1 /**************************************************************************
3 * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * 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, sub license, 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 portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
29 #include "main/glheader.h"
30 #include "main/context.h"
31 #include "main/condrender.h"
32 #include "main/samplerobj.h"
33 #include "main/state.h"
34 #include "main/enums.h"
36 #include "vbo/vbo_context.h"
37 #include "swrast/swrast.h"
38 #include "swrast_setup/swrast_setup.h"
41 #include "brw_defines.h"
42 #include "brw_context.h"
43 #include "brw_state.h"
45 #include "intel_batchbuffer.h"
47 #define FILE_DEBUG_FLAG DEBUG_PRIMS
49 static GLuint prim_to_hw_prim[GL_POLYGON+1] = {
63 static const GLenum reduced_prim[GL_POLYGON+1] = {
77 /* When the primitive changes, set a state bit and re-validate. Not
78 * the nicest and would rather deal with this by having all the
79 * programs be immune to the active primitive (ie. cope with all
80 * possibilities). That may not be realistic however.
82 static GLuint brw_set_prim(struct brw_context *brw,
83 const struct _mesa_prim *prim)
85 struct gl_context *ctx = &brw->intel.ctx;
86 GLenum mode = prim->mode;
88 DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim->mode));
90 /* Slight optimization to avoid the GS program when not needed:
92 if (mode == GL_QUAD_STRIP &&
93 ctx->Light.ShadeModel != GL_FLAT &&
94 ctx->Polygon.FrontMode == GL_FILL &&
95 ctx->Polygon.BackMode == GL_FILL)
96 mode = GL_TRIANGLE_STRIP;
98 if (prim->mode == GL_QUADS && prim->count == 4 &&
99 ctx->Light.ShadeModel != GL_FLAT &&
100 ctx->Polygon.FrontMode == GL_FILL &&
101 ctx->Polygon.BackMode == GL_FILL) {
102 mode = GL_TRIANGLE_FAN;
105 if (mode != brw->primitive) {
106 brw->primitive = mode;
107 brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
109 if (reduced_prim[mode] != brw->intel.reduced_primitive) {
110 brw->intel.reduced_primitive = reduced_prim[mode];
111 brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
115 return prim_to_hw_prim[mode];
118 static GLuint gen6_set_prim(struct brw_context *brw,
119 const struct _mesa_prim *prim)
121 DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim->mode));
123 if (prim->mode != brw->primitive) {
124 brw->primitive = prim->mode;
125 brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
128 return prim_to_hw_prim[mode];
132 static GLuint trim(GLenum prim, GLuint length)
134 if (prim == GL_QUAD_STRIP)
135 return length > 3 ? (length - length % 2) : 0;
136 else if (prim == GL_QUADS)
137 return length - length % 4;
143 static void brw_emit_prim(struct brw_context *brw,
144 const struct _mesa_prim *prim,
147 struct intel_context *intel = &brw->intel;
148 int verts_per_instance;
149 int vertex_access_type;
150 int start_vertex_location;
151 int base_vertex_location;
153 DBG("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
154 prim->start, prim->count);
156 start_vertex_location = prim->start;
157 base_vertex_location = prim->basevertex;
159 vertex_access_type = GEN4_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM;
160 start_vertex_location += brw->ib.start_vertex_offset;
161 base_vertex_location += brw->vb.start_vertex_bias;
163 vertex_access_type = GEN4_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL;
164 start_vertex_location += brw->vb.start_vertex_bias;
167 verts_per_instance = trim(prim->mode, prim->count);
169 /* If nothing to emit, just return. */
170 if (verts_per_instance == 0)
173 /* If we're set to always flush, do it before and after the primitive emit.
174 * We want to catch both missed flushes that hurt instruction/state cache
175 * and missed flushes of the render cache as it heads to other parts of
176 * the besides the draw code.
