wined3d: Use a hashmap to store the ffp shaders.
[wine/gsoc_dplay.git] / dlls / wined3d / glsl_shader.c
blob80c4411fbd844f484cc473107330c5a7d9d89f90
1 /*
2 * GLSL pixel and vertex shader implementation
4 * Copyright 2006 Jason Green
5 * Copyright 2006-2007 Henri Verbeet
6 * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
8 * This library is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * This library is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with this library; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
24 * D3D shader asm has swizzles on source parameters, and write masks for
25 * destination parameters. GLSL uses swizzles for both. The result of this is
26 * that for example "mov dst.xw, src.zyxw" becomes "dst.xw = src.zw" in GLSL.
27 * Ie, to generate a proper GLSL source swizzle, we need to take the D3D write
28 * mask for the destination parameter into account.
31 #include "config.h"
32 #include <stdio.h>
33 #include "wined3d_private.h"
35 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
36 WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
37 WINE_DECLARE_DEBUG_CHANNEL(d3d_caps);
39 #define GLINFO_LOCATION (*gl_info)
41 typedef struct {
42 char reg_name[150];
43 char mask_str[6];
44 } glsl_dst_param_t;
46 typedef struct {
47 char reg_name[150];
48 char param_str[100];
49 } glsl_src_param_t;
51 typedef struct {
52 const char *name;
53 DWORD coord_mask;
54 } glsl_sample_function_t;
56 /** Prints the GLSL info log which will contain error messages if they exist */
57 void print_glsl_info_log(WineD3D_GL_Info *gl_info, GLhandleARB obj) {
59 int infologLength = 0;
60 char *infoLog;
61 int i;
62 BOOL is_spam;
64 const char *spam[] = {
65 "Vertex shader was successfully compiled to run on hardware.\n", /* fglrx */
66 "Fragment shader was successfully compiled to run on hardware.\n", /* fglrx */
67 "Fragment shader(s) linked, vertex shader(s) linked. \n ", /* fglrx, with \n */
68 "Fragment shader(s) linked, vertex shader(s) linked.", /* fglrx, no \n */
69 "Vertex shader(s) linked, no fragment shader(s) defined. \n ", /* fglrx, with \n */
70 "Vertex shader(s) linked, no fragment shader(s) defined.", /* fglrx, no \n */
71 "Fragment shader was successfully compiled to run on hardware.\nWARNING: 0:1: extension 'GL_ARB_draw_buffers' is not supported",
72 "Fragment shader(s) linked, no vertex shader(s) defined.", /* fglrx, no \n */
73 "Fragment shader(s) linked, no vertex shader(s) defined. \n " /* fglrx, with \n */
76 GL_EXTCALL(glGetObjectParameterivARB(obj,
77 GL_OBJECT_INFO_LOG_LENGTH_ARB,
78 &infologLength));
80 /* A size of 1 is just a null-terminated string, so the log should be bigger than
81 * that if there are errors. */
82 if (infologLength > 1)
84 /* Fglrx doesn't terminate the string properly, but it tells us the proper length.
85 * So use HEAP_ZERO_MEMORY to avoid uninitialized bytes
87 infoLog = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, infologLength);
88 GL_EXTCALL(glGetInfoLogARB(obj, infologLength, NULL, infoLog));
89 is_spam = FALSE;
91 for(i = 0; i < sizeof(spam) / sizeof(spam[0]); i++) {
92 if(strcmp(infoLog, spam[i]) == 0) {
93 is_spam = TRUE;
94 break;
97 if(is_spam) {
98 TRACE("Spam received from GLSL shader #%u: %s\n", obj, debugstr_a(infoLog));
99 } else {
100 FIXME("Error received from GLSL shader #%u: %s\n", obj, debugstr_a(infoLog));
102 HeapFree(GetProcessHeap(), 0, infoLog);
107 * Loads (pixel shader) samplers
109 static void shader_glsl_load_psamplers(
110 WineD3D_GL_Info *gl_info,
111 IWineD3DStateBlock* iface,
112 GLhandleARB programId) {
114 IWineD3DStateBlockImpl* stateBlock = (IWineD3DStateBlockImpl*) iface;
115 GLhandleARB name_loc;
116 int i;
117 char sampler_name[20];
119 for (i = 0; i < MAX_FRAGMENT_SAMPLERS; ++i) {
120 snprintf(sampler_name, sizeof(sampler_name), "Psampler%d", i);
121 name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
122 if (name_loc != -1) {
123 int mapped_unit = stateBlock->wineD3DDevice->texUnitMap[i];
124 if (mapped_unit != -1 && mapped_unit < GL_LIMITS(fragment_samplers)) {
125 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
126 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
127 checkGLcall("glUniform1iARB");
128 } else {
129 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
135 static void shader_glsl_load_vsamplers(WineD3D_GL_Info *gl_info, IWineD3DStateBlock* iface, GLhandleARB programId) {
136 IWineD3DStateBlockImpl* stateBlock = (IWineD3DStateBlockImpl*) iface;
137 GLhandleARB name_loc;
138 char sampler_name[20];
139 int i;
141 for (i = 0; i < MAX_VERTEX_SAMPLERS; ++i) {
142 snprintf(sampler_name, sizeof(sampler_name), "Vsampler%d", i);
143 name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
144 if (name_loc != -1) {
145 int mapped_unit = stateBlock->wineD3DDevice->texUnitMap[MAX_FRAGMENT_SAMPLERS + i];
146 if (mapped_unit != -1 && mapped_unit < GL_LIMITS(combined_samplers)) {
147 TRACE("Loading %s for texture %d\n", sampler_name, mapped_unit);
148 GL_EXTCALL(glUniform1iARB(name_loc, mapped_unit));
149 checkGLcall("glUniform1iARB");
150 } else {
151 ERR("Trying to load sampler %s on unsupported unit %d\n", sampler_name, mapped_unit);
157 /**
158 * Loads floating point constants (aka uniforms) into the currently set GLSL program.
159 * When constant_list == NULL, it will load all the constants.
161 static void shader_glsl_load_constantsF(IWineD3DBaseShaderImpl* This, WineD3D_GL_Info *gl_info,
162 unsigned int max_constants, float* constants, GLhandleARB *constant_locations,
163 struct list *constant_list) {
164 constants_entry *constant;
165 local_constant* lconst;
166 GLhandleARB tmp_loc;
167 DWORD i, j, k;
168 DWORD *idx;
170 if (TRACE_ON(d3d_shader)) {
171 LIST_FOR_EACH_ENTRY(constant, constant_list, constants_entry, entry) {
172 idx = constant->idx;
173 j = constant->count;
174 while (j--) {
175 i = *idx++;
176 tmp_loc = constant_locations[i];
177 if (tmp_loc != -1) {
178 TRACE_(d3d_constants)("Loading constants %i: %f, %f, %f, %f\n", i,
179 constants[i * 4 + 0], constants[i * 4 + 1],
180 constants[i * 4 + 2], constants[i * 4 + 3]);
186 /* 1.X pshaders have the constants clamped to [-1;1] implicitly. */
187 if(WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) == 1 &&
188 shader_is_pshader_version(This->baseShader.hex_version)) {
189 float lcl_const[4];
191 LIST_FOR_EACH_ENTRY(constant, constant_list, constants_entry, entry) {
192 idx = constant->idx;
193 j = constant->count;
194 while (j--) {
195 i = *idx++;
196 tmp_loc = constant_locations[i];
197 if (tmp_loc != -1) {
198 /* We found this uniform name in the program - go ahead and send the data */
199 k = i * 4;
200 if(constants[k + 0] < -1.0) lcl_const[0] = -1.0;
201 else if(constants[k + 0] > 1.0) lcl_const[0] = 1.0;
202 else lcl_const[0] = constants[k + 0];
203 if(constants[k + 1] < -1.0) lcl_const[1] = -1.0;
204 else if(constants[k + 1] > 1.0) lcl_const[1] = 1.0;
205 else lcl_const[1] = constants[k + 1];
206 if(constants[k + 2] < -1.0) lcl_const[2] = -1.0;
207 else if(constants[k + 2] > 1.0) lcl_const[2] = 1.0;
208 else lcl_const[2] = constants[k + 2];
209 if(constants[k + 3] < -1.0) lcl_const[3] = -1.0;
210 else if(constants[k + 3] > 1.0) lcl_const[3] = 1.0;
211 else lcl_const[3] = constants[k + 3];
213 GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, lcl_const));
217 } else {
218 LIST_FOR_EACH_ENTRY(constant, constant_list, constants_entry, entry) {
219 idx = constant->idx;
220 j = constant->count;
221 while (j--) {
222 i = *idx++;
223 tmp_loc = constant_locations[i];
224 if (tmp_loc != -1) {
225 /* We found this uniform name in the program - go ahead and send the data */
226 GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, constants + (i * 4)));
231 checkGLcall("glUniform4fvARB()");
233 if(!This->baseShader.load_local_constsF) {
234 TRACE("No need to load local float constants for this shader\n");
235 return;
238 /* Load immediate constants */
239 if (TRACE_ON(d3d_shader)) {
240 LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
241 tmp_loc = constant_locations[lconst->idx];
242 if (tmp_loc != -1) {
243 GLfloat* values = (GLfloat*)lconst->value;
244 TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
245 values[0], values[1], values[2], values[3]);
249 /* Immediate constants are clamped to [-1;1] at shader creation time if needed */
250 LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
251 tmp_loc = constant_locations[lconst->idx];
252 if (tmp_loc != -1) {
253 /* We found this uniform name in the program - go ahead and send the data */
254 GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, (GLfloat*)lconst->value));
257 checkGLcall("glUniform4fvARB()");
260 /**
261 * Loads integer constants (aka uniforms) into the currently set GLSL program.
262 * When @constants_set == NULL, it will load all the constants.
264 static void shader_glsl_load_constantsI(
265 IWineD3DBaseShaderImpl* This,
266 WineD3D_GL_Info *gl_info,
267 GLhandleARB programId,
268 GLhandleARB locations[MAX_CONST_I],
269 unsigned max_constants,
270 int* constants,
271 BOOL* constants_set) {
273 int i;
274 struct list* ptr;
276 for (i=0; i<max_constants; ++i) {
277 if (NULL == constants_set || constants_set[i]) {
279 TRACE_(d3d_constants)("Loading constants %i: %i, %i, %i, %i\n",
280 i, constants[i*4], constants[i*4+1], constants[i*4+2], constants[i*4+3]);
282 /* We found this uniform name in the program - go ahead and send the data */
283 GL_EXTCALL(glUniform4ivARB(locations[i], 1, &constants[i*4]));
284 checkGLcall("glUniform4ivARB");
288 /* Load immediate constants */
289 ptr = list_head(&This->baseShader.constantsI);
290 while (ptr) {
291 local_constant* lconst = LIST_ENTRY(ptr, struct local_constant, entry);
292 unsigned int idx = lconst->idx;
293 GLint* values = (GLint*) lconst->value;
295 TRACE_(d3d_constants)("Loading local constants %i: %i, %i, %i, %i\n", idx,
296 values[0], values[1], values[2], values[3]);
298 /* We found this uniform name in the program - go ahead and send the data */
299 GL_EXTCALL(glUniform4ivARB(locations[idx], 1, values));
300 checkGLcall("glUniform4ivARB");
301 ptr = list_next(&This->baseShader.constantsI, ptr);
305 /**
306 * Loads boolean constants (aka uniforms) into the currently set GLSL program.
307 * When @constants_set == NULL, it will load all the constants.
309 static void shader_glsl_load_constantsB(
310 IWineD3DBaseShaderImpl* This,
311 WineD3D_GL_Info *gl_info,
312 GLhandleARB programId,
313 unsigned max_constants,
314 BOOL* constants,
315 BOOL* constants_set) {
317 GLhandleARB tmp_loc;
318 int i;
319 char tmp_name[8];
320 char is_pshader = shader_is_pshader_version(This->baseShader.hex_version);
321 const char* prefix = is_pshader? "PB":"VB";
322 struct list* ptr;
324 for (i=0; i<max_constants; ++i) {
325 if (NULL == constants_set || constants_set[i]) {
327 TRACE_(d3d_constants)("Loading constants %i: %i;\n", i, constants[i]);
329 /* TODO: Benchmark and see if it would be beneficial to store the
330 * locations of the constants to avoid looking up each time */
331 snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, i);
332 tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
333 if (tmp_loc != -1) {
334 /* We found this uniform name in the program - go ahead and send the data */
335 GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, &constants[i]));
336 checkGLcall("glUniform1ivARB");
341 /* Load immediate constants */
342 ptr = list_head(&This->baseShader.constantsB);
343 while (ptr) {
344 local_constant* lconst = LIST_ENTRY(ptr, struct local_constant, entry);
345 unsigned int idx = lconst->idx;
346 GLint* values = (GLint*) lconst->value;
348 TRACE_(d3d_constants)("Loading local constants %i: %i\n", idx, values[0]);
350 snprintf(tmp_name, sizeof(tmp_name), "%s[%i]", prefix, idx);
351 tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, tmp_name));
352 if (tmp_loc != -1) {
353 /* We found this uniform name in the program - go ahead and send the data */
354 GL_EXTCALL(glUniform1ivARB(tmp_loc, 1, values));
355 checkGLcall("glUniform1ivARB");
357 ptr = list_next(&This->baseShader.constantsB, ptr);
364 * Loads the app-supplied constants into the currently set GLSL program.
366 void shader_glsl_load_constants(
367 IWineD3DDevice* device,
368 char usePixelShader,
369 char useVertexShader) {
371 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) device;
372 struct shader_glsl_priv *priv = (struct shader_glsl_priv *)deviceImpl->shader_priv;
373 IWineD3DStateBlockImpl* stateBlock = deviceImpl->stateBlock;
374 WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
376 GLhandleARB *constant_locations;
377 struct list *constant_list;
378 GLhandleARB programId;
379 struct glsl_shader_prog_link *prog = priv->glsl_program;
380 unsigned int i;
382 if (!prog) {
383 /* No GLSL program set - nothing to do. */
384 return;
386 programId = prog->programId;
388 if (useVertexShader) {
389 IWineD3DBaseShaderImpl* vshader = (IWineD3DBaseShaderImpl*) stateBlock->vertexShader;
391 constant_locations = prog->vuniformF_locations;
392 constant_list = &stateBlock->set_vconstantsF;
394 /* Load DirectX 9 float constants/uniforms for vertex shader */
395 shader_glsl_load_constantsF(vshader, gl_info, GL_LIMITS(vshader_constantsF),
396 stateBlock->vertexShaderConstantF, constant_locations, constant_list);
398 /* Load DirectX 9 integer constants/uniforms for vertex shader */
399 shader_glsl_load_constantsI(vshader, gl_info, programId,
400 prog->vuniformI_locations, MAX_CONST_I,
401 stateBlock->vertexShaderConstantI,
402 stateBlock->changed.vertexShaderConstantsI);
404 /* Load DirectX 9 boolean constants/uniforms for vertex shader */
405 shader_glsl_load_constantsB(vshader, gl_info, programId, MAX_CONST_B,
406 stateBlock->vertexShaderConstantB,
407 stateBlock->changed.vertexShaderConstantsB);
409 /* Upload the position fixup params */
410 GL_EXTCALL(glUniform4fvARB(prog->posFixup_location, 1, &deviceImpl->posFixup[0]));
411 checkGLcall("glUniform4fvARB");
414 if (usePixelShader) {
416 IWineD3DBaseShaderImpl* pshader = (IWineD3DBaseShaderImpl*) stateBlock->pixelShader;
418 constant_locations = prog->puniformF_locations;
419 constant_list = &stateBlock->set_pconstantsF;
421 /* Load DirectX 9 float constants/uniforms for pixel shader */
422 shader_glsl_load_constantsF(pshader, gl_info, GL_LIMITS(pshader_constantsF),
423 stateBlock->pixelShaderConstantF, constant_locations, constant_list);
425 /* Load DirectX 9 integer constants/uniforms for pixel shader */
426 shader_glsl_load_constantsI(pshader, gl_info, programId,
427 prog->puniformI_locations, MAX_CONST_I,
428 stateBlock->pixelShaderConstantI,
429 stateBlock->changed.pixelShaderConstantsI);
431 /* Load DirectX 9 boolean constants/uniforms for pixel shader */
432 shader_glsl_load_constantsB(pshader, gl_info, programId, MAX_CONST_B,
433 stateBlock->pixelShaderConstantB,
434 stateBlock->changed.pixelShaderConstantsB);
436 /* Upload the environment bump map matrix if needed. The needsbumpmat member specifies the texture stage to load the matrix from.
437 * It can't be 0 for a valid texbem instruction.
439 for(i = 0; i < ((IWineD3DPixelShaderImpl *) pshader)->numbumpenvmatconsts; i++) {
440 IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pshader;
441 int stage = ps->luminanceconst[i].texunit;
443 float *data = (float *) &stateBlock->textureState[(int) ps->bumpenvmatconst[i].texunit][WINED3DTSS_BUMPENVMAT00];
444 GL_EXTCALL(glUniformMatrix2fvARB(prog->bumpenvmat_location[i], 1, 0, data));
445 checkGLcall("glUniformMatrix2fvARB");
447 /* texbeml needs the luminance scale and offset too. If texbeml is used, needsbumpmat
448 * is set too, so we can check that in the needsbumpmat check
450 if(ps->baseShader.reg_maps.luminanceparams[stage]) {
451 GLfloat *scale = (GLfloat *) &stateBlock->textureState[stage][WINED3DTSS_BUMPENVLSCALE];
452 GLfloat *offset = (GLfloat *) &stateBlock->textureState[stage][WINED3DTSS_BUMPENVLOFFSET];
454 GL_EXTCALL(glUniform1fvARB(prog->luminancescale_location[i], 1, scale));
455 checkGLcall("glUniform1fvARB");
456 GL_EXTCALL(glUniform1fvARB(prog->luminanceoffset_location[i], 1, offset));
457 checkGLcall("glUniform1fvARB");
461 if(((IWineD3DPixelShaderImpl *) pshader)->srgb_enabled &&
462 !((IWineD3DPixelShaderImpl *) pshader)->srgb_mode_hardcoded) {
463 float comparison[4];
464 float mul_low[4];
466 if(stateBlock->renderState[WINED3DRS_SRGBWRITEENABLE]) {
467 comparison[0] = srgb_cmp; comparison[1] = srgb_cmp;
468 comparison[2] = srgb_cmp; comparison[3] = srgb_cmp;
470 mul_low[0] = srgb_mul_low; mul_low[1] = srgb_mul_low;
471 mul_low[2] = srgb_mul_low; mul_low[3] = srgb_mul_low;
472 } else {
473 comparison[0] = 1.0 / 0.0; comparison[1] = 1.0 / 0.0;
474 comparison[2] = 1.0 / 0.0; comparison[3] = 1.0 / 0.0;
476 mul_low[0] = 1.0; mul_low[1] = 1.0;
477 mul_low[2] = 1.0; mul_low[3] = 1.0;
480 GL_EXTCALL(glUniform4fvARB(prog->srgb_comparison_location, 1, comparison));
481 GL_EXTCALL(glUniform4fvARB(prog->srgb_mul_low_location, 1, mul_low));
483 if(((IWineD3DPixelShaderImpl *) pshader)->vpos_uniform) {
484 float correction_params[4];
485 if(deviceImpl->render_offscreen) {
486 correction_params[0] = 0.0;
487 correction_params[1] = 1.0;
488 } else {
489 /* position is window relative, not viewport relative */
490 correction_params[0] = ((IWineD3DSurfaceImpl *) deviceImpl->render_targets[0])->currentDesc.Height;
491 correction_params[1] = -1.0;
493 GL_EXTCALL(glUniform4fvARB(prog->ycorrection_location, 1, correction_params));
498 /** Generate the variable & register declarations for the GLSL output target */
499 void shader_generate_glsl_declarations(
500 IWineD3DBaseShader *iface,
501 shader_reg_maps* reg_maps,
502 SHADER_BUFFER* buffer,
503 WineD3D_GL_Info* gl_info) {
505 IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
506 IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *) This->baseShader.device;
507 int i;
508 unsigned int extra_constants_needed = 0;
509 local_constant* lconst;
511 /* There are some minor differences between pixel and vertex shaders */
512 char pshader = shader_is_pshader_version(This->baseShader.hex_version);
513 char prefix = pshader ? 'P' : 'V';
515 /* Prototype the subroutines */
516 for (i = 0; i < This->baseShader.limits.label; i++) {
517 if (reg_maps->labels[i])
518 shader_addline(buffer, "void subroutine%u();\n", i);
521 /* Declare the constants (aka uniforms) */
522 if (This->baseShader.limits.constant_float > 0) {
523 unsigned max_constantsF = min(This->baseShader.limits.constant_float,
524 (pshader ? GL_LIMITS(pshader_constantsF) : GL_LIMITS(vshader_constantsF)));
525 shader_addline(buffer, "uniform vec4 %cC[%u];\n", prefix, max_constantsF);
528 if (This->baseShader.limits.constant_int > 0)
529 shader_addline(buffer, "uniform ivec4 %cI[%u];\n", prefix, This->baseShader.limits.constant_int);
531 if (This->baseShader.limits.constant_bool > 0)
532 shader_addline(buffer, "uniform bool %cB[%u];\n", prefix, This->baseShader.limits.constant_bool);
534 if(!pshader) {
535 shader_addline(buffer, "uniform vec4 posFixup;\n");
536 /* Predeclaration; This function is added at link time based on the pixel shader.
