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
7 * Copyright 2009 Henri Verbeet for CodeWeavers
9 * This library is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * This library is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with this library; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
25 * D3D shader asm has swizzles on source parameters, and write masks for
26 * destination parameters. GLSL uses swizzles for both. The result of this is
27 * that for example "mov dst.xw, src.zyxw" becomes "dst.xw = src.zw" in GLSL.
28 * Ie, to generate a proper GLSL source swizzle, we need to take the D3D write
29 * mask for the destination parameter into account.
35 #include "wined3d_private.h"
37 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
38 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
39 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
40 WINE_DECLARE_DEBUG_CHANNEL(d3d);
42 #define WINED3D_GLSL_SAMPLE_PROJECTED 0x1
43 #define WINED3D_GLSL_SAMPLE_RECT 0x2
44 #define WINED3D_GLSL_SAMPLE_LOD 0x4
45 #define WINED3D_GLSL_SAMPLE_GRAD 0x8
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 wined3d_shader_buffer shader_buffer;
85 struct wine_rb_tree program_lookup;
86 struct glsl_shader_prog_link *glsl_program;
87 struct constant_heap vconst_heap;
88 struct constant_heap pconst_heap;
90 GLhandleARB depth_blt_program_full[tex_type_count];
91 GLhandleARB depth_blt_program_masked[tex_type_count];
92 UINT next_constant_version;
95 /* Struct to maintain data about a linked GLSL program */
96 struct glsl_shader_prog_link {
97 struct wine_rb_entry program_lookup_entry;
98 struct list vshader_entry;
99 struct list pshader_entry;
100 GLhandleARB programId;
101 GLint *vuniformF_locations;
102 GLint *puniformF_locations;
103 GLint vuniformI_locations[MAX_CONST_I];
104 GLint puniformI_locations[MAX_CONST_I];
105 GLint posFixup_location;
106 GLint np2Fixup_location;
107 GLint bumpenvmat_location[MAX_TEXTURES];
108 GLint luminancescale_location[MAX_TEXTURES];
109 GLint luminanceoffset_location[MAX_TEXTURES];
110 GLint ycorrection_location;
111 GLenum vertex_color_clamp;
112 IWineD3DVertexShader *vshader;
113 IWineD3DPixelShader *pshader;
114 struct vs_compile_args vs_args;
115 struct ps_compile_args ps_args;
116 UINT constant_version;
117 const struct ps_np2fixup_info *np2Fixup_info;
121 IWineD3DVertexShader *vshader;
122 IWineD3DPixelShader *pshader;
123 struct ps_compile_args ps_args;
124 struct vs_compile_args vs_args;
125 } glsl_program_key_t;
127 struct shader_glsl_ctx_priv {
128 const struct vs_compile_args *cur_vs_args;
129 const struct ps_compile_args *cur_ps_args;
130 struct ps_np2fixup_info *cur_np2fixup_info;
133 struct glsl_ps_compiled_shader
135 struct ps_compile_args args;
136 struct ps_np2fixup_info np2fixup;
140 struct glsl_pshader_private
142 struct glsl_ps_compiled_shader *gl_shaders;
143 UINT num_gl_shaders, shader_array_size;
146 struct glsl_vs_compiled_shader
148 struct vs_compile_args args;
152 struct glsl_vshader_private
154 struct glsl_vs_compiled_shader *gl_shaders;
155 UINT num_gl_shaders, shader_array_size;
158 static const char *debug_gl_shader_type(GLenum type)
162 #define WINED3D_TO_STR(u) case u: return #u
163 WINED3D_TO_STR(GL_VERTEX_SHADER_ARB);
164 WINED3D_TO_STR(GL_GEOMETRY_SHADER_ARB);
165 WINED3D_TO_STR(GL_FRAGMENT_SHADER_ARB);
166 #undef WINED3D_TO_STR
168 return wine_dbg_sprintf("UNKNOWN(%#x)", type);
172 /* Extract a line from the info log.
173 * Note that this modifies the source string. */
174 static char *get_info_log_line(char **ptr)
179 if (!(q = strstr(p, "\n")))
181 if (!*p) return NULL;
191 /** Prints the GLSL info log which will contain error messages if they exist */
192 /* GL locking is done by the caller */
193 static void print_glsl_info_log(const struct wined3d_gl_info *gl_info, GLhandleARB obj)
195 int infologLength = 0;
200 static const char * const spam[] =
202 "Vertex shader was successfully compiled to run on hardware.\n", /* fglrx */
203 "Fragment shader was successfully compiled to run on hardware.\n", /* fglrx, with \n */
204 "Fragment shader was successfully compiled to run on hardware.", /* fglrx, no \n */
205 "Fragment shader(s) linked, vertex shader(s) linked. \n ", /* fglrx, with \n */
206 "Fragment shader(s) linked, vertex shader(s) linked. \n", /* fglrx, with \n */
207 "Fragment shader(s) linked, vertex shader(s) linked.", /* fglrx, no \n */
208 "Vertex shader(s) linked, no fragment shader(s) defined. \n ", /* fglrx, with \n */
209 "Vertex shader(s) linked, no fragment shader(s) defined. \n", /* fglrx, with \n */
210 "Vertex shader(s) linked, no fragment shader(s) defined.", /* fglrx, no \n */
211 "Fragment shader(s) linked, no vertex shader(s) defined. \n ", /* fglrx, with \n */
212 "Fragment shader(s) linked, no vertex shader(s) defined. \n", /* fglrx, with \n */
213 "Fragment shader(s) linked, no vertex shader(s) defined.", /* fglrx, no \n */
216 if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
218 GL_EXTCALL(glGetObjectParameterivARB(obj,
219 GL_OBJECT_INFO_LOG_LENGTH_ARB,
222 /* A size of 1 is just a null-terminated string, so the log should be bigger than
223 * that if there are errors. */
224 if (infologLength > 1)
228 /* Fglrx doesn't terminate the string properly, but it tells us the proper length.
229 * So use HEAP_ZERO_MEMORY to avoid uninitialized bytes
231 infoLog = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, infologLength);
232 GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
235 for (i = 0; i < sizeof(spam) / sizeof(*spam); ++i)
237 if (!strcmp(infoLog, spam[i]))
247 TRACE("Spam received from GLSL shader #%u:\n", obj);
248 while ((line = get_info_log_line(&ptr))) TRACE(" %s\n", line);
252 FIXME("Error received from GLSL shader #%u:\n", obj);
253 while ((line = get_info_log_line(&ptr))) FIXME(" %s\n", line);
255 HeapFree(GetProcessHeap(), 0, infoLog);
259 /* GL locking is done by the caller. */
260 static void shader_glsl_compile(const struct wined3d_gl_info *gl_info, GLhandleARB shader, const char *src)
262 TRACE("Compiling shader object %u.\n", shader);
263 GL_EXTCALL(glShaderSourceARB(shader, 1, &src, NULL));
264 checkGLcall("glShaderSourceARB");
265 GL_EXTCALL(glCompileShaderARB(shader));
266 checkGLcall("glCompileShaderARB");
267 print_glsl_info_log(gl_info, shader);
270 /* GL locking is done by the caller. */
271 static void shader_glsl_dump_program_source(const struct wined3d_gl_info *gl_info, GLhandleARB program)
273 GLint i, object_count, source_size = -1;
274 GLhandleARB *objects;
277 GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_ATTACHED_OBJECTS_ARB, &object_count));
278 objects = HeapAlloc(GetProcessHeap(), 0, object_count * sizeof(*objects));
281 ERR("Failed to allocate object array memory.\n");
285 GL_EXTCALL(glGetAttachedObjectsARB(program, object_count, NULL, objects));
286 for (i = 0; i < object_count; ++i)
291 GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_SHADER_SOURCE_LENGTH_ARB, &tmp));
293 if (source_size < tmp)
295 HeapFree(GetProcessHeap(), 0, source);
297 source = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, tmp);
300 ERR("Failed to allocate %d bytes for shader source.\n", tmp);
301 HeapFree(GetProcessHeap(), 0, objects);
307 FIXME("Object %u:\n", objects[i]);
308 GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_SUBTYPE_ARB, &tmp));
309 FIXME(" GL_OBJECT_SUBTYPE_ARB: %s.\n", debug_gl_shader_type(tmp));
310 GL_EXTCALL(glGetObjectParameterivARB(objects[i], GL_OBJECT_COMPILE_STATUS_ARB, &tmp));
311 FIXME(" GL_OBJECT_COMPILE_STATUS_ARB: %d.\n", tmp);
315 GL_EXTCALL(glGetShaderSourceARB(objects[i], source_size, NULL, source));
316 while ((line = get_info_log_line(&ptr))) FIXME(" %s\n", line);
320 HeapFree(GetProcessHeap(), 0, source);
321 HeapFree(GetProcessHeap(), 0, objects);
324 /* GL locking is done by the caller. */
325 static void shader_glsl_validate_link(const struct wined3d_gl_info *gl_info, GLhandleARB program)
329 if (!TRACE_ON(d3d_shader) && !FIXME_ON(d3d_shader)) return;
331 GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_TYPE_ARB, &tmp));
332 if (tmp == GL_PROGRAM_OBJECT_ARB)
334 GL_EXTCALL(glGetObjectParameterivARB(program, GL_OBJECT_LINK_STATUS_ARB, &tmp));
337 FIXME("Program %u link status invalid.\n", program);
338 shader_glsl_dump_program_source(gl_info, program);
342 print_glsl_info_log(gl_info, program);
346 * Loads (pixel shader) samplers
348 /* GL locking is done by the caller */
349 static void shader_glsl_load_psamplers(const struct wined3d_gl_info *gl_info,
350 DWORD *tex_unit_map, GLhandleARB programId)
354 char sampler_name[20];
356 for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
357 snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
358 name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
359 if (name_loc != -1) {
360 DWORD mapped_unit = tex_unit_map[i];
361 if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.fragment_samplers)
363 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
364 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
365 checkGLcall("glUniform1iARB");
367 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
373 /* GL locking is done by the caller */
374 static void shader_glsl_load_vsamplers(const struct wined3d_gl_info *gl_info,
375 DWORD *tex_unit_map, GLhandleARB programId)
378 char sampler_name[20];
381 for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
382 snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
383 name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
384 if (name_loc != -1) {
385 DWORD mapped_unit = tex_unit_map[MAX_FRAGMENT_SAMPLERS + i];
386 if (mapped_unit != WINED3D_UNMAPPED_STAGE && mapped_unit < gl_info->limits.combined_samplers)
388 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
389 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
390 checkGLcall("glUniform1iARB");
392 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
398 /* GL locking is done by the caller */
399 static inline void walk_constant_heap(const struct wined3d_gl_info *gl_info, const float *constants,
400 const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
403 unsigned int heap_idx = 1;
406 if (heap->entries[heap_idx].version <= version) return;
408 idx = heap->entries[heap_idx].idx;
409 if (constant_locations[idx] != -1) GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
410 stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
412 while (stack_idx >= 0)
414 /* Note that we fall through to the next case statement. */
415 switch(stack[stack_idx])
417 case HEAP_NODE_TRAVERSE_LEFT:
419 unsigned int left_idx = heap_idx << 1;
420 if (left_idx < heap->size && heap->entries[left_idx].version > version)
423 idx = heap->entries[heap_idx].idx;
424 if (constant_locations[idx] != -1)
425 GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
427 stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
428 stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
433 case HEAP_NODE_TRAVERSE_RIGHT:
435 unsigned int right_idx = (heap_idx << 1) + 1;
436 if (right_idx < heap->size && heap->entries[right_idx].version > version)
438 heap_idx = right_idx;
439 idx = heap->entries[heap_idx].idx;
440 if (constant_locations[idx] != -1)
441 GL_EXTCALL(glUniform4fvARB(constant_locations[idx], 1, &constants[idx * 4]));
443 stack[stack_idx++] = HEAP_NODE_POP;
444 stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
457 checkGLcall("walk_constant_heap()");
460 /* GL locking is done by the caller */
461 static inline void apply_clamped_constant(const struct wined3d_gl_info *gl_info, GLint location, const GLfloat *data)
463 GLfloat clamped_constant[4];
465 if (location == -1) return;
467 clamped_constant[0] = data[0] < -1.0f ? -1.0f : data[0] > 1.0f ? 1.0f : data[0];
468 clamped_constant[1] = data[1] < -1.0f ? -1.0f : data[1] > 1.0f ? 1.0f : data[1];
469 clamped_constant[2] = data[2] < -1.0f ? -1.0f : data[2] > 1.0f ? 1.0f : data[2];
470 clamped_constant[3] = data[3] < -1.0f ? -1.0f : data[3] > 1.0f ? 1.0f : data[3];
472 GL_EXTCALL(glUniform4fvARB(location, 1, clamped_constant));
475 /* GL locking is done by the caller */
476 static inline void walk_constant_heap_clamped(const struct wined3d_gl_info *gl_info, const float *constants,
477 const GLint *constant_locations, const struct constant_heap *heap, unsigned char *stack, DWORD version)
480 unsigned int heap_idx = 1;
483 if (heap->entries[heap_idx].version <= version) return;
485 idx = heap->entries[heap_idx].idx;
486 apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
487 stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
489 while (stack_idx >= 0)
491 /* Note that we fall through to the next case statement. */
492 switch(stack[stack_idx])
494 case HEAP_NODE_TRAVERSE_LEFT:
496 unsigned int left_idx = heap_idx << 1;
497 if (left_idx < heap->size && heap->entries[left_idx].version > version)
500 idx = heap->entries[heap_idx].idx;
501 apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
503 stack[stack_idx++] = HEAP_NODE_TRAVERSE_RIGHT;
504 stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
509 case HEAP_NODE_TRAVERSE_RIGHT:
511 unsigned int right_idx = (heap_idx << 1) + 1;
512 if (right_idx < heap->size && heap->entries[right_idx].version > version)
514 heap_idx = right_idx;
515 idx = heap->entries[heap_idx].idx;
516 apply_clamped_constant(gl_info, constant_locations[idx], &constants[idx * 4]);
518 stack[stack_idx++] = HEAP_NODE_POP;
519 stack[stack_idx] = HEAP_NODE_TRAVERSE_LEFT;
532 checkGLcall("walk_constant_heap_clamped()");
535 /* Loads floating point constants (aka uniforms) into the currently set GLSL program. */
536 /* GL locking is done by the caller */
537 static void shader_glsl_load_constantsF(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
538 const float *constants, const GLint *constant_locations, const struct constant_heap *heap,
539 unsigned char *stack, UINT version)
541 const local_constant *lconst;
543 /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
544 if (This->baseShader.reg_maps.shader_version.major == 1
545 && shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type))
546 walk_constant_heap_clamped(gl_info, constants, constant_locations, heap, stack, version);
548 walk_constant_heap(gl_info, constants, constant_locations, heap, stack, version);
550 if (!This->baseShader.load_local_constsF)
552 TRACE("No need to load local float constants for this shader\n");
556 /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
557 LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry)
559 GLint location = constant_locations[lconst->idx];
560 /* We found this uniform name in the program - go ahead and send the data */
561 if (location != -1) GL_EXTCALL(glUniform4fvARB(location, 1, (const GLfloat *)lconst->value));
563 checkGLcall("glUniform4fvARB()");
566 /* Loads integer constants (aka uniforms) into the currently set GLSL program. */
567 /* GL locking is done by the caller */
568 static void shader_glsl_load_constantsI(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
569 const GLint locations[MAX_CONST_I], const int *constants, WORD constants_set)
574 for (i = 0; constants_set; constants_set >>= 1, ++i)
576 if (!(constants_set & 1)) continue;
578 TRACE_(d3d_constants)("Loading constants %u: %i, %i, %i, %i\n",
579 i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
581 /* We found this uniform name in the program - go ahead and send the data */
582 GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
583 checkGLcall("glUniform4ivARB");
586 /* Load immediate constants */
587 ptr = list_head(&This->baseShader.constantsI);
589 const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
590 unsigned int idx = lconst->idx;
591 const GLint *values = (const GLint *)lconst->value;
593 TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
594 values[0], values[1], values[2], values[3]);
596 /* We found this uniform name in the program - go ahead and send the data */
597 GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
598 checkGLcall("glUniform4ivARB");
599 ptr = list_next(&This->baseShader.constantsI, ptr);
603 /* Loads boolean constants (aka uniforms) into the currently set GLSL program. */
604 /* GL locking is done by the caller */
605 static void shader_glsl_load_constantsB(IWineD3DBaseShaderImpl *This, const struct wined3d_gl_info *gl_info,
606 GLhandleARB programId, const BOOL *constants, WORD constants_set)
614 switch (This->baseShader.reg_maps.shader_version.type)
616 case WINED3D_SHADER_TYPE_VERTEX:
620 case WINED3D_SHADER_TYPE_GEOMETRY:
624 case WINED3D_SHADER_TYPE_PIXEL:
629 FIXME("Unknown shader type %#x.\n",
630 This->baseShader.reg_maps.shader_version.type);
635 /* TODO: Benchmark and see if it would be beneficial to store the
636 * locations of the constants to avoid looking up each time */
637 for (i = 0; constants_set; constants_set >>= 1, ++i)
639 if (!(constants_set & 1)) continue;
641 TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
643 /* TODO: Benchmark and see if it would be beneficial to store the
644 * locations of the constants to avoid looking up each time */
645 snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
646 tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
649 /* We found this uniform name in the program - go ahead and send the data */
650 GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
651 checkGLcall("glUniform1ivARB");
655 /* Load immediate constants */
656 ptr = list_head(&This->baseShader.constantsB);
658 const struct local_constant *lconst = LIST_ENTRY(ptr, const struct local_constant, entry);
659 unsigned int idx = lconst->idx;
660 const GLint *values = (const GLint *)lconst->value;
662 TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
664 snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
665 tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
667 /* We found this uniform name in the program - go ahead and send the data */
668 GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
669 checkGLcall("glUniform1ivARB");
671 ptr = list_next(&This->baseShader.constantsB, ptr);
675 static void reset_program_constant_version(struct wine_rb_entry *entry, void *context)
677 WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry)->constant_version = 0;
681 * Loads the texture dimensions for NP2 fixup into the currently set GLSL program.
683 /* GL locking is done by the caller (state handler) */
684 static void shader_glsl_load_np2fixup_constants(
685 IWineD3DDevice* device,
687 char useVertexShader) {
689 const IWineD3DDeviceImpl* deviceImpl = (const IWineD3DDeviceImpl*) device;
690 const struct glsl_shader_prog_link* prog = ((struct shader_glsl_priv *)(deviceImpl->shader_priv))->glsl_program;
693 /* No GLSL program set - nothing to do. */
697 if (!usePixelShader) {
698 /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
702 if (prog->ps_args.np2_fixup && -1 != prog->np2Fixup_location) {
703 const struct wined3d_gl_info *gl_info = &deviceImpl->adapter->gl_info;
704 const IWineD3DStateBlockImpl* stateBlock = (const IWineD3DStateBlockImpl*) deviceImpl->stateBlock;
706 UINT fixup = prog->ps_args.np2_fixup;
707 GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
709 for (i = 0; fixup; fixup >>= 1, ++i) {
710 const unsigned char idx = prog->np2Fixup_info->idx[i];
711 const IWineD3DBaseTextureImpl *tex = stateBlock->state.textures[i];
712 GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
716 ERR("Nonexistent texture is flagged for NP2 texcoord fixup.\n");
721 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
723 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
727 GL_EXTCALL(glUniform4fvARB(prog->np2Fixup_location, prog->np2Fixup_info->num_consts, np2fixup_constants));
732 * Loads the app-supplied constants into the currently set GLSL program.
