2 * GLSL pixel and vertex shader implementation
4 * Copyright 2006 Jason Green
5 * Copyright 2006-2007 Henri Verbeet
6 * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
8 * This library is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * This library is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with this library; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
24 * D3D shader asm has swizzles on source parameters, and write masks for
25 * destination parameters. GLSL uses swizzles for both. The result of this is
26 * that for example "mov dst.xw, src.zyxw" becomes "dst.xw = src.zw" in GLSL.
27 * Ie, to generate a proper GLSL source swizzle, we need to take the D3D write
28 * mask for the destination parameter into account.
34 #include "wined3d_private.h"
36 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
37 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
38 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
39 WINE_DECLARE_DEBUG_CHANNEL(d3d);
41 #define GLINFO_LOCATION (*gl_info)
43 #define WINED3D_GLSL_SAMPLE_PROJECTED 0x1
44 #define WINED3D_GLSL_SAMPLE_RECT 0x2
45 #define WINED3D_GLSL_SAMPLE_LOD 0x4
60 } glsl_sample_function_t;
64 HEAP_NODE_TRAVERSE_LEFT,
65 HEAP_NODE_TRAVERSE_RIGHT,
77 struct constant_entry *entries;
78 unsigned int *positions;
82 /* GLSL shader private data */
83 struct shader_glsl_priv {
84 struct hash_table_t *glsl_program_lookup;
85 struct glsl_shader_prog_link *glsl_program;
86 struct constant_heap vconst_heap;
87 struct constant_heap pconst_heap;
89 GLhandleARB depth_blt_program[tex_type_count];
90 UINT next_constant_version;
93 /* Struct to maintain data about a linked GLSL program */
94 struct glsl_shader_prog_link {
95 struct list vshader_entry;
96 struct list pshader_entry;
97 GLhandleARB programId;
98 GLint *vuniformF_locations;
99 GLint *puniformF_locations;
100 GLint vuniformI_locations[MAX_CONST_I];
101 GLint puniformI_locations[MAX_CONST_I];
102 GLint posFixup_location;
103 GLint rectFixup_location[MAX_FRAGMENT_SAMPLERS];
104 GLint bumpenvmat_location[MAX_TEXTURES];
105 GLint luminancescale_location[MAX_TEXTURES];
106 GLint luminanceoffset_location[MAX_TEXTURES];
107 GLint ycorrection_location;
108 GLenum vertex_color_clamp;
109 IWineD3DVertexShader *vshader;
110 IWineD3DPixelShader *pshader;
111 struct vs_compile_args vs_args;
112 struct ps_compile_args ps_args;
113 UINT constant_version;
117 IWineD3DVertexShader *vshader;
118 IWineD3DPixelShader *pshader;
119 struct ps_compile_args ps_args;
120 struct vs_compile_args vs_args;
121 } glsl_program_key_t;
124 /** Prints the GLSL info log which will contain error messages if they exist */
125 static void print_glsl_info_log(const WineD3D_GL_Info *gl_info, GLhandleARB obj)
127 int infologLength = 0;
132 static const char * const spam[] =
134 "Vertex shader was successfully compiled to run on hardware.\n", /* fglrx */
135 "Fragment shader was successfully compiled to run on hardware.\n", /* fglrx */
136 "Fragment shader(s) linked, vertex shader(s) linked. \n ", /* fglrx, with \n */
137 "Fragment shader(s) linked, vertex shader(s) linked.", /* fglrx, no \n */
138 "Vertex shader(s) linked, no fragment shader(s) defined. \n ", /* fglrx, with \n */
139 "Vertex shader(s) linked, no fragment shader(s) defined.", /* fglrx, no \n */
140 "Fragment shader was successfully compiled to run on hardware.\n"
141 "WARNING: 0:2: extension 'GL_ARB_draw_buffers' is not supported",
142 "Fragment shader(s) linked, no vertex shader(s) defined.", /* fglrx, no \n */
143 "Fragment shader(s) linked, no vertex shader(s) defined. \n ", /* fglrx, with \n */
144 "WARNING: 0:2: extension 'GL_ARB_draw_buffers' is not supported\n" /* MacOS ati */
147 if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
149 GL_EXTCALL(glGetObjectParameterivARB(obj,
150 GL_OBJECT_INFO_LOG_LENGTH_ARB,
153 /* A size of 1 is just a null-terminated string, so the log should be bigger than
154 * that if there are errors. */
155 if (infologLength > 1)
157 /* Fglrx doesn't terminate the string properly, but it tells us the proper length.
158 * So use HEAP_ZERO_MEMORY to avoid uninitialized bytes
160 infoLog = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, infologLength);
161 GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
164 for(i = 0; i < sizeof(spam) / sizeof(spam[0]); i++) {
165 if(strcmp(infoLog, spam[i]) == 0) {
171 TRACE("Spam received from GLSL shader #%u: %s\n", obj, debugstr_a(infoLog));
173 FIXME("Error received from GLSL shader #%u: %s\n", obj, debugstr_a(infoLog));
175 HeapFree(GetProcessHeap(), 0, infoLog);
180 * Loads (pixel shader) samplers
182 static void shader_glsl_load_psamplers(const WineD3D_GL_Info *gl_info, DWORD *tex_unit_map, GLhandleARB programId)
186 char sampler_name[20];
188 for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
189 snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
190 name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
191 if (name_loc != -1) {
192 DWORD mapped_unit = tex_unit_map[i];
193 if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < GL_LIMITS(fragment_samplers))
195 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
196 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
197 checkGLcall("glUniform1iARB");
199 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
205 static void shader_glsl_load_vsamplers(const WineD3D_GL_Info *gl_info, DWORD *tex_unit_map, GLhandleARB programId)
208 char sampler_name[20];
211 for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
212 snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
213 name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
214 if (name_loc != -1) {
215 DWORD mapped_unit = tex_unit_map[MAX_FRAGMENT_SAMPLERS + i];
216 if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < GL_LIMITS(combined_samplers))
218 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
219 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
220 checkGLcall("glUniform1iARB");
222 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
228 static inline void walk_constant_heap(const WineD3D_GL_Info *gl_info, const float *constants,
229 const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
232 unsigned int heap_idx = 1;
235 if (heap->entries[heap_idx].version <= version) return;
237 idx = heap->entries[heap_idx].idx;
238 if (constant_locations[idx] != -1) GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
239 stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
241 while (stack_idx >= 0)
243 /* Note that we fall through to the next case statement. */
244 switch(stack[stack_idx])
246 case HEAP_NODE_TRAVERSE_LEFT:
248 unsigned int left_idx = heap_idx << 1;
249 if (left_idx < heap->size && heap->entries[left_idx].version > version)
252 idx = heap->entries[heap_idx].idx;
253 if (constant_locations[idx] != -1)
254 GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
256 stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
257 stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
262 case HEAP_NODE_TRAVERSE_RIGHT:
264 unsigned int right_idx = (heap_idx << 1) + 1;
265 if (right_idx < heap->size && heap->entries[right_idx].version > version)
267 heap_idx = right_idx;
268 idx = heap->entries[heap_idx].idx;
269 if (constant_locations[idx] != -1)
270 GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
272 stack[stack_idx++] = HEAP_NODE_POP;
273 stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
286 checkGLcall("walk_constant_heap()");
289 static inline void apply_clamped_constant(const WineD3D_GL_Info *gl_info, GLint location, const GLfloat *data)
291 GLfloat clamped_constant[4];
293 if (location == -1) return;
295 clamped_constant[0] = data[0] < -1.0f ? -1.0f : data[0] > 1.0 ? 1.0 : data[0];
296 clamped_constant[1] = data[1] < -1.0f ? -1.0f : data[1] > 1.0 ? 1.0 : data[1];
297 clamped_constant[2] = data[2] < -1.0f ? -1.0f : data[2] > 1.0 ? 1.0 : data[2];
298 clamped_constant[3] = data[3] < -1.0f ? -1.0f : data[3] > 1.0 ? 1.0 : data[3];
300 GL_EXTCALL(glUniform4fvARB(location, 1, clamped_constant));
303 static inline void walk_constant_heap_clamped(const WineD3D_GL_Info *gl_info, const float *constants,
304 const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
307 unsigned int heap_idx = 1;
310 if (heap->entries[heap_idx].version <= version) return;
312 idx = heap->entries[heap_idx].idx;
313 apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
314 stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
316 while (stack_idx >= 0)
318 /* Note that we fall through to the next case statement. */
319 switch(stack[stack_idx])
321 case HEAP_NODE_TRAVERSE_LEFT:
323 unsigned int left_idx = heap_idx << 1;
324 if (left_idx < heap->size && heap->entries[left_idx].version > version)
327 idx = heap->entries[heap_idx].idx;
328 apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
330 stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
331 stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
336 case HEAP_NODE_TRAVERSE_RIGHT:
338 unsigned int right_idx = (heap_idx << 1) + 1;
339 if (right_idx < heap->size && heap->entries[right_idx].version > version)
341 heap_idx = right_idx;
342 idx = heap->entries[heap_idx].idx;
343 apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
345 stack[stack_idx++] = HEAP_NODE_POP;
346 stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
359 checkGLcall("walk_constant_heap_clamped()");
362 /* Loads floating point constants (aka uniforms) into the currently set GLSL program. */
363 static void shader_glsl_load_constantsF(IWineD3DBaseShaderImpl *This, const WineD3D_GL_Info *gl_info,
364 const float *constants, const GLint *constant_locations, const struct constant_heap *heap,
365 unsigned char *stack, UINT version)
367 const local_constant *lconst;
369 /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
370 if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.reg_maps.shader_version) == 1
371 && shader_is_pshader_version(This->baseShader.reg_maps.shader_version))
372 walk_constant_heap_clamped(gl_info, constants, constant_locations, heap, stack, version);
374 walk_constant_heap(gl_info, constants, constant_locations, heap, stack, version);
376 if (!This->baseShader.load_local_constsF)
378 TRACE("No need to load local float constants for this shader\n");
382 /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
383 LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry)
385 GLint location = constant_locations[lconst->idx];
386 /* We found this uniform name in the program - go ahead and send the data */
387 if (location != -1) GL_EXTCALL(glUniform4fvARB(location, 1, (const GLfloat *)lconst->value));
389 checkGLcall("glUniform4fvARB()");
392 /* Loads integer constants (aka uniforms) into the currently set GLSL program. */
393 static void shader_glsl_load_constantsI(IWineD3DBaseShaderImpl *This, const WineD3D_GL_Info *gl_info,
394 const GLint locations[MAX_CONST_I], const int *constants, WORD constants_set)
399 for (i = 0; constants_set; constants_set >>= 1, ++i)
401 if (!(constants_set & 1)) continue;
403 TRACE_(d3d_constants)("Loading constants %u: %i, %i, %i, %i\n",
404 i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
406 /* We found this uniform name in the program - go ahead and send the data */
407 GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
408 checkGLcall("glUniform4ivARB");
411 /* Load immediate constants */
412 ptr = list_head(&This->baseShader.constantsI);
414 const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
415 unsigned int idx = lconst->idx;
416 const GLint *values = (const GLint *)lconst->value;
418 TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
419 values[0], values[1], values[2], values[3]);
421 /* We found this uniform name in the program - go ahead and send the data */
422 GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
423 checkGLcall("glUniform4ivARB");
424 ptr = list_next(&This->baseShader.constantsI, ptr);
428 /* Loads boolean constants (aka uniforms) into the currently set GLSL program. */
429 static void shader_glsl_load_constantsB(IWineD3DBaseShaderImpl *This, const WineD3D_GL_Info *gl_info,
430 GLhandleARB programId, const BOOL *constants, WORD constants_set)
435 char is_pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version);
436 const char* prefix = is_pshader? "PB":"VB";
439 /* TODO: Benchmark and see if it would be beneficial to store the
440 * locations of the constants to avoid looking up each time */
441 for (i = 0; constants_set; constants_set >>= 1, ++i)
443 if (!(constants_set & 1)) continue;
445 TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
447 /* TODO: Benchmark and see if it would be beneficial to store the
448 * locations of the constants to avoid looking up each time */
449 snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
450 tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
453 /* We found this uniform name in the program - go ahead and send the data */
454 GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
455 checkGLcall("glUniform1ivARB");
459 /* Load immediate constants */
460 ptr = list_head(&This->baseShader.constantsB);
462 const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
463 unsigned int idx = lconst->idx;
464 const GLint *values = (const GLint *)lconst->value;
466 TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
468 snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
469 tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
471 /* We found this uniform name in the program - go ahead and send the data */
472 GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
473 checkGLcall("glUniform1ivARB");
475 ptr = list_next(&This->baseShader.constantsB, ptr);
479 static void reset_program_constant_version(void *value, void *context)
481 struct glsl_shader_prog_link *entry = value;
482 entry->constant_version = 0;
486 * Loads the app-supplied constants into the currently set GLSL program.
488 static void shader_glsl_load_constants(
489 IWineD3DDevice* device,
491 char useVertexShader) {
493 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) device;
494 struct shader_glsl_priv *priv = deviceImpl->shader_priv;
495 IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
496 const WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
498 GLhandleARB programId;
499 struct glsl_shader_prog_link *prog = priv->glsl_program;
500 UINT constant_version;
504 /* No GLSL program set - nothing to do. */
507 programId = prog->programId;
508 constant_version = prog->constant_version;
510 if (useVertexShader) {
511 IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
513 /* Load DirectX 9 float constants/uniforms for vertex shader */
514 shader_glsl_load_constantsF(vshader, gl_info, stateBlock->vertexShaderConstantF,
515 prog->vuniformF_locations, &priv->vconst_heap, priv->stack, constant_version);
517 /* Load DirectX 9 integer constants/uniforms for vertex shader */
518 if(vshader->baseShader.uses_int_consts) {
519 shader_glsl_load_constantsI(vshader, gl_info, prog->vuniformI_locations,
520 stateBlock->vertexShaderConstantI, stateBlock->changed.vertexShaderConstantsI);
523 /* Load DirectX 9 boolean constants/uniforms for vertex shader */
524 if(vshader->baseShader.uses_bool_consts) {
525 shader_glsl_load_constantsB(vshader, gl_info, programId,
526 stateBlock->vertexShaderConstantB, stateBlock->changed.vertexShaderConstantsB);
529 /* Upload the position fixup params */
530 GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, &deviceImpl->posFixup[0]));
531 checkGLcall("glUniform4fvARB");
534 if (usePixelShader) {
536 IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
538 /* Load DirectX 9 float constants/uniforms for pixel shader */
539 shader_glsl_load_constantsF(pshader, gl_info, stateBlock->pixelShaderConstantF,
540 prog->puniformF_locations, &priv->pconst_heap, priv->stack, constant_version);
542 /* Load DirectX 9 integer constants/uniforms for pixel shader */
543 if(pshader->baseShader.uses_int_consts) {
544 shader_glsl_load_constantsI(pshader, gl_info, prog->puniformI_locations,
545 stateBlock->pixelShaderConstantI, stateBlock->changed.pixelShaderConstantsI);
548 /* Load DirectX 9 boolean constants/uniforms for pixel shader */
549 if(pshader->baseShader.uses_bool_consts) {
550 shader_glsl_load_constantsB(pshader, gl_info, programId,
551 stateBlock->pixelShaderConstantB, stateBlock->changed.pixelShaderConstantsB);
554 /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
555 * It can't be 0 for a valid texbem instruction.