537 * VS 3.0 shaders have an array OUT[] the shader writes to, earlier versions don't have
538 * that. We know the input to the reorder function at vertex shader compile time, so
539 * we can deal with that. The reorder function for a 1.x and 2.x vertex shader can just
540 * read gl_FrontColor. The output depends on the pixel shader. The reorder function for a
541 * 1.x and 2.x pshader or for fixed function will write gl_FrontColor, and for a 3.0 shader
542 * it will write to the varying array. Here we depend on the shader optimizer on sorting that
543 * out. The nvidia driver only does that if the parameter is inout instead of out, hence the
544 * inout.
546 if(This->baseShader.hex_version >= WINED3DVS_VERSION(3, 0)) {
547 shader_addline(buffer, "void order_ps_input(in vec4[%u]);\n", MAX_REG_OUTPUT);
548 } else {
549 shader_addline(buffer, "void order_ps_input();\n");
551 } else {
552 IWineD3DPixelShaderImpl *ps_impl = (IWineD3DPixelShaderImpl *) This;
554 ps_impl->numbumpenvmatconsts = 0;
555 for(i = 0; i < (sizeof(reg_maps->bumpmat) / sizeof(reg_maps->bumpmat[0])); i++) {
556 if(!reg_maps->bumpmat[i]) {
557 continue;
560 ps_impl->bumpenvmatconst[(int) ps_impl->numbumpenvmatconsts].texunit = i;
561 shader_addline(buffer, "uniform mat2 bumpenvmat%d;\n", i);
563 if(reg_maps->luminanceparams) {
564 ps_impl->luminanceconst[(int) ps_impl->numbumpenvmatconsts].texunit = i;
565 shader_addline(buffer, "uniform float luminancescale%d;\n", i);
566 shader_addline(buffer, "uniform float luminanceoffset%d;\n", i);
567 extra_constants_needed++;
568 } else {
569 ps_impl->luminanceconst[(int) ps_impl->numbumpenvmatconsts].texunit = -1;
572 extra_constants_needed++;
573 ps_impl->numbumpenvmatconsts++;
576 if(device->stateBlock->renderState[WINED3DRS_SRGBWRITEENABLE]) {
577 ps_impl->srgb_enabled = 1;
578 if(This->baseShader.limits.constant_float + extra_constants_needed + 1 < GL_LIMITS(pshader_constantsF)) {
579 shader_addline(buffer, "uniform vec4 srgb_mul_low;\n");
580 shader_addline(buffer, "uniform vec4 srgb_comparison;\n");
581 ps_impl->srgb_mode_hardcoded = 0;
582 extra_constants_needed++;
583 } else {
584 ps_impl->srgb_mode_hardcoded = 1;
585 shader_addline(buffer, "const vec4 srgb_mul_low = vec4(%f, %f, %f, %f);\n",
586 srgb_mul_low, srgb_mul_low, srgb_mul_low, srgb_mul_low);
587 shader_addline(buffer, "const vec4 srgb_comparison = vec4(%f, %f, %f, %f);\n",
588 srgb_cmp, srgb_cmp, srgb_cmp, srgb_cmp);
590 } else {
591 IWineD3DPixelShaderImpl *ps_impl = (IWineD3DPixelShaderImpl *) This;
593 /* Do not write any srgb fixup into the shader to save shader size and processing time.
594 * As a consequence, we can't toggle srgb write on without recompilation
596 ps_impl->srgb_enabled = 0;
597 ps_impl->srgb_mode_hardcoded = 1;
599 if(reg_maps->vpos || reg_maps->usesdsy) {
600 if(This->baseShader.limits.constant_float + extra_constants_needed + 1 < GL_LIMITS(pshader_constantsF)) {
601 shader_addline(buffer, "uniform vec4 ycorrection;\n");
602 ((IWineD3DPixelShaderImpl *) This)->vpos_uniform = 1;
603 extra_constants_needed++;
604 } else {
605 /* This happens because we do not have proper tracking of the constant registers that are
606 * actually used, only the max limit of the shader version
608 FIXME("Cannot find a free uniform for vpos correction params\n");
609 shader_addline(buffer, "const vec4 ycorrection = vec4(%f, %f, 0.0, 0.0);\n",
610 device->render_offscreen ? 0.0 : ((IWineD3DSurfaceImpl *) device->render_targets[0])->currentDesc.Height,
611 device->render_offscreen ? 1.0 : -1.0);
613 shader_addline(buffer, "vec4 vpos;\n");
617 /* Declare texture samplers */
618 for (i = 0; i < This->baseShader.limits.sampler; i++) {
619 if (reg_maps->samplers[i]) {
621 DWORD stype = reg_maps->samplers[i] & WINED3DSP_TEXTURETYPE_MASK;
622 switch (stype) {
624 case WINED3DSTT_1D:
625 shader_addline(buffer, "uniform sampler1D %csampler%u;\n", prefix, i);
626 break;
627 case WINED3DSTT_2D:
628 if(device->stateBlock->textures[i] &&
629 IWineD3DBaseTexture_GetTextureDimensions(device->stateBlock->textures[i]) == GL_TEXTURE_RECTANGLE_ARB) {
630 shader_addline(buffer, "uniform sampler2DRect %csampler%u;\n", prefix, i);
631 } else {
632 shader_addline(buffer, "uniform sampler2D %csampler%u;\n", prefix, i);
634 break;
635 case WINED3DSTT_CUBE:
636 shader_addline(buffer, "uniform samplerCube %csampler%u;\n", prefix, i);
637 break;
638 case WINED3DSTT_VOLUME:
639 shader_addline(buffer, "uniform sampler3D %csampler%u;\n", prefix, i);
640 break;
641 default:
642 shader_addline(buffer, "uniform unsupported_sampler %csampler%u;\n", prefix, i);
643 FIXME("Unrecognized sampler type: %#x\n", stype);
644 break;
649 /* Declare address variables */
650 for (i = 0; i < This->baseShader.limits.address; i++) {
651 if (reg_maps->address[i])
652 shader_addline(buffer, "ivec4 A%d;\n", i);
655 /* Declare texture coordinate temporaries and initialize them */
656 for (i = 0; i < This->baseShader.limits.texcoord; i++) {
657 if (reg_maps->texcoord[i])
658 shader_addline(buffer, "vec4 T%u = gl_TexCoord[%u];\n", i, i);
661 /* Declare input register varyings. Only pixel shader, vertex shaders have that declared in the
662 * helper function shader that is linked in at link time
664 if(pshader && This->baseShader.hex_version >= WINED3DPS_VERSION(3, 0)) {
665 if(use_vs(device)) {
666 shader_addline(buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
667 } else {
668 /* TODO: Write a replacement shader for the fixed function vertex pipeline, so this isn't needed.
669 * For fixed function vertex processing + 3.0 pixel shader we need a separate function in the
670 * pixel shader that reads the fixed function color into the packed input registers.
672 shader_addline(buffer, "vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
676 /* Declare output register temporaries */
677 if(This->baseShader.limits.packed_output) {
678 shader_addline(buffer, "vec4 OUT[%u];\n", This->baseShader.limits.packed_output);
681 /* Declare temporary variables */
682 for(i = 0; i < This->baseShader.limits.temporary; i++) {
683 if (reg_maps->temporary[i])
684 shader_addline(buffer, "vec4 R%u;\n", i);
687 /* Declare attributes */
688 for (i = 0; i < This->baseShader.limits.attributes; i++) {
689 if (reg_maps->attributes[i])
690 shader_addline(buffer, "attribute vec4 attrib%i;\n", i);
693 /* Declare loop registers aLx */
694 for (i = 0; i < reg_maps->loop_depth; i++) {
695 shader_addline(buffer, "int aL%u;\n", i);
696 shader_addline(buffer, "int tmpInt%u;\n", i);
699 /* Temporary variables for matrix operations */
700 shader_addline(buffer, "vec4 tmp0;\n");
701 shader_addline(buffer, "vec4 tmp1;\n");
703 /* Local constants use a different name so they can be loaded once at shader link time
704 * They can't be hardcoded into the shader text via LC = {x, y, z, w}; because the
705 * float -> string conversion can cause precision loss.
707 if(!This->baseShader.load_local_constsF) {
708 LIST_FOR_EACH_ENTRY(lconst, &This->baseShader.constantsF, local_constant, entry) {
709 shader_addline(buffer, "uniform vec4 %cLC%u;\n", prefix, lconst->idx);
713 /* Start the main program */
714 shader_addline(buffer, "void main() {\n");
715 if(pshader && reg_maps->vpos) {
716 /* DirectX apps expect integer values, while OpenGL drivers add approximately 0.5. This causes
717 * off-by-one problems as spotted by the vPos d3d9 visual test. Unfortunately the ATI cards do
718 * not add exactly 0.5, but rather something like 0.49999999 or 0.50000001, which still causes
719 * precision troubles when we just substract 0.5.
721 * To deal with that just floor() the position. This will eliminate the fraction on all cards.
723 * TODO: Test how that behaves with multisampling once we can enable multisampling in winex11.
725 * An advantage of floor is that it works even if the driver doesn't add 1/2. It is somewhat
726 * questionable if 1.5, 2.5, ... are the proper values to return in gl_FragCoord, even though
727 * coordinates specify the pixel centers instead of the pixel corners. This code will behave
728 * correctly on drivers that returns integer values.
730 shader_addline(buffer, "vpos = floor(vec4(0, ycorrection[0], 0, 0) + gl_FragCoord * vec4(1, ycorrection[1], 1, 1));\n");
734 /*****************************************************************************
735 * Functions to generate GLSL strings from DirectX Shader bytecode begin here.
737 * For more information, see http://wiki.winehq.org/DirectX-Shaders
738 ****************************************************************************/
740 /* Prototypes */
741 static void shader_glsl_add_src_param(SHADER_OPCODE_ARG* arg, const DWORD param,
742 const DWORD addr_token, DWORD mask, glsl_src_param_t *src_param);
744 /** Used for opcode modifiers - They multiply the result by the specified amount */
745 static const char * const shift_glsl_tab[] = {
746 "", /* 0 (none) */
747 "2.0 * ", /* 1 (x2) */
748 "4.0 * ", /* 2 (x4) */
749 "8.0 * ", /* 3 (x8) */
750 "16.0 * ", /* 4 (x16) */
751 "32.0 * ", /* 5 (x32) */
752 "", /* 6 (x64) */
753 "", /* 7 (x128) */
754 "", /* 8 (d256) */
755 "", /* 9 (d128) */
756 "", /* 10 (d64) */
757 "", /* 11 (d32) */
758 "0.0625 * ", /* 12 (d16) */
759 "0.125 * ", /* 13 (d8) */
760 "0.25 * ", /* 14 (d4) */
761 "0.5 * " /* 15 (d2) */
764 /* Generate a GLSL parameter that does the input modifier computation and return the input register/mask to use */
765 static void shader_glsl_gen_modifier (
766 const DWORD instr,
767 const char *in_reg,
768 const char *in_regswizzle,
769 char *out_str) {
771 out_str[0] = 0;
773 if (instr == WINED3DSIO_TEXKILL)
774 return;
776 switch (instr & WINED3DSP_SRCMOD_MASK) {
777 case WINED3DSPSM_DZ: /* Need to handle this in the instructions itself (texld & texcrd). */
778 case WINED3DSPSM_DW:
779 case WINED3DSPSM_NONE:
780 sprintf(out_str, "%s%s", in_reg, in_regswizzle);
781 break;
782 case WINED3DSPSM_NEG:
783 sprintf(out_str, "-%s%s", in_reg, in_regswizzle);
784 break;
785 case WINED3DSPSM_NOT:
786 sprintf(out_str, "!%s%s", in_reg, in_regswizzle);
787 break;
788 case WINED3DSPSM_BIAS:
789 sprintf(out_str, "(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
790 break;
791 case WINED3DSPSM_BIASNEG:
792 sprintf(out_str, "-(%s%s - vec4(0.5)%s)", in_reg, in_regswizzle, in_regswizzle);
793 break;
794 case WINED3DSPSM_SIGN:
795 sprintf(out_str, "(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
796 break;
797 case WINED3DSPSM_SIGNNEG:
798 sprintf(out_str, "-(2.0 * (%s%s - 0.5))", in_reg, in_regswizzle);
799 break;
800 case WINED3DSPSM_COMP:
801 sprintf(out_str, "(1.0 - %s%s)", in_reg, in_regswizzle);
802 break;
803 case WINED3DSPSM_X2:
804 sprintf(out_str, "(2.0 * %s%s)", in_reg, in_regswizzle);
805 break;
806 case WINED3DSPSM_X2NEG:
807 sprintf(out_str, "-(2.0 * %s%s)", in_reg, in_regswizzle);
808 break;
809 case WINED3DSPSM_ABS:
810 sprintf(out_str, "abs(%s%s)", in_reg, in_regswizzle);
811 break;
812 case WINED3DSPSM_ABSNEG:
813 sprintf(out_str, "-abs(%s%s)", in_reg, in_regswizzle);
814 break;
815 default:
816 FIXME("Unhandled modifier %u\n", (instr & WINED3DSP_SRCMOD_MASK));
817 sprintf(out_str, "%s%s", in_reg, in_regswizzle);
821 /** Writes the GLSL variable name that corresponds to the register that the
822 * DX opcode parameter is trying to access */
823 static void shader_glsl_get_register_name(
824 const DWORD param,
825 const DWORD addr_token,
826 char* regstr,
827 BOOL* is_color,
828 SHADER_OPCODE_ARG* arg) {
830 /* oPos, oFog and oPts in D3D */
831 static const char * const hwrastout_reg_names[] = { "gl_Position", "gl_FogFragCoord", "gl_PointSize" };
833 DWORD reg = param & WINED3DSP_REGNUM_MASK;
834 DWORD regtype = shader_get_regtype(param);
835 IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) arg->shader;
836 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
837 WineD3D_GL_Info* gl_info = &deviceImpl->adapter->gl_info;
839 char pshader = shader_is_pshader_version(This->baseShader.hex_version);
840 char tmpStr[150];
842 *is_color = FALSE;
844 switch (regtype) {
845 case WINED3DSPR_TEMP:
846 sprintf(tmpStr, "R%u", reg);
847 break;
848 case WINED3DSPR_INPUT:
849 if (pshader) {
850 /* Pixel shaders >= 3.0 */
851 if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) >= 3) {
852 DWORD in_count = GL_LIMITS(glsl_varyings) / 4;
854 if (param & WINED3DSHADER_ADDRMODE_RELATIVE) {
855 glsl_src_param_t rel_param;
856 shader_glsl_add_src_param(arg, addr_token, 0, WINED3DSP_WRITEMASK_0, &rel_param);
858 /* Removing a + 0 would be an obvious optimization, but macos doesn't see the NOP
859 * operation there
861 if(((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg]) {
862 if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count) {
863 sprintf(tmpStr, "((%s + %u) > %d ? (%s + %u) > %d ? gl_SecondaryColor : gl_Color : IN[%s + %u])",
864 rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg], in_count - 1,
865 rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg], in_count,
866 rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg]);
867 } else {
868 sprintf(tmpStr, "IN[%s + %u]", rel_param.param_str, ((IWineD3DPixelShaderImpl *)This)->input_reg_map[reg]);
870 } else {
871 if (((IWineD3DPixelShaderImpl *)This)->declared_in_count > in_count) {
872 sprintf(tmpStr, "((%s) > %d ? (%s) > %d ? gl_SecondaryColor : gl_Color : IN[%s])",
873 rel_param.param_str, in_count - 1,
874 rel_param.param_str, in_count,
875 rel_param.param_str);
876 } else {
877 sprintf(tmpStr, "IN[%s]", rel_param.param_str);
880 } else {
881 DWORD idx = ((IWineD3DPixelShaderImpl *) This)->input_reg_map[reg];
882 if (idx == in_count) {
883 sprintf(tmpStr, "gl_Color");
884 } else if (idx == in_count + 1) {
885 sprintf(tmpStr, "gl_SecondaryColor");
886 } else {
887 sprintf(tmpStr, "IN[%u]", idx);
890 } else {
891 if (reg==0)
892 strcpy(tmpStr, "gl_Color");
893 else
894 strcpy(tmpStr, "gl_SecondaryColor");
896 } else {
897 if (vshader_input_is_color((IWineD3DVertexShader*) This, reg))
898 *is_color = TRUE;
899 sprintf(tmpStr, "attrib%u", reg);
901 break;
902 case WINED3DSPR_CONST:
904 const char prefix = pshader? 'P':'V';
906 /* Relative addressing */
907 if (param & WINED3DSHADER_ADDRMODE_RELATIVE) {
909 /* Relative addressing on shaders 2.0+ have a relative address token,
910 * prior to that, it was hard-coded as "A0.x" because there's only 1 register */
911 if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) >= 2) {
912 glsl_src_param_t rel_param;
913 shader_glsl_add_src_param(arg, addr_token, 0, WINED3DSP_WRITEMASK_0, &rel_param);
914 if(reg) {
915 sprintf(tmpStr, "%cC[%s + %u]", prefix, rel_param.param_str, reg);
916 } else {
917 sprintf(tmpStr, "%cC[%s]", prefix, rel_param.param_str);
919 } else {
920 if(reg) {
921 sprintf(tmpStr, "%cC[A0.x + %u]", prefix, reg);
922 } else {
923 sprintf(tmpStr, "%cC[A0.x]", prefix);
927 } else {
928 if(shader_constant_is_local(This, reg)) {
929 sprintf(tmpStr, "%cLC%u", prefix, reg);
930 } else {
931 sprintf(tmpStr, "%cC[%u]", prefix, reg);
935 break;
937 case WINED3DSPR_CONSTINT:
938 if (pshader)
939 sprintf(tmpStr, "PI[%u]", reg);
940 else
941 sprintf(tmpStr, "VI[%u]", reg);
942 break;
943 case WINED3DSPR_CONSTBOOL:
944 if (pshader)
945 sprintf(tmpStr, "PB[%u]", reg);
946 else
947 sprintf(tmpStr, "VB[%u]", reg);
948 break;
949 case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
950 if (pshader) {
951 sprintf(tmpStr, "T%u", reg);
952 } else {
953 sprintf(tmpStr, "A%u", reg);
955 break;
956 case WINED3DSPR_LOOP:
957 sprintf(tmpStr, "aL%u", This->baseShader.cur_loop_regno - 1);
958 break;
959 case WINED3DSPR_SAMPLER:
960 if (pshader)
961 sprintf(tmpStr, "Psampler%u", reg);
962 else
963 sprintf(tmpStr, "Vsampler%u", reg);
964 break;
965 case WINED3DSPR_COLOROUT:
966 if (reg >= GL_LIMITS(buffers)) {
967 WARN("Write to render target %u, only %d supported\n", reg, 4);
969 if (GL_SUPPORT(ARB_DRAW_BUFFERS)) {
970 sprintf(tmpStr, "gl_FragData[%u]", reg);
971 } else { /* On older cards with GLSL support like the GeforceFX there's only one buffer. */
972 sprintf(tmpStr, "gl_FragColor");
974 break;
975 case WINED3DSPR_RASTOUT:
976 sprintf(tmpStr, "%s", hwrastout_reg_names[reg]);
977 break;
978 case WINED3DSPR_DEPTHOUT:
979 sprintf(tmpStr, "gl_FragDepth");
980 break;
981 case WINED3DSPR_ATTROUT:
982 if (reg == 0) {
983 sprintf(tmpStr, "gl_FrontColor");
984 } else {
985 sprintf(tmpStr, "gl_FrontSecondaryColor");
987 break;
988 case WINED3DSPR_TEXCRDOUT:
989 /* Vertex shaders >= 3.0: WINED3DSPR_OUTPUT */
990 if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) >= 3)
991 sprintf(tmpStr, "OUT[%u]", reg);
992 else
993 sprintf(tmpStr, "gl_TexCoord[%u]", reg);
994 break;
995 case WINED3DSPR_MISCTYPE:
996 if (reg == 0) {
997 /* vPos */
998 sprintf(tmpStr, "vpos");
999 } else if (reg == 1){
1000 /* Note that gl_FrontFacing is a bool, while vFace is
1001 * a float for which the sign determines front/back
1003 sprintf(tmpStr, "(gl_FrontFacing ? 1.0 : -1.0)");
1004 } else {
1005 FIXME("Unhandled misctype register %d\n", reg);
1006 sprintf(tmpStr, "unrecognized_register");
1008 break;
1009 default:
1010 FIXME("Unhandled register name Type(%d)\n", regtype);
1011 sprintf(tmpStr, "unrecognized_register");
1012 break;
1015 strcat(regstr, tmpStr);
1018 /* Get the GLSL write mask for the destination register */
1019 static DWORD shader_glsl_get_write_mask(const DWORD param, char *write_mask) {
1020 char *ptr = write_mask;
1021 DWORD mask = param & WINED3DSP_WRITEMASK_ALL;
1023 if (shader_is_scalar(param)) {
1024 mask = WINED3DSP_WRITEMASK_0;
1025 } else {
1026 *ptr++ = '.';
1027 if (param & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
1028 if (param & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
1029 if (param & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
1030 if (param & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
1033 *ptr = '\0';
1035 return mask;
1038 static unsigned int shader_glsl_get_write_mask_size(DWORD write_mask) {
1039 unsigned int size = 0;
1041 if (write_mask & WINED3DSP_WRITEMASK_0) ++size;
1042 if (write_mask & WINED3DSP_WRITEMASK_1) ++size;
1043 if (write_mask & WINED3DSP_WRITEMASK_2) ++size;
1044 if (write_mask & WINED3DSP_WRITEMASK_3) ++size;
1046 return size;
1049 static void shader_glsl_get_swizzle(const DWORD param, BOOL fixup, DWORD mask, char *swizzle_str) {
1050 /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
1051 * but addressed as "rgba". To fix this we need to swap the register's x
1052 * and z components. */
1053 DWORD swizzle = (param & WINED3DSP_SWIZZLE_MASK) >> WINED3DSP_SWIZZLE_SHIFT;
1054 const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
1055 char *ptr = swizzle_str;
1057 if (!shader_is_scalar(param)) {
1058 *ptr++ = '.';
1059 /* swizzle bits fields: wwzzyyxx */
1060 if (mask & WINED3DSP_WRITEMASK_0) *ptr++ = swizzle_chars[swizzle & 0x03];
1061 if (mask & WINED3DSP_WRITEMASK_1) *ptr++ = swizzle_chars[(swizzle >> 2) & 0x03];
1062 if (mask & WINED3DSP_WRITEMASK_2) *ptr++ = swizzle_chars[(swizzle >> 4) & 0x03];
1063 if (mask & WINED3DSP_WRITEMASK_3) *ptr++ = swizzle_chars[(swizzle >> 6) & 0x03];
1066 *ptr = '\0';
1069 /* From a given parameter token, generate the corresponding GLSL string.
