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 2009-2010 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(void *shader_priv,
685 const struct wined3d_gl_info *gl_info, const struct wined3d_state *state)
687 struct shader_glsl_priv *glsl_priv = shader_priv;
688 const struct glsl_shader_prog_link *prog = glsl_priv->glsl_program;
690 /* No GLSL program set - nothing to do. */
693 /* NP2 texcoord fixup is (currently) only done for pixelshaders. */
694 if (!use_ps(state)) return;
696 if (prog->ps_args.np2_fixup && prog->np2Fixup_location != -1)
699 UINT fixup = prog->ps_args.np2_fixup;
700 GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
702 for (i = 0; fixup; fixup >>= 1, ++i)
704 const IWineD3DBaseTextureImpl *tex = state->textures[i];
705 const unsigned char idx = prog->np2Fixup_info->idx[i];
706 GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
710 ERR("Nonexistent texture is flagged for NP2 texcoord fixup.\n");
715 tex_dim[2] = tex->baseTexture.pow2Matrix[0]; tex_dim[3] = tex->baseTexture.pow2Matrix[5];
717 tex_dim[0] = tex->baseTexture.pow2Matrix[0]; tex_dim[1] = tex->baseTexture.pow2Matrix[5];
721 GL_EXTCALL(glUniform4fvARB(prog->np2Fixup_location, prog->np2Fixup_info->num_consts, np2fixup_constants));
726 * Loads the app-supplied constants into the currently set GLSL program.
728 /* GL locking is done by the caller (state handler) */
729 static void shader_glsl_load_constants(const struct wined3d_context *context,
730 char usePixelShader, char useVertexShader)
732 const struct wined3d_gl_info *gl_info = context->gl_info;
733 IWineD3DDeviceImpl *device = context->swapchain->device;
734 IWineD3DStateBlockImpl* stateBlock = device->stateBlock;
735 struct shader_glsl_priv *priv = device->shader_priv;
736 float position_fixup[4];
738 GLhandleARB programId;
739 struct glsl_shader_prog_link *prog = priv->glsl_program;
740 UINT constant_version;
744 /* No GLSL program set - nothing to do. */
747 programId = prog->programId;
748 constant_version = prog->constant_version;
752 IWineD3DBaseShaderImpl *vshader = (IWineD3DBaseShaderImpl *)stateBlock->state.vertex_shader;
754 /* Load DirectX 9 float constants/uniforms for vertex shader */
755 shader_glsl_load_constantsF(vshader, gl_info, stateBlock->state.vs_consts_f,
756 prog->vuniformF_locations, &priv->vconst_heap, priv->stack, constant_version);
758 /* Load DirectX 9 integer constants/uniforms for vertex shader */
759 shader_glsl_load_constantsI(vshader, gl_info, prog->vuniformI_locations, stateBlock->state.vs_consts_i,
760 stateBlock->changed.vertexShaderConstantsI & vshader->baseShader.reg_maps.integer_constants);
762 /* Load DirectX 9 boolean constants/uniforms for vertex shader */
763 shader_glsl_load_constantsB(vshader, gl_info, programId, stateBlock->state.vs_consts_b,
764 stateBlock->changed.vertexShaderConstantsB & vshader->baseShader.reg_maps.boolean_constants);
766 /* Upload the position fixup params */
767 shader_get_position_fixup(context, &stateBlock->state, position_fixup);
768 GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, position_fixup));
769 checkGLcall("glUniform4fvARB");
774 IWineD3DBaseShaderImpl *pshader = (IWineD3DBaseShaderImpl *)stateBlock->state.pixel_shader;
776 /* Load DirectX 9 float constants/uniforms for pixel shader */
777 shader_glsl_load_constantsF(pshader, gl_info, stateBlock->state.ps_consts_f,
778 prog->puniformF_locations, &priv->pconst_heap, priv->stack, constant_version);
780 /* Load DirectX 9 integer constants/uniforms for pixel shader */
781 shader_glsl_load_constantsI(pshader, gl_info, prog->puniformI_locations, stateBlock->state.ps_consts_i,
782 stateBlock->changed.pixelShaderConstantsI & pshader->baseShader.reg_maps.integer_constants);
784 /* Load DirectX 9 boolean constants/uniforms for pixel shader */
785 shader_glsl_load_constantsB(pshader, gl_info, programId, stateBlock->state.ps_consts_b,
786 stateBlock->changed.pixelShaderConstantsB & pshader->baseShader.reg_maps.boolean_constants);
788 /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
789 * It can't be 0 for a valid texbem instruction.
791 for(i = 0; i < MAX_TEXTURES; i++) {
794 if(prog->bumpenvmat_location[i] == -1) continue;
796 data = (const float *)&stateBlock->state.texture_states[i][WINED3DTSS_BUMPENVMAT00];
797 GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
798 checkGLcall("glUniformMatrix2fvARB");
800 /* texbeml needs the luminance scale and offset too. If texbeml
801 * is used, needsbumpmat is set too, so we can check that in the
802 * needsbumpmat check. */
803 if (prog->luminancescale_location[i] != -1)
805 const GLfloat *scale = (const GLfloat *)&stateBlock->state.texture_states[i][WINED3DTSS_BUMPENVLSCALE];
806 const GLfloat *offset = (const GLfloat *)&stateBlock->state.texture_states[i][WINED3DTSS_BUMPENVLOFFSET];
808 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
809 checkGLcall("glUniform1fvARB");
810 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
811 checkGLcall("glUniform1fvARB");
815 if(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
816 float correction_params[4];
818 if (context->render_offscreen)
820 correction_params[0] = 0.0f;
821 correction_params[1] = 1.0f;
823 /* position is window relative, not viewport relative */
824 correction_params[0] = context->current_rt->currentDesc.Height;
825 correction_params[1] = -1.0f;
827 GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
831 if (priv->next_constant_version == UINT_MAX)
833 TRACE("Max constant version reached, resetting to 0.\n");
834 wine_rb_for_each_entry(&priv->program_lookup, reset_program_constant_version, NULL);
835 priv->next_constant_version = 1;
839 prog->constant_version = priv->next_constant_version++;
843 static inline void update_heap_entry(struct constant_heap *heap, unsigned int idx,
844 unsigned int heap_idx, DWORD new_version)
846 struct constant_entry *entries = heap->entries;
847 unsigned int *positions = heap->positions;
848 unsigned int parent_idx;
852 parent_idx = heap_idx >> 1;
854 if (new_version <= entries[parent_idx].version) break;
856 entries[heap_idx] = entries[parent_idx];
857 positions[entries[parent_idx].idx] = heap_idx;
858 heap_idx = parent_idx;
861 entries[heap_idx].version = new_version;
862 entries[heap_idx].idx = idx;
863 positions[idx] = heap_idx;
866 static void shader_glsl_update_float_vertex_constants(IWineD3DDeviceImpl *device, UINT start, UINT count)
868 struct shader_glsl_priv *priv = device->shader_priv;
869 struct constant_heap *heap = &priv->vconst_heap;
872 for (i = start; i < count + start; ++i)
874 if (!device->stateBlock->changed.vertexShaderConstantsF[i])
875 update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
877 update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
881 static void shader_glsl_update_float_pixel_constants(IWineD3DDeviceImpl *device, UINT start, UINT count)
883 struct shader_glsl_priv *priv = device->shader_priv;
884 struct constant_heap *heap = &priv->pconst_heap;
887 for (i = start; i < count + start; ++i)
889 if (!device->stateBlock->changed.pixelShaderConstantsF[i])
890 update_heap_entry(heap, i, heap->size++, priv->next_constant_version);
892 update_heap_entry(heap, i, heap->positions[i], priv->next_constant_version);
896 static unsigned int vec4_varyings(DWORD shader_major, const struct wined3d_gl_info *gl_info)
898 unsigned int ret = gl_info->limits.glsl_varyings / 4;
899 /* 4.0 shaders do not write clip coords because d3d10 does not support user clipplanes */
900 if(shader_major > 3) return ret;
902 /* 3.0 shaders may need an extra varying for the clip coord on some cards(mostly dx10 ones) */
903 if (gl_info->quirks & WINED3D_QUIRK_GLSL_CLIP_VARYING) ret -= 1;
907 /** Generate the variable & register declarations for the GLSL output target */
908 static void shader_generate_glsl_declarations(const struct wined3d_context *context,
909 struct wined3d_shader_buffer *buffer, IWineD3DBaseShader *iface,
910 const struct wined3d_shader_reg_maps *reg_maps, struct shader_glsl_ctx_priv *ctx_priv)
912 IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
913 IWineD3DDeviceImpl *device = This->baseShader.device;
914 const struct wined3d_state *state = &device->stateBlock->state;
915 const struct ps_compile_args *ps_args = ctx_priv->cur_ps_args;
916 const struct wined3d_gl_info *gl_info = context->gl_info;
917 unsigned int i, extra_constants_needed = 0;
918 const local_constant *lconst;
921 /* There are some minor differences between pixel and vertex shaders */
922 char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
923 char prefix = pshader ? 'P' : 'V';
925 /* Prototype the subroutines */
926 for (i = 0, map = reg_maps->labels; map; map >>= 1, ++i)
928 if (map & 1) shader_addline(buffer, "void subroutine%u();\n", i);
931 /* Declare the constants (aka uniforms) */
932 if (This->baseShader.limits.constant_float > 0) {
933 unsigned max_constantsF;
934 /* Unless the shader uses indirect addressing, always declare the maximum array size and ignore that we need some
935 * uniforms privately. E.g. if GL supports 256 uniforms, and we need 2 for the pos fixup and immediate values, still
936 * declare VC[256]. If the shader needs more uniforms than we have it won't work in any case. If it uses less, the
937 * compiler will figure out which uniforms are really used and strip them out. This allows a shader to use c255 on
938 * a dx9 card, as long as it doesn't also use all the other constants.
940 * If the shader uses indirect addressing the compiler must assume that all declared uniforms are used. In this case,
941 * declare only the amount that we're assured to have.
943 * Thus we run into problems in these two cases:
944 * 1) The shader really uses more uniforms than supported
945 * 2) The shader uses indirect addressing, less constants than supported, but uses a constant index > #supported consts
949 /* No indirect addressing here. */
950 max_constantsF = gl_info->limits.glsl_ps_float_constants;
954 if (reg_maps->usesrelconstF)
956 /* Subtract the other potential uniforms from the max
957 * available (bools, ints, and 1 row of projection matrix).
958 * Subtract another uniform for immediate values, which have
959 * to be loaded via uniform by the driver as well. The shader
960 * code only uses 0.5, 2.0, 1.0, 128 and -128 in vertex
961 * shader code, so one vec4 should be enough. (Unfortunately
962 * the Nvidia driver doesn't store 128 and -128 in one float).
964 * Writing gl_ClipVertex requires one uniform for each
965 * clipplane as well. */
966 max_constantsF = gl_info->limits.glsl_vs_float_constants - 3;
967 if(ctx_priv->cur_vs_args->clip_enabled)
969 max_constantsF -= gl_info->limits.clipplanes;
971 max_constantsF -= count_bits(reg_maps->integer_constants);
972 /* Strictly speaking a bool only uses one scalar, but the nvidia(Linux) compiler doesn't pack them properly,
973 * so each scalar requires a full vec4. We could work around this by packing the booleans ourselves, but
974 * for now take this into account when calculating the number of available constants
976 max_constantsF -= count_bits(reg_maps->boolean_constants);
977 /* Set by driver quirks in directx.c */
978 max_constantsF -= gl_info->reserved_glsl_constants;
982 max_constantsF = gl_info->limits.glsl_vs_float_constants;
985 max_constantsF = min(This->baseShader.limits.constant_float, max_constantsF);
986 shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
989 /* Always declare the full set of constants, the compiler can remove the unused ones because d3d doesn't(yet)
990 * support indirect int and bool constant addressing. This avoids problems if the app uses e.g. i0 and i9.