734 /* GL locking is done by the caller (state handler) */
735 static void shader_glsl_load_constants(const struct wined3d_context *context,
736 char usePixelShader, char useVertexShader)
738 const struct wined3d_gl_info *gl_info = context->gl_info;
739 IWineD3DDeviceImpl *device = context->swapchain->device;
740 IWineD3DStateBlockImpl* stateBlock = device->stateBlock;
741 struct shader_glsl_priv *priv = device->shader_priv;
743 GLhandleARB programId;
744 struct glsl_shader_prog_link *prog = priv->glsl_program;
745 UINT constant_version;
749 /* No GLSL program set - nothing to do. */
752 programId = prog->programId;
753 constant_version = prog->constant_version;
757 IWineD3DBaseShaderImpl *vshader = (IWineD3DBaseShaderImpl *)stateBlock->state.vertex_shader;
759 /* Load DirectX 9 float constants/uniforms for vertex shader */
760 shader_glsl_load_constantsF(vshader, gl_info, stateBlock->state.vs_consts_f,
761 prog->vuniformF_locations, &priv->vconst_heap, priv->stack, constant_version);
763 /* Load DirectX 9 integer constants/uniforms for vertex shader */
764 shader_glsl_load_constantsI(vshader, gl_info, prog->vuniformI_locations, stateBlock->state.vs_consts_i,
765 stateBlock->changed.vertexShaderConstantsI & vshader->baseShader.reg_maps.integer_constants);
767 /* Load DirectX 9 boolean constants/uniforms for vertex shader */
768 shader_glsl_load_constantsB(vshader, gl_info, programId, stateBlock->state.vs_consts_b,
769 stateBlock->changed.vertexShaderConstantsB & vshader->baseShader.reg_maps.boolean_constants);
771 /* Upload the position fixup params */
772 GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, &device->posFixup[0]));
773 checkGLcall("glUniform4fvARB");
778 IWineD3DBaseShaderImpl *pshader = (IWineD3DBaseShaderImpl *)stateBlock->state.pixel_shader;
780 /* Load DirectX 9 float constants/uniforms for pixel shader */
781 shader_glsl_load_constantsF(pshader, gl_info, stateBlock->state.ps_consts_f,
782 prog->puniformF_locations, &priv->pconst_heap, priv->stack, constant_version);
784 /* Load DirectX 9 integer constants/uniforms for pixel shader */
785 shader_glsl_load_constantsI(pshader, gl_info, prog->puniformI_locations, stateBlock->state.ps_consts_i,
786 stateBlock->changed.pixelShaderConstantsI & pshader->baseShader.reg_maps.integer_constants);
788 /* Load DirectX 9 boolean constants/uniforms for pixel shader */
789 shader_glsl_load_constantsB(pshader, gl_info, programId, stateBlock->state.ps_consts_b,
790 stateBlock->changed.pixelShaderConstantsB & pshader->baseShader.reg_maps.boolean_constants);
792 /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
793 * It can't be 0 for a valid texbem instruction.
795 for(i = 0; i < MAX_TEXTURES; i++) {
798 if(prog->bumpenvmat_location[i] == -1) continue;
800 data = (const float *)&stateBlock->state.texture_states[i][WINED3DTSS_BUMPENVMAT00];
801 GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
802 checkGLcall("glUniformMatrix2fvARB");
804 /* texbeml needs the luminance scale and offset too. If texbeml
805 * is used, needsbumpmat is set too, so we can check that in the
806 * needsbumpmat check. */
807 if (prog->luminancescale_location[i] != -1)
809 const GLfloat *scale = (const GLfloat *)&stateBlock->state.texture_states[i][WINED3DTSS_BUMPENVLSCALE];
810 const GLfloat *offset = (const GLfloat *)&stateBlock->state.texture_states[i][WINED3DTSS_BUMPENVLOFFSET];
812 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
813 checkGLcall("glUniform1fvARB");
814 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
815 checkGLcall("glUniform1fvARB");
819 if(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
820 float correction_params[4];
822 if (context->render_offscreen)
824 correction_params[0] = 0.0f;
825 correction_params[1] = 1.0f;
827 /* position is window relative, not viewport relative */
828 correction_params[0] = context->current_rt->currentDesc.Height;
829 correction_params[1] = -1.0f;
831 GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
835 if (priv->next_constant_version == UINT_MAX)
837 TRACE("Max constant version reached, resetting to 0.\n");
838 wine_rb_for_each_entry(&priv->program_lookup, reset_program_constant_version, NULL);
839 priv->next_constant_version = 1;
843 prog->constant_version = priv->next_constant_version++;
847 static inline void update_heap_entry(struct constant_heap *heap, unsigned int idx,
848 unsigned int heap_idx, DWORD new_version)
850 struct constant_entry *entries = heap->entries;
851 unsigned int *positions = heap->positions;
852 unsigned int parent_idx;
856 parent_idx = heap_idx >> 1;
858 if (new_version <= entries[parent_idx].version) break;
860 entries[heap_idx] = entries[parent_idx];
861 positions[entries[parent_idx].idx] = heap_idx;
862 heap_idx = parent_idx;
865 entries[heap_idx].version = new_version;
866 entries[heap_idx].idx = idx;
867 positions[idx] = heap_idx;
870 static void shader_glsl_update_float_vertex_constants(IWineD3DDevice *iface, UINT start, UINT count)
872 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
873 struct shader_glsl_priv *priv = This->shader_priv;
874 struct constant_heap *heap = &priv->vconst_heap;
877 for (i = start; i < count + start; ++i)
879 if (!This->stateBlock->changed.vertexShaderConstantsF[i])
880 update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
882 update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
886 static void shader_glsl_update_float_pixel_constants(IWineD3DDevice *iface, UINT start, UINT count)
888 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
889 struct shader_glsl_priv *priv = This->shader_priv;
890 struct constant_heap *heap = &priv->pconst_heap;
893 for (i = start; i < count + start; ++i)
895 if (!This->stateBlock->changed.pixelShaderConstantsF[i])
896 update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
898 update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
902 static unsigned int vec4_varyings(DWORD shader_major, const struct wined3d_gl_info *gl_info)
904 unsigned int ret = gl_info->limits.glsl_varyings / 4;
905 /* 4.0 shaders do not write clip coords because d3d10 does not support user clipplanes */
906 if(shader_major > 3) return ret;
908 /* 3.0 shaders may need an extra varying for the clip coord on some cards(mostly dx10 ones) */
909 if (gl_info->quirks & WINED3D_QUIRK_GLSL_CLIP_VARYING) ret -= 1;
913 /** Generate the variable & register declarations for the GLSL output target */
914 static void shader_generate_glsl_declarations(const struct wined3d_context *context,
915 struct wined3d_shader_buffer *buffer, IWineD3DBaseShader *iface,
916 const shader_reg_maps *reg_maps, struct shader_glsl_ctx_priv *ctx_priv)
918 IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
919 IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
920 const struct ps_compile_args *ps_args = ctx_priv->cur_ps_args;
921 const struct wined3d_gl_info *gl_info = context->gl_info;
922 unsigned int i, extra_constants_needed = 0;
923 const local_constant *lconst;
926 /* There are some minor differences between pixel and vertex shaders */
927 char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
928 char prefix = pshader ? 'P' : 'V';
930 /* Prototype the subroutines */
931 for (i = 0, map = reg_maps->labels; map; map >>= 1, ++i)
933 if (map & 1) shader_addline(buffer, "void subroutine%u();\n", i);
936 /* Declare the constants (aka uniforms) */
937 if (This->baseShader.limits.constant_float > 0) {
938 unsigned max_constantsF;
939 /* Unless the shader uses indirect addressing, always declare the maximum array size and ignore that we need some
940 * uniforms privately. E.g. if GL supports 256 uniforms, and we need 2 for the pos fixup and immediate values, still
941 * declare VC[256]. If the shader needs more uniforms than we have it won't work in any case. If it uses less, the
942 * compiler will figure out which uniforms are really used and strip them out. This allows a shader to use c255 on
943 * a dx9 card, as long as it doesn't also use all the other constants.
945 * If the shader uses indirect addressing the compiler must assume that all declared uniforms are used. In this case,
946 * declare only the amount that we're assured to have.
948 * Thus we run into problems in these two cases:
949 * 1) The shader really uses more uniforms than supported
950 * 2) The shader uses indirect addressing, less constants than supported, but uses a constant index > #supported consts
954 /* No indirect addressing here. */
955 max_constantsF = gl_info->limits.glsl_ps_float_constants;
959 if(This->baseShader.reg_maps.usesrelconstF) {
960 /* Subtract the other potential uniforms from the max available (bools, ints, and 1 row of projection matrix).
961 * Subtract another uniform for immediate values, which have to be loaded via uniform by the driver as well.
962 * The shader code only uses 0.5, 2.0, 1.0, 128 and -128 in vertex shader code, so one vec4 should be enough
963 * (Unfortunately the Nvidia driver doesn't store 128 and -128 in one float).
965 * Writing gl_ClipVertex requires one uniform for each clipplane as well.
967 max_constantsF = gl_info->limits.glsl_vs_float_constants - 3;
968 if(ctx_priv->cur_vs_args->clip_enabled)
970 max_constantsF -= gl_info->limits.clipplanes;
972 max_constantsF -= count_bits(This->baseShader.reg_maps.integer_constants);
973 /* Strictly speaking a bool only uses one scalar, but the nvidia(Linux) compiler doesn't pack them properly,
974 * so each scalar requires a full vec4. We could work around this by packing the booleans ourselves, but
975 * for now take this into account when calculating the number of available constants
977 max_constantsF -= count_bits(This->baseShader.reg_maps.boolean_constants);
978 /* Set by driver quirks in directx.c */
979 max_constantsF -= gl_info->reserved_glsl_constants;
983 max_constantsF = gl_info->limits.glsl_vs_float_constants;
986 max_constantsF = min(This->baseShader.limits.constant_float, max_constantsF);
987 shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
990 /* Always declare the full set of constants, the compiler can remove the unused ones because d3d doesn't(yet)
991 * support indirect int and bool constant addressing. This avoids problems if the app uses e.g. i0 and i9.
993 if (This->baseShader.limits.constant_int > 0 && This->baseShader.reg_maps.integer_constants)
994 shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, This->baseShader.limits.constant_int);
996 if (This->baseShader.limits.constant_bool > 0 && This->baseShader.reg_maps.boolean_constants)
997 shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, This->baseShader.limits.constant_bool);
1001 shader_addline(buffer, "uniform vec4 posFixup;\n");
1002 shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
1006 for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
1008 if (!(map & 1)) continue;
1010 shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
1012 if (reg_maps->luminanceparams & (1 << i))
1014 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
1015 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
1016 extra_constants_needed++;
1019 extra_constants_needed++;
1022 if (ps_args->srgb_correction)
1024 shader_addline(buffer, "const vec4 srgb_const0 = vec4(%.8e, %.8e, %.8e, %.8e);\n",
1025 srgb_pow, srgb_mul_high, srgb_sub_high, srgb_mul_low);
1026 shader_addline(buffer, "const vec4 srgb_const1 = vec4(%.8e, 0.0, 0.0, 0.0);\n",
1029 if (reg_maps->vpos || reg_maps->usesdsy)
1031 if (This->baseShader.limits.constant_float + extra_constants_needed
1032 + 1 < gl_info->limits.glsl_ps_float_constants)
1034 shader_addline(buffer, "uniform vec4 ycorrection;\n");
1035 ((IWineD3DPixelShaderImpl *) This)->vpos_uniform = 1;
1036 extra_constants_needed++;
1038 /* This happens because we do not have proper tracking of the constant registers that are
1039 * actually used, only the max limit of the shader version
1041 FIXME("Cannot find a free uniform for vpos correction params\n");
1042 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
1043 context->render_offscreen ? 0.0f : device->render_targets[0]->currentDesc.Height,
1044 context->render_offscreen ? 1.0f : -1.0f);
1046 shader_addline(buffer, "vec4 vpos;\n");
1050 /* Declare texture samplers */
1051 for (i = 0; i < This->baseShader.limits.sampler; i++) {
1052 if (reg_maps->sampler_type[i])
1054 IWineD3DBaseTextureImpl *texture;
1056 switch (reg_maps->sampler_type[i])
1059 if (pshader && ps_args->shadow & (1 << i))
1060 shader_addline(buffer, "uniform sampler1DShadow %csampler%u;\n", prefix, i);
1062 shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
1065 texture = device->stateBlock->state.textures[i];
1066 if (pshader && ps_args->shadow & (1 << i))
1068 if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
1069 shader_addline(buffer, "uniform sampler2DRectShadow %csampler%u;\n", prefix, i);
1071 shader_addline(buffer, "uniform sampler2DShadow %csampler%u;\n", prefix, i);
1075 if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
1076 shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
1078 shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
1081 case WINED3DSTT_CUBE:
1082 if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported Cube shadow sampler.\n");
1083 shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
1085 case WINED3DSTT_VOLUME:
1086 if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported 3D shadow sampler.\n");
1087 shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
1090 shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
1091 FIXME("Unrecognized sampler type: %#x\n", reg_maps->sampler_type[i]);
1097 /* Declare uniforms for NP2 texcoord fixup:
1098 * This is NOT done inside the loop that declares the texture samplers since the NP2 fixup code
1099 * is currently only used for the GeforceFX series and when forcing the ARB_npot extension off.
1100 * Modern cards just skip the code anyway, so put it inside a separate loop. */
1101 if (pshader && ps_args->np2_fixup) {
1103 struct ps_np2fixup_info* const fixup = ctx_priv->cur_np2fixup_info;
1106 /* NP2/RECT textures in OpenGL use texcoords in the range [0,width]x[0,height]
1107 * while D3D has them in the (normalized) [0,1]x[0,1] range.
1108 * samplerNP2Fixup stores texture dimensions and is updated through
1109 * shader_glsl_load_np2fixup_constants when the sampler changes. */
1111 for (i = 0; i < This->baseShader.limits.sampler; ++i) {
1112 if (reg_maps->sampler_type[i]) {
1113 if (!(ps_args->np2_fixup & (1 << i))) continue;
1115 if (WINED3DSTT_2D != reg_maps->sampler_type[i]) {
1116 FIXME("Non-2D texture is flagged for NP2 texcoord fixup.\n");
1120 fixup->idx[i] = cur++;
1124 fixup->num_consts = (cur + 1) >> 1;
1125 shader_addline(buffer, "uniform vec4 %csamplerNP2Fixup[%u];\n", prefix, fixup->num_consts);
1128 /* Declare address variables */
1129 for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
1131 if (map & 1) shader_addline(buffer, "ivec4 A%u;\n", i);
1134 /* Declare texture coordinate temporaries and initialize them */
1135 for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
1137 if (map & 1) shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
1140 /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
1141 * helper function shader that is linked in at link time
1143 if (pshader && reg_maps->shader_version.major >= 3)
1145 if (use_vs(device->stateBlock))
1147 shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1149 /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
1150 * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
1151 * pixel shader that reads the fixed function color into the packed input registers.
1153 shader_addline(buffer, "vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1157 /* Declare output register temporaries */
1158 if(This->baseShader.limits.packed_output) {
1159 shader_addline(buffer, "vec4 OUT[%u];\n", This->baseShader.limits.packed_output);
1162 /* Declare temporary variables */
1163 for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
1165 if (map & 1) shader_addline(buffer, "vec4 R%u;\n", i);
1168 /* Declare attributes */
1169 if (reg_maps->shader_version.type == WINED3D_SHADER_TYPE_VERTEX)
1171 for (i = 0, map = reg_maps->input_registers; map; map >>= 1, ++i)
1173 if (map & 1) shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
1177 /* Declare loop registers aLx */
1178 for (i = 0; i < reg_maps->loop_depth; i++) {
1179 shader_addline(buffer, "int aL%u;\n", i);
1180 shader_addline(buffer, "int tmpInt%u;\n", i);
1183 /* Temporary variables for matrix operations */
1184 shader_addline(buffer, "vec4 tmp0;\n");
1185 shader_addline(buffer, "vec4 tmp1;\n");
1187 /* Local constants use a different name so they can be loaded once at shader link time
1188 * They can't be hardcoded into the shader text via LC = {x, y, z, w}; because the
1189 * float -> string conversion can cause precision loss.
1191 if(!This->baseShader.load_local_constsF) {
1192 LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
1193 shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
1197 /* Start the main program */
1198 shader_addline(buffer, "void main() {\n");
1199 if(pshader && reg_maps->vpos) {
1200 /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
1201 * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
1202 * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
1203 * precision troubles when we just substract 0.5.
1205 * To deal with that just floor() the position. This will eliminate the fraction on all cards.
1207 * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
1209 * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
1210 * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
1211 * coordinates specify the pixel centers instead of the pixel corners. This code will behave
1212 * correctly on drivers that returns integer values.
