2 * Copyright © 2011 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
26 #include "intel_batchbuffer.h"
27 #include "intel_fbo.h"
28 #include "intel_mipmap_tree.h"
30 #include "brw_context.h"
31 #include "brw_defines.h"
32 #include "brw_state.h"
34 #include "brw_blorp.h"
35 #include "gen6_blorp.h"
38 * \name Constants for BLORP VBO
41 #define GEN6_BLORP_NUM_VERTICES 3
42 #define GEN6_BLORP_NUM_VUE_ELEMS 8
43 #define GEN6_BLORP_VBO_SIZE (GEN6_BLORP_NUM_VERTICES \
44 * GEN6_BLORP_NUM_VUE_ELEMS \
49 gen6_blorp_emit_batch_head(struct brw_context *brw,
50 const brw_blorp_params *params)
52 struct gl_context *ctx = &brw->intel.ctx;
53 struct intel_context *intel = &brw->intel;
55 /* To ensure that the batch contains only the resolve, flush the batch
56 * before beginning and after finishing emitting the resolve packets.
58 * Ideally, we would not need to flush for the resolve op. But, I suspect
59 * that it's unsafe for CMD_PIPELINE_SELECT to occur multiple times in
60 * a single batch, and there is no safe way to ensure that other than by
61 * fencing the resolve with flushes. Ideally, we would just detect if
62 * a batch is in progress and do the right thing, but that would require
63 * the ability to *safely* access brw_context::state::dirty::brw
64 * outside of the brw_upload_state() codepath.
68 /* CMD_PIPELINE_SELECT
70 * Select the 3D pipeline, as opposed to the media pipeline.
74 OUT_BATCH(brw->CMD_PIPELINE_SELECT << 16);
81 * CMD_STATE_BASE_ADDRESS
83 * From the Sandy Bridge PRM, Volume 1, Part 1, Table STATE_BASE_ADDRESS:
84 * The following commands must be reissued following any change to the
87 * 3DSTATE_BINDING_TABLE_POINTERS
88 * 3DSTATE_SAMPLER_STATE_POINTERS
89 * 3DSTATE_VIEWPORT_STATE_POINTERS
90 * MEDIA_STATE_POINTERS
93 gen6_blorp_emit_state_base_address(struct brw_context *brw,
94 const brw_blorp_params *params)
96 struct intel_context *intel = &brw->intel;
99 OUT_BATCH(CMD_STATE_BASE_ADDRESS << 16 | (10 - 2));
100 OUT_BATCH(1); /* GeneralStateBaseAddressModifyEnable */
101 /* SurfaceStateBaseAddress */
102 OUT_RELOC(intel->batch.bo, I915_GEM_DOMAIN_SAMPLER, 0, 1);
103 /* DynamicStateBaseAddress */
104 OUT_RELOC(intel->batch.bo, (I915_GEM_DOMAIN_RENDER |
105 I915_GEM_DOMAIN_INSTRUCTION), 0, 1);
106 OUT_BATCH(1); /* IndirectObjectBaseAddress */
107 if (params->use_wm_prog) {
108 OUT_RELOC(brw->cache.bo, I915_GEM_DOMAIN_INSTRUCTION, 0,
109 1); /* Instruction base address: shader kernels */
111 OUT_BATCH(1); /* InstructionBaseAddress */
113 OUT_BATCH(1); /* GeneralStateUpperBound */
114 /* Dynamic state upper bound. Although the documentation says that
115 * programming it to zero will cause it to be ignored, that is a lie.
116 * If this isn't programmed to a real bound, the sampler border color
117 * pointer is rejected, causing border color to mysteriously fail.
119 OUT_BATCH(0xfffff001);
120 OUT_BATCH(1); /* IndirectObjectUpperBound*/
121 OUT_BATCH(1); /* InstructionAccessUpperBound */
127 gen6_blorp_emit_vertices(struct brw_context *brw,
128 const brw_blorp_params *params)
130 struct intel_context *intel = &brw->intel;
131 uint32_t vertex_offset;
133 /* Setup VBO for the rectangle primitive..
135 * A rectangle primitive (3DPRIM_RECTLIST) consists of only three
136 * vertices. The vertices reside in screen space with DirectX coordinates
137 * (that is, (0, 0) is the upper left corner).
144 * Since the VS is disabled, the clipper loads each VUE directly from
145 * the URB. This is controlled by the 3DSTATE_VERTEX_BUFFERS and
146 * 3DSTATE_VERTEX_ELEMENTS packets below. The VUE contents are as follows:
147 * dw0: Reserved, MBZ.