992 if (This->baseShader.limits.constant_int > 0 && reg_maps->integer_constants)
993 shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, This->baseShader.limits.constant_int);
995 if (This->baseShader.limits.constant_bool > 0 && reg_maps->boolean_constants)
996 shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, This->baseShader.limits.constant_bool);
1000 shader_addline(buffer, "uniform vec4 posFixup;\n");
1001 shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
1005 for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
1007 if (!(map & 1)) continue;
1009 shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
1011 if (reg_maps->luminanceparams & (1 << i))
1013 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
1014 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
1015 extra_constants_needed++;
1018 extra_constants_needed++;
1021 if (ps_args->srgb_correction)
1023 shader_addline(buffer, "const vec4 srgb_const0 = vec4(%.8e, %.8e, %.8e, %.8e);\n",
1024 srgb_pow, srgb_mul_high, srgb_sub_high, srgb_mul_low);
1025 shader_addline(buffer, "const vec4 srgb_const1 = vec4(%.8e, 0.0, 0.0, 0.0);\n",
1028 if (reg_maps->vpos || reg_maps->usesdsy)
1030 if (This->baseShader.limits.constant_float + extra_constants_needed
1031 + 1 < gl_info->limits.glsl_ps_float_constants)
1033 shader_addline(buffer, "uniform vec4 ycorrection;\n");
1034 ((IWineD3DPixelShaderImpl *) This)->vpos_uniform = 1;
1035 extra_constants_needed++;
1037 /* This happens because we do not have proper tracking of the constant registers that are
1038 * actually used, only the max limit of the shader version
1040 FIXME("Cannot find a free uniform for vpos correction params\n");
1041 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
1042 context->render_offscreen ? 0.0f : device->render_targets[0]->currentDesc.Height,
1043 context->render_offscreen ? 1.0f : -1.0f);
1045 shader_addline(buffer, "vec4 vpos;\n");
1049 /* Declare texture samplers */
1050 for (i = 0; i < This->baseShader.limits.sampler; i++) {
1051 if (reg_maps->sampler_type[i])
1053 IWineD3DBaseTextureImpl *texture;
1055 switch (reg_maps->sampler_type[i])
1058 if (pshader && ps_args->shadow & (1 << i))
1059 shader_addline(buffer, "uniform sampler1DShadow %csampler%u;\n", prefix, i);
1061 shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
1064 texture = state->textures[i];
1065 if (pshader && ps_args->shadow & (1 << i))
1067 if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
1068 shader_addline(buffer, "uniform sampler2DRectShadow %csampler%u;\n", prefix, i);
1070 shader_addline(buffer, "uniform sampler2DShadow %csampler%u;\n", prefix, i);
1074 if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
1075 shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
1077 shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
1080 case WINED3DSTT_CUBE:
1081 if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported Cube shadow sampler.\n");
1082 shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
1084 case WINED3DSTT_VOLUME:
1085 if (pshader && ps_args->shadow & (1 << i)) FIXME("Unsupported 3D shadow sampler.\n");
1086 shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
1089 shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
1090 FIXME("Unrecognized sampler type: %#x\n", reg_maps->sampler_type[i]);
1096 /* Declare uniforms for NP2 texcoord fixup:
1097 * This is NOT done inside the loop that declares the texture samplers since the NP2 fixup code
1098 * is currently only used for the GeforceFX series and when forcing the ARB_npot extension off.
1099 * Modern cards just skip the code anyway, so put it inside a separate loop. */
1100 if (pshader && ps_args->np2_fixup) {
1102 struct ps_np2fixup_info* const fixup = ctx_priv->cur_np2fixup_info;
1105 /* NP2/RECT textures in OpenGL use texcoords in the range [0,width]x[0,height]
1106 * while D3D has them in the (normalized) [0,1]x[0,1] range.
1107 * samplerNP2Fixup stores texture dimensions and is updated through
1108 * shader_glsl_load_np2fixup_constants when the sampler changes. */
1110 for (i = 0; i < This->baseShader.limits.sampler; ++i) {
1111 if (reg_maps->sampler_type[i]) {
1112 if (!(ps_args->np2_fixup & (1 << i))) continue;
1114 if (WINED3DSTT_2D != reg_maps->sampler_type[i]) {
1115 FIXME("Non-2D texture is flagged for NP2 texcoord fixup.\n");
1119 fixup->idx[i] = cur++;
1123 fixup->num_consts = (cur + 1) >> 1;
1124 shader_addline(buffer, "uniform vec4 %csamplerNP2Fixup[%u];\n", prefix, fixup->num_consts);
1127 /* Declare address variables */
1128 for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
1130 if (map & 1) shader_addline(buffer, "ivec4 A%u;\n", i);
1133 /* Declare texture coordinate temporaries and initialize them */
1134 for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
1136 if (map & 1) shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
1139 /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
1140 * helper function shader that is linked in at link time
1142 if (pshader && reg_maps->shader_version.major >= 3)
1146 shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1148 /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
1149 * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
1150 * pixel shader that reads the fixed function color into the packed input registers.
1152 shader_addline(buffer, "vec4 IN[%u];\n", vec4_varyings(reg_maps->shader_version.major, gl_info));
1156 /* Declare output register temporaries */
1157 if(This->baseShader.limits.packed_output) {
1158 shader_addline(buffer, "vec4 OUT[%u];\n", This->baseShader.limits.packed_output);
1161 /* Declare temporary variables */
1162 for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
1164 if (map & 1) shader_addline(buffer, "vec4 R%u;\n", i);
1167 /* Declare attributes */
1168 if (reg_maps->shader_version.type == WINED3D_SHADER_TYPE_VERTEX)
1170 for (i = 0, map = reg_maps->input_registers; map; map >>= 1, ++i)
1172 if (map & 1) shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
1176 /* Declare loop registers aLx */
1177 for (i = 0; i < reg_maps->loop_depth; i++) {
1178 shader_addline(buffer, "int aL%u;\n", i);
1179 shader_addline(buffer, "int tmpInt%u;\n", i);
1182 /* Temporary variables for matrix operations */
1183 shader_addline(buffer, "vec4 tmp0;\n");
1184 shader_addline(buffer, "vec4 tmp1;\n");
1186 /* Local constants use a different name so they can be loaded once at shader link time
1187 * They can't be hardcoded into the shader text via LC = {x, y, z, w}; because the
1188 * float -> string conversion can cause precision loss.
1190 if(!This->baseShader.load_local_constsF) {
1191 LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
1192 shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
1196 /* Start the main program */
1197 shader_addline(buffer, "void main() {\n");
1198 if(pshader && reg_maps->vpos) {
1199 /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
1200 * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
1201 * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
1202 * precision troubles when we just substract 0.5.
1204 * To deal with that just floor() the position. This will eliminate the fraction on all cards.
1206 * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
1208 * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
1209 * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
1210 * coordinates specify the pixel centers instead of the pixel corners. This code will behave
1211 * correctly on drivers that returns integer values.
1213 shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
1217 /*****************************************************************************
1218 * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
1220 * For more information, see http://wiki.winehq.org/DirectX-Shaders
1221 ****************************************************************************/
1224 static void shader_glsl_add_src_param(const struct wined3d_shader_instruction *ins,
1225 const struct wined3d_shader_src_param *wined3d_src, DWORD mask, glsl_src_param_t *glsl_src);
1227 /** Used for opcode modifiers - They multiply the result by the specified amount */
1228 static const char * const shift_glsl_tab[] = {
1230 "2.0 * ", /* 1 (x2) */
1231 "4.0 * ", /* 2 (x4) */
1232 "8.0 * ", /* 3 (x8) */
1233 "16.0 * ", /* 4 (x16) */
1234 "32.0 * ", /* 5 (x32) */
1241 "0.0625 * ", /* 12 (d16) */
1242 "0.125 * ", /* 13 (d8) */
1243 "0.25 * ", /* 14 (d4) */
1244 "0.5 * " /* 15 (d2) */
1247 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
1248 static void shader_glsl_gen_modifier(DWORD src_modifier, const char *in_reg, const char *in_regswizzle, char *out_str)
1252 switch (src_modifier)
1254 case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
1255 case WINED3DSPSM_DW:
1256 case WINED3DSPSM_NONE:
1257 sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1259 case WINED3DSPSM_NEG:
1260 sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
1262 case WINED3DSPSM_NOT:
1263 sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
1265 case WINED3DSPSM_BIAS:
1266 sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1268 case WINED3DSPSM_BIASNEG:
1269 sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
1271 case WINED3DSPSM_SIGN:
1272 sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1274 case WINED3DSPSM_SIGNNEG:
1275 sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
1277 case WINED3DSPSM_COMP:
1278 sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
1280 case WINED3DSPSM_X2:
1281 sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
1283 case WINED3DSPSM_X2NEG:
1284 sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
1286 case WINED3DSPSM_ABS:
1287 sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
1289 case WINED3DSPSM_ABSNEG:
1290 sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
1293 FIXME("Unhandled modifier %u\n", src_modifier);
1294 sprintf(out_str, "%s%s", in_reg, in_regswizzle);
1298 /** Writes the GLSL variable name that corresponds to the register that the
1299 * DX opcode parameter is trying to access */
1300 static void shader_glsl_get_register_name(const struct wined3d_shader_register *reg,
1301 char *register_name, BOOL *is_color, const struct wined3d_shader_instruction *ins)
1303 /* oPos, oFog and oPts in D3D */
1304 static const char * const hwrastout_reg_names[] = {"OUT[10]", "OUT[11].x", "OUT[11].y"};
1306 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
1307 const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
1308 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
1309 char pshader = shader_is_pshader_version(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 (reg_maps->shader_version.major >= 3)
1332 DWORD idx = ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg->idx];
1333 unsigned int in_count = vec4_varyings(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", ins->ctx->loop_state->current_reg - 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_SCALAR:
1480 sprintf(register_name, "%.8e", *(const float *)reg->immconst_data);
1483 case WINED3D_IMMCONST_VEC4:
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 struct wined3d_shader_loop_state *loop_state = ins->ctx->loop_state;
2828 glsl_src_param_t src1_param;
2829 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2830 const DWORD *control_values = NULL;
2831 const local_constant *constant;
2833 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2835 /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2836 * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2837 * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2840 if (ins->src[1].reg.type == WINED3DSPR_CONSTINT)
2842 LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
2843 if (constant->idx == ins->src[1].reg.idx)
2845 control_values = constant->value;
2853 struct wined3d_shader_loop_control loop_control;
2854 loop_control.count = control_values[0];
2855 loop_control.start = control_values[1];
2856 loop_control.step = (int)control_values[2];
2858 if (loop_control.step > 0)
2860 shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u < (%u * %d + %u); aL%u += %d) {\n",
2861 loop_state->current_depth, loop_control.start,
2862 loop_state->current_depth, loop_control.count, loop_control.step, loop_control.start,
2863 loop_state->current_depth, loop_control.step);
2865 else if (loop_control.step < 0)
2867 shader_addline(ins->ctx->buffer, "for (aL%u = %u; aL%u > (%u * %d + %u); aL%u += %d) {\n",
2868 loop_state->current_depth, loop_control.start,
2869 loop_state->current_depth, loop_control.count, loop_control.step, loop_control.start,
2870 loop_state->current_depth, loop_control.step);
2874 shader_addline(ins->ctx->buffer, "for (aL%u = %u, tmpInt%u = 0; tmpInt%u < %u; tmpInt%u++) {\n",
2875 loop_state->current_depth, loop_control.start, loop_state->current_depth,
2876 loop_state->current_depth, loop_control.count,
2877 loop_state->current_depth);
2880 shader_addline(ins->ctx->buffer,
2881 "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2882 loop_state->current_depth, loop_state->current_reg,
2883 src1_param.reg_name, loop_state->current_depth, src1_param.reg_name,
2884 loop_state->current_depth, loop_state->current_reg, src1_param.reg_name);
2887 ++loop_state->current_depth;
2888 ++loop_state->current_reg;
2891 static void shader_glsl_end(const struct wined3d_shader_instruction *ins)
2893 struct wined3d_shader_loop_state *loop_state = ins->ctx->loop_state;
2895 shader_addline(ins->ctx->buffer, "}\n");
2897 if (ins->handler_idx == WINED3DSIH_ENDLOOP)
2899 --loop_state->current_depth;
2900 --loop_state->current_reg;
2903 if (ins->handler_idx == WINED3DSIH_ENDREP)
2905 --loop_state->current_depth;
2909 static void shader_glsl_rep(const struct wined3d_shader_instruction *ins)
2911 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
2912 struct wined3d_shader_loop_state *loop_state = ins->ctx->loop_state;
2913 glsl_src_param_t src0_param;
2914 const DWORD *control_values = NULL;
2915 const local_constant *constant;
2917 /* Try to hardcode local values to help the GLSL compiler to unroll and optimize the loop */
2918 if (ins->src[0].reg.type == WINED3DSPR_CONSTINT)
2920 LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry)
2922 if (constant->idx == ins->src[0].reg.idx)
2924 control_values = constant->value;
2932 shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %d; tmpInt%d++) {\n",
2933 loop_state->current_depth, loop_state->current_depth,
2934 control_values[0], loop_state->current_depth);
2938 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2939 shader_addline(ins->ctx->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2940 loop_state->current_depth, loop_state->current_depth,
2941 src0_param.param_str, loop_state->current_depth);
2944 ++loop_state->current_depth;
2947 static void shader_glsl_if(const struct wined3d_shader_instruction *ins)
2949 glsl_src_param_t src0_param;
2951 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2952 shader_addline(ins->ctx->buffer, "if (%s) {\n", src0_param.param_str);
2955 static void shader_glsl_ifc(const struct wined3d_shader_instruction *ins)
2957 glsl_src_param_t src0_param;
2958 glsl_src_param_t src1_param;
2960 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2961 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2963 shader_addline(ins->ctx->buffer, "if (%s %s %s) {\n",
2964 src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2967 static void shader_glsl_else(const struct wined3d_shader_instruction *ins)
2969 shader_addline(ins->ctx->buffer, "} else {\n");
2972 static void shader_glsl_break(const struct wined3d_shader_instruction *ins)
2974 shader_addline(ins->ctx->buffer, "break;\n");
2977 /* FIXME: According to MSDN the compare is done per component. */
2978 static void shader_glsl_breakc(const struct wined3d_shader_instruction *ins)
2980 glsl_src_param_t src0_param;
2981 glsl_src_param_t src1_param;
2983 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0, &src0_param);
2984 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
2986 shader_addline(ins->ctx->buffer, "if (%s %s %s) break;\n",
2987 src0_param.param_str, shader_get_comp_op(ins->flags), src1_param.param_str);
2990 static void shader_glsl_label(const struct wined3d_shader_instruction *ins)
2992 shader_addline(ins->ctx->buffer, "}\n");
2993 shader_addline(ins->ctx->buffer, "void subroutine%u () {\n", ins->src[0].reg.idx);
2996 static void shader_glsl_call(const struct wined3d_shader_instruction *ins)
2998 shader_addline(ins->ctx->buffer, "subroutine%u();\n", ins->src[0].reg.idx);
3001 static void shader_glsl_callnz(const struct wined3d_shader_instruction *ins)
3003 glsl_src_param_t src1_param;
3005 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0, &src1_param);
3006 shader_addline(ins->ctx->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, ins->src[0].reg.idx);
3009 static void shader_glsl_ret(const struct wined3d_shader_instruction *ins)
3011 /* No-op. The closing } is written when a new function is started, and at the end of the shader. This
3012 * function only suppresses the unhandled instruction warning
3016 /*********************************************
3017 * Pixel Shader Specific Code begins here
3018 ********************************************/
3019 static void shader_glsl_tex(const struct wined3d_shader_instruction *ins)
3021 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3022 IWineD3DDeviceImpl *device = shader->baseShader.device;
3023 DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
3024 ins->ctx->reg_maps->shader_version.minor);
3025 glsl_sample_function_t sample_function;
3026 IWineD3DBaseTextureImpl *texture;
3027 DWORD sample_flags = 0;
3029 DWORD mask = 0, swizzle;
3031 /* 1.0-1.4: Use destination register as sampler source.