1214 shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
1218 /*****************************************************************************
1219 * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
1221 * For more information, see http://wiki.winehq.org/DirectX-Shaders
1222 ****************************************************************************/
1225 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1226 const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src);
1228 /** Used for opcode modifiers - They multiply the result by the specified amount */
1229 static const char * const shift_glsl_tab[] = {
1231 "2.0 * ", /* 1 (x2) */
1232 "4.0 * ", /* 2 (x4) */
1233 "8.0 * ", /* 3 (x8) */
1234 "16.0 * ", /* 4 (x16) */
1235 "32.0 * ", /* 5 (x32) */
1242 "0.0625 * ", /* 12 (d16) */
1243 "0.125 * ", /* 13 (d8) */
1244 "0.25 * ", /* 14 (d4) */
1245 "0.5 * " /* 15 (d2) */
1248 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
1249 static void shader_glsl_gen_modifier(DWORD src_modifier, const char *in_reg, const char *in_regswizzle, char *out_str)
1253 switch (src_modifier)
1255 case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
1256 case WINED3DSPSM_DW:
1257 case WINED3DSPSM_NONE:
1258 sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1260 case WINED3DSPSM_NEG:
1261 sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
1263 case WINED3DSPSM_NOT:
1264 sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
1266 case WINED3DSPSM_BIAS:
1267 sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1269 case WINED3DSPSM_BIASNEG:
1270 sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1272 case WINED3DSPSM_SIGN:
1273 sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1275 case WINED3DSPSM_SIGNNEG:
1276 sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1278 case WINED3DSPSM_COMP:
1279 sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
1281 case WINED3DSPSM_X2:
1282 sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
1284 case WINED3DSPSM_X2NEG:
1285 sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
1287 case WINED3DSPSM_ABS:
1288 sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
1290 case WINED3DSPSM_ABSNEG:
1291 sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
1294 FIXME("Unhandled modifier %u\n", src_modifier);
1295 sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1299 /** Writes the GLSL variable name that corresponds to the register that the
1300 * DX opcode parameter is trying to access */
1301 static void shader_glsl_get_register_name(const struct wined3d_shader_register *reg,
1302 char *register_name, BOOL *is_color, const struct wined3d_shader_instruction *ins)
1304 /* oPos, oFog and oPts in D3D */
1305 static const char * const hwrastout_reg_names[] = {"OUT[10]", "OUT[11].x", "OUT[11].y"};
1307 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1308 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
1309 char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
1315 case WINED3DSPR_TEMP:
1316 sprintf(register_name, "R%u", reg->idx);
1319 case WINED3DSPR_INPUT:
1320 /* vertex shaders */
1323 struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
1324 if (priv->cur_vs_args->swizzle_map & (1 << reg->idx)) *is_color = TRUE;
1325 sprintf(register_name, "attrib%u", reg->idx);
1329 /* pixel shaders >= 3.0 */
1330 if (This->baseShader.reg_maps.shader_version.major >= 3)
1332 DWORD idx = ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg->idx];
1333 unsigned int in_count = vec4_varyings(This->baseShader.reg_maps.shader_version.major, gl_info);
1337 glsl_src_param_t rel_param;
1339 shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1341 /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
1342 * operation there */
1345 if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count)
1347 sprintf(register_name,
1348 "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
1349 rel_param.param_str, idx, in_count - 1, rel_param.param_str, idx, in_count,
1350 rel_param.param_str, idx);
1354 sprintf(register_name, "IN[%s + %u]", rel_param.param_str, idx);
1359 if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count)
1361 sprintf(register_name, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
1362 rel_param.param_str, in_count - 1, rel_param.param_str, in_count,
1363 rel_param.param_str);
1367 sprintf(register_name, "IN[%s]", rel_param.param_str);
1373 if (idx == in_count) sprintf(register_name, "gl_Color");
1374 else if (idx == in_count + 1) sprintf(register_name, "gl_SecondaryColor");
1375 else sprintf(register_name, "IN[%u]", idx);
1380 if (!reg->idx) strcpy(register_name, "gl_Color");
1381 else strcpy(register_name, "gl_SecondaryColor");
1386 case WINED3DSPR_CONST:
1388 const char prefix = pshader ? 'P' : 'V';
1390 /* Relative addressing */
1393 glsl_src_param_t rel_param;
1394 shader_glsl_add_src_param(ins, reg->rel_addr, WINED3DSP_WRITEMASK_0, &rel_param);
1395 if (reg->idx) sprintf(register_name, "%cC[%s + %u]", prefix, rel_param.param_str, reg->idx);
1396 else sprintf(register_name, "%cC[%s]", prefix, rel_param.param_str);
1400 if (shader_constant_is_local(This, reg->idx))
1401 sprintf(register_name, "%cLC%u", prefix, reg->idx);
1403 sprintf(register_name, "%cC[%u]", prefix, reg->idx);
1408 case WINED3DSPR_CONSTINT:
1409 if (pshader) sprintf(register_name, "PI[%u]", reg->idx);
1410 else sprintf(register_name, "VI[%u]", reg->idx);
1413 case WINED3DSPR_CONSTBOOL:
1414 if (pshader) sprintf(register_name, "PB[%u]", reg->idx);
1415 else sprintf(register_name, "VB[%u]", reg->idx);
1418 case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
1419 if (pshader) sprintf(register_name, "T%u", reg->idx);
1420 else sprintf(register_name, "A%u", reg->idx);
1423 case WINED3DSPR_LOOP:
1424 sprintf(register_name, "aL%u", This->baseShader.cur_loop_regno - 1);
1427 case WINED3DSPR_SAMPLER:
1428 if (pshader) sprintf(register_name, "Psampler%u", reg->idx);
1429 else sprintf(register_name, "Vsampler%u", reg->idx);
1432 case WINED3DSPR_COLOROUT:
1433 if (reg->idx >= gl_info->limits.buffers)
1434 WARN("Write to render target %u, only %d supported.\n", reg->idx, gl_info->limits.buffers);
1436 sprintf(register_name, "gl_FragData[%u]", reg->idx);
1439 case WINED3DSPR_RASTOUT:
1440 sprintf(register_name, "%s", hwrastout_reg_names[reg->idx]);
1443 case WINED3DSPR_DEPTHOUT:
1444 sprintf(register_name, "gl_FragDepth");
1447 case WINED3DSPR_ATTROUT:
1448 if (!reg->idx) sprintf(register_name, "OUT[8]");
1449 else sprintf(register_name, "OUT[9]");
1452 case WINED3DSPR_TEXCRDOUT:
1453 /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
1454 sprintf(register_name, "OUT[%u]", reg->idx);
1457 case WINED3DSPR_MISCTYPE:
1461 sprintf(register_name, "vpos");
1463 else if (reg->idx == 1)
1465 /* Note that gl_FrontFacing is a bool, while vFace is
1466 * a float for which the sign determines front/back */
1467 sprintf(register_name, "(gl_FrontFacing ? 1.0 : -1.0)");
1471 FIXME("Unhandled misctype register %d\n", reg->idx);
1472 sprintf(register_name, "unrecognized_register");
1476 case WINED3DSPR_IMMCONST:
1477 switch (reg->immconst_type)
1479 case WINED3D_IMMCONST_FLOAT:
1480 sprintf(register_name, "%.8e", *(const float *)reg->immconst_data);
1483 case WINED3D_IMMCONST_FLOAT4:
1484 sprintf(register_name, "vec4(%.8e, %.8e, %.8e, %.8e)",
1485 *(const float *)®->immconst_data[0], *(const float *)®->immconst_data[1],
1486 *(const float *)®->immconst_data[2], *(const float *)®->immconst_data[3]);
1490 FIXME("Unhandled immconst type %#x\n", reg->immconst_type);
1491 sprintf(register_name, "<unhandled_immconst_type %#x>", reg->immconst_type);
1496 FIXME("Unhandled register name Type(%d)\n", reg->type);
1497 sprintf(register_name, "unrecognized_register");
1502 static void shader_glsl_write_mask_to_str(DWORD write_mask, char *str)
1505 if (write_mask & WINED3DSP_WRITEMASK_0) *str++ = 'x';
1506 if (write_mask & WINED3DSP_WRITEMASK_1) *str++ = 'y';
1507 if (write_mask & WINED3DSP_WRITEMASK_2) *str++ = 'z';
1508 if (write_mask & WINED3DSP_WRITEMASK_3) *str++ = 'w';
1512 /* Get the GLSL write mask for the destination register */
1513 static DWORD shader_glsl_get_write_mask(const struct wined3d_shader_dst_param *param, char *write_mask)
1515 DWORD mask = param->write_mask;
1517 if (shader_is_scalar(¶m->reg))
1519 mask = WINED3DSP_WRITEMASK_0;
1524 shader_glsl_write_mask_to_str(mask, write_mask);
1530 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1531 unsigned int size = 0;
1533 if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1534 if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1535 if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1536 if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1541 static void shader_glsl_swizzle_to_str(const DWORD swizzle, BOOL fixup, DWORD mask, char *str)
1543 /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1544 * but addressed as "rgba". To fix this we need to swap the register's x
1545 * and z components. */
1546 const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1549 /* swizzle bits fields: wwzzyyxx */
1550 if (mask & WINED3DSP_WRITEMASK_0) *str++ = swizzle_chars[swizzle & 0x03];
1551 if (mask & WINED3DSP_WRITEMASK_1) *str++ = swizzle_chars[(swizzle >> 2) & 0x03];
1552 if (mask & WINED3DSP_WRITEMASK_2) *str++ = swizzle_chars[(swizzle >> 4) & 0x03];
1553 if (mask & WINED3DSP_WRITEMASK_3) *str++ = swizzle_chars[(swizzle >> 6) & 0x03];
1557 static void shader_glsl_get_swizzle(const struct wined3d_shader_src_param *param,
1558 BOOL fixup, DWORD mask, char *swizzle_str)
1560 if (shader_is_scalar(¶m->reg))
1561 *swizzle_str = '\0';
1563 shader_glsl_swizzle_to_str(param->swizzle, fixup, mask, swizzle_str);
1566 /* From a given parameter token, generate the corresponding GLSL string.
1567 * Also, return the actual register name and swizzle in case the
1568 * caller needs this information as well. */
1569 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1570 const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src)
1572 BOOL is_color = FALSE;
1573 char swizzle_str[6];
1575 glsl_src->reg_name[0] = '\0';
1576 glsl_src->param_str[0] = '\0';
1577 swizzle_str[0] = '\0';
1579 shader_glsl_get_register_name(&wined3d_src->reg, glsl_src->reg_name, &is_color, ins);
1580 shader_glsl_get_swizzle(wined3d_src, is_color, mask, swizzle_str);
1581 shader_glsl_gen_modifier(wined3d_src->modifiers, glsl_src->reg_name, swizzle_str, glsl_src->param_str);
1584 /* From a given parameter token, generate the corresponding GLSL string.
1585 * Also, return the actual register name and swizzle in case the
1586 * caller needs this information as well. */
1587 static DWORD shader_glsl_add_dst_param(const struct wined3d_shader_instruction *ins,
1588 const struct wined3d_shader_dst_param *wined3d_dst, glsl_dst_param_t *glsl_dst)
1590 BOOL is_color = FALSE;
1592 glsl_dst->mask_str[0] = '\0';
1593 glsl_dst->reg_name[0] = '\0';
1595 shader_glsl_get_register_name(&wined3d_dst->reg, glsl_dst->reg_name, &is_color, ins);
1596 return shader_glsl_get_write_mask(wined3d_dst, glsl_dst->mask_str);
1599 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1600 static DWORD shader_glsl_append_dst_ext(struct wined3d_shader_buffer *buffer,
1601 const struct wined3d_shader_instruction *ins, const struct wined3d_shader_dst_param *dst)
1603 glsl_dst_param_t glsl_dst;
1606 mask = shader_glsl_add_dst_param(ins, dst, &glsl_dst);
1607 if (mask) shader_addline(buffer, "%s%s = %s(", glsl_dst.reg_name, glsl_dst.mask_str, shift_glsl_tab[dst->shift]);
1612 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1613 static DWORD shader_glsl_append_dst(struct wined3d_shader_buffer *buffer, const struct wined3d_shader_instruction *ins)
1615 return shader_glsl_append_dst_ext(buffer, ins, &ins->dst[0]);
1618 /** Process GLSL instruction modifiers */
1619 static void shader_glsl_add_instruction_modifiers(const struct wined3d_shader_instruction *ins)
1621 glsl_dst_param_t dst_param;
1624 if (!ins->dst_count) return;
1626 modifiers = ins->dst[0].modifiers;
1627 if (!modifiers) return;
1629 shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
1631 if (modifiers & WINED3DSPDM_SATURATE)
1633 /* _SAT means to clamp the value of the register to between 0 and 1 */
1634 shader_addline(ins->ctx->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1635 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1638 if (modifiers & WINED3DSPDM_MSAMPCENTROID)
1640 FIXME("_centroid modifier not handled\n");
1643 if (modifiers & WINED3DSPDM_PARTIALPRECISION)
1645 /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1649 static inline const char *shader_get_comp_op(DWORD op)
1652 case COMPARISON_GT: return ">";
1653 case COMPARISON_EQ: return "==";
1654 case COMPARISON_GE: return ">=";
1655 case COMPARISON_LT: return "<";
1656 case COMPARISON_NE: return "!=";
1657 case COMPARISON_LE: return "<=";
1659 FIXME("Unrecognized comparison value: %u\n", op);
1664 static void shader_glsl_get_sample_function(const struct wined3d_shader_context *ctx,
1665 DWORD sampler_idx, DWORD flags, glsl_sample_function_t *sample_function)
1667 WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ctx->reg_maps->sampler_type[sampler_idx];
1668 const struct wined3d_gl_info *gl_info = ctx->gl_info;
1669 BOOL shadow = shader_is_pshader_version(ctx->reg_maps->shader_version.type)
1670 && (((const struct shader_glsl_ctx_priv *)ctx->backend_data)->cur_ps_args->shadow & (1 << sampler_idx));
1671 BOOL projected = flags & WINED3D_GLSL_SAMPLE_PROJECTED;
1672 BOOL texrect = flags & WINED3D_GLSL_SAMPLE_RECT;
1673 BOOL lod = flags & WINED3D_GLSL_SAMPLE_LOD;
1674 BOOL grad = flags & WINED3D_GLSL_SAMPLE_GRAD;
1676 /* Note that there's no such thing as a projected cube texture. */
1677 switch(sampler_type) {
1683 sample_function->name = projected ? "shadow1DProjLod" : "shadow1DLod";
1687 if (gl_info->supported[EXT_GPU_SHADER4])
1688 sample_function->name = projected ? "shadow1DProjGrad" : "shadow1DGrad";
1689 else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1690 sample_function->name = projected ? "shadow1DProjGradARB" : "shadow1DGradARB";
1693 FIXME("Unsupported 1D shadow grad function.\n");
1694 sample_function->name = "unsupported1DGrad";
1699 sample_function->name = projected ? "shadow1DProj" : "shadow1D";
1701 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1707 sample_function->name = projected ? "texture1DProjLod" : "texture1DLod";
1711 if (gl_info->supported[EXT_GPU_SHADER4])
1712 sample_function->name = projected ? "texture1DProjGrad" : "texture1DGrad";
1713 else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1714 sample_function->name = projected ? "texture1DProjGradARB" : "texture1DGradARB";
1717 FIXME("Unsupported 1D grad function.\n");
1718 sample_function->name = "unsupported1DGrad";
1723 sample_function->name = projected ? "texture1DProj" : "texture1D";
1725 sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1736 sample_function->name = projected ? "shadow2DRectProjLod" : "shadow2DRectLod";
1740 if (gl_info->supported[EXT_GPU_SHADER4])
1741 sample_function->name = projected ? "shadow2DRectProjGrad" : "shadow2DRectGrad";
1742 else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1743 sample_function->name = projected ? "shadow2DRectProjGradARB" : "shadow2DRectGradARB";
1746 FIXME("Unsupported RECT shadow grad function.\n");
1747 sample_function->name = "unsupported2DRectGrad";
1752 sample_function->name = projected ? "shadow2DRectProj" : "shadow2DRect";
1759 sample_function->name = projected ? "shadow2DProjLod" : "shadow2DLod";
1763 if (gl_info->supported[EXT_GPU_SHADER4])
1764 sample_function->name = projected ? "shadow2DProjGrad" : "shadow2DGrad";
1765 else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1766 sample_function->name = projected ? "shadow2DProjGradARB" : "shadow2DGradARB";
1769 FIXME("Unsupported 2D shadow grad function.\n");
1770 sample_function->name = "unsupported2DGrad";
1775 sample_function->name = projected ? "shadow2DProj" : "shadow2D";
1778 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1786 sample_function->name = projected ? "texture2DRectProjLod" : "texture2DRectLod";
1790 if (gl_info->supported[EXT_GPU_SHADER4])
1791 sample_function->name = projected ? "texture2DRectProjGrad" : "texture2DRectGrad";
1792 else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1793 sample_function->name = projected ? "texture2DRectProjGradARB" : "texture2DRectGradARB";
1796 FIXME("Unsupported RECT grad function.\n");
1797 sample_function->name = "unsupported2DRectGrad";
1802 sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1809 sample_function->name = projected ? "texture2DProjLod" : "texture2DLod";
1813 if (gl_info->supported[EXT_GPU_SHADER4])
1814 sample_function->name = projected ? "texture2DProjGrad" : "texture2DGrad";
1815 else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1816 sample_function->name = projected ? "texture2DProjGradARB" : "texture2DGradARB";
1819 FIXME("Unsupported 2D grad function.\n");
1820 sample_function->name = "unsupported2DGrad";
1825 sample_function->name = projected ? "texture2DProj" : "texture2D";
1828 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1832 case WINED3DSTT_CUBE:
1835 FIXME("Unsupported Cube shadow function.\n");
1836 sample_function->name = "unsupportedCubeShadow";
1837 sample_function->coord_mask = 0;
1843 sample_function->name = "textureCubeLod";
1847 if (gl_info->supported[EXT_GPU_SHADER4])
1848 sample_function->name = "textureCubeGrad";
1849 else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1850 sample_function->name = "textureCubeGradARB";
1853 FIXME("Unsupported Cube grad function.\n");
1854 sample_function->name = "unsupportedCubeGrad";
1859 sample_function->name = "textureCube";
1861 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1865 case WINED3DSTT_VOLUME:
1868 FIXME("Unsupported 3D shadow function.\n");
1869 sample_function->name = "unsupported3DShadow";
1870 sample_function->coord_mask = 0;
1876 sample_function->name = projected ? "texture3DProjLod" : "texture3DLod";
1880 if (gl_info->supported[EXT_GPU_SHADER4])
1881 sample_function->name = projected ? "texture3DProjGrad" : "texture3DGrad";
1882 else if (gl_info->supported[ARB_SHADER_TEXTURE_LOD])
1883 sample_function->name = projected ? "texture3DProjGradARB" : "texture3DGradARB";
1886 FIXME("Unsupported 3D grad function.\n");
1887 sample_function->name = "unsupported3DGrad";
1892 sample_function->name = projected ? "texture3DProj" : "texture3D";
1894 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1899 sample_function->name = "";
1900 sample_function->coord_mask = 0;
1901 FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1906 static void shader_glsl_append_fixup_arg(char *arguments, const char *reg_name,
1907 BOOL sign_fixup, enum fixup_channel_source channel_source)
1909 switch(channel_source)
1911 case CHANNEL_SOURCE_ZERO:
1912 strcat(arguments, "0.0");
1915 case CHANNEL_SOURCE_ONE:
1916 strcat(arguments, "1.0");
1919 case CHANNEL_SOURCE_X:
1920 strcat(arguments, reg_name);
1921 strcat(arguments, ".x");
1924 case CHANNEL_SOURCE_Y:
1925 strcat(arguments, reg_name);
1926 strcat(arguments, ".y");
1929 case CHANNEL_SOURCE_Z:
1930 strcat(arguments, reg_name);
1931 strcat(arguments, ".z");
1934 case CHANNEL_SOURCE_W:
1935 strcat(arguments, reg_name);
1936 strcat(arguments, ".w");
1940 FIXME("Unhandled channel source %#x\n", channel_source);
1941 strcat(arguments, "undefined");
1945 if (sign_fixup) strcat(arguments, " * 2.0 - 1.0");
1948 static void shader_glsl_color_correction(const struct wined3d_shader_instruction *ins, struct color_fixup_desc fixup)
1950 struct wined3d_shader_dst_param dst;
1951 unsigned int mask_size, remaining;
1952 glsl_dst_param_t dst_param;
1953 char arguments[256];
1957 if (fixup.x_sign_fixup || fixup.x_source != CHANNEL_SOURCE_X) mask |= WINED3DSP_WRITEMASK_0;
1958 if (fixup.y_sign_fixup || fixup.y_source != CHANNEL_SOURCE_Y) mask |= WINED3DSP_WRITEMASK_1;
1959 if (fixup.z_sign_fixup || fixup.z_source != CHANNEL_SOURCE_Z) mask |= WINED3DSP_WRITEMASK_2;
1960 if (fixup.w_sign_fixup || fixup.w_source != CHANNEL_SOURCE_W) mask |= WINED3DSP_WRITEMASK_3;
1961 mask &= ins->dst[0].write_mask;
1963 if (!mask) return; /* Nothing to do */
1965 if (is_complex_fixup(fixup))
1967 enum complex_fixup complex_fixup = get_complex_fixup(fixup);
1968 FIXME("Complex fixup (%#x) not supported\n",complex_fixup);
1972 mask_size = shader_glsl_get_write_mask_size(mask);
1975 dst.write_mask = mask;
1976 shader_glsl_add_dst_param(ins, &dst, &dst_param);
1978 arguments[0] = '\0';
1979 remaining = mask_size;
1980 if (mask & WINED3DSP_WRITEMASK_0)
1982 shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.x_sign_fixup, fixup.x_source);
1983 if (--remaining) strcat(arguments, ", ");
1985 if (mask & WINED3DSP_WRITEMASK_1)
1987 shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.y_sign_fixup, fixup.y_source);
1988 if (--remaining) strcat(arguments, ", ");
1990 if (mask & WINED3DSP_WRITEMASK_2)
1992 shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.z_sign_fixup, fixup.z_source);
1993 if (--remaining) strcat(arguments, ", ");
1995 if (mask & WINED3DSP_WRITEMASK_3)
1997 shader_glsl_append_fixup_arg(arguments, dst_param.reg_name, fixup.w_sign_fixup, fixup.w_source);
1998 if (--remaining) strcat(arguments, ", ");
2003 shader_addline(ins->ctx->buffer, "%s%s = vec%u(%s);\n",
2004 dst_param.reg_name, dst_param.mask_str, mask_size, arguments);
2008 shader_addline(ins->ctx->buffer, "%s%s = %s;\n", dst_param.reg_name, dst_param.mask_str, arguments);
2012 static void PRINTF_ATTR(8, 9) shader_glsl_gen_sample_code(const struct wined3d_shader_instruction *ins,
2013 DWORD sampler, const glsl_sample_function_t *sample_function, DWORD swizzle,
2014 const char *dx, const char *dy,
2015 const char *bias, const char *coord_reg_fmt, ...)
