if(args->fog_src == VS_FOG_Z) {
4465 shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
4466 } else if (!reg_maps->fog) {
4467 shader_addline(buffer, "gl_FogFragCoord = 0.0;\n");
4468 }
4469
4470 /* Write the final position.
4471 *
4472 * OpenGL coordinates specify the center of the pixel while d3d coords specify
4473 * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
4474 * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
4475 * contains 1.0 to allow a mad.
4476 */
4477 shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
4478 shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
4479 if(args->clip_enabled) {
4480 shader_addline(buffer, "gl_ClipVertex = gl_Position;\n");
4481 }
4482
4483 /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
4484 *
4485 * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
4486 * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
4487 * which is the same as z = z * 2 - w.
4488 */
4489 shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
4490
4491 shader_addline(buffer, "}\n");
4492
4493 TRACE("Compiling shader object %p\n", (void *)(uintptr_t)shader_obj);
4494 GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
4495 GL_EXTCALL(glCompileShaderARB(shader_obj));
4496 shader_glsl_validate_compile_link(gl_info, shader_obj, FALSE);
4497
4498 return shader_obj;
4499}
4500
4501static GLhandleARB find_glsl_pshader(const struct wined3d_context *context,
4502 struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *shader,
4503 const struct ps_compile_args *args,
4504 UINT *inp2fixup_info
4505 )
4506{
4507 UINT i;
4508 DWORD new_size;
4509 struct glsl_ps_compiled_shader *new_array;
4510 struct glsl_pshader_private *shader_data;
4511 struct ps_np2fixup_info *np2fixup = NULL;
4512 GLhandleARB ret;
4513
4514 if (!shader->baseShader.backend_data)
4515 {
4516 shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4517 if (!shader->baseShader.backend_data)
4518 {
4519 ERR("Failed to allocate backend data.\n");
4520 return 0;
4521 }
4522 }
4523 shader_data = shader->baseShader.backend_data;
4524
4525 /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4526 * so a linear search is more performant than a hashmap or a binary search
4527 * (cache coherency etc)
4528 */
4529 for(i = 0; i < shader_data->num_gl_shaders; i++) {
4530 if(shader_data->gl_shaders[i].context==context
4531 && memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)) == 0) {
4532 if(args->np2_fixup) {
4533 *inp2fixup_info = i;
4534 }
4535 return shader_data->gl_shaders[i].prgId;
4536 }
4537 }
4538
4539 TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4540 if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4541 if (shader_data->num_gl_shaders)
4542 {
4543 new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4544 new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4545 new_size * sizeof(*shader_data->gl_shaders));
4546 } else {
4547 new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4548 new_size = 1;
4549 }
4550
4551 if(!new_array) {
4552 ERR("Out of memory\n");
4553 return 0;
4554 }
4555 shader_data->gl_shaders = new_array;
4556 shader_data->shader_array_size = new_size;
4557 }
4558
4559 shader_data->gl_shaders[shader_data->num_gl_shaders].context = context;
4560 shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4561
4562 memset(&shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup, 0, sizeof(struct ps_np2fixup_info));
4563 if (args->np2_fixup) np2fixup = &shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup;
4564
4565 pixelshader_update_samplers(&shader->baseShader.reg_maps,
4566 ((IWineD3DDeviceImpl *)shader->baseShader.device)->stateBlock->textures);
4567
4568 shader_buffer_clear(buffer);
4569 ret = shader_glsl_generate_pshader(context, buffer, shader, args, np2fixup);
4570 *inp2fixup_info = shader_data->num_gl_shaders;
4571 shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4572
4573 return ret;
4574}
4575
4576static inline BOOL vs_args_equal(const struct vs_compile_args *stored, const struct vs_compile_args *new,
4577 const DWORD use_map) {
4578 if((stored->swizzle_map & use_map) != new->swizzle_map) return FALSE;
4579 if((stored->clip_enabled) != new->clip_enabled) return FALSE;
4580 return stored->fog_src == new->fog_src;
4581}
4582
4583static GLhandleARB find_glsl_vshader(const struct wined3d_context *context,
4584 struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *shader,
4585 const struct vs_compile_args *args)
4586{
4587 UINT i;
4588 DWORD new_size;
4589 struct glsl_vs_compiled_shader *new_array;
4590 DWORD use_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.use_map;
4591 struct glsl_vshader_private *shader_data;
4592 GLhandleARB ret;
4593
4594 if (!shader->baseShader.backend_data)
4595 {
4596 shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4597 if (!shader->baseShader.backend_data)
4598 {
4599 ERR("Failed to allocate backend data.\n");
4600 return 0;
4601 }
4602 }
4603 shader_data = shader->baseShader.backend_data;
4604
4605 /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4606 * so a linear search is more performant than a hashmap or a binary search
4607 * (cache coherency etc)
4608 */
4609 for(i = 0; i < shader_data->num_gl_shaders; i++) {
4610 if(shader_data->gl_shaders[i].context==context
4611 && vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map)) {
4612 return shader_data->gl_shaders[i].prgId;
4613 }
4614 }
4615
4616 TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4617
4618 if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4619 if (shader_data->num_gl_shaders)
4620 {
4621 new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4622 new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4623 new_size * sizeof(*shader_data->gl_shaders));
4624 } else {
4625 new_array = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data->gl_shaders));
4626 new_size = 1;
4627 }
4628
4629 if(!new_array) {
4630 ERR("Out of memory\n");
4631 return 0;
4632 }
4633 shader_data->gl_shaders = new_array;
4634 shader_data->shader_array_size = new_size;
4635 }
4636
4637 shader_data->gl_shaders[shader_data->num_gl_shaders].context = context;
4638 shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4639
4640 shader_buffer_clear(buffer);
4641 ret = shader_glsl_generate_vshader(context, buffer, shader, args);
4642 shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4643
4644 return ret;
4645}
4646
4647/** Sets the GLSL program ID for the given pixel and vertex shader combination.
