2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
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:
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.
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.
26 **********************************************************************/
29 * Keith Whitwell <keith@tungstengraphics.com>
33 #include "main/api_exec.h"
34 #include "main/imports.h"
35 #include "main/macros.h"
36 #include "main/simple_list.h"
37 #include "main/version.h"
38 #include "main/vtxfmt.h"
40 #include "vbo/vbo_context.h"
42 #include "brw_context.h"
43 #include "brw_defines.h"
45 #include "brw_state.h"
47 #include "intel_fbo.h"
48 #include "intel_mipmap_tree.h"
49 #include "intel_regions.h"
50 #include "intel_span.h"
51 #include "intel_tex.h"
52 #include "intel_tex_obj.h"
54 #include "tnl/t_pipeline.h"
55 #include "glsl/ralloc.h"
57 /***************************************
58 * Mesa's Driver Functions
59 ***************************************/
62 brw_query_samples_for_format(struct gl_context *ctx, GLenum internalFormat,
65 struct intel_context *intel = intel_context(ctx);
83 static void brwInitDriverFunctions(struct intel_screen *screen,
84 struct dd_function_table *functions)
86 intelInitDriverFunctions( functions );
88 brwInitFragProgFuncs( functions );
89 brw_init_queryobj_functions(functions);
91 functions->QuerySamplesForFormat = brw_query_samples_for_format;
92 functions->BeginTransformFeedback = brw_begin_transform_feedback;
95 functions->EndTransformFeedback = gen7_end_transform_feedback;
97 functions->EndTransformFeedback = brw_end_transform_feedback;
101 brwCreateContext(int api,
102 const struct gl_config *mesaVis,
103 __DRIcontext *driContextPriv,
104 unsigned major_version,
105 unsigned minor_version,
108 void *sharedContextPrivate)
110 __DRIscreen *sPriv = driContextPriv->driScreenPriv;
111 struct intel_screen *screen = sPriv->driverPrivate;
112 struct dd_function_table functions;
115 struct brw_context *brw = rzalloc(NULL, struct brw_context);
117 printf("%s: failed to alloc context\n", __FUNCTION__);
118 *error = __DRI_CTX_ERROR_NO_MEMORY;
122 /* brwInitVtbl needs to know the chipset generation so that it can set the
125 brw->intel.gen = screen->gen;
129 brwInitDriverFunctions(screen, &functions);
131 struct intel_context *intel = &brw->intel;
132 struct gl_context *ctx = &intel->ctx;
134 if (!intelInitContext( intel, api, major_version, minor_version,
135 mesaVis, driContextPriv,
136 sharedContextPrivate, &functions,
138 printf("%s: failed to init intel context\n", __FUNCTION__);
143 brw_init_surface_formats(brw);
145 /* Initialize swrast, tnl driver tables: */
146 intelInitSpanFuncs(ctx);
148 TNLcontext *tnl = TNL_CONTEXT(ctx);
150 tnl->Driver.RunPipeline = _tnl_run_pipeline;
152 ctx->Const.MaxDualSourceDrawBuffers = 1;
153 ctx->Const.MaxDrawBuffers = BRW_MAX_DRAW_BUFFERS;
154 ctx->Const.MaxTextureImageUnits = BRW_MAX_TEX_UNIT;
155 ctx->Const.MaxTextureCoordUnits = 8; /* Mesa limit */
156 ctx->Const.MaxTextureUnits = MIN2(ctx->Const.MaxTextureCoordUnits,
157 ctx->Const.MaxTextureImageUnits);
158 ctx->Const.MaxVertexTextureImageUnits = BRW_MAX_TEX_UNIT;
159 ctx->Const.MaxCombinedTextureImageUnits =
160 ctx->Const.MaxVertexTextureImageUnits +
161 ctx->Const.MaxTextureImageUnits;
163 ctx->Const.MaxTextureLevels = 14; /* 8192 */
164 if (ctx->Const.MaxTextureLevels > MAX_TEXTURE_LEVELS)
165 ctx->Const.MaxTextureLevels = MAX_TEXTURE_LEVELS;
166 ctx->Const.Max3DTextureLevels = 9;
167 ctx->Const.MaxCubeTextureLevels = 12;
170 ctx->Const.MaxArrayTextureLayers = 2048;
172 ctx->Const.MaxArrayTextureLayers = 512;
174 ctx->Const.MaxTextureRectSize = (1<<12);
176 ctx->Const.MaxTextureMaxAnisotropy = 16.0;
178 /* Hardware only supports a limited number of transform feedback buffers.
179 * So we need to override the Mesa default (which is based only on software
182 ctx->Const.MaxTransformFeedbackBuffers = BRW_MAX_SOL_BUFFERS;
184 /* On Gen6, in the worst case, we use up one binding table entry per
185 * transform feedback component (see comments above the definition of
186 * BRW_MAX_SOL_BINDINGS, in brw_context.h), so we need to advertise a value
187 * for MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS equal to
188 * BRW_MAX_SOL_BINDINGS.