2017 const char *sampler_base;
2018 char dst_swizzle[6];
2019 struct color_fixup_desc fixup;
2020 BOOL np2_fixup = FALSE;
2023 shader_glsl_swizzle_to_str(swizzle, FALSE, ins->dst[0].write_mask, dst_swizzle);
2025 if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
2027 const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2028 fixup = priv->cur_ps_args->color_fixup[sampler];
2029 sampler_base = "Psampler";
2031 if(priv->cur_ps_args->np2_fixup & (1 << sampler)) {
2033 FIXME("Biased sampling from NP2 textures is unsupported\n");
2039 sampler_base = "Vsampler";
2040 fixup = COLOR_FIXUP_IDENTITY; /* FIXME: Vshader color fixup */
2043 shader_glsl_append_dst(ins->ctx->buffer, ins);
2045 shader_addline(ins->ctx->buffer, "%s(%s%u, ", sample_function->name, sampler_base, sampler);
2047 va_start(args, coord_reg_fmt);
2048 shader_vaddline(ins->ctx->buffer, coord_reg_fmt, args);
2052 shader_addline(ins->ctx->buffer, ", %s)%s);\n", bias, dst_swizzle);
2055 const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
2056 const unsigned char idx = priv->cur_np2fixup_info->idx[sampler];
2058 shader_addline(ins->ctx->buffer, " * PsamplerNP2Fixup[%u].%s)%s);\n", idx >> 1,
2059 (idx % 2) ? "zw" : "xy", dst_swizzle);
2060 } else if(dx && dy) {
2061 shader_addline(ins->ctx->buffer, ", %s, %s)%s);\n", dx, dy, dst_swizzle);
2063 shader_addline(ins->ctx->buffer, ")%s);\n", dst_swizzle);
2067 if(!is_identity_fixup(fixup)) {
2068 shader_glsl_color_correction(ins, fixup);
2072 /*****************************************************************************
2073 * Begin processing individual instruction opcodes
2074 ****************************************************************************/
2076 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
2077 static void shader_glsl_arith(const struct wined3d_shader_instruction *ins)
2079 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2080 glsl_src_param_t src0_param;
2081 glsl_src_param_t src1_param;
2085 /* Determine the GLSL operator to use based on the opcode */
2086 switch (ins->handler_idx)
2088 case WINED3DSIH_MUL: op = '*'; break;
2089 case WINED3DSIH_ADD: op = '+'; break;
2090 case WINED3DSIH_SUB: op = '-'; break;
2093 FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2097 write_mask = shader_glsl_append_dst(buffer, ins);
2098 shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2099 shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2100 shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
2103 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
2104 static void shader_glsl_mov(const struct wined3d_shader_instruction *ins)
2106 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
2107 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2108 glsl_src_param_t src0_param;
2111 write_mask = shader_glsl_append_dst(buffer, ins);
2112 shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2114 /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
2115 * shader versions WINED3DSIO_MOVA is used for this. */
2116 if (ins->ctx->reg_maps->shader_version.major == 1
2117 && !shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)
2118 && ins->dst[0].reg.type == WINED3DSPR_ADDR)
2120 /* This is a simple floor() */
2121 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2122 if (mask_size > 1) {
2123 shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
2125 shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
2128 else if(ins->handler_idx == WINED3DSIH_MOVA)
2130 /* We need to *round* to the nearest int here. */
2131 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2133 if (gl_info->supported[EXT_GPU_SHADER4])
2136 shader_addline(buffer, "ivec%d(round(%s)));\n", mask_size, src0_param.param_str);
2138 shader_addline(buffer, "int(round(%s)));\n", src0_param.param_str);
2143 shader_addline(buffer, "ivec%d(floor(abs(%s) + vec%d(0.5)) * sign(%s)));\n",
2144 mask_size, src0_param.param_str, mask_size, src0_param.param_str);
2146 shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n",
2147 src0_param.param_str, src0_param.param_str);
2152 shader_addline(buffer, "%s);\n", src0_param.param_str);
2156 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
2157 static void shader_glsl_dot(const struct wined3d_shader_instruction *ins)
2159 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2160 glsl_src_param_t src0_param;
2161 glsl_src_param_t src1_param;
2162 DWORD dst_write_mask, src_write_mask;
2163 unsigned int dst_size = 0;
2165 dst_write_mask = shader_glsl_append_dst(buffer, ins);
2166 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2168 /* dp3 works on vec3, dp4 on vec4 */
2169 if (ins->handler_idx == WINED3DSIH_DP4)
2171 src_write_mask = WINED3DSP_WRITEMASK_ALL;
2173 src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2176 shader_glsl_add_src_param(ins, &ins->src[0], src_write_mask, &src0_param);
2177 shader_glsl_add_src_param(ins, &ins->src[1], src_write_mask, &src1_param);
2180 shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
2182 shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
2186 /* Note that this instruction has some restrictions. The destination write mask
2187 * can't contain the w component, and the source swizzles have to be .xyzw */
2188 static void shader_glsl_cross(const struct wined3d_shader_instruction *ins)
2190 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2191 glsl_src_param_t src0_param;
2192 glsl_src_param_t src1_param;
2195 shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2196 shader_glsl_append_dst(ins->ctx->buffer, ins);
2197 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
2198 shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
2199 shader_addline(ins->ctx->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
2202 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
2203 * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
2204 * GLSL uses the value as-is. */
2205 static void shader_glsl_pow(const struct wined3d_shader_instruction *ins)
2207 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2208 glsl_src_param_t src0_param;
2209 glsl_src_param_t src1_param;
2210 DWORD dst_write_mask;
2211 unsigned int dst_size;
2213 dst_write_mask = shader_glsl_append_dst(buffer, ins);
2214 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2216 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2217 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2221 shader_addline(buffer, "vec%u(pow(abs(%s), %s)));\n",
2222 dst_size, src0_param.param_str, src1_param.param_str);
2226 shader_addline(buffer, "pow(abs(%s), %s));\n",
2227 src0_param.param_str, src1_param.param_str);
2231 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
2232 * Src0 is a scalar. Note that D3D uses the absolute of src0, while
2233 * GLSL uses the value as-is. */
2234 static void shader_glsl_log(const struct wined3d_shader_instruction *ins)
2236 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2237 glsl_src_param_t src0_param;
2238 DWORD dst_write_mask;
2239 unsigned int dst_size;
2241 dst_write_mask = shader_glsl_append_dst(buffer, ins);
2242 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
2244 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2248 shader_addline(buffer, "vec%u(log2(abs(%s))));\n",
2249 dst_size, src0_param.param_str);
2253 shader_addline(buffer, "log2(abs(%s)));\n",
2254 src0_param.param_str);
2258 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
2259 static void shader_glsl_map2gl(const struct wined3d_shader_instruction *ins)
2261 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2262 glsl_src_param_t src_param;
2263 const char *instruction;
2267 /* Determine the GLSL function to use based on the opcode */
2268 /* TODO: Possibly make this a table for faster lookups */
2269 switch (ins->handler_idx)
2271 case WINED3DSIH_MIN: instruction = "min"; break;
2272 case WINED3DSIH_MAX: instruction = "max"; break;
2273 case WINED3DSIH_ABS: instruction = "abs"; break;
2274 case WINED3DSIH_FRC: instruction = "fract"; break;
2275 case WINED3DSIH_EXP: instruction = "exp2"; break;
2276 case WINED3DSIH_DSX: instruction = "dFdx"; break;
2277 case WINED3DSIH_DSY: instruction = "ycorrection.y * dFdy"; break;
2278 default: instruction = "";
2279 FIXME("Opcode %#x not yet handled in GLSL\n", ins->handler_idx);
2283 write_mask = shader_glsl_append_dst(buffer, ins);
2285 shader_addline(buffer, "%s(", instruction);
2289 shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2290 shader_addline(buffer, "%s", src_param.param_str);
2291 for (i = 1; i < ins->src_count; ++i)
2293 shader_glsl_add_src_param(ins, &ins->src[i], write_mask, &src_param);
2294 shader_addline(buffer, ", %s", src_param.param_str);
2298 shader_addline(buffer, "));\n");
2301 static void shader_glsl_nrm(const struct wined3d_shader_instruction *ins)
2303 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2304 glsl_src_param_t src_param;
2305 unsigned int mask_size;
2309 write_mask = shader_glsl_get_write_mask(ins->dst, dst_mask);
2310 mask_size = shader_glsl_get_write_mask_size(write_mask);
2311 shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param);
2313 shader_addline(buffer, "tmp0.x = length(%s);\n", src_param.param_str);
2314 shader_glsl_append_dst(buffer, ins);
2317 shader_addline(buffer, "tmp0.x == 0.0 ? vec%u(0.0) : (%s / tmp0.x));\n",
2318 mask_size, src_param.param_str);
2322 shader_addline(buffer, "tmp0.x == 0.0 ? 0.0 : (%s / tmp0.x));\n",
2323 src_param.param_str);
2327 /** Process the WINED3DSIO_EXPP instruction in GLSL:
2328 * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
2329 * dst.x = 2^(floor(src))
2330 * dst.y = src - floor(src)
2331 * dst.z = 2^src (partial precision is allowed, but optional)
2333 * For 2.0 shaders, just do this (honoring writemask and swizzle):
2334 * dst = 2^src; (partial precision is allowed, but optional)
2336 static void shader_glsl_expp(const struct wined3d_shader_instruction *ins)
2338 glsl_src_param_t src_param;
2340 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src_param);
2342 if (ins->ctx->reg_maps->shader_version.major < 2)
2346 shader_addline(ins->ctx->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
2347 shader_addline(ins->ctx->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
2348 shader_addline(ins->ctx->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
2349 shader_addline(ins->ctx->buffer, "tmp0.w = 1.0;\n");
2351 shader_glsl_append_dst(ins->ctx->buffer, ins);
2352 shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2353 shader_addline(ins->ctx->buffer, "tmp0%s);\n", dst_mask);
2356 unsigned int mask_size;
2358 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2359 mask_size = shader_glsl_get_write_mask_size(write_mask);
2361 if (mask_size > 1) {
2362 shader_addline(ins->ctx->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
2364 shader_addline(ins->ctx->buffer, "exp2(%s));\n", src_param.param_str);
2369 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
2370 static void shader_glsl_rcp(const struct wined3d_shader_instruction *ins)
2372 glsl_src_param_t src_param;
2374 unsigned int mask_size;
2376 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2377 mask_size = shader_glsl_get_write_mask_size(write_mask);
2378 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2382 shader_addline(ins->ctx->buffer, "vec%u(1.0 / %s));\n",
2383 mask_size, src_param.param_str);
2387 shader_addline(ins->ctx->buffer, "1.0 / %s);\n",
2388 src_param.param_str);
2392 static void shader_glsl_rsq(const struct wined3d_shader_instruction *ins)
2394 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
2395 glsl_src_param_t src_param;
2397 unsigned int mask_size;
2399 write_mask = shader_glsl_append_dst(buffer, ins);
2400 mask_size = shader_glsl_get_write_mask_size(write_mask);
2402 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src_param);
2406 shader_addline(buffer, "vec%u(inversesqrt(abs(%s))));\n",
2407 mask_size, src_param.param_str);
2411 shader_addline(buffer, "inversesqrt(abs(%s)));\n",
2412 src_param.param_str);
2416 /** Process signed comparison opcodes in GLSL. */
2417 static void shader_glsl_compare(const struct wined3d_shader_instruction *ins)
2419 glsl_src_param_t src0_param;
2420 glsl_src_param_t src1_param;
2422 unsigned int mask_size;
2424 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2425 mask_size = shader_glsl_get_write_mask_size(write_mask);
2426 shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2427 shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2429 if (mask_size > 1) {
2430 const char *compare;
2432 switch(ins->handler_idx)
2434 case WINED3DSIH_SLT: compare = "lessThan"; break;
2435 case WINED3DSIH_SGE: compare = "greaterThanEqual"; break;
2436 default: compare = "";
2437 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2440 shader_addline(ins->ctx->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
2441 src0_param.param_str, src1_param.param_str);
2443 switch(ins->handler_idx)
2445 case WINED3DSIH_SLT:
2446 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
2447 * to return 0.0 but step returns 1.0 because step is not < x
2448 * An alternative is a bvec compare padded with an unused second component.
2449 * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
2450 * issue. Playing with not() is not possible either because not() does not accept
2453 shader_addline(ins->ctx->buffer, "(%s < %s) ? 1.0 : 0.0);\n",
2454 src0_param.param_str, src1_param.param_str);
2456 case WINED3DSIH_SGE:
2457 /* Here we can use the step() function and safe a conditional */
2458 shader_addline(ins->ctx->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
2461 FIXME("Can't handle opcode %#x\n", ins->handler_idx);
2467 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
2468 static void shader_glsl_cmp(const struct wined3d_shader_instruction *ins)
2470 glsl_src_param_t src0_param;
2471 glsl_src_param_t src1_param;
2472 glsl_src_param_t src2_param;
2473 DWORD write_mask, cmp_channel = 0;
2476 BOOL temp_destination = FALSE;
2478 if (shader_is_scalar(&ins->src[0].reg))
2480 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2482 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2483 shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2484 shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2486 shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2487 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2489 DWORD dst_mask = ins->dst[0].write_mask;
2490 struct wined3d_shader_dst_param dst = ins->dst[0];
2492 /* Cycle through all source0 channels */
2493 for (i=0; i<4; i++) {
2495 /* Find the destination channels which use the current source0 channel */
2496 for (j=0; j<4; j++) {
2497 if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2499 write_mask |= WINED3DSP_WRITEMASK_0 << j;
2500 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2503 dst.write_mask = dst_mask & write_mask;
2505 /* Splitting the cmp instruction up in multiple lines imposes a problem:
2506 * The first lines may overwrite source parameters of the following lines.
2507 * Deal with that by using a temporary destination register if needed
2509 if ((ins->src[0].reg.idx == ins->dst[0].reg.idx
2510 && ins->src[0].reg.type == ins->dst[0].reg.type)
2511 || (ins->src[1].reg.idx == ins->dst[0].reg.idx
2512 && ins->src[1].reg.type == ins->dst[0].reg.type)
2513 || (ins->src[2].reg.idx == ins->dst[0].reg.idx
2514 && ins->src[2].reg.type == ins->dst[0].reg.type))
2516 write_mask = shader_glsl_get_write_mask(&dst, mask_char);
2517 if (!write_mask) continue;
2518 shader_addline(ins->ctx->buffer, "tmp0%s = (", mask_char);
2519 temp_destination = TRUE;
2521 write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2522 if (!write_mask) continue;
2525 shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2526 shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2527 shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2529 shader_addline(ins->ctx->buffer, "%s >= 0.0 ? %s : %s);\n",
2530 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2533 if(temp_destination) {
2534 shader_glsl_get_write_mask(&ins->dst[0], mask_char);
2535 shader_glsl_append_dst(ins->ctx->buffer, ins);
2536 shader_addline(ins->ctx->buffer, "tmp0%s);\n", mask_char);
2542 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
2543 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
2544 * the compare is done per component of src0. */
2545 static void shader_glsl_cnd(const struct wined3d_shader_instruction *ins)
2547 struct wined3d_shader_dst_param dst;
2548 glsl_src_param_t src0_param;
2549 glsl_src_param_t src1_param;
2550 glsl_src_param_t src2_param;
2551 DWORD write_mask, cmp_channel = 0;
2554 DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
2555 ins->ctx->reg_maps->shader_version.minor);
2557 if (shader_version < WINED3D_SHADER_VERSION(1, 4))
2559 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2560 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2561 shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2562 shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2564 /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
2567 shader_addline(ins->ctx->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
2569 shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2570 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2574 /* Cycle through all source0 channels */
2575 dst_mask = ins->dst[0].write_mask;
2577 for (i=0; i<4; i++) {
2579 /* Find the destination channels which use the current source0 channel */
2580 for (j=0; j<4; j++) {
2581 if (((ins->src[0].swizzle >> (2 * j)) & 0x3) == i)
2583 write_mask |= WINED3DSP_WRITEMASK_0 << j;
2584 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
2588 dst.write_mask = dst_mask & write_mask;
2589 write_mask = shader_glsl_append_dst_ext(ins->ctx->buffer, ins, &dst);
2590 if (!write_mask) continue;
2592 shader_glsl_add_src_param(ins, &ins->src[0], cmp_channel, &src0_param);
2593 shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2594 shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2596 shader_addline(ins->ctx->buffer, "%s > 0.5 ? %s : %s);\n",
2597 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2601 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
2602 static void shader_glsl_mad(const struct wined3d_shader_instruction *ins)
2604 glsl_src_param_t src0_param;
2605 glsl_src_param_t src1_param;
2606 glsl_src_param_t src2_param;
2609 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2610 shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2611 shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2612 shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2613 shader_addline(ins->ctx->buffer, "(%s * %s) + %s);\n",
2614 src0_param.param_str, src1_param.param_str, src2_param.param_str);
2617 /* Handles transforming all WINED3DSIO_M?x? opcodes for
2618 Vertex shaders to GLSL codes */
2619 static void shader_glsl_mnxn(const struct wined3d_shader_instruction *ins)
2622 int nComponents = 0;
2623 struct wined3d_shader_dst_param tmp_dst = {{0}};
2624 struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
2625 struct wined3d_shader_instruction tmp_ins;
2627 memset(&tmp_ins, 0, sizeof(tmp_ins));
2629 /* Set constants for the temporary argument */
2630 tmp_ins.ctx = ins->ctx;
2631 tmp_ins.dst_count = 1;
2632 tmp_ins.dst = &tmp_dst;
2633 tmp_ins.src_count = 2;
2634 tmp_ins.src = tmp_src;
2636 switch(ins->handler_idx)
2638 case WINED3DSIH_M4x4:
2640 tmp_ins.handler_idx = WINED3DSIH_DP4;
2642 case WINED3DSIH_M4x3:
2644 tmp_ins.handler_idx = WINED3DSIH_DP4;
2646 case WINED3DSIH_M3x4:
2648 tmp_ins.handler_idx = WINED3DSIH_DP3;
2650 case WINED3DSIH_M3x3:
2652 tmp_ins.handler_idx = WINED3DSIH_DP3;
2654 case WINED3DSIH_M3x2:
2656 tmp_ins.handler_idx = WINED3DSIH_DP3;
2662 tmp_dst = ins->dst[0];
2663 tmp_src[0] = ins->src[0];
2664 tmp_src[1] = ins->src[1];
2665 for (i = 0; i < nComponents; ++i)
2667 tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
2668 shader_glsl_dot(&tmp_ins);
2669 ++tmp_src[1].reg.idx;
2674 The LRP instruction performs a component-wise linear interpolation
2675 between the second and third operands using the first operand as the
2676 blend factor. Equation: (dst = src2 + src0 * (src1 - src2))
2677 This is equivalent to mix(src2, src1, src0);
2679 static void shader_glsl_lrp(const struct wined3d_shader_instruction *ins)
2681 glsl_src_param_t src0_param;
2682 glsl_src_param_t src1_param;
2683 glsl_src_param_t src2_param;
2686 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2688 shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2689 shader_glsl_add_src_param(ins, &ins->src[1], write_mask, &src1_param);
2690 shader_glsl_add_src_param(ins, &ins->src[2], write_mask, &src2_param);
2692 shader_addline(ins->ctx->buffer, "mix(%s, %s, %s));\n",
2693 src2_param.param_str, src1_param.param_str, src0_param.param_str);
2696 /** Process the WINED3DSIO_LIT instruction in GLSL:
2697 * dst.x = dst.w = 1.0
2698 * dst.y = (src0.x > 0) ? src0.x
2699 * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
2700 * where src.w is clamped at +- 128
2702 static void shader_glsl_lit(const struct wined3d_shader_instruction *ins)
2704 glsl_src_param_t src0_param;
2705 glsl_src_param_t src1_param;
2706 glsl_src_param_t src3_param;
2709 shader_glsl_append_dst(ins->ctx->buffer, ins);
2710 shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2712 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2713 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src1_param);
2714 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &src3_param);
2716 /* The sdk specifies the instruction like this
2718 * if(src.x > 0.0) dst.y = src.x
2720 * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
2724 * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
2725 * dst.x = 1.0 ... No further explanation needed
2726 * dst.y = max(src.y, 0.0); ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
2727 * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0; ... 0 ^ power is 0, and otherwise we use y anyway
2728 * dst.w = 1.0. ... Nothing fancy.
