2 * GLSL pixel and vertex shader implementation
4 * Copyright 2006 Jason Green
5 * Copyright 2006-2007 Henri Verbeet
7 * This library is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * This library is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with this library; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
23 * D3D shader asm has swizzles on source parameters, and write masks for
24 * destination parameters. GLSL uses swizzles for both. The result of this is
25 * that for example "mov dst.xw, src.zyxw" becomes "dst.xw = src.zw" in GLSL.
26 * Ie, to generate a proper GLSL source swizzle, we need to take the D3D write
27 * mask for the destination parameter into account.
32 #include "wined3d_private.h"
34 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
35 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
36 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
38 #define GLINFO_LOCATION (*gl_info)
53 } glsl_sample_function_t;
55 /** Prints the GLSL info log which will contain error messages if they exist */
56 void print_glsl_info_log(WineD3D_GL_Info *gl_info, GLhandleARB obj) {
58 int infologLength = 0;
63 const char *spam[] = {
64 "Vertex shader was successfully compiled to run on hardware.\n", /* fglrx */
65 "Fragment shader was successfully compiled to run on hardware.\n", /* fglrx */
66 "Fragment shader(s) linked, vertex shader(s) linked.", /* fglrx, no \n */
67 "Vertex shader(s) linked, no fragment shader(s) defined.", /* fglrx, no \n */
68 "Fragment shader was successfully compiled to run on hardware.\nWARNING: 0:1: extension 'GL_ARB_draw_buffers' is not supported",
69 "Fragment shader(s) linked, no vertex shader(s) defined." /* fglrx, no \n */
72 GL_EXTCALL(glGetObjectParameterivARB(obj,
73 GL_OBJECT_INFO_LOG_LENGTH_ARB,
76 /* A size of 1 is just a null-terminated string, so the log should be bigger than
77 * that if there are errors. */
78 if (infologLength > 1)
80 /* Fglrx doesn't terminate the string properly, but it tells us the proper length.
81 * So use HEAP_ZERO_MEMORY to avoid uninitialized bytes
83 infoLog = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, infologLength);
84 GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
87 for(i = 0; i < sizeof(spam) / sizeof(spam[0]); i++) {
88 if(strcmp(infoLog, spam[i]) == 0) {
94 TRACE("Spam received from GLSL shader #%u: %s\n", obj, debugstr_a(infoLog));
96 FIXME("Error received from GLSL shader #%u: %s\n", obj, debugstr_a(infoLog));
98 HeapFree(GetProcessHeap(), 0, infoLog);
103 * Loads (pixel shader) samplers
105 static void shader_glsl_load_psamplers(
106 WineD3D_GL_Info *gl_info,
107 IWineD3DStateBlock* iface,
108 GLhandleARB programId) {
110 IWineD3DStateBlockImpl* stateBlock = (IWineD3DStateBlockImpl*) iface;
111 GLhandleARB name_loc;
113 char sampler_name[20];
115 for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
116 snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
117 name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
118 if (name_loc != -1) {
119 int mapped_unit = stateBlock->wineD3DDevice->texUnitMap[i];
120 if (mapped_unit != -1 && mapped_unit < GL_LIMITS(fragment_samplers)) {
121 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
122 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
123 checkGLcall("glUniform1iARB");
125 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
131 static void shader_glsl_load_vsamplers(WineD3D_GL_Info *gl_info, IWineD3DStateBlock* iface, GLhandleARB programId) {
132 IWineD3DStateBlockImpl* stateBlock = (IWineD3DStateBlockImpl*) iface;
133 GLhandleARB name_loc;
134 char sampler_name[20];
137 for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
138 snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
139 name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
140 if (name_loc != -1) {
141 int mapped_unit = stateBlock->wineD3DDevice->texUnitMap[MAX_FRAGMENT_SAMPLERS + i];
142 if (mapped_unit != -1 && mapped_unit < GL_LIMITS(combined_samplers)) {
143 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
144 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
145 checkGLcall("glUniform1iARB");
147 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
154 * Loads floating point constants (aka uniforms) into the currently set GLSL program.
155 * When constant_list == NULL, it will load all the constants.
157 static void shader_glsl_load_constantsF(IWineD3DBaseShaderImpl* This, WineD3D_GL_Info *gl_info,
158 unsigned int max_constants, float* constants, GLhandleARB *constant_locations,
159 struct list *constant_list) {
160 constants_entry *constant;
161 local_constant* lconst;
166 if (TRACE_ON(d3d_shader)) {
167 LIST_FOR_EACH_ENTRY(constant, constant_list, constants_entry, entry) {
172 tmp_loc = constant_locations[i];
174 TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
175 constants[i * 4 + 0], constants[i * 4 + 1],
176 constants[i * 4 + 2], constants[i * 4 + 3]);
182 /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
183 if(WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) == 1 &&
184 shader_is_pshader_version(This->baseShader.hex_version)) {
187 LIST_FOR_EACH_ENTRY(constant, constant_list, constants_entry, entry) {
192 tmp_loc = constant_locations[i];
194 /* We found this uniform name in the program - go ahead and send the data */
196 if(constants[k + 0] < -1.0) lcl_const[0] = -1.0;
197 else if(constants[k + 0] > 1.0) lcl_const[0] = 1.0;
198 else lcl_const[0] = constants[k + 0];
199 if(constants[k + 1] < -1.0) lcl_const[1] = -1.0;
200 else if(constants[k + 1] > 1.0) lcl_const[1] = 1.0;
201 else lcl_const[1] = constants[k + 1];
202 if(constants[k + 2] < -1.0) lcl_const[2] = -1.0;
203 else if(constants[k + 2] > 1.0) lcl_const[2] = 1.0;
204 else lcl_const[2] = constants[k + 2];
205 if(constants[k + 3] < -1.0) lcl_const[3] = -1.0;
206 else if(constants[k + 3] > 1.0) lcl_const[3] = 1.0;
207 else lcl_const[3] = constants[k + 3];
209 GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, lcl_const));
214 LIST_FOR_EACH_ENTRY(constant, constant_list, constants_entry, entry) {
219 tmp_loc = constant_locations[i];
221 /* We found this uniform name in the program - go ahead and send the data */
222 GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, constants + (i * 4)));
227 checkGLcall("glUniform4fvARB()");
229 if(!This->baseShader.load_local_constsF) {
230 TRACE("No need to load local float constants for this shader\n");
234 /* Load immediate constants */
235 if (TRACE_ON(d3d_shader)) {
236 LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
237 tmp_loc = constant_locations[lconst->idx];
239 GLfloat* values = (GLfloat*)lconst->value;
240 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
241 values[0], values[1], values[2], values[3]);
245 /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
246 LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
247 tmp_loc = constant_locations[lconst->idx];
249 /* We found this uniform name in the program - go ahead and send the data */
250 GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, (GLfloat*)lconst->value));
253 checkGLcall("glUniform4fvARB()");
257 * Loads integer constants (aka uniforms) into the currently set GLSL program.
258 * When @constants_set == NULL, it will load all the constants.
260 static void shader_glsl_load_constantsI(
261 IWineD3DBaseShaderImpl* This,
262 WineD3D_GL_Info *gl_info,
263 GLhandleARB programId,
264 GLhandleARB locations[MAX_CONST_I],
265 unsigned max_constants,
267 BOOL* constants_set) {
272 for (i=0; i<max_constants; ++i) {
273 if (NULL == constants_set || constants_set[i]) {
275 TRACE_(d3d_constants)("Loading constants %i: %i, %i, %i, %i\n",
276 i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
278 /* We found this uniform name in the program - go ahead and send the data */
279 GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
280 checkGLcall("glUniform4ivARB");
284 /* Load immediate constants */
285 ptr = list_head(&This->baseShader.constantsI);
287 local_constant* lconst = LIST_ENTRY(ptr, struct local_constant, entry);
288 unsigned int idx = lconst->idx;
289 GLint* values = (GLint*) lconst->value;
291 TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
292 values[0], values[1], values[2], values[3]);
294 /* We found this uniform name in the program - go ahead and send the data */
295 GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
296 checkGLcall("glUniform4ivARB");
297 ptr = list_next(&This->baseShader.constantsI, ptr);
302 * Loads boolean constants (aka uniforms) into the currently set GLSL program.
303 * When @constants_set == NULL, it will load all the constants.
305 static void shader_glsl_load_constantsB(
306 IWineD3DBaseShaderImpl* This,
307 WineD3D_GL_Info *gl_info,
308 GLhandleARB programId,
309 unsigned max_constants,
311 BOOL* constants_set) {
316 char is_pshader = shader_is_pshader_version(This->baseShader.hex_version);
317 const char* prefix = is_pshader? "PB":"VB";
320 for (i=0; i<max_constants; ++i) {
321 if (NULL == constants_set || constants_set[i]) {
323 TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
325 /* TODO: Benchmark and see if it would be beneficial to store the
326 * locations of the constants to avoid looking up each time */
327 snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
328 tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
330 /* We found this uniform name in the program - go ahead and send the data */
331 GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
332 checkGLcall("glUniform1ivARB");
337 /* Load immediate constants */
338 ptr = list_head(&This->baseShader.constantsB);
340 local_constant* lconst = LIST_ENTRY(ptr, struct local_constant, entry);
341 unsigned int idx = lconst->idx;
342 GLint* values = (GLint*) lconst->value;
344 TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
346 snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
347 tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
349 /* We found this uniform name in the program - go ahead and send the data */
350 GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
351 checkGLcall("glUniform1ivARB");
353 ptr = list_next(&This->baseShader.constantsB, ptr);
360 * Loads the app-supplied constants into the currently set GLSL program.
362 void shader_glsl_load_constants(
363 IWineD3DDevice* device,
365 char useVertexShader) {
367 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) device;
368 IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
369 WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
371 GLhandleARB *constant_locations;
372 struct list *constant_list;
373 GLhandleARB programId;
374 struct glsl_shader_prog_link *prog = stateBlock->glsl_program;
378 /* No GLSL program set - nothing to do. */
381 programId = prog->programId;
383 if (useVertexShader) {
384 IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
386 constant_locations = prog->vuniformF_locations;
387 constant_list = &stateBlock->set_vconstantsF;
389 /* Load DirectX 9 float constants/uniforms for vertex shader */
390 shader_glsl_load_constantsF(vshader, gl_info, GL_LIMITS(vshader_constantsF),
391 stateBlock->vertexShaderConstantF, constant_locations, constant_list);
393 /* Load DirectX 9 integer constants/uniforms for vertex shader */
394 shader_glsl_load_constantsI(vshader, gl_info, programId,
395 prog->vuniformI_locations, MAX_CONST_I,
396 stateBlock->vertexShaderConstantI,
397 stateBlock->changed.vertexShaderConstantsI);
399 /* Load DirectX 9 boolean constants/uniforms for vertex shader */
400 shader_glsl_load_constantsB(vshader, gl_info, programId, MAX_CONST_B,
401 stateBlock->vertexShaderConstantB,
402 stateBlock->changed.vertexShaderConstantsB);
404 /* Upload the position fixup params */
405 GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, &deviceImpl->posFixup[0]));
406 checkGLcall("glUniform4fvARB");
409 if (usePixelShader) {
411 IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
413 constant_locations = prog->puniformF_locations;
414 constant_list = &stateBlock->set_pconstantsF;
416 /* Load DirectX 9 float constants/uniforms for pixel shader */
417 shader_glsl_load_constantsF(pshader, gl_info, GL_LIMITS(pshader_constantsF),
418 stateBlock->pixelShaderConstantF, constant_locations, constant_list);
420 /* Load DirectX 9 integer constants/uniforms for pixel shader */
421 shader_glsl_load_constantsI(pshader, gl_info, programId,
422 prog->puniformI_locations, MAX_CONST_I,
423 stateBlock->pixelShaderConstantI,
424 stateBlock->changed.pixelShaderConstantsI);
426 /* Load DirectX 9 boolean constants/uniforms for pixel shader */
427 shader_glsl_load_constantsB(pshader, gl_info, programId, MAX_CONST_B,
428 stateBlock->pixelShaderConstantB,
429 stateBlock->changed.pixelShaderConstantsB);
431 /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
432 * It can't be 0 for a valid texbem instruction.
434 for(i = 0; i < ((IWineD3DPixelShaderImpl *) pshader)->numbumpenvmatconsts; i++) {
435 IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pshader;
436 int stage = ps->luminanceconst[i].texunit;
438 float *data = (float *) &stateBlock->textureState[(int) ps->bumpenvmatconst[i].texunit][WINED3DTSS_BUMPENVMAT00];
439 GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
440 checkGLcall("glUniformMatrix2fvARB");
442 /* texbeml needs the luminance scale and offset too. If texbeml is used, needsbumpmat
443 * is set too, so we can check that in the needsbumpmat check
445 if(ps->baseShader.reg_maps.luminanceparams[stage]) {
446 GLfloat *scale = (GLfloat *) &stateBlock->textureState[stage][WINED3DTSS_BUMPENVLSCALE];
447 GLfloat *offset = (GLfloat *) &stateBlock->textureState[stage][WINED3DTSS_BUMPENVLOFFSET];
449 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
450 checkGLcall("glUniform1fvARB");
451 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
452 checkGLcall("glUniform1fvARB");
456 if(((IWineD3DPixelShaderImpl *) pshader)->srgb_enabled &&
457 !((IWineD3DPixelShaderImpl *) pshader)->srgb_mode_hardcoded) {
461 if(stateBlock->renderState[WINED3DRS_SRGBWRITEENABLE]) {
462 comparison[0] = srgb_cmp; comparison[1] = srgb_cmp;
463 comparison[2] = srgb_cmp; comparison[3] = srgb_cmp;
465 mul_low[0] = srgb_mul_low; mul_low[1] = srgb_mul_low;
466 mul_low[2] = srgb_mul_low; mul_low[3] = srgb_mul_low;
468 comparison[0] = 1.0 / 0.0; comparison[1] = 1.0 / 0.0;
469 comparison[2] = 1.0 / 0.0; comparison[3] = 1.0 / 0.0;
471 mul_low[0] = 1.0; mul_low[1] = 1.0;
472 mul_low[2] = 1.0; mul_low[3] = 1.0;
475 GL_EXTCALL(glUniform4fvARB(prog->srgb_comparison_location, 1, comparison));
476 GL_EXTCALL(glUniform4fvARB(prog->srgb_mul_low_location, 1, mul_low));
478 if(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
479 float correction_params[4];
480 if(deviceImpl->render_offscreen) {
481 correction_params[0] = 0.0;
482 correction_params[1] = 1.0;
484 /* position is window relative, not viewport relative */
485 correction_params[0] = ((IWineD3DSurfaceImpl *) deviceImpl->render_targets[0])->currentDesc.Height;
486 correction_params[1] = -1.0;
488 GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
493 /** Generate the variable & register declarations for the GLSL output target */
494 void shader_generate_glsl_declarations(
495 IWineD3DBaseShader *iface,
496 shader_reg_maps* reg_maps,
497 SHADER_BUFFER* buffer,
498 WineD3D_GL_Info* gl_info) {
500 IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
501 IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
503 unsigned int extra_constants_needed = 0;
504 local_constant* lconst;
506 /* There are some minor differences between pixel and vertex shaders */
507 char pshader = shader_is_pshader_version(This->baseShader.hex_version);
508 char prefix = pshader ? 'P' : 'V';
510 /* Prototype the subroutines */
511 for (i = 0; i < This->baseShader.limits.label; i++) {
512 if (reg_maps->labels[i])
513 shader_addline(buffer, "void subroutine%u();\n", i);
516 /* Declare the constants (aka uniforms) */
517 if (This->baseShader.limits.constant_float > 0) {
518 unsigned max_constantsF = min(This->baseShader.limits.constant_float,
519 (pshader ? GL_LIMITS(pshader_constantsF) : GL_LIMITS(vshader_constantsF)));
520 shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
523 if (This->baseShader.limits.constant_int > 0)
524 shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, This->baseShader.limits.constant_int);
526 if (This->baseShader.limits.constant_bool > 0)
527 shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, This->baseShader.limits.constant_bool);
530 shader_addline(buffer, "uniform vec4 posFixup;\n");
531 /* Predeclaration; This function is added at link time based on the pixel shader.