148 * dw1: Render Target Array Index. The HiZ op does not use indexed
149 * vertices, so set the dword to 0.
150 * dw2: Viewport Index. The HiZ op disables viewport mapping and
151 * scissoring, so set the dword to 0.
152 * dw3: Point Width: The HiZ op does not emit the POINTLIST primitive, so
153 * set the dword to 0.
154 * dw4: Vertex Position X.
155 * dw5: Vertex Position Y.
156 * dw6: Vertex Position Z.
157 * dw7: Vertex Position W.
159 * For details, see the Sandybridge PRM, Volume 2, Part 1, Section 1.5.1
160 * "Vertex URB Entry (VUE) Formats".
165 const float vertices[GEN6_BLORP_VBO_SIZE] = {
166 /* v0 */ 0, 0, 0, 0, (float) params->x0, (float) params->y1, 0, 1,
167 /* v1 */ 0, 0, 0, 0, (float) params->x1, (float) params->y1, 0, 1,
168 /* v2 */ 0, 0, 0, 0, (float) params->x0, (float) params->y0, 0, 1,
171 vertex_data = (float *) brw_state_batch(brw, AUB_TRACE_VERTEX_BUFFER,
172 GEN6_BLORP_VBO_SIZE, 32,
174 memcpy(vertex_data, vertices, GEN6_BLORP_VBO_SIZE);
177 /* 3DSTATE_VERTEX_BUFFERS */
179 const int num_buffers = 1;
180 const int batch_length = 1 + 4 * num_buffers;
182 uint32_t dw0 = GEN6_VB0_ACCESS_VERTEXDATA |
183 (GEN6_BLORP_NUM_VUE_ELEMS * sizeof(float)) << BRW_VB0_PITCH_SHIFT;
186 dw0 |= GEN7_VB0_ADDRESS_MODIFYENABLE;
188 BEGIN_BATCH(batch_length);
189 OUT_BATCH((_3DSTATE_VERTEX_BUFFERS << 16) | (batch_length - 2));
192 OUT_RELOC(intel->batch.bo, I915_GEM_DOMAIN_VERTEX, 0,
195 OUT_RELOC(intel->batch.bo, I915_GEM_DOMAIN_VERTEX, 0,
196 vertex_offset + GEN6_BLORP_VBO_SIZE - 1);
201 /* 3DSTATE_VERTEX_ELEMENTS
203 * Fetch dwords 0 - 7 from each VUE. See the comments above where
204 * the vertex_bo is filled with data.
207 const int num_elements = 2;
208 const int batch_length = 1 + 2 * num_elements;
210 BEGIN_BATCH(batch_length);
211 OUT_BATCH((_3DSTATE_VERTEX_ELEMENTS << 16) | (batch_length - 2));
213 OUT_BATCH(GEN6_VE0_VALID |
214 BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT |
215 0 << BRW_VE0_SRC_OFFSET_SHIFT);
216 OUT_BATCH(BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_0_SHIFT |
217 BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_1_SHIFT |
218 BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_2_SHIFT |
219 BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_3_SHIFT);
221 OUT_BATCH(GEN6_VE0_VALID |
222 BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT |
223 16 << BRW_VE0_SRC_OFFSET_SHIFT);
224 OUT_BATCH(BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_0_SHIFT |
225 BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_1_SHIFT |
226 BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_2_SHIFT |
227 BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_3_SHIFT);
235 * Assign the entire URB to the VS. Even though the VS disabled, URB space
236 * is still needed because the clipper loads the VUE's from the URB. From
237 * the Sandybridge PRM, Volume 2, Part 1, Section 3DSTATE,
238 * Dword 1.15:0 "VS Number of URB Entries":
239 * This field is always used (even if VS Function Enable is DISABLED).
241 * The warning below appears in the PRM (Section 3DSTATE_URB), but we can
242 * safely ignore it because this batch contains only one draw call.