3032 * 2.0+: Use provided sampler source. */
3033 if (shader_version < WINED3D_SHADER_VERSION(2,0)) sampler_idx = ins->dst[0].reg.idx;
3034 else sampler_idx = ins->src[1].reg.idx;
3035 texture = device->stateBlock->state.textures[sampler_idx];
3037 if (shader_version < WINED3D_SHADER_VERSION(1,4))
3039 const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
3040 DWORD flags = (priv->cur_ps_args->tex_transform >> sampler_idx * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
3041 & WINED3D_PSARGS_TEXTRANSFORM_MASK;
3042 WINED3DSAMPLER_TEXTURE_TYPE sampler_type = ins->ctx->reg_maps->sampler_type[sampler_idx];
3044 /* Projected cube textures don't make a lot of sense, the resulting coordinates stay the same. */
3045 if (flags & WINED3D_PSARGS_PROJECTED && sampler_type != WINED3DSTT_CUBE) {
3046 sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3047 switch (flags & ~WINED3D_PSARGS_PROJECTED) {
3048 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3049 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
3050 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
3051 case WINED3DTTFF_COUNT4:
3052 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
3056 else if (shader_version < WINED3D_SHADER_VERSION(2,0))
3058 DWORD src_mod = ins->src[0].modifiers;
3060 if (src_mod == WINED3DSPSM_DZ) {
3061 sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3062 mask = WINED3DSP_WRITEMASK_2;
3063 } else if (src_mod == WINED3DSPSM_DW) {
3064 sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3065 mask = WINED3DSP_WRITEMASK_3;
3068 if (ins->flags & WINED3DSI_TEXLD_PROJECT)
3070 /* ps 2.0 texldp instruction always divides by the fourth component. */
3071 sample_flags |= WINED3D_GLSL_SAMPLE_PROJECTED;
3072 mask = WINED3DSP_WRITEMASK_3;
3076 if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
3077 sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3079 shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3080 mask |= sample_function.coord_mask;
3082 if (shader_version < WINED3D_SHADER_VERSION(2,0)) swizzle = WINED3DSP_NOSWIZZLE;
3083 else swizzle = ins->src[1].swizzle;
3085 /* 1.0-1.3: Use destination register as coordinate source.
3086 1.4+: Use provided coordinate source register. */
3087 if (shader_version < WINED3D_SHADER_VERSION(1,4))
3090 shader_glsl_write_mask_to_str(mask, coord_mask);
3091 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3092 "T%u%s", sampler_idx, coord_mask);
3094 glsl_src_param_t coord_param;
3095 shader_glsl_add_src_param(ins, &ins->src[0], mask, &coord_param);
3096 if (ins->flags & WINED3DSI_TEXLD_BIAS)
3098 glsl_src_param_t bias;
3099 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &bias);
3100 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, bias.param_str,
3101 "%s", coord_param.param_str);
3103 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, NULL,
3104 "%s", coord_param.param_str);
3109 static void shader_glsl_texldd(const struct wined3d_shader_instruction *ins)
3111 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3112 IWineD3DDeviceImpl *device = This->baseShader.device;
3113 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3114 glsl_sample_function_t sample_function;
3115 glsl_src_param_t coord_param, dx_param, dy_param;
3116 DWORD sample_flags = WINED3D_GLSL_SAMPLE_GRAD;
3118 DWORD swizzle = ins->src[1].swizzle;
3119 IWineD3DBaseTextureImpl *texture;
3121 if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4])
3123 FIXME("texldd used, but not supported by hardware. Falling back to regular tex\n");
3124 shader_glsl_tex(ins);
3128 sampler_idx = ins->src[1].reg.idx;
3129 texture = device->stateBlock->state.textures[sampler_idx];
3130 if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
3131 sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3133 shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3134 shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3135 shader_glsl_add_src_param(ins, &ins->src[2], sample_function.coord_mask, &dx_param);
3136 shader_glsl_add_src_param(ins, &ins->src[3], sample_function.coord_mask, &dy_param);
3138 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, dx_param.param_str, dy_param.param_str, NULL,
3139 "%s", coord_param.param_str);
3142 static void shader_glsl_texldl(const struct wined3d_shader_instruction *ins)
3144 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *)ins->ctx->shader;
3145 IWineD3DDeviceImpl *device = This->baseShader.device;
3146 const struct wined3d_gl_info *gl_info = ins->ctx->gl_info;
3147 glsl_sample_function_t sample_function;
3148 glsl_src_param_t coord_param, lod_param;
3149 DWORD sample_flags = WINED3D_GLSL_SAMPLE_LOD;
3151 DWORD swizzle = ins->src[1].swizzle;
3152 IWineD3DBaseTextureImpl *texture;
3154 sampler_idx = ins->src[1].reg.idx;
3155 texture = device->stateBlock->state.textures[sampler_idx];
3156 if (texture && texture->baseTexture.target == GL_TEXTURE_RECTANGLE_ARB)
3157 sample_flags |= WINED3D_GLSL_SAMPLE_RECT;
3159 shader_glsl_get_sample_function(ins->ctx, sampler_idx, sample_flags, &sample_function);
3160 shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &coord_param);
3162 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &lod_param);
3164 if (!gl_info->supported[ARB_SHADER_TEXTURE_LOD] && !gl_info->supported[EXT_GPU_SHADER4]
3165 && shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
3167 /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
3168 * However, they seem to work just fine in fragment shaders as well. */
3169 WARN("Using %s in fragment shader.\n", sample_function.name);
3171 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, swizzle, NULL, NULL, lod_param.param_str,
3172 "%s", coord_param.param_str);
3175 static void shader_glsl_texcoord(const struct wined3d_shader_instruction *ins)
3177 /* FIXME: Make this work for more than just 2D textures */
3178 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3179 DWORD write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3181 if (!(ins->ctx->reg_maps->shader_version.major == 1 && ins->ctx->reg_maps->shader_version.minor == 4))
3185 shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3186 shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n",
3187 ins->dst[0].reg.idx, dst_mask);
3189 DWORD reg = ins->src[0].reg.idx;
3190 DWORD src_mod = ins->src[0].modifiers;
3191 char dst_swizzle[6];
3193 shader_glsl_get_swizzle(&ins->src[0], FALSE, write_mask, dst_swizzle);
3195 if (src_mod == WINED3DSPSM_DZ) {
3196 glsl_src_param_t div_param;
3197 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3198 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &div_param);
3200 if (mask_size > 1) {
3201 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3203 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3205 } else if (src_mod == WINED3DSPSM_DW) {
3206 glsl_src_param_t div_param;
3207 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
3208 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_3, &div_param);
3210 if (mask_size > 1) {
3211 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
3213 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
3216 shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
3221 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
3222 * Take a 3-component dot product of the TexCoord[dstreg] and src,
3223 * then perform a 1D texture lookup from stage dstregnum, place into dst. */
3224 static void shader_glsl_texdp3tex(const struct wined3d_shader_instruction *ins)
3226 glsl_src_param_t src0_param;
3227 glsl_sample_function_t sample_function;
3228 DWORD sampler_idx = ins->dst[0].reg.idx;
3229 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3232 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3234 /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
3235 * scalar, and projected sampling would require 4.