532 * VS 3.0 shaders have an array OUT[] the shader writes to, earlier versions don't have
533 * that. We know the input to the reorder function at vertex shader compile time, so
534 * we can deal with that. The reorder function for a 1.x and 2.x vertex shader can just
535 * read gl_FrontColor. The output depends on the pixel shader. The reorder function for a
536 * 1.x and 2.x pshader or for fixed function will write gl_FrontColor, and for a 3.0 shader
537 * it will write to the varying array. Here we depend on the shader optimizer on sorting that
538 * out. The nvidia driver only does that if the parameter is inout instead of out, hence the
541 if(This->baseShader.hex_version >= WINED3DVS_VERSION(3, 0)) {
542 shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
544 shader_addline(buffer, "void order_ps_input();\n");
547 IWineD3DPixelShaderImpl *ps_impl = (IWineD3DPixelShaderImpl *) This;
549 ps_impl->numbumpenvmatconsts = 0;
550 for(i = 0; i < (sizeof(reg_maps->bumpmat) / sizeof(reg_maps->bumpmat[0])); i++) {
551 if(!reg_maps->bumpmat[i]) {
555 ps_impl->bumpenvmatconst[(int) ps_impl->numbumpenvmatconsts].texunit = i;
556 shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
558 if(reg_maps->luminanceparams) {
559 ps_impl->luminanceconst[(int) ps_impl->numbumpenvmatconsts].texunit = i;
560 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
561 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
562 extra_constants_needed++;
564 ps_impl->luminanceconst[(int) ps_impl->numbumpenvmatconsts].texunit = -1;
567 extra_constants_needed++;
568 ps_impl->numbumpenvmatconsts++;
571 if(device->stateBlock->renderState[WINED3DRS_SRGBWRITEENABLE]) {
572 ps_impl->srgb_enabled = 1;
573 if(This->baseShader.limits.constant_float + extra_constants_needed + 1 < GL_LIMITS(pshader_constantsF)) {
574 shader_addline(buffer, "uniform vec4 srgb_mul_low;\n");
575 shader_addline(buffer, "uniform vec4 srgb_comparison;\n");
576 ps_impl->srgb_mode_hardcoded = 0;
577 extra_constants_needed++;
579 ps_impl->srgb_mode_hardcoded = 1;
580 shader_addline(buffer, "const vec4 srgb_mul_low = vec4(%f, %f, %f, %f);\n",
581 srgb_mul_low, srgb_mul_low, srgb_mul_low, srgb_mul_low);
582 shader_addline(buffer, "const vec4 srgb_comparison = vec4(%f, %f, %f, %f);\n",
583 srgb_cmp, srgb_cmp, srgb_cmp, srgb_cmp);
586 IWineD3DPixelShaderImpl *ps_impl = (IWineD3DPixelShaderImpl *) This;
588 /* Do not write any srgb fixup into the shader to save shader size and processing time.
589 * As a consequence, we can't toggle srgb write on without recompilation
591 ps_impl->srgb_enabled = 0;
592 ps_impl->srgb_mode_hardcoded = 1;
594 if(reg_maps->vpos || reg_maps->usesdsy) {
595 if(This->baseShader.limits.constant_float + extra_constants_needed + 1 < GL_LIMITS(pshader_constantsF)) {
596 shader_addline(buffer, "uniform vec4 ycorrection;\n");
597 ((IWineD3DPixelShaderImpl *) This)->vpos_uniform = 1;
598 extra_constants_needed++;
600 /* This happens because we do not have proper tracking of the constant registers that are
601 * actually used, only the max limit of the shader version
603 FIXME("Cannot find a free uniform for vpos correction params\n");
604 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
605 device->render_offscreen ? 0.0 : ((IWineD3DSurfaceImpl *) device->render_targets[0])->currentDesc.Height,
606 device->render_offscreen ? 1.0 : -1.0);
608 shader_addline(buffer, "vec4 vpos;\n");
612 /* Declare texture samplers */
613 for (i = 0; i < This->baseShader.limits.sampler; i++) {
614 if (reg_maps->samplers[i]) {
616 DWORD stype = reg_maps->samplers[i] & WINED3DSP_TEXTURETYPE_MASK;
620 shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
623 if(device->stateBlock->textures[i] &&
624 IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[i]) == GL_TEXTURE_RECTANGLE_ARB) {
625 shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
627 shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
630 case WINED3DSTT_CUBE:
631 shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
633 case WINED3DSTT_VOLUME:
634 shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
637 shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
638 FIXME("Unrecognized sampler type: %#x\n", stype);
644 /* Declare address variables */
645 for (i = 0; i < This->baseShader.limits.address; i++) {
646 if (reg_maps->address[i])
647 shader_addline(buffer, "ivec4 A%d;\n", i);
650 /* Declare texture coordinate temporaries and initialize them */
651 for (i = 0; i < This->baseShader.limits.texcoord; i++) {
652 if (reg_maps->texcoord[i])
653 shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
656 /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
657 * helper function shader that is linked in at link time
659 if(pshader && This->baseShader.hex_version >= WINED3DPS_VERSION(3, 0)) {
661 shader_addline(buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
663 /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
664 * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
665 * pixel shader that reads the fixed function color into the packed input registers.
667 shader_addline(buffer, "vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
671 /* Declare output register temporaries */
672 if(This->baseShader.limits.packed_output) {
673 shader_addline(buffer, "vec4 OUT[%u];\n", This->baseShader.limits.packed_output);
676 /* Declare temporary variables */
677 for(i = 0; i < This->baseShader.limits.temporary; i++) {
678 if (reg_maps->temporary[i])
679 shader_addline(buffer, "vec4 R%u;\n", i);
682 /* Declare attributes */
683 for (i = 0; i < This->baseShader.limits.attributes; i++) {
684 if (reg_maps->attributes[i])
685 shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
688 /* Declare loop registers aLx */
689 for (i = 0; i < reg_maps->loop_depth; i++) {
690 shader_addline(buffer, "int aL%u;\n", i);
691 shader_addline(buffer, "int tmpInt%u;\n", i);
694 /* Temporary variables for matrix operations */
695 shader_addline(buffer, "vec4 tmp0;\n");
696 shader_addline(buffer, "vec4 tmp1;\n");
698 /* Hardcodable local constants */
699 if(!This->baseShader.load_local_constsF) {
700 LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
701 float *value = (float *) lconst->value;
702 shader_addline(buffer, "const vec4 LC%u = vec4(%f, %f, %f, %f);\n", lconst->idx,
703 value[0], value[1], value[2], value[3]);
707 /* Start the main program */
708 shader_addline(buffer, "void main() {\n");
709 if(pshader && reg_maps->vpos) {
710 /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
711 * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
712 * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
713 * precision troubles when we just substract 0.5.
715 * To deal with that just floor() the position. This will eliminate the fraction on all cards.
717 * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
719 * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
720 * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
721 * coordinates specify the pixel centers instead of the pixel corners. This code will behave
722 * correctly on drivers that returns integer values.
724 shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
728 /*****************************************************************************
729 * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
731 * For more information, see http://wiki.winehq.org/DirectX-Shaders
732 ****************************************************************************/
735 static void shader_glsl_add_src_param(SHADER_OPCODE_ARG* arg, const DWORD param,
736 const DWORD addr_token, DWORD mask, glsl_src_param_t *src_param);
738 /** Used for opcode modifiers - They multiply the result by the specified amount */
739 static const char * const shift_glsl_tab[] = {
741 "2.0 * ", /* 1 (x2) */
742 "4.0 * ", /* 2 (x4) */
743 "8.0 * ", /* 3 (x8) */
744 "16.0 * ", /* 4 (x16) */
745 "32.0 * ", /* 5 (x32) */
752 "0.0625 * ", /* 12 (d16) */
753 "0.125 * ", /* 13 (d8) */
754 "0.25 * ", /* 14 (d4) */
755 "0.5 * " /* 15 (d2) */
758 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
759 static void shader_glsl_gen_modifier (
762 const char *in_regswizzle,
767 if (instr == WINED3DSIO_TEXKILL)
770 switch (instr & WINED3DSP_SRCMOD_MASK) {
771 case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
773 case WINED3DSPSM_NONE:
774 sprintf(out_str, "%s%s", in_reg, in_regswizzle);
776 case WINED3DSPSM_NEG:
777 sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
779 case WINED3DSPSM_NOT:
780 sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
782 case WINED3DSPSM_BIAS:
783 sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
785 case WINED3DSPSM_BIASNEG:
786 sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
788 case WINED3DSPSM_SIGN:
789 sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
791 case WINED3DSPSM_SIGNNEG:
792 sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
794 case WINED3DSPSM_COMP:
795 sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
798 sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
800 case WINED3DSPSM_X2NEG:
801 sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
803 case WINED3DSPSM_ABS:
804 sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
806 case WINED3DSPSM_ABSNEG:
807 sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
810 FIXME("Unhandled modifier %u\n", (instr & WINED3DSP_SRCMOD_MASK));
811 sprintf(out_str, "%s%s", in_reg, in_regswizzle);
815 /** Writes the GLSL variable name that corresponds to the register that the
816 * DX opcode parameter is trying to access */
817 static void shader_glsl_get_register_name(
819 const DWORD addr_token,
822 SHADER_OPCODE_ARG* arg) {
824 /* oPos, oFog and oPts in D3D */
825 static const char * const hwrastout_reg_names[] = { "gl_Position", "gl_FogFragCoord", "gl_PointSize" };
827 DWORD reg = param & WINED3DSP_REGNUM_MASK;
828 DWORD regtype = shader_get_regtype(param);
829 IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) arg->shader;
830 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
831 WineD3D_GL_Info* gl_info = &deviceImpl->adapter->gl_info;
833 char pshader = shader_is_pshader_version(This->baseShader.hex_version);
839 case WINED3DSPR_TEMP:
840 sprintf(tmpStr, "R%u", reg);
842 case WINED3DSPR_INPUT:
844 /* Pixel shaders >= 3.0 */
845 if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) >= 3) {
846 if (param & WINED3DSHADER_ADDRMODE_RELATIVE) {
847 glsl_src_param_t rel_param;
848 shader_glsl_add_src_param(arg, addr_token, 0, WINED3DSP_WRITEMASK_0, &rel_param);
850 /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
853 if(((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg]) {
854 sprintf(tmpStr, "IN[%s + %u]", rel_param.param_str,
855 ((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg]);
857 sprintf(tmpStr, "IN[%s]", rel_param.param_str);
860 sprintf(tmpStr, "IN[%u]",
861 ((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg]);
865 strcpy(tmpStr, "gl_Color");
867 strcpy(tmpStr, "gl_SecondaryColor");
870 if (vshader_input_is_color((IWineD3DVertexShader*) This, reg))
872 sprintf(tmpStr, "attrib%u", reg);
875 case WINED3DSPR_CONST:
877 const char* prefix = pshader? "PC":"VC";
879 /* Relative addressing */
880 if (param & WINED3DSHADER_ADDRMODE_RELATIVE) {
882 /* Relative addressing on shaders 2.0+ have a relative address token,
883 * prior to that, it was hard-coded as "A0.x" because there's only 1 register */
884 if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) >= 2) {
885 glsl_src_param_t rel_param;
886 shader_glsl_add_src_param(arg, addr_token, 0, WINED3DSP_WRITEMASK_0, &rel_param);
888 sprintf(tmpStr, "%s[%s + %u]", prefix, rel_param.param_str, reg);
890 sprintf(tmpStr, "%s[%s]", prefix, rel_param.param_str);
894 sprintf(tmpStr, "%s[A0.x + %u]", prefix, reg);
896 sprintf(tmpStr, "%s[A0.x]", prefix);
901 if(shader_constant_is_local(This, reg)) {
902 sprintf(tmpStr, "LC%u", reg);
904 sprintf(tmpStr, "%s[%u]", prefix, reg);
910 case WINED3DSPR_CONSTINT:
912 sprintf(tmpStr, "PI[%u]", reg);
914 sprintf(tmpStr, "VI[%u]", reg);
916 case WINED3DSPR_CONSTBOOL:
918 sprintf(tmpStr, "PB[%u]", reg);
920 sprintf(tmpStr, "VB[%u]", reg);
922 case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
924 sprintf(tmpStr, "T%u", reg);
926 sprintf(tmpStr, "A%u", reg);
929 case WINED3DSPR_LOOP:
930 sprintf(tmpStr, "aL%u", This->baseShader.cur_loop_regno - 1);
932 case WINED3DSPR_SAMPLER:
934 sprintf(tmpStr, "Psampler%u", reg);
936 sprintf(tmpStr, "Vsampler%u", reg);
938 case WINED3DSPR_COLOROUT:
939 if (reg >= GL_LIMITS(buffers)) {
940 WARN("Write to render target %u, only %d supported\n", reg, 4);
942 if (GL_SUPPORT(ARB_DRAW_BUFFERS)) {
943 sprintf(tmpStr, "gl_FragData[%u]", reg);
944 } else { /* On older cards with GLSL support like the GeforceFX there's only one buffer. */
945 sprintf(tmpStr, "gl_FragColor");
948 case WINED3DSPR_RASTOUT:
949 sprintf(tmpStr, "%s", hwrastout_reg_names[reg]);
951 case WINED3DSPR_DEPTHOUT:
952 sprintf(tmpStr, "gl_FragDepth");
954 case WINED3DSPR_ATTROUT:
956 sprintf(tmpStr, "gl_FrontColor");
958 sprintf(tmpStr, "gl_FrontSecondaryColor");
961 case WINED3DSPR_TEXCRDOUT:
962 /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
963 if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) >= 3)
964 sprintf(tmpStr, "OUT[%u]", reg);
966 sprintf(tmpStr, "gl_TexCoord[%u]", reg);
968 case WINED3DSPR_MISCTYPE:
971 sprintf(tmpStr, "vpos");
972 } else if (reg == 1){
973 /* Note that gl_FrontFacing is a bool, while vFace is
974 * a float for which the sign determines front/back
976 sprintf(tmpStr, "(gl_FrontFacing ? 1.0 : -1.0)");
978 FIXME("Unhandled misctype register %d\n", reg);
979 sprintf(tmpStr, "unrecognized_register");
983 FIXME("Unhandled register name Type(%d)\n", regtype);
984 sprintf(tmpStr, "unrecognized_register");
988 strcat(regstr, tmpStr);
991 /* Get the GLSL write mask for the destination register */
992 static DWORD shader_glsl_get_write_mask(const DWORD param, char *write_mask) {
993 char *ptr = write_mask;
994 DWORD mask = param & WINED3DSP_WRITEMASK_ALL;
996 if (shader_is_scalar(param)) {
997 mask = WINED3DSP_WRITEMASK_0;
1000 if (param & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1001 if (param & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1002 if (param & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1003 if (param & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1011 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1012 unsigned int size = 0;
1014 if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1015 if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1016 if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1017 if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1022 static void shader_glsl_get_swizzle(const DWORD param, BOOL fixup, DWORD mask, char *swizzle_str) {
1023 /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1024 * but addressed as "rgba". To fix this we need to swap the register's x
1025 * and z components. */
1026 DWORD swizzle = (param & WINED3DSP_SWIZZLE_MASK) >> WINED3DSP_SWIZZLE_SHIFT;
1027 const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1028 char *ptr = swizzle_str;
1030 if (!shader_is_scalar(param)) {
1032 /* swizzle bits fields: wwzzyyxx */
1033 if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = swizzle_chars[swizzle & 0x03];
1034 if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = swizzle_chars[(swizzle >> 2) & 0x03];
1035 if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = swizzle_chars[(swizzle >> 4) & 0x03];
1036 if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = swizzle_chars[(swizzle >> 6) & 0x03];
1042 /* From a given parameter token, generate the corresponding GLSL string.