4648 * It sets the programId on the current StateBlock (because it should be called
4649 * inside of the DrawPrimitive() part of the render loop).
4650 *
4651 * If a program for the given combination does not exist, create one, and store
4652 * the program in the hash table. If it creates a program, it will link the
4653 * given objects, too.
4654 */
4655
4656/* GL locking is done by the caller */
4657static void set_glsl_shader_program(const struct wined3d_context *context,
4658 IWineD3DDeviceImpl *device, BOOL a_use_ps, BOOL a_use_vs)
4659{
4660 IWineD3DVertexShader *vshader = a_use_vs ? device->stateBlock->vertexShader : NULL;
4661 IWineD3DPixelShader *pshader = a_use_ps ? device->stateBlock->pixelShader : NULL;
4662 const struct wined3d_gl_info *gl_info = context->gl_info;
4663 struct shader_glsl_priv *priv = device->shader_priv;
4664 struct glsl_shader_prog_link *entry = NULL;
4665 GLhandleARB programId = 0;
4666 GLhandleARB reorder_shader_id = 0;
4667 unsigned int i;
4668 char glsl_name[8];
4669 struct ps_compile_args ps_compile_args;
4670 struct vs_compile_args vs_compile_args;
4671
4672 if (vshader) find_vs_compile_args((IWineD3DVertexShaderImpl *)vshader, device->stateBlock, &vs_compile_args);
4673 if (pshader) find_ps_compile_args((IWineD3DPixelShaderImpl *)pshader, device->stateBlock, &ps_compile_args);
4674
4675 entry = get_glsl_program_entry(priv, vshader, pshader, &vs_compile_args, &ps_compile_args, context);
4676 if (entry) {
4677 priv->glsl_program = entry;
4678 return;
4679 }
4680
4681 /* If we get to this point, then no matching program exists, so we create one */
4682 programId = GL_EXTCALL(glCreateProgramObjectARB());
4683 TRACE("Created new GLSL shader program %p\n", (void *)(uintptr_t)programId);
4684
4685 /* Create the entry */
4686 entry = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct glsl_shader_prog_link));
4687 entry->context = context;
4688 entry->programId = programId;
4689 entry->vshader = vshader;
4690 entry->pshader = pshader;
4691 entry->vs_args = vs_compile_args;
4692 entry->ps_args = ps_compile_args;
4693 entry->constant_version = 0;
4694 WINEFIXUPINFO_INIT(entry);
4695 /* Add the hash table entry */
4696 add_glsl_program_entry(priv, entry);
4697
4698 /* Set the current program */
4699 priv->glsl_program = entry;
4700
4701 /* Attach GLSL vshader */
4702 if (vshader)
4703 {
4704 GLhandleARB vshader_id = find_glsl_vshader(context, &priv->shader_buffer,
4705 (IWineD3DVertexShaderImpl *)vshader, &vs_compile_args);
4706 WORD map = ((IWineD3DBaseShaderImpl *)vshader)->baseShader.reg_maps.input_registers;
4707 char tmp_name[10];
4708
4709 reorder_shader_id = generate_param_reorder_function(&priv->shader_buffer, vshader, pshader, gl_info);
4710 TRACE("Attaching GLSL shader object %p to program %p\n", (void *)(uintptr_t)reorder_shader_id, (void *)(uintptr_t)programId);
4711 GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
4712 checkGLcall("glAttachObjectARB");
4713 /* Flag the reorder function for deletion, then it will be freed automatically when the program
4714 * is destroyed
4715 */
4716 GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
4717
4718 TRACE("Attaching GLSL shader object %p to program %p\n", (void *)(uintptr_t)vshader_id, (void *)(uintptr_t)programId);
4719 GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
4720 checkGLcall("glAttachObjectARB");
4721
4722 /* Bind vertex attributes to a corresponding index number to match
4723 * the same index numbers as ARB_vertex_programs (makes loading
4724 * vertex attributes simpler). With this method, we can use the
4725 * exact same code to load the attributes later for both ARB and
4726 * GLSL shaders.