243 * Because of URB corruption caused by allocating a previous GS unit
244 * URB entry to the VS unit, software is required to send a “GS NULL
245 * Fence” (Send URB fence with VS URB size == 1 and GS URB size == 0)
246 * plus a dummy DRAW call before any case where VS will be taking over
250 gen6_blorp_emit_urb_config(struct brw_context *brw,
251 const brw_blorp_params *params)
253 struct intel_context *intel = &brw->intel;
256 OUT_BATCH(_3DSTATE_URB << 16 | (3 - 2));
257 OUT_BATCH(brw->urb.max_vs_entries << GEN6_URB_VS_ENTRIES_SHIFT);
265 gen6_blorp_emit_blend_state(struct brw_context *brw,
266 const brw_blorp_params *params)
268 uint32_t cc_blend_state_offset;
270 struct gen6_blend_state *blend = (struct gen6_blend_state *)
271 brw_state_batch(brw, AUB_TRACE_BLEND_STATE,
272 sizeof(struct gen6_blend_state), 64,
273 &cc_blend_state_offset);
275 memset(blend, 0, sizeof(*blend));
277 blend->blend1.pre_blend_clamp_enable = 1;
278 blend->blend1.post_blend_clamp_enable = 1;
279 blend->blend1.clamp_range = BRW_RENDERTARGET_CLAMPRANGE_FORMAT;
281 blend->blend1.write_disable_r = false;
282 blend->blend1.write_disable_g = false;
283 blend->blend1.write_disable_b = false;
284 blend->blend1.write_disable_a = false;
286 return cc_blend_state_offset;
292 gen6_blorp_emit_cc_state(struct brw_context *brw,
293 const brw_blorp_params *params)
295 uint32_t cc_state_offset;
297 struct gen6_color_calc_state *cc = (struct gen6_color_calc_state *)
298 brw_state_batch(brw, AUB_TRACE_CC_STATE,
299 sizeof(gen6_color_calc_state), 64,
301 memset(cc, 0, sizeof(*cc));
303 return cc_state_offset;
308 * \param out_offset is relative to
309 * CMD_STATE_BASE_ADDRESS.DynamicStateBaseAddress.
312 gen6_blorp_emit_depth_stencil_state(struct brw_context *brw,
313 const brw_blorp_params *params)
315 uint32_t depthstencil_offset;
317 struct gen6_depth_stencil_state *state;
318 state = (struct gen6_depth_stencil_state *)
319 brw_state_batch(brw, AUB_TRACE_DEPTH_STENCIL_STATE,
321 &depthstencil_offset);
322 memset(state, 0, sizeof(*state));
324 /* See the following sections of the Sandy Bridge PRM, Volume 1, Part2:
325 * - 7.5.3.1 Depth Buffer Clear
326 * - 7.5.3.2 Depth Buffer Resolve
327 * - 7.5.3.3 Hierarchical Depth Buffer Resolve
329 state->ds2.depth_write_enable = 1;
330 if (params->hiz_op == GEN6_HIZ_OP_DEPTH_RESOLVE) {
331 state->ds2.depth_test_enable = 1;
332 state->ds2.depth_test_func = COMPAREFUNC_NEVER;
335 return depthstencil_offset;
339 /* 3DSTATE_CC_STATE_POINTERS
341 * The pointer offsets are relative to
342 * CMD_STATE_BASE_ADDRESS.DynamicStateBaseAddress.
344 * The HiZ op doesn't use BLEND_STATE or COLOR_CALC_STATE.
347 gen6_blorp_emit_cc_state_pointers(struct brw_context *brw,
348 const brw_blorp_params *params,
349 uint32_t cc_blend_state_offset,
350 uint32_t depthstencil_offset,
351 uint32_t cc_state_offset)
353 struct intel_context *intel = &brw->intel;
356 OUT_BATCH(_3DSTATE_CC_STATE_POINTERS << 16 | (4 - 2));
357 OUT_BATCH(cc_blend_state_offset | 1); /* BLEND_STATE offset */
358 OUT_BATCH(depthstencil_offset | 1); /* DEPTH_STENCIL_STATE offset */
359 OUT_BATCH(cc_state_offset | 1); /* COLOR_CALC_STATE offset */
364 /* WM push constants */
366 gen6_blorp_emit_wm_constants(struct brw_context *brw,
367 const brw_blorp_params *params)
369 uint32_t wm_push_const_offset;
371 void *constants = brw_state_batch(brw, AUB_TRACE_WM_CONSTANTS,
372 sizeof(params->wm_push_consts),
373 32, &wm_push_const_offset);
374 memcpy(constants, ¶ms->wm_push_consts,
375 sizeof(params->wm_push_consts));
377 return wm_push_const_offset;
381 /* SURFACE_STATE for renderbuffer or texture surface (see
382 * brw_update_renderbuffer_surface and brw_update_texture_surface)
385 gen6_blorp_emit_surface_state(struct brw_context *brw,
386 const brw_blorp_params *params,
387 const brw_blorp_surface_info *surface,
388 uint32_t read_domains, uint32_t write_domain)
390 uint32_t wm_surf_offset;
391 uint32_t width = surface->width;
392 uint32_t height = surface->height;
393 if (surface->num_samples > 1) {
394 /* Since gen6 uses INTEL_MSAA_LAYOUT_IMS, width and height are measured
395 * in samples. But SURFACE_STATE wants them in pixels, so we need to
396 * divide them each by 2.