2730 * So we still have one conditional in there. So do this:
2731 * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
2733 * 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),
2734 * 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.
2735 * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to
2737 shader_addline(ins->ctx->buffer,
2738 "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",
2739 src0_param.param_str, src1_param.param_str, src0_param.param_str, src3_param.param_str, dst_mask);
2742 /** Process the WINED3DSIO_DST instruction in GLSL:
2744 * dst.y = src0.x * src0.y
2748 static void shader_glsl_dst(const struct wined3d_shader_instruction *ins)
2750 glsl_src_param_t src0y_param;
2751 glsl_src_param_t src0z_param;
2752 glsl_src_param_t src1y_param;
2753 glsl_src_param_t src1w_param;
2756 shader_glsl_append_dst(ins->ctx->buffer, ins);
2757 shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
2759 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_1, &src0y_param);
2760 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &src0z_param);
2761 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_1, &src1y_param);
2762 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_3, &src1w_param);
2764 shader_addline(ins->ctx->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
2765 src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
2768 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
2769 * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
2770 * can handle it. But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
2772 * dst.x = cos(src0.?)
2773 * dst.y = sin(src0.?)
2777 static void shader_glsl_sincos(const struct wined3d_shader_instruction *ins)
2779 glsl_src_param_t src0_param;
2782 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2783 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2785 switch (write_mask) {
2786 case WINED3DSP_WRITEMASK_0:
2787 shader_addline(ins->ctx->buffer, "cos(%s));\n", src0_param.param_str);
2790 case WINED3DSP_WRITEMASK_1:
2791 shader_addline(ins->ctx->buffer, "sin(%s));\n", src0_param.param_str);
2794 case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
2795 shader_addline(ins->ctx->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
2799 ERR("Write mask should be .x, .y or .xy\n");
2804 /* sgn in vs_2_0 has 2 extra parameters(registers for temporary storage) which we don't use
2805 * here. But those extra parameters require a dedicated function for sgn, since map2gl would
2806 * generate invalid code
2808 static void shader_glsl_sgn(const struct wined3d_shader_instruction *ins)
2810 glsl_src_param_t src0_param;
2813 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
2814 shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src0_param);
2816 shader_addline(ins->ctx->buffer, "sign(%s));\n", src0_param.param_str);
2819 /** Process the WINED3DSIO_LOOP instruction in GLSL:
2820 * Start a for() loop where src1.y is the initial value of aL,
2821 * increment aL by src1.z for a total of src1.x iterations.
2822 * Need to use a temporary variable for this operation.
2824 /* FIXME: I don't think nested loops will work correctly this way. */
2825 static void shader_glsl_loop(const struct wined3d_shader_instruction *ins)
2827 glsl_src_param_t src1_param;
2828 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2829 const DWORD *control_values = NULL;
2830 const local_constant *constant;
2832 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2834 /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2835 * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2836 * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2839 if (ins->src[1].reg.type == WINED3DSPR_CONSTINT)
2841 LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
2842 if (constant->idx == ins->src[1].reg.idx)
2844 control_values = constant->value;
2852 struct wined3d_shader_loop_control loop_control;
2853 loop_control.count = control_values[0];
2854 loop_control.start = control_values[1];
2855 loop_control.step = (int)control_values[2];
2857 if (loop_control.step > 0)
2859 shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u < (%u * %d + %u); aL%u += %d) {\n",
2860 shader->baseShader.cur_loop_depth, loop_control.start,
2861 shader->baseShader.cur_loop_depth, loop_control.count, loop_control.step, loop_control.start,
2862 shader->baseShader.cur_loop_depth, loop_control.step);
2864 else if (loop_control.step < 0)
2866 shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u > (%u * %d + %u); aL%u += %d) {\n",
2867 shader->baseShader.cur_loop_depth, loop_control.start,
2868 shader->baseShader.cur_loop_depth, loop_control.count, loop_control.step, loop_control.start,
2869 shader->baseShader.cur_loop_depth, loop_control.step);
2873 shader_addline(ins->ctx->buffer, "for (aL%u = %u, tmpInt%u = 0; tmpInt%u < %u; tmpInt%u++) {\n",
2874 shader->baseShader.cur_loop_depth, loop_control.start, shader->baseShader.cur_loop_depth,
2875 shader->baseShader.cur_loop_depth, loop_control.count,
2876 shader->baseShader.cur_loop_depth);
2879 shader_addline(ins->ctx->buffer,
2880 "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2881 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno,
2882 src1_param.reg_name, shader->baseShader.cur_loop_depth, src1_param.reg_name,
2883 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno, src1_param.reg_name);
2886 shader->baseShader.cur_loop_depth++;
2887 shader->baseShader.cur_loop_regno++;
2890 static void shader_glsl_end(const struct wined3d_shader_instruction *ins)
2892 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2894 shader_addline(ins->ctx->buffer, "}\n");
2896 if (ins->handler_idx == WINED3DSIH_ENDLOOP)
2898 shader->baseShader.cur_loop_depth--;
2899 shader->baseShader.cur_loop_regno--;
2902 if (ins->handler_idx == WINED3DSIH_ENDREP)
2904 shader->baseShader.cur_loop_depth--;
2908 static void shader_glsl_rep(const struct wined3d_shader_instruction *ins)
2910 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2911 glsl_src_param_t src0_param;
2912 const DWORD *control_values = NULL;
2913 const local_constant *constant;
2915 /* Try to hardcode local values to help the GLSL compiler to unroll and optimize the loop */
2916 if (ins->src[0].reg.type == WINED3DSPR_CONSTINT)
2918 LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry)
2920 if (constant->idx == ins->src[0].reg.idx)
2922 control_values = constant->value;
2928 if(control_values) {
2929 shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %d; tmpInt%d++) {\n",
2930 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2931 control_values[0], shader->baseShader.cur_loop_depth);
2933 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2934 shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2935 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2936 src0_param.param_str, shader->baseShader.cur_loop_depth);
2938 shader->baseShader.cur_loop_depth++;
2941 static void shader_glsl_if(const struct wined3d_shader_instruction *ins)
2943 glsl_src_param_t src0_param;
2945 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2946 shader_addline(ins->ctx->buffer, "if (%s) {\n", src0_param.param_str);
2949 static void shader_glsl_ifc(const struct wined3d_shader_instruction *ins)
2951 glsl_src_param_t src0_param;
2952 glsl_src_param_t src1_param;
2954 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2955 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2957 shader_addline(ins->ctx->buffer, "if (%s %s %s) {\n",
2958 src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2961 static void shader_glsl_else(const struct wined3d_shader_instruction *ins)
2963 shader_addline(ins->ctx->buffer, "} else {\n");
2966 static void shader_glsl_break(const struct wined3d_shader_instruction *ins)
2968 shader_addline(ins->ctx->buffer, "break;\n");
2971 /* FIXME: According to MSDN the compare is done per component. */
2972 static void shader_glsl_breakc(const struct wined3d_shader_instruction *ins)
2974 glsl_src_param_t src0_param;
2975 glsl_src_param_t src1_param;
2977 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2978 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2980 shader_addline(ins->ctx->buffer, "if (%s %s %s) break;\n",
2981 src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2984 static void shader_glsl_label(const struct wined3d_shader_instruction *ins)
2986 shader_addline(ins->ctx->buffer, "}\n");
2987 shader_addline(ins->ctx->buffer, "void subroutine%u () {\n", ins->src[0].reg.idx);
2990 static void shader_glsl_call(const struct wined3d_shader_instruction *ins)
2992 shader_addline(ins->ctx->buffer, "subroutine%u();\n", ins->src[0].reg.idx);
2995 static void shader_glsl_callnz(const struct wined3d_shader_instruction *ins)
2997 glsl_src_param_t src1_param;
2999 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
3000 shader_addline(ins->ctx->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, ins->src[0].reg.idx);
3003 static void shader_glsl_ret(const struct wined3d_shader_instruction *ins)
3005 /* No-op. The closing } is written when a new function is started, and at the end of the shader. This
3006 * function only suppresses the unhandled instruction warning
3010 /*********************************************
3011 * Pixel Shader Specific Code begins here
3012 ********************************************/
3013 static void shader_glsl_tex(const struct wined3d_shader_instruction *ins)
3015 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3016 IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
3017 DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
3018 ins->ctx->reg_maps->shader_version.minor);
3019 glsl_sample_function_t sample_function;
3020 IWineD3DBaseTextureImpl *texture;
3021 DWORD sample_flags = 0;
3023 DWORD mask = 0, swizzle;
3025 /* 1.0-1.4: Use destination register as sampler source.
3026 * 2.0+: Use provided sampler source. */
3027 if (shader_version < WINED3D_SHADER_VERSION(2,0)) sampler_idx = ins->dst[0].reg.idx;
3028 else sampler_idx = ins->src[1].reg.idx;
3029 texture = deviceImpl->stateBlock->state.textures[sampler_idx];
3031 if (shader_version < WINED3D_SHADER_VERSION(1,4))
3033 DWORD flags = deviceImpl->stateBlock->state.texture_states[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
3034 WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3036 /* Projected cube textures don't make a lot of sense, the resulting coordinates stay the same. */
3037 if (flags & WINED3DTTFF_PROJECTED && sampler_type != WINED3DSTT_CUBE) {
3038 sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3039 switch (flags & ~WINED3DTTFF_PROJECTED) {
3040 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3041 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
3042 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
3043 case WINED3DTTFF_COUNT4:
3044 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
3048 else if (shader_version < WINED3D_SHADER_VERSION(2,0))
3050 DWORD src_mod = ins->src[0].modifiers;
3052 if (src_mod == WINED3DSPSM_DZ) {
3053 sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3054 mask = WINED3DSP_WRITEMASK_2;
3055 } else if (src_mod == WINED3DSPSM_DW) {
3056 sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3057 mask = WINED3DSP_WRITEMASK_3;
3060 if (ins->flags & WINED3DSI_TEXLD_PROJECT)
3062 /* ps 2.0 texldp instruction always divides by the fourth component. */
3063 sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3064 mask = WINED3DSP_WRITEMASK_3;
3068 if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
3069 sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3071 shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3072 mask |= sample_function.coord_mask;
3074 if (shader_version < WINED3D_SHADER_VERSION(2,0)) swizzle = WINED3DSP_NOSWIZZLE;
3075 else swizzle = ins->src[1].swizzle;
3077 /* 1.0-1.3: Use destination register as coordinate source.
3078 1.4+: Use provided coordinate source register. */
3079 if (shader_version < WINED3D_SHADER_VERSION(1,4))
3082 shader_glsl_write_mask_to_str(mask, coord_mask);
3083 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3084 "T%u%s", sampler_idx, coord_mask);
3086 glsl_src_param_t coord_param;
3087 shader_glsl_add_src_param(ins, &ins->src[0], mask, &coord_param);
3088 if (ins->flags & WINED3DSI_TEXLD_BIAS)
3090 glsl_src_param_t bias;
3091 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &bias);
3092 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, bias.param_str,
3093 "%s", coord_param.param_str);
3095 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3096 "%s", coord_param.param_str);
3101 static void shader_glsl_texldd(const struct wined3d_shader_instruction *ins)
3103 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3104 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
3105 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3106 glsl_sample_function_t sample_function;
3107 glsl_src_param_t coord_param, dx_param, dy_param;
3108 DWORD sample_flags = WINED3D_GLSL_SAMPLE_GRAD;
3110 DWORD swizzle = ins->src[1].swizzle;
3111 IWineD3DBaseTextureImpl *texture;
3113 if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4])
3115 FIXME("texldd used, but not supported by hardware. Falling back to regular tex\n");
3116 shader_glsl_tex(ins);
3120 sampler_idx = ins->src[1].reg.idx;
3121 texture = deviceImpl->stateBlock->state.textures[sampler_idx];
3122 if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
3123 sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3125 shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3126 shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3127 shader_glsl_add_src_param(ins, &ins->src[2], sample_function.coord_mask, &dx_param);
3128 shader_glsl_add_src_param(ins, &ins->src[3], sample_function.coord_mask, &dy_param);
3130 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, dx_param.param_str, dy_param.param_str, NULL,
3131 "%s", coord_param.param_str);
3134 static void shader_glsl_texldl(const struct wined3d_shader_instruction *ins)
3136 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3137 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
3138 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3139 glsl_sample_function_t sample_function;
3140 glsl_src_param_t coord_param, lod_param;
3141 DWORD sample_flags = WINED3D_GLSL_SAMPLE_LOD;
3143 DWORD swizzle = ins->src[1].swizzle;
3144 IWineD3DBaseTextureImpl *texture;
3146 sampler_idx = ins->src[1].reg.idx;
3147 texture = deviceImpl->stateBlock->state.textures[sampler_idx];
3148 if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
3149 sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3151 shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3152 shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3154 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &lod_param);
3156 if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4]
3157 && shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
3159 /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
3160 * However, they seem to work just fine in fragment shaders as well. */
3161 WARN("Using %s in fragment shader.\n", sample_function.name);
3163 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, lod_param.param_str,
3164 "%s", coord_param.param_str);
3167 static void shader_glsl_texcoord(const struct wined3d_shader_instruction *ins)
3169 /* FIXME: Make this work for more than just 2D textures */
3170 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3171 DWORD write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3173 if (!(ins->ctx->reg_maps->shader_version.major == 1 && ins->ctx->reg_maps->shader_version.minor == 4))
3177 shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3178 shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n",
3179 ins->dst[0].reg.idx, dst_mask);
3181 DWORD reg = ins->src[0].reg.idx;
3182 DWORD src_mod = ins->src[0].modifiers;
3183 char dst_swizzle[6];
3185 shader_glsl_get_swizzle(&ins->src[0], FALSE, write_mask, dst_swizzle);
3187 if (src_mod == WINED3DSPSM_DZ) {
3188 glsl_src_param_t div_param;
3189 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3190 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &div_param);
3192 if (mask_size > 1) {
3193 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3195 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3197 } else if (src_mod == WINED3DSPSM_DW) {
3198 glsl_src_param_t div_param;
3199 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3200 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &div_param);
3202 if (mask_size > 1) {
3203 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3205 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3208 shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
3213 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
3214 * Take a 3-component dot product of the TexCoord[dstreg] and src,
3215 * then perform a 1D texture lookup from stage dstregnum, place into dst. */
3216 static void shader_glsl_texdp3tex(const struct wined3d_shader_instruction *ins)
3218 glsl_src_param_t src0_param;
3219 glsl_sample_function_t sample_function;
3220 DWORD sampler_idx = ins->dst[0].reg.idx;
3221 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3224 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3226 /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
3227 * scalar, and projected sampling would require 4.