1043 * Also, return the actual register name and swizzle in case the
1044 * caller needs this information as well. */
1045 static void shader_glsl_add_src_param(SHADER_OPCODE_ARG* arg, const DWORD param,
1046 const DWORD addr_token, DWORD mask, glsl_src_param_t *src_param) {
1047 BOOL is_color = FALSE;
1048 char swizzle_str[6];
1050 src_param->reg_name[0] = '\0';
1051 src_param->param_str[0] = '\0';
1052 swizzle_str[0] = '\0';
1054 shader_glsl_get_register_name(param, addr_token, src_param->reg_name, &is_color, arg);
1056 shader_glsl_get_swizzle(param, is_color, mask, swizzle_str);
1057 shader_glsl_gen_modifier(param, src_param->reg_name, swizzle_str, src_param->param_str);
1060 /* From a given parameter token, generate the corresponding GLSL string.
1061 * Also, return the actual register name and swizzle in case the
1062 * caller needs this information as well. */
1063 static DWORD shader_glsl_add_dst_param(SHADER_OPCODE_ARG* arg, const DWORD param,
1064 const DWORD addr_token, glsl_dst_param_t *dst_param) {
1065 BOOL is_color = FALSE;
1067 dst_param->mask_str[0] = '\0';
1068 dst_param->reg_name[0] = '\0';
1070 shader_glsl_get_register_name(param, addr_token, dst_param->reg_name, &is_color, arg);
1071 return shader_glsl_get_write_mask(param, dst_param->mask_str);
1074 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1075 static DWORD shader_glsl_append_dst_ext(SHADER_BUFFER *buffer, SHADER_OPCODE_ARG *arg, const DWORD param) {
1076 glsl_dst_param_t dst_param;
1080 mask = shader_glsl_add_dst_param(arg, param, arg->dst_addr, &dst_param);
1083 shift = (param & WINED3DSP_DSTSHIFT_MASK) >> WINED3DSP_DSTSHIFT_SHIFT;
1084 shader_addline(buffer, "%s%s = %s(", dst_param.reg_name, dst_param.mask_str, shift_glsl_tab[shift]);
1090 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1091 static DWORD shader_glsl_append_dst(SHADER_BUFFER *buffer, SHADER_OPCODE_ARG *arg) {
1092 return shader_glsl_append_dst_ext(buffer, arg, arg->dst);
1095 /** Process GLSL instruction modifiers */
1096 void shader_glsl_add_instruction_modifiers(SHADER_OPCODE_ARG* arg) {
1098 DWORD mask = arg->dst & WINED3DSP_DSTMOD_MASK;
1100 if (arg->opcode->dst_token && mask != 0) {
1101 glsl_dst_param_t dst_param;
1103 shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
1105 if (mask & WINED3DSPDM_SATURATE) {
1106 /* _SAT means to clamp the value of the register to between 0 and 1 */
1107 shader_addline(arg->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1108 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1110 if (mask & WINED3DSPDM_MSAMPCENTROID) {
1111 FIXME("_centroid modifier not handled\n");
1113 if (mask & WINED3DSPDM_PARTIALPRECISION) {
1114 /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1119 static inline const char* shader_get_comp_op(
1120 const DWORD opcode) {
1122 DWORD op = (opcode & INST_CONTROLS_MASK) >> INST_CONTROLS_SHIFT;
1124 case COMPARISON_GT: return ">";
1125 case COMPARISON_EQ: return "==";
1126 case COMPARISON_GE: return ">=";
1127 case COMPARISON_LT: return "<";
1128 case COMPARISON_NE: return "!=";
1129 case COMPARISON_LE: return "<=";
1131 FIXME("Unrecognized comparison value: %u\n", op);
1136 static void shader_glsl_get_sample_function(DWORD sampler_type, BOOL projected, BOOL texrect, glsl_sample_function_t *sample_function) {
1137 /* Note that there's no such thing as a projected cube texture. */
1138 switch(sampler_type) {
1140 sample_function->name = projected ? "texture1DProj" : "texture1D";
1141 sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1145 sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1147 sample_function->name = projected ? "texture2DProj" : "texture2D";
1149 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1151 case WINED3DSTT_CUBE:
1152 sample_function->name = "textureCube";
1153 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1155 case WINED3DSTT_VOLUME:
1156 sample_function->name = projected ? "texture3DProj" : "texture3D";
1157 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1160 sample_function->name = "";
1161 FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1166 static void shader_glsl_color_correction(SHADER_OPCODE_ARG* arg) {
1167 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1168 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) shader->baseShader.device;
1169 WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
1170 glsl_dst_param_t dst_param;
1171 glsl_dst_param_t dst_param2;
1173 WINED3DFORMAT conversion_group;
1174 IWineD3DBaseTextureImpl *texture;
1175 DWORD mask, mask_size;
1177 BOOL recorded = FALSE;
1179 DWORD hex_version = shader->baseShader.hex_version;
1181 switch(arg->opcode->opcode) {
1182 case WINED3DSIO_TEX:
1183 if (hex_version < WINED3DPS_VERSION(2,0)) {
1184 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
1186 sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
1190 case WINED3DSIO_TEXLDL:
1191 FIXME("Add color fixup for vertex texture WINED3DSIO_TEXLDL\n");
1194 case WINED3DSIO_TEXDP3TEX:
1195 case WINED3DSIO_TEXM3x3TEX:
1196 case WINED3DSIO_TEXM3x3SPEC:
1197 case WINED3DSIO_TEXM3x3VSPEC:
1198 case WINED3DSIO_TEXBEM:
1199 case WINED3DSIO_TEXREG2AR:
1200 case WINED3DSIO_TEXREG2GB:
1201 case WINED3DSIO_TEXREG2RGB:
1202 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
1206 /* Not a texture sampling instruction, nothing to do */
1210 texture = (IWineD3DBaseTextureImpl *) deviceImpl->stateBlock->textures[sampler_idx];
1212 fmt = texture->resource.format;
1213 conversion_group = texture->baseTexture.shader_conversion_group;
1215 fmt = WINED3DFMT_UNKNOWN;
1216 conversion_group = WINED3DFMT_UNKNOWN;
1219 /* before doing anything, record the sampler with the format in the format conversion list,
1220 * but check if it's not there already
1222 for(i = 0; i < shader->baseShader.num_sampled_samplers; i++) {
1223 if(shader->baseShader.sampled_samplers[i] == sampler_idx) {
1229 shader->baseShader.sampled_samplers[shader->baseShader.num_sampled_samplers] = sampler_idx;
1230 shader->baseShader.num_sampled_samplers++;
1231 shader->baseShader.sampled_format[sampler_idx] = conversion_group;
1235 case WINED3DFMT_V8U8:
1236 case WINED3DFMT_V16U16:
1237 if(GL_SUPPORT(NV_TEXTURE_SHADER) ||
1238 (GL_SUPPORT(ATI_ENVMAP_BUMPMAP) && fmt == WINED3DFMT_V8U8)) {
1239 /* The 3rd channel returns 1.0 in d3d, but 0.0 in gl. Fix this while we're at it :-) */
1240 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_2, &dst_param);
1241 mask_size = shader_glsl_get_write_mask_size(mask);
1242 if(mask_size >= 3) {
1243 shader_addline(arg->buffer, "%s.%c = 1.0;\n", dst_param.reg_name, dst_param.mask_str[3]);
1246 /* Correct the sign, but leave the blue as it is - it was loaded correctly already */
1247 mask = shader_glsl_add_dst_param(arg, arg->dst,
1248 WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1,
1250 mask_size = shader_glsl_get_write_mask_size(mask);
1251 if(mask_size >= 2) {
1252 shader_addline(arg->buffer, "%s.%c%c = %s.%c%c * 2.0 - 1.0;\n",
1253 dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2],
1254 dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2]);
1255 } else if(mask_size == 1) {
1256 shader_addline(arg->buffer, "%s.%c = %s.%c * 2.0 - 1.0;\n", dst_param.reg_name, dst_param.mask_str[1],
1257 dst_param.reg_name, dst_param.mask_str[1]);
1262 case WINED3DFMT_X8L8V8U8:
1263 if(!GL_SUPPORT(NV_TEXTURE_SHADER)) {
1264 /* Red and blue are the signed channels, fix them up; Blue(=L) is correct already,
1265 * and a(X) is always 1.0
1267 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &dst_param);
1268 mask_size = shader_glsl_get_write_mask_size(mask);
1269 if(mask_size >= 2) {
1270 shader_addline(arg->buffer, "%s.%c%c = %s.%c%c * 2.0 - 1.0;\n",
1271 dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2],
1272 dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2]);
1273 } else if(mask_size == 1) {
1274 shader_addline(arg->buffer, "%s.%c = %s.%c * 2.0 - 1.0;\n",
1275 dst_param.reg_name, dst_param.mask_str[1],
1276 dst_param.reg_name, dst_param.mask_str[1]);
1281 case WINED3DFMT_L6V5U5:
1282 if(!GL_SUPPORT(NV_TEXTURE_SHADER)) {
1283 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &dst_param);
1284 mask_size = shader_glsl_get_write_mask_size(mask);
1285 shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_2, &dst_param2);
1286 if(mask_size >= 3) {
1287 /* Swap y and z (U and L), and do a sign conversion on x and the new y(V and U) */
1288 shader_addline(arg->buffer, "tmp0.g = %s.%c;\n",
1289 dst_param.reg_name, dst_param.mask_str[2]);
1290 shader_addline(arg->buffer, "%s.%c%c = %s.%c%c * 2.0 - 1.0;\n",
1291 dst_param.reg_name, dst_param.mask_str[2], dst_param.mask_str[1],
1292 dst_param2.reg_name, dst_param.mask_str[1], dst_param.mask_str[3]);
1293 shader_addline(arg->buffer, "%s.%c = tmp0.g;\n", dst_param.reg_name,
1294 dst_param.mask_str[3]);
1295 } else if(mask_size == 2) {
1296 /* This is bad: We have VL, but we need VU */
1297 FIXME("2 components sampled from a converted L6V5U5 texture\n");
1299 shader_addline(arg->buffer, "%s.%c = %s.%c * 2.0 - 1.0;\n",
1300 dst_param.reg_name, dst_param.mask_str[1],
1301 dst_param2.reg_name, dst_param.mask_str[1]);
1306 case WINED3DFMT_Q8W8V8U8:
1307 if(!GL_SUPPORT(NV_TEXTURE_SHADER)) {
1308 /* Correct the sign in all channels. The writemask just applies as-is, no
1309 * need for checking the mask size
1311 shader_glsl_add_dst_param(arg, arg->dst,
1312 WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 |
1313 WINED3DSP_WRITEMASK_2 | WINED3DSP_WRITEMASK_3,
1315 shader_addline(arg->buffer, "%s%s = %s%s * 2.0 - 1.0;\n", dst_param.reg_name, dst_param.mask_str,
1316 dst_param.reg_name, dst_param.mask_str);
1320 /* stupid compiler */
1326 /*****************************************************************************
1328 * Begin processing individual instruction opcodes
1330 ****************************************************************************/
1332 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
1333 void shader_glsl_arith(SHADER_OPCODE_ARG* arg) {
1334 CONST SHADER_OPCODE* curOpcode = arg->opcode;
1335 SHADER_BUFFER* buffer = arg->buffer;
1336 glsl_src_param_t src0_param;
1337 glsl_src_param_t src1_param;
1341 /* Determine the GLSL operator to use based on the opcode */
1342 switch (curOpcode->opcode) {
1343 case WINED3DSIO_MUL: op = '*'; break;
1344 case WINED3DSIO_ADD: op = '+'; break;
1345 case WINED3DSIO_SUB: op = '-'; break;
1348 FIXME("Opcode %s not yet handled in GLSL\n", curOpcode->name);
1352 write_mask = shader_glsl_append_dst(buffer, arg);
1353 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1354 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1355 shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
1358 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
1359 void shader_glsl_mov(SHADER_OPCODE_ARG* arg) {
1360 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1361 SHADER_BUFFER* buffer = arg->buffer;
1362 glsl_src_param_t src0_param;
1365 write_mask = shader_glsl_append_dst(buffer, arg);
1366 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1368 /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
1369 * shader versions WINED3DSIO_MOVA is used for this. */
1370 if ((WINED3DSHADER_VERSION_MAJOR(shader->baseShader.hex_version) == 1 &&
1371 !shader_is_pshader_version(shader->baseShader.hex_version) &&
1372 shader_get_regtype(arg->dst) == WINED3DSPR_ADDR)) {
1373 /* This is a simple floor() */
1374 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1375 if (mask_size > 1) {
1376 shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
1378 shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
1380 } else if(arg->opcode->opcode == WINED3DSIO_MOVA) {
1381 /* We need to *round* to the nearest int here. */
1382 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1383 if (mask_size > 1) {
1384 shader_addline(buffer, "ivec%d(floor(abs(%s) + vec%d(0.5)) * sign(%s)));\n", mask_size, src0_param.param_str, mask_size, src0_param.param_str);
1386 shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n", src0_param.param_str, src0_param.param_str);
1389 shader_addline(buffer, "%s);\n", src0_param.param_str);
1393 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
1394 void shader_glsl_dot(SHADER_OPCODE_ARG* arg) {
1395 CONST SHADER_OPCODE* curOpcode = arg->opcode;
1396 SHADER_BUFFER* buffer = arg->buffer;
1397 glsl_src_param_t src0_param;
1398 glsl_src_param_t src1_param;
1399 DWORD dst_write_mask, src_write_mask;
1400 unsigned int dst_size = 0;
1402 dst_write_mask = shader_glsl_append_dst(buffer, arg);
1403 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1405 /* dp3 works on vec3, dp4 on vec4 */
1406 if (curOpcode->opcode == WINED3DSIO_DP4) {
1407 src_write_mask = WINED3DSP_WRITEMASK_ALL;
1409 src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1412 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_write_mask, &src0_param);
1413 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_write_mask, &src1_param);
1416 shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1418 shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
1422 /* Note that this instruction has some restrictions. The destination write mask
1423 * can't contain the w component, and the source swizzles have to be .xyzw */
1424 void shader_glsl_cross(SHADER_OPCODE_ARG *arg) {
1425 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1426 glsl_src_param_t src0_param;
1427 glsl_src_param_t src1_param;
1430 shader_glsl_get_write_mask(arg->dst, dst_mask);
1431 shader_glsl_append_dst(arg->buffer, arg);
1432 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
1433 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_mask, &src1_param);
1434 shader_addline(arg->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
1437 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
1438 * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
1439 * GLSL uses the value as-is. */
1440 void shader_glsl_pow(SHADER_OPCODE_ARG *arg) {
1441 SHADER_BUFFER *buffer = arg->buffer;
1442 glsl_src_param_t src0_param;
1443 glsl_src_param_t src1_param;
1444 DWORD dst_write_mask;
1445 unsigned int dst_size;
1447 dst_write_mask = shader_glsl_append_dst(buffer, arg);
1448 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1450 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1451 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
1454 shader_addline(buffer, "vec%d(pow(abs(%s), %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1456 shader_addline(buffer, "pow(abs(%s), %s));\n", src0_param.param_str, src1_param.param_str);
1460 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
1461 * Src0 is a scalar. Note that D3D uses the absolute of src0, while