557 for(i = 0; i < ((IWineD3DPixelShaderImpl *) pshader)->numbumpenvmatconsts; i++) {
558 IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pshader;
559 int stage = ps->luminanceconst[i].texunit;
561 const float *data = (const float *)&stateBlock->textureState[(int)ps->bumpenvmatconst[i].texunit][WINED3DTSS_BUMPENVMAT00];
562 GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
563 checkGLcall("glUniformMatrix2fvARB");
565 /* texbeml needs the luminance scale and offset too. If texbeml is used, needsbumpmat
566 * is set too, so we can check that in the needsbumpmat check
568 if(ps->baseShader.reg_maps.luminanceparams[stage]) {
569 const GLfloat *scale = (const GLfloat *)&stateBlock->textureState[stage][WINED3DTSS_BUMPENVLSCALE];
570 const GLfloat *offset = (const GLfloat *)&stateBlock->textureState[stage][WINED3DTSS_BUMPENVLOFFSET];
572 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
573 checkGLcall("glUniform1fvARB");
574 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
575 checkGLcall("glUniform1fvARB");
579 if(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
580 float correction_params[4];
581 if(deviceImpl->render_offscreen) {
582 correction_params[0] = 0.0;
583 correction_params[1] = 1.0;
585 /* position is window relative, not viewport relative */
586 correction_params[0] = ((IWineD3DSurfaceImpl *) deviceImpl->render_targets[0])->currentDesc.Height;
587 correction_params[1] = -1.0;
589 GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
592 /* Constant loading for texture rect coord fixup. */
593 if (prog->ps_args.texrect_fixup) {
594 UINT fixup = prog->ps_args.texrect_fixup;
596 for (i = 0; fixup; fixup >>= 1, ++i) {
597 if (-1 != prog->rectFixup_location[i]) {
598 const IWineD3DBaseTextureImpl* const tex = (const IWineD3DBaseTextureImpl*) stateBlock->textures[i];
600 FIXME("Non-existant texture is flagged for NP2 texcoord fixup\n");
603 const float tex_dim[2] = {tex->baseTexture.pow2Matrix[0], tex->baseTexture.pow2Matrix[5]};
604 GL_EXTCALL(glUniform2fvARB(prog->rectFixup_location[i], 1, tex_dim));
611 if (priv->next_constant_version == UINT_MAX)
613 TRACE("Max constant version reached, resetting to 0.\n");
614 hash_table_for_each_entry(priv->glsl_program_lookup, reset_program_constant_version, NULL);
615 priv->next_constant_version = 1;
619 prog->constant_version = priv->next_constant_version++;
623 static inline void update_heap_entry(struct constant_heap *heap, unsigned int idx,
624 unsigned int heap_idx, DWORD new_version)
626 struct constant_entry *entries = heap->entries;
627 unsigned int *positions = heap->positions;
628 unsigned int parent_idx;
632 parent_idx = heap_idx >> 1;
634 if (new_version <= entries[parent_idx].version) break;
636 entries[heap_idx] = entries[parent_idx];
637 positions[entries[parent_idx].idx] = heap_idx;
638 heap_idx = parent_idx;
641 entries[heap_idx].version = new_version;
642 entries[heap_idx].idx = idx;
643 positions[idx] = heap_idx;
646 static void shader_glsl_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
648 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
649 struct shader_glsl_priv *priv = This->shader_priv;
650 struct constant_heap *heap = &priv->vconst_heap;
653 for (i = start; i < count + start; ++i)
655 if (!This->stateBlock->changed.vertexShaderConstantsF[i])
656 update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
658 update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
662 static void shader_glsl_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
664 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
665 struct shader_glsl_priv *priv = This->shader_priv;
666 struct constant_heap *heap = &priv->pconst_heap;
669 for (i = start; i < count + start; ++i)
671 if (!This->stateBlock->changed.pixelShaderConstantsF[i])
672 update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
674 update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
678 /** Generate the variable & register declarations for the GLSL output target */
679 static void shader_generate_glsl_declarations(IWineD3DBaseShader *iface, const shader_reg_maps *reg_maps,
680 SHADER_BUFFER *buffer, const WineD3D_GL_Info *gl_info,
681 const struct ps_compile_args *ps_args)
683 IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
684 IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
685 DWORD shader_version = reg_maps->shader_version;
686 unsigned int i, extra_constants_needed = 0;
687 const local_constant *lconst;
689 /* There are some minor differences between pixel and vertex shaders */
690 char pshader = shader_is_pshader_version(shader_version);
691 char prefix = pshader ? 'P' : 'V';
693 /* Prototype the subroutines */
694 for (i = 0; i < This->baseShader.limits.label; i++) {
695 if (reg_maps->labels[i])
696 shader_addline(buffer, "void subroutine%u();\n", i);
699 /* Declare the constants (aka uniforms) */
700 if (This->baseShader.limits.constant_float > 0) {
701 unsigned max_constantsF = min(This->baseShader.limits.constant_float,
702 (pshader ? GL_LIMITS(pshader_constantsF) : GL_LIMITS(vshader_constantsF)));
703 shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
706 if (This->baseShader.limits.constant_int > 0)
707 shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, This->baseShader.limits.constant_int);
709 if (This->baseShader.limits.constant_bool > 0)
710 shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, This->baseShader.limits.constant_bool);
713 shader_addline(buffer, "uniform vec4 posFixup;\n");
714 /* Predeclaration; This function is added at link time based on the pixel shader.
715 * VS 3.0 shaders have an array OUT[] the shader writes to, earlier versions don't have
716 * that. We know the input to the reorder function at vertex shader compile time, so
717 * we can deal with that. The reorder function for a 1.x and 2.x vertex shader can just
718 * read gl_FrontColor. The output depends on the pixel shader. The reorder function for a
719 * 1.x and 2.x pshader or for fixed function will write gl_FrontColor, and for a 3.0 shader
720 * it will write to the varying array. Here we depend on the shader optimizer on sorting that
721 * out. The nvidia driver only does that if the parameter is inout instead of out, hence the
724 if (shader_version >= WINED3DVS_VERSION(3, 0))
726 shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
728 shader_addline(buffer, "void order_ps_input();\n");
731 IWineD3DPixelShaderImpl *ps_impl = (IWineD3DPixelShaderImpl *) This;
733 ps_impl->numbumpenvmatconsts = 0;
734 for(i = 0; i < (sizeof(reg_maps->bumpmat) / sizeof(reg_maps->bumpmat[0])); i++) {
735 if(!reg_maps->bumpmat[i]) {
739 ps_impl->bumpenvmatconst[(int) ps_impl->numbumpenvmatconsts].texunit = i;
740 shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
742 if(reg_maps->luminanceparams) {
743 ps_impl->luminanceconst[(int) ps_impl->numbumpenvmatconsts].texunit = i;
744 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
745 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
746 extra_constants_needed++;
748 ps_impl->luminanceconst[(int) ps_impl->numbumpenvmatconsts].texunit = -1;
751 extra_constants_needed++;
752 ps_impl->numbumpenvmatconsts++;
755 if(ps_args->srgb_correction) {
756 shader_addline(buffer, "const vec4 srgb_mul_low = vec4(%f, %f, %f, %f);\n",
757 srgb_mul_low, srgb_mul_low, srgb_mul_low, srgb_mul_low);
758 shader_addline(buffer, "const vec4 srgb_comparison = vec4(%f, %f, %f, %f);\n",
759 srgb_cmp, srgb_cmp, srgb_cmp, srgb_cmp);
761 if(reg_maps->vpos || reg_maps->usesdsy) {
762 if(This->baseShader.limits.constant_float + extra_constants_needed + 1 < GL_LIMITS(pshader_constantsF)) {
763 shader_addline(buffer, "uniform vec4 ycorrection;\n");
764 ((IWineD3DPixelShaderImpl *) This)->vpos_uniform = 1;
765 extra_constants_needed++;
767 /* This happens because we do not have proper tracking of the constant registers that are
768 * actually used, only the max limit of the shader version
770 FIXME("Cannot find a free uniform for vpos correction params\n");
771 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
772 device->render_offscreen ? 0.0 : ((IWineD3DSurfaceImpl *) device->render_targets[0])->currentDesc.Height,
773 device->render_offscreen ? 1.0 : -1.0);
775 shader_addline(buffer, "vec4 vpos;\n");
779 /* Declare texture samplers */
780 for (i = 0; i < This->baseShader.limits.sampler; i++) {
781 if (reg_maps->samplers[i]) {
783 DWORD stype = reg_maps->samplers[i] & WINED3DSP_TEXTURETYPE_MASK;
787 shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
790 if(device->stateBlock->textures[i] &&
791 IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[i]) == GL_TEXTURE_RECTANGLE_ARB) {
792 shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
794 shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
797 if(ps_args->texrect_fixup & (1 << i)) {
798 /* RECT textures in OpenGL use texcoords in the range [0,width]x[0,height]
799 * while D3D has them in the (normalized) [0,1]x[0,1] range.
800 * samplerRectFixup stores texture dimensions and is updated through
801 * shader_glsl_load_constants when the sampler changes. */
802 shader_addline(buffer, "uniform vec2 %csamplerRectFixup%u;\n", prefix, i);
805 case WINED3DSTT_CUBE:
806 shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
808 case WINED3DSTT_VOLUME:
809 shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
812 shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
813 FIXME("Unrecognized sampler type: %#x\n", stype);
819 /* Declare address variables */
820 for (i = 0; i < This->baseShader.limits.address; i++) {
821 if (reg_maps->address[i])
822 shader_addline(buffer, "ivec4 A%d;\n", i);
825 /* Declare texture coordinate temporaries and initialize them */
826 for (i = 0; i < This->baseShader.limits.texcoord; i++) {
827 if (reg_maps->texcoord[i])
828 shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
831 /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
832 * helper function shader that is linked in at link time
834 if (pshader && shader_version >= WINED3DPS_VERSION(3, 0))
836 if (use_vs(device->stateBlock))
838 shader_addline(buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
840 /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
841 * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
842 * pixel shader that reads the fixed function color into the packed input registers.
844 shader_addline(buffer, "vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
848 /* Declare output register temporaries */
849 if(This->baseShader.limits.packed_output) {
850 shader_addline(buffer, "vec4 OUT[%u];\n", This->baseShader.limits.packed_output);
853 /* Declare temporary variables */
854 for(i = 0; i < This->baseShader.limits.temporary; i++) {
855 if (reg_maps->temporary[i])
856 shader_addline(buffer, "vec4 R%u;\n", i);
859 /* Declare attributes */
860 for (i = 0; i < This->baseShader.limits.attributes; i++) {
861 if (reg_maps->attributes[i])
862 shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
865 /* Declare loop registers aLx */
866 for (i = 0; i < reg_maps->loop_depth; i++) {
867 shader_addline(buffer, "int aL%u;\n", i);
868 shader_addline(buffer, "int tmpInt%u;\n", i);
871 /* Temporary variables for matrix operations */
872 shader_addline(buffer, "vec4 tmp0;\n");
873 shader_addline(buffer, "vec4 tmp1;\n");
875 /* Local constants use a different name so they can be loaded once at shader link time
876 * They can't be hardcoded into the shader text via LC = {x, y, z, w}; because the
877 * float -> string conversion can cause precision loss.
879 if(!This->baseShader.load_local_constsF) {
880 LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
881 shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
885 /* Start the main program */
886 shader_addline(buffer, "void main() {\n");
887 if(pshader && reg_maps->vpos) {
888 /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
889 * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
890 * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
891 * precision troubles when we just substract 0.5.
893 * To deal with that just floor() the position. This will eliminate the fraction on all cards.
895 * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
897 * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
898 * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
899 * coordinates specify the pixel centers instead of the pixel corners. This code will behave
900 * correctly on drivers that returns integer values.
902 shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
906 /*****************************************************************************
907 * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
909 * For more information, see http://wiki.winehq.org/DirectX-Shaders
910 ****************************************************************************/
913 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins, const DWORD param,
914 const DWORD addr_token, DWORD mask, glsl_src_param_t *src_param);
916 /** Used for opcode modifiers - They multiply the result by the specified amount */
917 static const char * const shift_glsl_tab[] = {
919 "2.0 * ", /* 1 (x2) */
920 "4.0 * ", /* 2 (x4) */
921 "8.0 * ", /* 3 (x8) */
922 "16.0 * ", /* 4 (x16) */
923 "32.0 * ", /* 5 (x32) */
930 "0.0625 * ", /* 12 (d16) */
931 "0.125 * ", /* 13 (d8) */
932 "0.25 * ", /* 14 (d4) */
933 "0.5 * " /* 15 (d2) */
936 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
937 static void shader_glsl_gen_modifier (
940 const char *in_regswizzle,
945 switch (instr & WINED3DSP_SRCMOD_MASK) {
946 case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
948 case WINED3DSPSM_NONE:
949 sprintf(out_str, "%s%s", in_reg, in_regswizzle);
951 case WINED3DSPSM_NEG:
952 sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
954 case WINED3DSPSM_NOT:
955 sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
957 case WINED3DSPSM_BIAS:
958 sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
960 case WINED3DSPSM_BIASNEG:
961 sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
963 case WINED3DSPSM_SIGN:
964 sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
966 case WINED3DSPSM_SIGNNEG:
967 sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
969 case WINED3DSPSM_COMP:
970 sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
973 sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
975 case WINED3DSPSM_X2NEG:
976 sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
978 case WINED3DSPSM_ABS:
979 sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
981 case WINED3DSPSM_ABSNEG:
982 sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
985 FIXME("Unhandled modifier %u\n", (instr & WINED3DSP_SRCMOD_MASK));
986 sprintf(out_str, "%s%s", in_reg, in_regswizzle);
990 /** Writes the GLSL variable name that corresponds to the register that the
991 * DX opcode parameter is trying to access */
992 static void shader_glsl_get_register_name(const DWORD param, const DWORD addr_token,
993 char *regstr, BOOL *is_color, const struct wined3d_shader_instruction *ins)
995 /* oPos, oFog and oPts in D3D */
996 static const char * const hwrastout_reg_names[] = { "gl_Position", "gl_FogFragCoord", "gl_PointSize" };
998 DWORD reg = param & WINED3DSP_REGNUM_MASK;
999 DWORD regtype = shader_get_regtype(param);
1000 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->shader;
1001 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
1002 const WineD3D_GL_Info* gl_info = &deviceImpl->adapter->gl_info;
1003 DWORD shader_version = This->baseShader.reg_maps.shader_version;
1004 char pshader = shader_is_pshader_version(shader_version);
1010 case WINED3DSPR_TEMP:
1011 sprintf(tmpStr, "R%u", reg);
1013 case WINED3DSPR_INPUT:
1015 /* Pixel shaders >= 3.0 */
1016 if (WINED3DSHADER_VERSION_MAJOR(shader_version) >= 3)
1018 DWORD in_count = GL_LIMITS(glsl_varyings) / 4;
1020 if (param & WINED3DSHADER_ADDRMODE_RELATIVE) {
1021 glsl_src_param_t rel_param;
1022 shader_glsl_add_src_param(ins, addr_token, 0, WINED3DSP_WRITEMASK_0, &rel_param);
1024 /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
1027 if(((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg]) {
1028 if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count) {
1029 sprintf(tmpStr, "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
1030 rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg], in_count - 1,
1031 rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg], in_count,
1032 rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg]);
1034 sprintf(tmpStr, "IN[%s + %u]", rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg]);
1037 if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count) {
1038 sprintf(tmpStr, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
1039 rel_param.param_str, in_count - 1,
1040 rel_param.param_str, in_count,
1041 rel_param.param_str);
1043 sprintf(tmpStr, "IN[%s]", rel_param.param_str);
1047 DWORD idx = ((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg];
1048 if (idx == in_count) {
1049 sprintf(tmpStr, "gl_Color");
1050 } else if (idx == in_count + 1) {
1051 sprintf(tmpStr, "gl_SecondaryColor");
1053 sprintf(tmpStr, "IN[%u]", idx);
1058 strcpy(tmpStr, "gl_Color");
1060 strcpy(tmpStr, "gl_SecondaryColor");
1063 if (((IWineD3DVertexShaderImpl *)This)->cur_args->swizzle_map & (1 << reg)) *is_color = TRUE;
1064 sprintf(tmpStr, "attrib%u", reg);
1067 case WINED3DSPR_CONST:
1069 const char prefix = pshader? 'P':'V';
1071 /* Relative addressing */
1072 if (param & WINED3DSHADER_ADDRMODE_RELATIVE) {
1074 /* Relative addressing on shaders 2.0+ have a relative address token,
1075 * prior to that, it was hard-coded as "A0.x" because there's only 1 register */
1076 if (WINED3DSHADER_VERSION_MAJOR(shader_version) >= 2)
1078 glsl_src_param_t rel_param;
1079 shader_glsl_add_src_param(ins, addr_token, 0, WINED3DSP_WRITEMASK_0, &rel_param);
1081 sprintf(tmpStr, "%cC[%s + %u]", prefix, rel_param.param_str, reg);
1083 sprintf(tmpStr, "%cC[%s]", prefix, rel_param.param_str);
1087 sprintf(tmpStr, "%cC[A0.x + %u]", prefix, reg);
1089 sprintf(tmpStr, "%cC[A0.x]", prefix);
1094 if(shader_constant_is_local(This, reg)) {
1095 sprintf(tmpStr, "%cLC%u", prefix, reg);
1097 sprintf(tmpStr, "%cC[%u]", prefix, reg);
1103 case WINED3DSPR_CONSTINT:
1105 sprintf(tmpStr, "PI[%u]", reg);
1107 sprintf(tmpStr, "VI[%u]", reg);
1109 case WINED3DSPR_CONSTBOOL:
1111 sprintf(tmpStr, "PB[%u]", reg);
1113 sprintf(tmpStr, "VB[%u]", reg);
1115 case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1117 sprintf(tmpStr, "T%u", reg);
1119 sprintf(tmpStr, "A%u", reg);
1122 case WINED3DSPR_LOOP:
1123 sprintf(tmpStr, "aL%u", This->baseShader.cur_loop_regno - 1);
1125 case WINED3DSPR_SAMPLER:
1127 sprintf(tmpStr, "Psampler%u", reg);
1129 sprintf(tmpStr, "Vsampler%u", reg);
1131 case WINED3DSPR_COLOROUT:
1132 if (reg >= GL_LIMITS(buffers)) {
1133 WARN("Write to render target %u, only %d supported\n", reg, 4);
1135 if (GL_SUPPORT(ARB_DRAW_BUFFERS)) {
1136 sprintf(tmpStr, "gl_FragData[%u]", reg);
1137 } else { /* On older cards with GLSL support like the GeforceFX there's only one buffer. */
1138 sprintf(tmpStr, "gl_FragColor");
1141 case WINED3DSPR_RASTOUT:
1142 sprintf(tmpStr, "%s", hwrastout_reg_names[reg]);
1144 case WINED3DSPR_DEPTHOUT:
1145 sprintf(tmpStr, "gl_FragDepth");
1147 case WINED3DSPR_ATTROUT:
1149 sprintf(tmpStr, "gl_FrontColor");
1151 sprintf(tmpStr, "gl_FrontSecondaryColor");
1154 case WINED3DSPR_TEXCRDOUT:
1155 /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
1156 if (WINED3DSHADER_VERSION_MAJOR(shader_version) >= 3) sprintf(tmpStr, "OUT[%u]", reg);
1157 else sprintf(tmpStr, "gl_TexCoord[%u]", reg);
1159 case WINED3DSPR_MISCTYPE:
1162 sprintf(tmpStr, "vpos");
1163 } else if (reg == 1){
1164 /* Note that gl_FrontFacing is a bool, while vFace is
1165 * a float for which the sign determines front/back
1167 sprintf(tmpStr, "(gl_FrontFacing ? 1.0 : -1.0)");
1169 FIXME("Unhandled misctype register %d\n", reg);
1170 sprintf(tmpStr, "unrecognized_register");
1174 FIXME("Unhandled register name Type(%d)\n", regtype);
1175 sprintf(tmpStr, "unrecognized_register");
1179 strcat(regstr, tmpStr);
1182 /* Get the GLSL write mask for the destination register */
1183 static DWORD shader_glsl_get_write_mask(const DWORD param, char *write_mask) {
1184 char *ptr = write_mask;
1185 DWORD mask = param & WINED3DSP_WRITEMASK_ALL;
1187 if (shader_is_scalar(param)) {
1188 mask = WINED3DSP_WRITEMASK_0;
1191 if (param & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1192 if (param & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1193 if (param & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1194 if (param & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1202 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1203 unsigned int size = 0;
1205 if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1206 if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1207 if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1208 if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1213 static void shader_glsl_get_swizzle(const DWORD param, BOOL fixup, DWORD mask, char *swizzle_str) {
1214 /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1215 * but addressed as "rgba". To fix this we need to swap the register's x
1216 * and z components. */
1217 DWORD swizzle = (param & WINED3DSP_SWIZZLE_MASK) >> WINED3DSP_SWIZZLE_SHIFT;
1218 const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1219 char *ptr = swizzle_str;
1221 if (!shader_is_scalar(param)) {
1223 /* swizzle bits fields: wwzzyyxx */
1224 if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = swizzle_chars[swizzle & 0x03];
1225 if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = swizzle_chars[(swizzle >> 2) & 0x03];
1226 if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = swizzle_chars[(swizzle >> 4) & 0x03];
1227 if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = swizzle_chars[(swizzle >> 6) & 0x03];
1233 /* From a given parameter token, generate the corresponding GLSL string.
