1 /*
2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
5
6 Permission is hereby granted, free of charge, to any person obtaining
7 a copy of this software and associated documentation files (the
8 "Software"), to deal in the Software without restriction, including
9 without limitation the rights to use, copy, modify, merge, publish,
10 distribute, sublicense, and/or sell copies of the Software, and to
11 permit persons to whom the Software is furnished to do so, subject to
12 the following conditions:
13
14 The above copyright notice and this permission notice (including the
15 next paragraph) shall be included in all copies or substantial
16 portions of the Software.
17
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25
26 **********************************************************************/
27 /*
28 * Authors:
29 * Keith Whitwell <keith@tungstengraphics.com>
30 */
31
32
33 #include "main/imports.h"
34 #include "main/macros.h"
35 #include "main/simple_list.h"
36
37 #include "vbo/vbo_context.h"
38
39 #include "brw_context.h"
40 #include "brw_defines.h"
41 #include "brw_draw.h"
42 #include "brw_state.h"
43
44 #include "intel_fbo.h"
45 #include "intel_mipmap_tree.h"
46 #include "intel_regions.h"
47 #include "intel_span.h"
48 #include "intel_tex.h"
49 #include "intel_tex_obj.h"
50
51 #include "tnl/t_pipeline.h"
52 #include "glsl/ralloc.h"
53
54 /***************************************
55 * Mesa's Driver Functions
56 ***************************************/
57
brwInitDriverFunctions(struct intel_screen * screen,struct dd_function_table * functions)58 static void brwInitDriverFunctions(struct intel_screen *screen,
59 struct dd_function_table *functions)
60 {
61 intelInitDriverFunctions( functions );
62
63 brwInitFragProgFuncs( functions );
64 brw_init_queryobj_functions(functions);
65
66 functions->BeginTransformFeedback = brw_begin_transform_feedback;
67
68 if (screen->gen >= 7)
69 functions->EndTransformFeedback = gen7_end_transform_feedback;
70 else
71 functions->EndTransformFeedback = brw_end_transform_feedback;
72 }
73
74 bool
brwCreateContext(int api,const struct gl_config * mesaVis,__DRIcontext * driContextPriv,unsigned major_version,unsigned minor_version,uint32_t flags,unsigned * error,void * sharedContextPrivate)75 brwCreateContext(int api,
76 const struct gl_config *mesaVis,
77 __DRIcontext *driContextPriv,
78 unsigned major_version,
79 unsigned minor_version,
80 uint32_t flags,
81 unsigned *error,
82 void *sharedContextPrivate)
83 {
84 __DRIscreen *sPriv = driContextPriv->driScreenPriv;
85 struct intel_screen *screen = sPriv->driverPrivate;
86 struct dd_function_table functions;
87 const unsigned req_version = major_version * 10 + minor_version;
88 unsigned max_supported_version = 0;
89 unsigned i;
90
91 #ifdef TEXTURE_FLOAT_ENABLED
92 bool has_texture_float = true;
93 #else
94 bool has_texture_float = false;
95 #endif
96
97 bool supports_gl30 = has_texture_float &&
98 (screen->gen == 6 ||
99 (screen->gen == 7 &&
100 screen->kernel_has_gen7_sol_reset));
101
102 /* Determine max_supported_version. */
103 switch (api) {
104 case API_OPENGL:
105 max_supported_version = supports_gl30 ? 30 : 21;
106 break;
107 case API_OPENGLES:
108 max_supported_version = 11;
109 break;
110 case API_OPENGLES2:
111 max_supported_version = 20;
112 break;
113 case API_OPENGL_CORE:
114 max_supported_version = supports_gl30 ? 31 : 0;
115 break;
116 default:
117 break;
118 }
119
120 if (max_supported_version == 0) {
121 *error = __DRI_CTX_ERROR_BAD_API;
122 return false;
123 } else if (req_version > max_supported_version) {
124 *error = __DRI_CTX_ERROR_BAD_VERSION;
125 return false;
126 }
127
128 struct brw_context *brw = rzalloc(NULL, struct brw_context);
129 if (!brw) {
130 printf("%s: failed to alloc context\n", __FUNCTION__);
131 *error = __DRI_CTX_ERROR_NO_MEMORY;
132 return false;
133 }
134
135 /* brwInitVtbl needs to know the chipset generation so that it can set the
136 * right pointers.