1070 * Also, return the actual register name and swizzle in case the
1071 * caller needs this information as well. */
1072 static void shader_glsl_add_src_param(SHADER_OPCODE_ARG* arg, const DWORD param,
1073 const DWORD addr_token, DWORD mask, glsl_src_param_t *src_param) {
1074 BOOL is_color = FALSE;
1075 char swizzle_str[6];
1077 src_param->reg_name[0] = '\0';
1078 src_param->param_str[0] = '\0';
1079 swizzle_str[0] = '\0';
1081 shader_glsl_get_register_name(param, addr_token, src_param->reg_name, &is_color, arg);
1083 shader_glsl_get_swizzle(param, is_color, mask, swizzle_str);
1084 shader_glsl_gen_modifier(param, src_param->reg_name, swizzle_str, src_param->param_str);
1087 /* From a given parameter token, generate the corresponding GLSL string.
1088 * Also, return the actual register name and swizzle in case the
1089 * caller needs this information as well. */
1090 static DWORD shader_glsl_add_dst_param(SHADER_OPCODE_ARG* arg, const DWORD param,
1091 const DWORD addr_token, glsl_dst_param_t *dst_param) {
1092 BOOL is_color = FALSE;
1094 dst_param->mask_str[0] = '\0';
1095 dst_param->reg_name[0] = '\0';
1097 shader_glsl_get_register_name(param, addr_token, dst_param->reg_name, &is_color, arg);
1098 return shader_glsl_get_write_mask(param, dst_param->mask_str);
1101 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1102 static DWORD shader_glsl_append_dst_ext(SHADER_BUFFER *buffer, SHADER_OPCODE_ARG *arg, const DWORD param) {
1103 glsl_dst_param_t dst_param;
1104 DWORD mask;
1105 int shift;
1107 mask = shader_glsl_add_dst_param(arg, param, arg->dst_addr, &dst_param);
1109 if(mask) {
1110 shift = (param & WINED3DSP_DSTSHIFT_MASK) >> WINED3DSP_DSTSHIFT_SHIFT;
1111 shader_addline(buffer, "%s%s = %s(", dst_param.reg_name, dst_param.mask_str, shift_glsl_tab[shift]);
1114 return mask;
1117 /* Append the destination part of the instruction to the buffer, return the effective write mask */
1118 static DWORD shader_glsl_append_dst(SHADER_BUFFER *buffer, SHADER_OPCODE_ARG *arg) {
1119 return shader_glsl_append_dst_ext(buffer, arg, arg->dst);
1122 /** Process GLSL instruction modifiers */
1123 void shader_glsl_add_instruction_modifiers(SHADER_OPCODE_ARG* arg) {
1125 DWORD mask = arg->dst & WINED3DSP_DSTMOD_MASK;
1127 if (arg->opcode->dst_token && mask != 0) {
1128 glsl_dst_param_t dst_param;
1130 shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
1132 if (mask & WINED3DSPDM_SATURATE) {
1133 /* _SAT means to clamp the value of the register to between 0 and 1 */
1134 shader_addline(arg->buffer, "%s%s = clamp(%s%s, 0.0, 1.0);\n", dst_param.reg_name,
1135 dst_param.mask_str, dst_param.reg_name, dst_param.mask_str);
1137 if (mask & WINED3DSPDM_MSAMPCENTROID) {
1138 FIXME("_centroid modifier not handled\n");
1140 if (mask & WINED3DSPDM_PARTIALPRECISION) {
1141 /* MSDN says this modifier can be safely ignored, so that's what we'll do. */
1146 static inline const char* shader_get_comp_op(
1147 const DWORD opcode) {
1149 DWORD op = (opcode & INST_CONTROLS_MASK) >> INST_CONTROLS_SHIFT;
1150 switch (op) {
1151 case COMPARISON_GT: return ">";
1152 case COMPARISON_EQ: return "==";
1153 case COMPARISON_GE: return ">=";
1154 case COMPARISON_LT: return "<";
1155 case COMPARISON_NE: return "!=";
1156 case COMPARISON_LE: return "<=";
1157 default:
1158 FIXME("Unrecognized comparison value: %u\n", op);
1159 return "(\?\?)";
1163 static void shader_glsl_get_sample_function(DWORD sampler_type, BOOL projected, BOOL texrect, glsl_sample_function_t *sample_function) {
1164 /* Note that there's no such thing as a projected cube texture. */
1165 switch(sampler_type) {
1166 case WINED3DSTT_1D:
1167 sample_function->name = projected ? "texture1DProj" : "texture1D";
1168 sample_function->coord_mask = WINED3DSP_WRITEMASK_0;
1169 break;
1170 case WINED3DSTT_2D:
1171 if(texrect) {
1172 sample_function->name = projected ? "texture2DRectProj" : "texture2DRect";
1173 } else {
1174 sample_function->name = projected ? "texture2DProj" : "texture2D";
1176 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1;
1177 break;
1178 case WINED3DSTT_CUBE:
1179 sample_function->name = "textureCube";
1180 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1181 break;
1182 case WINED3DSTT_VOLUME:
1183 sample_function->name = projected ? "texture3DProj" : "texture3D";
1184 sample_function->coord_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1185 break;
1186 default:
1187 sample_function->name = "";
1188 FIXME("Unrecognized sampler type: %#x;\n", sampler_type);
1189 break;
1193 static void shader_glsl_color_correction(SHADER_OPCODE_ARG* arg) {
1194 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1195 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) shader->baseShader.device;
1196 WineD3D_GL_Info *gl_info = &deviceImpl->adapter->gl_info;
1197 glsl_dst_param_t dst_param;
1198 glsl_dst_param_t dst_param2;
1199 WINED3DFORMAT fmt;
1200 WINED3DFORMAT conversion_group;
1201 IWineD3DBaseTextureImpl *texture;
1202 DWORD mask, mask_size;
1203 UINT i;
1204 BOOL recorded = FALSE;
1205 DWORD sampler_idx;
1206 DWORD hex_version = shader->baseShader.hex_version;
1208 switch(arg->opcode->opcode) {
1209 case WINED3DSIO_TEX:
1210 if (hex_version < WINED3DPS_VERSION(2,0)) {
1211 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
1212 } else {
1213 sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
1215 break;
1217 case WINED3DSIO_TEXLDL:
1218 FIXME("Add color fixup for vertex texture WINED3DSIO_TEXLDL\n");
1219 return;
1221 case WINED3DSIO_TEXDP3TEX:
1222 case WINED3DSIO_TEXM3x3TEX:
1223 case WINED3DSIO_TEXM3x3SPEC:
1224 case WINED3DSIO_TEXM3x3VSPEC:
1225 case WINED3DSIO_TEXBEM:
1226 case WINED3DSIO_TEXREG2AR:
1227 case WINED3DSIO_TEXREG2GB:
1228 case WINED3DSIO_TEXREG2RGB:
1229 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
1230 break;
1232 default:
1233 /* Not a texture sampling instruction, nothing to do */
1234 return;
1237 texture = (IWineD3DBaseTextureImpl *) deviceImpl->stateBlock->textures[sampler_idx];
1238 if(texture) {
1239 fmt = texture->resource.format;
1240 conversion_group = texture->baseTexture.shader_conversion_group;
1241 } else {
1242 fmt = WINED3DFMT_UNKNOWN;
1243 conversion_group = WINED3DFMT_UNKNOWN;
1246 /* before doing anything, record the sampler with the format in the format conversion list,
1247 * but check if it's not there already
1249 for(i = 0; i < shader->baseShader.num_sampled_samplers; i++) {
1250 if(shader->baseShader.sampled_samplers[i] == sampler_idx) {
1251 recorded = TRUE;
1252 break;
1255 if(!recorded) {
1256 shader->baseShader.sampled_samplers[shader->baseShader.num_sampled_samplers] = sampler_idx;
1257 shader->baseShader.num_sampled_samplers++;
1258 shader->baseShader.sampled_format[sampler_idx] = conversion_group;
1261 switch(fmt) {
1262 case WINED3DFMT_V8U8:
1263 case WINED3DFMT_V16U16:
1264 if(GL_SUPPORT(NV_TEXTURE_SHADER) ||
1265 (GL_SUPPORT(ATI_ENVMAP_BUMPMAP) && fmt == WINED3DFMT_V8U8)) {
1266 /* The 3rd channel returns 1.0 in d3d, but 0.0 in gl. Fix this while we're at it :-) */
1267 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_2, &dst_param);
1268 mask_size = shader_glsl_get_write_mask_size(mask);
1269 if(mask_size >= 3) {
1270 shader_addline(arg->buffer, "%s.%c = 1.0;\n", dst_param.reg_name, dst_param.mask_str[3]);
1272 } else {
1273 /* Correct the sign, but leave the blue as it is - it was loaded correctly already */
1274 mask = shader_glsl_add_dst_param(arg, arg->dst,
1275 WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1,
1276 &dst_param);
1277 mask_size = shader_glsl_get_write_mask_size(mask);
1278 if(mask_size >= 2) {
1279 shader_addline(arg->buffer, "%s.%c%c = %s.%c%c * 2.0 - 1.0;\n",
1280 dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2],
1281 dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2]);
1282 } else if(mask_size == 1) {
1283 shader_addline(arg->buffer, "%s.%c = %s.%c * 2.0 - 1.0;\n", dst_param.reg_name, dst_param.mask_str[1],
1284 dst_param.reg_name, dst_param.mask_str[1]);
1287 break;
1289 case WINED3DFMT_X8L8V8U8:
1290 if(!GL_SUPPORT(NV_TEXTURE_SHADER)) {
1291 /* Red and blue are the signed channels, fix them up; Blue(=L) is correct already,
1292 * and a(X) is always 1.0
1294 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &dst_param);
1295 mask_size = shader_glsl_get_write_mask_size(mask);
1296 if(mask_size >= 2) {
1297 shader_addline(arg->buffer, "%s.%c%c = %s.%c%c * 2.0 - 1.0;\n",
1298 dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2],
1299 dst_param.reg_name, dst_param.mask_str[1], dst_param.mask_str[2]);
1300 } else if(mask_size == 1) {
1301 shader_addline(arg->buffer, "%s.%c = %s.%c * 2.0 - 1.0;\n",
1302 dst_param.reg_name, dst_param.mask_str[1],
1303 dst_param.reg_name, dst_param.mask_str[1]);
1306 break;
1308 case WINED3DFMT_L6V5U5:
1309 if(!GL_SUPPORT(NV_TEXTURE_SHADER)) {
1310 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &dst_param);
1311 mask_size = shader_glsl_get_write_mask_size(mask);
1312 shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_2, &dst_param2);
1313 if(mask_size >= 3) {
1314 /* Swap y and z (U and L), and do a sign conversion on x and the new y(V and U) */
1315 shader_addline(arg->buffer, "tmp0.g = %s.%c;\n",
1316 dst_param.reg_name, dst_param.mask_str[2]);
1317 shader_addline(arg->buffer, "%s.%c%c = %s.%c%c * 2.0 - 1.0;\n",
1318 dst_param.reg_name, dst_param.mask_str[2], dst_param.mask_str[1],
1319 dst_param2.reg_name, dst_param.mask_str[1], dst_param.mask_str[3]);
1320 shader_addline(arg->buffer, "%s.%c = tmp0.g;\n", dst_param.reg_name,
1321 dst_param.mask_str[3]);
1322 } else if(mask_size == 2) {
1323 /* This is bad: We have VL, but we need VU */
1324 FIXME("2 components sampled from a converted L6V5U5 texture\n");
1325 } else {
1326 shader_addline(arg->buffer, "%s.%c = %s.%c * 2.0 - 1.0;\n",
1327 dst_param.reg_name, dst_param.mask_str[1],
1328 dst_param2.reg_name, dst_param.mask_str[1]);
1331 break;
1333 case WINED3DFMT_Q8W8V8U8:
1334 if(!GL_SUPPORT(NV_TEXTURE_SHADER)) {
1335 /* Correct the sign in all channels. The writemask just applies as-is, no
1336 * need for checking the mask size
1338 shader_glsl_add_dst_param(arg, arg->dst,
1339 WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 |
1340 WINED3DSP_WRITEMASK_2 | WINED3DSP_WRITEMASK_3,
1341 &dst_param);
1342 shader_addline(arg->buffer, "%s%s = %s%s * 2.0 - 1.0;\n", dst_param.reg_name, dst_param.mask_str,
1343 dst_param.reg_name, dst_param.mask_str);
1345 break;
1347 case WINED3DFMT_ATI2N:
1348 /* GL_ATI_texture_compression_3dc returns the two channels as luminance-alpha,
1349 * which means the first one is replicated accross .rgb, and the 2nd one is in
1350 * .a. We need the 2nd in .g
1352 * GL_EXT_texture_compression_rgtc returns the values in .rg, however, they
1353 * are swapped compared to d3d. So swap red and green.