3237 * It is a dependent read - not valid with conditional NP2 textures
3239 shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3240 mask_size = shader_glsl_get_write_mask_size(sample_function.coord_mask);
3245 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3246 "dot(gl_TexCoord[%u].xyz, %s)", sampler_idx, src0_param.param_str);
3250 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3251 "vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0)", sampler_idx, src0_param.param_str);
3255 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3256 "vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0)", sampler_idx, src0_param.param_str);
3260 FIXME("Unexpected mask size %u\n", mask_size);
3265 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
3266 * Take a 3-component dot product of the TexCoord[dstreg] and src. */
3267 static void shader_glsl_texdp3(const struct wined3d_shader_instruction *ins)
3269 glsl_src_param_t src0_param;
3270 DWORD dstreg = ins->dst[0].reg.idx;
3271 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3273 unsigned int mask_size;
3275 dst_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3276 mask_size = shader_glsl_get_write_mask_size(dst_mask);
3277 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3279 if (mask_size > 1) {
3280 shader_addline(ins->ctx->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
3282 shader_addline(ins->ctx->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
3286 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
3287 * Calculate the depth as dst.x / dst.y */
3288 static void shader_glsl_texdepth(const struct wined3d_shader_instruction *ins)
3290 glsl_dst_param_t dst_param;
3292 shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3294 /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
3295 * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
3296 * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
3297 * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
3300 shader_addline(ins->ctx->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n",
3301 dst_param.reg_name, dst_param.reg_name);
3304 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
3305 * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
3306 * Calculate tmp0.y = TexCoord[dstreg] . src.xyz; (tmp0.x has already been calculated)
3307 * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
3309 static void shader_glsl_texm3x2depth(const struct wined3d_shader_instruction *ins)
3311 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3312 DWORD dstreg = ins->dst[0].reg.idx;
3313 glsl_src_param_t src0_param;
3315 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3317 shader_addline(ins->ctx->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
3318 shader_addline(ins->ctx->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
3321 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
3322 * Calculate the 1st of a 2-row matrix multiplication. */
3323 static void shader_glsl_texm3x2pad(const struct wined3d_shader_instruction *ins)
3325 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3326 DWORD reg = ins->dst[0].reg.idx;
3327 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3328 glsl_src_param_t src0_param;
3330 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3331 shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3334 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
3335 * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
3336 static void shader_glsl_texm3x3pad(const struct wined3d_shader_instruction *ins)
3338 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3339 DWORD reg = ins->dst[0].reg.idx;
3340 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3341 struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3342 glsl_src_param_t src0_param;
3344 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3345 shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + tex_mx->current_row, reg, src0_param.param_str);
3346 tex_mx->texcoord_w[tex_mx->current_row++] = reg;
3349 static void shader_glsl_texm3x2tex(const struct wined3d_shader_instruction *ins)
3351 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3352 DWORD reg = ins->dst[0].reg.idx;
3353 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3354 glsl_src_param_t src0_param;
3355 glsl_sample_function_t sample_function;
3357 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3358 shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3360 shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3362 /* Sample the texture using the calculated coordinates */
3363 shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xy");
3366 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
3367 * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
3368 static void shader_glsl_texm3x3tex(const struct wined3d_shader_instruction *ins)
3370 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3371 struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3372 glsl_src_param_t src0_param;
3373 DWORD reg = ins->dst[0].reg.idx;
3374 glsl_sample_function_t sample_function;
3376 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3377 shader_addline(ins->ctx->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3379 /* Dependent read, not valid with conditional NP2 */
3380 shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3382 /* Sample the texture using the calculated coordinates */
3383 shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3385 tex_mx->current_row = 0;
3388 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
3389 * Perform the 3rd row of a 3x3 matrix multiply */
3390 static void shader_glsl_texm3x3(const struct wined3d_shader_instruction *ins)
3392 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3393 struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3394 glsl_src_param_t src0_param;
3396 DWORD reg = ins->dst[0].reg.idx;
3398 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3400 shader_glsl_append_dst(ins->ctx->buffer, ins);
3401 shader_glsl_get_write_mask(&ins->dst[0], dst_mask);
3402 shader_addline(ins->ctx->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
3404 tex_mx->current_row = 0;
3407 /* Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
3408 * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3409 static void shader_glsl_texm3x3spec(const struct wined3d_shader_instruction *ins)
3411 DWORD reg = ins->dst[0].reg.idx;
3412 glsl_src_param_t src0_param;
3413 glsl_src_param_t src1_param;
3414 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3415 struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3416 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3417 glsl_sample_function_t sample_function;
3419 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3420 shader_glsl_add_src_param(ins, &ins->src[1], src_mask, &src1_param);
3422 /* Perform the last matrix multiply operation */
3423 shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
3424 /* Reflection calculation */
3425 shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
3427 /* Dependent read, not valid with conditional NP2 */
3428 shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3430 /* Sample the texture */
3431 shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3433 tex_mx->current_row = 0;
3436 /* Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
3437 * Perform the final texture lookup based on the previous 2 3x3 matrix multiplies */
3438 static void shader_glsl_texm3x3vspec(const struct wined3d_shader_instruction *ins)
3440 DWORD reg = ins->dst[0].reg.idx;
3441 struct wined3d_shader_buffer *buffer = ins->ctx->buffer;
3442 struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
3443 glsl_src_param_t src0_param;
3444 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
3445 glsl_sample_function_t sample_function;
3447 shader_glsl_add_src_param(ins, &ins->src[0], src_mask, &src0_param);
3449 /* Perform the last matrix multiply operation */
3450 shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
3452 /* Construct the eye-ray vector from w coordinates */
3453 shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
3454 tex_mx->texcoord_w[0], tex_mx->texcoord_w[1], reg);
3455 shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
3457 /* Dependent read, not valid with conditional NP2 */
3458 shader_glsl_get_sample_function(ins->ctx, reg, 0, &sample_function);
3460 /* Sample the texture using the calculated coordinates */
3461 shader_glsl_gen_sample_code(ins, reg, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL, "tmp0.xyz");
3463 tex_mx->current_row = 0;
3466 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
3467 * Apply a fake bump map transform.
3468 * texbem is pshader <= 1.3 only, this saves a few version checks
3470 static void shader_glsl_texbem(const struct wined3d_shader_instruction *ins)
3472 const struct shader_glsl_ctx_priv *priv = ins->ctx->backend_data;
3473 glsl_sample_function_t sample_function;
3474 glsl_src_param_t coord_param;
3480 sampler_idx = ins->dst[0].reg.idx;
3481 flags = (priv->cur_ps_args->tex_transform >> sampler_idx * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
3482 & WINED3D_PSARGS_TEXTRANSFORM_MASK;
3484 /* Dependent read, not valid with conditional NP2 */
3485 shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3486 mask = sample_function.coord_mask;
3488 shader_glsl_write_mask_to_str(mask, coord_mask);
3490 /* with projective textures, texbem only divides the static texture coord, not the displacement,
3491 * so we can't let the GL handle this.
3493 if (flags & WINED3D_PSARGS_PROJECTED) {
3495 char coord_div_mask[3];
3496 switch (flags & ~WINED3D_PSARGS_PROJECTED) {
3497 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
3498 case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
3499 case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
3500 case WINED3DTTFF_COUNT4:
3501 case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
3503 shader_glsl_write_mask_to_str(div_mask, coord_div_mask);
3504 shader_addline(ins->ctx->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
3507 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &coord_param);
3509 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3510 "T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s", sampler_idx, coord_mask, sampler_idx,
3511 coord_param.param_str, coord_mask);
3513 if (ins->handler_idx == WINED3DSIH_TEXBEML)
3515 glsl_src_param_t luminance_param;
3516 glsl_dst_param_t dst_param;
3518 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_2, &luminance_param);
3519 shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3521 shader_addline(ins->ctx->buffer, "%s%s *= (%s * luminancescale%d + luminanceoffset%d);\n",
3522 dst_param.reg_name, dst_param.mask_str,
3523 luminance_param.param_str, sampler_idx, sampler_idx);
3527 static void shader_glsl_bem(const struct wined3d_shader_instruction *ins)
3529 glsl_src_param_t src0_param, src1_param;
3530 DWORD sampler_idx = ins->dst[0].reg.idx;
3532 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3533 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3535 shader_glsl_append_dst(ins->ctx->buffer, ins);
3536 shader_addline(ins->ctx->buffer, "%s + bumpenvmat%d * %s);\n",
3537 src0_param.param_str, sampler_idx, src1_param.param_str);
3540 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
3541 * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
3542 static void shader_glsl_texreg2ar(const struct wined3d_shader_instruction *ins)
3544 glsl_src_param_t src0_param;
3545 DWORD sampler_idx = ins->dst[0].reg.idx;
3546 glsl_sample_function_t sample_function;
3548 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3550 shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3551 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3552 "%s.wx", src0_param.reg_name);
3555 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
3556 * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
3557 static void shader_glsl_texreg2gb(const struct wined3d_shader_instruction *ins)
3559 glsl_src_param_t src0_param;
3560 DWORD sampler_idx = ins->dst[0].reg.idx;
3561 glsl_sample_function_t sample_function;
3563 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
3565 shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3566 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3567 "%s.yz", src0_param.reg_name);
3570 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
3571 * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
3572 static void shader_glsl_texreg2rgb(const struct wined3d_shader_instruction *ins)
3574 glsl_src_param_t src0_param;
3575 DWORD sampler_idx = ins->dst[0].reg.idx;
3576 glsl_sample_function_t sample_function;
3578 /* Dependent read, not valid with conditional NP2 */
3579 shader_glsl_get_sample_function(ins->ctx, sampler_idx, 0, &sample_function);
3580 shader_glsl_add_src_param(ins, &ins->src[0], sample_function.coord_mask, &src0_param);
3582 shader_glsl_gen_sample_code(ins, sampler_idx, &sample_function, WINED3DSP_NOSWIZZLE, NULL, NULL, NULL,
3583 "%s", src0_param.param_str);
3586 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
3587 * If any of the first 3 components are < 0, discard this pixel */
3588 static void shader_glsl_texkill(const struct wined3d_shader_instruction *ins)
3590 glsl_dst_param_t dst_param;
3592 /* The argument is a destination parameter, and no writemasks are allowed */
3593 shader_glsl_add_dst_param(ins, &ins->dst[0], &dst_param);
3594 if (ins->ctx->reg_maps->shader_version.major >= 2)
3596 /* 2.0 shaders compare all 4 components in texkill */
3597 shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
3599 /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
3600 * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
3601 * 4 components are defined, only the first 3 are used
3603 shader_addline(ins->ctx->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
3607 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
3608 * dst = dot2(src0, src1) + src2 */
3609 static void shader_glsl_dp2add(const struct wined3d_shader_instruction *ins)
3611 glsl_src_param_t src0_param;
3612 glsl_src_param_t src1_param;
3613 glsl_src_param_t src2_param;
3615 unsigned int mask_size;
3617 write_mask = shader_glsl_append_dst(ins->ctx->buffer, ins);
3618 mask_size = shader_glsl_get_write_mask_size(write_mask);
3620 shader_glsl_add_src_param(ins, &ins->src[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
3621 shader_glsl_add_src_param(ins, &ins->src[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
3622 shader_glsl_add_src_param(ins, &ins->src[2], WINED3DSP_WRITEMASK_0, &src2_param);
3624 if (mask_size > 1) {
3625 shader_addline(ins->ctx->buffer, "vec%d(dot(%s, %s) + %s));\n",
3626 mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
3628 shader_addline(ins->ctx->buffer, "dot(%s, %s) + %s);\n",
3629 src0_param.param_str, src1_param.param_str, src2_param.param_str);
3633 static void shader_glsl_input_pack(IWineD3DPixelShader *iface, struct wined3d_shader_buffer *buffer,
3634 const struct wined3d_shader_signature_element *input_signature, const struct wined3d_shader_reg_maps *reg_maps,
3635 enum vertexprocessing_mode vertexprocessing)
3638 IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
3639 WORD map = reg_maps->input_registers;
3641 for (i = 0; map; map >>= 1, ++i)
3643 const char *semantic_name;
3648 if (!(map & 1)) continue;
3650 semantic_name = input_signature[i].semantic_name;
3651 semantic_idx = input_signature[i].semantic_idx;