137 */
138 brw->intel.gen = screen->gen;
139
140 brwInitVtbl( brw );
141
142 brwInitDriverFunctions(screen, &functions);
143
144 struct intel_context *intel = &brw->intel;
145 struct gl_context *ctx = &intel->ctx;
146
147 if (!intelInitContext( intel, api, mesaVis, driContextPriv,
148 sharedContextPrivate, &functions )) {
149 printf("%s: failed to init intel context\n", __FUNCTION__);
150 *error = __DRI_CTX_ERROR_NO_MEMORY;
151 return false;
152 }
153
154 brw_init_surface_formats(brw);
155
156 /* Initialize swrast, tnl driver tables: */
157 intelInitSpanFuncs(ctx);
158
159 TNLcontext *tnl = TNL_CONTEXT(ctx);
160 if (tnl)
161 tnl->Driver.RunPipeline = _tnl_run_pipeline;
162
163 ctx->Const.MaxDualSourceDrawBuffers = 1;
164 ctx->Const.MaxDrawBuffers = BRW_MAX_DRAW_BUFFERS;
165 ctx->Const.MaxTextureImageUnits = BRW_MAX_TEX_UNIT;
166 ctx->Const.MaxTextureCoordUnits = 8; /* Mesa limit */
167 ctx->Const.MaxTextureUnits = MIN2(ctx->Const.MaxTextureCoordUnits,
168 ctx->Const.MaxTextureImageUnits);
169 ctx->Const.MaxVertexTextureImageUnits = BRW_MAX_TEX_UNIT;
170 ctx->Const.MaxCombinedTextureImageUnits =
171 ctx->Const.MaxVertexTextureImageUnits +
172 ctx->Const.MaxTextureImageUnits;
173
174 ctx->Const.MaxTextureLevels = 14; /* 8192 */
175 if (ctx->Const.MaxTextureLevels > MAX_TEXTURE_LEVELS)
176 ctx->Const.MaxTextureLevels = MAX_TEXTURE_LEVELS;
177 ctx->Const.Max3DTextureLevels = 9;
178 ctx->Const.MaxCubeTextureLevels = 12;
179
180 if (intel->gen >= 7)
181 ctx->Const.MaxArrayTextureLayers = 2048;
182 else
183 ctx->Const.MaxArrayTextureLayers = 512;
184
185 ctx->Const.MaxTextureRectSize = (1<<12);
186
187 ctx->Const.MaxTextureMaxAnisotropy = 16.0;
188
189 /* Hardware only supports a limited number of transform feedback buffers.
190 * So we need to override the Mesa default (which is based only on software
191 * limits).
192 */
193 ctx->Const.MaxTransformFeedbackBuffers = BRW_MAX_SOL_BUFFERS;
194
195 /* On Gen6, in the worst case, we use up one binding table entry per
196 * transform feedback component (see comments above the definition of
197 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value
198 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to
199 * BRW_MAX_SOL_BINDINGS.
200 *
201 * In "separate components" mode, we need to divide this value by
202 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries
203 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS.
204 */
205 ctx->Const.MaxTransformFeedbackInterleavedComponents = BRW_MAX_SOL_BINDINGS;
206 ctx->Const.MaxTransformFeedbackSeparateComponents =
207 BRW_MAX_SOL_BINDINGS / BRW_MAX_SOL_BUFFERS;
208
209 if (intel->gen == 6)
210 ctx->Const.MaxSamples = 4;
211 else if (intel->gen >= 7)
212 ctx->Const.MaxSamples = 8;
213
214 /* if conformance mode is set, swrast can handle any size AA point */
215 ctx->Const.MaxPointSizeAA = 255.0;
216
217 /* We want the GLSL compiler to emit code that uses condition codes */
218 for (i = 0; i <= MESA_SHADER_FRAGMENT; i++) {
219 ctx->ShaderCompilerOptions[i].MaxIfDepth = intel->gen < 6 ? 16 : UINT_MAX;
220 ctx->ShaderCompilerOptions[i].EmitCondCodes = true;
221 ctx->ShaderCompilerOptions[i].EmitNVTempInitialization = true;
222 ctx->ShaderCompilerOptions[i].EmitNoNoise = true;
223 ctx->ShaderCompilerOptions[i].EmitNoMainReturn = true;