1355 mask = shader_glsl_add_dst_param(arg, arg->dst, WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &dst_param);
1356 mask_size = shader_glsl_get_write_mask_size(mask);
1357 if(GL_SUPPORT(EXT_TEXTURE_COMPRESSION_RGTC)) {
1358 if(mask_size >= 2) {
1359 shader_addline(arg->buffer, "%s.%c%c = %s.%c%c;\n",
1360 dst_param.reg_name, dst_param.mask_str[1],
1361 dst_param.mask_str[2],
1362 dst_param.reg_name, dst_param.mask_str[2],
1363 dst_param.mask_str[1]);
1364 } else {
1365 FIXME("%u components sampled from a converted ATI2N texture\n", mask_size);
1367 } else {
1368 if(mask_size == 4) {
1369 /* Swap y and z (U and L), and do a sign conversion on x and the new y(V and U) */
1370 shader_addline(arg->buffer, "%s.%c = %s.%c;\n",
1371 dst_param.reg_name, dst_param.mask_str[2],
1372 dst_param.reg_name, dst_param.mask_str[4]);
1373 } else if(mask_size == 1) {
1374 /* Nothing to do */
1375 } else {
1376 FIXME("%u components sampled from a converted ATI2N texture\n", mask_size);
1377 /* This is bad: We have .r[gb], but we need .ra */
1380 break;
1382 /* stupid compiler */
1383 default:
1384 break;
1388 /*****************************************************************************
1390 * Begin processing individual instruction opcodes
1392 ****************************************************************************/
1394 /* Generate GLSL arithmetic functions (dst = src1 + src2) */
1395 void shader_glsl_arith(SHADER_OPCODE_ARG* arg) {
1396 CONST SHADER_OPCODE* curOpcode = arg->opcode;
1397 SHADER_BUFFER* buffer = arg->buffer;
1398 glsl_src_param_t src0_param;
1399 glsl_src_param_t src1_param;
1400 DWORD write_mask;
1401 char op;
1403 /* Determine the GLSL operator to use based on the opcode */
1404 switch (curOpcode->opcode) {
1405 case WINED3DSIO_MUL: op = '*'; break;
1406 case WINED3DSIO_ADD: op = '+'; break;
1407 case WINED3DSIO_SUB: op = '-'; break;
1408 default:
1409 op = ' ';
1410 FIXME("Opcode %s not yet handled in GLSL\n", curOpcode->name);
1411 break;
1414 write_mask = shader_glsl_append_dst(buffer, arg);
1415 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1416 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1417 shader_addline(buffer, "%s %c %s);\n", src0_param.param_str, op, src1_param.param_str);
1420 /* Process the WINED3DSIO_MOV opcode using GLSL (dst = src) */
1421 void shader_glsl_mov(SHADER_OPCODE_ARG* arg) {
1422 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1423 SHADER_BUFFER* buffer = arg->buffer;
1424 glsl_src_param_t src0_param;
1425 DWORD write_mask;
1427 write_mask = shader_glsl_append_dst(buffer, arg);
1428 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1430 /* In vs_1_1 WINED3DSIO_MOV can write to the address register. In later
1431 * shader versions WINED3DSIO_MOVA is used for this. */
1432 if ((WINED3DSHADER_VERSION_MAJOR(shader->baseShader.hex_version) == 1 &&
1433 !shader_is_pshader_version(shader->baseShader.hex_version) &&
1434 shader_get_regtype(arg->dst) == WINED3DSPR_ADDR)) {
1435 /* This is a simple floor() */
1436 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1437 if (mask_size > 1) {
1438 shader_addline(buffer, "ivec%d(floor(%s)));\n", mask_size, src0_param.param_str);
1439 } else {
1440 shader_addline(buffer, "int(floor(%s)));\n", src0_param.param_str);
1442 } else if(arg->opcode->opcode == WINED3DSIO_MOVA) {
1443 /* We need to *round* to the nearest int here. */
1444 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
1445 if (mask_size > 1) {
1446 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);
1447 } else {
1448 shader_addline(buffer, "int(floor(abs(%s) + 0.5) * sign(%s)));\n", src0_param.param_str, src0_param.param_str);
1450 } else {
1451 shader_addline(buffer, "%s);\n", src0_param.param_str);
1455 /* Process the dot product operators DP3 and DP4 in GLSL (dst = dot(src0, src1)) */
1456 void shader_glsl_dot(SHADER_OPCODE_ARG* arg) {
1457 CONST SHADER_OPCODE* curOpcode = arg->opcode;
1458 SHADER_BUFFER* buffer = arg->buffer;
1459 glsl_src_param_t src0_param;
1460 glsl_src_param_t src1_param;
1461 DWORD dst_write_mask, src_write_mask;
1462 unsigned int dst_size = 0;
1464 dst_write_mask = shader_glsl_append_dst(buffer, arg);
1465 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1467 /* dp3 works on vec3, dp4 on vec4 */
1468 if (curOpcode->opcode == WINED3DSIO_DP4) {
1469 src_write_mask = WINED3DSP_WRITEMASK_ALL;
1470 } else {
1471 src_write_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1474 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_write_mask, &src0_param);
1475 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_write_mask, &src1_param);
1477 if (dst_size > 1) {
1478 shader_addline(buffer, "vec%d(dot(%s, %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1479 } else {
1480 shader_addline(buffer, "dot(%s, %s));\n", src0_param.param_str, src1_param.param_str);
1484 /* Note that this instruction has some restrictions. The destination write mask
1485 * can't contain the w component, and the source swizzles have to be .xyzw */
1486 void shader_glsl_cross(SHADER_OPCODE_ARG *arg) {
1487 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
1488 glsl_src_param_t src0_param;
1489 glsl_src_param_t src1_param;
1490 char dst_mask[6];
1492 shader_glsl_get_write_mask(arg->dst, dst_mask);
1493 shader_glsl_append_dst(arg->buffer, arg);
1494 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
1495 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_mask, &src1_param);
1496 shader_addline(arg->buffer, "cross(%s, %s)%s);\n", src0_param.param_str, src1_param.param_str, dst_mask);
1499 /* Process the WINED3DSIO_POW instruction in GLSL (dst = |src0|^src1)
1500 * Src0 and src1 are scalars. Note that D3D uses the absolute of src0, while
1501 * GLSL uses the value as-is. */
1502 void shader_glsl_pow(SHADER_OPCODE_ARG *arg) {
1503 SHADER_BUFFER *buffer = arg->buffer;
1504 glsl_src_param_t src0_param;
1505 glsl_src_param_t src1_param;
1506 DWORD dst_write_mask;
1507 unsigned int dst_size;
1509 dst_write_mask = shader_glsl_append_dst(buffer, arg);
1510 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1512 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1513 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
1515 if (dst_size > 1) {
1516 shader_addline(buffer, "vec%d(pow(abs(%s), %s)));\n", dst_size, src0_param.param_str, src1_param.param_str);
1517 } else {
1518 shader_addline(buffer, "pow(abs(%s), %s));\n", src0_param.param_str, src1_param.param_str);
1522 /* Process the WINED3DSIO_LOG instruction in GLSL (dst = log2(|src0|))
1523 * Src0 is a scalar. Note that D3D uses the absolute of src0, while
1524 * GLSL uses the value as-is. */
1525 void shader_glsl_log(SHADER_OPCODE_ARG *arg) {
1526 SHADER_BUFFER *buffer = arg->buffer;
1527 glsl_src_param_t src0_param;
1528 DWORD dst_write_mask;
1529 unsigned int dst_size;
1531 dst_write_mask = shader_glsl_append_dst(buffer, arg);
1532 dst_size = shader_glsl_get_write_mask_size(dst_write_mask);
1534 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1536 if (dst_size > 1) {
1537 shader_addline(buffer, "vec%d(log2(abs(%s))));\n", dst_size, src0_param.param_str);
1538 } else {
1539 shader_addline(buffer, "log2(abs(%s)));\n", src0_param.param_str);
1543 /* Map the opcode 1-to-1 to the GL code (arg->dst = instruction(src0, src1, ...) */
1544 void shader_glsl_map2gl(SHADER_OPCODE_ARG* arg) {
1545 CONST SHADER_OPCODE* curOpcode = arg->opcode;
1546 SHADER_BUFFER* buffer = arg->buffer;
1547 glsl_src_param_t src_param;
1548 const char *instruction;
1549 char arguments[256];
1550 DWORD write_mask;
1551 unsigned i;
1553 /* Determine the GLSL function to use based on the opcode */
1554 /* TODO: Possibly make this a table for faster lookups */
1555 switch (curOpcode->opcode) {
1556 case WINED3DSIO_MIN: instruction = "min"; break;
1557 case WINED3DSIO_MAX: instruction = "max"; break;
1558 case WINED3DSIO_ABS: instruction = "abs"; break;
1559 case WINED3DSIO_FRC: instruction = "fract"; break;
1560 case WINED3DSIO_NRM: instruction = "normalize"; break;
1561 case WINED3DSIO_LOGP:
1562 case WINED3DSIO_LOG: instruction = "log2"; break;
1563 case WINED3DSIO_EXP: instruction = "exp2"; break;
1564 case WINED3DSIO_SGN: instruction = "sign"; break;
1565 case WINED3DSIO_DSX: instruction = "dFdx"; break;
1566 case WINED3DSIO_DSY: instruction = "ycorrection.y * dFdy"; break;
1567 default: instruction = "";
1568 FIXME("Opcode %s not yet handled in GLSL\n", curOpcode->name);
1569 break;
1572 write_mask = shader_glsl_append_dst(buffer, arg);
1574 arguments[0] = '\0';
1575 if (curOpcode->num_params > 0) {
1576 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src_param);
1577 strcat(arguments, src_param.param_str);
1578 for (i = 2; i < curOpcode->num_params; ++i) {
1579 strcat(arguments, ", ");
1580 shader_glsl_add_src_param(arg, arg->src[i-1], arg->src_addr[i-1], write_mask, &src_param);
1581 strcat(arguments, src_param.param_str);
1585 shader_addline(buffer, "%s(%s));\n", instruction, arguments);
1588 /** Process the WINED3DSIO_EXPP instruction in GLSL:
1589 * For shader model 1.x, do the following (and honor the writemask, so use a temporary variable):
1590 * dst.x = 2^(floor(src))
1591 * dst.y = src - floor(src)
1592 * dst.z = 2^src (partial precision is allowed, but optional)
1593 * dst.w = 1.0;
1594 * For 2.0 shaders, just do this (honoring writemask and swizzle):
1595 * dst = 2^src; (partial precision is allowed, but optional)
1597 void shader_glsl_expp(SHADER_OPCODE_ARG* arg) {
1598 IWineD3DBaseShaderImpl *shader = (IWineD3DBaseShaderImpl *)arg->shader;
1599 glsl_src_param_t src_param;
1601 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src_param);
1603 if (shader->baseShader.hex_version < WINED3DPS_VERSION(2,0)) {
1604 char dst_mask[6];
1606 shader_addline(arg->buffer, "tmp0.x = exp2(floor(%s));\n", src_param.param_str);
1607 shader_addline(arg->buffer, "tmp0.y = %s - floor(%s);\n", src_param.param_str, src_param.param_str);
1608 shader_addline(arg->buffer, "tmp0.z = exp2(%s);\n", src_param.param_str);
1609 shader_addline(arg->buffer, "tmp0.w = 1.0;\n");
1611 shader_glsl_append_dst(arg->buffer, arg);
1612 shader_glsl_get_write_mask(arg->dst, dst_mask);
1613 shader_addline(arg->buffer, "tmp0%s);\n", dst_mask);
1614 } else {
1615 DWORD write_mask;
1616 unsigned int mask_size;
1618 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1619 mask_size = shader_glsl_get_write_mask_size(write_mask);
1621 if (mask_size > 1) {
1622 shader_addline(arg->buffer, "vec%d(exp2(%s)));\n", mask_size, src_param.param_str);
1623 } else {
1624 shader_addline(arg->buffer, "exp2(%s));\n", src_param.param_str);
1629 /** Process the RCP (reciprocal or inverse) opcode in GLSL (dst = 1 / src) */
1630 void shader_glsl_rcp(SHADER_OPCODE_ARG* arg) {
1631 glsl_src_param_t src_param;
1632 DWORD write_mask;
1633 unsigned int mask_size;
1635 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1636 mask_size = shader_glsl_get_write_mask_size(write_mask);
1637 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src_param);
1639 if (mask_size > 1) {
1640 shader_addline(arg->buffer, "vec%d(1.0 / %s));\n", mask_size, src_param.param_str);
1641 } else {
1642 shader_addline(arg->buffer, "1.0 / %s);\n", src_param.param_str);
1646 void shader_glsl_rsq(SHADER_OPCODE_ARG* arg) {
1647 SHADER_BUFFER* buffer = arg->buffer;
1648 glsl_src_param_t src_param;
1649 DWORD write_mask;
1650 unsigned int mask_size;
1652 write_mask = shader_glsl_append_dst(buffer, arg);
1653 mask_size = shader_glsl_get_write_mask_size(write_mask);
1655 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src_param);
1657 if (mask_size > 1) {
1658 shader_addline(buffer, "vec%d(inversesqrt(%s)));\n", mask_size, src_param.param_str);
1659 } else {
1660 shader_addline(buffer, "inversesqrt(%s));\n", src_param.param_str);
1664 /** Process signed comparison opcodes in GLSL. */
1665 void shader_glsl_compare(SHADER_OPCODE_ARG* arg) {
1666 glsl_src_param_t src0_param;
1667 glsl_src_param_t src1_param;
1668 DWORD write_mask;
1669 unsigned int mask_size;
1671 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1672 mask_size = shader_glsl_get_write_mask_size(write_mask);
1673 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1674 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1676 if (mask_size > 1) {
1677 const char *compare;
1679 switch(arg->opcode->opcode) {
1680 case WINED3DSIO_SLT: compare = "lessThan"; break;
1681 case WINED3DSIO_SGE: compare = "greaterThanEqual"; break;
1682 default: compare = "";
1683 FIXME("Can't handle opcode %s\n", arg->opcode->name);
1686 shader_addline(arg->buffer, "vec%d(%s(%s, %s)));\n", mask_size, compare,
1687 src0_param.param_str, src1_param.param_str);
1688 } else {
1689 switch(arg->opcode->opcode) {
1690 case WINED3DSIO_SLT:
1691 /* Step(src0, src1) is not suitable here because if src0 == src1 SLT is supposed,
1692 * to return 0.0 but step returns 1.0 because step is not < x
1693 * An alternative is a bvec compare padded with an unused second component.
1694 * step(src1 * -1.0, src0 * -1.0) is not an option because it suffers from the same
1695 * issue. Playing with not() is not possible either because not() does not accept
1696 * a scalar.
1698 shader_addline(arg->buffer, "(%s < %s) ? 1.0 : 0.0);\n", src0_param.param_str, src1_param.param_str);
1699 break;
1700 case WINED3DSIO_SGE:
1701 /* Here we can use the step() function and safe a conditional */
1702 shader_addline(arg->buffer, "step(%s, %s));\n", src1_param.param_str, src0_param.param_str);
1703 break;
1704 default:
1705 FIXME("Can't handle opcode %s\n", arg->opcode->name);
1711 /** Process CMP instruction in GLSL (dst = src0 >= 0.0 ? src1 : src2), per channel */
1712 void shader_glsl_cmp(SHADER_OPCODE_ARG* arg) {
1713 glsl_src_param_t src0_param;
1714 glsl_src_param_t src1_param;
1715 glsl_src_param_t src2_param;
1716 DWORD write_mask, cmp_channel = 0;
1717 unsigned int i, j;
1718 char mask_char[6];
1719 BOOL temp_destination = FALSE;
1721 if(shader_is_scalar(arg->src[0])) {
1722 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1724 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
1725 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1726 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1728 shader_addline(arg->buffer, "%s >= 0.0 ? %s : %s);\n",
1729 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1730 } else {
1731 DWORD src0reg = arg->src[0] & WINED3DSP_REGNUM_MASK;
1732 DWORD src1reg = arg->src[1] & WINED3DSP_REGNUM_MASK;
1733 DWORD src2reg = arg->src[2] & WINED3DSP_REGNUM_MASK;
1734 DWORD src0regtype = shader_get_regtype(arg->src[0]);
1735 DWORD src1regtype = shader_get_regtype(arg->src[1]);
1736 DWORD src2regtype = shader_get_regtype(arg->src[2]);
1737 DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
1738 DWORD dstregtype = shader_get_regtype(arg->dst);
1740 /* Cycle through all source0 channels */
1741 for (i=0; i<4; i++) {
1742 write_mask = 0;
1743 /* Find the destination channels which use the current source0 channel */
1744 for (j=0; j<4; j++) {
1745 if ( ((arg->src[0] >> (WINED3DSP_SWIZZLE_SHIFT + 2*j)) & 0x3) == i ) {
1746 write_mask |= WINED3DSP_WRITEMASK_0 << j;
1747 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
1751 /* Splitting the cmp instruction up in multiple lines imposes a problem:
1752 * The first lines may overwrite source parameters of the following lines.
1753 * Deal with that by using a temporary destination register if needed
1755 if((src0reg == dstreg && src0regtype == dstregtype) ||
1756 (src1reg == dstreg && src1regtype == dstregtype) ||
1757 (src2reg == dstreg && src2regtype == dstregtype)) {
1759 write_mask = shader_glsl_get_write_mask(arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask), mask_char);
1760 if (!write_mask) continue;
1761 shader_addline(arg->buffer, "tmp0%s = (", mask_char);
1762 temp_destination = TRUE;
1763 } else {
1764 write_mask = shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask));
1765 if (!write_mask) continue;
1768 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], cmp_channel, &src0_param);
1769 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1770 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1772 shader_addline(arg->buffer, "%s >= 0.0 ? %s : %s);\n",
1773 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1776 if(temp_destination) {
1777 shader_glsl_get_write_mask(arg->dst, mask_char);
1778 shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst);
1779 shader_addline(arg->buffer, "tmp0%s);\n", mask_char);
1785 /** Process the CND opcode in GLSL (dst = (src0 > 0.5) ? src1 : src2) */
1786 /* For ps 1.1-1.3, only a single component of src0 is used. For ps 1.4
1787 * the compare is done per component of src0. */
1788 void shader_glsl_cnd(SHADER_OPCODE_ARG* arg) {
1789 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
1790 glsl_src_param_t src0_param;
1791 glsl_src_param_t src1_param;
1792 glsl_src_param_t src2_param;
1793 DWORD write_mask, cmp_channel = 0;
1794 unsigned int i, j;
1796 if (shader->baseShader.hex_version < WINED3DPS_VERSION(1, 4)) {
1797 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1798 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1799 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1800 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1802 /* Fun: The D3DSI_COISSUE flag changes the semantic of the cnd instruction for < 1.4 shaders */
1803 if(arg->opcode_token & WINED3DSI_COISSUE) {
1804 shader_addline(arg->buffer, "%s /* COISSUE! */);\n", src1_param.param_str);
1805 } else {
1806 shader_addline(arg->buffer, "%s > 0.5 ? %s : %s);\n",
1807 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1809 return;
1811 /* Cycle through all source0 channels */
1812 for (i=0; i<4; i++) {
1813 write_mask = 0;
1814 /* Find the destination channels which use the current source0 channel */
1815 for (j=0; j<4; j++) {
1816 if ( ((arg->src[0] >> (WINED3DSP_SWIZZLE_SHIFT + 2*j)) & 0x3) == i ) {
1817 write_mask |= WINED3DSP_WRITEMASK_0 << j;
1818 cmp_channel = WINED3DSP_WRITEMASK_0 << j;
1821 write_mask = shader_glsl_append_dst_ext(arg->buffer, arg, arg->dst & (~WINED3DSP_SWIZZLE_MASK | write_mask));
1822 if (!write_mask) continue;
1824 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], cmp_channel, &src0_param);
1825 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1826 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1828 shader_addline(arg->buffer, "%s > 0.5 ? %s : %s);\n",
1829 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1833 /** GLSL code generation for WINED3DSIO_MAD: Multiply the first 2 opcodes, then add the last */
1834 void shader_glsl_mad(SHADER_OPCODE_ARG* arg) {
1835 glsl_src_param_t src0_param;
1836 glsl_src_param_t src1_param;
1837 glsl_src_param_t src2_param;
1838 DWORD write_mask;
1840 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1841 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1842 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1843 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1844 shader_addline(arg->buffer, "(%s * %s) + %s);\n",
1845 src0_param.param_str, src1_param.param_str, src2_param.param_str);
1848 /** Handles transforming all WINED3DSIO_M?x? opcodes for
1849 Vertex shaders to GLSL codes */
1850 void shader_glsl_mnxn(SHADER_OPCODE_ARG* arg) {
1851 int i;
1852 int nComponents = 0;
1853 SHADER_OPCODE_ARG tmpArg;
1855 memset(&tmpArg, 0, sizeof(SHADER_OPCODE_ARG));
1857 /* Set constants for the temporary argument */
1858 tmpArg.shader = arg->shader;
1859 tmpArg.buffer = arg->buffer;
1860 tmpArg.src[0] = arg->src[0];
1861 tmpArg.src_addr[0] = arg->src_addr[0];
1862 tmpArg.src_addr[1] = arg->src_addr[1];
1863 tmpArg.reg_maps = arg->reg_maps;
1865 switch(arg->opcode->opcode) {
1866 case WINED3DSIO_M4x4:
1867 nComponents = 4;
1868 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP4);
1869 break;
1870 case WINED3DSIO_M4x3:
1871 nComponents = 3;
1872 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP4);
1873 break;
1874 case WINED3DSIO_M3x4:
1875 nComponents = 4;
1876 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
1877 break;
1878 case WINED3DSIO_M3x3:
1879 nComponents = 3;
1880 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
1881 break;
1882 case WINED3DSIO_M3x2:
1883 nComponents = 2;
1884 tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
1885 break;
1886 default:
1887 break;
1890 for (i = 0; i < nComponents; i++) {
1891 tmpArg.dst = ((arg->dst) & ~WINED3DSP_WRITEMASK_ALL)|(WINED3DSP_WRITEMASK_0<<i);
1892 tmpArg.src[1] = arg->src[1]+i;
1893 shader_glsl_dot(&tmpArg);
1898 The LRP instruction performs a component-wise linear interpolation
1899 between the second and third operands using the first operand as the
1900 blend factor. Equation: (dst = src2 + src0 * (src1 - src2))
1901 This is equivalent to mix(src2, src1, src0);
1903 void shader_glsl_lrp(SHADER_OPCODE_ARG* arg) {
1904 glsl_src_param_t src0_param;
1905 glsl_src_param_t src1_param;
1906 glsl_src_param_t src2_param;
1907 DWORD write_mask;
1909 write_mask = shader_glsl_append_dst(arg->buffer, arg);
1911 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
1912 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
1913 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);
1915 shader_addline(arg->buffer, "mix(%s, %s, %s));\n",
1916 src2_param.param_str, src1_param.param_str, src0_param.param_str);
1919 /** Process the WINED3DSIO_LIT instruction in GLSL:
1920 * dst.x = dst.w = 1.0
1921 * dst.y = (src0.x > 0) ? src0.x
1922 * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
1923 * where src.w is clamped at +- 128
1925 void shader_glsl_lit(SHADER_OPCODE_ARG* arg) {
1926 glsl_src_param_t src0_param;
1927 glsl_src_param_t src1_param;
1928 glsl_src_param_t src3_param;
1929 char dst_mask[6];
1931 shader_glsl_append_dst(arg->buffer, arg);
1932 shader_glsl_get_write_mask(arg->dst, dst_mask);
1934 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
1935 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_1, &src1_param);
1936 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src3_param);
1938 /* The sdk specifies the instruction like this
1939 * dst.x = 1.0;
1940 * if(src.x > 0.0) dst.y = src.x
1941 * else dst.y = 0.0.