190 * In "separate components" mode, we need to divide this value by
191 * BRW_MAX_SOL_BUFFERS, so that the total number of binding table entries
192 * used up by all buffers will not exceed BRW_MAX_SOL_BINDINGS.
194 ctx->Const.MaxTransformFeedbackInterleavedComponents = BRW_MAX_SOL_BINDINGS;
195 ctx->Const.MaxTransformFeedbackSeparateComponents =
196 BRW_MAX_SOL_BINDINGS / BRW_MAX_SOL_BUFFERS;
199 ctx->Const.MaxSamples = 4;
200 else if (intel->gen >= 7)
201 ctx->Const.MaxSamples = 8;
203 /* if conformance mode is set, swrast can handle any size AA point */
204 ctx->Const.MaxPointSizeAA = 255.0;
206 /* We want the GLSL compiler to emit code that uses condition codes */
207 for (i = 0; i <= MESA_SHADER_FRAGMENT; i++) {
208 ctx->ShaderCompilerOptions[i].MaxIfDepth = intel->gen < 6 ? 16 : UINT_MAX;
209 ctx->ShaderCompilerOptions[i].EmitCondCodes = true;
210 ctx->ShaderCompilerOptions[i].EmitNoNoise = true;
211 ctx->ShaderCompilerOptions[i].EmitNoMainReturn = true;
212 ctx->ShaderCompilerOptions[i].EmitNoIndirectInput = true;
213 ctx->ShaderCompilerOptions[i].EmitNoIndirectOutput = true;
215 ctx->ShaderCompilerOptions[i].EmitNoIndirectUniform =
216 (i == MESA_SHADER_FRAGMENT);
217 ctx->ShaderCompilerOptions[i].EmitNoIndirectTemp =
218 (i == MESA_SHADER_FRAGMENT);
219 ctx->ShaderCompilerOptions[i].LowerClipDistance = true;
222 ctx->Const.VertexProgram.MaxNativeInstructions = (16 * 1024);
223 ctx->Const.VertexProgram.MaxAluInstructions = 0;
224 ctx->Const.VertexProgram.MaxTexInstructions = 0;
225 ctx->Const.VertexProgram.MaxTexIndirections = 0;
226 ctx->Const.VertexProgram.MaxNativeAluInstructions = 0;
227 ctx->Const.VertexProgram.MaxNativeTexInstructions = 0;
228 ctx->Const.VertexProgram.MaxNativeTexIndirections = 0;
229 ctx->Const.VertexProgram.MaxNativeAttribs = 16;
230 ctx->Const.VertexProgram.MaxNativeTemps = 256;
231 ctx->Const.VertexProgram.MaxNativeAddressRegs = 1;
232 ctx->Const.VertexProgram.MaxNativeParameters = 1024;
233 ctx->Const.VertexProgram.MaxEnvParams =
234 MIN2(ctx->Const.VertexProgram.MaxNativeParameters,
235 ctx->Const.VertexProgram.MaxEnvParams);
237 ctx->Const.FragmentProgram.MaxNativeInstructions = (1 * 1024);
238 ctx->Const.FragmentProgram.MaxNativeAluInstructions = (1 * 1024);
239 ctx->Const.FragmentProgram.MaxNativeTexInstructions = (1 * 1024);
240 ctx->Const.FragmentProgram.MaxNativeTexIndirections = (1 * 1024);
241 ctx->Const.FragmentProgram.MaxNativeAttribs = 12;
242 ctx->Const.FragmentProgram.MaxNativeTemps = 256;
243 ctx->Const.FragmentProgram.MaxNativeAddressRegs = 0;
244 ctx->Const.FragmentProgram.MaxNativeParameters = 1024;
245 ctx->Const.FragmentProgram.MaxEnvParams =
246 MIN2(ctx->Const.FragmentProgram.MaxNativeParameters,
247 ctx->Const.FragmentProgram.MaxEnvParams);
249 /* Fragment shaders use real, 32-bit twos-complement integers for all
252 ctx->Const.FragmentProgram.LowInt.RangeMin = 31;
253 ctx->Const.FragmentProgram.LowInt.RangeMax = 30;
254 ctx->Const.FragmentProgram.LowInt.Precision = 0;
255 ctx->Const.FragmentProgram.HighInt = ctx->Const.FragmentProgram.MediumInt
256 = ctx->Const.FragmentProgram.LowInt;
258 /* Gen6 converts quads to polygon in beginning of 3D pipeline,
259 but we're not sure how it's actually done for vertex order,
260 that affect provoking vertex decision. Always use last vertex
261 convention for quad primitive which works as expected for now. */
263 ctx->Const.QuadsFollowProvokingVertexConvention = false;
265 ctx->Const.QueryCounterBits.Timestamp = 36;