1234 * Also, return the actual register name and swizzle in case the
1235 * caller needs this information as well. */
1236 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1237 const DWORD param, const DWORD addr_token, DWORD mask, glsl_src_param_t *src_param)
1239 BOOL is_color = FALSE;
1240 char swizzle_str[6];
1242 src_param->reg_name[0] = '\0';
1243 src_param->param_str[0] = '\0';
1244 swizzle_str[0] = '\0';
1246 shader_glsl_get_register_name(param, addr_token, src_param->reg_name, &is_color, ins);
1248 shader_glsl_get_swizzle(param, is_color, mask, swizzle_str);
1249 shader_glsl_gen_modifier(param, src_param->reg_name, swizzle_str, src_param->param_str);
1252 /* From a given parameter token, generate the corresponding GLSL string.
1253 * Also, return the actual register name and swizzle in case the
1254 * caller needs this information as well. */
1255 static DWORD shader_glsl_add_dst_param(const struct wined3d_shader_instruction *ins,
1256 const DWORD param, const DWORD addr_token, glsl_dst_param_t *dst_param)
1258 BOOL is_color = FALSE;
1260 dst_param->mask_str[0] = '\0';
1261 dst_param->reg_name[0] = '\0';
1263 shader_glsl_get_register_name(param, addr_token, dst_param->reg_name, &is_color, ins);
1264 return shader_glsl_get_write_mask(param, dst_param->mask_str);
1267 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1268 static DWORD shader_glsl_append_dst_ext(SHADER_BUFFER *buffer,
1269 const struct wined3d_shader_instruction *ins, const DWORD param)
1271 glsl_dst_param_t dst_param;
1275 mask = shader_glsl_add_dst_param(ins, param, ins->dst[0].addr_token, &dst_param);
1278 shift = (param & WINED3DSP_DSTSHIFT_MASK) >> WINED3DSP_DSTSHIFT_SHIFT;
1279 shader_addline(buffer, "%s%s = %s(", dst_param.reg_name, dst_param.mask_str, shift_glsl_tab[shift]);
1285 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1286 static DWORD shader_glsl_append_dst(SHADER_BUFFER *buffer, const struct wined3d_shader_instruction *ins)
1288 return shader_glsl_append_dst_ext(buffer, ins, ins->dst[0].token);
1291 /** Process GLSL instruction modifiers */
1292 void shader_glsl_add_instruction_modifiers(const struct wined3d_shader_instruction *ins)
1294 glsl_dst_param_t dst_param;
1297 if (!ins->dst_count) return;
1299 mask = ins->dst[0].token & WINED3DSP_DSTMOD_MASK;
1302 shader_glsl_add_dst_param(ins, ins->dst[0].token, 0, &dst_param);
1304 if (mask & WINED3DSPDM_SATURATE)
1306 /* _SAT means to clamp the value of the register to between 0 and 1 */
1307 shader_addline(ins->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1308 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1311 if (mask & WINED3DSPDM_MSAMPCENTROID)
1313 FIXME("_centroid modifier not handled\n");
1316 if (mask & WINED3DSPDM_PARTIALPRECISION)
1318 /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1322 static inline const char *shader_get_comp_op(DWORD flags)
1324 DWORD op = (flags & INST_CONTROLS_MASK) >> INST_CONTROLS_SHIFT;
1326 case COMPARISON_GT: return ">";
1327 case COMPARISON_EQ: return "==";
1328 case COMPARISON_GE: return ">=";
1329 case COMPARISON_LT: return "<";
1330 case COMPARISON_NE: return "!=";
1331 case COMPARISON_LE: return "<=";
1333 FIXME("Unrecognized comparison value: %u\n", op);
1338 static void shader_glsl_get_sample_function(DWORD sampler_type, DWORD flags, glsl_sample_function_t *sample_function)
1340 BOOL projected = flags & WINED3D_GLSL_SAMPLE_PROJECTED;
1341 BOOL texrect = flags & WINED3D_GLSL_SAMPLE_RECT;
1342 BOOL lod = flags & WINED3D_GLSL_SAMPLE_LOD;
1344 /* Note that there's no such thing as a projected cube texture. */
1345 switch(sampler_type) {
1348 sample_function->name = projected ? "texture1DProjLod" : "texture1DLod";
1350 sample_function->name = projected ? "texture1DProj" : "texture1D";
1352 sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1357 sample_function->name = projected ? "texture2DRectProjLod" : "texture2DRectLod";
1359 sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1363 sample_function->name = projected ? "texture2DProjLod" : "texture2DLod";
1365 sample_function->name = projected ? "texture2DProj" : "texture2D";
1368 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1370 case WINED3DSTT_CUBE:
1372 sample_function->name = "textureCubeLod";
1374 sample_function->name = "textureCube";
1376 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1378 case WINED3DSTT_VOLUME:
1380 sample_function->name = projected ? "texture3DProjLod" : "texture3DLod";
1382 sample_function->name = projected ? "texture3DProj" : "texture3D";
1384 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1387 sample_function->name = "";
1388 sample_function->coord_mask = 0;
1389 FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1394 static void shader_glsl_append_fixup_arg(char *arguments, const char *reg_name,
1395 BOOL sign_fixup, enum fixup_channel_source channel_source)
1397 switch(channel_source)
1399 case CHANNEL_SOURCE_ZERO:
1400 strcat(arguments, "0.0");
1403 case CHANNEL_SOURCE_ONE:
1404 strcat(arguments, "1.0");
1407 case CHANNEL_SOURCE_X:
1408 strcat(arguments, reg_name);
1409 strcat(arguments, ".x");
1412 case CHANNEL_SOURCE_Y:
1413 strcat(arguments, reg_name);
1414 strcat(arguments, ".y");
1417 case CHANNEL_SOURCE_Z:
1418 strcat(arguments, reg_name);
1419 strcat(arguments, ".z");
1422 case CHANNEL_SOURCE_W:
1423 strcat(arguments, reg_name);
1424 strcat(arguments, ".w");
1428 FIXME("Unhandled channel source %#x\n", channel_source);
1429 strcat(arguments, "undefined");
1433 if (sign_fixup) strcat(arguments, " * 2.0 - 1.0");
1436 static void shader_glsl_color_correction(const struct wined3d_shader_instruction *ins, struct color_fixup_desc fixup)
1438 unsigned int mask_size, remaining;
1439 glsl_dst_param_t dst_param;
1440 char arguments[256];
1445 if (fixup.x_sign_fixup || fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1446 if (fixup.y_sign_fixup || fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1447 if (fixup.z_sign_fixup || fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1448 if (fixup.w_sign_fixup || fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1449 mask &= ins->dst[0].token;
1451 if (!mask) return; /* Nothing to do */
1453 if (is_yuv_fixup(fixup))
1455 enum yuv_fixup yuv_fixup = get_yuv_fixup(fixup);
1456 FIXME("YUV fixup (%#x) not supported\n", yuv_fixup);
1460 mask_size = shader_glsl_get_write_mask_size(mask);
1462 dst_param.mask_str[0] = '\0';
1463 shader_glsl_get_write_mask(mask, dst_param.mask_str);
1465 dst_param.reg_name[0] = '\0';
1466 shader_glsl_get_register_name(ins->dst[0].token, ins->dst[0].addr_token, dst_param.reg_name, &dummy, ins);
1468 arguments[0] = '\0';
1469 remaining = mask_size;
1470 if (mask & WINED3DSP_WRITEMASK_0)
1472 shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.x_sign_fixup, fixup.x_source);
1473 if (--remaining) strcat(arguments, ", ");
1475 if (mask & WINED3DSP_WRITEMASK_1)
1477 shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.y_sign_fixup, fixup.y_source);
1478 if (--remaining) strcat(arguments, ", ");
1480 if (mask & WINED3DSP_WRITEMASK_2)
1482 shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.z_sign_fixup, fixup.z_source);
1483 if (--remaining) strcat(arguments, ", ");
1485 if (mask & WINED3DSP_WRITEMASK_3)
1487 shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.w_sign_fixup, fixup.w_source);
1488 if (--remaining) strcat(arguments, ", ");
1493 shader_addline(ins->buffer, "%s%s = vec%u(%s);\n",
1494 dst_param.reg_name, dst_param.mask_str, mask_size, arguments);
1498 shader_addline(ins->buffer, "%s%s = %s;\n", dst_param.reg_name, dst_param.mask_str, arguments);
1502 static void PRINTF_ATTR(6, 7) shader_glsl_gen_sample_code(const struct wined3d_shader_instruction *ins,
1503 DWORD sampler, const glsl_sample_function_t *sample_function, DWORD swizzle,
1504 const char *bias, const char *coord_reg_fmt, ...)
1506 const char *sampler_base;
1507 char dst_swizzle[6];
1508 struct color_fixup_desc fixup;
1509 BOOL rect_fixup = FALSE;
1512 shader_glsl_get_swizzle(swizzle, FALSE, ins->dst[0].token, dst_swizzle);
1514 if (shader_is_pshader_version(ins->reg_maps->shader_version))
1516 IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->shader;
1517 fixup = This->cur_args->color_fixup[sampler];
1518 sampler_base = "Psampler";
1520 if(This->cur_args->texrect_fixup & (1 << sampler)) {
1522 FIXME("Biased sampling from RECT textures is unsupported\n");
1528 sampler_base = "Vsampler";
1529 fixup = COLOR_FIXUP_IDENTITY; /* FIXME: Vshader color fixup */
1532 shader_glsl_append_dst(ins->buffer, ins);
1534 shader_addline(ins->buffer, "%s(%s%u, ", sample_function->name, sampler_base, sampler);
1536 va_start(args, coord_reg_fmt);
1537 shader_vaddline(ins->buffer, coord_reg_fmt, args);
1541 shader_addline(ins->buffer, ", %s)%s);\n", bias, dst_swizzle);
1544 shader_addline(ins->buffer, " * PsamplerRectFixup%u)%s);\n", sampler, dst_swizzle);
1546 shader_addline(ins->buffer, ")%s);\n", dst_swizzle);
1550 if(!is_identity_fixup(fixup)) {
1551 shader_glsl_color_correction(ins, fixup);
1555 /*****************************************************************************
1557 * Begin processing individual instruction opcodes
1559 ****************************************************************************/
1561 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
1562 static void shader_glsl_arith(const struct wined3d_shader_instruction *ins)
1564 SHADER_BUFFER *buffer = ins->buffer;
1565 glsl_src_param_t src0_param;
1566 glsl_src_param_t src1_param;
1570 /* Determine the GLSL operator to use based on the opcode */
1571 switch (ins->handler_idx)
1573 case WINED3DSIH_MUL: op = '*'; break;
1574 case WINED3DSIH_ADD: op = '+'; break;
1575 case WINED3DSIH_SUB: op = '-'; break;
1578 FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
1582 write_mask = shader_glsl_append_dst(buffer, ins);
1583 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], write_mask, &src0_param);
1584 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1], write_mask, &src1_param);
1585 shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
1588 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
1589 static void shader_glsl_mov(const struct wined3d_shader_instruction *ins)
1591 SHADER_BUFFER *buffer = ins->buffer;
1592 glsl_src_param_t src0_param;
1595 write_mask = shader_glsl_append_dst(buffer, ins);
1596 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], write_mask, &src0_param);
1598 /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
1599 * shader versions WINED3DSIO_MOVA is used for this. */
1600 if ((WINED3DSHADER_VERSION_MAJOR(ins->reg_maps->shader_version) == 1
1601 && !shader_is_pshader_version(ins->reg_maps->shader_version)
1602 && shader_get_regtype(ins->dst[0].token) == WINED3DSPR_ADDR))
1604 /* This is a simple floor() */
1605 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1606 if (mask_size > 1) {
1607 shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
1609 shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
1612 else if(ins->handler_idx == WINED3DSIH_MOVA)
1614 /* We need to *round* to the nearest int here. */
1615 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1616 if (mask_size > 1) {
1617 shader_addline(buffer, "ivec%d(floor(abs(%s) + vec%d(0.5)) * sign(%s)));\n", mask_size, src0_param.param_str, mask_size, src0_param.param_str);
1619 shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n", src0_param.param_str, src0_param.param_str);
1622 shader_addline(buffer, "%s);\n", src0_param.param_str);
1626 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
1627 static void shader_glsl_dot(const struct wined3d_shader_instruction *ins)
1629 SHADER_BUFFER *buffer = ins->buffer;
1630 glsl_src_param_t src0_param;
1631 glsl_src_param_t src1_param;
1632 DWORD dst_write_mask, src_write_mask;
1633 unsigned int dst_size = 0;
1635 dst_write_mask = shader_glsl_append_dst(buffer, ins);
1636 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1638 /* dp3 works on vec3, dp4 on vec4 */
1639 if (ins->handler_idx == WINED3DSIH_DP4)
1641 src_write_mask = WINED3DSP_WRITEMASK_ALL;
1643 src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1646 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], src_write_mask, &src0_param);
1647 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1], src_write_mask, &src1_param);
1650 shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1652 shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
1656 /* Note that this instruction has some restrictions. The destination write mask
1657 * can't contain the w component, and the source swizzles have to be .xyzw */
1658 static void shader_glsl_cross(const struct wined3d_shader_instruction *ins)
1660 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1661 glsl_src_param_t src0_param;
1662 glsl_src_param_t src1_param;
1665 shader_glsl_get_write_mask(ins->dst[0].token, dst_mask);
1666 shader_glsl_append_dst(ins->buffer, ins);
1667 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], src_mask, &src0_param);
1668 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1], src_mask, &src1_param);
1669 shader_addline(ins->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
1672 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
1673 * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
1674 * GLSL uses the value as-is. */
1675 static void shader_glsl_pow(const struct wined3d_shader_instruction *ins)
1677 SHADER_BUFFER *buffer = ins->buffer;
1678 glsl_src_param_t src0_param;
1679 glsl_src_param_t src1_param;
1680 DWORD dst_write_mask;
1681 unsigned int dst_size;
1683 dst_write_mask = shader_glsl_append_dst(buffer, ins);
1684 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1686 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1687 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
1690 shader_addline(buffer, "vec%d(pow(abs(%s), %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1692 shader_addline(buffer, "pow(abs(%s), %s));\n", src0_param.param_str, src1_param.param_str);
1696 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
1697 * Src0 is a scalar. Note that D3D uses the absolute of src0, while
1698 * GLSL uses the value as-is. */
1699 static void shader_glsl_log(const struct wined3d_shader_instruction *ins)
1701 SHADER_BUFFER *buffer = ins->buffer;
1702 glsl_src_param_t src0_param;
1703 DWORD dst_write_mask;
1704 unsigned int dst_size;
1706 dst_write_mask = shader_glsl_append_dst(buffer, ins);
1707 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1709 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1712 shader_addline(buffer, "vec%d(log2(abs(%s))));\n", dst_size, src0_param.param_str);
1714 shader_addline(buffer, "log2(abs(%s)));\n", src0_param.param_str);
1718 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
1719 static void shader_glsl_map2gl(const struct wined3d_shader_instruction *ins)
1721 SHADER_BUFFER *buffer = ins->buffer;
1722 glsl_src_param_t src_param;
1723 const char *instruction;
1727 /* Determine the GLSL function to use based on the opcode */
1728 /* TODO: Possibly make this a table for faster lookups */
1729 switch (ins->handler_idx)
1731 case WINED3DSIH_MIN: instruction = "min"; break;
1732 case WINED3DSIH_MAX: instruction = "max"; break;
1733 case WINED3DSIH_ABS: instruction = "abs"; break;
1734 case WINED3DSIH_FRC: instruction = "fract"; break;
1735 case WINED3DSIH_NRM: instruction = "normalize"; break;
1736 case WINED3DSIH_EXP: instruction = "exp2"; break;
1737 case WINED3DSIH_SGN: instruction = "sign"; break;
1738 case WINED3DSIH_DSX: instruction = "dFdx"; break;
1739 case WINED3DSIH_DSY: instruction = "ycorrection.y * dFdy"; break;
1740 default: instruction = "";
1741 FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
1745 write_mask = shader_glsl_append_dst(buffer, ins);
1747 shader_addline(buffer, "%s(", instruction);
1751 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], write_mask, &src_param);
1752 shader_addline(buffer, "%s", src_param.param_str);
1753 for (i = 1; i < ins->src_count; ++i)
1755 shader_glsl_add_src_param(ins, ins->src[i], ins->src_addr[i], write_mask, &src_param);
1756 shader_addline(buffer, ", %s", src_param.param_str);
1760 shader_addline(buffer, "));\n");
1763 /** Process the WINED3DSIO_EXPP instruction in GLSL:
1764 * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
1765 * dst.x = 2^(floor(src))
1766 * dst.y = src - floor(src)
1767 * dst.z = 2^src (partial precision is allowed, but optional)
1769 * For 2.0 shaders, just do this (honoring writemask and swizzle):
1770 * dst = 2^src; (partial precision is allowed, but optional)
1772 static void shader_glsl_expp(const struct wined3d_shader_instruction *ins)
1774 glsl_src_param_t src_param;
1776 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_0, &src_param);
1778 if (ins->reg_maps->shader_version < WINED3DPS_VERSION(2,0))
1782 shader_addline(ins->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
1783 shader_addline(ins->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
1784 shader_addline(ins->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
1785 shader_addline(ins->buffer, "tmp0.w = 1.0;\n");
1787 shader_glsl_append_dst(ins->buffer, ins);
1788 shader_glsl_get_write_mask(ins->dst[0].token, dst_mask);
1789 shader_addline(ins->buffer, "tmp0%s);\n", dst_mask);
1792 unsigned int mask_size;
1794 write_mask = shader_glsl_append_dst(ins->buffer, ins);
1795 mask_size = shader_glsl_get_write_mask_size(write_mask);
1797 if (mask_size > 1) {
1798 shader_addline(ins->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
1800 shader_addline(ins->buffer, "exp2(%s));\n", src_param.param_str);
1805 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
1806 static void shader_glsl_rcp(const struct wined3d_shader_instruction *ins)
1808 glsl_src_param_t src_param;
1810 unsigned int mask_size;
1812 write_mask = shader_glsl_append_dst(ins->buffer, ins);
1813 mask_size = shader_glsl_get_write_mask_size(write_mask);
1814 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_3, &src_param);
1816 if (mask_size > 1) {
1817 shader_addline(ins->buffer, "vec%d(1.0 / %s));\n", mask_size, src_param.param_str);
1819 shader_addline(ins->buffer, "1.0 / %s);\n", src_param.param_str);
1823 static void shader_glsl_rsq(const struct wined3d_shader_instruction *ins)
1825 SHADER_BUFFER *buffer = ins->buffer;
1826 glsl_src_param_t src_param;
1828 unsigned int mask_size;
1830 write_mask = shader_glsl_append_dst(buffer, ins);
1831 mask_size = shader_glsl_get_write_mask_size(write_mask);
1833 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_3, &src_param);
1835 if (mask_size > 1) {
1836 shader_addline(buffer, "vec%d(inversesqrt(%s)));\n", mask_size, src_param.param_str);
1838 shader_addline(buffer, "inversesqrt(%s));\n", src_param.param_str);
1842 /** Process signed comparison opcodes in GLSL. */
1843 static void shader_glsl_compare(const struct wined3d_shader_instruction *ins)
1845 glsl_src_param_t src0_param;
1846 glsl_src_param_t src1_param;
1848 unsigned int mask_size;
1850 write_mask = shader_glsl_append_dst(ins->buffer, ins);
1851 mask_size = shader_glsl_get_write_mask_size(write_mask);
1852 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], write_mask, &src0_param);
1853 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1], write_mask, &src1_param);
1855 if (mask_size > 1) {
1856 const char *compare;
1858 switch(ins->handler_idx)
1860 case WINED3DSIH_SLT: compare = "lessThan"; break;
1861 case WINED3DSIH_SGE: compare = "greaterThanEqual"; break;
1862 default: compare = "";
1863 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
1866 shader_addline(ins->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
1867 src0_param.param_str, src1_param.param_str);
1869 switch(ins->handler_idx)
1871 case WINED3DSIH_SLT:
1872 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
1873 * to return 0.0 but step returns 1.0 because step is not < x
1874 * An alternative is a bvec compare padded with an unused second component.