1942 * if(src.x > 0.0 && src.y > 0.0) dst.z = pow(src.y, power);
1943 * else dst.z = 0.0;
1944 * dst.w = 1.0;
1946 * Obviously that has quite a few conditionals in it which we don't like. So the first step is this:
1947 * dst.x = 1.0 ... No further explanation needed
1948 * dst.y = max(src.y, 0.0); ... If x < 0.0, use 0.0, otherwise x. Same as the conditional
1949 * dst.z = x > 0.0 ? pow(max(y, 0.0), p) : 0; ... 0 ^ power is 0, and otherwise we use y anyway
1950 * dst.w = 1.0. ... Nothing fancy.
1952 * So we still have one conditional in there. So do this:
1953 * dst.z = pow(max(0.0, src.y) * step(0.0, src.x), power);
1955 * 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),
1956 * 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.
1957 * if both x and y are > 0, we get pow(y * 1.0, power), as it is supposed to
1959 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",
1960 src0_param.param_str, src1_param.param_str, src0_param.param_str, src3_param.param_str, dst_mask);
1963 /** Process the WINED3DSIO_DST instruction in GLSL:
1964 * dst.x = 1.0
1965 * dst.y = src0.x * src0.y
1966 * dst.z = src0.z
1967 * dst.w = src1.w
1969 void shader_glsl_dst(SHADER_OPCODE_ARG* arg) {
1970 glsl_src_param_t src0y_param;
1971 glsl_src_param_t src0z_param;
1972 glsl_src_param_t src1y_param;
1973 glsl_src_param_t src1w_param;
1974 char dst_mask[6];
1976 shader_glsl_append_dst(arg->buffer, arg);
1977 shader_glsl_get_write_mask(arg->dst, dst_mask);
1979 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_1, &src0y_param);
1980 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &src0z_param);
1981 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_1, &src1y_param);
1982 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_3, &src1w_param);
1984 shader_addline(arg->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
1985 src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
1988 /** Process the WINED3DSIO_SINCOS instruction in GLSL:
1989 * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
1990 * can handle it. But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
1992 * dst.x = cos(src0.?)
1993 * dst.y = sin(src0.?)
1994 * dst.z = dst.z
1995 * dst.w = dst.w
1997 void shader_glsl_sincos(SHADER_OPCODE_ARG* arg) {
1998 glsl_src_param_t src0_param;
1999 DWORD write_mask;
2001 write_mask = shader_glsl_append_dst(arg->buffer, arg);
2002 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2004 switch (write_mask) {
2005 case WINED3DSP_WRITEMASK_0:
2006 shader_addline(arg->buffer, "cos(%s));\n", src0_param.param_str);
2007 break;
2009 case WINED3DSP_WRITEMASK_1:
2010 shader_addline(arg->buffer, "sin(%s));\n", src0_param.param_str);
2011 break;
2013 case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
2014 shader_addline(arg->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
2015 break;
2017 default:
2018 ERR("Write mask should be .x, .y or .xy\n");
2019 break;
2023 /** Process the WINED3DSIO_LOOP instruction in GLSL:
2024 * Start a for() loop where src1.y is the initial value of aL,
2025 * increment aL by src1.z for a total of src1.x iterations.
2026 * Need to use a temporary variable for this operation.
2028 /* FIXME: I don't think nested loops will work correctly this way. */
2029 void shader_glsl_loop(SHADER_OPCODE_ARG* arg) {
2030 glsl_src_param_t src1_param;
2031 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
2032 DWORD regtype = shader_get_regtype(arg->src[1]);
2033 DWORD reg = arg->src[1] & WINED3DSP_REGNUM_MASK;
2034 const DWORD *control_values = NULL;
2035 local_constant *constant;
2037 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
2039 /* Try to hardcode the loop control parameters if possible. Direct3D 9 class hardware doesn't support real
2040 * varying indexing, but Microsoft designed this feature for Shader model 2.x+. If the loop control is
2041 * known at compile time, the GLSL compiler can unroll the loop, and replace indirect addressing with direct
2042 * addressing.
2044 if(regtype == WINED3DSPR_CONSTINT) {
2045 LIST_FOR_EACH_ENTRY(constant, &shader->baseShader.constantsI, local_constant, entry) {
2046 if(constant->idx == reg) {
2047 control_values = constant->value;
2048 break;
2053 if(control_values) {
2054 if(control_values[2] > 0) {
2055 shader_addline(arg->buffer, "for (aL%u = %d; aL%u < (%d * %d + %d); aL%u += %d) {\n",
2056 shader->baseShader.cur_loop_depth, control_values[1],
2057 shader->baseShader.cur_loop_depth, control_values[0], control_values[2], control_values[1],
2058 shader->baseShader.cur_loop_depth, control_values[2]);
2059 } else if(control_values[2] == 0) {
2060 shader_addline(arg->buffer, "for (aL%u = %d, tmpInt%u = 0; tmpInt%u < %d; tmpInt%u++) {\n",
2061 shader->baseShader.cur_loop_depth, control_values[1], shader->baseShader.cur_loop_depth,
2062 shader->baseShader.cur_loop_depth, control_values[0],
2063 shader->baseShader.cur_loop_depth);
2064 } else {
2065 shader_addline(arg->buffer, "for (aL%u = %d; aL%u > (%d * %d + %d); aL%u += %d) {\n",
2066 shader->baseShader.cur_loop_depth, control_values[1],
2067 shader->baseShader.cur_loop_depth, control_values[0], control_values[2], control_values[1],
2068 shader->baseShader.cur_loop_depth, control_values[2]);
2070 } else {
2071 shader_addline(arg->buffer, "for (tmpInt%u = 0, aL%u = %s.y; tmpInt%u < %s.x; tmpInt%u++, aL%u += %s.z) {\n",
2072 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno,
2073 src1_param.reg_name, shader->baseShader.cur_loop_depth, src1_param.reg_name,
2074 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_regno, src1_param.reg_name);
2077 shader->baseShader.cur_loop_depth++;
2078 shader->baseShader.cur_loop_regno++;
2081 void shader_glsl_end(SHADER_OPCODE_ARG* arg) {
2082 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
2084 shader_addline(arg->buffer, "}\n");
2086 if(arg->opcode->opcode == WINED3DSIO_ENDLOOP) {
2087 shader->baseShader.cur_loop_depth--;
2088 shader->baseShader.cur_loop_regno--;
2090 if(arg->opcode->opcode == WINED3DSIO_ENDREP) {
2091 shader->baseShader.cur_loop_depth--;
2095 void shader_glsl_rep(SHADER_OPCODE_ARG* arg) {
2096 IWineD3DBaseShaderImpl* shader = (IWineD3DBaseShaderImpl*) arg->shader;
2097 glsl_src_param_t src0_param;
2099 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2100 shader_addline(arg->buffer, "for (tmpInt%d = 0; tmpInt%d < %s; tmpInt%d++) {\n",
2101 shader->baseShader.cur_loop_depth, shader->baseShader.cur_loop_depth,
2102 src0_param.param_str, shader->baseShader.cur_loop_depth);
2103 shader->baseShader.cur_loop_depth++;
2106 void shader_glsl_if(SHADER_OPCODE_ARG* arg) {
2107 glsl_src_param_t src0_param;
2109 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2110 shader_addline(arg->buffer, "if (%s) {\n", src0_param.param_str);
2113 void shader_glsl_ifc(SHADER_OPCODE_ARG* arg) {
2114 glsl_src_param_t src0_param;
2115 glsl_src_param_t src1_param;
2117 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2118 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2120 shader_addline(arg->buffer, "if (%s %s %s) {\n",
2121 src0_param.param_str, shader_get_comp_op(arg->opcode_token), src1_param.param_str);
2124 void shader_glsl_else(SHADER_OPCODE_ARG* arg) {
2125 shader_addline(arg->buffer, "} else {\n");
2128 void shader_glsl_break(SHADER_OPCODE_ARG* arg) {
2129 shader_addline(arg->buffer, "break;\n");
2132 /* FIXME: According to MSDN the compare is done per component. */
2133 void shader_glsl_breakc(SHADER_OPCODE_ARG* arg) {
2134 glsl_src_param_t src0_param;
2135 glsl_src_param_t src1_param;
2137 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
2138 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2140 shader_addline(arg->buffer, "if (%s %s %s) break;\n",
2141 src0_param.param_str, shader_get_comp_op(arg->opcode_token), src1_param.param_str);
2144 void shader_glsl_label(SHADER_OPCODE_ARG* arg) {
2146 DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2147 shader_addline(arg->buffer, "}\n");
2148 shader_addline(arg->buffer, "void subroutine%u () {\n", snum);
2151 void shader_glsl_call(SHADER_OPCODE_ARG* arg) {
2152 DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2153 shader_addline(arg->buffer, "subroutine%u();\n", snum);
2156 void shader_glsl_callnz(SHADER_OPCODE_ARG* arg) {
2157 glsl_src_param_t src1_param;
2159 DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
2160 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
2161 shader_addline(arg->buffer, "if (%s) subroutine%u();\n", src1_param.param_str, snum);
2164 /*********************************************
2165 * Pixel Shader Specific Code begins here
2166 ********************************************/
2167 void pshader_glsl_tex(SHADER_OPCODE_ARG* arg) {
2168 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2169 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2170 DWORD hex_version = This->baseShader.hex_version;
2171 char dst_swizzle[6];
2172 glsl_sample_function_t sample_function;
2173 DWORD sampler_type;
2174 DWORD sampler_idx;
2175 BOOL projected, texrect = FALSE;
2176 DWORD mask = 0;
2178 /* All versions have a destination register */
2179 shader_glsl_append_dst(arg->buffer, arg);
2181 /* 1.0-1.4: Use destination register as sampler source.
2182 * 2.0+: Use provided sampler source. */
2183 if (hex_version < WINED3DPS_VERSION(1,4)) {
2184 DWORD flags;
2186 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2187 flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2189 if (flags & WINED3DTTFF_PROJECTED) {
2190 projected = TRUE;
2191 switch (flags & ~WINED3DTTFF_PROJECTED) {
2192 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2193 case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
2194 case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
2195 case WINED3DTTFF_COUNT4:
2196 case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
2198 } else {
2199 projected = FALSE;
2201 } else if (hex_version < WINED3DPS_VERSION(2,0)) {
2202 DWORD src_mod = arg->src[0] & WINED3DSP_SRCMOD_MASK;
2203 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2205 if (src_mod == WINED3DSPSM_DZ) {
2206 projected = TRUE;
2207 mask = WINED3DSP_WRITEMASK_2;
2208 } else if (src_mod == WINED3DSPSM_DW) {
2209 projected = TRUE;
2210 mask = WINED3DSP_WRITEMASK_3;
2211 } else {
2212 projected = FALSE;
2214 } else {
2215 sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
2216 if(arg->opcode_token & WINED3DSI_TEXLD_PROJECT) {
2217 /* ps 2.0 texldp instruction always divides by the fourth component. */
2218 projected = TRUE;
2219 mask = WINED3DSP_WRITEMASK_3;
2220 } else {
2221 projected = FALSE;
2225 if(deviceImpl->stateBlock->textures[sampler_idx] &&
2226 IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2227 texrect = TRUE;
2230 sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2231 shader_glsl_get_sample_function(sampler_type, projected, texrect, &sample_function);
2232 mask |= sample_function.coord_mask;
2234 if (hex_version < WINED3DPS_VERSION(2,0)) {
2235 shader_glsl_get_write_mask(arg->dst, dst_swizzle);
2236 } else {
2237 shader_glsl_get_swizzle(arg->src[1], FALSE, arg->dst, dst_swizzle);
2240 /* 1.0-1.3: Use destination register as coordinate source.
2241 1.4+: Use provided coordinate source register. */
2242 if (hex_version < WINED3DPS_VERSION(1,4)) {
2243 char coord_mask[6];
2244 shader_glsl_get_write_mask(mask, coord_mask);
2245 shader_addline(arg->buffer, "%s(Psampler%u, T%u%s)%s);\n",
2246 sample_function.name, sampler_idx, sampler_idx, coord_mask, dst_swizzle);
2247 } else {
2248 glsl_src_param_t coord_param;
2249 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], mask, &coord_param);
2250 if(arg->opcode_token & WINED3DSI_TEXLD_BIAS) {
2251 glsl_src_param_t bias;
2252 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &bias);
2254 shader_addline(arg->buffer, "%s(Psampler%u, %s, %s)%s);\n",
2255 sample_function.name, sampler_idx, coord_param.param_str,
2256 bias.param_str, dst_swizzle);
2257 } else {
2258 shader_addline(arg->buffer, "%s(Psampler%u, %s)%s);\n",
2259 sample_function.name, sampler_idx, coord_param.param_str, dst_swizzle);
2264 void shader_glsl_texldl(SHADER_OPCODE_ARG* arg) {
2265 IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*)arg->shader;
2266 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2267 glsl_sample_function_t sample_function;
2268 glsl_src_param_t coord_param, lod_param;
2269 char dst_swizzle[6];
2270 DWORD sampler_type;
2271 DWORD sampler_idx;
2272 BOOL texrect = FALSE;
2274 shader_glsl_append_dst(arg->buffer, arg);
2275 shader_glsl_get_swizzle(arg->src[1], FALSE, arg->dst, dst_swizzle);
2277 sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
2278 sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2279 if(deviceImpl->stateBlock->textures[sampler_idx] &&
2280 IWineD3DBaseTexture_GetTextureDimensions(deviceImpl->stateBlock->textures[sampler_idx]) == GL_TEXTURE_RECTANGLE_ARB) {
2281 texrect = TRUE;
2283 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);
2285 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &lod_param);
2287 if (shader_is_pshader_version(This->baseShader.hex_version)) {
2288 /* The GLSL spec claims the Lod sampling functions are only supported in vertex shaders.
2289 * However, they seem to work just fine in fragment shaders as well. */
2290 WARN("Using %sLod in fragment shader.\n", sample_function.name);
2291 shader_addline(arg->buffer, "%sLod(Psampler%u, %s, %s)%s);\n",
2292 sample_function.name, sampler_idx, coord_param.param_str, lod_param.param_str, dst_swizzle);
2293 } else {
2294 shader_addline(arg->buffer, "%sLod(Vsampler%u, %s, %s)%s);\n",
2295 sample_function.name, sampler_idx, coord_param.param_str, lod_param.param_str, dst_swizzle);
2299 void pshader_glsl_texcoord(SHADER_OPCODE_ARG* arg) {
2301 /* FIXME: Make this work for more than just 2D textures */
2303 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2304 SHADER_BUFFER* buffer = arg->buffer;
2305 DWORD hex_version = This->baseShader.hex_version;
2306 DWORD write_mask;
2307 char dst_mask[6];
2309 write_mask = shader_glsl_append_dst(arg->buffer, arg);
2310 shader_glsl_get_write_mask(write_mask, dst_mask);
2312 if (hex_version != WINED3DPS_VERSION(1,4)) {
2313 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2314 shader_addline(buffer, "clamp(gl_TexCoord[%u], 0.0, 1.0)%s);\n", reg, dst_mask);
2315 } else {
2316 DWORD reg = arg->src[0] & WINED3DSP_REGNUM_MASK;
2317 DWORD src_mod = arg->src[0] & WINED3DSP_SRCMOD_MASK;
2318 char dst_swizzle[6];
2320 shader_glsl_get_swizzle(arg->src[0], FALSE, write_mask, dst_swizzle);
2322 if (src_mod == WINED3DSPSM_DZ) {
2323 glsl_src_param_t div_param;
2324 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2325 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &div_param);
2327 if (mask_size > 1) {
2328 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2329 } else {
2330 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2332 } else if (src_mod == WINED3DSPSM_DW) {
2333 glsl_src_param_t div_param;
2334 unsigned int mask_size = shader_glsl_get_write_mask_size(write_mask);
2335 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &div_param);
2337 if (mask_size > 1) {
2338 shader_addline(buffer, "gl_TexCoord[%u]%s / vec%d(%s));\n", reg, dst_swizzle, mask_size, div_param.param_str);
2339 } else {
2340 shader_addline(buffer, "gl_TexCoord[%u]%s / %s);\n", reg, dst_swizzle, div_param.param_str);
2342 } else {
2343 shader_addline(buffer, "gl_TexCoord[%u]%s);\n", reg, dst_swizzle);
2348 /** Process the WINED3DSIO_TEXDP3TEX instruction in GLSL:
2349 * Take a 3-component dot product of the TexCoord[dstreg] and src,
2350 * then perform a 1D texture lookup from stage dstregnum, place into dst. */
2351 void pshader_glsl_texdp3tex(SHADER_OPCODE_ARG* arg) {
2352 glsl_src_param_t src0_param;
2353 char dst_mask[6];
2354 glsl_sample_function_t sample_function;
2355 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2356 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2357 DWORD sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2359 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2361 shader_glsl_append_dst(arg->buffer, arg);
2362 shader_glsl_get_write_mask(arg->dst, dst_mask);
2364 /* Do I have to take care about the projected bit? I don't think so, since the dp3 returns only one
2365 * scalar, and projected sampling would require 4.
2367 * It is a dependent read - not valid with conditional NP2 textures
2369 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2371 switch(count_bits(sample_function.coord_mask)) {
2372 case 1:
2373 shader_addline(arg->buffer, "%s(Psampler%u, dot(gl_TexCoord[%u].xyz, %s))%s);\n",
2374 sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2375 break;
2377 case 2:
2378 shader_addline(arg->buffer, "%s(Psampler%u, vec2(dot(gl_TexCoord[%u].xyz, %s), 0.0))%s);\n",
2379 sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2380 break;
2382 case 3:
2383 shader_addline(arg->buffer, "%s(Psampler%u, vec3(dot(gl_TexCoord[%u].xyz, %s), 0.0, 0.0))%s);\n",
2384 sample_function.name, sampler_idx, sampler_idx, src0_param.param_str, dst_mask);
2385 break;
2386 default:
2387 FIXME("Unexpected mask bitcount %d\n", count_bits(sample_function.coord_mask));
2391 /** Process the WINED3DSIO_TEXDP3 instruction in GLSL:
2392 * Take a 3-component dot product of the TexCoord[dstreg] and src. */
2393 void pshader_glsl_texdp3(SHADER_OPCODE_ARG* arg) {
2394 glsl_src_param_t src0_param;
2395 DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
2396 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2397 DWORD dst_mask;
2398 unsigned int mask_size;
2400 dst_mask = shader_glsl_append_dst(arg->buffer, arg);
2401 mask_size = shader_glsl_get_write_mask_size(dst_mask);
2402 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2404 if (mask_size > 1) {
2405 shader_addline(arg->buffer, "vec%d(dot(T%u.xyz, %s)));\n", mask_size, dstreg, src0_param.param_str);
2406 } else {
2407 shader_addline(arg->buffer, "dot(T%u.xyz, %s));\n", dstreg, src0_param.param_str);
2411 /** Process the WINED3DSIO_TEXDEPTH instruction in GLSL:
2412 * Calculate the depth as dst.x / dst.y */
2413 void pshader_glsl_texdepth(SHADER_OPCODE_ARG* arg) {
2414 glsl_dst_param_t dst_param;
2416 shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
2418 /* Tests show that texdepth never returns anything below 0.0, and that r5.y is clamped to 1.0.