4727 *
4728 * We have to do this here because we need to know the Program ID
4729 * in order to make the bindings work, and it has to be done prior
4730 * to linking the GLSL program. */
4731 for (i = 0; map; map >>= 1, ++i)
4732 {
4733 if (!(map & 1)) continue;
4734
4735 snprintf(tmp_name, sizeof(tmp_name), "attrib%u", i);
4736 GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
4737 }
4738 checkGLcall("glBindAttribLocationARB");
4739
4740 list_add_head(&((IWineD3DBaseShaderImpl *)vshader)->baseShader.linked_programs, &entry->vshader_entry);
4741 }
4742#ifdef VBOX_WITH_VMSVGA
4743 else
4744 if (device->strided_streams.position_transformed)
4745 {
4746 GLhandleARB passthrough_vshader_id;
4747
4748 passthrough_vshader_id = generate_passthrough_vshader(gl_info);
4749 TRACE("Attaching GLSL shader object %p to program %p\n", (void *)(uintptr_t)passthrough_vshader_id, (void *)(uintptr_t)programId);
4750 GL_EXTCALL(glAttachObjectARB(programId, passthrough_vshader_id));
4751 checkGLcall("glAttachObjectARB");
4752 /* Flag the reorder function for deletion, then it will be freed automatically when the program
4753 * is destroyed
4754 */
4755 GL_EXTCALL(glDeleteObjectARB(passthrough_vshader_id));
4756 }
4757#endif
4758
4759
4760 /* Attach GLSL pshader */
4761 if (pshader)
4762 {
4763 GLhandleARB pshader_id = find_glsl_pshader(context, &priv->shader_buffer,
4764 (IWineD3DPixelShaderImpl *)pshader, &ps_compile_args,
4765 &entry->inp2Fixup_info
4766 );
4767 TRACE("Attaching GLSL shader object %p to program %p\n", (void *)(uintptr_t)pshader_id, (void *)(uintptr_t)programId);
4768 GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
4769 checkGLcall("glAttachObjectARB");
4770
4771 list_add_head(&((IWineD3DBaseShaderImpl *)pshader)->baseShader.linked_programs, &entry->pshader_entry);
4772 }
4773
4774 /* Link the program */
4775 TRACE("Linking GLSL shader program %p\n", (void *)(uintptr_t)programId);
4776 GL_EXTCALL(glLinkProgramARB(programId));
4777 shader_glsl_validate_compile_link(gl_info, programId, TRUE);
4778
4779 entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY,
4780 sizeof(GLhandleARB) * gl_info->limits.glsl_vs_float_constants);
4781 for (i = 0; i < gl_info->limits.glsl_vs_float_constants; ++i)
4782 {
4783 snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
4784 entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4785 }
4786 for (i = 0; i < MAX_CONST_I; ++i)
4787 {
4788 snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
4789 entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4790 }
4791 entry->puniformF_locations = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY,
4792 sizeof(GLhandleARB) * gl_info->limits.glsl_ps_float_constants);
4793 for (i = 0; i < gl_info->limits.glsl_ps_float_constants; ++i)
4794 {
4795 snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
4796 entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4797 }
4798 for (i = 0; i < MAX_CONST_I; ++i)
4799 {
4800 snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
4801 entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4802 }
4803
4804 if(pshader) {
4805 char name[32];
4806
4807 for(i = 0; i < MAX_TEXTURES; i++) {
4808 sprintf(name, "bumpenvmat%u", i);
4809 entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4810 sprintf(name, "luminancescale%u", i);
4811 entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4812 sprintf(name, "luminanceoffset%u", i);
4813 entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4814 }
4815
4816 if (ps_compile_args.np2_fixup) {
4817 if (WINEFIXUPINFO_ISVALID(entry)) {
4818 entry->np2Fixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "PsamplerNP2Fixup"));
4819 } else {
4820 FIXME("NP2 texcoord fixup needed for this pixelshader, but no fixup uniform found.\n");
4821 }
4822 }
4823 }
4824
4825 entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
4826 entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
4827 checkGLcall("Find glsl program uniform locations");
4828
4829 if (pshader
4830 && ((IWineD3DPixelShaderImpl *)pshader)->baseShader.reg_maps.shader_version.major >= 3
4831 && ((IWineD3DPixelShaderImpl *)pshader)->declared_in_count > vec4_varyings(3, gl_info))
4832 {
4833 TRACE("Shader %p needs vertex color clamping disabled\n", (void *)(uintptr_t)programId);
4834 entry->vertex_color_clamp = GL_FALSE;
4835 } else {
4836 entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
4837 }
4838
4839 /* Set the shader to allow uniform loading on it */
4840 GL_EXTCALL(glUseProgramObjectARB(programId));
4841 checkGLcall("glUseProgramObjectARB(programId)");
4842
4843#ifdef DEBUG_misha
4844 {
4845 GLint programIdTest = -1;
4846 glGetIntegerv(GL_CURRENT_PROGRAM, &programIdTest);
4847 Assert(programIdTest == programId);
4848 }
4849#endif
4850
4851 /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
4852 * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
4853 * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
4854 * vertex shader with fixed function pixel processing is used we make sure that the card
4855 * supports enough samplers to allow the max number of vertex samplers with all possible
4856 * fixed function fragment processing setups. So once the program is linked these samplers
4857 * won't change.
4858 */
4859 if (vshader) shader_glsl_load_vsamplers(gl_info, device->texUnitMap, programId);
4860 if (pshader) shader_glsl_load_psamplers(gl_info, device->texUnitMap, programId);
4861
4862 /* If the local constants do not have to be loaded with the environment constants,
4863 * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
4864 * later
4865 */