401 struct intel_region *region = surface->mt->region;
402 uint32_t tile_x, tile_y;
404 uint32_t *surf = (uint32_t *)
405 brw_state_batch(brw, AUB_TRACE_SURFACE_STATE, 6 * 4, 32,
408 surf[0] = (BRW_SURFACE_2D << BRW_SURFACE_TYPE_SHIFT |
409 BRW_SURFACE_MIPMAPLAYOUT_BELOW << BRW_SURFACE_MIPLAYOUT_SHIFT |
410 BRW_SURFACE_CUBEFACE_ENABLES |
411 surface->brw_surfaceformat << BRW_SURFACE_FORMAT_SHIFT);
414 surf[1] = (surface->compute_tile_offsets(&tile_x, &tile_y) +
417 surf[2] = (0 << BRW_SURFACE_LOD_SHIFT |
418 (width - 1) << BRW_SURFACE_WIDTH_SHIFT |
419 (height - 1) << BRW_SURFACE_HEIGHT_SHIFT);
421 uint32_t tiling = surface->map_stencil_as_y_tiled
422 ? BRW_SURFACE_TILED | BRW_SURFACE_TILED_Y
423 : brw_get_surface_tiling_bits(region->tiling);
424 uint32_t pitch_bytes = region->pitch * region->cpp;
425 if (surface->map_stencil_as_y_tiled)
428 0 << BRW_SURFACE_DEPTH_SHIFT |
429 (pitch_bytes - 1) << BRW_SURFACE_PITCH_SHIFT);
431 surf[4] = brw_get_surface_num_multisamples(surface->num_samples);
433 /* Note that the low bits of these fields are missing, so
434 * there's the possibility of getting in trouble.
436 assert(tile_x % 4 == 0);
437 assert(tile_y % 2 == 0);
438 surf[5] = ((tile_x / 4) << BRW_SURFACE_X_OFFSET_SHIFT |
439 (tile_y / 2) << BRW_SURFACE_Y_OFFSET_SHIFT |
440 (surface->mt->align_h == 4 ?
441 BRW_SURFACE_VERTICAL_ALIGN_ENABLE : 0));
443 /* Emit relocation to surface contents */
444 drm_intel_bo_emit_reloc(brw->intel.batch.bo,
447 surf[1] - region->bo->offset,
448 read_domains, write_domain);
450 return wm_surf_offset;
454 /* BINDING_TABLE. See brw_wm_binding_table(). */
456 gen6_blorp_emit_binding_table(struct brw_context *brw,
457 const brw_blorp_params *params,
458 uint32_t wm_surf_offset_renderbuffer,
459 uint32_t wm_surf_offset_texture)
461 uint32_t wm_bind_bo_offset;
462 uint32_t *bind = (uint32_t *)
463 brw_state_batch(brw, AUB_TRACE_BINDING_TABLE,
465 BRW_BLORP_NUM_BINDING_TABLE_ENTRIES,
468 bind[BRW_BLORP_RENDERBUFFER_BINDING_TABLE_INDEX] =
469 wm_surf_offset_renderbuffer;
470 bind[BRW_BLORP_TEXTURE_BINDING_TABLE_INDEX] = wm_surf_offset_texture;
472 return wm_bind_bo_offset;
477 * SAMPLER_STATE. See brw_update_sampler_state().
480 gen6_blorp_emit_sampler_state(struct brw_context *brw,
481 const brw_blorp_params *params)
483 uint32_t sampler_offset;
485 struct brw_sampler_state *sampler = (struct brw_sampler_state *)
486 brw_state_batch(brw, AUB_TRACE_SAMPLER_STATE,
487 sizeof(struct brw_sampler_state),
488 32, &sampler_offset);
489 memset(sampler, 0, sizeof(*sampler));
491 sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
492 sampler->ss0.mip_filter = BRW_MIPFILTER_NONE;
493 sampler->ss0.mag_filter = BRW_MAPFILTER_LINEAR;
495 sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
496 sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
497 sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
499 sampler->ss0.min_mag_neq = 1;
503 sampler->ss0.lod_bias = 0;
505 sampler->ss0.lod_preclamp = 1; /* OpenGL mode */
506 sampler->ss0.default_color_mode = 0; /* OpenGL/DX10 mode */
508 /* Set BaseMipLevel, MaxLOD, MinLOD:
510 * XXX: I don't think that using firstLevel, lastLevel works,
511 * because we always setup the surface state as if firstLevel ==
512 * level zero. Probably have to subtract firstLevel from each of
515 sampler->ss0.base_level = U_FIXED(0, 1);
517 sampler->ss1.max_lod = U_FIXED(0, 6);
518 sampler->ss1.min_lod = U_FIXED(0, 6);
520 sampler->ss3.non_normalized_coord = 1;
522 sampler->ss3.address_round |= BRW_ADDRESS_ROUNDING_ENABLE_U_MIN |
523 BRW_ADDRESS_ROUNDING_ENABLE_V_MIN |
524 BRW_ADDRESS_ROUNDING_ENABLE_R_MIN;
525 sampler->ss3.address_round |= BRW_ADDRESS_ROUNDING_ENABLE_U_MAG |
526 BRW_ADDRESS_ROUNDING_ENABLE_V_MAG |
527 BRW_ADDRESS_ROUNDING_ENABLE_R_MAG;
529 return sampler_offset;
534 * 3DSTATE_SAMPLER_STATE_POINTERS. See upload_sampler_state_pointers().