3229 * It is a dependent read - not valid with conditional NP2 textures
3231 shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3232 mask_size = shader_glsl_get_write_mask_size(sample_function.coord_mask);
3237 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3238 "dot(gl_TexCoord[%u].xyz, %s)", sampler_idx, src0_param.param_str);
3242 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3243 "vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0)", sampler_idx, src0_param.param_str);
3247 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3248 "vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0)", sampler_idx, src0_param.param_str);
3252 FIXME("Unexpected mask size %u\n", mask_size);
3257 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
3258 * Take a 3-component dot product of the TexCoord[dstreg] and src. */
3259 static void shader_glsl_texdp3(const struct wined3d_shader_instruction *ins)
3261 glsl_src_param_t src0_param;
3262 DWORD dstreg = ins->dst[0].reg.idx;
3263 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3265 unsigned int mask_size;
3267 dst_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3268 mask_size = shader_glsl_get_write_mask_size(dst_mask);
3269 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3271 if (mask_size > 1) {
3272 shader_addline(ins->ctx->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
3274 shader_addline(ins->ctx->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
3278 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
3279 * Calculate the depth as dst.x / dst.y */
3280 static void shader_glsl_texdepth(const struct wined3d_shader_instruction *ins)
3282 glsl_dst_param_t dst_param;
3284 shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3286 /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
3287 * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
3288 * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
3289 * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
3292 shader_addline(ins->ctx->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n",
3293 dst_param.reg_name, dst_param.reg_name);
3296 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
3297 * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
3298 * Calculate tmp0.y = TexCoord[dstreg] . src.xyz; (tmp0.x has already been calculated)
3299 * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
3301 static void shader_glsl_texm3x2depth(const struct wined3d_shader_instruction *ins)
3303 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3304 DWORD dstreg = ins->dst[0].reg.idx;
3305 glsl_src_param_t src0_param;
3307 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3309 shader_addline(ins->ctx->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
3310 shader_addline(ins->ctx->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
3313 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
3314 * Calculate the 1st of a 2-row matrix multiplication. */
3315 static void shader_glsl_texm3x2pad(const struct wined3d_shader_instruction *ins)
3317 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3318 DWORD reg = ins->dst[0].reg.idx;
3319 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3320 glsl_src_param_t src0_param;
3322 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3323 shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3326 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
3327 * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
3328 static void shader_glsl_texm3x3pad(const struct wined3d_shader_instruction *ins)
3330 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3331 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3332 DWORD reg = ins->dst[0].reg.idx;
3333 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3334 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3335 glsl_src_param_t src0_param;
3337 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3338 shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + current_state->current_row, reg, src0_param.param_str);
3339 current_state->texcoord_w[current_state->current_row++] = reg;
3342 static void shader_glsl_texm3x2tex(const struct wined3d_shader_instruction *ins)
3344 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3345 DWORD reg = ins->dst[0].reg.idx;
3346 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3347 glsl_src_param_t src0_param;
3348 glsl_sample_function_t sample_function;
3350 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3351 shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3353 shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3355 /* Sample the texture using the calculated coordinates */
3356 shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xy");
3359 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
3360 * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
3361 static void shader_glsl_texm3x3tex(const struct wined3d_shader_instruction *ins)
3363 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3364 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3365 SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
3366 glsl_src_param_t src0_param;
3367 DWORD reg = ins->dst[0].reg.idx;
3368 glsl_sample_function_t sample_function;
3370 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3371 shader_addline(ins->ctx->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3373 /* Dependent read, not valid with conditional NP2 */
3374 shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3376 /* Sample the texture using the calculated coordinates */
3377 shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3379 current_state->current_row = 0;
3382 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
3383 * Perform the 3rd row of a 3x3 matrix multiply */
3384 static void shader_glsl_texm3x3(const struct wined3d_shader_instruction *ins)
3386 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3387 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3388 SHADER_PARSE_STATE *current_state = &shader->baseShader.parse_state;
3389 glsl_src_param_t src0_param;
3391 DWORD reg = ins->dst[0].reg.idx;
3393 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3395 shader_glsl_append_dst(ins->ctx->buffer, ins);
3396 shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3397 shader_addline(ins->ctx->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
3399 current_state->current_row = 0;
3402 /* Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
3403 * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3404 static void shader_glsl_texm3x3spec(const struct wined3d_shader_instruction *ins)
3406 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3407 DWORD reg = ins->dst[0].reg.idx;
3408 glsl_src_param_t src0_param;
3409 glsl_src_param_t src1_param;
3410 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3411 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3412 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3413 glsl_sample_function_t sample_function;
3415 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3416 shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
3418 /* Perform the last matrix multiply operation */
3419 shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3420 /* Reflection calculation */
3421 shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
3423 /* Dependent read, not valid with conditional NP2 */
3424 shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3426 /* Sample the texture */
3427 shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3429 current_state->current_row = 0;
3432 /* Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
3433 * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3434 static void shader_glsl_texm3x3vspec(const struct wined3d_shader_instruction *ins)
3436 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3437 DWORD reg = ins->dst[0].reg.idx;
3438 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3439 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
3440 glsl_src_param_t src0_param;
3441 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3442 glsl_sample_function_t sample_function;
3444 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3446 /* Perform the last matrix multiply operation */
3447 shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
3449 /* Construct the eye-ray vector from w coordinates */
3450 shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
3451 current_state->texcoord_w[0], current_state->texcoord_w[1], reg);
3452 shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
3454 /* Dependent read, not valid with conditional NP2 */
3455 shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3457 /* Sample the texture using the calculated coordinates */
3458 shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3460 current_state->current_row = 0;
3463 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
3464 * Apply a fake bump map transform.
3465 * texbem is pshader <= 1.3 only, this saves a few version checks
3467 static void shader_glsl_texbem(const struct wined3d_shader_instruction *ins)
3469 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3470 IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *)shader->baseShader.device;
3471 glsl_sample_function_t sample_function;
3472 glsl_src_param_t coord_param;
3478 sampler_idx = ins->dst[0].reg.idx;
3479 flags = deviceImpl->stateBlock->state.texture_states[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
3481 /* Dependent read, not valid with conditional NP2 */
3482 shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3483 mask = sample_function.coord_mask;
3485 shader_glsl_write_mask_to_str(mask, coord_mask);
3487 /* with projective textures, texbem only divides the static texture coord, not the displacement,
3488 * so we can't let the GL handle this.
3490 if (flags & WINED3DTTFF_PROJECTED) {
3492 char coord_div_mask[3];
3493 switch (flags & ~WINED3DTTFF_PROJECTED) {
3494 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3495 case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
3496 case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
3497 case WINED3DTTFF_COUNT4:
3498 case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
3500 shader_glsl_write_mask_to_str(div_mask, coord_div_mask);
3501 shader_addline(ins->ctx->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
3504 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &coord_param);
3506 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3507 "T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s", sampler_idx, coord_mask, sampler_idx,
3508 coord_param.param_str, coord_mask);
3510 if (ins->handler_idx == WINED3DSIH_TEXBEML)
3512 glsl_src_param_t luminance_param;
3513 glsl_dst_param_t dst_param;
3515 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &luminance_param);
3516 shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3518 shader_addline(ins->ctx->buffer, "%s%s *= (%s * luminancescale%d + luminanceoffset%d);\n",
3519 dst_param.reg_name, dst_param.mask_str,
3520 luminance_param.param_str, sampler_idx, sampler_idx);
3524 static void shader_glsl_bem(const struct wined3d_shader_instruction *ins)
3526 glsl_src_param_t src0_param, src1_param;
3527 DWORD sampler_idx = ins->dst[0].reg.idx;
3529 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3530 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3532 shader_glsl_append_dst(ins->ctx->buffer, ins);
3533 shader_addline(ins->ctx->buffer, "%s + bumpenvmat%d * %s);\n",
3534 src0_param.param_str, sampler_idx, src1_param.param_str);
3537 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
3538 * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
3539 static void shader_glsl_texreg2ar(const struct wined3d_shader_instruction *ins)
3541 glsl_src_param_t src0_param;
3542 DWORD sampler_idx = ins->dst[0].reg.idx;
3543 glsl_sample_function_t sample_function;
3545 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3547 shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3548 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3549 "%s.wx", src0_param.reg_name);
3552 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
3553 * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
3554 static void shader_glsl_texreg2gb(const struct wined3d_shader_instruction *ins)
3556 glsl_src_param_t src0_param;
3557 DWORD sampler_idx = ins->dst[0].reg.idx;
3558 glsl_sample_function_t sample_function;
3560 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3562 shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3563 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3564 "%s.yz", src0_param.reg_name);
3567 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
3568 * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
3569 static void shader_glsl_texreg2rgb(const struct wined3d_shader_instruction *ins)
3571 glsl_src_param_t src0_param;
3572 DWORD sampler_idx = ins->dst[0].reg.idx;
3573 glsl_sample_function_t sample_function;
3575 /* Dependent read, not valid with conditional NP2 */
3576 shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3577 shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &src0_param);
3579 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3580 "%s", src0_param.param_str);
3583 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
3584 * If any of the first 3 components are < 0, discard this pixel */
3585 static void shader_glsl_texkill(const struct wined3d_shader_instruction *ins)
3587 glsl_dst_param_t dst_param;
3589 /* The argument is a destination parameter, and no writemasks are allowed */
3590 shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3591 if (ins->ctx->reg_maps->shader_version.major >= 2)
3593 /* 2.0 shaders compare all 4 components in texkill */
3594 shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
3596 /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
3597 * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
3598 * 4 components are defined, only the first 3 are used
3600 shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
3604 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
3605 * dst = dot2(src0, src1) + src2 */
3606 static void shader_glsl_dp2add(const struct wined3d_shader_instruction *ins)
3608 glsl_src_param_t src0_param;
3609 glsl_src_param_t src1_param;
3610 glsl_src_param_t src2_param;
3612 unsigned int mask_size;
3614 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3615 mask_size = shader_glsl_get_write_mask_size(write_mask);
3617 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3618 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3619 shader_glsl_add_src_param(ins, &ins->src[2], WINED3DSP_WRITEMASK_0, &src2_param);
3621 if (mask_size > 1) {
3622 shader_addline(ins->ctx->buffer, "vec%d(dot(%s, %s) + %s));\n",
3623 mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
3625 shader_addline(ins->ctx->buffer, "dot(%s, %s) + %s);\n",
3626 src0_param.param_str, src1_param.param_str, src2_param.param_str);
3630 static void shader_glsl_input_pack(IWineD3DPixelShader *iface, struct wined3d_shader_buffer *buffer,
3631 const struct wined3d_shader_signature_element *input_signature, const struct shader_reg_maps *reg_maps,
3632 enum vertexprocessing_mode vertexprocessing)
3635 IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
3636 WORD map = reg_maps->input_registers;
3638 for (i = 0; map; map >>= 1, ++i)
3640 const char *semantic_name;
3645 if (!(map & 1)) continue;
3647 semantic_name = input_signature[i].semantic_name;
3648 semantic_idx = input_signature[i].semantic_idx;
3649 shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3651 if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3653 if (semantic_idx < 8 && vertexprocessing == pretransformed)
3654 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
3655 This->input_reg_map[i], reg_mask, semantic_idx, reg_mask);
3657 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3658 This->input_reg_map[i], reg_mask, reg_mask);
3660 else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3663 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
3664 This->input_reg_map[i], reg_mask, reg_mask);
3665 else if (semantic_idx == 1)
3666 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
3667 This->input_reg_map[i], reg_mask, reg_mask);
3669 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3670 This->input_reg_map[i], reg_mask, reg_mask);
3674 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3675 This->input_reg_map[i], reg_mask, reg_mask);
3680 /*********************************************
3681 * Vertex Shader Specific Code begins here
3682 ********************************************/
3684 static void add_glsl_program_entry(struct shader_glsl_priv *priv, struct glsl_shader_prog_link *entry) {
3685 glsl_program_key_t key;
3687 key.vshader = entry->vshader;
3688 key.pshader = entry->pshader;
3689 key.vs_args = entry->vs_args;
3690 key.ps_args = entry->ps_args;
3692 if (wine_rb_put(&priv->program_lookup, &key, &entry->program_lookup_entry) == -1)
3694 ERR("Failed to insert program entry.\n");
3698 static struct glsl_shader_prog_link *get_glsl_program_entry(struct shader_glsl_priv *priv,
3699 IWineD3DVertexShader *vshader, IWineD3DPixelShader *pshader, struct vs_compile_args *vs_args,
3700 struct ps_compile_args *ps_args) {
3701 struct wine_rb_entry *entry;
3702 glsl_program_key_t key;
3704 key.vshader = vshader;
3705 key.pshader = pshader;
3706 key.vs_args = *vs_args;
3707 key.ps_args = *ps_args;
3709 entry = wine_rb_get(&priv->program_lookup, &key);
3710 return entry ? WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry) : NULL;
3713 /* GL locking is done by the caller */
3714 static void delete_glsl_program_entry(struct shader_glsl_priv *priv, const struct wined3d_gl_info *gl_info,
3715 struct glsl_shader_prog_link *entry)
3717 glsl_program_key_t key;
3719 key.vshader = entry->vshader;
3720 key.pshader = entry->pshader;
3721 key.vs_args = entry->vs_args;
3722 key.ps_args = entry->ps_args;
3723 wine_rb_remove(&priv->program_lookup, &key);
3725 GL_EXTCALL(glDeleteObjectARB(entry->programId));
3726 if (entry->vshader) list_remove(&entry->vshader_entry);
3727 if (entry->pshader) list_remove(&entry->pshader_entry);
3728 HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
3729 HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
3730 HeapFree(GetProcessHeap(), 0, entry);
3733 static void handle_ps3_input(struct wined3d_shader_buffer *buffer, const struct wined3d_gl_info *gl_info, const DWORD *map,
3734 const struct wined3d_shader_signature_element *input_signature, const struct shader_reg_maps *reg_maps_in,
3735 const struct wined3d_shader_signature_element *output_signature, const struct shader_reg_maps *reg_maps_out)
3738 const char *semantic_name_in;
3739 UINT semantic_idx_in;
3742 unsigned int in_count = vec4_varyings(3, gl_info);
3744 char destination[50];
3745 WORD input_map, output_map;
3747 set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
3749 input_map = reg_maps_in->input_registers;
3750 for (i = 0; input_map; input_map >>= 1, ++i)
3752 if (!(input_map & 1)) continue;
3755 /* Declared, but not read register */
3756 if (in_idx == ~0U) continue;
3757 if (in_idx >= (in_count + 2))
3759 FIXME("More input varyings declared than supported, expect issues.\n");
3763 if (in_idx == in_count) {
3764 sprintf(destination, "gl_FrontColor");
3765 } else if (in_idx == in_count + 1) {
3766 sprintf(destination, "gl_FrontSecondaryColor");
3768 sprintf(destination, "IN[%u]", in_idx);
3771 semantic_name_in = input_signature[i].semantic_name;
3772 semantic_idx_in = input_signature[i].semantic_idx;
3775 output_map = reg_maps_out->output_registers;
3776 for (j = 0; output_map; output_map >>= 1, ++j)
3780 if (!(output_map & 1)
3781 || semantic_idx_in != output_signature[j].semantic_idx
3782 || strcmp(semantic_name_in, output_signature[j].semantic_name)
3783 || !(mask = input_signature[i].mask & output_signature[j].mask))
3787 shader_glsl_write_mask_to_str(mask, reg_mask);
3789 shader_addline(buffer, "%s%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3790 destination, reg_mask, j, reg_mask);
3794 for (i = 0; i < in_count + 2; ++i)
3798 if (!set[i] || set[i] == WINED3DSP_WRITEMASK_ALL)
3801 if (set[i] == ~0U) set[i] = 0;
3804 if (!(set[i] & WINED3DSP_WRITEMASK_0)) reg_mask[size++] = 'x';
3805 if (!(set[i] & WINED3DSP_WRITEMASK_1)) reg_mask[size++] = 'y';
3806 if (!(set[i] & WINED3DSP_WRITEMASK_2)) reg_mask[size++] = 'z';
3807 if (!(set[i] & WINED3DSP_WRITEMASK_3)) reg_mask[size++] = 'w';
3808 reg_mask[size] = '\0';
3810 if (i == in_count) sprintf(destination, "gl_FrontColor");
3811 else if (i == in_count + 1) sprintf(destination, "gl_FrontSecondaryColor");
3812 else sprintf(destination, "IN[%u]", i);
3814 if (size == 1) shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
3815 else shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
3818 HeapFree(GetProcessHeap(), 0, set);
3821 /* GL locking is done by the caller */
3822 static GLhandleARB generate_param_reorder_function(struct wined3d_shader_buffer *buffer,
3823 IWineD3DVertexShader *vertexshader, IWineD3DPixelShader *pixelshader, const struct wined3d_gl_info *gl_info)
3825 GLhandleARB ret = 0;
3826 IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) vertexshader;
3827 IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pixelshader;
3828 DWORD ps_major = ps ? ps->baseShader.reg_maps.shader_version.major : 0;
3830 const char *semantic_name;
3833 const struct wined3d_shader_signature_element *output_signature = vs->baseShader.output_signature;
3834 WORD map = vs->baseShader.reg_maps.output_registers;
3836 shader_buffer_clear(buffer);
3838 shader_addline(buffer, "#version 120\n");
3839 shader_addline(buffer, "const float FLT_MAX = 1e38;\n");
3843 shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3845 for (i = 0; map; map >>= 1, ++i)
3849 if (!(map & 1)) continue;
3851 semantic_name = output_signature[i].semantic_name;
3852 semantic_idx = output_signature[i].semantic_idx;
3853 write_mask = output_signature[i].mask;
3854 shader_glsl_write_mask_to_str(write_mask, reg_mask);
3856 if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3859 shader_addline(buffer, "gl_FrontColor%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3860 reg_mask, i, reg_mask);
3861 else if (semantic_idx == 1)
3862 shader_addline(buffer, "gl_FrontSecondaryColor%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3863 reg_mask, i, reg_mask);
3865 else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3867 shader_addline(buffer, "gl_Position%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3868 reg_mask, i, reg_mask);
3870 else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3872 if (semantic_idx < 8)
3874 if (!(gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) || ps_major > 0)
3875 write_mask |= WINED3DSP_WRITEMASK_3;
3877 shader_addline(buffer, "gl_TexCoord[%u]%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3878 semantic_idx, reg_mask, i, reg_mask);
3879 if (!(write_mask & WINED3DSP_WRITEMASK_3))
3880 shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", semantic_idx);
3883 else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3885 shader_addline(buffer, "gl_PointSize = clamp(OUT[%u].%c, -FLT_MAX, FLT_MAX);\n", i, reg_mask[1]);
3887 else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3889 shader_addline(buffer, "gl_FogFragCoord = clamp(OUT[%u].%c, -FLT_MAX, FLT_MAX);\n", i, reg_mask[1]);
3892 shader_addline(buffer, "}\n");
3897 /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3898 shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(3, gl_info));
3899 shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3901 /* First, sort out position and point size. Those are not passed to the pixel shader */
3902 for (i = 0; map; map >>= 1, ++i)
3904 if (!(map & 1)) continue;
3906 semantic_name = output_signature[i].semantic_name;
3907 shader_glsl_write_mask_to_str(output_signature[i].mask, reg_mask);
3909 if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3911 shader_addline(buffer, "gl_Position%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3912 reg_mask, i, reg_mask);
3914 else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3916 shader_addline(buffer, "gl_PointSize = clamp(OUT[%u].%c, -FLT_MAX, FLT_MAX);\n", i, reg_mask[1]);
3920 /* Then, fix the pixel shader input */
3921 handle_ps3_input(buffer, gl_info, ps->input_reg_map, ps->baseShader.input_signature,
3922 &ps->baseShader.reg_maps, output_signature, &vs->baseShader.reg_maps);
3924 shader_addline(buffer, "}\n");
3927 ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3928 checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3929 shader_glsl_compile(gl_info, ret, buffer->buffer);
3934 /* GL locking is done by the caller */
3935 static void hardcode_local_constants(IWineD3DBaseShaderImpl *shader, const struct wined3d_gl_info *gl_info,
3936 GLhandleARB programId, char prefix)
3938 const local_constant *lconst;
3943 LIST_FOR_EACH_ENTRY(lconst, &shader->baseShader.constantsF, local_constant, entry) {
3944 value = (const float *)lconst->value;
3945 snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
3946 tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3947 GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
3949 checkGLcall("Hardcoding local constants");
3952 /* GL locking is done by the caller */
3953 static GLuint shader_glsl_generate_pshader(const struct wined3d_context *context,
3954 struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *This,
3955 const struct ps_compile_args *args, struct ps_np2fixup_info *np2fixup_info)
3957 const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3958 const struct wined3d_gl_info *gl_info = context->gl_info;
3959 CONST DWORD *function = This->baseShader.function;
3960 struct shader_glsl_ctx_priv priv_ctx;
3962 /* Create the hw GLSL shader object and assign it as the shader->prgId */
3963 GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3965 memset(&priv_ctx, 0, sizeof(priv_ctx));
3966 priv_ctx.cur_ps_args = args;
3967 priv_ctx.cur_np2fixup_info = np2fixup_info;
3969 shader_addline(buffer, "#version 120\n");
3971 if (gl_info->supported[ARB_SHADER_TEXTURE_LOD] && reg_maps->usestexldd)
3973 shader_addline(buffer, "#extension GL_ARB_shader_texture_lod : enable\n");
3975 if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
3977 /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
3978 * drivers write a warning if we don't do so
3980 shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
3982 if (gl_info->supported[EXT_GPU_SHADER4])
3984 shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
3987 /* Base Declarations */
3988 shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShader *)This, reg_maps, &priv_ctx);
3990 /* Pack 3.0 inputs */
3991 if (reg_maps->shader_version.major >= 3 && args->vp_mode != vertexshader)
3993 shader_glsl_input_pack((IWineD3DPixelShader *) This, buffer,
3994 This->baseShader.input_signature, reg_maps, args->vp_mode);
3997 /* Base Shader Body */
3998 shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
4000 /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
4001 if (reg_maps->shader_version.major < 2)
4003 /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
4004 shader_addline(buffer, "gl_FragData[0] = R0;\n");
4007 if (args->srgb_correction)
4009 shader_addline(buffer, "tmp0.xyz = pow(gl_FragData[0].xyz, vec3(srgb_const0.x));\n");
4010 shader_addline(buffer, "tmp0.xyz = tmp0.xyz * vec3(srgb_const0.y) - vec3(srgb_const0.z);\n");
4011 shader_addline(buffer, "tmp1.xyz = gl_FragData[0].xyz * vec3(srgb_const0.w);\n");
4012 shader_addline(buffer, "bvec3 srgb_compare = lessThan(gl_FragData[0].xyz, vec3(srgb_const1.x));\n");
4013 shader_addline(buffer, "gl_FragData[0].xyz = mix(tmp0.xyz, tmp1.xyz, vec3(srgb_compare));\n");
4014 shader_addline(buffer, "gl_FragData[0] = clamp(gl_FragData[0], 0.0, 1.0);\n");
4016 /* Pixel shader < 3.0 do not replace the fog stage.