4866 if (pshader && !((IWineD3DBaseShaderImpl *)pshader)->baseShader.load_local_constsF)
4867 {
4868 hardcode_local_constants((IWineD3DBaseShaderImpl *) pshader, gl_info, programId, 'P');
4869 }
4870 if (vshader && !((IWineD3DBaseShaderImpl *)vshader)->baseShader.load_local_constsF)
4871 {
4872 hardcode_local_constants((IWineD3DBaseShaderImpl *) vshader, gl_info, programId, 'V');
4873 }
4874}
4875
4876/* GL locking is done by the caller */
4877static GLhandleARB create_glsl_blt_shader(const struct wined3d_gl_info *gl_info, enum tex_types tex_type)
4878{
4879 GLhandleARB program_id;
4880 GLhandleARB vshader_id, pshader_id;
4881 static const char *blt_vshader[] =
4882 {
4883 "#version 120\n"
4884 "void main(void)\n"
4885 "{\n"
4886 " gl_Position = gl_Vertex;\n"
4887 " gl_FrontColor = vec4(1.0);\n"
4888 " gl_TexCoord[0] = gl_MultiTexCoord0;\n"
4889 "}\n"
4890 };
4891
4892 static const char *blt_pshaders[tex_type_count] =
4893 {
4894 /* tex_1d */
4895 NULL,
4896 /* tex_2d */
4897 "#version 120\n"
4898 "uniform sampler2D sampler;\n"
4899 "void main(void)\n"
4900 "{\n"
4901 " gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4902 "}\n",
4903 /* tex_3d */
4904 NULL,
4905 /* tex_cube */
4906 "#version 120\n"
4907 "uniform samplerCube sampler;\n"
4908 "void main(void)\n"
4909 "{\n"
4910 " gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4911 "}\n",
4912 /* tex_rect */
4913 "#version 120\n"
4914 "#extension GL_ARB_texture_rectangle : enable\n"
4915 "uniform sampler2DRect sampler;\n"
4916 "void main(void)\n"
4917 "{\n"
4918 " gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4919 "}\n",
4920 };
4921
4922 if (!blt_pshaders[tex_type])
4923 {
4924 FIXME("tex_type %#x not supported\n", tex_type);
4925 tex_type = tex_2d;
4926 }
4927
4928 vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4929 GL_EXTCALL(glShaderSourceARB(vshader_id, 1, blt_vshader, NULL));
4930 GL_EXTCALL(glCompileShaderARB(vshader_id));
4931 shader_glsl_validate_compile_link(gl_info, vshader_id, FALSE);
4932
4933 pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
4934 GL_EXTCALL(glShaderSourceARB(pshader_id, 1, &blt_pshaders[tex_type], NULL));
4935 GL_EXTCALL(glCompileShaderARB(pshader_id));
4936
4937 shader_glsl_validate_compile_link(gl_info, vshader_id, FALSE);
4938
4939 program_id = GL_EXTCALL(glCreateProgramObjectARB());
4940 GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
4941 GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
4942 GL_EXTCALL(glLinkProgramARB(program_id));
4943 shader_glsl_validate_compile_link(gl_info, program_id, TRUE);
4944
4945 /* Once linked we can mark the shaders for deletion. They will be deleted once the program
4946 * is destroyed
4947 */
4948 GL_EXTCALL(glDeleteObjectARB(vshader_id));
4949 GL_EXTCALL(glDeleteObjectARB(pshader_id));
4950 return program_id;
4951}
4952
4953/* GL locking is done by the caller */
4954static void shader_glsl_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4955{
4956 const struct wined3d_gl_info *gl_info = context->gl_info;
4957 IWineD3DDeviceImpl *device = context_get_device(context);
4958 struct shader_glsl_priv *priv = device->shader_priv;
4959 GLhandleARB program_id = 0;
4960 GLenum old_vertex_color_clamp, current_vertex_color_clamp;
4961
4962 old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4963
4964 if (useVS || usePS) set_glsl_shader_program(context, device, usePS, useVS);
4965 else priv->glsl_program = NULL;
4966
4967 current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4968
4969 if (old_vertex_color_clamp != current_vertex_color_clamp)
4970 {
4971 if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4972 {
4973 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
4974 checkGLcall("glClampColorARB");
4975 }
4976 else
4977 {
4978 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4979 }
4980 }
4981
4982 program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4983 if (program_id) TRACE("Using GLSL program %p\n", (void *)(uintptr_t)program_id);
4984 GL_EXTCALL(glUseProgramObjectARB(program_id));
4985 checkGLcall("glUseProgramObjectARB");
4986#ifdef DEBUG_misha
4987 {
4988 GLint programIdTest = -1;
4989 glGetIntegerv(GL_CURRENT_PROGRAM, &programIdTest);
4990 Assert(programIdTest == program_id);
4991 }
4992#endif
4993
4994 /* In case that NP2 texcoord fixup data is found for the selected program, trigger a reload of the
4995 * constants. This has to be done because it can't be guaranteed that sampler() (from state.c) is
4996 * called between selecting the shader and using it, which results in wrong fixup for some frames. */
4997 if (priv->glsl_program && WINEFIXUPINFO_ISVALID(priv->glsl_program))
4998 {
4999 shader_glsl_load_np2fixup_constants((IWineD3DDevice *)device, usePS, useVS);
5000 }
5001}
5002
5003/* GL locking is done by the caller */
5004static void shader_glsl_select_depth_blt(IWineD3DDevice *iface, enum tex_types tex_type) {
5005 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
5006 const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
5007 struct shader_glsl_priv *priv = This->shader_priv;
5008 GLhandleARB *blt_program = &priv->depth_blt_program[tex_type];
5009
5010 if (!*blt_program) {
5011 GLint loc;
5012 *blt_program = create_glsl_blt_shader(gl_info, tex_type);
5013 loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "sampler"));
5014 GL_EXTCALL(glUseProgramObjectARB(*blt_program));
5015#ifdef DEBUG_misha
5016 {
5017 GLint programIdTest = -1;
5018 glGetIntegerv(GL_CURRENT_PROGRAM, &programIdTest);
5019 Assert(programIdTest == *blt_program);
5020 }
5021#endif
5022 GL_EXTCALL(glUniform1iARB(loc, 0));
5023 } else {
5024 GL_EXTCALL(glUseProgramObjectARB(*blt_program));