3652 shader_glsl_write_mask_to_str(input_signature[i].mask, reg_mask);
3654 if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3656 if (semantic_idx < 8 && vertexprocessing == pretransformed)
3657 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
3658 This->input_reg_map[i], reg_mask, semantic_idx, reg_mask);
3660 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3661 This->input_reg_map[i], reg_mask, reg_mask);
3663 else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3666 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
3667 This->input_reg_map[i], reg_mask, reg_mask);
3668 else if (semantic_idx == 1)
3669 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
3670 This->input_reg_map[i], reg_mask, reg_mask);
3672 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3673 This->input_reg_map[i], reg_mask, reg_mask);
3677 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
3678 This->input_reg_map[i], reg_mask, reg_mask);
3683 /*********************************************
3684 * Vertex Shader Specific Code begins here
3685 ********************************************/
3687 static void add_glsl_program_entry(struct shader_glsl_priv *priv, struct glsl_shader_prog_link *entry) {
3688 glsl_program_key_t key;
3690 key.vshader = entry->vshader;
3691 key.pshader = entry->pshader;
3692 key.vs_args = entry->vs_args;
3693 key.ps_args = entry->ps_args;
3695 if (wine_rb_put(&priv->program_lookup, &key, &entry->program_lookup_entry) == -1)
3697 ERR("Failed to insert program entry.\n");
3701 static struct glsl_shader_prog_link *get_glsl_program_entry(struct shader_glsl_priv *priv,
3702 IWineD3DVertexShader *vshader, IWineD3DPixelShader *pshader, struct vs_compile_args *vs_args,
3703 struct ps_compile_args *ps_args) {
3704 struct wine_rb_entry *entry;
3705 glsl_program_key_t key;
3707 key.vshader = vshader;
3708 key.pshader = pshader;
3709 key.vs_args = *vs_args;
3710 key.ps_args = *ps_args;
3712 entry = wine_rb_get(&priv->program_lookup, &key);
3713 return entry ? WINE_RB_ENTRY_VALUE(entry, struct glsl_shader_prog_link, program_lookup_entry) : NULL;
3716 /* GL locking is done by the caller */
3717 static void delete_glsl_program_entry(struct shader_glsl_priv *priv, const struct wined3d_gl_info *gl_info,
3718 struct glsl_shader_prog_link *entry)
3720 glsl_program_key_t key;
3722 key.vshader = entry->vshader;
3723 key.pshader = entry->pshader;
3724 key.vs_args = entry->vs_args;
3725 key.ps_args = entry->ps_args;
3726 wine_rb_remove(&priv->program_lookup, &key);
3728 GL_EXTCALL(glDeleteObjectARB(entry->programId));
3729 if (entry->vshader) list_remove(&entry->vshader_entry);
3730 if (entry->pshader) list_remove(&entry->pshader_entry);
3731 HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
3732 HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
3733 HeapFree(GetProcessHeap(), 0, entry);
3736 static void handle_ps3_input(struct wined3d_shader_buffer *buffer,
3737 const struct wined3d_gl_info *gl_info, const DWORD *map,
3738 const struct wined3d_shader_signature_element *input_signature,
3739 const struct wined3d_shader_reg_maps *reg_maps_in,
3740 const struct wined3d_shader_signature_element *output_signature,
3741 const struct wined3d_shader_reg_maps *reg_maps_out)
3744 const char *semantic_name_in;
3745 UINT semantic_idx_in;
3748 unsigned int in_count = vec4_varyings(3, gl_info);
3750 char destination[50];
3751 WORD input_map, output_map;
3753 set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
3755 input_map = reg_maps_in->input_registers;
3756 for (i = 0; input_map; input_map >>= 1, ++i)
3758 if (!(input_map & 1)) continue;
3761 /* Declared, but not read register */
3762 if (in_idx == ~0U) continue;
3763 if (in_idx >= (in_count + 2))
3765 FIXME("More input varyings declared than supported, expect issues.\n");
3769 if (in_idx == in_count) {
3770 sprintf(destination, "gl_FrontColor");
3771 } else if (in_idx == in_count + 1) {
3772 sprintf(destination, "gl_FrontSecondaryColor");
3774 sprintf(destination, "IN[%u]", in_idx);
3777 semantic_name_in = input_signature[i].semantic_name;
3778 semantic_idx_in = input_signature[i].semantic_idx;
3781 output_map = reg_maps_out->output_registers;
3782 for (j = 0; output_map; output_map >>= 1, ++j)
3786 if (!(output_map & 1)
3787 || semantic_idx_in != output_signature[j].semantic_idx
3788 || strcmp(semantic_name_in, output_signature[j].semantic_name)
3789 || !(mask = input_signature[i].mask & output_signature[j].mask))
3793 shader_glsl_write_mask_to_str(mask, reg_mask);
3795 shader_addline(buffer, "%s%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3796 destination, reg_mask, j, reg_mask);
3800 for (i = 0; i < in_count + 2; ++i)
3804 if (!set[i] || set[i] == WINED3DSP_WRITEMASK_ALL)
3807 if (set[i] == ~0U) set[i] = 0;
3810 if (!(set[i] & WINED3DSP_WRITEMASK_0)) reg_mask[size++] = 'x';
3811 if (!(set[i] & WINED3DSP_WRITEMASK_1)) reg_mask[size++] = 'y';
3812 if (!(set[i] & WINED3DSP_WRITEMASK_2)) reg_mask[size++] = 'z';
3813 if (!(set[i] & WINED3DSP_WRITEMASK_3)) reg_mask[size++] = 'w';
3814 reg_mask[size] = '\0';
3816 if (i == in_count) sprintf(destination, "gl_FrontColor");
3817 else if (i == in_count + 1) sprintf(destination, "gl_FrontSecondaryColor");
3818 else sprintf(destination, "IN[%u]", i);
3820 if (size == 1) shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
3821 else shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
3824 HeapFree(GetProcessHeap(), 0, set);
3827 /* GL locking is done by the caller */
3828 static GLhandleARB generate_param_reorder_function(struct wined3d_shader_buffer *buffer,
3829 IWineD3DVertexShader *vertexshader, IWineD3DPixelShader *pixelshader, const struct wined3d_gl_info *gl_info)
3831 GLhandleARB ret = 0;
3832 IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) vertexshader;
3833 IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pixelshader;
3834 DWORD ps_major = ps ? ps->baseShader.reg_maps.shader_version.major : 0;
3836 const char *semantic_name;
3839 const struct wined3d_shader_signature_element *output_signature = vs->baseShader.output_signature;
3840 WORD map = vs->baseShader.reg_maps.output_registers;
3842 shader_buffer_clear(buffer);
3844 shader_addline(buffer, "#version 120\n");
3845 shader_addline(buffer, "const float FLT_MAX = 1e38;\n");
3849 shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3851 for (i = 0; map; map >>= 1, ++i)
3855 if (!(map & 1)) continue;
3857 semantic_name = output_signature[i].semantic_name;
3858 semantic_idx = output_signature[i].semantic_idx;
3859 write_mask = output_signature[i].mask;
3860 shader_glsl_write_mask_to_str(write_mask, reg_mask);
3862 if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_COLOR))
3865 shader_addline(buffer, "gl_FrontColor%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3866 reg_mask, i, reg_mask);
3867 else if (semantic_idx == 1)
3868 shader_addline(buffer, "gl_FrontSecondaryColor%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3869 reg_mask, i, reg_mask);
3871 else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3873 shader_addline(buffer, "gl_Position%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3874 reg_mask, i, reg_mask);
3876 else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_TEXCOORD))
3878 if (semantic_idx < 8)
3880 if (!(gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) || ps_major > 0)
3881 write_mask |= WINED3DSP_WRITEMASK_3;
3883 shader_addline(buffer, "gl_TexCoord[%u]%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3884 semantic_idx, reg_mask, i, reg_mask);
3885 if (!(write_mask & WINED3DSP_WRITEMASK_3))
3886 shader_addline(buffer, "gl_TexCoord[%u].w = 1.0;\n", semantic_idx);
3889 else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3891 shader_addline(buffer, "gl_PointSize = clamp(OUT[%u].%c, -FLT_MAX, FLT_MAX);\n", i, reg_mask[1]);
3893 else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_FOG))
3895 shader_addline(buffer, "gl_FogFragCoord = clamp(OUT[%u].%c, -FLT_MAX, FLT_MAX);\n", i, reg_mask[1]);
3898 shader_addline(buffer, "}\n");
3903 /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3904 shader_addline(buffer, "varying vec4 IN[%u];\n", vec4_varyings(3, gl_info));
3905 shader_addline(buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3907 /* First, sort out position and point size. Those are not passed to the pixel shader */
3908 for (i = 0; map; map >>= 1, ++i)
3910 if (!(map & 1)) continue;
3912 semantic_name = output_signature[i].semantic_name;
3913 shader_glsl_write_mask_to_str(output_signature[i].mask, reg_mask);
3915 if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_POSITION))
3917 shader_addline(buffer, "gl_Position%s = clamp(OUT[%u]%s, -FLT_MAX, FLT_MAX);\n",
3918 reg_mask, i, reg_mask);
3920 else if (shader_match_semantic(semantic_name, WINED3DDECLUSAGE_PSIZE))
3922 shader_addline(buffer, "gl_PointSize = clamp(OUT[%u].%c, -FLT_MAX, FLT_MAX);\n", i, reg_mask[1]);
3926 /* Then, fix the pixel shader input */
3927 handle_ps3_input(buffer, gl_info, ps->input_reg_map, ps->baseShader.input_signature,
3928 &ps->baseShader.reg_maps, output_signature, &vs->baseShader.reg_maps);
3930 shader_addline(buffer, "}\n");
3933 ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3934 checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3935 shader_glsl_compile(gl_info, ret, buffer->buffer);
3940 /* GL locking is done by the caller */
3941 static void hardcode_local_constants(IWineD3DBaseShaderImpl *shader, const struct wined3d_gl_info *gl_info,
3942 GLhandleARB programId, char prefix)
3944 const local_constant *lconst;
3949 LIST_FOR_EACH_ENTRY(lconst, &shader->baseShader.constantsF, local_constant, entry) {
3950 value = (const float *)lconst->value;
3951 snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
3952 tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3953 GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
3955 checkGLcall("Hardcoding local constants");
3958 /* GL locking is done by the caller */
3959 static GLuint shader_glsl_generate_pshader(const struct wined3d_context *context,
3960 struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *This,
3961 const struct ps_compile_args *args, struct ps_np2fixup_info *np2fixup_info)
3963 const struct wined3d_shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
3964 const struct wined3d_gl_info *gl_info = context->gl_info;
3965 CONST DWORD *function = This->baseShader.function;
3966 struct shader_glsl_ctx_priv priv_ctx;
3968 /* Create the hw GLSL shader object and assign it as the shader->prgId */
3969 GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3971 memset(&priv_ctx, 0, sizeof(priv_ctx));
3972 priv_ctx.cur_ps_args = args;
3973 priv_ctx.cur_np2fixup_info = np2fixup_info;
3975 shader_addline(buffer, "#version 120\n");
3977 if (gl_info->supported[ARB_SHADER_TEXTURE_LOD] && reg_maps->usestexldd)
3979 shader_addline(buffer, "#extension GL_ARB_shader_texture_lod : enable\n");
3981 if (gl_info->supported[ARB_TEXTURE_RECTANGLE])
3983 /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
3984 * drivers write a warning if we don't do so
3986 shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
3988 if (gl_info->supported[EXT_GPU_SHADER4])
3990 shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
3993 /* Base Declarations */
3994 shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShader *)This, reg_maps, &priv_ctx);
3996 /* Pack 3.0 inputs */
3997 if (reg_maps->shader_version.major >= 3 && args->vp_mode != vertexshader)
3999 shader_glsl_input_pack((IWineD3DPixelShader *) This, buffer,
4000 This->baseShader.input_signature, reg_maps, args->vp_mode);
4003 /* Base Shader Body */
4004 shader_generate_main((IWineD3DBaseShader *)This, buffer, reg_maps, function, &priv_ctx);
4006 /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
4007 if (reg_maps->shader_version.major < 2)
4009 /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
4010 shader_addline(buffer, "gl_FragData[0] = R0;\n");
4013 if (args->srgb_correction)
4015 shader_addline(buffer, "tmp0.xyz = pow(gl_FragData[0].xyz, vec3(srgb_const0.x));\n");
4016 shader_addline(buffer, "tmp0.xyz = tmp0.xyz * vec3(srgb_const0.y) - vec3(srgb_const0.z);\n");
4017 shader_addline(buffer, "tmp1.xyz = gl_FragData[0].xyz * vec3(srgb_const0.w);\n");
4018 shader_addline(buffer, "bvec3 srgb_compare = lessThan(gl_FragData[0].xyz, vec3(srgb_const1.x));\n");
4019 shader_addline(buffer, "gl_FragData[0].xyz = mix(tmp0.xyz, tmp1.xyz, vec3(srgb_compare));\n");
4020 shader_addline(buffer, "gl_FragData[0] = clamp(gl_FragData[0], 0.0, 1.0);\n");
4022 /* Pixel shader < 3.0 do not replace the fog stage.
4023 * This implements linear fog computation and blending.
4024 * TODO: non linear fog
4025 * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
4026 * -1/(e-s) and e/(e-s) respectively.
4028 if (reg_maps->shader_version.major < 3)
4031 case FOG_OFF: break;
4033 shader_addline(buffer, "float fogstart = -1.0 / (gl_Fog.end - gl_Fog.start);\n");
4034 shader_addline(buffer, "float fogend = gl_Fog.end * -fogstart;\n");
4035 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * fogstart + fogend, 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) */
4040 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * 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");
4045 /* Fog = e^(-(gl_Fog.density * gl_FogFragCoord)^2) */
4046 shader_addline(buffer, "float Fog = exp(-gl_Fog.density * gl_Fog.density * gl_FogFragCoord * gl_FogFragCoord);\n");
4047 shader_addline(buffer, "Fog = clamp(Fog, 0.0, 1.0);\n");
4048 shader_addline(buffer, "gl_FragData[0].xyz = mix(gl_Fog.color.xyz, gl_FragData[0].xyz, Fog);\n");
4053 shader_addline(buffer, "}\n");
4055 TRACE("Compiling shader object %u\n", shader_obj);
4056 shader_glsl_compile(gl_info, shader_obj, buffer->buffer);
4058 /* Store the shader object */
4062 /* GL locking is done by the caller */
4063 static GLuint shader_glsl_generate_vshader(const struct wined3d_context *context,
4064 struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *This,
4065 const struct vs_compile_args *args)
4067 const struct wined3d_shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
4068 const struct wined3d_gl_info *gl_info = context->gl_info;
4069 CONST DWORD *function = This->baseShader.function;
4070 struct shader_glsl_ctx_priv priv_ctx;
4072 /* Create the hw GLSL shader program and assign it as the shader->prgId */
4073 GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4075 shader_addline(buffer, "#version 120\n");
4077 if (gl_info->supported[EXT_GPU_SHADER4])
4079 shader_addline(buffer, "#extension GL_EXT_gpu_shader4 : enable\n");
4082 memset(&priv_ctx, 0, sizeof(priv_ctx));
4083 priv_ctx.cur_vs_args = args;
4085 /* Base Declarations */
4086 shader_generate_glsl_declarations(context, buffer, (IWineD3DBaseShader *)This, reg_maps, &priv_ctx);
4088 /* Base Shader Body */
4089 shader_generate_main((IWineD3DBaseShader*)This, buffer, reg_maps, function, &priv_ctx);
4091 /* Unpack outputs */
4092 shader_addline(buffer, "order_ps_input(OUT);\n");
4094 /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
4095 * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
4096 * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
4097 * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
4099 if(args->fog_src == VS_FOG_Z) {
4100 shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
4101 } else if (!reg_maps->fog) {
4102 shader_addline(buffer, "gl_FogFragCoord = 0.0;\n");
4105 /* Write the final position.