2419 * Negative input is accepted, -0.25 / -0.5 returns 0.5. GL should clamp gl_FragDepth to [0;1], but
2420 * this doesn't always work, so clamp the results manually. Whether or not the x value is clamped at 1
2421 * too is irrelevant, since if x = 0, any y value < 1.0 (and > 1.0 is not allowed) results in a result
2422 * >= 1.0 or < 0.0
2424 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);
2427 /** Process the WINED3DSIO_TEXM3X2DEPTH instruction in GLSL:
2428 * Last row of a 3x2 matrix multiply, use the result to calculate the depth:
2429 * Calculate tmp0.y = TexCoord[dstreg] . src.xyz; (tmp0.x has already been calculated)
2430 * depth = (tmp0.y == 0.0) ? 1.0 : tmp0.x / tmp0.y
2432 void pshader_glsl_texm3x2depth(SHADER_OPCODE_ARG* arg) {
2433 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2434 DWORD dstreg = arg->dst & WINED3DSP_REGNUM_MASK;
2435 glsl_src_param_t src0_param;
2437 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2439 shader_addline(arg->buffer, "tmp0.y = dot(T%u.xyz, %s);\n", dstreg, src0_param.param_str);
2440 shader_addline(arg->buffer, "gl_FragDepth = (tmp0.y == 0.0) ? 1.0 : clamp(tmp0.x / tmp0.y, 0.0, 1.0);\n");
2443 /** Process the WINED3DSIO_TEXM3X2PAD instruction in GLSL
2444 * Calculate the 1st of a 2-row matrix multiplication. */
2445 void pshader_glsl_texm3x2pad(SHADER_OPCODE_ARG* arg) {
2446 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2447 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2448 SHADER_BUFFER* buffer = arg->buffer;
2449 glsl_src_param_t src0_param;
2451 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2452 shader_addline(buffer, "tmp0.x = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2455 /** Process the WINED3DSIO_TEXM3X3PAD instruction in GLSL
2456 * Calculate the 1st or 2nd row of a 3-row matrix multiplication. */
2457 void pshader_glsl_texm3x3pad(SHADER_OPCODE_ARG* arg) {
2459 IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2460 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2461 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2462 SHADER_BUFFER* buffer = arg->buffer;
2463 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2464 glsl_src_param_t src0_param;
2466 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2467 shader_addline(buffer, "tmp0.%c = dot(T%u.xyz, %s);\n", 'x' + current_state->current_row, reg, src0_param.param_str);
2468 current_state->texcoord_w[current_state->current_row++] = reg;
2471 void pshader_glsl_texm3x2tex(SHADER_OPCODE_ARG* arg) {
2472 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2473 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2474 SHADER_BUFFER* buffer = arg->buffer;
2475 glsl_src_param_t src0_param;
2476 char dst_mask[6];
2478 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2479 shader_addline(buffer, "tmp0.y = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2481 shader_glsl_append_dst(buffer, arg);
2482 shader_glsl_get_write_mask(arg->dst, dst_mask);
2484 /* Sample the texture using the calculated coordinates */
2485 shader_addline(buffer, "texture2D(Psampler%u, tmp0.xy)%s);\n", reg, dst_mask);
2488 /** Process the WINED3DSIO_TEXM3X3TEX instruction in GLSL
2489 * Perform the 3rd row of a 3x3 matrix multiply, then sample the texture using the calculated coordinates */
2490 void pshader_glsl_texm3x3tex(SHADER_OPCODE_ARG* arg) {
2491 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2492 glsl_src_param_t src0_param;
2493 char dst_mask[6];
2494 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2495 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2496 SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2497 DWORD sampler_type = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2498 glsl_sample_function_t sample_function;
2500 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2501 shader_addline(arg->buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2503 shader_glsl_append_dst(arg->buffer, arg);
2504 shader_glsl_get_write_mask(arg->dst, dst_mask);
2505 /* Dependent read, not valid with conditional NP2 */
2506 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2508 /* Sample the texture using the calculated coordinates */
2509 shader_addline(arg->buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2511 current_state->current_row = 0;
2514 /** Process the WINED3DSIO_TEXM3X3 instruction in GLSL
2515 * Perform the 3rd row of a 3x3 matrix multiply */
2516 void pshader_glsl_texm3x3(SHADER_OPCODE_ARG* arg) {
2517 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2518 glsl_src_param_t src0_param;
2519 char dst_mask[6];
2520 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2521 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2522 SHADER_PARSE_STATE* current_state = &This->baseShader.parse_state;
2524 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2526 shader_glsl_append_dst(arg->buffer, arg);
2527 shader_glsl_get_write_mask(arg->dst, dst_mask);
2528 shader_addline(arg->buffer, "vec4(tmp0.xy, dot(T%u.xyz, %s), 1.0)%s);\n", reg, src0_param.param_str, dst_mask);
2530 current_state->current_row = 0;
2533 /** Process the WINED3DSIO_TEXM3X3SPEC instruction in GLSL
2534 * Peform the final texture lookup based on the previous 2 3x3 matrix multiplies */
2535 void pshader_glsl_texm3x3spec(SHADER_OPCODE_ARG* arg) {
2537 IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2538 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2539 glsl_src_param_t src0_param;
2540 glsl_src_param_t src1_param;
2541 char dst_mask[6];
2542 SHADER_BUFFER* buffer = arg->buffer;
2543 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2544 DWORD stype = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2545 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2546 glsl_sample_function_t sample_function;
2548 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2549 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], src_mask, &src1_param);
2551 /* Perform the last matrix multiply operation */
2552 shader_addline(buffer, "tmp0.z = dot(T%u.xyz, %s);\n", reg, src0_param.param_str);
2553 /* Reflection calculation */
2554 shader_addline(buffer, "tmp0.xyz = -reflect((%s), normalize(tmp0.xyz));\n", src1_param.param_str);
2556 shader_glsl_append_dst(buffer, arg);
2557 shader_glsl_get_write_mask(arg->dst, dst_mask);
2558 /* Dependent read, not valid with conditional NP2 */
2559 shader_glsl_get_sample_function(stype, FALSE, FALSE, &sample_function);
2561 /* Sample the texture */
2562 shader_addline(buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2564 current_state->current_row = 0;
2567 /** Process the WINED3DSIO_TEXM3X3VSPEC instruction in GLSL
2568 * Peform the final texture lookup based on the previous 2 3x3 matrix multiplies */
2569 void pshader_glsl_texm3x3vspec(SHADER_OPCODE_ARG* arg) {
2571 IWineD3DPixelShaderImpl* shader = (IWineD3DPixelShaderImpl*) arg->shader;
2572 DWORD reg = arg->dst & WINED3DSP_REGNUM_MASK;
2573 SHADER_BUFFER* buffer = arg->buffer;
2574 SHADER_PARSE_STATE* current_state = &shader->baseShader.parse_state;
2575 glsl_src_param_t src0_param;
2576 char dst_mask[6];
2577 DWORD src_mask = WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1 | WINED3DSP_WRITEMASK_2;
2578 DWORD sampler_type = arg->reg_maps->samplers[reg] & WINED3DSP_TEXTURETYPE_MASK;
2579 glsl_sample_function_t sample_function;
2581 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], src_mask, &src0_param);
2583 /* Perform the last matrix multiply operation */
2584 shader_addline(buffer, "tmp0.z = dot(vec3(T%u), vec3(%s));\n", reg, src0_param.param_str);
2586 /* Construct the eye-ray vector from w coordinates */
2587 shader_addline(buffer, "tmp1.xyz = normalize(vec3(gl_TexCoord[%u].w, gl_TexCoord[%u].w, gl_TexCoord[%u].w));\n",
2588 current_state->texcoord_w[0], current_state->texcoord_w[1], reg);
2589 shader_addline(buffer, "tmp0.xyz = -reflect(tmp1.xyz, normalize(tmp0.xyz));\n");
2591 shader_glsl_append_dst(buffer, arg);
2592 shader_glsl_get_write_mask(arg->dst, dst_mask);
2593 /* Dependent read, not valid with conditional NP2 */
2594 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2596 /* Sample the texture using the calculated coordinates */
2597 shader_addline(buffer, "%s(Psampler%u, tmp0.xyz)%s);\n", sample_function.name, reg, dst_mask);
2599 current_state->current_row = 0;
2602 /** Process the WINED3DSIO_TEXBEM instruction in GLSL.
2603 * Apply a fake bump map transform.
2604 * texbem is pshader <= 1.3 only, this saves a few version checks
2606 void pshader_glsl_texbem(SHADER_OPCODE_ARG* arg) {
2607 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2608 IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
2609 char dst_swizzle[6];
2610 glsl_sample_function_t sample_function;
2611 glsl_src_param_t coord_param;
2612 DWORD sampler_type;
2613 DWORD sampler_idx;
2614 DWORD mask;
2615 DWORD flags;
2616 char coord_mask[6];
2618 sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2619 flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];
2621 sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2622 /* Dependent read, not valid with conditional NP2 */
2623 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2624 mask = sample_function.coord_mask;
2626 shader_glsl_get_write_mask(arg->dst, dst_swizzle);
2628 shader_glsl_get_write_mask(mask, coord_mask);
2630 /* with projective textures, texbem only divides the static texture coord, not the displacement,
2631 * so we can't let the GL handle this.
2633 if (flags & WINED3DTTFF_PROJECTED) {
2634 DWORD div_mask=0;
2635 char coord_div_mask[3];
2636 switch (flags & ~WINED3DTTFF_PROJECTED) {
2637 case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
2638 case WINED3DTTFF_COUNT2: div_mask = WINED3DSP_WRITEMASK_1; break;
2639 case WINED3DTTFF_COUNT3: div_mask = WINED3DSP_WRITEMASK_2; break;
2640 case WINED3DTTFF_COUNT4:
2641 case WINED3DTTFF_DISABLE: div_mask = WINED3DSP_WRITEMASK_3; break;
2643 shader_glsl_get_write_mask(div_mask, coord_div_mask);
2644 shader_addline(arg->buffer, "T%u%s /= T%u%s;\n", sampler_idx, coord_mask, sampler_idx, coord_div_mask);
2647 shader_glsl_append_dst(arg->buffer, arg);
2648 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &coord_param);
2649 if(arg->opcode->opcode == WINED3DSIO_TEXBEML) {
2650 glsl_src_param_t luminance_param;
2651 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &luminance_param);
2652 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",
2653 sample_function.name, sampler_idx, sampler_idx, coord_mask, sampler_idx, coord_param.param_str, coord_mask,
2654 luminance_param.param_str, sampler_idx, sampler_idx, dst_swizzle);
2655 } else {
2656 shader_addline(arg->buffer, "%s(Psampler%u, T%u%s + vec4(bumpenvmat%d * %s, 0.0, 0.0)%s )%s);\n",
2657 sample_function.name, sampler_idx, sampler_idx, coord_mask, sampler_idx, coord_param.param_str, coord_mask, dst_swizzle);
2661 void pshader_glsl_bem(SHADER_OPCODE_ARG* arg) {
2662 glsl_src_param_t src0_param, src1_param;
2663 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2665 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &src0_param);
2666 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0|WINED3DSP_WRITEMASK_1, &src1_param);
2668 shader_glsl_append_dst(arg->buffer, arg);
2669 shader_addline(arg->buffer, "%s + bumpenvmat%d * %s);\n",
2670 src0_param.param_str, sampler_idx, src1_param.param_str);
2673 /** Process the WINED3DSIO_TEXREG2AR instruction in GLSL
2674 * Sample 2D texture at dst using the alpha & red (wx) components of src as texture coordinates */
2675 void pshader_glsl_texreg2ar(SHADER_OPCODE_ARG* arg) {
2677 glsl_src_param_t src0_param;
2678 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2679 char dst_mask[6];
2681 shader_glsl_append_dst(arg->buffer, arg);
2682 shader_glsl_get_write_mask(arg->dst, dst_mask);
2683 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2685 shader_addline(arg->buffer, "texture2D(Psampler%u, %s.wx)%s);\n", sampler_idx, src0_param.reg_name, dst_mask);
2688 /** Process the WINED3DSIO_TEXREG2GB instruction in GLSL
2689 * Sample 2D texture at dst using the green & blue (yz) components of src as texture coordinates */
2690 void pshader_glsl_texreg2gb(SHADER_OPCODE_ARG* arg) {
2691 glsl_src_param_t src0_param;
2692 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2693 char dst_mask[6];
2695 shader_glsl_append_dst(arg->buffer, arg);
2696 shader_glsl_get_write_mask(arg->dst, dst_mask);
2697 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_ALL, &src0_param);
2699 shader_addline(arg->buffer, "texture2D(Psampler%u, %s.yz)%s);\n", sampler_idx, src0_param.reg_name, dst_mask);
2702 /** Process the WINED3DSIO_TEXREG2RGB instruction in GLSL
2703 * Sample texture at dst using the rgb (xyz) components of src as texture coordinates */
2704 void pshader_glsl_texreg2rgb(SHADER_OPCODE_ARG* arg) {
2705 glsl_src_param_t src0_param;
2706 char dst_mask[6];
2707 DWORD sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
2708 DWORD sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
2709 glsl_sample_function_t sample_function;
2711 shader_glsl_append_dst(arg->buffer, arg);
2712 shader_glsl_get_write_mask(arg->dst, dst_mask);
2713 /* Dependent read, not valid with conditional NP2 */
2714 shader_glsl_get_sample_function(sampler_type, FALSE, FALSE, &sample_function);
2715 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], sample_function.coord_mask, &src0_param);
2717 shader_addline(arg->buffer, "%s(Psampler%u, %s)%s);\n", sample_function.name, sampler_idx, src0_param.param_str, dst_mask);
2720 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
2721 * If any of the first 3 components are < 0, discard this pixel */
2722 void pshader_glsl_texkill(SHADER_OPCODE_ARG* arg) {
2723 IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
2724 DWORD hex_version = This->baseShader.hex_version;
2725 glsl_dst_param_t dst_param;
2727 /* The argument is a destination parameter, and no writemasks are allowed */
2728 shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
2729 if((hex_version >= WINED3DPS_VERSION(2,0))) {
2730 /* 2.0 shaders compare all 4 components in texkill */
2731 shader_addline(arg->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
2732 } else {
2733 /* 1.X shaders only compare the first 3 components, probably due to the nature of the texkill
2734 * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
2735 * 4 components are defined, only the first 3 are used
2737 shader_addline(arg->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
2741 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.
2742 * dst = dot2(src0, src1) + src2 */
2743 void pshader_glsl_dp2add(SHADER_OPCODE_ARG* arg) {
2744 glsl_src_param_t src0_param;
2745 glsl_src_param_t src1_param;
2746 glsl_src_param_t src2_param;
2747 DWORD write_mask;
2748 unsigned int mask_size;
2750 write_mask = shader_glsl_append_dst(arg->buffer, arg);
2751 mask_size = shader_glsl_get_write_mask_size(write_mask);
2753 shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src0_param);
2754 shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1, &src1_param);
2755 shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], WINED3DSP_WRITEMASK_0, &src2_param);
2757 if (mask_size > 1) {
2758 shader_addline(arg->buffer, "vec%d(dot(%s, %s) + %s));\n", mask_size, src0_param.param_str, src1_param.param_str, src2_param.param_str);
2759 } else {
2760 shader_addline(arg->buffer, "dot(%s, %s) + %s);\n", src0_param.param_str, src1_param.param_str, src2_param.param_str);
2764 void pshader_glsl_input_pack(
2765 SHADER_BUFFER* buffer,
2766 semantic* semantics_in,
2767 IWineD3DPixelShader *iface) {
2769 unsigned int i;
2770 IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *) iface;
2772 for (i = 0; i < MAX_REG_INPUT; i++) {
2774 DWORD usage_token = semantics_in[i].usage;
2775 DWORD register_token = semantics_in[i].reg;
2776 DWORD usage, usage_idx;
2777 char reg_mask[6];
2779 /* Uninitialized */
2780 if (!usage_token) continue;
2781 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2782 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2783 shader_glsl_get_write_mask(register_token, reg_mask);
2785 switch(usage) {
2787 case WINED3DDECLUSAGE_TEXCOORD:
2788 if(usage_idx < 8 && This->vertexprocessing == pretransformed) {
2789 shader_addline(buffer, "IN[%u]%s = gl_TexCoord[%u]%s;\n",
2790 This->input_reg_map[i], reg_mask, usage_idx, reg_mask);
2791 } else {
2792 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2793 This->input_reg_map[i], reg_mask, reg_mask);
2795 break;
2797 case WINED3DDECLUSAGE_COLOR:
2798 if (usage_idx == 0)
2799 shader_addline(buffer, "IN[%u]%s = vec4(gl_Color)%s;\n",
2800 This->input_reg_map[i], reg_mask, reg_mask);
2801 else if (usage_idx == 1)
2802 shader_addline(buffer, "IN[%u]%s = vec4(gl_SecondaryColor)%s;\n",
2803 This->input_reg_map[i], reg_mask, reg_mask);
2804 else
2805 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2806 This->input_reg_map[i], reg_mask, reg_mask);
2807 break;
2809 default:
2810 shader_addline(buffer, "IN[%u]%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2811 This->input_reg_map[i], reg_mask, reg_mask);
2816 /*********************************************
2817 * Vertex Shader Specific Code begins here
2818 ********************************************/
2820 static void add_glsl_program_entry(struct shader_glsl_priv *priv, struct glsl_shader_prog_link *entry) {
2821 glsl_program_key_t *key;
2823 key = HeapAlloc(GetProcessHeap(), 0, sizeof(glsl_program_key_t));
2824 key->vshader = entry->vshader;
2825 key->pshader = entry->pshader;
2827 hash_table_put(priv->glsl_program_lookup, key, entry);
2830 static struct glsl_shader_prog_link *get_glsl_program_entry(struct shader_glsl_priv *priv,
2831 GLhandleARB vshader, GLhandleARB pshader) {
2832 glsl_program_key_t key;
2834 key.vshader = vshader;
2835 key.pshader = pshader;
2837 return (struct glsl_shader_prog_link *)hash_table_get(priv->glsl_program_lookup, &key);
2840 void delete_glsl_program_entry(struct shader_glsl_priv *priv, WineD3D_GL_Info *gl_info, struct glsl_shader_prog_link *entry) {
2841 glsl_program_key_t *key;
2843 key = HeapAlloc(GetProcessHeap(), 0, sizeof(glsl_program_key_t));
2844 key->vshader = entry->vshader;
2845 key->pshader = entry->pshader;
2846 hash_table_remove(priv->glsl_program_lookup, key);
2848 GL_EXTCALL(glDeleteObjectARB(entry->programId));
2849 if (entry->vshader) list_remove(&entry->vshader_entry);
2850 if (entry->pshader) list_remove(&entry->pshader_entry);
2851 HeapFree(GetProcessHeap(), 0, entry->vuniformF_locations);
2852 HeapFree(GetProcessHeap(), 0, entry->puniformF_locations);
2853 HeapFree(GetProcessHeap(), 0, entry);
2856 static void handle_ps3_input(SHADER_BUFFER *buffer, semantic *semantics_in, semantic *semantics_out, WineD3D_GL_Info *gl_info, DWORD *map) {
2857 unsigned int i, j;
2858 DWORD usage_token, usage_token_out;
2859 DWORD register_token, register_token_out;
2860 DWORD usage, usage_idx, usage_out, usage_idx_out;
2861 DWORD *set;
2862 DWORD in_idx;
2863 DWORD in_count = GL_LIMITS(glsl_varyings) / 4;
2864 char reg_mask[6], reg_mask_out[6];
2865 char destination[50];
2867 set = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(*set) * (in_count + 2));
2869 if (!semantics_out) {
2870 /* Save gl_FrontColor & gl_FrontSecondaryColor before overwriting them. */
2871 shader_addline(buffer, "vec4 front_color = gl_FrontColor;\n");
2872 shader_addline(buffer, "vec4 front_secondary_color = gl_FrontSecondaryColor;\n");
2875 for(i = 0; i < MAX_REG_INPUT; i++) {
2876 usage_token = semantics_in[i].usage;
2877 if (!usage_token) continue;
2879 in_idx = map[i];
2880 if (in_idx >= (in_count + 2)) {
2881 FIXME("More input varyings declared than supported, expect issues\n");
2882 continue;
2883 } else if(map[i] == -1) {
2884 /* Declared, but not read register */
2885 continue;
2888 if (in_idx == in_count) {
2889 sprintf(destination, "gl_FrontColor");
2890 } else if (in_idx == in_count + 1) {
2891 sprintf(destination, "gl_FrontSecondaryColor");
2892 } else {
2893 sprintf(destination, "IN[%u]", in_idx);
2896 register_token = semantics_in[i].reg;
2898 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2899 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2900 set[map[i]] = shader_glsl_get_write_mask(register_token, reg_mask);
2902 if(!semantics_out) {
2903 switch(usage) {
2904 case WINED3DDECLUSAGE_COLOR:
2905 if (usage_idx == 0)
2906 shader_addline(buffer, "%s%s = front_color%s;\n",
2907 destination, reg_mask, reg_mask);
2908 else if (usage_idx == 1)
2909 shader_addline(buffer, "%s%s = front_secondary_color%s;\n",
2910 destination, reg_mask, reg_mask);
2911 else
2912 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2913 destination, reg_mask, reg_mask);
2914 break;
2916 case WINED3DDECLUSAGE_TEXCOORD:
2917 if (usage_idx < 8) {
2918 shader_addline(buffer, "%s%s = gl_TexCoord[%u]%s;\n",
2919 destination, reg_mask, usage_idx, reg_mask);
2920 } else {
2921 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2922 destination, reg_mask, reg_mask);
2924 break;
2926 case WINED3DDECLUSAGE_FOG:
2927 shader_addline(buffer, "%s%s = vec4(gl_FogFragCoord, 0.0, 0.0, 0.0)%s;\n",
2928 destination, reg_mask, reg_mask);
2929 break;
2931 default:
2932 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2933 destination, reg_mask, reg_mask);
2935 } else {
2936 BOOL found = FALSE;
2937 for(j = 0; j < MAX_REG_OUTPUT; j++) {
2938 usage_token_out = semantics_out[j].usage;
2939 if (!usage_token_out) continue;
2940 register_token_out = semantics_out[j].reg;
2942 usage_out = (usage_token_out & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
2943 usage_idx_out = (usage_token_out & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
2944 shader_glsl_get_write_mask(register_token_out, reg_mask_out);
2946 if(usage == usage_out &&
2947 usage_idx == usage_idx_out) {
2948 shader_addline(buffer, "%s%s = OUT[%u]%s;\n",
2949 destination, reg_mask, j, reg_mask);
2950 found = TRUE;
2953 if(!found) {
2954 shader_addline(buffer, "%s%s = vec4(0.0, 0.0, 0.0, 0.0)%s;\n",
2955 destination, reg_mask, reg_mask);
2960 /* This is solely to make the compiler / linker happy and avoid warning about undefined
2961 * varyings. It shouldn't result in any real code executed on the GPU, since all read
2962 * input varyings are assigned above, if the optimizer works properly.