5025#ifdef DEBUG_misha
5026 {
5027 GLint programIdTest = -1;
5028 glGetIntegerv(GL_CURRENT_PROGRAM, &programIdTest);
5029 Assert(programIdTest == *blt_program);
5030 }
5031#endif
5032 }
5033}
5034
5035/* GL locking is done by the caller */
5036static void shader_glsl_deselect_depth_blt(IWineD3DDevice *iface) {
5037 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
5038 const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
5039 struct shader_glsl_priv *priv = This->shader_priv;
5040 GLhandleARB program_id;
5041
5042 program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
5043 if (program_id) TRACE("Using GLSL program %p\n", (void *)(uintptr_t)program_id);
5044
5045 GL_EXTCALL(glUseProgramObjectARB(program_id));
5046 checkGLcall("glUseProgramObjectARB");
5047#ifdef DEBUG_misha
5048 {
5049 GLint programIdTest = -1;
5050 glGetIntegerv(GL_CURRENT_PROGRAM, &programIdTest);
5051 Assert(programIdTest == program_id);
5052 }
5053#endif
5054}
5055
5056static void shader_glsl_destroy(IWineD3DBaseShader *iface) {
5057 const struct list *linked_programs;
5058 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) iface;
5059 IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
5060 struct shader_glsl_priv *priv = device->shader_priv;
5061 const struct wined3d_gl_info *gl_info;
5062 struct wined3d_context *context;
5063
5064 /* Note: Do not use QueryInterface here to find out which shader type this is because this code
5065 * can be called from IWineD3DBaseShader::Release
5066 */
5067 char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
5068
5069 if(pshader) {
5070 struct glsl_pshader_private *shader_data;
5071 shader_data = This->baseShader.backend_data;
5072 if(!shader_data || shader_data->num_gl_shaders == 0)
5073 {
5074 HeapFree(GetProcessHeap(), 0, shader_data);
5075 This->baseShader.backend_data = NULL;
5076 return;
5077 }
5078
5079 context = context_acquire(device, NULL, CTXUSAGE_RESOURCELOAD);
5080 gl_info = context->gl_info;
5081
5082 if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->pshader == iface)
5083 {
5084 ENTER_GL();
5085 shader_glsl_select(context, FALSE, FALSE);
5086 LEAVE_GL();
5087 }
5088 } else {
5089 struct glsl_vshader_private *shader_data;
5090 shader_data = This->baseShader.backend_data;
5091 if(!shader_data || shader_data->num_gl_shaders == 0)
5092 {
5093 HeapFree(GetProcessHeap(), 0, shader_data);
5094 This->baseShader.backend_data = NULL;
5095 return;
5096 }
5097
5098 context = context_acquire(device, NULL, CTXUSAGE_RESOURCELOAD);
5099 gl_info = context->gl_info;
5100
5101 if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->vshader == iface)
5102 {
5103 ENTER_GL();
5104 shader_glsl_select(context, FALSE, FALSE);
5105 LEAVE_GL();
5106 }
5107 }
5108
5109 linked_programs = &This->baseShader.linked_programs;
5110
5111 TRACE("Deleting linked programs\n");
5112 if (linked_programs->next) {
5113 struct glsl_shader_prog_link *entry, *entry2;
5114
5115 ENTER_GL();
5116 if(pshader) {
5117 LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
5118 delete_glsl_program_entry(priv, gl_info, entry);
5119 }
5120 } else {
5121 LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
5122 delete_glsl_program_entry(priv, gl_info, entry);
5123 }
5124 }
5125 LEAVE_GL();
5126 }
5127
5128 if(pshader) {
5129 UINT i;
5130 struct glsl_pshader_private *shader_data = This->baseShader.backend_data;
5131
5132 ENTER_GL();
5133 for(i = 0; i < shader_data->num_gl_shaders; i++) {
5134 if (shader_data->gl_shaders[i].context==context_get_current())
5135 {
5136 TRACE("deleting pshader %p\n", (void *)(uintptr_t)shader_data->gl_shaders[i].prgId);
5137 GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
5138 checkGLcall("glDeleteObjectARB");
5139 }
5140 else
5141 {
5142 WARN("Attempting to delete pshader %p created in ctx %p from ctx %p\n",
5143 (void *)(uintptr_t)shader_data->gl_shaders[i].prgId, shader_data->gl_shaders[i].context, context_get_current());
5144 }
5145 }
5146 LEAVE_GL();
5147 HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
5148 }
5149 else
5150 {
5151 UINT i;
5152 struct glsl_vshader_private *shader_data = This->baseShader.backend_data;
5153
5154 ENTER_GL();
5155 for(i = 0; i < shader_data->num_gl_shaders; i++) {
5156 if (shader_data->gl_shaders[i].context==context_get_current())
5157 {
5158 TRACE("deleting vshader %p\n", (void *)(uintptr_t)shader_data->gl_shaders[i].prgId);
5159 GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
5160 checkGLcall("glDeleteObjectARB");
5161 }
5162 else
5163 {
5164 WARN("Attempting to delete vshader %p created in ctx %p from ctx %p\n",
5165 (void *)(uintptr_t)shader_data->gl_shaders[i].prgId, shader_data->gl_shaders[i].context, context_get_current());
5166 }
5167 }
5168 LEAVE_GL();
5169 HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
5170 }
5171
5172 HeapFree(GetProcessHeap(), 0, This->baseShader.backend_data);
5173 This->baseShader.backend_data = NULL;
5174
5175 context_release(context);
5176}
5177
5178static int glsl_program_key_compare(const void *key, const struct wine_rb_entry *entry)
5179{
5180 const glsl_program_key_t *k = key;
5181 const struct glsl_shader_prog_link *prog = WINE_RB_ENTRY_VALUE(entry,
5182 const struct glsl_shader_prog_link, program_lookup_entry);
5183 int cmp;
5184
5185 if (k->context > prog->context) return 1;
5186 else if (k->context < prog->context) return -1;
5187
5188 if (k->vshader > prog->vshader) return 1;
5189 else if (k->vshader < prog->vshader) return -1;
5190
5191 if (k->pshader > prog->pshader) return 1;
5192 else if (k->pshader < prog->pshader) return -1;
5193