4017 * This implements linear fog computation and blending.
4018 * TODO: non linear fog
4019 * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
4020 * -1/(e-s) and e/(e-s) respectively.
4022 if (reg_maps->shader_version.major < 3)
4025 case FOG_OFF: break;
4027 shader_addline(buffer, "float fogstart = -1.0 / (gl_Fog.end - gl_Fog.start);\n");
4028 shader_addline(buffer, "float fogend = gl_Fog.end * -fogstart;\n");
4029 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * fogstart + fogend, 0.0, 1.0);\n");
4030 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4033 /* Fog = e^(-gl_Fog.density * gl_FogFragCoord) */
4034 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_FogFragCoord);\n");
4035 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4036 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4039 /* Fog = e^(-(gl_Fog.density * gl_FogFragCoord)^2) */
4040 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_Fog.density * gl_FogFragCoord * gl_FogFragCoord);\n");
4041 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4042 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4047 shader_addline(buffer, "}\n");
4049 TRACE("Compiling shader object %u\n", shader_obj);
4050 shader_glsl_compile(gl_info, shader_obj, buffer->buffer);
4052 /* Store the shader object */
4056 /* GL locking is done by the caller */
4057 static GLuint shader_glsl_generate_vshader(const struct wined3d_context *context,
4058 struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *This,
4059 const struct vs_compile_args *args)
4061 const struct shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
4062 const struct wined3d_gl_info *gl_info = context->gl_info;
4063 CONST DWORD *function = This->baseShader.function;
4064 struct shader_glsl_ctx_priv priv_ctx;
4066 /* Create the hw GLSL shader program and assign it as the shader->prgId */
4067 GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4069 shader_addline(buffer, "#version 120\n");
4071 if (gl_info->supported[EXT_GPU_SHADER4])
4073 shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
4076 memset(&priv_ctx, 0, sizeof(priv_ctx));
4077 priv_ctx.cur_vs_args = args;
4079 /* Base Declarations */
4080 shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShader *)This, reg_maps, &priv_ctx);
4082 /* Base Shader Body */
4083 shader_generate_main((IWineD3DBaseShader*)This, buffer, reg_maps, function, &priv_ctx);
4085 /* Unpack outputs */
4086 shader_addline(buffer, "order_ps_input(OUT);\n");
4088 /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
4089 * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
4090 * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
4091 * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
4093 if(args->fog_src == VS_FOG_Z) {
4094 shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
4095 } else if (!reg_maps->fog) {
4096 shader_addline(buffer, "gl_FogFragCoord = 0.0;\n");
4099 /* Write the final position.
4101 * OpenGL coordinates specify the center of the pixel while d3d coords specify
4102 * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
4103 * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
4104 * contains 1.0 to allow a mad.
4106 shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
4107 shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
4108 if(args->clip_enabled) {
4109 shader_addline(buffer, "gl_ClipVertex = gl_Position;\n");
4112 /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
4114 * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
4115 * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
4116 * which is the same as z = z * 2 - w.
4118 shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
4120 shader_addline(buffer, "}\n");
4122 TRACE("Compiling shader object %u\n", shader_obj);
4123 shader_glsl_compile(gl_info, shader_obj, buffer->buffer);
4128 static GLhandleARB find_glsl_pshader(const struct wined3d_context *context,
4129 struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *shader,
4130 const struct ps_compile_args *args, const struct ps_np2fixup_info **np2fixup_info)
4132 struct wined3d_state *state = &((IWineD3DDeviceImpl *)shader->baseShader.device)->stateBlock->state;
4135 struct glsl_ps_compiled_shader *new_array;
4136 struct glsl_pshader_private *shader_data;
4137 struct ps_np2fixup_info *np2fixup = NULL;
4140 if (!shader->baseShader.backend_data)
4142 shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4143 if (!shader->baseShader.backend_data)
4145 ERR("Failed to allocate backend data.\n");
4149 shader_data = shader->baseShader.backend_data;
4151 /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4152 * so a linear search is more performant than a hashmap or a binary search
4153 * (cache coherency etc)
4155 for (i = 0; i < shader_data->num_gl_shaders; ++i)
4157 if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4159 if (args->np2_fixup) *np2fixup_info = &shader_data->gl_shaders[i].np2fixup;
4160 return shader_data->gl_shaders[i].prgId;
4164 TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4165 if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4166 if (shader_data->num_gl_shaders)
4168 new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4169 new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4170 new_size * sizeof(*shader_data->gl_shaders));
4172 new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4177 ERR("Out of memory\n");
4180 shader_data->gl_shaders = new_array;
4181 shader_data->shader_array_size = new_size;
4184 shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4186 memset(&shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup, 0, sizeof(struct ps_np2fixup_info));
4187 if (args->np2_fixup) np2fixup = &shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup;
4189 pixelshader_update_samplers(&shader->baseShader.reg_maps, (IWineD3DBaseTexture **)state->textures);
4191 shader_buffer_clear(buffer);
4192 ret = shader_glsl_generate_pshader(context, buffer, shader, args, np2fixup);
4193 shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4194 *np2fixup_info = np2fixup;
4199 static inline BOOL vs_args_equal(const struct vs_compile_args *stored, const struct vs_compile_args *new,
4200 const DWORD use_map) {
4201 if((stored->swizzle_map & use_map) != new->swizzle_map) return FALSE;
4202 if((stored->clip_enabled) != new->clip_enabled) return FALSE;
4203 return stored->fog_src == new->fog_src;
4206 static GLhandleARB find_glsl_vshader(const struct wined3d_context *context,
4207 struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *shader,
4208 const struct vs_compile_args *args)
4212 struct glsl_vs_compiled_shader *new_array;
4213 DWORD use_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.use_map;
4214 struct glsl_vshader_private *shader_data;
4217 if (!shader->baseShader.backend_data)
4219 shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4220 if (!shader->baseShader.backend_data)
4222 ERR("Failed to allocate backend data.\n");
4226 shader_data = shader->baseShader.backend_data;
4228 /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4229 * so a linear search is more performant than a hashmap or a binary search
4230 * (cache coherency etc)
4232 for(i = 0; i < shader_data->num_gl_shaders; i++) {
4233 if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map)) {
4234 return shader_data->gl_shaders[i].prgId;
4238 TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4240 if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4241 if (shader_data->num_gl_shaders)
4243 new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4244 new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4245 new_size * sizeof(*shader_data->gl_shaders));
4247 new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4252 ERR("Out of memory\n");
4255 shader_data->gl_shaders = new_array;
4256 shader_data->shader_array_size = new_size;
4259 shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4261 shader_buffer_clear(buffer);
4262 ret = shader_glsl_generate_vshader(context, buffer, shader, args);
4263 shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4268 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
4269 * It sets the programId on the current StateBlock (because it should be called
4270 * inside of the DrawPrimitive() part of the render loop).
4272 * If a program for the given combination does not exist, create one, and store
4273 * the program in the hash table. If it creates a program, it will link the
4274 * given objects, too.
4277 /* GL locking is done by the caller */
4278 static void set_glsl_shader_program(const struct wined3d_context *context,
4279 IWineD3DDeviceImpl *device, BOOL use_ps, BOOL use_vs)
4281 IWineD3DVertexShader *vshader = use_vs ? (IWineD3DVertexShader *)device->stateBlock->state.vertex_shader : NULL;
4282 IWineD3DPixelShader *pshader = use_ps ? (IWineD3DPixelShader *)device->stateBlock->state.pixel_shader : NULL;
4283 const struct wined3d_gl_info *gl_info = context->gl_info;
4284 struct shader_glsl_priv *priv = device->shader_priv;
4285 struct glsl_shader_prog_link *entry = NULL;
4286 GLhandleARB programId = 0;
4287 GLhandleARB reorder_shader_id = 0;
4290 struct ps_compile_args ps_compile_args;
4291 struct vs_compile_args vs_compile_args;
4293 if (vshader) find_vs_compile_args((IWineD3DVertexShaderImpl *)vshader, device->stateBlock, &vs_compile_args);
4294 if (pshader) find_ps_compile_args((IWineD3DPixelShaderImpl *)pshader, device->stateBlock, &ps_compile_args);
4296 entry = get_glsl_program_entry(priv, vshader, pshader, &vs_compile_args, &ps_compile_args);
4298 priv->glsl_program = entry;
4302 /* If we get to this point, then no matching program exists, so we create one */
4303 programId = GL_EXTCALL(glCreateProgramObjectARB());
4304 TRACE("Created new GLSL shader program %u\n", programId);
4306 /* Create the entry */
4307 entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
4308 entry->programId = programId;
4309 entry->vshader = vshader;
4310 entry->pshader = pshader;
4311 entry->vs_args = vs_compile_args;
4312 entry->ps_args = ps_compile_args;
4313 entry->constant_version = 0;
4314 entry->np2Fixup_info = NULL;
4315 /* Add the hash table entry */
4316 add_glsl_program_entry(priv, entry);
4318 /* Set the current program */
4319 priv->glsl_program = entry;
4321 /* Attach GLSL vshader */
4324 GLhandleARB vshader_id = find_glsl_vshader(context, &priv->shader_buffer,
4325 (IWineD3DVertexShaderImpl *)vshader, &vs_compile_args);
4326 WORD map = ((IWineD3DBaseShaderImpl *)vshader)->baseShader.reg_maps.input_registers;
4329 reorder_shader_id = generate_param_reorder_function(&priv->shader_buffer, vshader, pshader, gl_info);
4330 TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
4331 GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
4332 checkGLcall("glAttachObjectARB");
4333 /* Flag the reorder function for deletion, then it will be freed automatically when the program
4336 GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
4338 TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
4339 GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
4340 checkGLcall("glAttachObjectARB");
4342 /* Bind vertex attributes to a corresponding index number to match
4343 * the same index numbers as ARB_vertex_programs (makes loading
4344 * vertex attributes simpler). With this method, we can use the
4345 * exact same code to load the attributes later for both ARB and
4348 * We have to do this here because we need to know the Program ID
4349 * in order to make the bindings work, and it has to be done prior
4350 * to linking the GLSL program. */
4351 for (i = 0; map; map >>= 1, ++i)
4353 if (!(map & 1)) continue;
4355 snprintf(tmp_name, sizeof(tmp_name), "attrib%u", i);
4356 GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
4358 checkGLcall("glBindAttribLocationARB");
4360 list_add_head(&((IWineD3DBaseShaderImpl *)vshader)->baseShader.linked_programs, &entry->vshader_entry);
4363 /* Attach GLSL pshader */
4366 GLhandleARB pshader_id = find_glsl_pshader(context, &priv->shader_buffer,
4367 (IWineD3DPixelShaderImpl *)pshader, &ps_compile_args, &entry->np2Fixup_info);
4368 TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
4369 GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
4370 checkGLcall("glAttachObjectARB");
4372 list_add_head(&((IWineD3DBaseShaderImpl *)pshader)->baseShader.linked_programs, &entry->pshader_entry);
4375 /* Link the program */
4376 TRACE("Linking GLSL shader program %u\n", programId);
4377 GL_EXTCALL(glLinkProgramARB(programId));
4378 shader_glsl_validate_link(gl_info, programId);
4380 entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4381 sizeof(GLhandleARB) * gl_info->limits.glsl_vs_float_constants);
4382 for (i = 0; i < gl_info->limits.glsl_vs_float_constants; ++i)
4384 snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
4385 entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4387 for (i = 0; i < MAX_CONST_I; ++i)
4389 snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
4390 entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4392 entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4393 sizeof(GLhandleARB) * gl_info->limits.glsl_ps_float_constants);
4394 for (i = 0; i < gl_info->limits.glsl_ps_float_constants; ++i)
4396 snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
4397 entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4399 for (i = 0; i < MAX_CONST_I; ++i)
4401 snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
4402 entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4408 for(i = 0; i < MAX_TEXTURES; i++) {
4409 sprintf(name, "bumpenvmat%u", i);
4410 entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4411 sprintf(name, "luminancescale%u", i);
4412 entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4413 sprintf(name, "luminanceoffset%u", i);
4414 entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4417 if (ps_compile_args.np2_fixup) {
4418 if (entry->np2Fixup_info) {
4419 entry->np2Fixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "PsamplerNP2Fixup"));
4421 FIXME("NP2 texcoord fixup needed for this pixelshader, but no fixup uniform found.\n");
4426 entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
4427 entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
4428 checkGLcall("Find glsl program uniform locations");
4431 && ((IWineD3DPixelShaderImpl *)pshader)->baseShader.reg_maps.shader_version.major >= 3
4432 && ((IWineD3DPixelShaderImpl *)pshader)->declared_in_count > vec4_varyings(3, gl_info))
4434 TRACE("Shader %d needs vertex color clamping disabled\n", programId);
4435 entry->vertex_color_clamp = GL_FALSE;
4437 entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
4440 /* Set the shader to allow uniform loading on it */
4441 GL_EXTCALL(glUseProgramObjectARB(programId));
4442 checkGLcall("glUseProgramObjectARB(programId)");
4444 /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
4445 * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
4446 * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
4447 * vertex shader with fixed function pixel processing is used we make sure that the card
4448 * supports enough samplers to allow the max number of vertex samplers with all possible
4449 * fixed function fragment processing setups. So once the program is linked these samplers
4452 if (vshader) shader_glsl_load_vsamplers(gl_info, device->texUnitMap, programId);
4453 if (pshader) shader_glsl_load_psamplers(gl_info, device->texUnitMap, programId);
4455 /* If the local constants do not have to be loaded with the environment constants,
4456 * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
4459 if (pshader && !((IWineD3DBaseShaderImpl *)pshader)->baseShader.load_local_constsF)
4461 hardcode_local_constants((IWineD3DBaseShaderImpl *) pshader, gl_info, programId, 'P');
4463 if (vshader && !((IWineD3DBaseShaderImpl *)vshader)->baseShader.load_local_constsF)
4465 hardcode_local_constants((IWineD3DBaseShaderImpl *) vshader, gl_info, programId, 'V');
4469 /* GL locking is done by the caller */
4470 static GLhandleARB create_glsl_blt_shader(const struct wined3d_gl_info *gl_info, enum tex_types tex_type, BOOL masked)
4472 GLhandleARB program_id;
4473 GLhandleARB vshader_id, pshader_id;
4474 const char *blt_pshader;
4476 static const char *blt_vshader =
4480 " gl_Position = gl_Vertex;\n"
4481 " gl_FrontColor = vec4(1.0);\n"
4482 " gl_TexCoord[0] = gl_MultiTexCoord0;\n"
4485 static const char * const blt_pshaders_full[tex_type_count] =
4491 "uniform sampler2D sampler;\n"
4494 " gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4500 "uniform samplerCube sampler;\n"
4503 " gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4507 "#extension GL_ARB_texture_rectangle : enable\n"
4508 "uniform sampler2DRect sampler;\n"
4511 " gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4515 static const char * const blt_pshaders_masked[tex_type_count] =
4521 "uniform sampler2D sampler;\n"
4522 "uniform vec4 mask;\n"
4525 " if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4526 " gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4532 "uniform samplerCube sampler;\n"
4533 "uniform vec4 mask;\n"
4536 " if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4537 " gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4541 "#extension GL_ARB_texture_rectangle : enable\n"
4542 "uniform sampler2DRect sampler;\n"
4543 "uniform vec4 mask;\n"
4546 " if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4547 " gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4551 blt_pshader = masked ? blt_pshaders_masked[tex_type] : blt_pshaders_full[tex_type];
4554 FIXME("tex_type %#x not supported\n", tex_type);
4558 vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4559 shader_glsl_compile(gl_info, vshader_id, blt_vshader);
4561 pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
4562 shader_glsl_compile(gl_info, pshader_id, blt_pshader);
4564 program_id = GL_EXTCALL(glCreateProgramObjectARB());
4565 GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
4566 GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
4567 GL_EXTCALL(glLinkProgramARB(program_id));
4569 shader_glsl_validate_link(gl_info, program_id);