1875 * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
1876 * issue. Playing with not() is not possible either because not() does not accept
1879 shader_addline(ins->buffer, "(%s < %s) ? 1.0 : 0.0);\n", src0_param.param_str, src1_param.param_str);
1881 case WINED3DSIH_SGE:
1882 /* Here we can use the step() function and safe a conditional */
1883 shader_addline(ins->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
1886 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
1892 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
1893 static void shader_glsl_cmp(const struct wined3d_shader_instruction *ins)
1895 glsl_src_param_t src0_param;
1896 glsl_src_param_t src1_param;
1897 glsl_src_param_t src2_param;
1898 DWORD write_mask, cmp_channel = 0;
1901 BOOL temp_destination = FALSE;
1903 if (shader_is_scalar(ins->src[0]))
1905 write_mask = shader_glsl_append_dst(ins->buffer, ins);
1907 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
1908 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1], write_mask, &src1_param);
1909 shader_glsl_add_src_param(ins, ins->src[2], ins->src_addr[2], write_mask, &src2_param);
1911 shader_addline(ins->buffer, "%s >= 0.0 ? %s : %s);\n",
1912 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1914 DWORD src0reg = ins->src[0] & WINED3DSP_REGNUM_MASK;
1915 DWORD src1reg = ins->src[1] & WINED3DSP_REGNUM_MASK;
1916 DWORD src2reg = ins->src[2] & WINED3DSP_REGNUM_MASK;
1917 DWORD src0regtype = shader_get_regtype(ins->src[0]);
1918 DWORD src1regtype = shader_get_regtype(ins->src[1]);
1919 DWORD src2regtype = shader_get_regtype(ins->src[2]);
1920 DWORD dstreg = ins->dst[0].register_idx;
1921 DWORD dstregtype = shader_get_regtype(ins->dst[0].token);
1923 /* Cycle through all source0 channels */
1924 for (i=0; i<4; i++) {
1926 /* Find the destination channels which use the current source0 channel */
1927 for (j=0; j<4; j++) {
1928 if (((ins->src[0] >> (WINED3DSP_SWIZZLE_SHIFT + 2 * j)) & 0x3) == i)
1930 write_mask |= WINED3DSP_WRITEMASK_0 << j;
1931 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
1935 /* Splitting the cmp instruction up in multiple lines imposes a problem:
1936 * The first lines may overwrite source parameters of the following lines.
1937 * Deal with that by using a temporary destination register if needed
1939 if((src0reg == dstreg && src0regtype == dstregtype) ||
1940 (src1reg == dstreg && src1regtype == dstregtype) ||
1941 (src2reg == dstreg && src2regtype == dstregtype)) {
1943 write_mask = shader_glsl_get_write_mask(ins->dst[0].token & (~WINED3DSP_SWIZZLE_MASK | write_mask),
1945 if (!write_mask) continue;
1946 shader_addline(ins->buffer, "tmp0%s = (", mask_char);
1947 temp_destination = TRUE;
1949 write_mask = shader_glsl_append_dst_ext(ins->buffer, ins,
1950 ins->dst[0].token & (~WINED3DSP_SWIZZLE_MASK | write_mask));
1951 if (!write_mask) continue;
1954 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], cmp_channel, &src0_param);
1955 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1], write_mask, &src1_param);
1956 shader_glsl_add_src_param(ins, ins->src[2], ins->src_addr[2], write_mask, &src2_param);
1958 shader_addline(ins->buffer, "%s >= 0.0 ? %s : %s);\n",
1959 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1962 if(temp_destination) {
1963 shader_glsl_get_write_mask(ins->dst[0].token, mask_char);
1964 shader_glsl_append_dst_ext(ins->buffer, ins, ins->dst[0].token);
1965 shader_addline(ins->buffer, "tmp0%s);\n", mask_char);
1971 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
1972 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
1973 * the compare is done per component of src0. */
1974 static void shader_glsl_cnd(const struct wined3d_shader_instruction *ins)
1976 glsl_src_param_t src0_param;
1977 glsl_src_param_t src1_param;
1978 glsl_src_param_t src2_param;
1979 DWORD write_mask, cmp_channel = 0;
1982 if (ins->reg_maps->shader_version < WINED3DPS_VERSION(1, 4))
1984 write_mask = shader_glsl_append_dst(ins->buffer, ins);
1985 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1986 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1], write_mask, &src1_param);
1987 shader_glsl_add_src_param(ins, ins->src[2], ins->src_addr[2], write_mask, &src2_param);
1989 /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
1992 shader_addline(ins->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
1994 shader_addline(ins->buffer, "%s > 0.5 ? %s : %s);\n",
1995 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1999 /* Cycle through all source0 channels */
2000 for (i=0; i<4; i++) {
2002 /* Find the destination channels which use the current source0 channel */
2003 for (j=0; j<4; j++) {
2004 if (((ins->src[0] >> (WINED3DSP_SWIZZLE_SHIFT + 2 * j)) & 0x3) == i)
2006 write_mask |= WINED3DSP_WRITEMASK_0 << j;
2007 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2010 write_mask = shader_glsl_append_dst_ext(ins->buffer, ins,
2011 ins->dst[0].token & (~WINED3DSP_SWIZZLE_MASK | write_mask));
2012 if (!write_mask) continue;
2014 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], cmp_channel, &src0_param);
2015 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1], write_mask, &src1_param);
2016 shader_glsl_add_src_param(ins, ins->src[2], ins->src_addr[2], write_mask, &src2_param);
2018 shader_addline(ins->buffer, "%s > 0.5 ? %s : %s);\n",
2019 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2023 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
2024 static void shader_glsl_mad(const struct wined3d_shader_instruction *ins)
2026 glsl_src_param_t src0_param;
2027 glsl_src_param_t src1_param;
2028 glsl_src_param_t src2_param;
2031 write_mask = shader_glsl_append_dst(ins->buffer, ins);
2032 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], write_mask, &src0_param);
2033 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1], write_mask, &src1_param);
2034 shader_glsl_add_src_param(ins, ins->src[2], ins->src_addr[2], write_mask, &src2_param);
2035 shader_addline(ins->buffer, "(%s * %s) + %s);\n",
2036 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2039 /** Handles transforming all WINED3DSIO_M?x? opcodes for
2040 Vertex shaders to GLSL codes */
2041 static void shader_glsl_mnxn(const struct wined3d_shader_instruction *ins)
2044 int nComponents = 0;
2045 struct wined3d_shader_dst_param tmp_dst = {0};
2046 struct wined3d_shader_instruction tmp_ins;
2048 memset(&tmp_ins, 0, sizeof(tmp_ins));
2050 /* Set constants for the temporary argument */
2051 tmp_ins.shader = ins->shader;
2052 tmp_ins.buffer = ins->buffer;
2053 tmp_ins.src[0] = ins->src[0];
2054 tmp_ins.src_addr[0] = ins->src_addr[0];
2055 tmp_ins.src_addr[1] = ins->src_addr[1];
2056 tmp_ins.reg_maps = ins->reg_maps;
2057 tmp_ins.dst_count = 1;
2058 tmp_ins.dst = &tmp_dst;
2059 tmp_ins.src_count = 2;
2061 switch(ins->handler_idx)
2063 case WINED3DSIH_M4x4:
2065 tmp_ins.handler_idx = WINED3DSIH_DP4;
2067 case WINED3DSIH_M4x3:
2069 tmp_ins.handler_idx = WINED3DSIH_DP4;
2071 case WINED3DSIH_M3x4:
2073 tmp_ins.handler_idx = WINED3DSIH_DP3;
2075 case WINED3DSIH_M3x3:
2077 tmp_ins.handler_idx = WINED3DSIH_DP3;
2079 case WINED3DSIH_M3x2:
2081 tmp_ins.handler_idx = WINED3DSIH_DP3;
2087 for (i = 0; i < nComponents; ++i)
2089 tmp_dst.register_idx = ins->dst[0].register_idx;
2090 tmp_dst.token = ((ins->dst[0].token) & ~WINED3DSP_WRITEMASK_ALL) | (WINED3DSP_WRITEMASK_0 << i);
2091 tmp_ins.src[1] = ins->src[1] + i;
2092 shader_glsl_dot(&tmp_ins);
2097 The LRP instruction performs a component-wise linear interpolation
2098 between the second and third operands using the first operand as the
2099 blend factor. Equation: (dst = src2 + src0 * (src1 - src2))
2100 This is equivalent to mix(src2, src1, src0);
2102 static void shader_glsl_lrp(const struct wined3d_shader_instruction *ins)
2104 glsl_src_param_t src0_param;
2105 glsl_src_param_t src1_param;
2106 glsl_src_param_t src2_param;
2109 write_mask = shader_glsl_append_dst(ins->buffer, ins);
2111 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], write_mask, &src0_param);
2112 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1], write_mask, &src1_param);
2113 shader_glsl_add_src_param(ins, ins->src[2], ins->src_addr[2], write_mask, &src2_param);
2115 shader_addline(ins->buffer, "mix(%s, %s, %s));\n",
2116 src2_param.param_str, src1_param.param_str, src0_param.param_str);
2119 /** Process the WINED3DSIO_LIT instruction in GLSL:
2120 * dst.x = dst.w = 1.0
2121 * dst.y = (src0.x > 0) ? src0.x
2122 * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
2123 * where src.w is clamped at +- 128
2125 static void shader_glsl_lit(const struct wined3d_shader_instruction *ins)
2127 glsl_src_param_t src0_param;
2128 glsl_src_param_t src1_param;
2129 glsl_src_param_t src3_param;
2132 shader_glsl_append_dst(ins->buffer, ins);
2133 shader_glsl_get_write_mask(ins->dst[0].token, dst_mask);
2135 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2136 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_1, &src1_param);
2137 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_3, &src3_param);
2139 /* The sdk specifies the instruction like this
2141 * if(src.x > 0.0) dst.y = src.x
2143 * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
2147 * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
2148 * dst.x = 1.0 ... No further explanation needed
2149 * dst.y = max(src.y, 0.0); ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
2150 * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0; ... 0 ^ power is 0, and otherwise we use y anyway
2151 * dst.w = 1.0. ... Nothing fancy.
2153 * So we still have one conditional in there. So do this:
2154 * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
2156 * 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),
2157 * 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.
2158 * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to
2160 shader_addline(ins->buffer,
2161 "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",
2162 src0_param.param_str, src1_param.param_str, src0_param.param_str, src3_param.param_str, dst_mask);
2165 /** Process the WINED3DSIO_DST instruction in GLSL:
2167 * dst.y = src0.x * src0.y
2171 static void shader_glsl_dst(const struct wined3d_shader_instruction *ins)
2173 glsl_src_param_t src0y_param;
2174 glsl_src_param_t src0z_param;
2175 glsl_src_param_t src1y_param;
2176 glsl_src_param_t src1w_param;
2179 shader_glsl_append_dst(ins->buffer, ins);
2180 shader_glsl_get_write_mask(ins->dst[0].token, dst_mask);
2182 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_1, &src0y_param);
2183 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_2, &src0z_param);
2184 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1], WINED3DSP_WRITEMASK_1, &src1y_param);
2185 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1], WINED3DSP_WRITEMASK_3, &src1w_param);
2187 shader_addline(ins->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
2188 src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
2191 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
2192 * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
2193 * can handle it. But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
2195 * dst.x = cos(src0.?)
2196 * dst.y = sin(src0.?)
2200 static void shader_glsl_sincos(const struct wined3d_shader_instruction *ins)
2202 glsl_src_param_t src0_param;
2205 write_mask = shader_glsl_append_dst(ins->buffer, ins);
2206 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2208 switch (write_mask) {
2209 case WINED3DSP_WRITEMASK_0:
2210 shader_addline(ins->buffer, "cos(%s));\n", src0_param.param_str);
2213 case WINED3DSP_WRITEMASK_1:
2214 shader_addline(ins->buffer, "sin(%s));\n", src0_param.param_str);
2217 case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
2218 shader_addline(ins->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
2222 ERR("Write mask should be .x, .y or .xy\n");
2227 /** Process the WINED3DSIO_LOOP instruction in GLSL:
2228 * Start a for() loop where src1.y is the initial value of aL,
2229 * increment aL by src1.z for a total of src1.x iterations.
2230 * Need to use a temporary variable for this operation.