1462 * GLSL uses the value as-is. */
1463 void shader_glsl_log(SHADER_OPCODE_ARG *arg) {
1464 SHADER_BUFFER *buffer = arg->buffer;
1465 glsl_src_param_t src0_param;
1466 DWORD dst_write_mask;
1467 unsigned int dst_size;
1469 dst_write_mask = shader_glsl_append_dst(buffer, arg);
1470 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1472 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1475 shader_addline(buffer, "vec%d(log2(abs(%s))));\n", dst_size, src0_param.param_str);
1477 shader_addline(buffer, "log2(abs(%s)));\n", src0_param.param_str);
1481 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
1482 void shader_glsl_map2gl(SHADER_OPCODE_ARG* arg) {
1483 CONST SHADER_OPCODE* curOpcode = arg->opcode;
1484 SHADER_BUFFER* buffer = arg->buffer;
1485 glsl_src_param_t src_param;
1486 const char *instruction;
1487 char arguments[256];
1491 /* Determine the GLSL function to use based on the opcode */
1492 /* TODO: Possibly make this a table for faster lookups */
1493 switch (curOpcode->opcode) {
1494 case WINED3DSIO_MIN: instruction = "min"; break;
1495 case WINED3DSIO_MAX: instruction = "max"; break;
1496 case WINED3DSIO_ABS: instruction = "abs"; break;
1497 case WINED3DSIO_FRC: instruction = "fract"; break;
1498 case WINED3DSIO_NRM: instruction = "normalize"; break;
1499 case WINED3DSIO_LOGP:
1500 case WINED3DSIO_LOG: instruction = "log2"; break;
1501 case WINED3DSIO_EXP: instruction = "exp2"; break;
1502 case WINED3DSIO_SGN: instruction = "sign"; break;
1503 case WINED3DSIO_DSX: instruction = "dFdx"; break;
1504 case WINED3DSIO_DSY: instruction = "ycorrection.y * dFdy"; break;
1505 default: instruction = "";
1506 FIXME("Opcode %s not yet handled in GLSL\n", curOpcode->name);
1510 write_mask = shader_glsl_append_dst(buffer, arg);
1512 arguments[0] = '\0';
1513 if (curOpcode->num_params > 0) {
1514 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src_param);
1515 strcat(arguments, src_param.param_str);
1516 for (i = 2; i < curOpcode->num_params; ++i) {
1517 strcat(arguments, ", ");
1518 shader_glsl_add_src_param(arg, arg->src[i-1], arg->src_addr[i-1], write_mask, &src_param);
1519 strcat(arguments, src_param.param_str);
1523 shader_addline(buffer, "%s(%s));\n", instruction, arguments);
1526 /** Process the WINED3DSIO_EXPP instruction in GLSL:
1527 * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
1528 * dst.x = 2^(floor(src))
1529 * dst.y = src - floor(src)
1530 * dst.z = 2^src (partial precision is allowed, but optional)
1532 * For 2.0 shaders, just do this (honoring writemask and swizzle):
1533 * dst = 2^src; (partial precision is allowed, but optional)
1535 void shader_glsl_expp(SHADER_OPCODE_ARG* arg) {
1536 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)arg->shader;
1537 glsl_src_param_t src_param;
1539 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src_param);
1541 if (shader->baseShader.hex_version < WINED3DPS_VERSION(2,0)) {
1544 shader_addline(arg->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
1545 shader_addline(arg->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
1546 shader_addline(arg->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
1547 shader_addline(arg->buffer, "tmp0.w = 1.0;\n");
1549 shader_glsl_append_dst(arg->buffer, arg);
1550 shader_glsl_get_write_mask(arg->dst, dst_mask);
1551 shader_addline(arg->buffer, "tmp0%s);\n", dst_mask);
1554 unsigned int mask_size;
1556 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1557 mask_size = shader_glsl_get_write_mask_size(write_mask);
1559 if (mask_size > 1) {
1560 shader_addline(arg->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
1562 shader_addline(arg->buffer, "exp2(%s));\n", src_param.param_str);
1567 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
1568 void shader_glsl_rcp(SHADER_OPCODE_ARG* arg) {
1569 glsl_src_param_t src_param;
1571 unsigned int mask_size;
1573 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1574 mask_size = shader_glsl_get_write_mask_size(write_mask);
1575 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src_param);
1577 if (mask_size > 1) {
1578 shader_addline(arg->buffer, "vec%d(1.0 / %s));\n", mask_size, src_param.param_str);
1580 shader_addline(arg->buffer, "1.0 / %s);\n", src_param.param_str);
1584 void shader_glsl_rsq(SHADER_OPCODE_ARG* arg) {
1585 SHADER_BUFFER* buffer = arg->buffer;
1586 glsl_src_param_t src_param;
1588 unsigned int mask_size;
1590 write_mask = shader_glsl_append_dst(buffer, arg);
1591 mask_size = shader_glsl_get_write_mask_size(write_mask);
1593 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src_param);
1595 if (mask_size > 1) {
1596 shader_addline(buffer, "vec%d(inversesqrt(%s)));\n", mask_size, src_param.param_str);
1598 shader_addline(buffer, "inversesqrt(%s));\n", src_param.param_str);
1602 /** Process signed comparison opcodes in GLSL. */
1603 void shader_glsl_compare(SHADER_OPCODE_ARG* arg) {
1604 glsl_src_param_t src0_param;
1605 glsl_src_param_t src1_param;
1607 unsigned int mask_size;
1609 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1610 mask_size = shader_glsl_get_write_mask_size(write_mask);
1611 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1612 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1614 if (mask_size > 1) {
1615 const char *compare;
1617 switch(arg->opcode->opcode) {
1618 case WINED3DSIO_SLT: compare = "lessThan"; break;
1619 case WINED3DSIO_SGE: compare = "greaterThanEqual"; break;
1620 default: compare = "";
1621 FIXME("Can't handle opcode %s\n", arg->opcode->name);
1624 shader_addline(arg->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
1625 src0_param.param_str, src1_param.param_str);
1627 switch(arg->opcode->opcode) {
1628 case WINED3DSIO_SLT:
1629 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
1630 * to return 0.0 but step returns 1.0 because step is not < x
1631 * An alternative is a bvec compare padded with an unused second component.
1632 * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
1633 * issue. Playing with not() is not possible either because not() does not accept
1636 shader_addline(arg->buffer, "(%s < %s) ? 1.0 : 0.0);\n", src0_param.param_str, src1_param.param_str);
1638 case WINED3DSIO_SGE:
1639 /* Here we can use the step() function and safe a conditional */
1640 shader_addline(arg->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
1643 FIXME("Can't handle opcode %s\n", arg->opcode->name);
1649 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
1650 void shader_glsl_cmp(SHADER_OPCODE_ARG* arg) {
1651 glsl_src_param_t src0_param;
1652 glsl_src_param_t src1_param;
1653 glsl_src_param_t src2_param;
1654 DWORD write_mask, cmp_channel = 0;
1657 BOOL temp_destination = FALSE;
1659 if(shader_is_scalar(arg->src[0])) {
1660 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1662 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
1663 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1664 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1666 shader_addline(arg->buffer, "%s >= 0.0 ? %s : %s);\n",
1667 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1669 DWORD src0reg = arg->src[0] & WINED3DSP_REGNUM_MASK;
1670 DWORD src1reg = arg->src[1] & WINED3DSP_REGNUM_MASK;
1671 DWORD src2reg = arg->src[2] & WINED3DSP_REGNUM_MASK;
1672 DWORD src0regtype = shader_get_regtype(arg->src[0]);
1673 DWORD src1regtype = shader_get_regtype(arg->src[1]);
1674 DWORD src2regtype = shader_get_regtype(arg->src[2]);
1675 DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
1676 DWORD dstregtype = shader_get_regtype(arg->dst);
1678 /* Cycle through all source0 channels */
1679 for (i=0; i<4; i++) {
1681 /* Find the destination channels which use the current source0 channel */
1682 for (j=0; j<4; j++) {
1683 if ( ((arg->src[0] >> (WINED3DSP_SWIZZLE_SHIFT + 2*j)) & 0x3) == i ) {
1684 write_mask |= WINED3DSP_WRITEMASK_0 << j;
1685 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
1689 /* Splitting the cmp instruction up in multiple lines imposes a problem:
1690 * The first lines may overwrite source parameters of the following lines.
1691 * Deal with that by using a temporary destination register if needed
1693 if((src0reg == dstreg && src0regtype == dstregtype) ||
1694 (src1reg == dstreg && src1regtype == dstregtype) ||
1695 (src2reg == dstreg && src2regtype == dstregtype)) {
1697 write_mask = shader_glsl_get_write_mask(arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask), mask_char);
1698 if (!write_mask) continue;
1699 shader_addline(arg->buffer, "tmp0%s = (", mask_char);
1700 temp_destination = TRUE;
1702 write_mask = shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask));
1703 if (!write_mask) continue;
1706 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], cmp_channel, &src0_param);
1707 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1708 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1710 shader_addline(arg->buffer, "%s >= 0.0 ? %s : %s);\n",
1711 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1714 if(temp_destination) {
1715 shader_glsl_get_write_mask(arg->dst, mask_char);
1716 shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst);
1717 shader_addline(arg->buffer, "tmp0%s);\n", mask_char);
1723 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
1724 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
1725 * the compare is done per component of src0. */
1726 void shader_glsl_cnd(SHADER_OPCODE_ARG* arg) {
1727 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1728 glsl_src_param_t src0_param;
1729 glsl_src_param_t src1_param;
1730 glsl_src_param_t src2_param;
1731 DWORD write_mask, cmp_channel = 0;
1734 if (shader->baseShader.hex_version < WINED3DPS_VERSION(1, 4)) {
1735 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1736 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1737 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1738 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1740 /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
1741 if(arg->opcode_token & WINED3DSI_COISSUE) {
1742 shader_addline(arg->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
1744 shader_addline(arg->buffer, "%s > 0.5 ? %s : %s);\n",
1745 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1749 /* Cycle through all source0 channels */
1750 for (i=0; i<4; i++) {
1752 /* Find the destination channels which use the current source0 channel */
1753 for (j=0; j<4; j++) {
1754 if ( ((arg->src[0] >> (WINED3DSP_SWIZZLE_SHIFT + 2*j)) & 0x3) == i ) {
1755 write_mask |= WINED3DSP_WRITEMASK_0 << j;
1756 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
1759 write_mask = shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask));
1760 if (!write_mask) continue;
1762 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], cmp_channel, &src0_param);
1763 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1764 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1766 shader_addline(arg->buffer, "%s > 0.5 ? %s : %s);\n",
1767 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1771 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
1772 void shader_glsl_mad(SHADER_OPCODE_ARG* arg) {
1773 glsl_src_param_t src0_param;
1774 glsl_src_param_t src1_param;
1775 glsl_src_param_t src2_param;
1778 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1779 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1780 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1781 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1782 shader_addline(arg->buffer, "(%s * %s) + %s);\n",
1783 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1786 /** Handles transforming all WINED3DSIO_M?x? opcodes for
1787 Vertex shaders to GLSL codes */
1788 void shader_glsl_mnxn(SHADER_OPCODE_ARG* arg) {
1790 int nComponents = 0;
1791 SHADER_OPCODE_ARG tmpArg;
1793 memset(&tmpArg, 0, sizeof(SHADER_OPCODE_ARG));
1795 /* Set constants for the temporary argument */
1796 tmpArg.shader = arg->shader;
1797 tmpArg.buffer = arg->buffer;
1798 tmpArg.src[0] = arg->src[0];
1799 tmpArg.src_addr[0] = arg->src_addr[0];
1800 tmpArg.src_addr[1] = arg->src_addr[1];
1801 tmpArg.reg_maps = arg->reg_maps;
1803 switch(arg->opcode->opcode) {
1804 case WINED3DSIO_M4x4:
1806 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP4);
1808 case WINED3DSIO_M4x3:
1810 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP4);
1812 case WINED3DSIO_M3x4:
1814 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
1816 case WINED3DSIO_M3x3:
1818 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
1820 case WINED3DSIO_M3x2:
1822 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
1828 for (i = 0; i < nComponents; i++) {
1829 tmpArg.dst = ((arg->dst) & ~WINED3DSP_WRITEMASK_ALL)|(WINED3DSP_WRITEMASK_0<<i);
1830 tmpArg.src[1] = arg->src[1]+i;
1831 shader_glsl_dot(&tmpArg);
1836 The LRP instruction performs a component-wise linear interpolation
1837 between the second and third operands using the first operand as the
1838 blend factor. Equation: (dst = src2 + src0 * (src1 - src2))
1839 This is equivalent to mix(src2, src1, src0);
1841 void shader_glsl_lrp(SHADER_OPCODE_ARG* arg) {
1842 glsl_src_param_t src0_param;
1843 glsl_src_param_t src1_param;
1844 glsl_src_param_t src2_param;
1847 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1849 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1850 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1851 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1853 shader_addline(arg->buffer, "mix(%s, %s, %s));\n",
1854 src2_param.param_str, src1_param.param_str, src0_param.param_str);
1857 /** Process the WINED3DSIO_LIT instruction in GLSL:
1858 * dst.x = dst.w = 1.0
1859 * dst.y = (src0.x > 0) ? src0.x
1860 * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
1861 * where src.w is clamped at +- 128
1863 void shader_glsl_lit(SHADER_OPCODE_ARG* arg) {
1864 glsl_src_param_t src0_param;
1865 glsl_src_param_t src1_param;
1866 glsl_src_param_t src3_param;
1869 shader_glsl_append_dst(arg->buffer, arg);
1870 shader_glsl_get_write_mask(arg->dst, dst_mask);
1872 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1873 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_1, &src1_param);
1874 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src3_param);
1876 /* The sdk specifies the instruction like this
1878 * if(src.x > 0.0) dst.y = src.x
1880 * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
1884 * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
1885 * dst.x = 1.0 ... No further explanation needed
1886 * dst.y = max(src.y, 0.0); ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
1887 * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0; ... 0 ^ power is 0, and otherwise we use y anyway
1888 * dst.w = 1.0. ... Nothing fancy.