5194 if (k->vshader && (cmp = memcmp(&k->vs_args, &prog->vs_args, sizeof(prog->vs_args)))) return cmp;
5195 if (k->pshader && (cmp = memcmp(&k->ps_args, &prog->ps_args, sizeof(prog->ps_args)))) return cmp;
5196
5197 return 0;
5198}
5199
5200static BOOL constant_heap_init(struct constant_heap *heap, unsigned int constant_count)
5201{
5202#ifndef VBOX
5203 SIZE_T size = (constant_count + 1) * sizeof(*heap->entries) + constant_count * sizeof(*heap->positions);
5204 void *mem = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, size);
5205#else
5206 SIZE_T size;
5207 void *mem;
5208
5209 /* Don't trash the heap if the input is bogus. */
5210 if (constant_count == 0)
5211 constant_count = 1;
5212
5213 size = (constant_count + 1) * sizeof(*heap->entries) + constant_count * sizeof(*heap->positions);
5214 mem = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, size);
5215#endif
5216
5217 if (!mem)
5218 {
5219 ERR("Failed to allocate memory\n");
5220 return FALSE;
5221 }
5222
5223 heap->entries = mem;
5224 heap->entries[1].version = 0;
5225 heap->positions = (unsigned int *)(heap->entries + constant_count + 1);
5226 heap->size = 1;
5227
5228 return TRUE;
5229}
5230
5231static void constant_heap_free(struct constant_heap *heap)
5232{
5233 HeapFree(GetProcessHeap(), 0, heap->entries);
5234}
5235
5236static const struct wine_rb_functions wined3d_glsl_program_rb_functions =
5237{
5238 wined3d_rb_alloc,
5239 wined3d_rb_realloc,
5240 wined3d_rb_free,
5241 glsl_program_key_compare,
5242};
5243
5244static HRESULT shader_glsl_alloc(IWineD3DDevice *iface) {
5245 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
5246 const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
5247 struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
5248 SIZE_T stack_size = wined3d_log2i(max(gl_info->limits.glsl_vs_float_constants,
5249 gl_info->limits.glsl_ps_float_constants)) + 1;
5250
5251 if (!shader_buffer_init(&priv->shader_buffer))
5252 {
5253 ERR("Failed to initialize shader buffer.\n");
5254 goto fail;
5255 }
5256
5257 priv->stack = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, stack_size * sizeof(*priv->stack));
5258 if (!priv->stack)
5259 {
5260 ERR("Failed to allocate memory.\n");
5261 goto fail;
5262 }
5263 if (!constant_heap_init(&priv->vconst_heap, gl_info->limits.glsl_vs_float_constants))
5264 {
5265 ERR("Failed to initialize vertex shader constant heap\n");
5266 goto fail;
5267 }
5268 if (!constant_heap_init(&priv->pconst_heap, gl_info->limits.glsl_ps_float_constants))
5269 {
5270 ERR("Failed to initialize pixel shader constant heap\n");
5271 goto fail;
5272 }
5273
5274 if (wine_rb_init(&priv->program_lookup, &wined3d_glsl_program_rb_functions) == -1)
5275 {
5276 ERR("Failed to initialize rbtree.\n");
5277 goto fail;
5278 }
5279
5280 priv->next_constant_version = 1;
5281
5282 This->shader_priv = priv;
5283 return WINED3D_OK;
5284
5285fail:
5286 constant_heap_free(&priv->pconst_heap);
5287 constant_heap_free(&priv->vconst_heap);
5288 HeapFree(GetProcessHeap(), 0, priv->stack);
5289 shader_buffer_free(&priv->shader_buffer);
5290 HeapFree(GetProcessHeap(), 0, priv);
5291 return E_OUTOFMEMORY;
5292}
5293
5294/* Context activation is done by the caller. */
5295static void shader_glsl_free(IWineD3DDevice *iface) {
5296 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
5297 const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
5298 struct shader_glsl_priv *priv = This->shader_priv;
5299 int i;
5300
5301 ENTER_GL();
5302 for (i = 0; i < tex_type_count; ++i)
5303 {
5304 if (priv->depth_blt_program[i])
5305 {
5306 GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program[i]));
5307 }
5308 }
5309 LEAVE_GL();
5310
5311 wine_rb_destroy(&priv->program_lookup, NULL, NULL);
5312 constant_heap_free(&priv->pconst_heap);
5313 constant_heap_free(&priv->vconst_heap);
5314 HeapFree(GetProcessHeap(), 0, priv->stack);
5315 shader_buffer_free(&priv->shader_buffer);
5316
5317 HeapFree(GetProcessHeap(), 0, This->shader_priv);
5318 This->shader_priv = NULL;
5319}
5320
5321static BOOL shader_glsl_dirty_const(IWineD3DDevice *iface) {
5322 /* TODO: GL_EXT_bindable_uniform can be used to share constants across shaders */
5323 return FALSE;
5324}
5325
5326static void shader_glsl_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *pCaps)
5327{
5328 /* Nvidia Geforce6/7 or Ati R4xx/R5xx cards with GLSL support, support VS 3.0 but older Nvidia/Ati
5329 * models with GLSL support only support 2.0. In case of nvidia we can detect VS 2.0 support based
5330 * on the version of NV_vertex_program.
5331 * For Ati cards there's no way using glsl (it abstracts the lowlevel info away) and also not
5332 * using ARB_vertex_program. It is safe to assume that when a card supports pixel shader 2.0 it
5333 * supports vertex shader 2.0 too and the way around. We can detect ps2.0 using the maximum number
5334 * of native instructions, so use that here. For more info see the pixel shader versioning code below.
5335 */
5336 if ((gl_info->supported[NV_VERTEX_PROGRAM2] && !gl_info->supported[NV_VERTEX_PROGRAM3])
5337 || gl_info->limits.arb_ps_instructions <= 512
5338 || gl_info->limits.glsl_vs_float_constants < 256)
5339 pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
5340 else
5341 pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
5342 TRACE_(d3d_caps)("Hardware vertex shader version %d.%d enabled (GLSL)\n", (pCaps->VertexShaderVersion >> 8) & 0xff, pCaps->VertexShaderVersion & 0xff);
5343 pCaps->MaxVertexShaderConst = gl_info->limits.glsl_vs_float_constants;
5344
5345 /* Older DX9-class videocards (GeforceFX / Radeon >9500/X*00) only support pixel shader 2.0/2.0a/2.0b.
