1 | /*
|
---|
2 | * shaders implementation
|
---|
3 | *
|
---|
4 | * Copyright 2002-2003 Jason Edmeades
|
---|
5 | * Copyright 2002-2003 Raphael Junqueira
|
---|
6 | * Copyright 2004 Christian Costa
|
---|
7 | * Copyright 2005 Oliver Stieber
|
---|
8 | * Copyright 2006 Ivan Gyurdiev
|
---|
9 | * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
|
---|
10 | *
|
---|
11 | * This library is free software; you can redistribute it and/or
|
---|
12 | * modify it under the terms of the GNU Lesser General Public
|
---|
13 | * License as published by the Free Software Foundation; either
|
---|
14 | * version 2.1 of the License, or (at your option) any later version.
|
---|
15 | *
|
---|
16 | * This library is distributed in the hope that it will be useful,
|
---|
17 | * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
18 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
---|
19 | * Lesser General Public License for more details.
|
---|
20 | *
|
---|
21 | * You should have received a copy of the GNU Lesser General Public
|
---|
22 | * License along with this library; if not, write to the Free Software
|
---|
23 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
|
---|
24 | */
|
---|
25 |
|
---|
26 | /*
|
---|
27 | * Sun LGPL Disclaimer: For the avoidance of doubt, except that if any license choice
|
---|
28 | * other than GPL or LGPL is available it will apply instead, Sun elects to use only
|
---|
29 | * the Lesser General Public License version 2.1 (LGPLv2) at this time for any software where
|
---|
30 | * a choice of LGPL license versions is made available with the language indicating
|
---|
31 | * that LGPLv2 or any later version may be used, or where a choice of which version
|
---|
32 | * of the LGPL is applied is otherwise unspecified.
|
---|
33 | */
|
---|
34 |
|
---|
35 | #include "config.h"
|
---|
36 |
|
---|
37 | #include <math.h>
|
---|
38 | #include <stdio.h>
|
---|
39 |
|
---|
40 | #include "wined3d_private.h"
|
---|
41 |
|
---|
42 | WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
|
---|
43 |
|
---|
44 | #define GLINFO_LOCATION ((IWineD3DDeviceImpl *)This->baseShader.device)->adapter->gl_info
|
---|
45 |
|
---|
46 | static void vshader_set_limits(IWineD3DVertexShaderImpl *This)
|
---|
47 | {
|
---|
48 | DWORD shader_version = WINED3D_SHADER_VERSION(This->baseShader.reg_maps.shader_version.major,
|
---|
49 | This->baseShader.reg_maps.shader_version.minor);
|
---|
50 |
|
---|
51 | This->baseShader.limits.texcoord = 0;
|
---|
52 | This->baseShader.limits.attributes = 16;
|
---|
53 | This->baseShader.limits.packed_input = 0;
|
---|
54 |
|
---|
55 | switch (shader_version)
|
---|
56 | {
|
---|
57 | case WINED3D_SHADER_VERSION(1,0):
|
---|
58 | case WINED3D_SHADER_VERSION(1,1):
|
---|
59 | This->baseShader.limits.temporary = 12;
|
---|
60 | This->baseShader.limits.constant_bool = 0;
|
---|
61 | This->baseShader.limits.constant_int = 0;
|
---|
62 | This->baseShader.limits.address = 1;
|
---|
63 | This->baseShader.limits.packed_output = 0;
|
---|
64 | This->baseShader.limits.sampler = 0;
|
---|
65 | This->baseShader.limits.label = 0;
|
---|
66 | /* TODO: vs_1_1 has a minimum of 96 constants. What happens if a vs_1_1 shader is used
|
---|
67 | * on a vs_3_0 capable card that has 256 constants? */
|
---|
68 | This->baseShader.limits.constant_float = min(256, GL_LIMITS(vshader_constantsF));
|
---|
69 | break;
|
---|
70 |
|
---|
71 | case WINED3D_SHADER_VERSION(2,0):
|
---|
72 | case WINED3D_SHADER_VERSION(2,1):
|
---|