1890 * So we still have one conditional in there. So do this:
1891 * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
1893 * 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),
1894 * 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.
1895 * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to
1897 shader_addline(arg->buffer, "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",
1898 src0_param.param_str, src1_param.param_str, src0_param.param_str, src3_param.param_str, dst_mask);
1901 /** Process the WINED3DSIO_DST instruction in GLSL:
1903 * dst.y = src0.x * src0.y
1907 void shader_glsl_dst(SHADER_OPCODE_ARG* arg) {
1908 glsl_src_param_t src0y_param;
1909 glsl_src_param_t src0z_param;
1910 glsl_src_param_t src1y_param;
1911 glsl_src_param_t src1w_param;
1914 shader_glsl_append_dst(arg->buffer, arg);
1915 shader_glsl_get_write_mask(arg->dst, dst_mask);
1917 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_1, &src0y_param);
1918 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &src0z_param);
1919 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_1, &src1y_param);
1920 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_3, &src1w_param);
1922 shader_addline(arg->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
1923 src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
1926 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
1927 * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
1928 * can handle it. But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
1930 * dst.x = cos(src0.?)
1931 * dst.y = sin(src0.?)
1935 void shader_glsl_sincos(SHADER_OPCODE_ARG* arg) {
1936 glsl_src_param_t src0_param;
1939 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1940 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1942 switch (write_mask) {
1943 case WINED3DSP_WRITEMASK_0:
1944 shader_addline(arg->buffer, "cos(%s));\n", src0_param.param_str);
1947 case WINED3DSP_WRITEMASK_1:
1948 shader_addline(arg->buffer, "sin(%s));\n", src0_param.param_str);
1951 case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
1952 shader_addline(arg->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
1956 ERR("Write mask should be .x, .y or .xy\n");
1961 /** Process the WINED3DSIO_LOOP instruction in GLSL:
1962 * Start a for() loop where src1.y is the initial value of aL,
1963 * increment aL by src1.z for a total of src1.x iterations.
1964 * Need to use a temporary variable for this operation.
1966 /* FIXME: I don't think nested loops will work correctly this way. */
1967 void shader_glsl_loop(SHADER_OPCODE_ARG* arg) {
1968 glsl_src_param_t src1_param;
1969 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1970 DWORD regtype = shader_get_regtype(arg->src[1]);
1971 DWORD reg = arg->src[1] & WINED3DSP_REGNUM_MASK;
1972 const DWORD *control_values = NULL;
1973 local_constant *constant;
1975 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
1977 /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
1978 * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
1979 * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
1982 if(regtype == WINED3DSPR_CONSTINT) {
1983 LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
1984 if(constant->idx == reg) {
1985 control_values = constant->value;
1991 if(control_values) {
1992 if(control_values[2] > 0) {
1993 shader_addline(arg->buffer, "for (aL%u = %d; aL%u < (%d * %d + %d); aL%u += %d) {\n",
1994 shader->baseShader.cur_loop_depth, control_values[1],
1995 shader->baseShader.cur_loop_depth, control_values[0], control_values[2], control_values[1],
1996 shader->baseShader.cur_loop_depth, control_values[2]);
1997 } else if(control_values[2] == 0) {
1998 shader_addline(arg->buffer, "for (aL%u = %d, tmpInt%u = 0; tmpInt%u < %d; tmpInt%u++) {\n",
1999 shader->baseShader.cur_loop_depth, control_values[1], shader->baseShader.cur_loop_depth,
2000 shader->baseShader.cur_loop_depth, control_values[0],
2001 shader->baseShader.cur_loop_depth);
2003 shader_addline(arg->buffer, "for (aL%u = %d; aL%u > (%d * %d + %d); aL%u += %d) {\n",
2004 shader->baseShader.cur_loop_depth, control_values[1],
2005 shader->baseShader.cur_loop_depth, control_values[0], control_values[2], control_values[1],
2006 shader->baseShader.cur_loop_depth, control_values[2]);
2009 shader_addline(arg->buffer, "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2010 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno,
2011 src1_param.reg_name, shader->baseShader.cur_loop_depth, src1_param.reg_name,
2012 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno, src1_param.reg_name);
2015 shader->baseShader.cur_loop_depth++;
2016 shader->baseShader.cur_loop_regno++;
2019 void shader_glsl_end(SHADER_OPCODE_ARG* arg) {
2020 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
2022 shader_addline(arg->buffer, "}\n");
2024 if(arg->opcode->opcode == WINED3DSIO_ENDLOOP) {
2025 shader->baseShader.cur_loop_depth--;
2026 shader->baseShader.cur_loop_regno--;
2028 if(arg->opcode->opcode == WINED3DSIO_ENDREP) {
2029 shader->baseShader.cur_loop_depth--;
2033 void shader_glsl_rep(SHADER_OPCODE_ARG* arg) {
2034 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
2035 glsl_src_param_t src0_param;
2037 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2038 shader_addline(arg->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2039 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2040 src0_param.param_str, shader->baseShader.cur_loop_depth);
2041 shader->baseShader.cur_loop_depth++;
2044 void shader_glsl_if(SHADER_OPCODE_ARG* arg) {
2045 glsl_src_param_t src0_param;
2047 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2048 shader_addline(arg->buffer, "if (%s) {\n", src0_param.param_str);
2051 void shader_glsl_ifc(SHADER_OPCODE_ARG* arg) {
2052 glsl_src_param_t src0_param;
2053 glsl_src_param_t src1_param;
2055 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2056 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2058 shader_addline(arg->buffer, "if (%s %s %s) {\n",
2059 src0_param.param_str, shader_get_comp_op(arg->opcode_token), src1_param.param_str);
2062 void shader_glsl_else(SHADER_OPCODE_ARG* arg) {
2063 shader_addline(arg->buffer, "} else {\n");
2066 void shader_glsl_break(SHADER_OPCODE_ARG* arg) {
2067 shader_addline(arg->buffer, "break;\n");
2070 /* FIXME: According to MSDN the compare is done per component. */
2071 void shader_glsl_breakc(SHADER_OPCODE_ARG* arg) {
2072 glsl_src_param_t src0_param;
2073 glsl_src_param_t src1_param;
2075 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2076 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2078 shader_addline(arg->buffer, "if (%s %s %s) break;\n",
2079 src0_param.param_str, shader_get_comp_op(arg->opcode_token), src1_param.param_str);
2082 void shader_glsl_label(SHADER_OPCODE_ARG* arg) {
2084 DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2085 shader_addline(arg->buffer, "}\n");
2086 shader_addline(arg->buffer, "void subroutine%u () {\n", snum);
2089 void shader_glsl_call(SHADER_OPCODE_ARG* arg) {
2090 DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2091 shader_addline(arg->buffer, "subroutine%u();\n", snum);
2094 void shader_glsl_callnz(SHADER_OPCODE_ARG* arg) {
2095 glsl_src_param_t src1_param;
2097 DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2098 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2099 shader_addline(arg->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, snum);
2102 /*********************************************
2103 * Pixel Shader Specific Code begins here
2104 ********************************************/
2105 void pshader_glsl_tex(SHADER_OPCODE_ARG* arg) {
2106 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2107 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2108 DWORD hex_version = This->baseShader.hex_version;
2109 char dst_swizzle[6];
2110 glsl_sample_function_t sample_function;
2113 BOOL projected, texrect = FALSE;
2116 /* All versions have a destination register */
2117 shader_glsl_append_dst(arg->buffer, arg);
2119 /* 1.0-1.4: Use destination register as sampler source.
2120 * 2.0+: Use provided sampler source. */
2121 if (hex_version < WINED3DPS_VERSION(1,4)) {
2124 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2125 flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2127 if (flags & WINED3DTTFF_PROJECTED) {
2129 switch (flags & ~WINED3DTTFF_PROJECTED) {
2130 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2131 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
2132 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
2133 case WINED3DTTFF_COUNT4:
2134 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
2139 } else if (hex_version < WINED3DPS_VERSION(2,0)) {
2140 DWORD src_mod = arg->src[0] & WINED3DSP_SRCMOD_MASK;
2141 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2143 if (src_mod == WINED3DSPSM_DZ) {
2145 mask = WINED3DSP_WRITEMASK_2;
2146 } else if (src_mod == WINED3DSPSM_DW) {
2148 mask = WINED3DSP_WRITEMASK_3;
2153 sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
2154 if(arg->opcode_token & WINED3DSI_TEXLD_PROJECT) {
2155 /* ps 2.0 texldp instruction always divides by the fourth component. */
2157 mask = WINED3DSP_WRITEMASK_3;
2163 if(deviceImpl->stateBlock->textures[sampler_idx] &&
2164 IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2168 sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2169 shader_glsl_get_sample_function(sampler_type, projected, texrect, &sample_function);
2170 mask |= sample_function.coord_mask;
2172 if (hex_version < WINED3DPS_VERSION(2,0)) {
2173 shader_glsl_get_write_mask(arg->dst, dst_swizzle);
2175 shader_glsl_get_swizzle(arg->src[1], FALSE, arg->dst, dst_swizzle);
2178 /* 1.0-1.3: Use destination register as coordinate source.
2179 1.4+: Use provided coordinate source register. */
2180 if (hex_version < WINED3DPS_VERSION(1,4)) {
2182 shader_glsl_get_write_mask(mask, coord_mask);
2183 shader_addline(arg->buffer, "%s(Psampler%u, T%u%s)%s);\n",
2184 sample_function.name, sampler_idx, sampler_idx, coord_mask, dst_swizzle);
2186 glsl_src_param_t coord_param;
2187 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], mask, &coord_param);
2188 if(arg->opcode_token & WINED3DSI_TEXLD_BIAS) {
2189 glsl_src_param_t bias;
2190 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &bias);
2192 shader_addline(arg->buffer, "%s(Psampler%u, %s, %s)%s);\n",
2193 sample_function.name, sampler_idx, coord_param.param_str,
2194 bias.param_str, dst_swizzle);
2196 shader_addline(arg->buffer, "%s(Psampler%u, %s)%s);\n",
2197 sample_function.name, sampler_idx, coord_param.param_str, dst_swizzle);
2202 void shader_glsl_texldl(SHADER_OPCODE_ARG* arg) {
2203 IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*)arg->shader;
2204 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2205 glsl_sample_function_t sample_function;
2206 glsl_src_param_t coord_param, lod_param;
2207 char dst_swizzle[6];
2210 BOOL texrect = FALSE;
2212 shader_glsl_append_dst(arg->buffer, arg);
2213 shader_glsl_get_swizzle(arg->src[1], FALSE, arg->dst, dst_swizzle);
2215 sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
2216 sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2217 if(deviceImpl->stateBlock->textures[sampler_idx] &&
2218 IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2221 shader_glsl_get_sample_function(sampler_type, FALSE, texrect, &sample_function); shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], sample_function.coord_mask, &coord_param);
2223 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &lod_param);
2225 if (shader_is_pshader_version(This->baseShader.hex_version)) {
2226 /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
2227 * However, they seem to work just fine in fragment shaders as well. */
2228 WARN("Using %sLod in fragment shader.\n", sample_function.name);
2229 shader_addline(arg->buffer, "%sLod(Psampler%u, %s, %s)%s);\n",
2230 sample_function.name, sampler_idx, coord_param.param_str, lod_param.param_str, dst_swizzle);
2232 shader_addline(arg->buffer, "%sLod(Vsampler%u, %s, %s)%s);\n",
2233 sample_function.name, sampler_idx, coord_param.param_str, lod_param.param_str, dst_swizzle);
2237 void pshader_glsl_texcoord(SHADER_OPCODE_ARG* arg) {
2239 /* FIXME: Make this work for more than just 2D textures */
2241 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2242 SHADER_BUFFER* buffer = arg->buffer;
2243 DWORD hex_version = This->baseShader.hex_version;
2247 write_mask = shader_glsl_append_dst(arg->buffer, arg);
2248 shader_glsl_get_write_mask(write_mask, dst_mask);
2250 if (hex_version != WINED3DPS_VERSION(1,4)) {
2251 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2252 shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n", reg, dst_mask);
2254 DWORD reg = arg->src[0] & WINED3DSP_REGNUM_MASK;
2255 DWORD src_mod = arg->src[0] & WINED3DSP_SRCMOD_MASK;
2256 char dst_swizzle[6];
2258 shader_glsl_get_swizzle(arg->src[0], FALSE, write_mask, dst_swizzle);
2260 if (src_mod == WINED3DSPSM_DZ) {
2261 glsl_src_param_t div_param;
2262 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2263 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &div_param);
2265 if (mask_size > 1) {
2266 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2268 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2270 } else if (src_mod == WINED3DSPSM_DW) {
2271 glsl_src_param_t div_param;
2272 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2273 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &div_param);
2275 if (mask_size > 1) {
2276 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2278 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2281 shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
2286 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
2287 * Take a 3-component dot product of the TexCoord[dstreg] and src,
2288 * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2289 void pshader_glsl_texdp3tex(SHADER_OPCODE_ARG* arg) {
2290 glsl_src_param_t src0_param;
2292 glsl_sample_function_t sample_function;
2293 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2294 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2295 DWORD sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2297 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2299 shader_glsl_append_dst(arg->buffer, arg);
2300 shader_glsl_get_write_mask(arg->dst, dst_mask);
2302 /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
2303 * scalar, and projected sampling would require 4.