224 ctx->ShaderCompilerOptions[i].EmitNoIndirectInput = true;
225 ctx->ShaderCompilerOptions[i].EmitNoIndirectOutput = true;
226
227 ctx->ShaderCompilerOptions[i].EmitNoIndirectUniform =
228 (i == MESA_SHADER_FRAGMENT);
229 ctx->ShaderCompilerOptions[i].EmitNoIndirectTemp =
230 (i == MESA_SHADER_FRAGMENT);
231 ctx->ShaderCompilerOptions[i].LowerClipDistance = true;
232 }
233
234 ctx->Const.VertexProgram.MaxNativeInstructions = (16 * 1024);
235 ctx->Const.VertexProgram.MaxAluInstructions = 0;
236 ctx->Const.VertexProgram.MaxTexInstructions = 0;
237 ctx->Const.VertexProgram.MaxTexIndirections = 0;
238 ctx->Const.VertexProgram.MaxNativeAluInstructions = 0;
239 ctx->Const.VertexProgram.MaxNativeTexInstructions = 0;
240 ctx->Const.VertexProgram.MaxNativeTexIndirections = 0;
241 ctx->Const.VertexProgram.MaxNativeAttribs = 16;
242 ctx->Const.VertexProgram.MaxNativeTemps = 256;
243 ctx->Const.VertexProgram.MaxNativeAddressRegs = 1;
244 ctx->Const.VertexProgram.MaxNativeParameters = 1024;
245 ctx->Const.VertexProgram.MaxEnvParams =
246 MIN2(ctx->Const.VertexProgram.MaxNativeParameters,
247 ctx->Const.VertexProgram.MaxEnvParams);
248
249 ctx->Const.FragmentProgram.MaxNativeInstructions = (16 * 1024);
250 ctx->Const.FragmentProgram.MaxNativeAluInstructions = (16 * 1024);
251 ctx->Const.FragmentProgram.MaxNativeTexInstructions = (16 * 1024);
252 ctx->Const.FragmentProgram.MaxNativeTexIndirections = (16 * 1024);
253 ctx->Const.FragmentProgram.MaxNativeAttribs = 12;
254 ctx->Const.FragmentProgram.MaxNativeTemps = 256;
255 ctx->Const.FragmentProgram.MaxNativeAddressRegs = 0;
256 ctx->Const.FragmentProgram.MaxNativeParameters = 1024;
257 ctx->Const.FragmentProgram.MaxEnvParams =
258 MIN2(ctx->Const.FragmentProgram.MaxNativeParameters,
259 ctx->Const.FragmentProgram.MaxEnvParams);
260
261 /* Fragment shaders use real, 32-bit twos-complement integers for all
262 * integer types.
263 */
264 ctx->Const.FragmentProgram.LowInt.RangeMin = 31;
265 ctx->Const.FragmentProgram.LowInt.RangeMax = 30;
266 ctx->Const.FragmentProgram.LowInt.Precision = 0;
267 ctx->Const.FragmentProgram.HighInt = ctx->Const.FragmentProgram.MediumInt
268 = ctx->Const.FragmentProgram.LowInt;
269
270 /* Gen6 converts quads to polygon in beginning of 3D pipeline,
271 but we're not sure how it's actually done for vertex order,
272 that affect provoking vertex decision. Always use last vertex
273 convention for quad primitive which works as expected for now. */
274 if (intel->gen >= 6)
275 ctx->Const.QuadsFollowProvokingVertexConvention = false;
276
277 ctx->Const.QueryCounterBits.Timestamp = 36;
278
279 if (intel->is_g4x || intel->gen >= 5) {
280 brw->CMD_VF_STATISTICS = GM45_3DSTATE_VF_STATISTICS;
281 brw->CMD_PIPELINE_SELECT = CMD_PIPELINE_SELECT_GM45;
282 brw->has_surface_tile_offset = true;
283 if (intel->gen < 6)
284 brw->has_compr4 = true;
285 brw->has_aa_line_parameters = true;
286 brw->has_pln = true;
287 } else {
288 brw->CMD_VF_STATISTICS = GEN4_3DSTATE_VF_STATISTICS;
289 brw->CMD_PIPELINE_SELECT = CMD_PIPELINE_SELECT_965;
290 }
291
292 /* WM maximum threads is number of EUs times number of threads per EU. */
293 assert(intel->gen <= 7);
294
295 if (intel->is_haswell) {
296 if (intel->gt == 1) {
297 brw->max_wm_threads = 102;