73 | This->baseShader.limits.temporary = 12;
|
---|
74 | This->baseShader.limits.constant_bool = 16;
|
---|
75 | This->baseShader.limits.constant_int = 16;
|
---|
76 | This->baseShader.limits.address = 1;
|
---|
77 | This->baseShader.limits.packed_output = 0;
|
---|
78 | This->baseShader.limits.sampler = 0;
|
---|
79 | This->baseShader.limits.label = 16;
|
---|
80 | This->baseShader.limits.constant_float = min(256, GL_LIMITS(vshader_constantsF));
|
---|
81 | break;
|
---|
82 |
|
---|
83 | case WINED3D_SHADER_VERSION(4,0):
|
---|
84 | FIXME("Using 3.0 limits for 4.0 shader\n");
|
---|
85 | /* Fall through */
|
---|
86 |
|
---|
87 | case WINED3D_SHADER_VERSION(3,0):
|
---|
88 | This->baseShader.limits.temporary = 32;
|
---|
89 | This->baseShader.limits.constant_bool = 32;
|
---|
90 | This->baseShader.limits.constant_int = 32;
|
---|
91 | This->baseShader.limits.address = 1;
|
---|
92 | This->baseShader.limits.packed_output = 12;
|
---|
93 | This->baseShader.limits.sampler = 4;
|
---|
94 | This->baseShader.limits.label = 16; /* FIXME: 2048 */
|
---|
95 | /* DX10 cards on Windows advertise a d3d9 constant limit of 256 even though they are capable
|
---|
96 | * of supporting much more(GL drivers advertise 1024). d3d9.dll and d3d8.dll clamp the
|
---|
97 | * wined3d-advertised maximum. Clamp the constant limit for <= 3.0 shaders to 256.s
|
---|
98 | * use constant buffers */
|
---|
99 | This->baseShader.limits.constant_float = min(256, GL_LIMITS(vshader_constantsF));
|
---|
100 | break;
|
---|
101 |
|
---|
102 | default:
|
---|
103 | This->baseShader.limits.temporary = 12;
|
---|
104 | This->baseShader.limits.constant_bool = 16;
|
---|
105 | This->baseShader.limits.constant_int = 16;
|
---|
106 | This->baseShader.limits.address = 1;
|
---|
107 | This->baseShader.limits.packed_output = 0;
|
---|
108 | This->baseShader.limits.sampler = 0;
|
---|
109 | This->baseShader.limits.label = 16;
|
---|
110 | This->baseShader.limits.constant_float = min(256, GL_LIMITS(vshader_constantsF));
|
---|
111 | FIXME("Unrecognized vertex shader version %u.%u\n",
|
---|
112 | This->baseShader.reg_maps.shader_version.major,
|
---|
113 | This->baseShader.reg_maps.shader_version.minor);
|
---|
114 | }
|
---|
115 | }
|
---|
116 |
|
---|
117 | static BOOL match_usage(BYTE usage1, BYTE usage_idx1, BYTE usage2, BYTE usage_idx2) {
|
---|
118 | if (usage_idx1 != usage_idx2) return FALSE;
|
---|
119 | if (usage1 == usage2) return TRUE;
|
---|
120 | if (usage1 == WINED3DDECLUSAGE_POSITION && usage2 == WINED3DDECLUSAGE_POSITIONT) return TRUE;
|
---|
121 | if (usage2 == WINED3DDECLUSAGE_POSITION && usage1 == WINED3DDECLUSAGE_POSITIONT) return TRUE;
|
---|
122 |
|
---|
123 | return FALSE;
|
---|
124 | }
|
---|
125 |
|
---|
126 | BOOL vshader_get_input(IWineD3DVertexShader* iface, BYTE usage_req, BYTE usage_idx_req, unsigned int *regnum)
|
---|
127 | {
|
---|
128 | IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
|
---|
129 | WORD map = This->baseShader.reg_maps.input_registers;
|
---|
130 | unsigned int i;
|
---|
131 |
|
---|
132 | for (i = 0; map; map >>= 1, ++i)
|
---|
133 | {
|
---|
134 | if (!(map & 1)) continue;
|
---|
135 |
|
---|
136 | if (match_usage(This->attributes[i].usage,
|
---|
137 | This->attributes[i].usage_idx, usage_req, usage_idx_req))
|
---|
138 | {
|
---|
139 | *regnum = i;
|
---|
140 | return TRUE;
|
---|
141 | }
|
---|
142 | }
|
---|
143 | return FALSE;
|
---|
144 | }
|
---|
145 |
|
---|
146 | /* *******************************************
|
---|
147 | IWineD3DVertexShader IUnknown parts follow