2305 * It is a dependent read - not valid with conditional NP2 textures
2307 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2309 switch(count_bits(sample_function.coord_mask)) {
2311 shader_addline(arg->buffer, "%s(Psampler%u, dot(gl_TexCoord[%u].xyz, %s))%s);\n",
2312 sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2316 shader_addline(arg->buffer, "%s(Psampler%u, vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0))%s);\n",
2317 sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2321 shader_addline(arg->buffer, "%s(Psampler%u, vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0))%s);\n",
2322 sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2325 FIXME("Unexpected mask bitcount %d\n", count_bits(sample_function.coord_mask));
2329 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
2330 * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2331 void pshader_glsl_texdp3(SHADER_OPCODE_ARG* arg) {
2332 glsl_src_param_t src0_param;
2333 DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
2334 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2336 unsigned int mask_size;
2338 dst_mask = shader_glsl_append_dst(arg->buffer, arg);
2339 mask_size = shader_glsl_get_write_mask_size(dst_mask);
2340 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2342 if (mask_size > 1) {
2343 shader_addline(arg->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
2345 shader_addline(arg->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
2349 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
2350 * Calculate the depth as dst.x / dst.y */
2351 void pshader_glsl_texdepth(SHADER_OPCODE_ARG* arg) {
2352 glsl_dst_param_t dst_param;
2354 shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
2356 /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
2357 * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
2358 * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
2359 * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
2362 shader_addline(arg->buffer, "gl_FragDepth = clamp((%s.x / min(%s.y, 1.0)), 0.0, 1.0);\n", dst_param.reg_name, dst_param.reg_name);
2365 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
2366 * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2367 * Calculate tmp0.y = TexCoord[dstreg] . src.xyz; (tmp0.x has already been calculated)
2368 * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2370 void pshader_glsl_texm3x2depth(SHADER_OPCODE_ARG* arg) {
2371 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2372 DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
2373 glsl_src_param_t src0_param;
2375 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2377 shader_addline(arg->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
2378 shader_addline(arg->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
2381 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
2382 * Calculate the 1st of a 2-row matrix multiplication. */
2383 void pshader_glsl_texm3x2pad(SHADER_OPCODE_ARG* arg) {
2384 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2385 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2386 SHADER_BUFFER* buffer = arg->buffer;
2387 glsl_src_param_t src0_param;
2389 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2390 shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2393 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
2394 * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
2395 void pshader_glsl_texm3x3pad(SHADER_OPCODE_ARG* arg) {
2397 IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2398 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2399 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2400 SHADER_BUFFER* buffer = arg->buffer;
2401 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2402 glsl_src_param_t src0_param;
2404 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2405 shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + current_state->current_row, reg, src0_param.param_str);
2406 current_state->texcoord_w[current_state->current_row++] = reg;
2409 void pshader_glsl_texm3x2tex(SHADER_OPCODE_ARG* arg) {
2410 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2411 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2412 SHADER_BUFFER* buffer = arg->buffer;
2413 glsl_src_param_t src0_param;
2416 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2417 shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2419 shader_glsl_append_dst(buffer, arg);
2420 shader_glsl_get_write_mask(arg->dst, dst_mask);
2422 /* Sample the texture using the calculated coordinates */
2423 shader_addline(buffer, "texture2D(Psampler%u, tmp0.xy)%s);\n", reg, dst_mask);
2426 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
2427 * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
2428 void pshader_glsl_texm3x3tex(SHADER_OPCODE_ARG* arg) {
2429 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2430 glsl_src_param_t src0_param;
2432 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2433 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2434 SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2435 DWORD sampler_type = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2436 glsl_sample_function_t sample_function;
2438 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2439 shader_addline(arg->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2441 shader_glsl_append_dst(arg->buffer, arg);
2442 shader_glsl_get_write_mask(arg->dst, dst_mask);
2443 /* Dependent read, not valid with conditional NP2 */
2444 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2446 /* Sample the texture using the calculated coordinates */
2447 shader_addline(arg->buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2449 current_state->current_row = 0;
2452 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
2453 * Perform the 3rd row of a 3x3 matrix multiply */
2454 void pshader_glsl_texm3x3(SHADER_OPCODE_ARG* arg) {
2455 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2456 glsl_src_param_t src0_param;
2458 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2459 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2460 SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2462 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2464 shader_glsl_append_dst(arg->buffer, arg);
2465 shader_glsl_get_write_mask(arg->dst, dst_mask);
2466 shader_addline(arg->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
2468 current_state->current_row = 0;
2471 /** Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
2472 * Peform the final texture lookup based on the previous 2 3x3 matrix multiplies */
2473 void pshader_glsl_texm3x3spec(SHADER_OPCODE_ARG* arg) {
2475 IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2476 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2477 glsl_src_param_t src0_param;
2478 glsl_src_param_t src1_param;
2480 SHADER_BUFFER* buffer = arg->buffer;
2481 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2482 DWORD stype = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2483 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2484 glsl_sample_function_t sample_function;
2486 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2487 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_mask, &src1_param);
2489 /* Perform the last matrix multiply operation */
2490 shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2491 /* Reflection calculation */
2492 shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
2494 shader_glsl_append_dst(buffer, arg);
2495 shader_glsl_get_write_mask(arg->dst, dst_mask);
2496 /* Dependent read, not valid with conditional NP2 */
2497 shader_glsl_get_sample_function(stype, FALSE, FALSE, &sample_function);
2499 /* Sample the texture */
2500 shader_addline(buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2502 current_state->current_row = 0;
2505 /** Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
2506 * Peform the final texture lookup based on the previous 2 3x3 matrix multiplies */
2507 void pshader_glsl_texm3x3vspec(SHADER_OPCODE_ARG* arg) {
2509 IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2510 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2511 SHADER_BUFFER* buffer = arg->buffer;
2512 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2513 glsl_src_param_t src0_param;
2515 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2516 DWORD sampler_type = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2517 glsl_sample_function_t sample_function;
2519 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2521 /* Perform the last matrix multiply operation */
2522 shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
2524 /* Construct the eye-ray vector from w coordinates */
2525 shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
2526 current_state->texcoord_w[0], current_state->texcoord_w[1], reg);
2527 shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
2529 shader_glsl_append_dst(buffer, arg);
2530 shader_glsl_get_write_mask(arg->dst, dst_mask);
2531 /* Dependent read, not valid with conditional NP2 */
2532 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2534 /* Sample the texture using the calculated coordinates */
2535 shader_addline(buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2537 current_state->current_row = 0;
2540 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
2541 * Apply a fake bump map transform.
2542 * texbem is pshader <= 1.3 only, this saves a few version checks
2544 void pshader_glsl_texbem(SHADER_OPCODE_ARG* arg) {
2545 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2546 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2547 char dst_swizzle[6];
2548 glsl_sample_function_t sample_function;
2549 glsl_src_param_t coord_param;
2556 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2557 flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2559 sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2560 /* Dependent read, not valid with conditional NP2 */
2561 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2562 mask = sample_function.coord_mask;
2564 shader_glsl_get_write_mask(arg->dst, dst_swizzle);
2566 shader_glsl_get_write_mask(mask, coord_mask);
2568 /* with projective textures, texbem only divides the static texture coord, not the displacement,
2569 * so we can't let the GL handle this.
2571 if (flags & WINED3DTTFF_PROJECTED) {
2573 char coord_div_mask[3];
2574 switch (flags & ~WINED3DTTFF_PROJECTED) {
2575 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2576 case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
2577 case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
2578 case WINED3DTTFF_COUNT4:
2579 case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
2581 shader_glsl_get_write_mask(div_mask, coord_div_mask);
2582 shader_addline(arg->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
2585 shader_glsl_append_dst(arg->buffer, arg);
2586 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &coord_param);
2587 if(arg->opcode->opcode == WINED3DSIO_TEXBEML) {
2588 glsl_src_param_t luminance_param;
2589 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &luminance_param);
2590 shader_addline(arg->buffer, "(%s(Psampler%u, T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s )*(%s * luminancescale%d + luminanceoffset%d))%s);\n",
2591 sample_function.name, sampler_idx, sampler_idx, coord_mask, sampler_idx, coord_param.param_str, coord_mask,
2592 luminance_param.param_str, sampler_idx, sampler_idx, dst_swizzle);
2594 shader_addline(arg->buffer, "%s(Psampler%u, T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s )%s);\n",
2595 sample_function.name, sampler_idx, sampler_idx, coord_mask, sampler_idx, coord_param.param_str, coord_mask, dst_swizzle);
2599 void pshader_glsl_bem(SHADER_OPCODE_ARG* arg) {
2600 glsl_src_param_t src0_param, src1_param;
2601 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2603 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &src0_param);
2604 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &src1_param);
2606 shader_glsl_append_dst(arg->buffer, arg);
2607 shader_addline(arg->buffer, "%s + bumpenvmat%d * %s);\n",
2608 src0_param.param_str, sampler_idx, src1_param.param_str);
2611 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
2612 * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
2613 void pshader_glsl_texreg2ar(SHADER_OPCODE_ARG* arg) {
2615 glsl_src_param_t src0_param;
2616 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2619 shader_glsl_append_dst(arg->buffer, arg);
2620 shader_glsl_get_write_mask(arg->dst, dst_mask);
2621 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2623 shader_addline(arg->buffer, "texture2D(Psampler%u, %s.wx)%s);\n", sampler_idx, src0_param.reg_name, dst_mask);
2626 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
2627 * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
2628 void pshader_glsl_texreg2gb(SHADER_OPCODE_ARG* arg) {
2629 glsl_src_param_t src0_param;
2630 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2633 shader_glsl_append_dst(arg->buffer, arg);
2634 shader_glsl_get_write_mask(arg->dst, dst_mask);
2635 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2637 shader_addline(arg->buffer, "texture2D(Psampler%u, %s.yz)%s);\n", sampler_idx, src0_param.reg_name, dst_mask);
2640 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
2641 * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
2642 void pshader_glsl_texreg2rgb(SHADER_OPCODE_ARG* arg) {
2643 glsl_src_param_t src0_param;
2645 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2646 DWORD sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2647 glsl_sample_function_t sample_function;
2649 shader_glsl_append_dst(arg->buffer, arg);
2650 shader_glsl_get_write_mask(arg->dst, dst_mask);
2651 /* Dependent read, not valid with conditional NP2 */
2652 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2653 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], sample_function.coord_mask, &src0_param);
2655 shader_addline(arg->buffer, "%s(Psampler%u, %s)%s);\n", sample_function.name, sampler_idx, src0_param.param_str, dst_mask);
2658 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
2659 * If any of the first 3 components are < 0, discard this pixel */
2660 void pshader_glsl_texkill(SHADER_OPCODE_ARG* arg) {
2661 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2662 DWORD hex_version = This->baseShader.hex_version;
2663 glsl_dst_param_t dst_param;
2665 /* The argument is a destination parameter, and no writemasks are allowed */
2666 shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
2667 if((hex_version >= WINED3DPS_VERSION(2,0))) {
2668 /* 2.0 shaders compare all 4 components in texkill */
2669 shader_addline(arg->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
2671 /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
2672 * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
2673 * 4 components are defined, only the first 3 are used
2675 shader_addline(arg->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
2679 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
2680 * dst = dot2(src0, src1) + src2 */
2681 void pshader_glsl_dp2add(SHADER_OPCODE_ARG* arg) {
2682 glsl_src_param_t src0_param;
2683 glsl_src_param_t src1_param;
2684 glsl_src_param_t src2_param;
2686 unsigned int mask_size;
2688 write_mask = shader_glsl_append_dst(arg->buffer, arg);
2689 mask_size = shader_glsl_get_write_mask_size(write_mask);
2691 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
2692 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
2693 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], WINED3DSP_WRITEMASK_0, &src2_param);
2695 shader_addline(arg->buffer, "dot(%s, %s) + %s);\n", src0_param.param_str, src1_param.param_str, src2_param.param_str);
2698 void pshader_glsl_input_pack(
2699 SHADER_BUFFER* buffer,
2700 semantic* semantics_in,
2701 IWineD3DPixelShader *iface) {
2704 IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *) iface;
2706 for (i = 0; i < MAX_REG_INPUT; i++) {
2708 DWORD usage_token = semantics_in[i].usage;
2709 DWORD register_token = semantics_in[i].reg;
2710 DWORD usage, usage_idx;
2714 if (!usage_token) continue;
2715 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2716 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2717 shader_glsl_get_write_mask(register_token, reg_mask);
2721 case WINED3DDECLUSAGE_TEXCOORD:
2722 if(usage_idx < 8 && This->vertexprocessing == pretransformed) {
2723 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
2724 This->input_reg_map[i], reg_mask, usage_idx, reg_mask);
2726 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2727 This->input_reg_map[i], reg_mask, reg_mask);
2731 case WINED3DDECLUSAGE_COLOR:
2733 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
2734 This->input_reg_map[i], reg_mask, reg_mask);
2735 else if (usage_idx == 1)
2736 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
2737 This->input_reg_map[i], reg_mask, reg_mask);
2739 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2740 This->input_reg_map[i], reg_mask, reg_mask);
2744 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2745 This->input_reg_map[i], reg_mask, reg_mask);
2750 /*********************************************
2751 * Vertex Shader Specific Code begins here
2752 ********************************************/
2754 static void add_glsl_program_entry(IWineD3DDeviceImpl *device, struct glsl_shader_prog_link *entry) {
2755 glsl_program_key_t *key;
2757 key = HeapAlloc(GetProcessHeap(), 0, sizeof(glsl_program_key_t));
2758 key->vshader = entry->vshader;
2759 key->pshader = entry->pshader;
2761 hash_table_put(device->glsl_program_lookup, key, entry);
2764 static struct glsl_shader_prog_link *get_glsl_program_entry(IWineD3DDeviceImpl *device,
2765 GLhandleARB vshader, GLhandleARB pshader) {
2766 glsl_program_key_t key;
2768 key.vshader = vshader;
2769 key.pshader = pshader;
2771 return (struct glsl_shader_prog_link *)hash_table_get(device->glsl_program_lookup, &key);
2774 void delete_glsl_program_entry(IWineD3DDevice *iface, struct glsl_shader_prog_link *entry) {
2775 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
2776 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
2777 glsl_program_key_t *key;
2779 key = HeapAlloc(GetProcessHeap(), 0, sizeof(glsl_program_key_t));
2780 key->vshader = entry->vshader;
2781 key->pshader = entry->pshader;
2782 hash_table_remove(This->glsl_program_lookup, key);
2784 GL_EXTCALL(glDeleteObjectARB(entry->programId));
2785 if (entry->vshader) list_remove(&entry->vshader_entry);
2786 if (entry->pshader) list_remove(&entry->pshader_entry);
2787 HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
2788 HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
2789 HeapFree(GetProcessHeap(), 0, entry);
2792 static void handle_ps3_input(SHADER_BUFFER *buffer, semantic *semantics_in, semantic *semantics_out, WineD3D_GL_Info *gl_info, DWORD *map) {
2794 DWORD usage_token, usage_token_out;
2795 DWORD register_token, register_token_out;
2796 DWORD usage, usage_idx, usage_out, usage_idx_out;
2798 char reg_mask[6], reg_mask_out[6];
2800 set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (GL_LIMITS(glsl_varyings) / 4));
2802 for(i = 0; i < MAX_REG_INPUT; i++) {
2803 usage_token = semantics_in[i].usage;
2804 if (!usage_token) continue;
2805 if(map[i] >= (GL_LIMITS(glsl_varyings) / 4)) {
2806 FIXME("More input varyings declared than supported, expect issues\n");
2808 } else if(map[i] == -1) {
2809 /* Declared, but not read register */
2812 register_token = semantics_in[i].reg;
2814 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2815 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2816 set[map[i]] = shader_glsl_get_write_mask(register_token, reg_mask);
2818 if(!semantics_out) {
2820 case WINED3DDECLUSAGE_COLOR:
2822 shader_addline(buffer, "IN[%u]%s = gl_FrontColor%s;\n",
2823 map[i], reg_mask, reg_mask);
2824 else if (usage_idx == 1)
2825 shader_addline(buffer, "IN[%u]%s = gl_FrontSecondaryColor%s;\n",
2826 map[i], reg_mask, reg_mask);
2828 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2829 map[i], reg_mask, reg_mask);
2832 case WINED3DDECLUSAGE_TEXCOORD:
2833 if (usage_idx < 8) {
2834 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
2835 map[i], reg_mask, usage_idx, reg_mask);
2837 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2838 map[i], reg_mask, reg_mask);
2842 case WINED3DDECLUSAGE_FOG:
2843 shader_addline(buffer, "IN[%u]%s = vec4(gl_FogFragCoord, 0.0, 0.0, 0.0)%s;\n",
2844 map[i], reg_mask, reg_mask);
2848 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2849 map[i], reg_mask, reg_mask);
2853 for(j = 0; j < MAX_REG_OUTPUT; j++) {
2854 usage_token_out = semantics_out[j].usage;
2855 if (!usage_token_out) continue;
2856 register_token_out = semantics_out[j].reg;
2858 usage_out = (usage_token_out & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2859 usage_idx_out = (usage_token_out & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2860 shader_glsl_get_write_mask(register_token_out, reg_mask_out);
2862 if(usage == usage_out &&
2863 usage_idx == usage_idx_out) {
2864 shader_addline(buffer, "IN[%u]%s = OUT[%u]%s;\n",
2865 map[i], reg_mask, j, reg_mask);
2870 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2871 map[i], reg_mask, reg_mask);
2876 /* This is solely to make the compiler / linker happy and avoid warning about undefined
2877 * varyings. It shouldn't result in any real code executed on the GPU, since all read
2878 * input varyings are assigned above, if the optimizer works properly.