4571 /* Once linked we can mark the shaders for deletion. They will be deleted once the program
4574 GL_EXTCALL(glDeleteObjectARB(vshader_id));
4575 GL_EXTCALL(glDeleteObjectARB(pshader_id));
4579 /* GL locking is done by the caller */
4580 static void shader_glsl_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4582 const struct wined3d_gl_info *gl_info = context->gl_info;
4583 IWineD3DDeviceImpl *device = context->swapchain->device;
4584 struct shader_glsl_priv *priv = device->shader_priv;
4585 GLhandleARB program_id = 0;
4586 GLenum old_vertex_color_clamp, current_vertex_color_clamp;
4588 old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4590 if (useVS || usePS) set_glsl_shader_program(context, device, usePS, useVS);
4591 else priv->glsl_program = NULL;
4593 current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4595 if (old_vertex_color_clamp != current_vertex_color_clamp)
4597 if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4599 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
4600 checkGLcall("glClampColorARB");
4604 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4608 program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4609 if (program_id) TRACE("Using GLSL program %u\n", program_id);
4610 GL_EXTCALL(glUseProgramObjectARB(program_id));
4611 checkGLcall("glUseProgramObjectARB");
4613 /* In case that NP2 texcoord fixup data is found for the selected program, trigger a reload of the
4614 * constants. This has to be done because it can't be guaranteed that sampler() (from state.c) is
4615 * called between selecting the shader and using it, which results in wrong fixup for some frames. */
4616 if (priv->glsl_program && priv->glsl_program->np2Fixup_info)
4618 shader_glsl_load_np2fixup_constants((IWineD3DDevice *)device, usePS, useVS);
4622 /* GL locking is done by the caller */
4623 static void shader_glsl_select_depth_blt(IWineD3DDevice *iface,
4624 enum tex_types tex_type, const SIZE *ds_mask_size)
4626 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4627 const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4628 BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4629 struct shader_glsl_priv *priv = This->shader_priv;
4630 GLhandleARB *blt_program;
4633 blt_program = masked ? &priv->depth_blt_program_masked[tex_type] : &priv->depth_blt_program_full[tex_type];
4636 *blt_program = create_glsl_blt_shader(gl_info, tex_type, masked);
4637 loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "sampler"));
4638 GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4639 GL_EXTCALL(glUniform1iARB(loc, 0));
4643 GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4648 loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "mask"));
4649 GL_EXTCALL(glUniform4fARB(loc, 0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy));
4653 /* GL locking is done by the caller */
4654 static void shader_glsl_deselect_depth_blt(IWineD3DDevice *iface) {
4655 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4656 const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4657 struct shader_glsl_priv *priv = This->shader_priv;
4658 GLhandleARB program_id;
4660 program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4661 if (program_id) TRACE("Using GLSL program %u\n", program_id);
4663 GL_EXTCALL(glUseProgramObjectARB(program_id));
4664 checkGLcall("glUseProgramObjectARB");
4667 static void shader_glsl_destroy(IWineD3DBaseShader *iface) {
4668 const struct list *linked_programs;
4669 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) iface;
4670 IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
4671 struct shader_glsl_priv *priv = device->shader_priv;
4672 const struct wined3d_gl_info *gl_info;
4673 struct wined3d_context *context;
4675 /* Note: Do not use QueryInterface here to find out which shader type this is because this code
4676 * can be called from IWineD3DBaseShader::Release
4678 char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
4681 struct glsl_pshader_private *shader_data;
4682 shader_data = This->baseShader.backend_data;
4683 if (!shader_data || !shader_data->num_gl_shaders)
4685 HeapFree(GetProcessHeap(), 0, shader_data);
4686 This->baseShader.backend_data = NULL;
4690 context = context_acquire(device, NULL);
4691 gl_info = context->gl_info;
4693 if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->pshader == iface)
4696 shader_glsl_select(context, FALSE, FALSE);
4700 struct glsl_vshader_private *shader_data;
4701 shader_data = This->baseShader.backend_data;
4702 if (!shader_data || !shader_data->num_gl_shaders)
4704 HeapFree(GetProcessHeap(), 0, shader_data);
4705 This->baseShader.backend_data = NULL;
4709 context = context_acquire(device, NULL);
4710 gl_info = context->gl_info;
4712 if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->vshader == iface)
4715 shader_glsl_select(context, FALSE, FALSE);
4720 linked_programs = &This->baseShader.linked_programs;
4722 TRACE("Deleting linked programs\n");
4723 if (linked_programs->next) {
4724 struct glsl_shader_prog_link *entry, *entry2;
4728 LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
4729 delete_glsl_program_entry(priv, gl_info, entry);
4732 LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
4733 delete_glsl_program_entry(priv, gl_info, entry);
4741 struct glsl_pshader_private *shader_data = This->baseShader.backend_data;
4744 for(i = 0; i < shader_data->num_gl_shaders; i++) {
4745 TRACE("deleting pshader %u\n", shader_data->gl_shaders[i].prgId);
4746 GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4747 checkGLcall("glDeleteObjectARB");
4750 HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4755 struct glsl_vshader_private *shader_data = This->baseShader.backend_data;
4758 for(i = 0; i < shader_data->num_gl_shaders; i++) {
4759 TRACE("deleting vshader %u\n", shader_data->gl_shaders[i].prgId);
4760 GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4761 checkGLcall("glDeleteObjectARB");
4764 HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4767 HeapFree(GetProcessHeap(), 0, This->baseShader.backend_data);
4768 This->baseShader.backend_data = NULL;
4770 context_release(context);
4773 static int glsl_program_key_compare(const void *key, const struct wine_rb_entry *entry)
4775 const glsl_program_key_t *k = key;
4776 const struct glsl_shader_prog_link *prog = WINE_RB_ENTRY_VALUE(entry,
4777 const struct glsl_shader_prog_link, program_lookup_entry);
4780 if (k->vshader > prog->vshader) return 1;
4781 else if (k->vshader < prog->vshader) return -1;
4783 if (k->pshader > prog->pshader) return 1;
4784 else if (k->pshader < prog->pshader) return -1;
4786 if (k->vshader && (cmp = memcmp(&k->vs_args, &prog->vs_args, sizeof(prog->vs_args)))) return cmp;
4787 if (k->pshader && (cmp = memcmp(&k->ps_args, &prog->ps_args, sizeof(prog->ps_args)))) return cmp;
4792 static BOOL constant_heap_init(struct constant_heap *heap, unsigned int constant_count)
4794 SIZE_T size = (constant_count + 1) * sizeof(*heap->entries) + constant_count * sizeof(*heap->positions);
4795 void *mem = HeapAlloc(GetProcessHeap(), 0, size);
4799 ERR("Failed to allocate memory\n");
4803 heap->entries = mem;
4804 heap->entries[1].version = 0;
4805 heap->positions = (unsigned int *)(heap->entries + constant_count + 1);
4811 static void constant_heap_free(struct constant_heap *heap)
4813 HeapFree(GetProcessHeap(), 0, heap->entries);
4816 static const struct wine_rb_functions wined3d_glsl_program_rb_functions =
4821 glsl_program_key_compare,
4824 static HRESULT shader_glsl_alloc(IWineD3DDevice *iface) {
4825 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4826 const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4827 struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
4828 SIZE_T stack_size = wined3d_log2i(max(gl_info->limits.glsl_vs_float_constants,
4829 gl_info->limits.glsl_ps_float_constants)) + 1;
4831 if (!shader_buffer_init(&priv->shader_buffer))
4833 ERR("Failed to initialize shader buffer.\n");
4837 priv->stack = HeapAlloc(GetProcessHeap(), 0, stack_size * sizeof(*priv->stack));
4840 ERR("Failed to allocate memory.\n");
4844 if (!constant_heap_init(&priv->vconst_heap, gl_info->limits.glsl_vs_float_constants))
4846 ERR("Failed to initialize vertex shader constant heap\n");
4850 if (!constant_heap_init(&priv->pconst_heap, gl_info->limits.glsl_ps_float_constants))
4852 ERR("Failed to initialize pixel shader constant heap\n");
4856 if (wine_rb_init(&priv->program_lookup, &wined3d_glsl_program_rb_functions) == -1)
4858 ERR("Failed to initialize rbtree.\n");
4862 priv->next_constant_version = 1;
4864 This->shader_priv = priv;
4868 constant_heap_free(&priv->pconst_heap);
4869 constant_heap_free(&priv->vconst_heap);
4870 HeapFree(GetProcessHeap(), 0, priv->stack);
4871 shader_buffer_free(&priv->shader_buffer);
4872 HeapFree(GetProcessHeap(), 0, priv);
4873 return E_OUTOFMEMORY;
4876 /* Context activation is done by the caller. */
4877 static void shader_glsl_free(IWineD3DDevice *iface) {
4878 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
4879 const struct wined3d_gl_info *gl_info = &This->adapter->gl_info;
4880 struct shader_glsl_priv *priv = This->shader_priv;
4884 for (i = 0; i < tex_type_count; ++i)
4886 if (priv->depth_blt_program_full[i])
4888 GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_full[i]));
4890 if (priv->depth_blt_program_masked[i])
4892 GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_masked[i]));
4897 wine_rb_destroy(&priv->program_lookup, NULL, NULL);
4898 constant_heap_free(&priv->pconst_heap);
4899 constant_heap_free(&priv->vconst_heap);
4900 HeapFree(GetProcessHeap(), 0, priv->stack);
4901 shader_buffer_free(&priv->shader_buffer);
4903 HeapFree(GetProcessHeap(), 0, This->shader_priv);
4904 This->shader_priv = NULL;
4907 static BOOL shader_glsl_dirty_const(IWineD3DDevice *iface) {
4908 /* TODO: GL_EXT_bindable_uniform can be used to share constants across shaders */
4912 static void shader_glsl_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *pCaps)
4914 /* Nvidia Geforce6/7 or Ati R4xx/R5xx cards with GLSL support, support VS 3.0 but older Nvidia/Ati
4915 * models with GLSL support only support 2.0. In case of nvidia we can detect VS 2.0 support based
4916 * on the version of NV_vertex_program.
4917 * For Ati cards there's no way using glsl (it abstracts the lowlevel info away) and also not
4918 * using ARB_vertex_program. It is safe to assume that when a card supports pixel shader 2.0 it
4919 * supports vertex shader 2.0 too and the way around. We can detect ps2.0 using the maximum number
4920 * of native instructions, so use that here. For more info see the pixel shader versioning code below.
4922 if ((gl_info->supported[NV_VERTEX_PROGRAM2] && !gl_info->supported[NV_VERTEX_PROGRAM3])
4923 || gl_info->limits.arb_ps_instructions <= 512)
4924 pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4926 pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4927 TRACE_(d3d_caps)("Hardware vertex shader version %d.%d enabled (GLSL)\n", (pCaps->VertexShaderVersion >> 8) & 0xff, pCaps->VertexShaderVersion & 0xff);
4928 pCaps->MaxVertexShaderConst = gl_info->limits.glsl_vs_float_constants;
4930 /* Older DX9-class videocards (GeforceFX / Radeon >9500/X*00) only support pixel shader 2.0/2.0a/2.0b.
4931 * In OpenGL the extensions related to GLSL abstract lowlevel GL info away which is needed
4932 * to distinguish between 2.0 and 3.0 (and 2.0a/2.0b). In case of Nvidia we use their fragment
4933 * program extensions. On other hardware including ATI GL_ARB_fragment_program offers the info
4934 * in max native instructions. Intel and others also offer the info in this extension but they
4935 * don't support GLSL (at least on Windows).
4937 * PS2.0 requires at least 96 instructions, 2.0a/2.0b go up to 512. Assume that if the number
4938 * of instructions is 512 or less we have to do with ps2.0 hardware.
4939 * 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.
4941 if ((gl_info->supported[NV_FRAGMENT_PROGRAM] && !gl_info->supported[NV_FRAGMENT_PROGRAM2])
4942 || gl_info->limits.arb_ps_instructions <= 512)
4943 pCaps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
4945 pCaps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
4947 pCaps->MaxPixelShaderConst = gl_info->limits.glsl_ps_float_constants;
4949 /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
4950 * Direct3D minimum requirement.
4952 * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
4953 * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
4955 * The problem is that the refrast clamps temporary results in the shader to
4956 * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
4957 * then applications may miss the clamping behavior. On the other hand, if it is smaller,
4958 * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
4959 * offer a way to query this.
4961 pCaps->PixelShader1xMaxValue = 8.0;
4962 TRACE_(d3d_caps)("Hardware pixel shader version %d.%d enabled (GLSL)\n", (pCaps->PixelShaderVersion >> 8) & 0xff, pCaps->PixelShaderVersion & 0xff);
4964 pCaps->VSClipping = TRUE;
4967 static BOOL shader_glsl_color_fixup_supported(struct color_fixup_desc fixup)
4969 if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4971 TRACE("Checking support for fixup:\n");
4972 dump_color_fixup_desc(fixup);
4975 /* We support everything except YUV conversions. */
4976 if (!is_complex_fixup(fixup))
4982 TRACE("[FAILED]\n");
4986 static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4988 /* WINED3DSIH_ABS */ shader_glsl_map2gl,
4989 /* WINED3DSIH_ADD */ shader_glsl_arith,
4990 /* WINED3DSIH_BEM */ shader_glsl_bem,
4991 /* WINED3DSIH_BREAK */ shader_glsl_break,
4992 /* WINED3DSIH_BREAKC */ shader_glsl_breakc,
4993 /* WINED3DSIH_BREAKP */ NULL,
4994 /* WINED3DSIH_CALL */ shader_glsl_call,
4995 /* WINED3DSIH_CALLNZ */ shader_glsl_callnz,
4996 /* WINED3DSIH_CMP */ shader_glsl_cmp,
4997 /* WINED3DSIH_CND */ shader_glsl_cnd,
4998 /* WINED3DSIH_CRS */ shader_glsl_cross,
4999 /* WINED3DSIH_CUT */ NULL,
5000 /* WINED3DSIH_DCL */ NULL,
5001 /* WINED3DSIH_DEF */ NULL,
5002 /* WINED3DSIH_DEFB */ NULL,
5003 /* WINED3DSIH_DEFI */ NULL,
5004 /* WINED3DSIH_DP2ADD */ shader_glsl_dp2add,
5005 /* WINED3DSIH_DP3 */ shader_glsl_dot,
5006 /* WINED3DSIH_DP4 */ shader_glsl_dot,
5007 /* WINED3DSIH_DST */ shader_glsl_dst,
5008 /* WINED3DSIH_DSX */ shader_glsl_map2gl,
5009 /* WINED3DSIH_DSY */ shader_glsl_map2gl,
5010 /* WINED3DSIH_ELSE */ shader_glsl_else,
5011 /* WINED3DSIH_EMIT */ NULL,
5012 /* WINED3DSIH_ENDIF */ shader_glsl_end,
5013 /* WINED3DSIH_ENDLOOP */ shader_glsl_end,
5014 /* WINED3DSIH_ENDREP */ shader_glsl_end,
5015 /* WINED3DSIH_EXP */ shader_glsl_map2gl,
5016 /* WINED3DSIH_EXPP */ shader_glsl_expp,
5017 /* WINED3DSIH_FRC */ shader_glsl_map2gl,
5018 /* WINED3DSIH_IADD */ NULL,
5019 /* WINED3DSIH_IF */ shader_glsl_if,
5020 /* WINED3DSIH_IFC */ shader_glsl_ifc,
5021 /* WINED3DSIH_IGE */ NULL,
5022 /* WINED3DSIH_LABEL */ shader_glsl_label,
5023 /* WINED3DSIH_LIT */ shader_glsl_lit,
5024 /* WINED3DSIH_LOG */ shader_glsl_log,
5025 /* WINED3DSIH_LOGP */ shader_glsl_log,
5026 /* WINED3DSIH_LOOP */ shader_glsl_loop,
5027 /* WINED3DSIH_LRP */ shader_glsl_lrp,
5028 /* WINED3DSIH_LT */ NULL,
5029 /* WINED3DSIH_M3x2 */ shader_glsl_mnxn,
5030 /* WINED3DSIH_M3x3 */ shader_glsl_mnxn,
5031 /* WINED3DSIH_M3x4 */ shader_glsl_mnxn,
5032 /* WINED3DSIH_M4x3 */ shader_glsl_mnxn,
5033 /* WINED3DSIH_M4x4 */ shader_glsl_mnxn,
5034 /* WINED3DSIH_MAD */ shader_glsl_mad,
5035 /* WINED3DSIH_MAX */ shader_glsl_map2gl,
5036 /* WINED3DSIH_MIN */ shader_glsl_map2gl,
5037 /* WINED3DSIH_MOV */ shader_glsl_mov,
5038 /* WINED3DSIH_MOVA */ shader_glsl_mov,
5039 /* WINED3DSIH_MUL */ shader_glsl_arith,
5040 /* WINED3DSIH_NOP */ NULL,
5041 /* WINED3DSIH_NRM */ shader_glsl_nrm,
5042 /* WINED3DSIH_PHASE */ NULL,
5043 /* WINED3DSIH_POW */ shader_glsl_pow,
5044 /* WINED3DSIH_RCP */ shader_glsl_rcp,
5045 /* WINED3DSIH_REP */ shader_glsl_rep,
5046 /* WINED3DSIH_RET */ shader_glsl_ret,
5047 /* WINED3DSIH_RSQ */ shader_glsl_rsq,
5048 /* WINED3DSIH_SETP */ NULL,
5049 /* WINED3DSIH_SGE */ shader_glsl_compare,
5050 /* WINED3DSIH_SGN */ shader_glsl_sgn,
5051 /* WINED3DSIH_SINCOS */ shader_glsl_sincos,
5052 /* WINED3DSIH_SLT */ shader_glsl_compare,
5053 /* WINED3DSIH_SUB */ shader_glsl_arith,
5054 /* WINED3DSIH_TEX */ shader_glsl_tex,
5055 /* WINED3DSIH_TEXBEM */ shader_glsl_texbem,
5056 /* WINED3DSIH_TEXBEML */ shader_glsl_texbem,
5057 /* WINED3DSIH_TEXCOORD */ shader_glsl_texcoord,
5058 /* WINED3DSIH_TEXDEPTH */ shader_glsl_texdepth,
5059 /* WINED3DSIH_TEXDP3 */ shader_glsl_texdp3,
5060 /* WINED3DSIH_TEXDP3TEX */ shader_glsl_texdp3tex,
5061 /* WINED3DSIH_TEXKILL */ shader_glsl_texkill,
5062 /* WINED3DSIH_TEXLDD */ shader_glsl_texldd,
5063 /* WINED3DSIH_TEXLDL */ shader_glsl_texldl,
5064 /* WINED3DSIH_TEXM3x2DEPTH */ shader_glsl_texm3x2depth,
5065 /* WINED3DSIH_TEXM3x2PAD */ shader_glsl_texm3x2pad,
5066 /* WINED3DSIH_TEXM3x2TEX */ shader_glsl_texm3x2tex,
5067 /* WINED3DSIH_TEXM3x3 */ shader_glsl_texm3x3,
5068 /* WINED3DSIH_TEXM3x3DIFF */ NULL,
5069 /* WINED3DSIH_TEXM3x3PAD */ shader_glsl_texm3x3pad,
5070 /* WINED3DSIH_TEXM3x3SPEC */ shader_glsl_texm3x3spec,
5071 /* WINED3DSIH_TEXM3x3TEX */ shader_glsl_texm3x3tex,
5072 /* WINED3DSIH_TEXM3x3VSPEC */ shader_glsl_texm3x3vspec,
5073 /* WINED3DSIH_TEXREG2AR */ shader_glsl_texreg2ar,
5074 /* WINED3DSIH_TEXREG2GB */ shader_glsl_texreg2gb,
5075 /* WINED3DSIH_TEXREG2RGB */ shader_glsl_texreg2rgb,
5078 static void shader_glsl_handle_instruction(const struct wined3d_shader_instruction *ins) {
5079 SHADER_HANDLER hw_fct;
5081 /* Select handler */
5082 hw_fct = shader_glsl_instruction_handler_table[ins->handler_idx];
5084 /* Unhandled opcode */
5087 FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5092 shader_glsl_add_instruction_modifiers(ins);
5095 const shader_backend_t glsl_shader_backend = {
5096 shader_glsl_handle_instruction,
5098 shader_glsl_select_depth_blt,
5099 shader_glsl_deselect_depth_blt,
5100 shader_glsl_update_float_vertex_constants,
5101 shader_glsl_update_float_pixel_constants,
5102 shader_glsl_load_constants,
5103 shader_glsl_load_np2fixup_constants,
5104 shader_glsl_destroy,
5107 shader_glsl_dirty_const,
5108 shader_glsl_get_caps,
5109 shader_glsl_color_fixup_supported,