5346 * In OpenGL the extensions related to GLSL abstract lowlevel GL info away which is needed
5347 * to distinguish between 2.0 and 3.0 (and 2.0a/2.0b). In case of Nvidia we use their fragment
5348 * program extensions. On other hardware including ATI GL_ARB_fragment_program offers the info
5349 * in max native instructions. Intel and others also offer the info in this extension but they
5350 * don't support GLSL (at least on Windows).
5351 *
5352 * PS2.0 requires at least 96 instructions, 2.0a/2.0b go up to 512. Assume that if the number
5353 * of instructions is 512 or less we have to do with ps2.0 hardware.
5354 * NOTE: ps3.0 hardware requires 512 or more instructions but ati and nvidia offer 'enough' (1024 vs 4096) on their most basic ps3.0 hardware.
5355 */
5356 if ((gl_info->supported[NV_FRAGMENT_PROGRAM] && !gl_info->supported[NV_FRAGMENT_PROGRAM2])
5357 || gl_info->limits.arb_ps_instructions <= 512
5358 || gl_info->limits.glsl_vs_float_constants < 256)
5359 pCaps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
5360 else
5361 pCaps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
5362
5363 pCaps->MaxPixelShaderConst = gl_info->limits.glsl_ps_float_constants;
5364
5365 /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
5366 * Direct3D minimum requirement.
5367 *
5368 * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
5369 * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
5370 *
5371 * The problem is that the refrast clamps temporary results in the shader to
5372 * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
5373 * then applications may miss the clamping behavior. On the other hand, if it is smaller,
5374 * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
5375 * offer a way to query this.
5376 */
5377 pCaps->PixelShader1xMaxValue = 8.0;
5378 TRACE_(d3d_caps)("Hardware pixel shader version %d.%d enabled (GLSL)\n", (pCaps->PixelShaderVersion >> 8) & 0xff, pCaps->PixelShaderVersion & 0xff);
5379
5380 pCaps->VSClipping = TRUE;
5381}
5382
5383static BOOL shader_glsl_color_fixup_supported(struct color_fixup_desc fixup)
5384{
5385 if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
5386 {
5387 TRACE("Checking support for fixup:\n");
5388 dump_color_fixup_desc(fixup);
5389 }
5390
5391 /* We support everything except YUV conversions. */
5392 if (!is_complex_fixup(fixup))
5393 {
5394 TRACE("[OK]\n");
5395 return TRUE;
5396 }
5397
5398 TRACE("[FAILED]\n");
5399 return FALSE;
5400}
5401
5402static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
5403{
5404 /* WINED3DSIH_ABS */ shader_glsl_map2gl,
5405 /* WINED3DSIH_ADD */ shader_glsl_arith,
5406 /* WINED3DSIH_BEM */ shader_glsl_bem,
5407 /* WINED3DSIH_BREAK */ shader_glsl_break,
5408 /* WINED3DSIH_BREAKC */ shader_glsl_breakc,
5409 /* WINED3DSIH_BREAKP */ NULL,
5410 /* WINED3DSIH_CALL */ shader_glsl_call,
5411 /* WINED3DSIH_CALLNZ */ shader_glsl_callnz,
5412 /* WINED3DSIH_CMP */ shader_glsl_cmp,
5413 /* WINED3DSIH_CND */ shader_glsl_cnd,
5414 /* WINED3DSIH_CRS */ shader_glsl_cross,
5415 /* WINED3DSIH_CUT */ NULL,
5416 /* WINED3DSIH_DCL */ NULL,
5417 /* WINED3DSIH_DEF */ NULL,
5418 /* WINED3DSIH_DEFB */ NULL,
5419 /* WINED3DSIH_DEFI */ NULL,
5420 /* WINED3DSIH_DP2ADD */ shader_glsl_dp2add,
5421 /* WINED3DSIH_DP3 */ shader_glsl_dot,
5422 /* WINED3DSIH_DP4 */ shader_glsl_dot,
5423 /* WINED3DSIH_DST */ shader_glsl_dst,
5424 /* WINED3DSIH_DSX */ shader_glsl_map2gl,
5425 /* WINED3DSIH_DSY */ shader_glsl_map2gl,
5426 /* WINED3DSIH_ELSE */ shader_glsl_else,
5427 /* WINED3DSIH_EMIT */ NULL,
5428 /* WINED3DSIH_ENDIF */ shader_glsl_end,
5429 /* WINED3DSIH_ENDLOOP */ shader_glsl_end,
5430 /* WINED3DSIH_ENDREP */ shader_glsl_end,
5431 /* WINED3DSIH_EXP */ shader_glsl_map2gl,
5432 /* WINED3DSIH_EXPP */ shader_glsl_expp,
5433 /* WINED3DSIH_FRC */ shader_glsl_map2gl,
5434 /* WINED3DSIH_IADD */ NULL,
5435 /* WINED3DSIH_IF */ shader_glsl_if,
5436 /* WINED3DSIH_IFC */ shader_glsl_ifc,
5437 /* WINED3DSIH_IGE */ NULL,
5438 /* WINED3DSIH_LABEL */ shader_glsl_label,
5439 /* WINED3DSIH_LIT */ shader_glsl_lit,