2232 /* FIXME: I don't think nested loops will work correctly this way. */
2233 static void shader_glsl_loop(const struct wined3d_shader_instruction *ins)
2235 glsl_src_param_t src1_param;
2236 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->shader;
2237 DWORD regtype = shader_get_regtype(ins->src[1]);
2238 DWORD reg = ins->src[1] & WINED3DSP_REGNUM_MASK;
2239 const DWORD *control_values = NULL;
2240 const local_constant *constant;
2242 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2244 /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2245 * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2246 * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2249 if(regtype == WINED3DSPR_CONSTINT) {
2250 LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
2251 if(constant->idx == reg) {
2252 control_values = constant->value;
2258 if(control_values) {
2259 if(control_values[2] > 0) {
2260 shader_addline(ins->buffer, "for (aL%u = %d; aL%u < (%d * %d + %d); aL%u += %d) {\n",
2261 shader->baseShader.cur_loop_depth, control_values[1],
2262 shader->baseShader.cur_loop_depth, control_values[0], control_values[2], control_values[1],
2263 shader->baseShader.cur_loop_depth, control_values[2]);
2264 } else if(control_values[2] == 0) {
2265 shader_addline(ins->buffer, "for (aL%u = %d, tmpInt%u = 0; tmpInt%u < %d; tmpInt%u++) {\n",
2266 shader->baseShader.cur_loop_depth, control_values[1], shader->baseShader.cur_loop_depth,
2267 shader->baseShader.cur_loop_depth, control_values[0],
2268 shader->baseShader.cur_loop_depth);
2270 shader_addline(ins->buffer, "for (aL%u = %d; aL%u > (%d * %d + %d); aL%u += %d) {\n",
2271 shader->baseShader.cur_loop_depth, control_values[1],
2272 shader->baseShader.cur_loop_depth, control_values[0], control_values[2], control_values[1],
2273 shader->baseShader.cur_loop_depth, control_values[2]);
2276 shader_addline(ins->buffer, "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2277 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno,
2278 src1_param.reg_name, shader->baseShader.cur_loop_depth, src1_param.reg_name,
2279 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno, src1_param.reg_name);
2282 shader->baseShader.cur_loop_depth++;
2283 shader->baseShader.cur_loop_regno++;
2286 static void shader_glsl_end(const struct wined3d_shader_instruction *ins)
2288 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->shader;
2290 shader_addline(ins->buffer, "}\n");
2292 if (ins->handler_idx == WINED3DSIH_ENDLOOP)
2294 shader->baseShader.cur_loop_depth--;
2295 shader->baseShader.cur_loop_regno--;
2298 if (ins->handler_idx == WINED3DSIH_ENDREP)
2300 shader->baseShader.cur_loop_depth--;
2304 static void shader_glsl_rep(const struct wined3d_shader_instruction *ins)
2306 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->shader;
2307 glsl_src_param_t src0_param;
2309 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2310 shader_addline(ins->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2311 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2312 src0_param.param_str, shader->baseShader.cur_loop_depth);
2313 shader->baseShader.cur_loop_depth++;
2316 static void shader_glsl_if(const struct wined3d_shader_instruction *ins)
2318 glsl_src_param_t src0_param;
2320 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2321 shader_addline(ins->buffer, "if (%s) {\n", src0_param.param_str);
2324 static void shader_glsl_ifc(const struct wined3d_shader_instruction *ins)
2326 glsl_src_param_t src0_param;
2327 glsl_src_param_t src1_param;
2329 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2330 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2332 shader_addline(ins->buffer, "if (%s %s %s) {\n",
2333 src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2336 static void shader_glsl_else(const struct wined3d_shader_instruction *ins)
2338 shader_addline(ins->buffer, "} else {\n");
2341 static void shader_glsl_break(const struct wined3d_shader_instruction *ins)
2343 shader_addline(ins->buffer, "break;\n");
2346 /* FIXME: According to MSDN the compare is done per component. */
2347 static void shader_glsl_breakc(const struct wined3d_shader_instruction *ins)
2349 glsl_src_param_t src0_param;
2350 glsl_src_param_t src1_param;
2352 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2353 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2355 shader_addline(ins->buffer, "if (%s %s %s) break;\n",
2356 src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2359 static void shader_glsl_label(const struct wined3d_shader_instruction *ins)
2362 DWORD snum = (ins->src[0]) & WINED3DSP_REGNUM_MASK;
2363 shader_addline(ins->buffer, "}\n");
2364 shader_addline(ins->buffer, "void subroutine%u () {\n", snum);
2367 static void shader_glsl_call(const struct wined3d_shader_instruction *ins)
2369 DWORD snum = (ins->src[0]) & WINED3DSP_REGNUM_MASK;
2370 shader_addline(ins->buffer, "subroutine%u();\n", snum);
2373 static void shader_glsl_callnz(const struct wined3d_shader_instruction *ins)
2375 glsl_src_param_t src1_param;
2377 DWORD snum = (ins->src[0]) & WINED3DSP_REGNUM_MASK;
2378 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2379 shader_addline(ins->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, snum);
2382 /*********************************************
2383 * Pixel Shader Specific Code begins here
2384 ********************************************/
2385 static void pshader_glsl_tex(const struct wined3d_shader_instruction *ins)
2387 IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->shader;
2388 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2389 DWORD shader_version = ins->reg_maps->shader_version;
2390 glsl_sample_function_t sample_function;
2391 DWORD sample_flags = 0;
2394 DWORD mask = 0, swizzle;
2396 /* 1.0-1.4: Use destination register as sampler source.
2397 * 2.0+: Use provided sampler source. */
2398 if (shader_version < WINED3DPS_VERSION(2,0)) sampler_idx = ins->dst[0].register_idx;
2399 else sampler_idx = ins->src[1] & WINED3DSP_REGNUM_MASK;
2400 sampler_type = ins->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2402 if (shader_version < WINED3DPS_VERSION(1,4))
2404 DWORD flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2406 /* Projected cube textures don't make a lot of sense, the resulting coordinates stay the same. */
2407 if (flags & WINED3DTTFF_PROJECTED && sampler_type != WINED3DSTT_CUBE) {
2408 sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2409 switch (flags & ~WINED3DTTFF_PROJECTED) {
2410 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2411 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
2412 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
2413 case WINED3DTTFF_COUNT4:
2414 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
2418 else if (shader_version < WINED3DPS_VERSION(2,0))
2420 DWORD src_mod = ins->src[0] & WINED3DSP_SRCMOD_MASK;
2422 if (src_mod == WINED3DSPSM_DZ) {
2423 sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2424 mask = WINED3DSP_WRITEMASK_2;
2425 } else if (src_mod == WINED3DSPSM_DW) {
2426 sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2427 mask = WINED3DSP_WRITEMASK_3;
2430 if (ins->flags & WINED3DSI_TEXLD_PROJECT)
2432 /* ps 2.0 texldp instruction always divides by the fourth component. */
2433 sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
2434 mask = WINED3DSP_WRITEMASK_3;
2438 if(deviceImpl->stateBlock->textures[sampler_idx] &&
2439 IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2440 sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
2443 shader_glsl_get_sample_function(sampler_type, sample_flags, &sample_function);
2444 mask |= sample_function.coord_mask;
2446 if (shader_version < WINED3DPS_VERSION(2,0)) swizzle = WINED3DVS_NOSWIZZLE;
2447 else swizzle = ins->src[1] & WINED3DSP_SWIZZLE_MASK;
2449 /* 1.0-1.3: Use destination register as coordinate source.
2450 1.4+: Use provided coordinate source register. */
2451 if (shader_version < WINED3DPS_VERSION(1,4))
2454 shader_glsl_get_write_mask(mask, coord_mask);
2455 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL,
2456 "T%u%s", sampler_idx, coord_mask);
2458 glsl_src_param_t coord_param;
2459 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], mask, &coord_param);
2460 if (ins->flags & WINED3DSI_TEXLD_BIAS)
2462 glsl_src_param_t bias;
2463 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_3, &bias);
2464 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, bias.param_str,
2465 "%s", coord_param.param_str);
2467 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL,
2468 "%s", coord_param.param_str);
2473 static void shader_glsl_texldl(const struct wined3d_shader_instruction *ins)
2475 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->shader;
2476 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2477 glsl_sample_function_t sample_function;
2478 glsl_src_param_t coord_param, lod_param;
2479 DWORD sample_flags = WINED3D_GLSL_SAMPLE_LOD;
2482 DWORD swizzle = ins->src[1] & WINED3DSP_SWIZZLE_MASK;
2484 sampler_idx = ins->src[1] & WINED3DSP_REGNUM_MASK;
2485 sampler_type = ins->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2486 if(deviceImpl->stateBlock->textures[sampler_idx] &&
2487 IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2488 sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
2490 shader_glsl_get_sample_function(sampler_type, sample_flags, &sample_function);
2491 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], sample_function.coord_mask, &coord_param);
2493 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_3, &lod_param);
2495 if (shader_is_pshader_version(ins->reg_maps->shader_version))
2497 /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
2498 * However, they seem to work just fine in fragment shaders as well. */
2499 WARN("Using %s in fragment shader.\n", sample_function.name);
2501 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, lod_param.param_str,
2502 "%s", coord_param.param_str);
2505 static void pshader_glsl_texcoord(const struct wined3d_shader_instruction *ins)
2507 /* FIXME: Make this work for more than just 2D textures */
2508 SHADER_BUFFER *buffer = ins->buffer;
2512 write_mask = shader_glsl_append_dst(ins->buffer, ins);
2513 shader_glsl_get_write_mask(write_mask, dst_mask);
2515 if (ins->reg_maps->shader_version != WINED3DPS_VERSION(1,4))
2517 DWORD reg = ins->dst[0].register_idx;
2518 shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n", reg, dst_mask);
2520 DWORD reg = ins->src[0] & WINED3DSP_REGNUM_MASK;
2521 DWORD src_mod = ins->src[0] & WINED3DSP_SRCMOD_MASK;
2522 char dst_swizzle[6];
2524 shader_glsl_get_swizzle(ins->src[0], FALSE, write_mask, dst_swizzle);
2526 if (src_mod == WINED3DSPSM_DZ) {
2527 glsl_src_param_t div_param;
2528 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2529 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_2, &div_param);
2531 if (mask_size > 1) {
2532 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2534 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2536 } else if (src_mod == WINED3DSPSM_DW) {
2537 glsl_src_param_t div_param;
2538 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2539 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_3, &div_param);
2541 if (mask_size > 1) {
2542 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2544 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2547 shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
2552 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
2553 * Take a 3-component dot product of the TexCoord[dstreg] and src,
2554 * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2555 static void pshader_glsl_texdp3tex(const struct wined3d_shader_instruction *ins)
2557 glsl_src_param_t src0_param;
2558 glsl_sample_function_t sample_function;
2559 DWORD sampler_idx = ins->dst[0].register_idx;
2560 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2561 DWORD sampler_type = ins->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2564 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], src_mask, &src0_param);
2566 /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
2567 * scalar, and projected sampling would require 4.
2569 * It is a dependent read - not valid with conditional NP2 textures
2571 shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
2572 mask_size = shader_glsl_get_write_mask_size(sample_function.coord_mask);
2577 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DVS_NOSWIZZLE, NULL,
2578 "dot(gl_TexCoord[%u].xyz, %s)", sampler_idx, src0_param.param_str);
2582 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DVS_NOSWIZZLE, NULL,
2583 "vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0)", sampler_idx, src0_param.param_str);
2587 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DVS_NOSWIZZLE, NULL,
2588 "vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0)", sampler_idx, src0_param.param_str);
2592 FIXME("Unexpected mask size %u\n", mask_size);
2597 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
2598 * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2599 static void pshader_glsl_texdp3(const struct wined3d_shader_instruction *ins)
2601 glsl_src_param_t src0_param;
2602 DWORD dstreg = ins->dst[0].register_idx;
2603 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2605 unsigned int mask_size;
2607 dst_mask = shader_glsl_append_dst(ins->buffer, ins);
2608 mask_size = shader_glsl_get_write_mask_size(dst_mask);
2609 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], src_mask, &src0_param);
2611 if (mask_size > 1) {
2612 shader_addline(ins->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
2614 shader_addline(ins->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
2618 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
2619 * Calculate the depth as dst.x / dst.y */
2620 static void pshader_glsl_texdepth(const struct wined3d_shader_instruction *ins)
2622 glsl_dst_param_t dst_param;
2624 shader_glsl_add_dst_param(ins, ins->dst[0].token, 0, &dst_param);
2626 /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
2627 * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
2628 * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
2629 * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
2632 shader_addline(ins->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n",
2633 dst_param.reg_name, dst_param.reg_name);
2636 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
2637 * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2638 * Calculate tmp0.y = TexCoord[dstreg] . src.xyz; (tmp0.x has already been calculated)
2639 * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2641 static void pshader_glsl_texm3x2depth(const struct wined3d_shader_instruction *ins)
2643 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2644 DWORD dstreg = ins->dst[0].register_idx;
2645 glsl_src_param_t src0_param;
2647 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], src_mask, &src0_param);
2649 shader_addline(ins->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
2650 shader_addline(ins->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
2653 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
2654 * Calculate the 1st of a 2-row matrix multiplication. */
2655 static void pshader_glsl_texm3x2pad(const struct wined3d_shader_instruction *ins)
2657 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2658 DWORD reg = ins->dst[0].register_idx;
2659 SHADER_BUFFER *buffer = ins->buffer;
2660 glsl_src_param_t src0_param;
2662 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], src_mask, &src0_param);
2663 shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2666 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
2667 * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
2668 static void pshader_glsl_texm3x3pad(const struct wined3d_shader_instruction *ins)
2670 IWineD3DPixelShaderImpl *shader = (IWineD3DPixelShaderImpl *)ins->shader;
2671 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2672 DWORD reg = ins->dst[0].register_idx;
2673 SHADER_BUFFER *buffer = ins->buffer;
2674 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2675 glsl_src_param_t src0_param;
2677 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], src_mask, &src0_param);
2678 shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + current_state->current_row, reg, src0_param.param_str);
2679 current_state->texcoord_w[current_state->current_row++] = reg;
2682 static void pshader_glsl_texm3x2tex(const struct wined3d_shader_instruction *ins)
2684 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2685 DWORD reg = ins->dst[0].register_idx;
2686 SHADER_BUFFER *buffer = ins->buffer;
2687 glsl_src_param_t src0_param;
2688 DWORD sampler_type = ins->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2689 glsl_sample_function_t sample_function;
2691 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], src_mask, &src0_param);
2692 shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2694 shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
2696 /* Sample the texture using the calculated coordinates */
2697 shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DVS_NOSWIZZLE, NULL, "tmp0.xy");
2700 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
2701 * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
2702 static void pshader_glsl_texm3x3tex(const struct wined3d_shader_instruction *ins)
2704 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2705 glsl_src_param_t src0_param;
2706 DWORD reg = ins->dst[0].register_idx;
2707 IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->shader;
2708 SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2709 DWORD sampler_type = ins->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2710 glsl_sample_function_t sample_function;
2712 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], src_mask, &src0_param);
2713 shader_addline(ins->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2715 /* Dependent read, not valid with conditional NP2 */
2716 shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
2718 /* Sample the texture using the calculated coordinates */
2719 shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DVS_NOSWIZZLE, NULL, "tmp0.xyz");
2721 current_state->current_row = 0;
2724 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
2725 * Perform the 3rd row of a 3x3 matrix multiply */
2726 static void pshader_glsl_texm3x3(const struct wined3d_shader_instruction *ins)
2728 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2729 glsl_src_param_t src0_param;
2731 DWORD reg = ins->dst[0].register_idx;
2732 IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->shader;
2733 SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2735 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], src_mask, &src0_param);
2737 shader_glsl_append_dst(ins->buffer, ins);
2738 shader_glsl_get_write_mask(ins->dst[0].token, dst_mask);
2739 shader_addline(ins->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
2741 current_state->current_row = 0;
2744 /** Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
2745 * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
2746 static void pshader_glsl_texm3x3spec(const struct wined3d_shader_instruction *ins)
2748 IWineD3DPixelShaderImpl *shader = (IWineD3DPixelShaderImpl *)ins->shader;
2749 DWORD reg = ins->dst[0].register_idx;
2750 glsl_src_param_t src0_param;
2751 glsl_src_param_t src1_param;
2752 SHADER_BUFFER *buffer = ins->buffer;
2753 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2754 DWORD stype = ins->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2755 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2756 glsl_sample_function_t sample_function;
2758 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], src_mask, &src0_param);
2759 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1], src_mask, &src1_param);
2761 /* Perform the last matrix multiply operation */
2762 shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2763 /* Reflection calculation */
2764 shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
2766 /* Dependent read, not valid with conditional NP2 */
2767 shader_glsl_get_sample_function(stype, 0, &sample_function);
2769 /* Sample the texture */
2770 shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DVS_NOSWIZZLE, NULL, "tmp0.xyz");
2772 current_state->current_row = 0;
2775 /** Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
2776 * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
2777 static void pshader_glsl_texm3x3vspec(const struct wined3d_shader_instruction *ins)
2779 IWineD3DPixelShaderImpl *shader = (IWineD3DPixelShaderImpl *)ins->shader;
2780 DWORD reg = ins->dst[0].register_idx;
2781 SHADER_BUFFER *buffer = ins->buffer;
2782 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2783 glsl_src_param_t src0_param;
2784 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2785 DWORD sampler_type = ins->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2786 glsl_sample_function_t sample_function;
2788 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], src_mask, &src0_param);
2790 /* Perform the last matrix multiply operation */
2791 shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
2793 /* Construct the eye-ray vector from w coordinates */
2794 shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
2795 current_state->texcoord_w[0], current_state->texcoord_w[1], reg);
2796 shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
2798 /* Dependent read, not valid with conditional NP2 */
2799 shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
2801 /* Sample the texture using the calculated coordinates */
2802 shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DVS_NOSWIZZLE, NULL, "tmp0.xyz");
2804 current_state->current_row = 0;
2807 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
2808 * Apply a fake bump map transform.
2809 * texbem is pshader <= 1.3 only, this saves a few version checks
2811 static void pshader_glsl_texbem(const struct wined3d_shader_instruction *ins)
2813 IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)ins->shader;
2814 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2815 glsl_sample_function_t sample_function;
2816 glsl_src_param_t coord_param;
2823 sampler_idx = ins->dst[0].register_idx;
2824 flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2826 sampler_type = ins->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2827 /* Dependent read, not valid with conditional NP2 */
2828 shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
2829 mask = sample_function.coord_mask;
2831 shader_glsl_get_write_mask(mask, coord_mask);
2833 /* with projective textures, texbem only divides the static texture coord, not the displacement,
2834 * so we can't let the GL handle this.