4107 * OpenGL coordinates specify the center of the pixel while d3d coords specify
4108 * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
4109 * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
4110 * contains 1.0 to allow a mad.
4112 shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
4113 shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
4114 if(args->clip_enabled) {
4115 shader_addline(buffer, "gl_ClipVertex = gl_Position;\n");
4118 /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
4120 * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
4121 * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
4122 * which is the same as z = z * 2 - w.
4124 shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
4126 shader_addline(buffer, "}\n");
4128 TRACE("Compiling shader object %u\n", shader_obj);
4129 shader_glsl_compile(gl_info, shader_obj, buffer->buffer);
4134 static GLhandleARB find_glsl_pshader(const struct wined3d_context *context,
4135 struct wined3d_shader_buffer *buffer, IWineD3DPixelShaderImpl *shader,
4136 const struct ps_compile_args *args, const struct ps_np2fixup_info **np2fixup_info)
4138 struct wined3d_state *state = &shader->baseShader.device->stateBlock->state;
4141 struct glsl_ps_compiled_shader *new_array;
4142 struct glsl_pshader_private *shader_data;
4143 struct ps_np2fixup_info *np2fixup = NULL;
4146 if (!shader->baseShader.backend_data)
4148 shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4149 if (!shader->baseShader.backend_data)
4151 ERR("Failed to allocate backend data.\n");
4155 shader_data = shader->baseShader.backend_data;
4157 /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4158 * so a linear search is more performant than a hashmap or a binary search
4159 * (cache coherency etc)
4161 for (i = 0; i < shader_data->num_gl_shaders; ++i)
4163 if (!memcmp(&shader_data->gl_shaders[i].args, args, sizeof(*args)))
4165 if (args->np2_fixup) *np2fixup_info = &shader_data->gl_shaders[i].np2fixup;
4166 return shader_data->gl_shaders[i].prgId;
4170 TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4171 if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4172 if (shader_data->num_gl_shaders)
4174 new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4175 new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4176 new_size * sizeof(*shader_data->gl_shaders));
4178 new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4183 ERR("Out of memory\n");
4186 shader_data->gl_shaders = new_array;
4187 shader_data->shader_array_size = new_size;
4190 shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4192 memset(&shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup, 0, sizeof(struct ps_np2fixup_info));
4193 if (args->np2_fixup) np2fixup = &shader_data->gl_shaders[shader_data->num_gl_shaders].np2fixup;
4195 pixelshader_update_samplers(&shader->baseShader.reg_maps, (IWineD3DBaseTexture **)state->textures);
4197 shader_buffer_clear(buffer);
4198 ret = shader_glsl_generate_pshader(context, buffer, shader, args, np2fixup);
4199 shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4200 *np2fixup_info = np2fixup;
4205 static inline BOOL vs_args_equal(const struct vs_compile_args *stored, const struct vs_compile_args *new,
4206 const DWORD use_map) {
4207 if((stored->swizzle_map & use_map) != new->swizzle_map) return FALSE;
4208 if((stored->clip_enabled) != new->clip_enabled) return FALSE;
4209 return stored->fog_src == new->fog_src;
4212 static GLhandleARB find_glsl_vshader(const struct wined3d_context *context,
4213 struct wined3d_shader_buffer *buffer, IWineD3DVertexShaderImpl *shader,
4214 const struct vs_compile_args *args)
4218 struct glsl_vs_compiled_shader *new_array;
4219 DWORD use_map = shader->baseShader.device->strided_streams.use_map;
4220 struct glsl_vshader_private *shader_data;
4223 if (!shader->baseShader.backend_data)
4225 shader->baseShader.backend_data = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*shader_data));
4226 if (!shader->baseShader.backend_data)
4228 ERR("Failed to allocate backend data.\n");
4232 shader_data = shader->baseShader.backend_data;
4234 /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
4235 * so a linear search is more performant than a hashmap or a binary search
4236 * (cache coherency etc)
4238 for(i = 0; i < shader_data->num_gl_shaders; i++) {
4239 if(vs_args_equal(&shader_data->gl_shaders[i].args, args, use_map)) {
4240 return shader_data->gl_shaders[i].prgId;
4244 TRACE("No matching GL shader found for shader %p, compiling a new shader.\n", shader);
4246 if(shader_data->shader_array_size == shader_data->num_gl_shaders) {
4247 if (shader_data->num_gl_shaders)
4249 new_size = shader_data->shader_array_size + max(1, shader_data->shader_array_size / 2);
4250 new_array = HeapReAlloc(GetProcessHeap(), 0, shader_data->gl_shaders,
4251 new_size * sizeof(*shader_data->gl_shaders));
4253 new_array = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader_data->gl_shaders));
4258 ERR("Out of memory\n");
4261 shader_data->gl_shaders = new_array;
4262 shader_data->shader_array_size = new_size;
4265 shader_data->gl_shaders[shader_data->num_gl_shaders].args = *args;
4267 shader_buffer_clear(buffer);
4268 ret = shader_glsl_generate_vshader(context, buffer, shader, args);
4269 shader_data->gl_shaders[shader_data->num_gl_shaders++].prgId = ret;
4274 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
4275 * It sets the programId on the current StateBlock (because it should be called
4276 * inside of the DrawPrimitive() part of the render loop).
4278 * If a program for the given combination does not exist, create one, and store
4279 * the program in the hash table. If it creates a program, it will link the
4280 * given objects, too.
4283 /* GL locking is done by the caller */
4284 static void set_glsl_shader_program(const struct wined3d_context *context,
4285 IWineD3DDeviceImpl *device, BOOL use_ps, BOOL use_vs)
4287 const struct wined3d_state *state = &device->stateBlock->state;
4288 IWineD3DVertexShader *vshader = use_vs ? (IWineD3DVertexShader *)state->vertex_shader : NULL;
4289 IWineD3DPixelShader *pshader = use_ps ? (IWineD3DPixelShader *)state->pixel_shader : NULL;
4290 const struct wined3d_gl_info *gl_info = context->gl_info;
4291 struct shader_glsl_priv *priv = device->shader_priv;
4292 struct glsl_shader_prog_link *entry = NULL;
4293 GLhandleARB programId = 0;
4294 GLhandleARB reorder_shader_id = 0;
4297 struct ps_compile_args ps_compile_args;
4298 struct vs_compile_args vs_compile_args;
4300 if (vshader) find_vs_compile_args(state, (IWineD3DVertexShaderImpl *)vshader, &vs_compile_args);
4301 if (pshader) find_ps_compile_args(state, (IWineD3DPixelShaderImpl *)pshader, &ps_compile_args);
4303 entry = get_glsl_program_entry(priv, vshader, pshader, &vs_compile_args, &ps_compile_args);
4305 priv->glsl_program = entry;
4309 /* If we get to this point, then no matching program exists, so we create one */
4310 programId = GL_EXTCALL(glCreateProgramObjectARB());
4311 TRACE("Created new GLSL shader program %u\n", programId);
4313 /* Create the entry */
4314 entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
4315 entry->programId = programId;
4316 entry->vshader = vshader;
4317 entry->pshader = pshader;
4318 entry->vs_args = vs_compile_args;
4319 entry->ps_args = ps_compile_args;
4320 entry->constant_version = 0;
4321 entry->np2Fixup_info = NULL;
4322 /* Add the hash table entry */
4323 add_glsl_program_entry(priv, entry);
4325 /* Set the current program */
4326 priv->glsl_program = entry;
4328 /* Attach GLSL vshader */
4331 GLhandleARB vshader_id = find_glsl_vshader(context, &priv->shader_buffer,
4332 (IWineD3DVertexShaderImpl *)vshader, &vs_compile_args);
4333 WORD map = ((IWineD3DBaseShaderImpl *)vshader)->baseShader.reg_maps.input_registers;
4336 reorder_shader_id = generate_param_reorder_function(&priv->shader_buffer, vshader, pshader, gl_info);
4337 TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
4338 GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
4339 checkGLcall("glAttachObjectARB");
4340 /* Flag the reorder function for deletion, then it will be freed automatically when the program
4343 GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
4345 TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
4346 GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
4347 checkGLcall("glAttachObjectARB");
4349 /* Bind vertex attributes to a corresponding index number to match
4350 * the same index numbers as ARB_vertex_programs (makes loading
4351 * vertex attributes simpler). With this method, we can use the
4352 * exact same code to load the attributes later for both ARB and
4355 * We have to do this here because we need to know the Program ID
4356 * in order to make the bindings work, and it has to be done prior
4357 * to linking the GLSL program. */
4358 for (i = 0; map; map >>= 1, ++i)
4360 if (!(map & 1)) continue;
4362 snprintf(tmp_name, sizeof(tmp_name), "attrib%u", i);
4363 GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
4365 checkGLcall("glBindAttribLocationARB");
4367 list_add_head(&((IWineD3DBaseShaderImpl *)vshader)->baseShader.linked_programs, &entry->vshader_entry);
4370 /* Attach GLSL pshader */
4373 GLhandleARB pshader_id = find_glsl_pshader(context, &priv->shader_buffer,
4374 (IWineD3DPixelShaderImpl *)pshader, &ps_compile_args, &entry->np2Fixup_info);
4375 TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
4376 GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
4377 checkGLcall("glAttachObjectARB");
4379 list_add_head(&((IWineD3DBaseShaderImpl *)pshader)->baseShader.linked_programs, &entry->pshader_entry);
4382 /* Link the program */
4383 TRACE("Linking GLSL shader program %u\n", programId);
4384 GL_EXTCALL(glLinkProgramARB(programId));
4385 shader_glsl_validate_link(gl_info, programId);
4387 entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4388 sizeof(GLhandleARB) * gl_info->limits.glsl_vs_float_constants);
4389 for (i = 0; i < gl_info->limits.glsl_vs_float_constants; ++i)
4391 snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
4392 entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4394 for (i = 0; i < MAX_CONST_I; ++i)
4396 snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
4397 entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4399 entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0,
4400 sizeof(GLhandleARB) * gl_info->limits.glsl_ps_float_constants);
4401 for (i = 0; i < gl_info->limits.glsl_ps_float_constants; ++i)
4403 snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
4404 entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4406 for (i = 0; i < MAX_CONST_I; ++i)
4408 snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
4409 entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
4415 for(i = 0; i < MAX_TEXTURES; i++) {
4416 sprintf(name, "bumpenvmat%u", i);
4417 entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4418 sprintf(name, "luminancescale%u", i);
4419 entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4420 sprintf(name, "luminanceoffset%u", i);
4421 entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
4424 if (ps_compile_args.np2_fixup) {
4425 if (entry->np2Fixup_info) {
4426 entry->np2Fixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "PsamplerNP2Fixup"));
4428 FIXME("NP2 texcoord fixup needed for this pixelshader, but no fixup uniform found.\n");
4433 entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
4434 entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
4435 checkGLcall("Find glsl program uniform locations");
4438 && ((IWineD3DPixelShaderImpl *)pshader)->baseShader.reg_maps.shader_version.major >= 3
4439 && ((IWineD3DPixelShaderImpl *)pshader)->declared_in_count > vec4_varyings(3, gl_info))
4441 TRACE("Shader %d needs vertex color clamping disabled\n", programId);
4442 entry->vertex_color_clamp = GL_FALSE;
4444 entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
4447 /* Set the shader to allow uniform loading on it */
4448 GL_EXTCALL(glUseProgramObjectARB(programId));
4449 checkGLcall("glUseProgramObjectARB(programId)");
4451 /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
4452 * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
4453 * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
4454 * vertex shader with fixed function pixel processing is used we make sure that the card
4455 * supports enough samplers to allow the max number of vertex samplers with all possible
4456 * fixed function fragment processing setups. So once the program is linked these samplers
4459 if (vshader) shader_glsl_load_vsamplers(gl_info, device->texUnitMap, programId);
4460 if (pshader) shader_glsl_load_psamplers(gl_info, device->texUnitMap, programId);