298 brw->max_vs_threads = 70;
299 brw->urb.size = 128;
300 brw->urb.max_vs_entries = 640;
301 brw->urb.max_gs_entries = 256;
302 } else if (intel->gt == 2) {
303 brw->max_wm_threads = 204;
304 brw->max_vs_threads = 280;
305 brw->urb.size = 256;
306 brw->urb.max_vs_entries = 1664;
307 brw->urb.max_gs_entries = 640;
308 }
309 } else if (intel->gen == 7) {
310 if (intel->gt == 1) {
311 brw->max_wm_threads = 48;
312 brw->max_vs_threads = 36;
313 brw->max_gs_threads = 36;
314 brw->urb.size = 128;
315 brw->urb.max_vs_entries = 512;
316 brw->urb.max_gs_entries = 192;
317 } else if (intel->gt == 2) {
318 brw->max_wm_threads = 172;
319 brw->max_vs_threads = 128;
320 brw->max_gs_threads = 128;
321 brw->urb.size = 256;
322 brw->urb.max_vs_entries = 704;
323 brw->urb.max_gs_entries = 320;
324 } else {
325 assert(!"Unknown gen7 device.");
326 }
327 } else if (intel->gen == 6) {
328 if (intel->gt == 2) {
329 brw->max_wm_threads = 80;
330 brw->max_vs_threads = 60;
331 brw->max_gs_threads = 60;
332 brw->urb.size = 64; /* volume 5c.5 section 5.1 */
333 brw->urb.max_vs_entries = 256; /* volume 2a (see 3DSTATE_URB) */
334 brw->urb.max_gs_entries = 256;
335 } else {
336 brw->max_wm_threads = 40;
337 brw->max_vs_threads = 24;
338 brw->max_gs_threads = 21; /* conservative; 24 if rendering disabled */
339 brw->urb.size = 32; /* volume 5c.5 section 5.1 */
340 brw->urb.max_vs_entries = 256; /* volume 2a (see 3DSTATE_URB) */
341 brw->urb.max_gs_entries = 256;
342 }
343 brw->urb.gen6_gs_previously_active = false;
344 } else if (intel->gen == 5) {
345 brw->urb.size = 1024;
346 brw->max_vs_threads = 72;
347 brw->max_gs_threads = 32;
348 brw->max_wm_threads = 12 * 6;
349 } else if (intel->is_g4x) {
350 brw->urb.size = 384;
351 brw->max_vs_threads = 32;
352 brw->max_gs_threads = 2;
353 brw->max_wm_threads = 10 * 5;
354 } else if (intel->gen < 6) {
355 brw->urb.size = 256;
356 brw->max_vs_threads = 16;
357 brw->max_gs_threads = 2;
358 brw->max_wm_threads = 8 * 4;
359 brw->has_negative_rhw_bug = true;
360 }
361
362 if (intel->gen <= 7) {
363 brw->needs_unlit_centroid_workaround = true;
364 }
365
366 brw->prim_restart.in_progress = false;
367 brw->prim_restart.enable_cut_index = false;
368 intel->hw_ctx = drm_intel_gem_context_create(intel->bufmgr);
369
370 brw_init_state( brw );
371
372 brw->curbe.last_buf = calloc(1, 4096);
373 brw->curbe.next_buf = calloc(1, 4096);
374
375 brw->state.dirty.mesa = ~0;
376 brw->state.dirty.brw = ~0;
377
378 brw->emit_state_always = 0;
379
380 intel->batch.need_workaround_flush = true;
381
382 ctx->VertexProgram._MaintainTnlProgram = true;
383 ctx->FragmentProgram._MaintainTexEnvProgram = true;
384
385 brw_draw_init( brw );
386
387 brw->precompile = driQueryOptionb(&intel->optionCache, "shader_precompile");
388
389 ctx->Const.NativeIntegers = true;
390 ctx->Const.UniformBooleanTrue = 1;
391
392 ctx->Const.ForceGLSLExtensionsWarn = driQueryOptionb(&intel->optionCache, "force_glsl_extensions_warn");
393
394 ctx->Const.ContextFlags = 0;
395 if ((flags & __DRI_CTX_FLAG_FORWARD_COMPATIBLE) != 0)
396 ctx->Const.ContextFlags |= GL_CONTEXT_FLAG_FORWARD_COMPATIBLE_BIT;
397
398 if ((flags & __DRI_CTX_FLAG_DEBUG) != 0)
399 ctx->Const.ContextFlags |= GL_CONTEXT_FLAG_DEBUG_BIT;
400
401 return true;
402 }
403
404