2836 if (flags & WINED3DTTFF_PROJECTED) {
2838 char coord_div_mask[3];
2839 switch (flags & ~WINED3DTTFF_PROJECTED) {
2840 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2841 case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
2842 case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
2843 case WINED3DTTFF_COUNT4:
2844 case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
2846 shader_glsl_get_write_mask(div_mask, coord_div_mask);
2847 shader_addline(ins->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
2850 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0],
2851 WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &coord_param);
2853 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DVS_NOSWIZZLE, NULL,
2854 "T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s", sampler_idx, coord_mask, sampler_idx,
2855 coord_param.param_str, coord_mask);
2857 if (ins->handler_idx == WINED3DSIH_TEXBEML)
2859 glsl_src_param_t luminance_param;
2860 glsl_dst_param_t dst_param;
2862 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_2, &luminance_param);
2863 shader_glsl_add_dst_param(ins, ins->dst[0].token, ins->dst[0].addr_token, &dst_param);
2865 shader_addline(ins->buffer, "%s%s *= (%s * luminancescale%d + luminanceoffset%d);\n",
2866 dst_param.reg_name, dst_param.mask_str,
2867 luminance_param.param_str, sampler_idx, sampler_idx);
2871 static void pshader_glsl_bem(const struct wined3d_shader_instruction *ins)
2873 glsl_src_param_t src0_param, src1_param;
2874 DWORD sampler_idx = ins->dst[0].register_idx;
2876 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0],
2877 WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
2878 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1],
2879 WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
2881 shader_glsl_append_dst(ins->buffer, ins);
2882 shader_addline(ins->buffer, "%s + bumpenvmat%d * %s);\n",
2883 src0_param.param_str, sampler_idx, src1_param.param_str);
2886 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
2887 * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
2888 static void pshader_glsl_texreg2ar(const struct wined3d_shader_instruction *ins)
2890 glsl_src_param_t src0_param;
2891 DWORD sampler_idx = ins->dst[0].register_idx;
2892 DWORD sampler_type = ins->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2893 glsl_sample_function_t sample_function;
2895 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2897 shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
2898 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DVS_NOSWIZZLE, NULL,
2899 "%s.wx", src0_param.reg_name);
2902 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
2903 * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
2904 static void pshader_glsl_texreg2gb(const struct wined3d_shader_instruction *ins)
2906 glsl_src_param_t src0_param;
2907 DWORD sampler_idx = ins->dst[0].register_idx;
2908 DWORD sampler_type = ins->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2909 glsl_sample_function_t sample_function;
2911 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2913 shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
2914 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DVS_NOSWIZZLE, NULL,
2915 "%s.yz", src0_param.reg_name);
2918 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
2919 * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
2920 static void pshader_glsl_texreg2rgb(const struct wined3d_shader_instruction *ins)
2922 glsl_src_param_t src0_param;
2923 DWORD sampler_idx = ins->dst[0].register_idx;
2924 DWORD sampler_type = ins->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2925 glsl_sample_function_t sample_function;
2927 /* Dependent read, not valid with conditional NP2 */
2928 shader_glsl_get_sample_function(sampler_type, 0, &sample_function);
2929 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0], sample_function.coord_mask, &src0_param);
2931 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DVS_NOSWIZZLE, NULL,
2932 "%s", src0_param.param_str);
2935 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
2936 * If any of the first 3 components are < 0, discard this pixel */
2937 static void pshader_glsl_texkill(const struct wined3d_shader_instruction *ins)
2939 glsl_dst_param_t dst_param;
2941 /* The argument is a destination parameter, and no writemasks are allowed */
2942 shader_glsl_add_dst_param(ins, ins->dst[0].token, 0, &dst_param);
2943 if ((ins->reg_maps->shader_version >= WINED3DPS_VERSION(2,0)))
2945 /* 2.0 shaders compare all 4 components in texkill */
2946 shader_addline(ins->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
2948 /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
2949 * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
2950 * 4 components are defined, only the first 3 are used
2952 shader_addline(ins->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
2956 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
2957 * dst = dot2(src0, src1) + src2 */
2958 static void pshader_glsl_dp2add(const struct wined3d_shader_instruction *ins)
2960 glsl_src_param_t src0_param;
2961 glsl_src_param_t src1_param;
2962 glsl_src_param_t src2_param;
2964 unsigned int mask_size;
2966 write_mask = shader_glsl_append_dst(ins->buffer, ins);
2967 mask_size = shader_glsl_get_write_mask_size(write_mask);
2969 shader_glsl_add_src_param(ins, ins->src[0], ins->src_addr[0],
2970 WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
2971 shader_glsl_add_src_param(ins, ins->src[1], ins->src_addr[1],
2972 WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
2973 shader_glsl_add_src_param(ins, ins->src[2], ins->src_addr[2],
2974 WINED3DSP_WRITEMASK_0, &src2_param);
2976 if (mask_size > 1) {
2977 shader_addline(ins->buffer, "vec%d(dot(%s, %s) + %s));\n",
2978 mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
2980 shader_addline(ins->buffer, "dot(%s, %s) + %s);\n",
2981 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2985 static void pshader_glsl_input_pack(SHADER_BUFFER* buffer, const struct semantic* semantics_in,
2986 IWineD3DPixelShader *iface, enum vertexprocessing_mode vertexprocessing)
2989 IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *) iface;
2991 for (i = 0; i < MAX_REG_INPUT; i++) {
2993 DWORD usage_token = semantics_in[i].usage;
2994 DWORD register_token = semantics_in[i].reg;
2995 DWORD usage, usage_idx;
2999 if (!usage_token) continue;
3000 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
3001 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
3002 shader_glsl_get_write_mask(register_token, reg_mask);
3006 case WINED3DDECLUSAGE_TEXCOORD:
3007 if(usage_idx < 8 && vertexprocessing == pretransformed) {
3008 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
3009 This->input_reg_map[i], reg_mask, usage_idx, reg_mask);
3011 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3012 This->input_reg_map[i], reg_mask, reg_mask);
3016 case WINED3DDECLUSAGE_COLOR:
3018 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
3019 This->input_reg_map[i], reg_mask, reg_mask);
3020 else if (usage_idx == 1)
3021 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
3022 This->input_reg_map[i], reg_mask, reg_mask);
3024 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3025 This->input_reg_map[i], reg_mask, reg_mask);
3029 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3030 This->input_reg_map[i], reg_mask, reg_mask);
3035 /*********************************************
3036 * Vertex Shader Specific Code begins here
3037 ********************************************/
3039 static void add_glsl_program_entry(struct shader_glsl_priv *priv, struct glsl_shader_prog_link *entry) {
3040 glsl_program_key_t *key;
3042 key = HeapAlloc(GetProcessHeap(), 0, sizeof(glsl_program_key_t));
3043 key->vshader = entry->vshader;
3044 key->pshader = entry->pshader;
3045 key->vs_args = entry->vs_args;
3046 key->ps_args = entry->ps_args;
3048 hash_table_put(priv->glsl_program_lookup, key, entry);
3051 static struct glsl_shader_prog_link *get_glsl_program_entry(struct shader_glsl_priv *priv,
3052 IWineD3DVertexShader *vshader, IWineD3DPixelShader *pshader, struct vs_compile_args *vs_args,
3053 struct ps_compile_args *ps_args) {
3054 glsl_program_key_t key;
3056 key.vshader = vshader;
3057 key.pshader = pshader;
3058 key.vs_args = *vs_args;
3059 key.ps_args = *ps_args;
3061 return hash_table_get(priv->glsl_program_lookup, &key);
3064 static void delete_glsl_program_entry(struct shader_glsl_priv *priv, const WineD3D_GL_Info *gl_info,
3065 struct glsl_shader_prog_link *entry)
3067 glsl_program_key_t *key;
3069 key = HeapAlloc(GetProcessHeap(), 0, sizeof(glsl_program_key_t));
3070 key->vshader = entry->vshader;
3071 key->pshader = entry->pshader;
3072 key->vs_args = entry->vs_args;
3073 key->ps_args = entry->ps_args;
3074 hash_table_remove(priv->glsl_program_lookup, key);
3076 GL_EXTCALL(glDeleteObjectARB(entry->programId));
3077 if (entry->vshader) list_remove(&entry->vshader_entry);
3078 if (entry->pshader) list_remove(&entry->pshader_entry);
3079 HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
3080 HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
3081 HeapFree(GetProcessHeap(), 0, entry);
3084 static void handle_ps3_input(SHADER_BUFFER *buffer, const struct semantic *semantics_in,
3085 const struct semantic *semantics_out, const WineD3D_GL_Info *gl_info, const DWORD *map)
3088 DWORD usage_token, usage_token_out;
3089 DWORD register_token, register_token_out;
3090 DWORD usage, usage_idx, usage_out, usage_idx_out;
3093 DWORD in_count = GL_LIMITS(glsl_varyings) / 4;
3094 char reg_mask[6], reg_mask_out[6];
3095 char destination[50];
3097 set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
3099 if (!semantics_out) {
3100 /* Save gl_FrontColor & gl_FrontSecondaryColor before overwriting them. */
3101 shader_addline(buffer, "vec4 front_color = gl_FrontColor;\n");
3102 shader_addline(buffer, "vec4 front_secondary_color = gl_FrontSecondaryColor;\n");
3105 for(i = 0; i < MAX_REG_INPUT; i++) {
3106 usage_token = semantics_in[i].usage;
3107 if (!usage_token) continue;
3110 if (in_idx >= (in_count + 2)) {
3111 FIXME("More input varyings declared than supported, expect issues\n");
3114 else if (map[i] == ~0U)
3116 /* Declared, but not read register */
3120 if (in_idx == in_count) {
3121 sprintf(destination, "gl_FrontColor");
3122 } else if (in_idx == in_count + 1) {
3123 sprintf(destination, "gl_FrontSecondaryColor");
3125 sprintf(destination, "IN[%u]", in_idx);
3128 register_token = semantics_in[i].reg;
3130 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
3131 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
3132 set[map[i]] = shader_glsl_get_write_mask(register_token, reg_mask);
3134 if(!semantics_out) {
3136 case WINED3DDECLUSAGE_COLOR:
3138 shader_addline(buffer, "%s%s = front_color%s;\n",
3139 destination, reg_mask, reg_mask);
3140 else if (usage_idx == 1)
3141 shader_addline(buffer, "%s%s = front_secondary_color%s;\n",
3142 destination, reg_mask, reg_mask);
3144 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3145 destination, reg_mask, reg_mask);
3148 case WINED3DDECLUSAGE_TEXCOORD:
3149 if (usage_idx < 8) {
3150 shader_addline(buffer, "%s%s = gl_TexCoord[%u]%s;\n",
3151 destination, reg_mask, usage_idx, reg_mask);
3153 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3154 destination, reg_mask, reg_mask);
3158 case WINED3DDECLUSAGE_FOG:
3159 shader_addline(buffer, "%s%s = vec4(gl_FogFragCoord, 0.0, 0.0, 0.0)%s;\n",
3160 destination, reg_mask, reg_mask);
3164 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3165 destination, reg_mask, reg_mask);
3169 for(j = 0; j < MAX_REG_OUTPUT; j++) {
3170 usage_token_out = semantics_out[j].usage;
3171 if (!usage_token_out) continue;
3172 register_token_out = semantics_out[j].reg;
3174 usage_out = (usage_token_out & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
3175 usage_idx_out = (usage_token_out & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
3176 shader_glsl_get_write_mask(register_token_out, reg_mask_out);
3178 if(usage == usage_out &&
3179 usage_idx == usage_idx_out) {
3180 shader_addline(buffer, "%s%s = OUT[%u]%s;\n",
3181 destination, reg_mask, j, reg_mask);
3186 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3187 destination, reg_mask, reg_mask);
3192 /* This is solely to make the compiler / linker happy and avoid warning about undefined
3193 * varyings. It shouldn't result in any real code executed on the GPU, since all read
3194 * input varyings are assigned above, if the optimizer works properly.
3196 for(i = 0; i < in_count + 2; i++) {
3197 if(set[i] != WINED3DSP_WRITEMASK_ALL) {
3198 unsigned int size = 0;
3199 memset(reg_mask, 0, sizeof(reg_mask));
3200 if(!(set[i] & WINED3DSP_WRITEMASK_0)) {
3201 reg_mask[size] = 'x';
3204 if(!(set[i] & WINED3DSP_WRITEMASK_1)) {
3205 reg_mask[size] = 'y';
3208 if(!(set[i] & WINED3DSP_WRITEMASK_2)) {
3209 reg_mask[size] = 'z';
3212 if(!(set[i] & WINED3DSP_WRITEMASK_3)) {
3213 reg_mask[size] = 'w';
3217 if (i == in_count) {
3218 sprintf(destination, "gl_FrontColor");
3219 } else if (i == in_count + 1) {
3220 sprintf(destination, "gl_FrontSecondaryColor");
3222 sprintf(destination, "IN[%u]", i);
3226 shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
3228 shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
3233 HeapFree(GetProcessHeap(), 0, set);
3236 static GLhandleARB generate_param_reorder_function(IWineD3DVertexShader *vertexshader,
3237 IWineD3DPixelShader *pixelshader, const WineD3D_GL_Info *gl_info)
3239 GLhandleARB ret = 0;
3240 IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) vertexshader;
3241 IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pixelshader;
3242 IWineD3DDeviceImpl *device;
3243 DWORD vs_major = WINED3DSHADER_VERSION_MAJOR(vs->baseShader.reg_maps.shader_version);
3244 DWORD ps_major = ps ? WINED3DSHADER_VERSION_MAJOR(ps->baseShader.reg_maps.shader_version) : 0;
3246 SHADER_BUFFER buffer;
3248 DWORD register_token;
3249 DWORD usage, usage_idx, writemask;
3251 const struct semantic *semantics_out, *semantics_in;
3253 shader_buffer_init(&buffer);
3255 shader_addline(&buffer, "#version 120\n");
3257 if(vs_major < 3 && ps_major < 3) {
3258 /* That one is easy: The vertex shader writes to the builtin varyings, the pixel shader reads from them.
3259 * Take care about the texcoord .w fixup though if we're using the fixed function fragment pipeline
3261 device = (IWineD3DDeviceImpl *) vs->baseShader.device;
3262 if((GLINFO_LOCATION).set_texcoord_w && ps_major == 0 && vs_major > 0 &&
3263 !device->frag_pipe->ffp_proj_control) {
3264 shader_addline(&buffer, "void order_ps_input() {\n");
3265 for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
3266 if(vs->baseShader.reg_maps.texcoord_mask[i] != 0 &&
3267 vs->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
3268 shader_addline(&buffer, "gl_TexCoord[%u].w = 1.0;\n", i);
3271 shader_addline(&buffer, "}\n");
3273 shader_addline(&buffer, "void order_ps_input() { /* do nothing */ }\n");
3275 } else if(ps_major < 3 && vs_major >= 3) {
3276 /* The vertex shader writes to its own varyings, the pixel shader needs them in the builtin ones */
3277 semantics_out = vs->semantics_out;
3279 shader_addline(&buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3280 for(i = 0; i < MAX_REG_OUTPUT; i++) {
3281 usage_token = semantics_out[i].usage;
3282 if (!usage_token) continue;
3283 register_token = semantics_out[i].reg;
3285 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
3286 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
3287 writemask = shader_glsl_get_write_mask(register_token, reg_mask);
3290 case WINED3DDECLUSAGE_COLOR:
3292 shader_addline(&buffer, "gl_FrontColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3293 else if (usage_idx == 1)
3294 shader_addline(&buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3297 case WINED3DDECLUSAGE_POSITION:
3298 shader_addline(&buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3301 case WINED3DDECLUSAGE_TEXCOORD:
3302 if (usage_idx < 8) {
3303 if(!(GLINFO_LOCATION).set_texcoord_w || ps_major > 0) writemask |= WINED3DSP_WRITEMASK_3;
3305 shader_addline(&buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
3306 usage_idx, reg_mask, i, reg_mask);
3307 if(!(writemask & WINED3DSP_WRITEMASK_3)) {
3308 shader_addline(&buffer, "gl_TexCoord[%u].w = 1.0;\n", usage_idx);
3313 case WINED3DDECLUSAGE_PSIZE:
3314 shader_addline(&buffer, "gl_PointSize = OUT[%u].x;\n", i);
3317 case WINED3DDECLUSAGE_FOG:
3318 shader_addline(&buffer, "gl_FogFragCoord = OUT[%u].%c;\n", i, reg_mask[1]);
3325 shader_addline(&buffer, "}\n");
3327 } else if(ps_major >= 3 && vs_major >= 3) {
3328 semantics_out = vs->semantics_out;
3329 semantics_in = ps->semantics_in;
3331 /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3332 shader_addline(&buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
3333 shader_addline(&buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3335 /* First, sort out position and point size. Those are not passed to the pixel shader */
3336 for(i = 0; i < MAX_REG_OUTPUT; i++) {
3337 usage_token = semantics_out[i].usage;
3338 if (!usage_token) continue;
3339 register_token = semantics_out[i].reg;
3341 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
3342 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
3343 shader_glsl_get_write_mask(register_token, reg_mask);
3346 case WINED3DDECLUSAGE_POSITION:
3347 shader_addline(&buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3350 case WINED3DDECLUSAGE_PSIZE:
3351 shader_addline(&buffer, "gl_PointSize = OUT[%u].x;\n", i);
3359 /* Then, fix the pixel shader input */
3360 handle_ps3_input(&buffer, semantics_in, semantics_out, gl_info, ps->input_reg_map);
3362 shader_addline(&buffer, "}\n");
3363 } else if(ps_major >= 3 && vs_major < 3) {
3364 semantics_in = ps->semantics_in;
3366 shader_addline(&buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
3367 shader_addline(&buffer, "void order_ps_input() {\n");
3368 /* The vertex shader wrote to the builtin varyings. There is no need to figure out position and
3369 * point size, but we depend on the optimizers kindness to find out that the pixel shader doesn't
3370 * read gl_TexCoord and gl_ColorX, otherwise we'll run out of varyings
3372 handle_ps3_input(&buffer, semantics_in, NULL, gl_info, ps->input_reg_map);
3373 shader_addline(&buffer, "}\n");
3375 ERR("Unexpected vertex and pixel shader version condition: vs: %d, ps: %d\n", vs_major, ps_major);
3378 ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3379 checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3380 GL_EXTCALL(glShaderSourceARB(ret, 1, (const char**)&buffer.buffer, NULL));
3381 checkGLcall("glShaderSourceARB(ret, 1, &buffer.buffer, NULL)");
3382 GL_EXTCALL(glCompileShaderARB(ret));
3383 checkGLcall("glCompileShaderARB(ret)");
3385 shader_buffer_free(&buffer);
3389 static void hardcode_local_constants(IWineD3DBaseShaderImpl *shader, const WineD3D_GL_Info *gl_info,
3390 GLhandleARB programId, char prefix)
3392 const local_constant *lconst;
3397 LIST_FOR_EACH_ENTRY(lconst, &shader->baseShader.constantsF, local_constant, entry) {
3398 value = (const float *)lconst->value;
3399 snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
3400 tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3401 GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
3403 checkGLcall("Hardcoding local constants\n");
3406 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
3407 * It sets the programId on the current StateBlock (because it should be called
3408 * inside of the DrawPrimitive() part of the render loop).