|
---|
148 | ******************************************* */
|
---|
149 | static HRESULT WINAPI IWineD3DVertexShaderImpl_QueryInterface(IWineD3DVertexShader *iface, REFIID riid, LPVOID *ppobj) {
|
---|
150 | TRACE("iface %p, riid %s, ppobj %p\n", iface, debugstr_guid(riid), ppobj);
|
---|
151 |
|
---|
152 | if (IsEqualGUID(riid, &IID_IWineD3DVertexShader)
|
---|
153 | || IsEqualGUID(riid, &IID_IWineD3DBaseShader)
|
---|
154 | || IsEqualGUID(riid, &IID_IWineD3DBase)
|
---|
155 | || IsEqualGUID(riid, &IID_IUnknown))
|
---|
156 | {
|
---|
157 | IUnknown_AddRef(iface);
|
---|
158 | *ppobj = iface;
|
---|
159 | return S_OK;
|
---|
160 | }
|
---|
161 |
|
---|
162 | WARN("%s not implemented, returning E_NOINTERFACE\n", debugstr_guid(riid));
|
---|
163 |
|
---|
164 | *ppobj = NULL;
|
---|
165 | return E_NOINTERFACE;
|
---|
166 | }
|
---|
167 |
|
---|
168 | static ULONG WINAPI IWineD3DVertexShaderImpl_AddRef(IWineD3DVertexShader *iface) {
|
---|
169 | IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
|
---|
170 | ULONG refcount = InterlockedIncrement(&This->baseShader.ref);
|
---|
171 |
|
---|
172 | TRACE("%p increasing refcount to %u\n", This, refcount);
|
---|
173 |
|
---|
174 | return refcount;
|
---|
175 | }
|
---|
176 |
|
---|
177 | static ULONG WINAPI IWineD3DVertexShaderImpl_Release(IWineD3DVertexShader *iface) {
|
---|
178 | IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
|
---|
179 | ULONG refcount = InterlockedDecrement(&This->baseShader.ref);
|
---|
180 |
|
---|
181 | TRACE("%p decreasing refcount to %u\n", This, refcount);
|
---|
182 |
|
---|
183 | if (!refcount)
|
---|
184 | {
|
---|
185 | shader_cleanup((IWineD3DBaseShader *)iface);
|
---|
186 | HeapFree(GetProcessHeap(), 0, This);
|
---|
187 | }
|
---|
188 |
|
---|
189 | return refcount;
|
---|
190 | }
|
---|
191 |
|
---|
192 | /* *******************************************
|
---|
193 | IWineD3DVertexShader IWineD3DVertexShader parts follow
|
---|
194 | ******************************************* */
|
---|
195 |
|
---|
196 | static HRESULT WINAPI IWineD3DVertexShaderImpl_GetParent(IWineD3DVertexShader *iface, IUnknown** parent){
|
---|
197 | IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
|
---|
198 |
|
---|
199 | *parent = This->parent;
|
---|
200 | IUnknown_AddRef(*parent);
|
---|
201 | TRACE("(%p) : returning %p\n", This, *parent);
|
---|
202 | return WINED3D_OK;
|
---|
203 | }
|
---|
204 |
|
---|
205 | static HRESULT WINAPI IWineD3DVertexShaderImpl_GetDevice(IWineD3DVertexShader* iface, IWineD3DDevice **pDevice){
|
---|
206 | IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)iface;
|
---|
207 | IWineD3DDevice_AddRef(This->baseShader.device);
|
---|
208 | *pDevice = This->baseShader.device;
|
---|
209 | TRACE("(%p) returning %p\n", This, *pDevice);
|
---|
210 | return WINED3D_OK;
|
---|
211 | }
|
---|
212 |
|
---|
213 | static HRESULT WINAPI IWineD3DVertexShaderImpl_GetFunction(IWineD3DVertexShader* impl, VOID* pData, UINT* pSizeOfData) {
|
---|
214 | IWineD3DVertexShaderImpl *This = (IWineD3DVertexShaderImpl *)impl;
|
---|
215 | TRACE("(%p) : pData(%p), pSizeOfData(%p)\n", This, pData, pSizeOfData);
|
---|
216 |
|
---|
217 | if (NULL == pData) {
|
---|
218 | *pSizeOfData = This->baseShader.functionLength;
|
---|
219 | return WINED3D_OK;
|
---|
220 | }
|
---|
221 | if (*pSizeOfData < This->baseShader.functionLength) {
|
---|
222 | /* MSDN claims (for d3d8 at least) that if *pSizeOfData is smaller
|
---|
223 | * than the required size we should write the required size and