2880 for(i = 0; i < GL_LIMITS(glsl_varyings) / 4; i++) {
2881 if(set[i] != WINED3DSP_WRITEMASK_ALL) {
2882 unsigned int size = 0;
2883 memset(reg_mask, 0, sizeof(reg_mask));
2884 if(!(set[i] & WINED3DSP_WRITEMASK_0)) {
2885 reg_mask[size] = 'x';
2888 if(!(set[i] & WINED3DSP_WRITEMASK_1)) {
2889 reg_mask[size] = 'y';
2892 if(!(set[i] & WINED3DSP_WRITEMASK_2)) {
2893 reg_mask[size] = 'z';
2896 if(!(set[i] & WINED3DSP_WRITEMASK_3)) {
2897 reg_mask[size] = 'w';
2902 shader_addline(buffer, "IN[%u].%s = 0.0;\n", i, reg_mask);
2905 shader_addline(buffer, "IN[%u].%s = vec2(0.0, 0.0);\n", i, reg_mask);
2908 shader_addline(buffer, "IN[%u].%s = vec3(0.0, 0.0, 0.0);\n", i, reg_mask);
2911 shader_addline(buffer, "IN[%u].%s = vec4(0.0, 0.0, 0.0, 0.0);\n", i, reg_mask);
2917 HeapFree(GetProcessHeap(), 0, set);
2920 static GLhandleARB generate_param_reorder_function(IWineD3DVertexShader *vertexshader,
2921 IWineD3DPixelShader *pixelshader,
2922 WineD3D_GL_Info *gl_info) {
2923 GLhandleARB ret = 0;
2924 IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) vertexshader;
2925 IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pixelshader;
2926 DWORD vs_major = vs ? WINED3DSHADER_VERSION_MAJOR(vs->baseShader.hex_version) : 0;
2927 DWORD ps_major = ps ? WINED3DSHADER_VERSION_MAJOR(ps->baseShader.hex_version) : 0;
2929 SHADER_BUFFER buffer;
2931 DWORD register_token;
2932 DWORD usage, usage_idx, writemask;
2934 semantic *semantics_out, *semantics_in;
2936 buffer.buffer = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, SHADER_PGMSIZE);
2939 buffer.newline = TRUE;
2941 if(vs_major < 3 && ps_major < 3) {
2942 /* That one is easy: The vertex shader writes to the builtin varyings, the pixel shader reads from them.
2943 * Take care about the texcoord .w fixup though if we're using the fixed function fragment pipeline
2945 if((GLINFO_LOCATION).set_texcoord_w && ps_major == 0 && vs_major > 0) {
2946 shader_addline(&buffer, "void order_ps_input() {\n");
2947 for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
2948 if(vs->baseShader.reg_maps.texcoord_mask[i] != 0 &&
2949 vs->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
2950 shader_addline(&buffer, "gl_TexCoord[%u].w = 1.0;\n", i);
2953 shader_addline(&buffer, "}\n");
2955 shader_addline(&buffer, "void order_ps_input() { /* do nothing */ }\n");
2957 } else if(ps_major < 3 && vs_major >= 3) {
2958 /* The vertex shader writes to its own varyings, the pixel shader needs them in the builtin ones */
2959 semantics_out = vs->semantics_out;
2961 shader_addline(&buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
2962 for(i = 0; i < MAX_REG_OUTPUT; i++) {
2963 usage_token = semantics_out[i].usage;
2964 if (!usage_token) continue;
2965 register_token = semantics_out[i].reg;
2967 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2968 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2969 writemask = shader_glsl_get_write_mask(register_token, reg_mask);
2972 case WINED3DDECLUSAGE_COLOR:
2974 shader_addline(&buffer, "gl_FrontColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
2975 else if (usage_idx == 1)
2976 shader_addline(&buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
2979 case WINED3DDECLUSAGE_POSITION:
2980 shader_addline(&buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
2983 case WINED3DDECLUSAGE_TEXCOORD:
2984 if (usage_idx < 8) {
2985 if(!(GLINFO_LOCATION).set_texcoord_w || ps_major > 0) writemask |= WINED3DSP_WRITEMASK_3;
2987 shader_addline(&buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
2988 usage_idx, reg_mask, i, reg_mask);
2989 if(!(writemask & WINED3DSP_WRITEMASK_3)) {
2990 shader_addline(&buffer, "gl_TexCoord[%u].w = 1.0;\n", usage_idx);
2995 case WINED3DDECLUSAGE_PSIZE:
2996 shader_addline(&buffer, "gl_PointSize = OUT[%u].x;\n", i);
2999 case WINED3DDECLUSAGE_FOG:
3000 shader_addline(&buffer, "gl_FogFragCoord = OUT[%u].%c;\n", i, reg_mask[1]);
3007 shader_addline(&buffer, "}\n");
3009 } else if(ps_major >= 3 && vs_major >= 3) {
3010 semantics_out = vs->semantics_out;
3011 semantics_in = ps->semantics_in;
3013 /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3014 shader_addline(&buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
3015 shader_addline(&buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3017 /* First, sort out position and point size. Those are not passed to the pixel shader */
3018 for(i = 0; i < MAX_REG_OUTPUT; i++) {
3019 usage_token = semantics_out[i].usage;
3020 if (!usage_token) continue;
3021 register_token = semantics_out[i].reg;
3023 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
3024 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
3025 shader_glsl_get_write_mask(register_token, reg_mask);
3028 case WINED3DDECLUSAGE_POSITION:
3029 shader_addline(&buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3032 case WINED3DDECLUSAGE_PSIZE:
3033 shader_addline(&buffer, "gl_PointSize = OUT[%u].x;\n", i);
3041 /* Then, fix the pixel shader input */
3042 handle_ps3_input(&buffer, semantics_in, semantics_out, gl_info, ps->input_reg_map);
3044 shader_addline(&buffer, "}\n");
3045 } else if(ps_major >= 3 && vs_major < 3) {
3046 semantics_in = ps->semantics_in;
3048 shader_addline(&buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
3049 shader_addline(&buffer, "void order_ps_input() {\n");
3050 /* The vertex shader wrote to the builtin varyings. There is no need to figure out position and
3051 * point size, but we depend on the optimizers kindness to find out that the pixel shader doesn't
3052 * read gl_TexCoord and gl_ColorX, otherwise we'll run out of varyings
3054 handle_ps3_input(&buffer, semantics_in, NULL, gl_info, ps->input_reg_map);
3055 shader_addline(&buffer, "}\n");
3057 ERR("Unexpected vertex and pixel shader version condition: vs: %d, ps: %d\n", vs_major, ps_major);
3060 ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3061 checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3062 GL_EXTCALL(glShaderSourceARB(ret, 1, (const char**)&buffer.buffer, NULL));
3063 checkGLcall("glShaderSourceARB(ret, 1, (const char**)&buffer.buffer, NULL)");
3064 GL_EXTCALL(glCompileShaderARB(ret));
3065 checkGLcall("glCompileShaderARB(ret)");
3067 HeapFree(GetProcessHeap(), 0, buffer.buffer);
3071 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
3072 * It sets the programId on the current StateBlock (because it should be called
3073 * inside of the DrawPrimitive() part of the render loop).
3075 * If a program for the given combination does not exist, create one, and store
3076 * the program in the hash table. If it creates a program, it will link the
3077 * given objects, too.
3079 static void set_glsl_shader_program(IWineD3DDevice *iface, BOOL use_ps, BOOL use_vs) {
3080 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3081 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3082 IWineD3DPixelShader *pshader = This->stateBlock->pixelShader;
3083 IWineD3DVertexShader *vshader = This->stateBlock->vertexShader;
3084 struct glsl_shader_prog_link *entry = NULL;
3085 GLhandleARB programId = 0;
3086 GLhandleARB reorder_shader_id = 0;
3090 GLhandleARB vshader_id = use_vs ? ((IWineD3DBaseShaderImpl*)vshader)->baseShader.prgId : 0;
3091 GLhandleARB pshader_id = use_ps ? ((IWineD3DBaseShaderImpl*)pshader)->baseShader.prgId : 0;
3092 entry = get_glsl_program_entry(This, vshader_id, pshader_id);
3094 This->stateBlock->glsl_program = entry;
3098 /* If we get to this point, then no matching program exists, so we create one */
3099 programId = GL_EXTCALL(glCreateProgramObjectARB());
3100 TRACE("Created new GLSL shader program %u\n", programId);
3102 /* Create the entry */
3103 entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
3104 entry->programId = programId;
3105 entry->vshader = vshader_id;
3106 entry->pshader = pshader_id;
3107 /* Add the hash table entry */
3108 add_glsl_program_entry(This, entry);
3110 /* Set the current program */
3111 This->stateBlock->glsl_program = entry;
3113 /* Attach GLSL vshader */
3115 int max_attribs = 16; /* TODO: Will this always be the case? It is at the moment... */
3118 reorder_shader_id = generate_param_reorder_function(vshader, pshader, gl_info);
3119 TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
3120 GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
3121 checkGLcall("glAttachObjectARB");
3122 /* Flag the reorder function for deletion, then it will be freed automatically when the program
3125 GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
3127 TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
3128 GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
3129 checkGLcall("glAttachObjectARB");
3131 /* Bind vertex attributes to a corresponding index number to match
3132 * the same index numbers as ARB_vertex_programs (makes loading
3133 * vertex attributes simpler). With this method, we can use the
3134 * exact same code to load the attributes later for both ARB and
3137 * We have to do this here because we need to know the Program ID
3138 * in order to make the bindings work, and it has to be done prior
3139 * to linking the GLSL program. */
3140 for (i = 0; i < max_attribs; ++i) {
3141 if (((IWineD3DBaseShaderImpl*)vshader)->baseShader.reg_maps.attributes[i]) {
3142 snprintf(tmp_name, sizeof(tmp_name), "attrib%i", i);
3143 GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
3146 checkGLcall("glBindAttribLocationARB");
3148 list_add_head(&((IWineD3DBaseShaderImpl *)vshader)->baseShader.linked_programs, &entry->vshader_entry);
3151 /* Attach GLSL pshader */
3153 TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
3154 GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
3155 checkGLcall("glAttachObjectARB");
3157 list_add_head(&((IWineD3DBaseShaderImpl *)pshader)->baseShader.linked_programs, &entry->pshader_entry);
3160 /* Link the program */
3161 TRACE("Linking GLSL shader program %u\n", programId);
3162 GL_EXTCALL(glLinkProgramARB(programId));
3163 print_glsl_info_log(&GLINFO_LOCATION, programId);
3165 entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0, sizeof(GLhandleARB) * GL_LIMITS(vshader_constantsF));
3166 for (i = 0; i < GL_LIMITS(vshader_constantsF); ++i) {
3167 snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
3168 entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3170 for (i = 0; i < MAX_CONST_I; ++i) {
3171 snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
3172 entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3174 entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0, sizeof(GLhandleARB) * GL_LIMITS(pshader_constantsF));
3175 for (i = 0; i < GL_LIMITS(pshader_constantsF); ++i) {
3176 snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
3177 entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3179 for (i = 0; i < MAX_CONST_I; ++i) {
3180 snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
3181 entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3185 for(i = 0; i < ((IWineD3DPixelShaderImpl*)pshader)->numbumpenvmatconsts; i++) {
3187 sprintf(name, "bumpenvmat%d", ((IWineD3DPixelShaderImpl*)pshader)->bumpenvmatconst[i].texunit);
3188 entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3189 sprintf(name, "luminancescale%d", ((IWineD3DPixelShaderImpl*)pshader)->luminanceconst[i].texunit);
3190 entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3191 sprintf(name, "luminanceoffset%d", ((IWineD3DPixelShaderImpl*)pshader)->luminanceconst[i].texunit);
3192 entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3197 entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
3198 entry->srgb_comparison_location = GL_EXTCALL(glGetUniformLocationARB(programId, "srgb_comparison"));
3199 entry->srgb_mul_low_location = GL_EXTCALL(glGetUniformLocationARB(programId, "srgb_mul_low"));
3200 entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
3201 checkGLcall("Find glsl program uniform locations");
3203 /* Set the shader to allow uniform loading on it */
3204 GL_EXTCALL(glUseProgramObjectARB(programId));
3205 checkGLcall("glUseProgramObjectARB(programId)");
3207 /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
3208 * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
3209 * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
3210 * vertex shader with fixed function pixel processing is used we make sure that the card
3211 * supports enough samplers to allow the max number of vertex samplers with all possible
3212 * fixed function fragment processing setups. So once the program is linked these samplers
3216 /* Load vertex shader samplers */
3217 shader_glsl_load_vsamplers(gl_info, (IWineD3DStateBlock*)This->stateBlock, programId);
3220 /* Load pixel shader samplers */
3221 shader_glsl_load_psamplers(gl_info, (IWineD3DStateBlock*)This->stateBlock, programId);
3225 static GLhandleARB create_glsl_blt_shader(WineD3D_GL_Info *gl_info) {
3226 GLhandleARB program_id;
3227 GLhandleARB vshader_id, pshader_id;
3228 const char *blt_vshader[] = {
3231 " gl_Position = gl_Vertex;\n"
3232 " gl_FrontColor = vec4(1.0);\n"
3233 " gl_TexCoord[0].x = (gl_Vertex.x * 0.5) + 0.5;\n"
3234 " gl_TexCoord[0].y = (-gl_Vertex.y * 0.5) + 0.5;\n"
3238 const char *blt_pshader[] = {
3239 "uniform sampler2D sampler;\n"
3242 " gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
3246 vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3247 GL_EXTCALL(glShaderSourceARB(vshader_id, 1, blt_vshader, NULL));
3248 GL_EXTCALL(glCompileShaderARB(vshader_id));
3250 pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3251 GL_EXTCALL(glShaderSourceARB(pshader_id, 1, blt_pshader, NULL));
3252 GL_EXTCALL(glCompileShaderARB(pshader_id));
3254 program_id = GL_EXTCALL(glCreateProgramObjectARB());
3255 GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
3256 GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
3257 GL_EXTCALL(glLinkProgramARB(program_id));
3259 print_glsl_info_log(&GLINFO_LOCATION, program_id);
3261 /* Once linked we can mark the shaders for deletion. They will be deleted once the program