4462 /* If the local constants do not have to be loaded with the environment constants,
4463 * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
4466 if (pshader && !((IWineD3DBaseShaderImpl *)pshader)->baseShader.load_local_constsF)
4468 hardcode_local_constants((IWineD3DBaseShaderImpl *) pshader, gl_info, programId, 'P');
4470 if (vshader && !((IWineD3DBaseShaderImpl *)vshader)->baseShader.load_local_constsF)
4472 hardcode_local_constants((IWineD3DBaseShaderImpl *) vshader, gl_info, programId, 'V');
4476 /* GL locking is done by the caller */
4477 static GLhandleARB create_glsl_blt_shader(const struct wined3d_gl_info *gl_info, enum tex_types tex_type, BOOL masked)
4479 GLhandleARB program_id;
4480 GLhandleARB vshader_id, pshader_id;
4481 const char *blt_pshader;
4483 static const char *blt_vshader =
4487 " gl_Position = gl_Vertex;\n"
4488 " gl_FrontColor = vec4(1.0);\n"
4489 " gl_TexCoord[0] = gl_MultiTexCoord0;\n"
4492 static const char * const blt_pshaders_full[tex_type_count] =
4498 "uniform sampler2D sampler;\n"
4501 " gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4507 "uniform samplerCube sampler;\n"
4510 " gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4514 "#extension GL_ARB_texture_rectangle : enable\n"
4515 "uniform sampler2DRect sampler;\n"
4518 " gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4522 static const char * const blt_pshaders_masked[tex_type_count] =
4528 "uniform sampler2D sampler;\n"
4529 "uniform vec4 mask;\n"
4532 " if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4533 " gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
4539 "uniform samplerCube sampler;\n"
4540 "uniform vec4 mask;\n"
4543 " if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4544 " gl_FragDepth = textureCube(sampler, gl_TexCoord[0].xyz).x;\n"
4548 "#extension GL_ARB_texture_rectangle : enable\n"
4549 "uniform sampler2DRect sampler;\n"
4550 "uniform vec4 mask;\n"
4553 " if (all(lessThan(gl_FragCoord.xy, mask.zw))) discard;\n"
4554 " gl_FragDepth = texture2DRect(sampler, gl_TexCoord[0].xy).x;\n"
4558 blt_pshader = masked ? blt_pshaders_masked[tex_type] : blt_pshaders_full[tex_type];
4561 FIXME("tex_type %#x not supported\n", tex_type);
4565 vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
4566 shader_glsl_compile(gl_info, vshader_id, blt_vshader);
4568 pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
4569 shader_glsl_compile(gl_info, pshader_id, blt_pshader);
4571 program_id = GL_EXTCALL(glCreateProgramObjectARB());
4572 GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
4573 GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
4574 GL_EXTCALL(glLinkProgramARB(program_id));
4576 shader_glsl_validate_link(gl_info, program_id);
4578 /* Once linked we can mark the shaders for deletion. They will be deleted once the program
4581 GL_EXTCALL(glDeleteObjectARB(vshader_id));
4582 GL_EXTCALL(glDeleteObjectARB(pshader_id));
4586 /* GL locking is done by the caller */
4587 static void shader_glsl_select(const struct wined3d_context *context, BOOL usePS, BOOL useVS)
4589 const struct wined3d_gl_info *gl_info = context->gl_info;
4590 IWineD3DDeviceImpl *device = context->swapchain->device;
4591 struct shader_glsl_priv *priv = device->shader_priv;
4592 GLhandleARB program_id = 0;
4593 GLenum old_vertex_color_clamp, current_vertex_color_clamp;
4595 old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4597 if (useVS || usePS) set_glsl_shader_program(context, device, usePS, useVS);
4598 else priv->glsl_program = NULL;
4600 current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
4602 if (old_vertex_color_clamp != current_vertex_color_clamp)
4604 if (gl_info->supported[ARB_COLOR_BUFFER_FLOAT])
4606 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
4607 checkGLcall("glClampColorARB");
4611 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
4615 program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4616 if (program_id) TRACE("Using GLSL program %u\n", program_id);
4617 GL_EXTCALL(glUseProgramObjectARB(program_id));
4618 checkGLcall("glUseProgramObjectARB");
4620 /* In case that NP2 texcoord fixup data is found for the selected program, trigger a reload of the
4621 * constants. This has to be done because it can't be guaranteed that sampler() (from state.c) is
4622 * called between selecting the shader and using it, which results in wrong fixup for some frames. */
4623 if (priv->glsl_program && priv->glsl_program->np2Fixup_info)
4625 shader_glsl_load_np2fixup_constants(priv, gl_info, &device->stateBlock->state);
4629 /* GL locking is done by the caller */
4630 static void shader_glsl_select_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info,
4631 enum tex_types tex_type, const SIZE *ds_mask_size)
4633 BOOL masked = ds_mask_size->cx && ds_mask_size->cy;
4634 struct shader_glsl_priv *priv = shader_priv;
4635 GLhandleARB *blt_program;
4638 blt_program = masked ? &priv->depth_blt_program_masked[tex_type] : &priv->depth_blt_program_full[tex_type];
4641 *blt_program = create_glsl_blt_shader(gl_info, tex_type, masked);
4642 loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "sampler"));
4643 GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4644 GL_EXTCALL(glUniform1iARB(loc, 0));
4648 GL_EXTCALL(glUseProgramObjectARB(*blt_program));
4653 loc = GL_EXTCALL(glGetUniformLocationARB(*blt_program, "mask"));
4654 GL_EXTCALL(glUniform4fARB(loc, 0.0f, 0.0f, (float)ds_mask_size->cx, (float)ds_mask_size->cy));
4658 /* GL locking is done by the caller */
4659 static void shader_glsl_deselect_depth_blt(void *shader_priv, const struct wined3d_gl_info *gl_info)
4661 struct shader_glsl_priv *priv = shader_priv;
4662 GLhandleARB program_id;
4664 program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
4665 if (program_id) TRACE("Using GLSL program %u\n", program_id);
4667 GL_EXTCALL(glUseProgramObjectARB(program_id));
4668 checkGLcall("glUseProgramObjectARB");
4671 static void shader_glsl_destroy(IWineD3DBaseShader *iface) {
4672 const struct list *linked_programs;
4673 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) iface;
4674 IWineD3DDeviceImpl *device = This->baseShader.device;
4675 struct shader_glsl_priv *priv = device->shader_priv;
4676 const struct wined3d_gl_info *gl_info;
4677 struct wined3d_context *context;
4679 /* Note: Do not use QueryInterface here to find out which shader type this is because this code
4680 * can be called from IWineD3DBaseShader::Release
4682 char pshader = shader_is_pshader_version(This->baseShader.reg_maps.shader_version.type);
4685 struct glsl_pshader_private *shader_data;
4686 shader_data = This->baseShader.backend_data;
4687 if (!shader_data || !shader_data->num_gl_shaders)
4689 HeapFree(GetProcessHeap(), 0, shader_data);
4690 This->baseShader.backend_data = NULL;
4694 context = context_acquire(device, NULL);
4695 gl_info = context->gl_info;
4697 if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->pshader == iface)
4700 shader_glsl_select(context, FALSE, FALSE);
4704 struct glsl_vshader_private *shader_data;
4705 shader_data = This->baseShader.backend_data;
4706 if (!shader_data || !shader_data->num_gl_shaders)
4708 HeapFree(GetProcessHeap(), 0, shader_data);
4709 This->baseShader.backend_data = NULL;
4713 context = context_acquire(device, NULL);
4714 gl_info = context->gl_info;
4716 if (priv->glsl_program && (IWineD3DBaseShader *)priv->glsl_program->vshader == iface)
4719 shader_glsl_select(context, FALSE, FALSE);
4724 linked_programs = &This->baseShader.linked_programs;
4726 TRACE("Deleting linked programs\n");
4727 if (linked_programs->next) {
4728 struct glsl_shader_prog_link *entry, *entry2;
4732 LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
4733 delete_glsl_program_entry(priv, gl_info, entry);
4736 LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
4737 delete_glsl_program_entry(priv, gl_info, entry);
4745 struct glsl_pshader_private *shader_data = This->baseShader.backend_data;
4748 for(i = 0; i < shader_data->num_gl_shaders; i++) {
4749 TRACE("deleting pshader %u\n", shader_data->gl_shaders[i].prgId);
4750 GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4751 checkGLcall("glDeleteObjectARB");
4754 HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4759 struct glsl_vshader_private *shader_data = This->baseShader.backend_data;
4762 for(i = 0; i < shader_data->num_gl_shaders; i++) {
4763 TRACE("deleting vshader %u\n", shader_data->gl_shaders[i].prgId);
4764 GL_EXTCALL(glDeleteObjectARB(shader_data->gl_shaders[i].prgId));
4765 checkGLcall("glDeleteObjectARB");
4768 HeapFree(GetProcessHeap(), 0, shader_data->gl_shaders);
4771 HeapFree(GetProcessHeap(), 0, This->baseShader.backend_data);
4772 This->baseShader.backend_data = NULL;
4774 context_release(context);
4777 static int glsl_program_key_compare(const void *key, const struct wine_rb_entry *entry)
4779 const glsl_program_key_t *k = key;
4780 const struct glsl_shader_prog_link *prog = WINE_RB_ENTRY_VALUE(entry,
4781 const struct glsl_shader_prog_link, program_lookup_entry);
4784 if (k->vshader > prog->vshader) return 1;
4785 else if (k->vshader < prog->vshader) return -1;
4787 if (k->pshader > prog->pshader) return 1;
4788 else if (k->pshader < prog->pshader) return -1;
4790 if (k->vshader && (cmp = memcmp(&k->vs_args, &prog->vs_args, sizeof(prog->vs_args)))) return cmp;
4791 if (k->pshader && (cmp = memcmp(&k->ps_args, &prog->ps_args, sizeof(prog->ps_args)))) return cmp;
4796 static BOOL constant_heap_init(struct constant_heap *heap, unsigned int constant_count)
4798 SIZE_T size = (constant_count + 1) * sizeof(*heap->entries) + constant_count * sizeof(*heap->positions);
4799 void *mem = HeapAlloc(GetProcessHeap(), 0, size);
4803 ERR("Failed to allocate memory\n");
4807 heap->entries = mem;
4808 heap->entries[1].version = 0;
4809 heap->positions = (unsigned int *)(heap->entries + constant_count + 1);
4815 static void constant_heap_free(struct constant_heap *heap)
4817 HeapFree(GetProcessHeap(), 0, heap->entries);
4820 static const struct wine_rb_functions wined3d_glsl_program_rb_functions =
4825 glsl_program_key_compare,
4828 static HRESULT shader_glsl_alloc(IWineD3DDeviceImpl *device)
4830 const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4831 struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
4832 SIZE_T stack_size = wined3d_log2i(max(gl_info->limits.glsl_vs_float_constants,
4833 gl_info->limits.glsl_ps_float_constants)) + 1;
4835 if (!shader_buffer_init(&priv->shader_buffer))
4837 ERR("Failed to initialize shader buffer.\n");
4841 priv->stack = HeapAlloc(GetProcessHeap(), 0, stack_size * sizeof(*priv->stack));
4844 ERR("Failed to allocate memory.\n");
4848 if (!constant_heap_init(&priv->vconst_heap, gl_info->limits.glsl_vs_float_constants))
4850 ERR("Failed to initialize vertex shader constant heap\n");
4854 if (!constant_heap_init(&priv->pconst_heap, gl_info->limits.glsl_ps_float_constants))
4856 ERR("Failed to initialize pixel shader constant heap\n");
4860 if (wine_rb_init(&priv->program_lookup, &wined3d_glsl_program_rb_functions) == -1)
4862 ERR("Failed to initialize rbtree.\n");
4866 priv->next_constant_version = 1;
4868 device->shader_priv = priv;
4872 constant_heap_free(&priv->pconst_heap);
4873 constant_heap_free(&priv->vconst_heap);
4874 HeapFree(GetProcessHeap(), 0, priv->stack);
4875 shader_buffer_free(&priv->shader_buffer);
4876 HeapFree(GetProcessHeap(), 0, priv);
4877 return E_OUTOFMEMORY;
4880 /* Context activation is done by the caller. */
4881 static void shader_glsl_free(IWineD3DDeviceImpl *device)
4883 const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
4884 struct shader_glsl_priv *priv = device->shader_priv;
4888 for (i = 0; i < tex_type_count; ++i)
4890 if (priv->depth_blt_program_full[i])
4892 GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_full[i]));
4894 if (priv->depth_blt_program_masked[i])
4896 GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_program_masked[i]));
4901 wine_rb_destroy(&priv->program_lookup, NULL, NULL);
4902 constant_heap_free(&priv->pconst_heap);
4903 constant_heap_free(&priv->vconst_heap);
4904 HeapFree(GetProcessHeap(), 0, priv->stack);
4905 shader_buffer_free(&priv->shader_buffer);
4907 HeapFree(GetProcessHeap(), 0, device->shader_priv);
4908 device->shader_priv = NULL;
4911 static BOOL shader_glsl_dirty_const(void)
4913 /* TODO: GL_EXT_bindable_uniform can be used to share constants across shaders */
4917 static void shader_glsl_get_caps(const struct wined3d_gl_info *gl_info, struct shader_caps *pCaps)
4919 /* Nvidia Geforce6/7 or Ati R4xx/R5xx cards with GLSL support, support VS 3.0 but older Nvidia/Ati
4920 * models with GLSL support only support 2.0. In case of nvidia we can detect VS 2.0 support based
4921 * on the version of NV_vertex_program.