|
---|
224 | * return D3DERR_MOREDATA. That's not actually true. */
|
---|
225 | return WINED3DERR_INVALIDCALL;
|
---|
226 | }
|
---|
227 |
|
---|
228 | TRACE("(%p) : GetFunction copying to %p\n", This, pData);
|
---|
229 | memcpy(pData, This->baseShader.function, This->baseShader.functionLength);
|
---|
230 |
|
---|
231 | return WINED3D_OK;
|
---|
232 | }
|
---|
233 |
|
---|
234 | /* Note that for vertex shaders CompileShader isn't called until the
|
---|
235 | * shader is first used. The reason for this is that we need the vertex
|
---|
236 | * declaration the shader will be used with in order to determine if
|
---|
237 | * the data in a register is of type D3DCOLOR, and needs swizzling. */
|
---|
238 | static HRESULT WINAPI IWineD3DVertexShaderImpl_SetFunction(IWineD3DVertexShader *iface,
|
---|
239 | const DWORD *pFunction, const struct wined3d_shader_signature *output_signature)
|
---|
240 | {
|
---|
241 | IWineD3DVertexShaderImpl *This =(IWineD3DVertexShaderImpl *)iface;
|
---|
242 | IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *) This->baseShader.device;
|
---|
243 | const struct wined3d_shader_frontend *fe;
|
---|
244 | unsigned int i;
|
---|
245 | HRESULT hr;
|
---|
246 | shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
|
---|
247 |
|
---|
248 | TRACE("(%p) : pFunction %p\n", iface, pFunction);
|
---|
249 |
|
---|
250 | fe = shader_select_frontend(*pFunction);
|
---|
251 | if (!fe)
|
---|
252 | {
|
---|
253 | FIXME("Unable to find frontend for shader.\n");
|
---|
254 | return WINED3DERR_INVALIDCALL;
|
---|
255 | }
|
---|
256 | This->baseShader.frontend = fe;
|
---|
257 | This->baseShader.frontend_data = fe->shader_init(pFunction, output_signature);
|
---|
258 | if (!This->baseShader.frontend_data)
|
---|
259 | {
|
---|
260 | FIXME("Failed to initialize frontend.\n");
|
---|
261 | return WINED3DERR_INVALIDCALL;
|
---|
262 | }
|
---|
263 |
|
---|
264 | /* First pass: trace shader */
|
---|
265 | if (TRACE_ON(d3d_shader)) shader_trace_init(fe, This->baseShader.frontend_data, pFunction);
|
---|
266 |
|
---|
267 | /* Initialize immediate constant lists */
|
---|
268 | list_init(&This->baseShader.constantsF);
|
---|
269 | list_init(&This->baseShader.constantsB);
|
---|
270 | list_init(&This->baseShader.constantsI);
|
---|
271 |
|
---|
272 | /* Second pass: figure out registers used, semantics, etc.. */
|
---|
273 | This->min_rel_offset = GL_LIMITS(vshader_constantsF);
|
---|
274 | This->max_rel_offset = 0;
|
---|
275 | hr = shader_get_registers_used((IWineD3DBaseShader*) This, fe,
|
---|
276 | reg_maps, This->attributes, NULL, This->output_signature,
|
---|
277 | pFunction, GL_LIMITS(vshader_constantsF));
|
---|
278 | if (hr != WINED3D_OK) return hr;
|
---|
279 |
|
---|
280 | if (output_signature)
|
---|
281 | {
|
---|
282 | for (i = 0; i < output_signature->element_count; ++i)
|
---|
283 | {
|
---|
284 | struct wined3d_shader_signature_element *e = &output_signature->elements[i];
|
---|
285 | reg_maps->output_registers |= 1 << e->register_idx;
|
---|
286 | This->output_signature[e->register_idx] = *e;
|
---|
287 | }
|
---|
288 | }
|
---|
289 |
|
---|
290 | vshader_set_limits(This);
|
---|
291 |
|
---|
292 | if (deviceImpl->vs_selected_mode == SHADER_ARB
|
---|
293 | && ((GLINFO_LOCATION).quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT)
|
---|
294 | && This->min_rel_offset <= This->max_rel_offset)
|
---|
295 | {
|
---|
296 | if(This->max_rel_offset - This->min_rel_offset > 127) {
|
---|