267 if (intel->is_g4x || intel->gen >= 5) {
268 brw->CMD_VF_STATISTICS = GM45_3DSTATE_VF_STATISTICS;
269 brw->CMD_PIPELINE_SELECT = CMD_PIPELINE_SELECT_GM45;
270 brw->has_surface_tile_offset = true;
272 brw->has_compr4 = true;
273 brw->has_aa_line_parameters = true;
276 brw->CMD_VF_STATISTICS = GEN4_3DSTATE_VF_STATISTICS;
277 brw->CMD_PIPELINE_SELECT = CMD_PIPELINE_SELECT_965;
280 /* WM maximum threads is number of EUs times number of threads per EU. */
281 if (intel->gen >= 7) {
282 if (intel->gt == 1) {
283 brw->max_wm_threads = 48;
284 brw->max_vs_threads = 36;
285 brw->max_gs_threads = 36;
287 brw->urb.max_vs_entries = 512;
288 brw->urb.max_gs_entries = 192;
289 } else if (intel->gt == 2) {
290 brw->max_wm_threads = 172;
291 brw->max_vs_threads = 128;
292 brw->max_gs_threads = 128;
294 brw->urb.max_vs_entries = 704;
295 brw->urb.max_gs_entries = 320;
297 assert(!"Unknown gen7 device.");
299 } else if (intel->gen == 6) {
300 if (intel->gt == 2) {
301 brw->max_wm_threads = 80;
302 brw->max_vs_threads = 60;
303 brw->max_gs_threads = 60;
304 brw->urb.size = 64; /* volume 5c.5 section 5.1 */
305 brw->urb.max_vs_entries = 256; /* volume 2a (see 3DSTATE_URB) */
306 brw->urb.max_gs_entries = 256;
308 brw->max_wm_threads = 40;
309 brw->max_vs_threads = 24;
310 brw->max_gs_threads = 21; /* conservative; 24 if rendering disabled */
311 brw->urb.size = 32; /* volume 5c.5 section 5.1 */
312 brw->urb.max_vs_entries = 256; /* volume 2a (see 3DSTATE_URB) */
313 brw->urb.max_gs_entries = 256;
315 brw->urb.gen6_gs_previously_active = false;
316 } else if (intel->gen == 5) {
317 brw->urb.size = 1024;
318 brw->max_vs_threads = 72;
319 brw->max_gs_threads = 32;
320 brw->max_wm_threads = 12 * 6;
321 } else if (intel->is_g4x) {
323 brw->max_vs_threads = 32;
324 brw->max_gs_threads = 2;
325 brw->max_wm_threads = 10 * 5;
326 } else if (intel->gen < 6) {
328 brw->max_vs_threads = 16;
329 brw->max_gs_threads = 2;
330 brw->max_wm_threads = 8 * 4;
331 brw->has_negative_rhw_bug = true;
334 if (intel->gen <= 7) {
335 brw->needs_unlit_centroid_workaround = true;
338 brw->prim_restart.in_progress = false;
339 brw->prim_restart.enable_cut_index = false;
340 intel->hw_ctx = drm_intel_gem_context_create(intel->bufmgr);
342 brw_init_state( brw );
344 brw->curbe.last_buf = calloc(1, 4096);
345 brw->curbe.next_buf = calloc(1, 4096);
347 brw->state.dirty.mesa = ~0;
348 brw->state.dirty.brw = ~0;
350 brw->emit_state_always = 0;
352 intel->batch.need_workaround_flush = true;
354 ctx->VertexProgram._MaintainTnlProgram = true;
355 ctx->FragmentProgram._MaintainTexEnvProgram = true;
357 brw_draw_init( brw );
359 brw->precompile = driQueryOptionb(&intel->optionCache, "shader_precompile");
361 ctx->Const.NativeIntegers = true;
362 ctx->Const.UniformBooleanTrue = 1;
364 ctx->Const.ForceGLSLExtensionsWarn = driQueryOptionb(&intel->optionCache, "force_glsl_extensions_warn");
366 ctx->Const.DisableGLSLLineContinuations = driQueryOptionb(&intel->optionCache, "disable_glsl_line_continuations");
368 ctx->Const.ContextFlags = 0;
369 if ((flags & __DRI_CTX_FLAG_FORWARD_COMPATIBLE) != 0)
370 ctx->Const.ContextFlags |= GL_CONTEXT_FLAG_FORWARD_COMPATIBLE_BIT;
372 if ((flags & __DRI_CTX_FLAG_DEBUG) != 0)
373 ctx->Const.ContextFlags |= GL_CONTEXT_FLAG_DEBUG_BIT;
375 brw_fs_alloc_reg_sets(brw);
377 if (INTEL_DEBUG & DEBUG_SHADER_TIME)
378 brw_init_shader_time(brw);
380 _mesa_compute_version(ctx);
382 _mesa_initialize_dispatch_tables(ctx);
383 _mesa_initialize_vbo_vtxfmt(ctx);