2964 for(i = 0; i < in_count + 2; i++) {
2965 if(set[i] != WINED3DSP_WRITEMASK_ALL) {
2966 unsigned int size = 0;
2967 memset(reg_mask, 0, sizeof(reg_mask));
2968 if(!(set[i] & WINED3DSP_WRITEMASK_0)) {
2969 reg_mask[size] = 'x';
2970 size++;
2972 if(!(set[i] & WINED3DSP_WRITEMASK_1)) {
2973 reg_mask[size] = 'y';
2974 size++;
2976 if(!(set[i] & WINED3DSP_WRITEMASK_2)) {
2977 reg_mask[size] = 'z';
2978 size++;
2980 if(!(set[i] & WINED3DSP_WRITEMASK_3)) {
2981 reg_mask[size] = 'w';
2982 size++;
2985 if (i == in_count) {
2986 sprintf(destination, "gl_FrontColor");
2987 } else if (i == in_count + 1) {
2988 sprintf(destination, "gl_FrontSecondaryColor");
2989 } else {
2990 sprintf(destination, "IN[%u]", i);
2993 if (size == 1) {
2994 shader_addline(buffer, "%s.%s = 0.0;\n", destination, reg_mask);
2995 } else {
2996 shader_addline(buffer, "%s.%s = vec%u(0.0);\n", destination, reg_mask, size);
3001 HeapFree(GetProcessHeap(), 0, set);
3004 static GLhandleARB generate_param_reorder_function(IWineD3DVertexShader *vertexshader,
3005 IWineD3DPixelShader *pixelshader,
3006 WineD3D_GL_Info *gl_info) {
3007 GLhandleARB ret = 0;
3008 IWineD3DVertexShaderImpl *vs = (IWineD3DVertexShaderImpl *) vertexshader;
3009 IWineD3DPixelShaderImpl *ps = (IWineD3DPixelShaderImpl *) pixelshader;
3010 DWORD vs_major = vs ? WINED3DSHADER_VERSION_MAJOR(vs->baseShader.hex_version) : 0;
3011 DWORD ps_major = ps ? WINED3DSHADER_VERSION_MAJOR(ps->baseShader.hex_version) : 0;
3012 unsigned int i;
3013 SHADER_BUFFER buffer;
3014 DWORD usage_token;
3015 DWORD register_token;
3016 DWORD usage, usage_idx, writemask;
3017 char reg_mask[6];
3018 semantic *semantics_out, *semantics_in;
3020 buffer.buffer = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, SHADER_PGMSIZE);
3021 buffer.bsize = 0;
3022 buffer.lineNo = 0;
3023 buffer.newline = TRUE;
3025 shader_addline(&buffer, "#version 120\n");
3027 if(vs_major < 3 && ps_major < 3) {
3028 /* That one is easy: The vertex shader writes to the builtin varyings, the pixel shader reads from them.
3029 * Take care about the texcoord .w fixup though if we're using the fixed function fragment pipeline
3031 if((GLINFO_LOCATION).set_texcoord_w && ps_major == 0 && vs_major > 0) {
3032 shader_addline(&buffer, "void order_ps_input() {\n");
3033 for(i = 0; i < min(8, MAX_REG_TEXCRD); i++) {
3034 if(vs->baseShader.reg_maps.texcoord_mask[i] != 0 &&
3035 vs->baseShader.reg_maps.texcoord_mask[i] != WINED3DSP_WRITEMASK_ALL) {
3036 shader_addline(&buffer, "gl_TexCoord[%u].w = 1.0;\n", i);
3039 shader_addline(&buffer, "}\n");
3040 } else {
3041 shader_addline(&buffer, "void order_ps_input() { /* do nothing */ }\n");
3043 } else if(ps_major < 3 && vs_major >= 3) {
3044 /* The vertex shader writes to its own varyings, the pixel shader needs them in the builtin ones */
3045 semantics_out = vs->semantics_out;
3047 shader_addline(&buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3048 for(i = 0; i < MAX_REG_OUTPUT; i++) {
3049 usage_token = semantics_out[i].usage;
3050 if (!usage_token) continue;
3051 register_token = semantics_out[i].reg;
3053 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
3054 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
3055 writemask = shader_glsl_get_write_mask(register_token, reg_mask);
3057 switch(usage) {
3058 case WINED3DDECLUSAGE_COLOR:
3059 if (usage_idx == 0)
3060 shader_addline(&buffer, "gl_FrontColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3061 else if (usage_idx == 1)
3062 shader_addline(&buffer, "gl_FrontSecondaryColor%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3063 break;
3065 case WINED3DDECLUSAGE_POSITION:
3066 shader_addline(&buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3067 break;
3069 case WINED3DDECLUSAGE_TEXCOORD:
3070 if (usage_idx < 8) {
3071 if(!(GLINFO_LOCATION).set_texcoord_w || ps_major > 0) writemask |= WINED3DSP_WRITEMASK_3;
3073 shader_addline(&buffer, "gl_TexCoord[%u]%s = OUT[%u]%s;\n",
3074 usage_idx, reg_mask, i, reg_mask);
3075 if(!(writemask & WINED3DSP_WRITEMASK_3)) {
3076 shader_addline(&buffer, "gl_TexCoord[%u].w = 1.0;\n", usage_idx);
3079 break;
3081 case WINED3DDECLUSAGE_PSIZE:
3082 shader_addline(&buffer, "gl_PointSize = OUT[%u].x;\n", i);
3083 break;
3085 case WINED3DDECLUSAGE_FOG:
3086 shader_addline(&buffer, "gl_FogFragCoord = OUT[%u].%c;\n", i, reg_mask[1]);
3087 break;
3089 default:
3090 break;
3093 shader_addline(&buffer, "}\n");
3095 } else if(ps_major >= 3 && vs_major >= 3) {
3096 semantics_out = vs->semantics_out;
3097 semantics_in = ps->semantics_in;
3099 /* This one is tricky: a 3.0 pixel shader reads from a 3.0 vertex shader */
3100 shader_addline(&buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
3101 shader_addline(&buffer, "void order_ps_input(in vec4 OUT[%u]) {\n", MAX_REG_OUTPUT);
3103 /* First, sort out position and point size. Those are not passed to the pixel shader */
3104 for(i = 0; i < MAX_REG_OUTPUT; i++) {
3105 usage_token = semantics_out[i].usage;
3106 if (!usage_token) continue;
3107 register_token = semantics_out[i].reg;
3109 usage = (usage_token & WINED3DSP_DCL_USAGE_MASK) >> WINED3DSP_DCL_USAGE_SHIFT;
3110 usage_idx = (usage_token & WINED3DSP_DCL_USAGEINDEX_MASK) >> WINED3DSP_DCL_USAGEINDEX_SHIFT;
3111 shader_glsl_get_write_mask(register_token, reg_mask);
3113 switch(usage) {
3114 case WINED3DDECLUSAGE_POSITION:
3115 shader_addline(&buffer, "gl_Position%s = OUT[%u]%s;\n", reg_mask, i, reg_mask);
3116 break;
3118 case WINED3DDECLUSAGE_PSIZE:
3119 shader_addline(&buffer, "gl_PointSize = OUT[%u].x;\n", i);
3120 break;
3122 default:
3123 break;
3127 /* Then, fix the pixel shader input */
3128 handle_ps3_input(&buffer, semantics_in, semantics_out, gl_info, ps->input_reg_map);
3130 shader_addline(&buffer, "}\n");
3131 } else if(ps_major >= 3 && vs_major < 3) {
3132 semantics_in = ps->semantics_in;
3134 shader_addline(&buffer, "varying vec4 IN[%u];\n", GL_LIMITS(glsl_varyings) / 4);
3135 shader_addline(&buffer, "void order_ps_input() {\n");
3136 /* The vertex shader wrote to the builtin varyings. There is no need to figure out position and
3137 * point size, but we depend on the optimizers kindness to find out that the pixel shader doesn't
3138 * read gl_TexCoord and gl_ColorX, otherwise we'll run out of varyings
3140 handle_ps3_input(&buffer, semantics_in, NULL, gl_info, ps->input_reg_map);
3141 shader_addline(&buffer, "}\n");
3142 } else {
3143 ERR("Unexpected vertex and pixel shader version condition: vs: %d, ps: %d\n", vs_major, ps_major);
3146 ret = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3147 checkGLcall("glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB)");
3148 GL_EXTCALL(glShaderSourceARB(ret, 1, (const char**)&buffer.buffer, NULL));
3149 checkGLcall("glShaderSourceARB(ret, 1, (const char**)&buffer.buffer, NULL)");
3150 GL_EXTCALL(glCompileShaderARB(ret));
3151 checkGLcall("glCompileShaderARB(ret)");
3153 HeapFree(GetProcessHeap(), 0, buffer.buffer);
3154 return ret;
3157 static void hardcode_local_constants(IWineD3DBaseShaderImpl *shader, WineD3D_GL_Info *gl_info, GLhandleARB programId, char prefix) {
3158 local_constant* lconst;
3159 GLuint tmp_loc;
3160 float *value;
3161 char glsl_name[8];
3163 LIST_FOR_EACH_ENTRY(lconst, &shader->baseShader.constantsF, local_constant, entry) {
3164 value = (float *) lconst->value;
3165 snprintf(glsl_name, sizeof(glsl_name), "%cLC%u", prefix, lconst->idx);
3166 tmp_loc = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3167 GL_EXTCALL(glUniform4fvARB(tmp_loc, 1, value));
3169 checkGLcall("Hardcoding local constants\n");
3172 /** Sets the GLSL program ID for the given pixel and vertex shader combination.
3173 * It sets the programId on the current StateBlock (because it should be called
3174 * inside of the DrawPrimitive() part of the render loop).
3176 * If a program for the given combination does not exist, create one, and store
3177 * the program in the hash table. If it creates a program, it will link the
3178 * given objects, too.
3180 static void set_glsl_shader_program(IWineD3DDevice *iface, BOOL use_ps, BOOL use_vs) {
3181 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3182 struct shader_glsl_priv *priv = (struct shader_glsl_priv *)This->shader_priv;
3183 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3184 IWineD3DPixelShader *pshader = This->stateBlock->pixelShader;
3185 IWineD3DVertexShader *vshader = This->stateBlock->vertexShader;
3186 struct glsl_shader_prog_link *entry = NULL;
3187 GLhandleARB programId = 0;
3188 GLhandleARB reorder_shader_id = 0;
3189 int i;
3190 char glsl_name[8];
3192 GLhandleARB vshader_id = use_vs ? ((IWineD3DBaseShaderImpl*)vshader)->baseShader.prgId : 0;
3193 GLhandleARB pshader_id = use_ps ? ((IWineD3DBaseShaderImpl*)pshader)->baseShader.prgId : 0;
3194 entry = get_glsl_program_entry(priv, vshader_id, pshader_id);
3195 if (entry) {
3196 priv->glsl_program = entry;
3197 return;
3200 /* If we get to this point, then no matching program exists, so we create one */
3201 programId = GL_EXTCALL(glCreateProgramObjectARB());
3202 TRACE("Created new GLSL shader program %u\n", programId);
3204 /* Create the entry */
3205 entry = HeapAlloc(GetProcessHeap(), 0, sizeof(struct glsl_shader_prog_link));
3206 entry->programId = programId;
3207 entry->vshader = vshader_id;
3208 entry->pshader = pshader_id;
3209 /* Add the hash table entry */
3210 add_glsl_program_entry(priv, entry);
3212 /* Set the current program */
3213 priv->glsl_program = entry;
3215 /* Attach GLSL vshader */
3216 if (vshader_id) {
3217 int max_attribs = 16; /* TODO: Will this always be the case? It is at the moment... */
3218 char tmp_name[10];
3220 reorder_shader_id = generate_param_reorder_function(vshader, pshader, gl_info);
3221 TRACE("Attaching GLSL shader object %u to program %u\n", reorder_shader_id, programId);
3222 GL_EXTCALL(glAttachObjectARB(programId, reorder_shader_id));
3223 checkGLcall("glAttachObjectARB");
3224 /* Flag the reorder function for deletion, then it will be freed automatically when the program
3225 * is destroyed
3227 GL_EXTCALL(glDeleteObjectARB(reorder_shader_id));
3229 TRACE("Attaching GLSL shader object %u to program %u\n", vshader_id, programId);
3230 GL_EXTCALL(glAttachObjectARB(programId, vshader_id));
3231 checkGLcall("glAttachObjectARB");
3233 /* Bind vertex attributes to a corresponding index number to match
3234 * the same index numbers as ARB_vertex_programs (makes loading
3235 * vertex attributes simpler). With this method, we can use the
3236 * exact same code to load the attributes later for both ARB and
3237 * GLSL shaders.
3239 * We have to do this here because we need to know the Program ID
3240 * in order to make the bindings work, and it has to be done prior
3241 * to linking the GLSL program. */
3242 for (i = 0; i < max_attribs; ++i) {
3243 if (((IWineD3DBaseShaderImpl*)vshader)->baseShader.reg_maps.attributes[i]) {
3244 snprintf(tmp_name, sizeof(tmp_name), "attrib%i", i);
3245 GL_EXTCALL(glBindAttribLocationARB(programId, i, tmp_name));
3248 checkGLcall("glBindAttribLocationARB");
3250 list_add_head(&((IWineD3DBaseShaderImpl *)vshader)->baseShader.linked_programs, &entry->vshader_entry);
3253 /* Attach GLSL pshader */
3254 if (pshader_id) {
3255 TRACE("Attaching GLSL shader object %u to program %u\n", pshader_id, programId);
3256 GL_EXTCALL(glAttachObjectARB(programId, pshader_id));
3257 checkGLcall("glAttachObjectARB");
3259 list_add_head(&((IWineD3DBaseShaderImpl *)pshader)->baseShader.linked_programs, &entry->pshader_entry);
3262 /* Link the program */
3263 TRACE("Linking GLSL shader program %u\n", programId);
3264 GL_EXTCALL(glLinkProgramARB(programId));
3265 print_glsl_info_log(&GLINFO_LOCATION, programId);
3267 entry->vuniformF_locations = HeapAlloc(GetProcessHeap(), 0, sizeof(GLhandleARB) * GL_LIMITS(vshader_constantsF));
3268 for (i = 0; i < GL_LIMITS(vshader_constantsF); ++i) {
3269 snprintf(glsl_name, sizeof(glsl_name), "VC[%i]", i);
3270 entry->vuniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3272 for (i = 0; i < MAX_CONST_I; ++i) {
3273 snprintf(glsl_name, sizeof(glsl_name), "VI[%i]", i);
3274 entry->vuniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3276 entry->puniformF_locations = HeapAlloc(GetProcessHeap(), 0, sizeof(GLhandleARB) * GL_LIMITS(pshader_constantsF));
3277 for (i = 0; i < GL_LIMITS(pshader_constantsF); ++i) {
3278 snprintf(glsl_name, sizeof(glsl_name), "PC[%i]", i);
3279 entry->puniformF_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3281 for (i = 0; i < MAX_CONST_I; ++i) {
3282 snprintf(glsl_name, sizeof(glsl_name), "PI[%i]", i);
3283 entry->puniformI_locations[i] = GL_EXTCALL(glGetUniformLocationARB(programId, glsl_name));
3286 if(pshader) {
3287 for(i = 0; i < ((IWineD3DPixelShaderImpl*)pshader)->numbumpenvmatconsts; i++) {
3288 char name[32];
3289 sprintf(name, "bumpenvmat%d", ((IWineD3DPixelShaderImpl*)pshader)->bumpenvmatconst[i].texunit);
3290 entry->bumpenvmat_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3291 sprintf(name, "luminancescale%d", ((IWineD3DPixelShaderImpl*)pshader)->luminanceconst[i].texunit);
3292 entry->luminancescale_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3293 sprintf(name, "luminanceoffset%d", ((IWineD3DPixelShaderImpl*)pshader)->luminanceconst[i].texunit);
3294 entry->luminanceoffset_location[i] = GL_EXTCALL(glGetUniformLocationARB(programId, name));
3299 entry->posFixup_location = GL_EXTCALL(glGetUniformLocationARB(programId, "posFixup"));
3300 entry->srgb_comparison_location = GL_EXTCALL(glGetUniformLocationARB(programId, "srgb_comparison"));
3301 entry->srgb_mul_low_location = GL_EXTCALL(glGetUniformLocationARB(programId, "srgb_mul_low"));
3302 entry->ycorrection_location = GL_EXTCALL(glGetUniformLocationARB(programId, "ycorrection"));
3303 checkGLcall("Find glsl program uniform locations");
3305 if (pshader && WINED3DSHADER_VERSION_MAJOR(((IWineD3DPixelShaderImpl *)pshader)->baseShader.hex_version) >= 3
3306 && ((IWineD3DPixelShaderImpl *)pshader)->declared_in_count > GL_LIMITS(glsl_varyings) / 4) {
3307 TRACE("Shader %d needs vertex color clamping disabled\n", programId);
3308 entry->vertex_color_clamp = GL_FALSE;
3309 } else {
3310 entry->vertex_color_clamp = GL_FIXED_ONLY_ARB;
3313 /* Set the shader to allow uniform loading on it */
3314 GL_EXTCALL(glUseProgramObjectARB(programId));
3315 checkGLcall("glUseProgramObjectARB(programId)");
3317 /* Load the vertex and pixel samplers now. The function that finds the mappings makes sure
3318 * that it stays the same for each vertexshader-pixelshader pair(=linked glsl program). If
3319 * a pshader with fixed function pipeline is used there are no vertex samplers, and if a
3320 * vertex shader with fixed function pixel processing is used we make sure that the card
3321 * supports enough samplers to allow the max number of vertex samplers with all possible
3322 * fixed function fragment processing setups. So once the program is linked these samplers
3323 * won't change.