4922 * For Ati cards there's no way using glsl (it abstracts the lowlevel info away) and also not
4923 * using ARB_vertex_program. It is safe to assume that when a card supports pixel shader 2.0 it
4924 * supports vertex shader 2.0 too and the way around. We can detect ps2.0 using the maximum number
4925 * of native instructions, so use that here. For more info see the pixel shader versioning code below.
4927 if ((gl_info->supported[NV_VERTEX_PROGRAM2] && !gl_info->supported[NV_VERTEX_PROGRAM3])
4928 || gl_info->limits.arb_ps_instructions <= 512)
4929 pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
4931 pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
4932 TRACE_(d3d_caps)("Hardware vertex shader version %d.%d enabled (GLSL)\n", (pCaps->VertexShaderVersion >> 8) & 0xff, pCaps->VertexShaderVersion & 0xff);
4933 pCaps->MaxVertexShaderConst = gl_info->limits.glsl_vs_float_constants;
4935 /* Older DX9-class videocards (GeforceFX / Radeon >9500/X*00) only support pixel shader 2.0/2.0a/2.0b.
4936 * In OpenGL the extensions related to GLSL abstract lowlevel GL info away which is needed
4937 * to distinguish between 2.0 and 3.0 (and 2.0a/2.0b). In case of Nvidia we use their fragment
4938 * program extensions. On other hardware including ATI GL_ARB_fragment_program offers the info
4939 * in max native instructions. Intel and others also offer the info in this extension but they
4940 * don't support GLSL (at least on Windows).
4942 * PS2.0 requires at least 96 instructions, 2.0a/2.0b go up to 512. Assume that if the number
4943 * of instructions is 512 or less we have to do with ps2.0 hardware.
4944 * 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.
4946 if ((gl_info->supported[NV_FRAGMENT_PROGRAM] && !gl_info->supported[NV_FRAGMENT_PROGRAM2])
4947 || gl_info->limits.arb_ps_instructions <= 512)
4948 pCaps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
4950 pCaps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
4952 pCaps->MaxPixelShaderConst = gl_info->limits.glsl_ps_float_constants;
4954 /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
4955 * Direct3D minimum requirement.
4957 * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
4958 * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
4960 * The problem is that the refrast clamps temporary results in the shader to
4961 * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
4962 * then applications may miss the clamping behavior. On the other hand, if it is smaller,
4963 * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
4964 * offer a way to query this.
4966 pCaps->PixelShader1xMaxValue = 8.0;
4967 TRACE_(d3d_caps)("Hardware pixel shader version %d.%d enabled (GLSL)\n", (pCaps->PixelShaderVersion >> 8) & 0xff, pCaps->PixelShaderVersion & 0xff);
4969 pCaps->VSClipping = TRUE;
4972 static BOOL shader_glsl_color_fixup_supported(struct color_fixup_desc fixup)
4974 if (TRACE_ON(d3d_shader) && TRACE_ON(d3d))
4976 TRACE("Checking support for fixup:\n");
4977 dump_color_fixup_desc(fixup);
4980 /* We support everything except YUV conversions. */
4981 if (!is_complex_fixup(fixup))
4987 TRACE("[FAILED]\n");
4991 static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TABLE_SIZE] =
4993 /* WINED3DSIH_ABS */ shader_glsl_map2gl,
4994 /* WINED3DSIH_ADD */ shader_glsl_arith,
4995 /* WINED3DSIH_AND */ NULL,
4996 /* WINED3DSIH_BEM */ shader_glsl_bem,
4997 /* WINED3DSIH_BREAK */ shader_glsl_break,
4998 /* WINED3DSIH_BREAKC */ shader_glsl_breakc,
4999 /* WINED3DSIH_BREAKP */ NULL,
5000 /* WINED3DSIH_CALL */ shader_glsl_call,
5001 /* WINED3DSIH_CALLNZ */ shader_glsl_callnz,
5002 /* WINED3DSIH_CMP */ shader_glsl_cmp,
5003 /* WINED3DSIH_CND */ shader_glsl_cnd,
5004 /* WINED3DSIH_CRS */ shader_glsl_cross,
5005 /* WINED3DSIH_CUT */ NULL,
5006 /* WINED3DSIH_DCL */ NULL,
5007 /* WINED3DSIH_DEF */ NULL,
5008 /* WINED3DSIH_DEFB */ NULL,
5009 /* WINED3DSIH_DEFI */ NULL,
5010 /* WINED3DSIH_DIV */ NULL,
5011 /* WINED3DSIH_DP2ADD */ shader_glsl_dp2add,
5012 /* WINED3DSIH_DP3 */ shader_glsl_dot,
5013 /* WINED3DSIH_DP4 */ shader_glsl_dot,
5014 /* WINED3DSIH_DST */ shader_glsl_dst,
5015 /* WINED3DSIH_DSX */ shader_glsl_map2gl,
5016 /* WINED3DSIH_DSY */ shader_glsl_map2gl,
5017 /* WINED3DSIH_ELSE */ shader_glsl_else,
5018 /* WINED3DSIH_EMIT */ NULL,
5019 /* WINED3DSIH_ENDIF */ shader_glsl_end,
5020 /* WINED3DSIH_ENDLOOP */ shader_glsl_end,
5021 /* WINED3DSIH_ENDREP */ shader_glsl_end,
5022 /* WINED3DSIH_EXP */ shader_glsl_map2gl,
5023 /* WINED3DSIH_EXPP */ shader_glsl_expp,
5024 /* WINED3DSIH_FRC */ shader_glsl_map2gl,
5025 /* WINED3DSIH_FTOI */ NULL,
5026 /* WINED3DSIH_IADD */ NULL,
5027 /* WINED3DSIH_IEQ */ NULL,
5028 /* WINED3DSIH_IF */ shader_glsl_if,
5029 /* WINED3DSIH_IFC */ shader_glsl_ifc,
5030 /* WINED3DSIH_IGE */ NULL,
5031 /* WINED3DSIH_IMUL */ NULL,
5032 /* WINED3DSIH_ITOF */ NULL,
5033 /* WINED3DSIH_LABEL */ shader_glsl_label,
5034 /* WINED3DSIH_LD */ NULL,
5035 /* WINED3DSIH_LIT */ shader_glsl_lit,
5036 /* WINED3DSIH_LOG */ shader_glsl_log,
5037 /* WINED3DSIH_LOGP */ shader_glsl_log,
5038 /* WINED3DSIH_LOOP */ shader_glsl_loop,
5039 /* WINED3DSIH_LRP */ shader_glsl_lrp,
5040 /* WINED3DSIH_LT */ NULL,
5041 /* WINED3DSIH_M3x2 */ shader_glsl_mnxn,
5042 /* WINED3DSIH_M3x3 */ shader_glsl_mnxn,
5043 /* WINED3DSIH_M3x4 */ shader_glsl_mnxn,
5044 /* WINED3DSIH_M4x3 */ shader_glsl_mnxn,
5045 /* WINED3DSIH_M4x4 */ shader_glsl_mnxn,
5046 /* WINED3DSIH_MAD */ shader_glsl_mad,
5047 /* WINED3DSIH_MAX */ shader_glsl_map2gl,
5048 /* WINED3DSIH_MIN */ shader_glsl_map2gl,
5049 /* WINED3DSIH_MOV */ shader_glsl_mov,
5050 /* WINED3DSIH_MOVA */ shader_glsl_mov,
5051 /* WINED3DSIH_MOVC */ NULL,
5052 /* WINED3DSIH_MUL */ shader_glsl_arith,
5053 /* WINED3DSIH_NOP */ NULL,
5054 /* WINED3DSIH_NRM */ shader_glsl_nrm,
5055 /* WINED3DSIH_PHASE */ NULL,
5056 /* WINED3DSIH_POW */ shader_glsl_pow,
5057 /* WINED3DSIH_RCP */ shader_glsl_rcp,
5058 /* WINED3DSIH_REP */ shader_glsl_rep,
5059 /* WINED3DSIH_RET */ shader_glsl_ret,
5060 /* WINED3DSIH_RSQ */ shader_glsl_rsq,
5061 /* WINED3DSIH_SAMPLE */ NULL,
5062 /* WINED3DSIH_SAMPLE_GRAD */ NULL,
5063 /* WINED3DSIH_SAMPLE_LOD */ NULL,
5064 /* WINED3DSIH_SETP */ NULL,
5065 /* WINED3DSIH_SGE */ shader_glsl_compare,
5066 /* WINED3DSIH_SGN */ shader_glsl_sgn,
5067 /* WINED3DSIH_SINCOS */ shader_glsl_sincos,
5068 /* WINED3DSIH_SLT */ shader_glsl_compare,
5069 /* WINED3DSIH_SQRT */ NULL,
5070 /* WINED3DSIH_SUB */ shader_glsl_arith,
5071 /* WINED3DSIH_TEX */ shader_glsl_tex,
5072 /* WINED3DSIH_TEXBEM */ shader_glsl_texbem,
5073 /* WINED3DSIH_TEXBEML */ shader_glsl_texbem,
5074 /* WINED3DSIH_TEXCOORD */ shader_glsl_texcoord,
5075 /* WINED3DSIH_TEXDEPTH */ shader_glsl_texdepth,
5076 /* WINED3DSIH_TEXDP3 */ shader_glsl_texdp3,
5077 /* WINED3DSIH_TEXDP3TEX */ shader_glsl_texdp3tex,
5078 /* WINED3DSIH_TEXKILL */ shader_glsl_texkill,
5079 /* WINED3DSIH_TEXLDD */ shader_glsl_texldd,
5080 /* WINED3DSIH_TEXLDL */ shader_glsl_texldl,
5081 /* WINED3DSIH_TEXM3x2DEPTH */ shader_glsl_texm3x2depth,
5082 /* WINED3DSIH_TEXM3x2PAD */ shader_glsl_texm3x2pad,
5083 /* WINED3DSIH_TEXM3x2TEX */ shader_glsl_texm3x2tex,
5084 /* WINED3DSIH_TEXM3x3 */ shader_glsl_texm3x3,
5085 /* WINED3DSIH_TEXM3x3DIFF */ NULL,
5086 /* WINED3DSIH_TEXM3x3PAD */ shader_glsl_texm3x3pad,
5087 /* WINED3DSIH_TEXM3x3SPEC */ shader_glsl_texm3x3spec,
5088 /* WINED3DSIH_TEXM3x3TEX */ shader_glsl_texm3x3tex,
5089 /* WINED3DSIH_TEXM3x3VSPEC */ shader_glsl_texm3x3vspec,
5090 /* WINED3DSIH_TEXREG2AR */ shader_glsl_texreg2ar,
5091 /* WINED3DSIH_TEXREG2GB */ shader_glsl_texreg2gb,
5092 /* WINED3DSIH_TEXREG2RGB */ shader_glsl_texreg2rgb,
5093 /* WINED3DSIH_UTOF */ NULL,
5096 static void shader_glsl_handle_instruction(const struct wined3d_shader_instruction *ins) {
5097 SHADER_HANDLER hw_fct;
5099 /* Select handler */
5100 hw_fct = shader_glsl_instruction_handler_table[ins->handler_idx];
5102 /* Unhandled opcode */
5105 FIXME("Backend can't handle opcode %#x\n", ins->handler_idx);
5110 shader_glsl_add_instruction_modifiers(ins);
5113 const shader_backend_t glsl_shader_backend = {
5114 shader_glsl_handle_instruction,
5116 shader_glsl_select_depth_blt,
5117 shader_glsl_deselect_depth_blt,
5118 shader_glsl_update_float_vertex_constants,
5119 shader_glsl_update_float_pixel_constants,
5120 shader_glsl_load_constants,
5121 shader_glsl_load_np2fixup_constants,
5122 shader_glsl_destroy,
5125 shader_glsl_dirty_const,
5126 shader_glsl_get_caps,
5127 shader_glsl_color_fixup_supported,