3264 GL_EXTCALL(glDeleteObjectARB(vshader_id));
3265 GL_EXTCALL(glDeleteObjectARB(pshader_id));
3269 static void shader_glsl_select(IWineD3DDevice *iface, BOOL usePS, BOOL useVS) {
3270 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3271 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3272 GLhandleARB program_id = 0;
3274 if (useVS || usePS) set_glsl_shader_program(iface, usePS, useVS);
3275 else This->stateBlock->glsl_program = NULL;
3277 program_id = This->stateBlock->glsl_program ? This->stateBlock->glsl_program->programId : 0;
3278 if (program_id) TRACE("Using GLSL program %u\n", program_id);
3279 GL_EXTCALL(glUseProgramObjectARB(program_id));
3280 checkGLcall("glUseProgramObjectARB");
3283 static void shader_glsl_select_depth_blt(IWineD3DDevice *iface) {
3284 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3285 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3286 struct shader_glsl_priv *priv = (struct shader_glsl_priv *) This->shader_priv;
3287 static GLhandleARB loc = -1;
3289 if (!priv->depth_blt_glsl_program_id) {
3290 priv->depth_blt_glsl_program_id = create_glsl_blt_shader(gl_info);
3291 loc = GL_EXTCALL(glGetUniformLocationARB(priv->depth_blt_glsl_program_id, "sampler"));
3294 GL_EXTCALL(glUseProgramObjectARB(priv->depth_blt_glsl_program_id));
3295 GL_EXTCALL(glUniform1iARB(loc, 0));
3298 static void shader_glsl_destroy_depth_blt(IWineD3DDevice *iface) {
3299 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3300 struct shader_glsl_priv *priv = (struct shader_glsl_priv *) This->shader_priv;
3301 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3303 if(priv->depth_blt_glsl_program_id) {
3304 GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_glsl_program_id));
3305 priv->depth_blt_glsl_program_id = 0;
3309 static void shader_glsl_cleanup(IWineD3DDevice *iface) {
3310 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3311 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3312 GL_EXTCALL(glUseProgramObjectARB(0));
3315 static void shader_glsl_destroy(IWineD3DBaseShader *iface) {
3316 struct list *linked_programs;
3317 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) iface;
3318 WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *) This->baseShader.device)->adapter->gl_info;
3320 /* Note: Do not use QueryInterface here to find out which shader type this is because this code
3321 * can be called from IWineD3DBaseShader::Release
3323 char pshader = shader_is_pshader_version(This->baseShader.hex_version);
3325 if(This->baseShader.prgId == 0) return;
3326 linked_programs = &This->baseShader.linked_programs;
3328 TRACE("Deleting linked programs\n");
3329 if (linked_programs->next) {
3330 struct glsl_shader_prog_link *entry, *entry2;
3333 LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
3334 delete_glsl_program_entry(This->baseShader.device, entry);
3337 LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
3338 delete_glsl_program_entry(This->baseShader.device, entry);
3343 TRACE("Deleting shader object %u\n", This->baseShader.prgId);
3344 GL_EXTCALL(glDeleteObjectARB(This->baseShader.prgId));
3345 checkGLcall("glDeleteObjectARB");
3346 This->baseShader.prgId = 0;
3347 This->baseShader.is_compiled = FALSE;
3350 static unsigned int glsl_program_key_hash(void *key) {
3351 glsl_program_key_t *k = (glsl_program_key_t *)key;
3353 unsigned int hash = k->vshader | k->pshader << 16;
3354 hash += ~(hash << 15);
3355 hash ^= (hash >> 10);
3356 hash += (hash << 3);
3357 hash ^= (hash >> 6);
3358 hash += ~(hash << 11);
3359 hash ^= (hash >> 16);
3364 static BOOL glsl_program_key_compare(void *keya, void *keyb) {
3365 glsl_program_key_t *ka = (glsl_program_key_t *)keya;
3366 glsl_program_key_t *kb = (glsl_program_key_t *)keyb;
3368 return ka->vshader == kb->vshader && ka->pshader == kb->pshader;
3371 static HRESULT shader_glsl_alloc(IWineD3DDevice *iface) {
3372 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3373 This->shader_priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
3374 This->glsl_program_lookup = hash_table_create(&glsl_program_key_hash, &glsl_program_key_compare);
3378 static void shader_glsl_free(IWineD3DDevice *iface) {
3379 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3380 HeapFree(GetProcessHeap(), 0, This->shader_priv);
3381 This->shader_priv = NULL;
3384 static BOOL shader_glsl_dirty_const(IWineD3DDevice *iface) {
3385 /* TODO: GL_EXT_bindable_uniform can be used to share constants across shaders */
3389 static void shader_glsl_generate_pshader(IWineD3DPixelShader *iface, SHADER_BUFFER *buffer) {
3390 IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
3391 shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3392 CONST DWORD *function = This->baseShader.function;
3393 const char *fragcolor;
3394 WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3396 /* Create the hw GLSL shader object and assign it as the baseShader.prgId */
3397 GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3399 if (GL_SUPPORT(ARB_DRAW_BUFFERS)) {
3400 shader_addline(buffer, "#extension GL_ARB_draw_buffers : enable\n");
3402 if (GL_SUPPORT(ARB_TEXTURE_RECTANGLE)) {
3403 /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
3404 * drivers write a warning if we don't do so
3406 shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
3409 /* Base Declarations */
3410 shader_generate_glsl_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION);
3412 /* Pack 3.0 inputs */
3413 if (This->baseShader.hex_version >= WINED3DPS_VERSION(3,0)) {
3415 if(((IWineD3DDeviceImpl *) This->baseShader.device)->strided_streams.u.s.position_transformed) {
3416 This->vertexprocessing = pretransformed;
3417 pshader_glsl_input_pack(buffer, This->semantics_in, iface);
3418 } else if(!use_vs((IWineD3DDeviceImpl *) This->baseShader.device)) {
3419 This->vertexprocessing = fixedfunction;
3420 pshader_glsl_input_pack(buffer, This->semantics_in, iface);
3422 This->vertexprocessing = vertexshader;
3426 /* Base Shader Body */
3427 shader_generate_main( (IWineD3DBaseShader*) This, buffer, reg_maps, function);
3429 /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
3430 if (This->baseShader.hex_version < WINED3DPS_VERSION(2,0)) {
3431 /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
3432 if(GL_SUPPORT(ARB_DRAW_BUFFERS))
3433 shader_addline(buffer, "gl_FragData[0] = R0;\n");
3435 shader_addline(buffer, "gl_FragColor = R0;\n");
3438 if(GL_SUPPORT(ARB_DRAW_BUFFERS)) {
3439 fragcolor = "gl_FragData[0]";
3441 fragcolor = "gl_FragColor";
3443 if(This->srgb_enabled) {
3444 shader_addline(buffer, "tmp0.xyz = pow(%s.xyz, vec3(%f, %f, %f)) * vec3(%f, %f, %f) - vec3(%f, %f, %f);\n",
3445 fragcolor, srgb_pow, srgb_pow, srgb_pow, srgb_mul_high, srgb_mul_high, srgb_mul_high,
3446 srgb_sub_high, srgb_sub_high, srgb_sub_high);
3447 shader_addline(buffer, "tmp1.xyz = %s.xyz * srgb_mul_low.xyz;\n", fragcolor);
3448 shader_addline(buffer, "%s.x = %s.x < srgb_comparison.x ? tmp1.x : tmp0.x;\n", fragcolor, fragcolor);
3449 shader_addline(buffer, "%s.y = %s.y < srgb_comparison.y ? tmp1.y : tmp0.y;\n", fragcolor, fragcolor);
3450 shader_addline(buffer, "%s.z = %s.z < srgb_comparison.z ? tmp1.z : tmp0.z;\n", fragcolor, fragcolor);
3451 shader_addline(buffer, "%s = clamp(%s, 0.0, 1.0);\n", fragcolor, fragcolor);
3453 /* Pixel shader < 3.0 do not replace the fog stage.
3454 * This implements linear fog computation and blending.
3455 * TODO: non linear fog
3456 * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
3457 * -1/(e-s) and e/(e-s) respectively.
3459 if(This->baseShader.hex_version < WINED3DPS_VERSION(3,0)) {
3460 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * gl_Fog.start + gl_Fog.end, 0.0, 1.0);\n");
3461 shader_addline(buffer, "%s.xyz = mix(gl_Fog.color.xyz, %s.xyz, Fog);\n", fragcolor, fragcolor);
3464 shader_addline(buffer, "}\n");
3466 TRACE("Compiling shader object %u\n", shader_obj);
3467 GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
3468 GL_EXTCALL(glCompileShaderARB(shader_obj));
3469 print_glsl_info_log(&GLINFO_LOCATION, shader_obj);
3471 /* Store the shader object */
3472 This->baseShader.prgId = shader_obj;
3475 static void shader_glsl_generate_vshader(IWineD3DVertexShader *iface, SHADER_BUFFER *buffer) {
3476 IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
3477 shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3478 CONST DWORD *function = This->baseShader.function;
3479 WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3481 /* Create the hw GLSL shader program and assign it as the baseShader.prgId */
3482 GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3484 /* Base Declarations */
3485 shader_generate_glsl_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION);
3487 /* Base Shader Body */
3488 shader_generate_main( (IWineD3DBaseShader*) This, buffer, reg_maps, function);
3490 /* Unpack 3.0 outputs */
3491 if (This->baseShader.hex_version >= WINED3DVS_VERSION(3,0)) {
3492 shader_addline(buffer, "order_ps_input(OUT);\n");
3494 shader_addline(buffer, "order_ps_input();\n");
3497 /* If this shader doesn't use fog copy the z coord to the fog coord so that we can use table fog */
3499 shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
3501 /* Write the final position.
3503 * OpenGL coordinates specify the center of the pixel while d3d coords specify
3504 * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3505 * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3506 * contains 1.0 to allow a mad.
3508 shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
3509 shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
3511 /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3513 * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
3514 * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
3515 * which is the same as z = z / 2 - w.
3517 shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
3519 shader_addline(buffer, "}\n");
3521 TRACE("Compiling shader object %u\n", shader_obj);
3522 GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
3523 GL_EXTCALL(glCompileShaderARB(shader_obj));
3524 print_glsl_info_log(&GLINFO_LOCATION, shader_obj);
3526 /* Store the shader object */
3527 This->baseShader.prgId = shader_obj;
3530 static void shader_glsl_get_caps(WINED3DDEVTYPE devtype, WineD3D_GL_Info *gl_info, struct shader_caps *pCaps) {
3531 /* We don't have a GLSL fixed function pipeline yet, so let the none backend set its caps,
3532 * then overwrite the shader specific ones
3534 none_shader_backend.shader_get_caps(devtype, gl_info, pCaps);
3536 /* Nvidia Geforce6/7 or Ati R4xx/R5xx cards with GLSL support, support VS 3.0 but older Nvidia/Ati
3537 * models with GLSL support only support 2.0. In case of nvidia we can detect VS 2.0 support using
3538 * vs_nv_version which is based on NV_vertex_program.
3539 * For Ati cards there's no way using glsl (it abstracts the lowlevel info away) and also not
3540 * using ARB_vertex_program. It is safe to assume that when a card supports pixel shader 2.0 it
3541 * supports vertex shader 2.0 too and the way around. We can detect ps2.0 using the maximum number
3542 * of native instructions, so use that here. For more info see the pixel shader versioning code below.
3544 if((GLINFO_LOCATION.vs_nv_version == VS_VERSION_20) || (GLINFO_LOCATION.ps_arb_max_instructions <= 512))
3545 pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
3547 pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
3548 TRACE_(d3d_caps)("Hardware vertex shader version %d.%d enabled (GLSL)\n", (pCaps->VertexShaderVersion >> 8) & 0xff, pCaps->VertexShaderVersion & 0xff);
3549 pCaps->MaxVertexShaderConst = GL_LIMITS(vshader_constantsF);
3551 /* Older DX9-class videocards (GeforceFX / Radeon >9500/X*00) only support pixel shader 2.0/2.0a/2.0b.
3552 * In OpenGL the extensions related to GLSL abstract lowlevel GL info away which is needed
3553 * to distinguish between 2.0 and 3.0 (and 2.0a/2.0b). In case of Nvidia we use their fragment
3554 * program extensions. On other hardware including ATI GL_ARB_fragment_program offers the info
3555 * in max native instructions. Intel and others also offer the info in this extension but they
3556 * don't support GLSL (at least on Windows).
3558 * PS2.0 requires at least 96 instructions, 2.0a/2.0b go up to 512. Assume that if the number
3559 * of instructions is 512 or less we have to do with ps2.0 hardware.
3560 * 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.
3562 if((GLINFO_LOCATION.ps_nv_version == PS_VERSION_20) || (GLINFO_LOCATION.ps_arb_max_instructions <= 512))
3563 pCaps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
3565 pCaps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
3567 /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
3568 * Direct3D minimum requirement.
3570 * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
3571 * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
3573 * The problem is that the refrast clamps temporary results in the shader to
3574 * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
3575 * then applications may miss the clamping behavior. On the other hand, if it is smaller,
3576 * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
3577 * offer a way to query this.
3579 pCaps->PixelShader1xMaxValue = 8.0;
3580 TRACE_(d3d_caps)("Hardware pixel shader version %d.%d enabled (GLSL)\n", (pCaps->PixelShaderVersion >> 8) & 0xff, pCaps->PixelShaderVersion & 0xff);
3583 static void shader_glsl_load_init(void) {}
3585 const shader_backend_t glsl_shader_backend = {
3586 &shader_glsl_select,
3587 &shader_glsl_select_depth_blt,
3588 &shader_glsl_destroy_depth_blt,
3589 &shader_glsl_load_constants,
3590 &shader_glsl_cleanup,
3591 &shader_glsl_color_correction,
3592 &shader_glsl_destroy,
3595 &shader_glsl_dirty_const,
3596 &shader_glsl_generate_pshader,
3597 &shader_glsl_generate_vshader,
3598 &shader_glsl_get_caps,
3599 &shader_glsl_load_init,