5440 /* WINED3DSIH_LOG */ shader_glsl_log,
5441 /* WINED3DSIH_LOGP */ shader_glsl_log,
5442 /* WINED3DSIH_LOOP */ shader_glsl_loop,
5443 /* WINED3DSIH_LRP */ shader_glsl_lrp,
5444 /* WINED3DSIH_LT */ NULL,
5445 /* WINED3DSIH_M3x2 */ shader_glsl_mnxn,
5446 /* WINED3DSIH_M3x3 */ shader_glsl_mnxn,
5447 /* WINED3DSIH_M3x4 */ shader_glsl_mnxn,
5448 /* WINED3DSIH_M4x3 */ shader_glsl_mnxn,
5449 /* WINED3DSIH_M4x4 */ shader_glsl_mnxn,
5450 /* WINED3DSIH_MAD */ shader_glsl_mad,
5451 /* WINED3DSIH_MAX */ shader_glsl_map2gl,
5452 /* WINED3DSIH_MIN */ shader_glsl_map2gl,
5453 /* WINED3DSIH_MOV */ shader_glsl_mov,
5454 /* WINED3DSIH_MOVA */ shader_glsl_mov,
5455 /* WINED3DSIH_MUL */ shader_glsl_arith,
5456 /* WINED3DSIH_NOP */ NULL,
5457 /* WINED3DSIH_NRM */ shader_glsl_nrm,
5458 /* WINED3DSIH_PHASE */ NULL,
5459 /* WINED3DSIH_POW */ shader_glsl_pow,
5460 /* WINED3DSIH_RCP */ shader_glsl_rcp,
5461 /* WINED3DSIH_REP */ shader_glsl_rep,
5462 /* WINED3DSIH_RET */ shader_glsl_ret,
5463 /* WINED3DSIH_RSQ */ shader_glsl_rsq,
5464#ifdef VBOX_WITH_VMSVGA
5465 /* WINED3DSIH_SETP */ shader_glsl_setp,
5466#else
5467 /* WINED3DSIH_SETP */ NULL,
5468#endif
5469 /* WINED3DSIH_SGE */ shader_glsl_compare,
5470 /* WINED3DSIH_SGN */ shader_glsl_sgn,
5471 /* WINED3DSIH_SINCOS */ shader_glsl_sincos,
5472 /* WINED3DSIH_SLT */ shader_glsl_compare,
5473 /* WINED3DSIH_SUB */ shader_glsl_arith,
5474 /* WINED3DSIH_TEX */ shader_glsl_tex,
5475 /* WINED3DSIH_TEXBEM */ shader_glsl_texbem,
5476 /* WINED3DSIH_TEXBEML */ shader_glsl_texbem,
5477 /* WINED3DSIH_TEXCOORD */ shader_glsl_texcoord,
5478 /* WINED3DSIH_TEXDEPTH */ shader_glsl_texdepth,
5479 /* WINED3DSIH_TEXDP3 */ shader_glsl_texdp3,
5480 /* WINED3DSIH_TEXDP3TEX */ shader_glsl_texdp3tex,
5481 /* WINED3DSIH_TEXKILL */ shader_glsl_texkill,
5482 /* WINED3DSIH_TEXLDD */ shader_glsl_texldd,
5483 /* WINED3DSIH_TEXLDL */ shader_glsl_texldl,
5484 /* WINED3DSIH_TEXM3x2DEPTH */ shader_glsl_texm3x2depth,
5485 /* WINED3DSIH_TEXM3x2PAD */ shader_glsl_texm3x2pad,
5486 /* WINED3DSIH_TEXM3x2TEX */ shader_glsl_texm3x2tex,
5487 /* WINED3DSIH_TEXM3x3 */ shader_glsl_texm3x3,
5488 /* WINED3DSIH_TEXM3x3DIFF */ NULL,
5489 /* WINED3DSIH_TEXM3x3PAD */ shader_glsl_texm3x3pad,
5490 /* WINED3DSIH_TEXM3x3SPEC */ shader_glsl_texm3x3spec,
5491 /* WINED3DSIH_TEXM3x3TEX */ shader_glsl_texm3x3tex,
5492 /* WINED3DSIH_TEXM3x3VSPEC */ shader_glsl_texm3x3vspec,
5493 /* WINED3DSIH_TEXREG2AR */ shader_glsl_texreg2ar,
5494 /* WINED3DSIH_TEXREG2GB */ shader_glsl_texreg2gb,
5495 /* WINED3DSIH_TEXREG2RGB */ shader_glsl_texreg2rgb,
5496};
5497
5498static void shader_glsl_handle_instruction(const struct wined3d_shader_instruction *ins) {
5499 SHADER_HANDLER hw_fct;
5500
5501 /* Select handler */
5502 hw_fct = shader_glsl_instruction_handler_table[ins->handler_idx];
5503
5504 /* Unhandled opcode */
5505 if (!hw_fct)
5506 {
5507 FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5508 return;
5509 }
5510 hw_fct(ins);
5511
5512 shader_glsl_add_instruction_modifiers(ins);
5513}
5514
5515const shader_backend_t glsl_shader_backend = {
5516 shader_glsl_handle_instruction,
5517 shader_glsl_select,
5518 shader_glsl_select_depth_blt,
5519 shader_glsl_deselect_depth_blt,
5520 shader_glsl_update_float_vertex_constants,
5521 shader_glsl_update_float_pixel_constants,
5522 shader_glsl_load_constants,
5523 shader_glsl_load_np2fixup_constants,
5524 shader_glsl_destroy,
5525 shader_glsl_alloc,
5526 shader_glsl_free,
5527 shader_glsl_dirty_const,
5528 shader_glsl_get_caps,
5529 shader_glsl_color_fixup_supported,
5530};
5531
5532#if defined(VBOXWINEDBG_SHADERS) || defined(VBOX_WINE_WITH_PROFILE)
5533void vboxWDbgPrintF(char * szString, ...)
5534{
5535 char szBuffer[4096*2] = {0};
5536 va_list pArgList;
5537 va_start(pArgList, szString);
5538 _vsnprintf(szBuffer, sizeof(szBuffer) / sizeof(szBuffer[0]), szString, pArgList);
5539 va_end(pArgList);
5540
5541 OutputDebugStringA(szBuffer);
5542}
5543#endif
Note: See TracBrowser for help on using the repository browser.

© 2024 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy Automated Access Etiquette