178 if (intel->always_flush_cache) {
179 intel_batchbuffer_emit_mi_flush(intel);
183 OUT_BATCH(CMD_3D_PRIM << 16 | (6 - 2) |
184 hw_prim << GEN4_3DPRIM_TOPOLOGY_TYPE_SHIFT |
186 OUT_BATCH(verts_per_instance);
187 OUT_BATCH(start_vertex_location);
188 OUT_BATCH(1); // instance count
189 OUT_BATCH(0); // start instance location
190 OUT_BATCH(base_vertex_location);
193 intel->batch.need_workaround_flush = true;
195 if (intel->always_flush_cache) {
196 intel_batchbuffer_emit_mi_flush(intel);
200 static void gen7_emit_prim(struct brw_context *brw,
201 const struct _mesa_prim *prim,
204 struct intel_context *intel = &brw->intel;
205 int verts_per_instance;
206 int vertex_access_type;
207 int start_vertex_location;
208 int base_vertex_location;
210 DBG("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
211 prim->start, prim->count);
213 start_vertex_location = prim->start;
214 base_vertex_location = prim->basevertex;
216 vertex_access_type = GEN7_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM;
217 start_vertex_location += brw->ib.start_vertex_offset;
218 base_vertex_location += brw->vb.start_vertex_bias;
220 vertex_access_type = GEN7_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL;
221 start_vertex_location += brw->vb.start_vertex_bias;
224 verts_per_instance = trim(prim->mode, prim->count);
226 /* If nothing to emit, just return. */
227 if (verts_per_instance == 0)
230 /* If we're set to always flush, do it before and after the primitive emit.
231 * We want to catch both missed flushes that hurt instruction/state cache
232 * and missed flushes of the render cache as it heads to other parts of
233 * the besides the draw code.
235 if (intel->always_flush_cache) {
236 intel_batchbuffer_emit_mi_flush(intel);
240 OUT_BATCH(CMD_3D_PRIM << 16 | (7 - 2));
241 OUT_BATCH(hw_prim | vertex_access_type);
242 OUT_BATCH(verts_per_instance);
243 OUT_BATCH(start_vertex_location);
244 OUT_BATCH(1); // instance count
245 OUT_BATCH(0); // start instance location
246 OUT_BATCH(base_vertex_location);
249 if (intel->always_flush_cache) {
250 intel_batchbuffer_emit_mi_flush(intel);
255 static void brw_merge_inputs( struct brw_context *brw,
256 const struct gl_client_array *arrays[])
258 struct brw_vertex_info old = brw->vb.info;
261 for (i = 0; i < brw->vb.nr_buffers; i++) {
262 drm_intel_bo_unreference(brw->vb.buffers[i].bo);
263 brw->vb.buffers[i].bo = NULL;
265 brw->vb.nr_buffers = 0;
267 memset(&brw->vb.info, 0, sizeof(brw->vb.info));
269 for (i = 0; i < VERT_ATTRIB_MAX; i++) {
270 brw->vb.inputs[i].buffer = -1;
271 brw->vb.inputs[i].glarray = arrays[i];
272 brw->vb.inputs[i].attrib = (gl_vert_attrib) i;
274 if (arrays[i]->StrideB != 0)
275 brw->vb.info.sizes[i/16] |= (brw->vb.inputs[i].glarray->Size - 1) <<
279 /* Raise statechanges if input sizes have changed. */
280 if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
281 brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
284 /* May fail if out of video memory for texture or vbo upload, or on
285 * fallback conditions.
287 static GLboolean brw_try_draw_prims( struct gl_context *ctx,
288 const struct gl_client_array *arrays[],
289 const struct _mesa_prim *prim,
291 const struct _mesa_index_buffer *ib,
295 struct intel_context *intel = intel_context(ctx);
296 struct brw_context *brw = brw_context(ctx);
297 GLboolean retval = GL_FALSE;
298 GLboolean warn = GL_FALSE;
302 _mesa_update_state( ctx );
304 /* We have to validate the textures *before* checking for fallbacks;
305 * otherwise, the software fallback won't be able to rely on the
306 * texture state, the firstLevel and lastLevel fields won't be
307 * set in the intel texture object (they'll both be 0), and the
308 * software fallback will segfault if it attempts to access any
309 * texture level other than level 0.
311 brw_validate_textures( brw );
313 /* Bind all inputs, derive varying and size information:
315 brw_merge_inputs( brw, arrays );
318 brw->state.dirty.brw |= BRW_NEW_INDICES;
320 brw->vb.min_index = min_index;
321 brw->vb.max_index = max_index;
322 brw->state.dirty.brw |= BRW_NEW_VERTICES;
324 /* Have to validate state quite late. Will rebuild tnl_program,
325 * which depends on varying information.
327 * Note this is where brw->vs->prog_data.inputs_read is calculated,
328 * so can't access it earlier.