3410 * If a program for the given combination does not exist, create one, and store
3411 * the program in the hash table. If it creates a program, it will link the
3412 * given objects, too.
3414 static void set_glsl_shader_program(IWineD3DDevice *iface, BOOL use_ps, BOOL use_vs) {
3415 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3416 struct shader_glsl_priv *priv = This->shader_priv;
3417 const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3418 IWineD3DPixelShader *pshader = This->stateBlock->pixelShader;
3419 IWineD3DVertexShader *vshader = This->stateBlock->vertexShader;
3420 struct glsl_shader_prog_link *entry = NULL;
3421 GLhandleARB programId = 0;
3422 GLhandleARB reorder_shader_id = 0;
3425 GLhandleARB vshader_id, pshader_id;
3426 struct ps_compile_args ps_compile_args;
3427 struct vs_compile_args vs_compile_args;
3430 find_vs_compile_args((IWineD3DVertexShaderImpl*)This->stateBlock->vertexShader, This->stateBlock, &vs_compile_args);
3432 /* FIXME: Do we really have to spend CPU cycles to generate a few zeroed bytes? */
3433 memset(&vs_compile_args, 0, sizeof(vs_compile_args));
3436 find_ps_compile_args((IWineD3DPixelShaderImpl*)This->stateBlock->pixelShader, This->stateBlock, &ps_compile_args);
3438 /* FIXME: Do we really have to spend CPU cycles to generate a few zeroed bytes? */
3439 memset(&ps_compile_args, 0, sizeof(ps_compile_args));
3441 entry = get_glsl_program_entry(priv, vshader, pshader, &vs_compile_args, &ps_compile_args);
3443 priv->glsl_program = entry;
3447 /* If we get to this point, then no matching program exists, so we create one */
3448 programId = GL_EXTCALL(glCreateProgramObjectARB());
3449 TRACE("Created new GLSL shader program %u\n", programId);
3451 /* Create the entry */
3452 entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
3453 entry->programId = programId;
3454 entry->vshader = vshader;
3455 entry->pshader = pshader;
3456 entry->vs_args = vs_compile_args;
3457 entry->ps_args = ps_compile_args;
3458 entry->constant_version = 0;
3459 /* Add the hash table entry */
3460 add_glsl_program_entry(priv, entry);
3462 /* Set the current program */
3463 priv->glsl_program = entry;
3466 vshader_id = find_gl_vshader((IWineD3DVertexShaderImpl *) vshader, &vs_compile_args);
3471 /* Attach GLSL vshader */
3473 const unsigned int max_attribs = 16; /* TODO: Will this always be the case? It is at the moment... */
3476 reorder_shader_id = generate_param_reorder_function(vshader, pshader, gl_info);
3477 TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
3478 GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
3479 checkGLcall("glAttachObjectARB");
3480 /* Flag the reorder function for deletion, then it will be freed automatically when the program
3483 GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
3485 TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
3486 GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
3487 checkGLcall("glAttachObjectARB");
3489 /* Bind vertex attributes to a corresponding index number to match
3490 * the same index numbers as ARB_vertex_programs (makes loading
3491 * vertex attributes simpler). With this method, we can use the
3492 * exact same code to load the attributes later for both ARB and
3495 * We have to do this here because we need to know the Program ID
3496 * in order to make the bindings work, and it has to be done prior
3497 * to linking the GLSL program. */
3498 for (i = 0; i < max_attribs; ++i) {
3499 if (((IWineD3DBaseShaderImpl*)vshader)->baseShader.reg_maps.attributes[i]) {
3500 snprintf(tmp_name, sizeof(tmp_name), "attrib%i", i);
3501 GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
3504 checkGLcall("glBindAttribLocationARB");
3506 list_add_head(&((IWineD3DBaseShaderImpl *)vshader)->baseShader.linked_programs, &entry->vshader_entry);
3510 pshader_id = find_gl_pshader((IWineD3DPixelShaderImpl *) pshader, &ps_compile_args);
3515 /* Attach GLSL pshader */
3517 TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
3518 GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
3519 checkGLcall("glAttachObjectARB");
3521 list_add_head(&((IWineD3DBaseShaderImpl *)pshader)->baseShader.linked_programs, &entry->pshader_entry);
3524 /* Link the program */
3525 TRACE("Linking GLSL shader program %u\n", programId);
3526 GL_EXTCALL(glLinkProgramARB(programId));
3527 print_glsl_info_log(&GLINFO_LOCATION, programId);
3529 entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0, sizeof(GLhandleARB) * GL_LIMITS(vshader_constantsF));
3530 for (i = 0; i < GL_LIMITS(vshader_constantsF); ++i) {
3531 snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
3532 entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3534 for (i = 0; i < MAX_CONST_I; ++i) {
3535 snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
3536 entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3538 entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0, sizeof(GLhandleARB) * GL_LIMITS(pshader_constantsF));
3539 for (i = 0; i < GL_LIMITS(pshader_constantsF); ++i) {
3540 snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
3541 entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3543 for (i = 0; i < MAX_CONST_I; ++i) {
3544 snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
3545 entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3549 for(i = 0; i < ((IWineD3DPixelShaderImpl*)pshader)->numbumpenvmatconsts; i++) {
3551 sprintf(name, "bumpenvmat%d", ((IWineD3DPixelShaderImpl*)pshader)->bumpenvmatconst[i].texunit);
3552 entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3553 sprintf(name, "luminancescale%d", ((IWineD3DPixelShaderImpl*)pshader)->luminanceconst[i].texunit);
3554 entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3555 sprintf(name, "luminanceoffset%d", ((IWineD3DPixelShaderImpl*)pshader)->luminanceconst[i].texunit);
3556 entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3560 if (use_ps && ps_compile_args.texrect_fixup) {
3562 for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
3563 if (ps_compile_args.texrect_fixup & (1 << i)) {
3564 sprintf(name, "PsamplerRectFixup%u", i);
3565 entry->rectFixup_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3567 entry->rectFixup_location[i] = -1;
3572 entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
3573 entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
3574 checkGLcall("Find glsl program uniform locations");
3577 && WINED3DSHADER_VERSION_MAJOR(((IWineD3DPixelShaderImpl *)pshader)->baseShader.reg_maps.shader_version) >= 3
3578 && ((IWineD3DPixelShaderImpl *)pshader)->declared_in_count > GL_LIMITS(glsl_varyings) / 4)
3580 TRACE("Shader %d needs vertex color clamping disabled\n", programId);
3581 entry->vertex_color_clamp = GL_FALSE;
3583 entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
3586 /* Set the shader to allow uniform loading on it */
3587 GL_EXTCALL(glUseProgramObjectARB(programId));
3588 checkGLcall("glUseProgramObjectARB(programId)");
3590 /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
3591 * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
3592 * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
3593 * vertex shader with fixed function pixel processing is used we make sure that the card
3594 * supports enough samplers to allow the max number of vertex samplers with all possible
3595 * fixed function fragment processing setups. So once the program is linked these samplers
3599 /* Load vertex shader samplers */
3600 shader_glsl_load_vsamplers(gl_info, This->texUnitMap, programId);
3603 /* Load pixel shader samplers */
3604 shader_glsl_load_psamplers(gl_info, This->texUnitMap, programId);
3607 /* If the local constants do not have to be loaded with the environment constants,
3608 * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
3611 if(pshader && !((IWineD3DPixelShaderImpl*)pshader)->baseShader.load_local_constsF) {
3612 hardcode_local_constants((IWineD3DBaseShaderImpl *) pshader, gl_info, programId, 'P');
3614 if(vshader && !((IWineD3DVertexShaderImpl*)vshader)->baseShader.load_local_constsF) {
3615 hardcode_local_constants((IWineD3DBaseShaderImpl *) vshader, gl_info, programId, 'V');
3619 static GLhandleARB create_glsl_blt_shader(const WineD3D_GL_Info *gl_info, enum tex_types tex_type)
3621 GLhandleARB program_id;
3622 GLhandleARB vshader_id, pshader_id;
3623 static const char *blt_vshader[] =
3628 " gl_Position = gl_Vertex;\n"
3629 " gl_FrontColor = vec4(1.0);\n"
3630 " gl_TexCoord[0] = gl_MultiTexCoord0;\n"
3634 static const char *blt_pshaders[tex_type_count] =
3640 "uniform sampler2D sampler;\n"
3643 " gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
3649 "uniform samplerCube sampler;\n"
3652 " gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
3656 "#extension GL_ARB_texture_rectangle : enable\n"
3657 "uniform sampler2DRect sampler;\n"
3660 " gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
3664 if (!blt_pshaders[tex_type])
3666 FIXME("tex_type %#x not supported\n", tex_type);
3670 vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3671 GL_EXTCALL(glShaderSourceARB(vshader_id, 1, blt_vshader, NULL));
3672 GL_EXTCALL(glCompileShaderARB(vshader_id));
3674 pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3675 GL_EXTCALL(glShaderSourceARB(pshader_id, 1, &blt_pshaders[tex_type], NULL));
3676 GL_EXTCALL(glCompileShaderARB(pshader_id));
3678 program_id = GL_EXTCALL(glCreateProgramObjectARB());
3679 GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
3680 GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
3681 GL_EXTCALL(glLinkProgramARB(program_id));
3683 print_glsl_info_log(&GLINFO_LOCATION, program_id);
3685 /* Once linked we can mark the shaders for deletion. They will be deleted once the program
3688 GL_EXTCALL(glDeleteObjectARB(vshader_id));
3689 GL_EXTCALL(glDeleteObjectARB(pshader_id));
3693 static void shader_glsl_select(IWineD3DDevice *iface, BOOL usePS, BOOL useVS) {
3694 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3695 struct shader_glsl_priv *priv = This->shader_priv;
3696 const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3697 GLhandleARB program_id = 0;
3698 GLenum old_vertex_color_clamp, current_vertex_color_clamp;
3700 old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
3702 if (useVS || usePS) set_glsl_shader_program(iface, usePS, useVS);
3703 else priv->glsl_program = NULL;
3705 current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
3707 if (old_vertex_color_clamp != current_vertex_color_clamp) {
3708 if (GL_SUPPORT(ARB_COLOR_BUFFER_FLOAT)) {
3709 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
3710 checkGLcall("glClampColorARB");
3712 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
3716 program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
3717 if (program_id) TRACE("Using GLSL program %u\n", program_id);
3718 GL_EXTCALL(glUseProgramObjectARB(program_id));
3719 checkGLcall("glUseProgramObjectARB");
3722 static void shader_glsl_select_depth_blt(IWineD3DDevice *iface, enum tex_types tex_type) {
3723 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3724 const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3725 struct shader_glsl_priv *priv = This->shader_priv;
3726 GLhandleARB *blt_program = &priv->depth_blt_program[tex_type];
3728 if (!*blt_program) {
3730 *blt_program = create_glsl_blt_shader(gl_info, tex_type);
3731 loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "sampler"));
3732 GL_EXTCALL(glUseProgramObjectARB(*blt_program));
3733 GL_EXTCALL(glUniform1iARB(loc, 0));
3735 GL_EXTCALL(glUseProgramObjectARB(*blt_program));
3739 static void shader_glsl_deselect_depth_blt(IWineD3DDevice *iface) {
3740 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3741 const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3742 struct shader_glsl_priv *priv = This->shader_priv;
3743 GLhandleARB program_id;
3745 program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
3746 if (program_id) TRACE("Using GLSL program %u\n", program_id);
3748 GL_EXTCALL(glUseProgramObjectARB(program_id));
3749 checkGLcall("glUseProgramObjectARB");
3752 static void shader_glsl_destroy(IWineD3DBaseShader *iface) {
3753 const struct list *linked_programs;
3754 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) iface;
3755 IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
3756 struct shader_glsl_priv *priv = device->shader_priv;
3757 const WineD3D_GL_Info *gl_info = &device->adapter->gl_info;
3758 IWineD3DPixelShaderImpl *ps = NULL;
3759 IWineD3DVertexShaderImpl *vs = NULL;
3761 /* Note: Do not use QueryInterface here to find out which shader type this is because this code
3762 * can be called from IWineD3DBaseShader::Release
3764 char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version);
3767 ps = (IWineD3DPixelShaderImpl *) This;
3768 if(ps->num_gl_shaders == 0) return;
3769 if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->pshader == iface)
3770 shader_glsl_select(This->baseShader.device, FALSE, FALSE);
3772 vs = (IWineD3DVertexShaderImpl *) This;
3773 if(vs->num_gl_shaders == 0) return;
3774 if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->vshader == iface)
3775 shader_glsl_select(This->baseShader.device, FALSE, FALSE);
3778 linked_programs = &This->baseShader.linked_programs;
3780 TRACE("Deleting linked programs\n");
3781 if (linked_programs->next) {
3782 struct glsl_shader_prog_link *entry, *entry2;
3785 LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
3786 delete_glsl_program_entry(priv, gl_info, entry);
3789 LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
3790 delete_glsl_program_entry(priv, gl_info, entry);
3799 for(i = 0; i < ps->num_gl_shaders; i++) {
3800 TRACE("deleting pshader %u\n", ps->gl_shaders[i].prgId);
3801 GL_EXTCALL(glDeleteObjectARB(ps->gl_shaders[i].prgId));
3802 checkGLcall("glDeleteObjectARB");
3805 HeapFree(GetProcessHeap(), 0, ps->gl_shaders);
3806 ps->gl_shaders = NULL;
3807 ps->num_gl_shaders = 0;
3808 ps->shader_array_size = 0;
3813 for(i = 0; i < vs->num_gl_shaders; i++) {
3814 TRACE("deleting vshader %u\n", vs->gl_shaders[i].prgId);
3815 GL_EXTCALL(glDeleteObjectARB(vs->gl_shaders[i].prgId));
3816 checkGLcall("glDeleteObjectARB");
3819 HeapFree(GetProcessHeap(), 0, vs->gl_shaders);
3820 vs->gl_shaders = NULL;
3821 vs->num_gl_shaders = 0;
3822 vs->shader_array_size = 0;
3826 static unsigned int glsl_program_key_hash(const void *key)
3828 const glsl_program_key_t *k = key;
3830 unsigned int hash = ((DWORD_PTR) k->vshader) | ((DWORD_PTR) k->pshader) << 16;
3831 hash += ~(hash << 15);
3832 hash ^= (hash >> 10);
3833 hash += (hash << 3);
3834 hash ^= (hash >> 6);
3835 hash += ~(hash << 11);
3836 hash ^= (hash >> 16);
3841 static BOOL glsl_program_key_compare(const void *keya, const void *keyb)
3843 const glsl_program_key_t *ka = keya;
3844 const glsl_program_key_t *kb = keyb;
3846 return ka->vshader == kb->vshader && ka->pshader == kb->pshader &&
3847 (memcmp(&ka->ps_args, &kb->ps_args, sizeof(kb->ps_args)) == 0) &&
3848 (memcmp(&ka->vs_args, &kb->vs_args, sizeof(kb->vs_args)) == 0);
3851 static BOOL constant_heap_init(struct constant_heap *heap, unsigned int constant_count)
3853 SIZE_T size = (constant_count + 1) * sizeof(*heap->entries) + constant_count * sizeof(*heap->positions);
3854 void *mem = HeapAlloc(GetProcessHeap(), 0, size);
3858 ERR("Failed to allocate memory\n");
3862 heap->entries = mem;
3863 heap->entries[1].version = 0;
3864 heap->positions = (unsigned int *)(heap->entries + constant_count + 1);
3870 static void constant_heap_free(struct constant_heap *heap)
3872 HeapFree(GetProcessHeap(), 0, heap->entries);
3875 static HRESULT shader_glsl_alloc(IWineD3DDevice *iface) {
3876 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3877 const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3878 struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
3879 SIZE_T stack_size = wined3d_log2i(max(GL_LIMITS(vshader_constantsF), GL_LIMITS(pshader_constantsF))) + 1;
3881 priv->stack = HeapAlloc(GetProcessHeap(), 0, stack_size * sizeof(*priv->stack));
3884 ERR("Failed to allocate memory.\n");
3885 HeapFree(GetProcessHeap(), 0, priv);
3886 return E_OUTOFMEMORY;
3889 if (!constant_heap_init(&priv->vconst_heap, GL_LIMITS(vshader_constantsF)))
3891 ERR("Failed to initialize vertex shader constant heap\n");
3892 HeapFree(GetProcessHeap(), 0, priv->stack);
3893 HeapFree(GetProcessHeap(), 0, priv);
3894 return E_OUTOFMEMORY;
3897 if (!constant_heap_init(&priv->pconst_heap, GL_LIMITS(pshader_constantsF)))
3899 ERR("Failed to initialize pixel shader constant heap\n");
3900 constant_heap_free(&priv->vconst_heap);
3901 HeapFree(GetProcessHeap(), 0, priv->stack);
3902 HeapFree(GetProcessHeap(), 0, priv);
3903 return E_OUTOFMEMORY;
3906 priv->glsl_program_lookup = hash_table_create(glsl_program_key_hash, glsl_program_key_compare);
3907 priv->next_constant_version = 1;
3909 This->shader_priv = priv;
3913 static void shader_glsl_free(IWineD3DDevice *iface) {
3914 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3915 const WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3916 struct shader_glsl_priv *priv = This->shader_priv;
3919 for (i = 0; i < tex_type_count; ++i)
3921 if (priv->depth_blt_program[i])
3923 GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program[i]));
3927 hash_table_destroy(priv->glsl_program_lookup, NULL, NULL);
3928 constant_heap_free(&priv->pconst_heap);
3929 constant_heap_free(&priv->vconst_heap);
3931 HeapFree(GetProcessHeap(), 0, This->shader_priv);
3932 This->shader_priv = NULL;
3935 static BOOL shader_glsl_dirty_const(IWineD3DDevice *iface) {
3936 /* TODO: GL_EXT_bindable_uniform can be used to share constants across shaders */
3940 static GLuint shader_glsl_generate_pshader(IWineD3DPixelShader *iface, SHADER_BUFFER *buffer, const struct ps_compile_args *args) {
3941 IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
3942 const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3943 CONST DWORD *function = This->baseShader.function;
3944 const char *fragcolor;
3945 const WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3947 /* Create the hw GLSL shader object and assign it as the shader->prgId */
3948 GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3950 shader_addline(buffer, "#version 120\n");
3952 if (GL_SUPPORT(ARB_DRAW_BUFFERS)) {
3953 shader_addline(buffer, "#extension GL_ARB_draw_buffers : enable\n");
3955 if (GL_SUPPORT(ARB_TEXTURE_RECTANGLE)) {
3956 /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
3957 * drivers write a warning if we don't do so
3959 shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
3962 /* Base Declarations */
3963 shader_generate_glsl_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION, args);
3965 /* Pack 3.0 inputs */
3966 if (reg_maps->shader_version >= WINED3DPS_VERSION(3,0) && args->vp_mode != vertexshader) {
3967 pshader_glsl_input_pack(buffer, This->semantics_in, iface, args->vp_mode);
3970 /* Base Shader Body */
3971 shader_generate_main( (IWineD3DBaseShader*) This, buffer, reg_maps, function);
3973 /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
3974 if (reg_maps->shader_version < WINED3DPS_VERSION(2,0))
3976 /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
3977 if(GL_SUPPORT(ARB_DRAW_BUFFERS))
3978 shader_addline(buffer, "gl_FragData[0] = R0;\n");
3980 shader_addline(buffer, "gl_FragColor = R0;\n");
3983 if(GL_SUPPORT(ARB_DRAW_BUFFERS)) {
3984 fragcolor = "gl_FragData[0]";
3986 fragcolor = "gl_FragColor";
3988 if(args->srgb_correction) {
3989 shader_addline(buffer, "tmp0.xyz = pow(%s.xyz, vec3(%f, %f, %f)) * vec3(%f, %f, %f) - vec3(%f, %f, %f);\n",
3990 fragcolor, srgb_pow, srgb_pow, srgb_pow, srgb_mul_high, srgb_mul_high, srgb_mul_high,
3991 srgb_sub_high, srgb_sub_high, srgb_sub_high);
3992 shader_addline(buffer, "tmp1.xyz = %s.xyz * srgb_mul_low.xyz;\n", fragcolor);
3993 shader_addline(buffer, "%s.x = %s.x < srgb_comparison.x ? tmp1.x : tmp0.x;\n", fragcolor, fragcolor);
3994 shader_addline(buffer, "%s.y = %s.y < srgb_comparison.y ? tmp1.y : tmp0.y;\n", fragcolor, fragcolor);
3995 shader_addline(buffer, "%s.z = %s.z < srgb_comparison.z ? tmp1.z : tmp0.z;\n", fragcolor, fragcolor);
3996 shader_addline(buffer, "%s = clamp(%s, 0.0, 1.0);\n", fragcolor, fragcolor);
3998 /* Pixel shader < 3.0 do not replace the fog stage.