297 | FIXME("The difference between the minimum and maximum relative offset is > 127\n");
|
---|
298 | FIXME("Which this OpenGL implementation does not support. Try using GLSL\n");
|
---|
299 | FIXME("Min: %d, Max: %d\n", This->min_rel_offset, This->max_rel_offset);
|
---|
300 | } else if(This->max_rel_offset - This->min_rel_offset > 63) {
|
---|
301 | This->rel_offset = This->min_rel_offset + 63;
|
---|
302 | } else if(This->max_rel_offset > 63) {
|
---|
303 | This->rel_offset = This->min_rel_offset;
|
---|
304 | } else {
|
---|
305 | This->rel_offset = 0;
|
---|
306 | }
|
---|
307 | }
|
---|
308 | This->baseShader.load_local_constsF = This->baseShader.reg_maps.usesrelconstF && !list_empty(&This->baseShader.constantsF);
|
---|
309 |
|
---|
310 | /* copy the function ... because it will certainly be released by application */
|
---|
311 | This->baseShader.function = HeapAlloc(GetProcessHeap(), 0, This->baseShader.functionLength);
|
---|
312 | if (!This->baseShader.function) return E_OUTOFMEMORY;
|
---|
313 | memcpy(This->baseShader.function, pFunction, This->baseShader.functionLength);
|
---|
314 |
|
---|
315 | return WINED3D_OK;
|
---|
316 | }
|
---|
317 |
|
---|
318 | /* Set local constants for d3d8 shaders */
|
---|
319 | static HRESULT WINAPI IWIneD3DVertexShaderImpl_SetLocalConstantsF(IWineD3DVertexShader *iface,
|
---|
320 | UINT start_idx, const float *src_data, UINT count) {
|
---|
321 | IWineD3DVertexShaderImpl *This =(IWineD3DVertexShaderImpl *)iface;
|
---|
322 | UINT i, end_idx;
|
---|
323 |
|
---|
324 | TRACE("(%p) : start_idx %u, src_data %p, count %u\n", This, start_idx, src_data, count);
|
---|
325 |
|
---|
326 | end_idx = start_idx + count;
|
---|
327 | if (end_idx > GL_LIMITS(vshader_constantsF)) {
|
---|
328 | WARN("end_idx %u > float constants limit %u\n", end_idx, GL_LIMITS(vshader_constantsF));
|
---|
329 | end_idx = GL_LIMITS(vshader_constantsF);
|
---|
330 | }
|
---|
331 |
|
---|
332 | for (i = start_idx; i < end_idx; ++i) {
|
---|
333 | local_constant* lconst = HeapAlloc(GetProcessHeap(), 0, sizeof(local_constant));
|
---|
334 | if (!lconst) return E_OUTOFMEMORY;
|
---|
335 |
|
---|
336 | lconst->idx = i;
|
---|
337 | memcpy(lconst->value, src_data + (i - start_idx) * 4 /* 4 components */, 4 * sizeof(float));
|
---|
338 | list_add_head(&This->baseShader.constantsF, &lconst->entry);
|
---|
339 | }
|
---|
340 |
|
---|
341 | return WINED3D_OK;
|
---|
342 | }
|
---|
343 |
|
---|
344 | const IWineD3DVertexShaderVtbl IWineD3DVertexShader_Vtbl =
|
---|
345 | {
|
---|
346 | /*** IUnknown methods ***/
|
---|
347 | IWineD3DVertexShaderImpl_QueryInterface,
|
---|
348 | IWineD3DVertexShaderImpl_AddRef,
|
---|
349 | IWineD3DVertexShaderImpl_Release,
|
---|
350 | /*** IWineD3DBase methods ***/
|
---|
351 | IWineD3DVertexShaderImpl_GetParent,
|
---|
352 | /*** IWineD3DBaseShader methods ***/
|
---|
353 | IWineD3DVertexShaderImpl_SetFunction,
|
---|
354 | /*** IWineD3DVertexShader methods ***/
|
---|
355 | IWineD3DVertexShaderImpl_GetDevice,
|
---|
356 | IWineD3DVertexShaderImpl_GetFunction,
|
---|
357 | IWIneD3DVertexShaderImpl_SetLocalConstantsF
|
---|
358 | };
|
---|
359 |
|
---|
360 | void find_vs_compile_args(IWineD3DVertexShaderImpl *shader, IWineD3DStateBlockImpl *stateblock, struct vs_compile_args *args) {
|
---|
361 | args->fog_src = stateblock->renderState[WINED3DRS_FOGTABLEMODE] == WINED3DFOG_NONE ? VS_FOG_COORD : VS_FOG_Z;
|
---|
362 | args->swizzle_map = ((IWineD3DDeviceImpl *)shader->baseShader.device)->strided_streams.swizzle_map;
|
---|
363 | }
|
---|