331 intel_prepare_render(intel);
333 for (i = 0; i < nr_prims; i++) {
335 int estimated_max_prim_size;
337 estimated_max_prim_size = 512; /* batchbuffer commands */
338 estimated_max_prim_size += (BRW_MAX_TEX_UNIT *
339 (sizeof(struct brw_sampler_state) +
340 sizeof(struct gen5_sampler_default_color)));
341 estimated_max_prim_size += 1024; /* gen6 VS push constants */
342 estimated_max_prim_size += 1024; /* gen6 WM push constants */
343 estimated_max_prim_size += 512; /* misc. pad */
345 /* Flush the batch if it's approaching full, so that we don't wrap while
346 * we've got validated state that needs to be in the same batch as the
349 intel_batchbuffer_require_space(intel, estimated_max_prim_size, false);
352 hw_prim = brw_set_prim(brw, &prim[i]);
354 hw_prim = gen6_set_prim(brw, &prim[i]);
356 if (brw->state.dirty.brw) {
357 brw_validate_state(brw);
359 /* Various fallback checks: */
360 if (brw->intel.Fallback)
363 /* Check that we can fit our state in with our existing batchbuffer, or
366 if (dri_bufmgr_check_aperture_space(brw->state.validated_bos,
367 brw->state.validated_bo_count)) {
368 static GLboolean warned;
369 intel_batchbuffer_flush(intel);
371 /* Validate the state after we flushed the batch (which would have
372 * changed the set of dirty state). If we still fail to
373 * check_aperture, warn of what's happening, but attempt to continue
374 * on since it may succeed anyway, and the user would probably rather
375 * see a failure and a warning than a fallback.
377 brw_validate_state(brw);
379 dri_bufmgr_check_aperture_space(brw->state.validated_bos,
380 brw->state.validated_bo_count)) {
386 intel->no_batch_wrap = GL_TRUE;
387 brw_upload_state(brw);
391 gen7_emit_prim(brw, &prim[i], hw_prim);
393 brw_emit_prim(brw, &prim[i], hw_prim);
395 intel->no_batch_wrap = GL_FALSE;
400 if (intel->always_flush_batch)
401 intel_batchbuffer_flush(intel);
404 brw_state_cache_check_size(brw);
407 fprintf(stderr, "i965: Single primitive emit potentially exceeded "
408 "available aperture space\n");
411 DBG("%s failed\n", __FUNCTION__);
416 void brw_draw_prims( struct gl_context *ctx,
417 const struct gl_client_array *arrays[],
418 const struct _mesa_prim *prim,
420 const struct _mesa_index_buffer *ib,
421 GLboolean index_bounds_valid,
427 if (!_mesa_check_conditional_render(ctx))
430 if (!vbo_all_varyings_in_vbos(arrays)) {
431 if (!index_bounds_valid)
432 vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index);
434 /* Decide if we want to rebase. If so we end up recursing once
435 * only into this function.
437 if (min_index != 0 && !vbo_any_varyings_in_vbos(arrays)) {
438 vbo_rebase_prims(ctx, arrays,
440 ib, min_index, max_index,
446 /* Make a first attempt at drawing:
448 retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
450 /* Otherwise, we really are out of memory. Pass the drawing
451 * command to the software tnl module and which will in turn call
452 * swrast to do the drawing.
455 _swsetup_Wakeup(ctx);
457 _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
462 void brw_draw_init( struct brw_context *brw )
464 struct gl_context *ctx = &brw->intel.ctx;
465 struct vbo_context *vbo = vbo_context(ctx);
468 /* Register our drawing function:
470 vbo->draw_prims = brw_draw_prims;
472 for (i = 0; i < VERT_ATTRIB_MAX; i++)
473 brw->vb.inputs[i].buffer = -1;
474 brw->vb.nr_buffers = 0;
475 brw->vb.nr_enabled = 0;
478 void brw_draw_destroy( struct brw_context *brw )
482 for (i = 0; i < brw->vb.nr_buffers; i++) {
483 drm_intel_bo_unreference(brw->vb.buffers[i].bo);
484 brw->vb.buffers[i].bo = NULL;
486 brw->vb.nr_buffers = 0;
488 for (i = 0; i < brw->vb.nr_enabled; i++) {
489 brw->vb.enabled[i]->buffer = -1;
491 brw->vb.nr_enabled = 0;
493 drm_intel_bo_unreference(brw->ib.bo);