3999 * This implements linear fog computation and blending.
4000 * TODO: non linear fog
4001 * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
4002 * -1/(e-s) and e/(e-s) respectively.
4004 if(reg_maps->shader_version < WINED3DPS_VERSION(3,0)) {
4006 case FOG_OFF: break;
4008 shader_addline(buffer, "float fogstart = -1.0 / (gl_Fog.end - gl_Fog.start);\n");
4009 shader_addline(buffer, "float fogend = gl_Fog.end * -fogstart;\n");
4010 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * fogstart + fogend, 0.0, 1.0);\n");
4011 shader_addline(buffer, "%s.xyz = mix(gl_Fog.color.xyz, %s.xyz, Fog);\n", fragcolor, fragcolor);
4014 /* Fog = e^(-gl_Fog.density * gl_FogFragCoord) */
4015 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_FogFragCoord);\n");
4016 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4017 shader_addline(buffer, "%s.xyz = mix(gl_Fog.color.xyz, %s.xyz, Fog);\n", fragcolor, fragcolor);
4020 /* Fog = e^(-(gl_Fog.density * gl_FogFragCoord)^2) */
4021 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_Fog.density * gl_FogFragCoord * gl_FogFragCoord);\n");
4022 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4023 shader_addline(buffer, "%s.xyz = mix(gl_Fog.color.xyz, %s.xyz, Fog);\n", fragcolor, fragcolor);
4028 shader_addline(buffer, "}\n");
4030 TRACE("Compiling shader object %u\n", shader_obj);
4031 GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
4032 GL_EXTCALL(glCompileShaderARB(shader_obj));
4033 print_glsl_info_log(&GLINFO_LOCATION, shader_obj);
4035 /* Store the shader object */
4039 static GLuint shader_glsl_generate_vshader(IWineD3DVertexShader *iface, SHADER_BUFFER *buffer, const struct vs_compile_args *args) {
4040 IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
4041 const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
4042 CONST DWORD *function = This->baseShader.function;
4043 const WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
4045 /* Create the hw GLSL shader program and assign it as the shader->prgId */
4046 GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4048 shader_addline(buffer, "#version 120\n");
4050 /* Base Declarations */
4051 shader_generate_glsl_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION, NULL);
4053 /* Base Shader Body */
4054 shader_generate_main( (IWineD3DBaseShader*) This, buffer, reg_maps, function);
4056 /* Unpack 3.0 outputs */
4057 if (reg_maps->shader_version >= WINED3DVS_VERSION(3,0)) shader_addline(buffer, "order_ps_input(OUT);\n");
4058 else shader_addline(buffer, "order_ps_input();\n");
4060 /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
4061 * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
4062 * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
4063 * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
4065 if(args->fog_src == VS_FOG_Z) {
4066 shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
4067 } else if (!reg_maps->fog) {
4068 shader_addline(buffer, "gl_FogFragCoord = 0.0;\n");
4071 /* Write the final position.
4073 * OpenGL coordinates specify the center of the pixel while d3d coords specify
4074 * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
4075 * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
4076 * contains 1.0 to allow a mad.
4078 shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
4079 shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
4081 /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
4083 * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
4084 * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
4085 * which is the same as z = z * 2 - w.
4087 shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
4089 shader_addline(buffer, "}\n");
4091 TRACE("Compiling shader object %u\n", shader_obj);
4092 GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
4093 GL_EXTCALL(glCompileShaderARB(shader_obj));
4094 print_glsl_info_log(&GLINFO_LOCATION, shader_obj);
4099 static void shader_glsl_get_caps(WINED3DDEVTYPE devtype, const WineD3D_GL_Info *gl_info, struct shader_caps *pCaps)
4101 /* Nvidia Geforce6/7 or Ati R4xx/R5xx cards with GLSL support, support VS 3.0 but older Nvidia/Ati
4102 * models with GLSL support only support 2.0. In case of nvidia we can detect VS 2.0 support using
4103 * vs_nv_version which is based on NV_vertex_program.
4104 * For Ati cards there's no way using glsl (it abstracts the lowlevel info away) and also not
4105 * using ARB_vertex_program. It is safe to assume that when a card supports pixel shader 2.0 it
4106 * supports vertex shader 2.0 too and the way around. We can detect ps2.0 using the maximum number
4107 * of native instructions, so use that here. For more info see the pixel shader versioning code below.
4109 if((GLINFO_LOCATION.vs_nv_version == VS_VERSION_20) || (GLINFO_LOCATION.ps_arb_max_instructions <= 512))
4110 pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4112 pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4113 TRACE_(d3d_caps)("Hardware vertex shader version %d.%d enabled (GLSL)\n", (pCaps->VertexShaderVersion >> 8) & 0xff, pCaps->VertexShaderVersion & 0xff);
4114 pCaps->MaxVertexShaderConst = GL_LIMITS(vshader_constantsF);
4116 /* Older DX9-class videocards (GeforceFX / Radeon >9500/X*00) only support pixel shader 2.0/2.0a/2.0b.
4117 * In OpenGL the extensions related to GLSL abstract lowlevel GL info away which is needed
4118 * to distinguish between 2.0 and 3.0 (and 2.0a/2.0b). In case of Nvidia we use their fragment
4119 * program extensions. On other hardware including ATI GL_ARB_fragment_program offers the info
4120 * in max native instructions. Intel and others also offer the info in this extension but they
4121 * don't support GLSL (at least on Windows).
4123 * PS2.0 requires at least 96 instructions, 2.0a/2.0b go up to 512. Assume that if the number
4124 * of instructions is 512 or less we have to do with ps2.0 hardware.
4125 * 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.
4127 if((GLINFO_LOCATION.ps_nv_version == PS_VERSION_20) || (GLINFO_LOCATION.ps_arb_max_instructions <= 512))
4128 pCaps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
4130 pCaps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
4132 /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
4133 * Direct3D minimum requirement.
4135 * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
4136 * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
4138 * The problem is that the refrast clamps temporary results in the shader to
4139 * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
4140 * then applications may miss the clamping behavior. On the other hand, if it is smaller,
4141 * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
4142 * offer a way to query this.
4144 pCaps->PixelShader1xMaxValue = 8.0;
4145 TRACE_(d3d_caps)("Hardware pixel shader version %d.%d enabled (GLSL)\n", (pCaps->PixelShaderVersion >> 8) & 0xff, pCaps->PixelShaderVersion & 0xff);
4148 static BOOL shader_glsl_color_fixup_supported(struct color_fixup_desc fixup)
4150 if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4152 TRACE("Checking support for fixup:\n");
4153 dump_color_fixup_desc(fixup);
4156 /* We support everything except YUV conversions. */
4157 if (!is_yuv_fixup(fixup))
4163 TRACE("[FAILED]\n");
4167 static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4169 /* WINED3DSIH_ABS */ shader_glsl_map2gl,
4170 /* WINED3DSIH_ADD */ shader_glsl_arith,
4171 /* WINED3DSIH_BEM */ pshader_glsl_bem,
4172 /* WINED3DSIH_BREAK */ shader_glsl_break,
4173 /* WINED3DSIH_BREAKC */ shader_glsl_breakc,
4174 /* WINED3DSIH_BREAKP */ NULL,
4175 /* WINED3DSIH_CALL */ shader_glsl_call,
4176 /* WINED3DSIH_CALLNZ */ shader_glsl_callnz,
4177 /* WINED3DSIH_CMP */ shader_glsl_cmp,
4178 /* WINED3DSIH_CND */ shader_glsl_cnd,
4179 /* WINED3DSIH_CRS */ shader_glsl_cross,
4180 /* WINED3DSIH_DCL */ NULL,
4181 /* WINED3DSIH_DEF */ NULL,
4182 /* WINED3DSIH_DEFB */ NULL,
4183 /* WINED3DSIH_DEFI */ NULL,
4184 /* WINED3DSIH_DP2ADD */ pshader_glsl_dp2add,
4185 /* WINED3DSIH_DP3 */ shader_glsl_dot,
4186 /* WINED3DSIH_DP4 */ shader_glsl_dot,
4187 /* WINED3DSIH_DST */ shader_glsl_dst,
4188 /* WINED3DSIH_DSX */ shader_glsl_map2gl,
4189 /* WINED3DSIH_DSY */ shader_glsl_map2gl,
4190 /* WINED3DSIH_ELSE */ shader_glsl_else,
4191 /* WINED3DSIH_ENDIF */ shader_glsl_end,
4192 /* WINED3DSIH_ENDLOOP */ shader_glsl_end,
4193 /* WINED3DSIH_ENDREP */ shader_glsl_end,
4194 /* WINED3DSIH_EXP */ shader_glsl_map2gl,
4195 /* WINED3DSIH_EXPP */ shader_glsl_expp,
4196 /* WINED3DSIH_FRC */ shader_glsl_map2gl,
4197 /* WINED3DSIH_IF */ shader_glsl_if,
4198 /* WINED3DSIH_IFC */ shader_glsl_ifc,
4199 /* WINED3DSIH_LABEL */ shader_glsl_label,
4200 /* WINED3DSIH_LIT */ shader_glsl_lit,
4201 /* WINED3DSIH_LOG */ shader_glsl_log,
4202 /* WINED3DSIH_LOGP */ shader_glsl_log,
4203 /* WINED3DSIH_LOOP */ shader_glsl_loop,
4204 /* WINED3DSIH_LRP */ shader_glsl_lrp,
4205 /* WINED3DSIH_M3x2 */ shader_glsl_mnxn,
4206 /* WINED3DSIH_M3x3 */ shader_glsl_mnxn,
4207 /* WINED3DSIH_M3x4 */ shader_glsl_mnxn,
4208 /* WINED3DSIH_M4x3 */ shader_glsl_mnxn,
4209 /* WINED3DSIH_M4x4 */ shader_glsl_mnxn,
4210 /* WINED3DSIH_MAD */ shader_glsl_mad,
4211 /* WINED3DSIH_MAX */ shader_glsl_map2gl,
4212 /* WINED3DSIH_MIN */ shader_glsl_map2gl,
4213 /* WINED3DSIH_MOV */ shader_glsl_mov,
4214 /* WINED3DSIH_MOVA */ shader_glsl_mov,
4215 /* WINED3DSIH_MUL */ shader_glsl_arith,
4216 /* WINED3DSIH_NOP */ NULL,
4217 /* WINED3DSIH_NRM */ shader_glsl_map2gl,
4218 /* WINED3DSIH_PHASE */ NULL,
4219 /* WINED3DSIH_POW */ shader_glsl_pow,
4220 /* WINED3DSIH_RCP */ shader_glsl_rcp,
4221 /* WINED3DSIH_REP */ shader_glsl_rep,
4222 /* WINED3DSIH_RET */ NULL,
4223 /* WINED3DSIH_RSQ */ shader_glsl_rsq,
4224 /* WINED3DSIH_SETP */ NULL,
4225 /* WINED3DSIH_SGE */ shader_glsl_compare,
4226 /* WINED3DSIH_SGN */ shader_glsl_map2gl,
4227 /* WINED3DSIH_SINCOS */ shader_glsl_sincos,
4228 /* WINED3DSIH_SLT */ shader_glsl_compare,
4229 /* WINED3DSIH_SUB */ shader_glsl_arith,
4230 /* WINED3DSIH_TEX */ pshader_glsl_tex,
4231 /* WINED3DSIH_TEXBEM */ pshader_glsl_texbem,
4232 /* WINED3DSIH_TEXBEML */ pshader_glsl_texbem,
4233 /* WINED3DSIH_TEXCOORD */ pshader_glsl_texcoord,
4234 /* WINED3DSIH_TEXDEPTH */ pshader_glsl_texdepth,
4235 /* WINED3DSIH_TEXDP3 */ pshader_glsl_texdp3,
4236 /* WINED3DSIH_TEXDP3TEX */ pshader_glsl_texdp3tex,
4237 /* WINED3DSIH_TEXKILL */ pshader_glsl_texkill,
4238 /* WINED3DSIH_TEXLDD */ NULL,
4239 /* WINED3DSIH_TEXLDL */ shader_glsl_texldl,
4240 /* WINED3DSIH_TEXM3x2DEPTH */ pshader_glsl_texm3x2depth,
4241 /* WINED3DSIH_TEXM3x2PAD */ pshader_glsl_texm3x2pad,
4242 /* WINED3DSIH_TEXM3x2TEX */ pshader_glsl_texm3x2tex,
4243 /* WINED3DSIH_TEXM3x3 */ pshader_glsl_texm3x3,
4244 /* WINED3DSIH_TEXM3x3DIFF */ NULL,
4245 /* WINED3DSIH_TEXM3x3PAD */ pshader_glsl_texm3x3pad,
4246 /* WINED3DSIH_TEXM3x3SPEC */ pshader_glsl_texm3x3spec,
4247 /* WINED3DSIH_TEXM3x3TEX */ pshader_glsl_texm3x3tex,
4248 /* WINED3DSIH_TEXM3x3VSPEC */ pshader_glsl_texm3x3vspec,
4249 /* WINED3DSIH_TEXREG2AR */ pshader_glsl_texreg2ar,
4250 /* WINED3DSIH_TEXREG2GB */ pshader_glsl_texreg2gb,
4251 /* WINED3DSIH_TEXREG2RGB */ pshader_glsl_texreg2rgb,
4254 const shader_backend_t glsl_shader_backend = {
4255 shader_glsl_instruction_handler_table,
4257 shader_glsl_select_depth_blt,
4258 shader_glsl_deselect_depth_blt,
4259 shader_glsl_update_float_vertex_constants,
4260 shader_glsl_update_float_pixel_constants,
4261 shader_glsl_load_constants,
4262 shader_glsl_destroy,
4265 shader_glsl_dirty_const,
4266 shader_glsl_generate_pshader,
4267 shader_glsl_generate_vshader,
4268 shader_glsl_get_caps,
4269 shader_glsl_color_fixup_supported,