3325 if(vshader_id) {
3326 /* Load vertex shader samplers */
3327 shader_glsl_load_vsamplers(gl_info, (IWineD3DStateBlock*)This->stateBlock, programId);
3329 if(pshader_id) {
3330 /* Load pixel shader samplers */
3331 shader_glsl_load_psamplers(gl_info, (IWineD3DStateBlock*)This->stateBlock, programId);
3334 /* If the local constants do not have to be loaded with the environment constants,
3335 * load them now to have them hardcoded in the GLSL program. This saves some CPU cycles
3336 * later
3338 if(pshader && !((IWineD3DPixelShaderImpl*)pshader)->baseShader.load_local_constsF) {
3339 hardcode_local_constants((IWineD3DBaseShaderImpl *) pshader, gl_info, programId, 'P');
3341 if(vshader && !((IWineD3DVertexShaderImpl*)vshader)->baseShader.load_local_constsF) {
3342 hardcode_local_constants((IWineD3DBaseShaderImpl *) vshader, gl_info, programId, 'V');
3346 static GLhandleARB create_glsl_blt_shader(WineD3D_GL_Info *gl_info) {
3347 GLhandleARB program_id;
3348 GLhandleARB vshader_id, pshader_id;
3349 const char *blt_vshader[] = {
3350 "#version 120\n"
3351 "void main(void)\n"
3352 "{\n"
3353 " gl_Position = gl_Vertex;\n"
3354 " gl_FrontColor = vec4(1.0);\n"
3355 " gl_TexCoord[0].x = (gl_Vertex.x * 0.5) + 0.5;\n"
3356 " gl_TexCoord[0].y = (-gl_Vertex.y * 0.5) + 0.5;\n"
3357 "}\n"
3360 const char *blt_pshader[] = {
3361 "#version 120\n"
3362 "uniform sampler2D sampler;\n"
3363 "void main(void)\n"
3364 "{\n"
3365 " gl_FragDepth = texture2D(sampler, gl_TexCoord[0].xy).x;\n"
3366 "}\n"
3369 vshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3370 GL_EXTCALL(glShaderSourceARB(vshader_id, 1, blt_vshader, NULL));
3371 GL_EXTCALL(glCompileShaderARB(vshader_id));
3373 pshader_id = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3374 GL_EXTCALL(glShaderSourceARB(pshader_id, 1, blt_pshader, NULL));
3375 GL_EXTCALL(glCompileShaderARB(pshader_id));
3377 program_id = GL_EXTCALL(glCreateProgramObjectARB());
3378 GL_EXTCALL(glAttachObjectARB(program_id, vshader_id));
3379 GL_EXTCALL(glAttachObjectARB(program_id, pshader_id));
3380 GL_EXTCALL(glLinkProgramARB(program_id));
3382 print_glsl_info_log(&GLINFO_LOCATION, program_id);
3384 /* Once linked we can mark the shaders for deletion. They will be deleted once the program
3385 * is destroyed
3387 GL_EXTCALL(glDeleteObjectARB(vshader_id));
3388 GL_EXTCALL(glDeleteObjectARB(pshader_id));
3389 return program_id;
3392 static void shader_glsl_select(IWineD3DDevice *iface, BOOL usePS, BOOL useVS) {
3393 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3394 struct shader_glsl_priv *priv = (struct shader_glsl_priv *)This->shader_priv;
3395 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3396 GLhandleARB program_id = 0;
3397 GLenum old_vertex_color_clamp, current_vertex_color_clamp;
3399 old_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
3401 if (useVS || usePS) set_glsl_shader_program(iface, usePS, useVS);
3402 else priv->glsl_program = NULL;
3404 current_vertex_color_clamp = priv->glsl_program ? priv->glsl_program->vertex_color_clamp : GL_FIXED_ONLY_ARB;
3406 if (old_vertex_color_clamp != current_vertex_color_clamp) {
3407 if (GL_SUPPORT(ARB_COLOR_BUFFER_FLOAT)) {
3408 GL_EXTCALL(glClampColorARB(GL_CLAMP_VERTEX_COLOR_ARB, current_vertex_color_clamp));
3409 checkGLcall("glClampColorARB");
3410 } else {
3411 FIXME("vertex color clamp needs to be changed, but extension not supported.\n");
3415 program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
3416 if (program_id) TRACE("Using GLSL program %u\n", program_id);
3417 GL_EXTCALL(glUseProgramObjectARB(program_id));
3418 checkGLcall("glUseProgramObjectARB");
3421 static void shader_glsl_select_depth_blt(IWineD3DDevice *iface) {
3422 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3423 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3424 struct shader_glsl_priv *priv = (struct shader_glsl_priv *) This->shader_priv;
3425 static GLhandleARB loc = -1;
3427 if (!priv->depth_blt_glsl_program_id) {
3428 priv->depth_blt_glsl_program_id = create_glsl_blt_shader(gl_info);
3429 loc = GL_EXTCALL(glGetUniformLocationARB(priv->depth_blt_glsl_program_id, "sampler"));
3432 GL_EXTCALL(glUseProgramObjectARB(priv->depth_blt_glsl_program_id));
3433 GL_EXTCALL(glUniform1iARB(loc, 0));
3436 static void shader_glsl_deselect_depth_blt(IWineD3DDevice *iface) {
3437 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3438 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3439 struct shader_glsl_priv *priv = (struct shader_glsl_priv *) This->shader_priv;
3440 GLhandleARB program_id;
3442 program_id = priv->glsl_program ? priv->glsl_program->programId : 0;
3443 if (program_id) TRACE("Using GLSL program %u\n", program_id);
3445 GL_EXTCALL(glUseProgramObjectARB(program_id));
3446 checkGLcall("glUseProgramObjectARB");
3449 static void shader_glsl_cleanup(IWineD3DDevice *iface) {
3450 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3451 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3452 GL_EXTCALL(glUseProgramObjectARB(0));
3455 static void shader_glsl_destroy(IWineD3DBaseShader *iface) {
3456 struct list *linked_programs;
3457 IWineD3DBaseShaderImpl *This = (IWineD3DBaseShaderImpl *) iface;
3458 IWineD3DDeviceImpl *device = (IWineD3DDeviceImpl *)This->baseShader.device;
3459 struct shader_glsl_priv *priv = (struct shader_glsl_priv *)device->shader_priv;
3460 WineD3D_GL_Info *gl_info = &device->adapter->gl_info;
3462 /* Note: Do not use QueryInterface here to find out which shader type this is because this code
3463 * can be called from IWineD3DBaseShader::Release
3465 char pshader = shader_is_pshader_version(This->baseShader.hex_version);
3467 if(This->baseShader.prgId == 0) return;
3468 linked_programs = &This->baseShader.linked_programs;
3470 TRACE("Deleting linked programs\n");
3471 if (linked_programs->next) {
3472 struct glsl_shader_prog_link *entry, *entry2;
3474 if(pshader) {
3475 LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, pshader_entry) {
3476 delete_glsl_program_entry(priv, gl_info, entry);
3478 } else {
3479 LIST_FOR_EACH_ENTRY_SAFE(entry, entry2, linked_programs, struct glsl_shader_prog_link, vshader_entry) {
3480 delete_glsl_program_entry(priv, gl_info, entry);
3485 TRACE("Deleting shader object %u\n", This->baseShader.prgId);
3486 GL_EXTCALL(glDeleteObjectARB(This->baseShader.prgId));
3487 checkGLcall("glDeleteObjectARB");
3488 This->baseShader.prgId = 0;
3489 This->baseShader.is_compiled = FALSE;
3492 static unsigned int glsl_program_key_hash(void *key) {
3493 glsl_program_key_t *k = (glsl_program_key_t *)key;
3495 unsigned int hash = k->vshader | k->pshader << 16;
3496 hash += ~(hash << 15);
3497 hash ^= (hash >> 10);
3498 hash += (hash << 3);
3499 hash ^= (hash >> 6);
3500 hash += ~(hash << 11);
3501 hash ^= (hash >> 16);
3503 return hash;
3506 static BOOL glsl_program_key_compare(void *keya, void *keyb) {
3507 glsl_program_key_t *ka = (glsl_program_key_t *)keya;
3508 glsl_program_key_t *kb = (glsl_program_key_t *)keyb;
3510 return ka->vshader == kb->vshader && ka->pshader == kb->pshader;
3513 static HRESULT shader_glsl_alloc(IWineD3DDevice *iface) {
3514 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3515 struct shader_glsl_priv *priv = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(struct shader_glsl_priv));
3516 priv->glsl_program_lookup = hash_table_create(glsl_program_key_hash, glsl_program_key_compare);
3517 This->shader_priv = priv;
3518 return WINED3D_OK;
3521 static void shader_glsl_free(IWineD3DDevice *iface) {
3522 IWineD3DDeviceImpl *This = (IWineD3DDeviceImpl *)iface;
3523 WineD3D_GL_Info *gl_info = &This->adapter->gl_info;
3524 struct shader_glsl_priv *priv = (struct shader_glsl_priv *)This->shader_priv;
3526 if(priv->depth_blt_glsl_program_id) {
3527 GL_EXTCALL(glDeleteObjectARB(priv->depth_blt_glsl_program_id));
3530 hash_table_destroy(priv->glsl_program_lookup, NULL, NULL);
3532 HeapFree(GetProcessHeap(), 0, This->shader_priv);
3533 This->shader_priv = NULL;
3536 static BOOL shader_glsl_dirty_const(IWineD3DDevice *iface) {
3537 /* TODO: GL_EXT_bindable_uniform can be used to share constants across shaders */
3538 return FALSE;
3541 static void shader_glsl_generate_pshader(IWineD3DPixelShader *iface, SHADER_BUFFER *buffer) {
3542 IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
3543 shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3544 CONST DWORD *function = This->baseShader.function;
3545 const char *fragcolor;
3546 WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3548 /* Create the hw GLSL shader object and assign it as the baseShader.prgId */
3549 GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_FRAGMENT_SHADER_ARB));
3551 shader_addline(buffer, "#version 120\n");
3553 if (GL_SUPPORT(ARB_DRAW_BUFFERS)) {
3554 shader_addline(buffer, "#extension GL_ARB_draw_buffers : enable\n");
3556 if (GL_SUPPORT(ARB_TEXTURE_RECTANGLE)) {
3557 /* The spec says that it doesn't have to be explicitly enabled, but the nvidia
3558 * drivers write a warning if we don't do so
3560 shader_addline(buffer, "#extension GL_ARB_texture_rectangle : enable\n");
3563 /* Base Declarations */
3564 shader_generate_glsl_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION);
3566 /* Pack 3.0 inputs */
3567 if (This->baseShader.hex_version >= WINED3DPS_VERSION(3,0)) {
3569 if(((IWineD3DDeviceImpl *) This->baseShader.device)->strided_streams.u.s.position_transformed) {
3570 This->vertexprocessing = pretransformed;
3571 pshader_glsl_input_pack(buffer, This->semantics_in, iface);
3572 } else if(!use_vs((IWineD3DDeviceImpl *) This->baseShader.device)) {
3573 This->vertexprocessing = fixedfunction;
3574 pshader_glsl_input_pack(buffer, This->semantics_in, iface);
3575 } else {
3576 This->vertexprocessing = vertexshader;
3580 /* Base Shader Body */
3581 shader_generate_main( (IWineD3DBaseShader*) This, buffer, reg_maps, function);
3583 /* Pixel shaders < 2.0 place the resulting color in R0 implicitly */
3584 if (This->baseShader.hex_version < WINED3DPS_VERSION(2,0)) {
3585 /* Some older cards like GeforceFX ones don't support multiple buffers, so also not gl_FragData */
3586 if(GL_SUPPORT(ARB_DRAW_BUFFERS))
3587 shader_addline(buffer, "gl_FragData[0] = R0;\n");
3588 else
3589 shader_addline(buffer, "gl_FragColor = R0;\n");
3592 if(GL_SUPPORT(ARB_DRAW_BUFFERS)) {
3593 fragcolor = "gl_FragData[0]";
3594 } else {
3595 fragcolor = "gl_FragColor";
3597 if(This->srgb_enabled) {
3598 shader_addline(buffer, "tmp0.xyz = pow(%s.xyz, vec3(%f, %f, %f)) * vec3(%f, %f, %f) - vec3(%f, %f, %f);\n",
3599 fragcolor, srgb_pow, srgb_pow, srgb_pow, srgb_mul_high, srgb_mul_high, srgb_mul_high,
3600 srgb_sub_high, srgb_sub_high, srgb_sub_high);
3601 shader_addline(buffer, "tmp1.xyz = %s.xyz * srgb_mul_low.xyz;\n", fragcolor);
3602 shader_addline(buffer, "%s.x = %s.x < srgb_comparison.x ? tmp1.x : tmp0.x;\n", fragcolor, fragcolor);
3603 shader_addline(buffer, "%s.y = %s.y < srgb_comparison.y ? tmp1.y : tmp0.y;\n", fragcolor, fragcolor);
3604 shader_addline(buffer, "%s.z = %s.z < srgb_comparison.z ? tmp1.z : tmp0.z;\n", fragcolor, fragcolor);
3605 shader_addline(buffer, "%s = clamp(%s, 0.0, 1.0);\n", fragcolor, fragcolor);
3607 /* Pixel shader < 3.0 do not replace the fog stage.
3608 * This implements linear fog computation and blending.
3609 * TODO: non linear fog
3610 * NOTE: gl_Fog.start and gl_Fog.end don't hold fog start s and end e but
3611 * -1/(e-s) and e/(e-s) respectively.
3613 if(This->baseShader.hex_version < WINED3DPS_VERSION(3,0)) {
3614 shader_addline(buffer, "float Fog = clamp(gl_FogFragCoord * gl_Fog.start + gl_Fog.end, 0.0, 1.0);\n");
3615 shader_addline(buffer, "%s.xyz = mix(gl_Fog.color.xyz, %s.xyz, Fog);\n", fragcolor, fragcolor);
3618 shader_addline(buffer, "}\n");
3620 TRACE("Compiling shader object %u\n", shader_obj);
3621 GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
3622 GL_EXTCALL(glCompileShaderARB(shader_obj));
3623 print_glsl_info_log(&GLINFO_LOCATION, shader_obj);
3625 /* Store the shader object */
3626 This->baseShader.prgId = shader_obj;
3629 static void shader_glsl_generate_vshader(IWineD3DVertexShader *iface, SHADER_BUFFER *buffer) {
3630 IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
3631 shader_reg_maps* reg_maps = &This->baseShader.reg_maps;
3632 CONST DWORD *function = This->baseShader.function;
3633 WineD3D_GL_Info *gl_info = &((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info;
3635 /* Create the hw GLSL shader program and assign it as the baseShader.prgId */
3636 GLhandleARB shader_obj = GL_EXTCALL(glCreateShaderObjectARB(GL_VERTEX_SHADER_ARB));
3638 shader_addline(buffer, "#version 120\n");
3640 /* Base Declarations */
3641 shader_generate_glsl_declarations( (IWineD3DBaseShader*) This, reg_maps, buffer, &GLINFO_LOCATION);
3643 /* Base Shader Body */
3644 shader_generate_main( (IWineD3DBaseShader*) This, buffer, reg_maps, function);
3646 /* Unpack 3.0 outputs */
3647 if (This->baseShader.hex_version >= WINED3DVS_VERSION(3,0)) {
3648 shader_addline(buffer, "order_ps_input(OUT);\n");
3649 } else {
3650 shader_addline(buffer, "order_ps_input();\n");
3653 /* If this shader doesn't use fog copy the z coord to the fog coord so that we can use table fog */
3654 if (!reg_maps->fog)
3655 shader_addline(buffer, "gl_FogFragCoord = gl_Position.z;\n");
3657 /* Write the final position.
3659 * OpenGL coordinates specify the center of the pixel while d3d coords specify
3660 * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
3661 * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
3662 * contains 1.0 to allow a mad.
3664 shader_addline(buffer, "gl_Position.y = gl_Position.y * posFixup.y;\n");
3665 shader_addline(buffer, "gl_Position.xy += posFixup.zw * gl_Position.ww;\n");
3667 /* Z coord [0;1]->[-1;1] mapping, see comment in transform_projection in state.c
3669 * Basically we want (in homogeneous coordinates) z = z * 2 - 1. However, shaders are run
3670 * before the homogeneous divide, so we have to take the w into account: z = ((z / w) * 2 - 1) * w,
3671 * which is the same as z = z / 2 - w.
3673 shader_addline(buffer, "gl_Position.z = gl_Position.z * 2.0 - gl_Position.w;\n");
3675 shader_addline(buffer, "}\n");
3677 TRACE("Compiling shader object %u\n", shader_obj);
3678 GL_EXTCALL(glShaderSourceARB(shader_obj, 1, (const char**)&buffer->buffer, NULL));
3679 GL_EXTCALL(glCompileShaderARB(shader_obj));
3680 print_glsl_info_log(&GLINFO_LOCATION, shader_obj);
3682 /* Store the shader object */
3683 This->baseShader.prgId = shader_obj;
3686 static void shader_glsl_get_caps(WINED3DDEVTYPE devtype, WineD3D_GL_Info *gl_info, struct shader_caps *pCaps) {
3687 /* Nvidia Geforce6/7 or Ati R4xx/R5xx cards with GLSL support, support VS 3.0 but older Nvidia/Ati
3688 * models with GLSL support only support 2.0. In case of nvidia we can detect VS 2.0 support using
3689 * vs_nv_version which is based on NV_vertex_program.
3690 * For Ati cards there's no way using glsl (it abstracts the lowlevel info away) and also not
3691 * using ARB_vertex_program. It is safe to assume that when a card supports pixel shader 2.0 it
3692 * supports vertex shader 2.0 too and the way around. We can detect ps2.0 using the maximum number
3693 * of native instructions, so use that here. For more info see the pixel shader versioning code below.
3695 if((GLINFO_LOCATION.vs_nv_version == VS_VERSION_20) || (GLINFO_LOCATION.ps_arb_max_instructions <= 512))
3696 pCaps->VertexShaderVersion = WINED3DVS_VERSION(2,0);
3697 else
3698 pCaps->VertexShaderVersion = WINED3DVS_VERSION(3,0);
3699 TRACE_(d3d_caps)("Hardware vertex shader version %d.%d enabled (GLSL)\n", (pCaps->VertexShaderVersion >> 8) & 0xff, pCaps->VertexShaderVersion & 0xff);
3700 pCaps->MaxVertexShaderConst = GL_LIMITS(vshader_constantsF);
3702 /* Older DX9-class videocards (GeforceFX / Radeon >9500/X*00) only support pixel shader 2.0/2.0a/2.0b.
3703 * In OpenGL the extensions related to GLSL abstract lowlevel GL info away which is needed
3704 * to distinguish between 2.0 and 3.0 (and 2.0a/2.0b). In case of Nvidia we use their fragment
3705 * program extensions. On other hardware including ATI GL_ARB_fragment_program offers the info
3706 * in max native instructions. Intel and others also offer the info in this extension but they
3707 * don't support GLSL (at least on Windows).
3709 * PS2.0 requires at least 96 instructions, 2.0a/2.0b go up to 512. Assume that if the number
3710 * of instructions is 512 or less we have to do with ps2.0 hardware.
3711 * 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.
3713 if((GLINFO_LOCATION.ps_nv_version == PS_VERSION_20) || (GLINFO_LOCATION.ps_arb_max_instructions <= 512))
3714 pCaps->PixelShaderVersion = WINED3DPS_VERSION(2,0);
3715 else
3716 pCaps->PixelShaderVersion = WINED3DPS_VERSION(3,0);
3718 /* FIXME: The following line is card dependent. -8.0 to 8.0 is the
3719 * Direct3D minimum requirement.
3721 * Both GL_ARB_fragment_program and GLSL require a "maximum representable magnitude"
3722 * of colors to be 2^10, and 2^32 for other floats. Should we use 1024 here?
3724 * The problem is that the refrast clamps temporary results in the shader to
3725 * [-MaxValue;+MaxValue]. If the card's max value is bigger than the one we advertize here,
3726 * then applications may miss the clamping behavior. On the other hand, if it is smaller,
3727 * the shader will generate incorrect results too. Unfortunately, GL deliberately doesn't
3728 * offer a way to query this.
3730 pCaps->PixelShader1xMaxValue = 8.0;
3731 TRACE_(d3d_caps)("Hardware pixel shader version %d.%d enabled (GLSL)\n", (pCaps->PixelShaderVersion >> 8) & 0xff, pCaps->PixelShaderVersion & 0xff);
3734 const shader_backend_t glsl_shader_backend = {
3735 shader_glsl_select,
3736 shader_glsl_select_depth_blt,
3737 shader_glsl_deselect_depth_blt,
3738 shader_glsl_load_constants,
3739 shader_glsl_cleanup,
3740 shader_glsl_color_correction,
3741 shader_glsl_destroy,
3742 shader_glsl_alloc,
3743 shader_glsl_free,
3744 shader_glsl_dirty_const,
3745 shader_glsl_generate_pshader,
3746 shader_glsl_generate_vshader,
3747 shader_glsl_get_caps,