537 gen6_blorp_emit_sampler_state_pointers(struct brw_context *brw,
538 const brw_blorp_params *params,
539 uint32_t sampler_offset)
541 struct intel_context *intel = &brw->intel;
544 OUT_BATCH(_3DSTATE_SAMPLER_STATE_POINTERS << 16 |
545 VS_SAMPLER_STATE_CHANGE |
546 GS_SAMPLER_STATE_CHANGE |
547 PS_SAMPLER_STATE_CHANGE |
549 OUT_BATCH(0); /* VS */
550 OUT_BATCH(0); /* GS */
551 OUT_BATCH(sampler_offset);
558 * Disable vertex shader.
561 gen6_blorp_emit_vs_disable(struct brw_context *brw,
562 const brw_blorp_params *params)
564 struct intel_context *intel = &brw->intel;
566 if (intel->gen == 6) {
567 /* From the BSpec, Volume 2a, Part 3 "Vertex Shader", Section
568 * 3DSTATE_VS, Dword 5.0 "VS Function Enable":
570 * [DevSNB] A pipeline flush must be programmed prior to a
571 * 3DSTATE_VS command that causes the VS Function Enable to
572 * toggle. Pipeline flush can be executed by sending a PIPE_CONTROL
573 * command with CS stall bit set and a post sync operation.
575 intel_emit_post_sync_nonzero_flush(intel);
579 OUT_BATCH(_3DSTATE_VS << 16 | (6 - 2));
591 * Disable the geometry shader.
594 gen6_blorp_emit_gs_disable(struct brw_context *brw,
595 const brw_blorp_params *params)
597 struct intel_context *intel = &brw->intel;
600 OUT_BATCH(_3DSTATE_GS << 16 | (7 - 2));
613 * Disable the clipper.
615 * The BLORP op emits a rectangle primitive, which requires clipping to
616 * be disabled. From page 10 of the Sandy Bridge PRM Volume 2 Part 1
617 * Section 1.3 "3D Primitives Overview":
619 * Either the CLIP unit should be DISABLED, or the CLIP unit's Clip
620 * Mode should be set to a value other than CLIPMODE_NORMAL.
622 * Also disable perspective divide. This doesn't change the clipper's
623 * output, but does spare a few electrons.
626 gen6_blorp_emit_clip_disable(struct brw_context *brw,
627 const brw_blorp_params *params)
629 struct intel_context *intel = &brw->intel;
632 OUT_BATCH(_3DSTATE_CLIP << 16 | (4 - 2));
634 OUT_BATCH(GEN6_CLIP_PERSPECTIVE_DIVIDE_DISABLE);
642 * Disable ViewportTransformEnable (dw2.1)
644 * From the SandyBridge PRM, Volume 2, Part 1, Section 1.3, "3D
645 * Primitives Overview":
646 * RECTLIST: Viewport Mapping must be DISABLED (as is typical with the
647 * use of screen- space coordinates).
649 * A solid rectangle must be rendered, so set FrontFaceFillMode (dw2.4:3)
650 * and BackFaceFillMode (dw2.5:6) to SOLID(0).
652 * From the Sandy Bridge PRM, Volume 2, Part 1, Section
653 * 6.4.1.1 3DSTATE_SF, Field FrontFaceFillMode:
654 * SOLID: Any triangle or rectangle object found to be front-facing
655 * is rendered as a solid object. This setting is required when
656 * (rendering rectangle (RECTLIST) objects.
659 gen6_blorp_emit_sf_config(struct brw_context *brw,
660 const brw_blorp_params *params)
662 struct intel_context *intel = &brw->intel;
665 OUT_BATCH(_3DSTATE_SF << 16 | (20 - 2));
666 OUT_BATCH((1 - 1) << GEN6_SF_NUM_OUTPUTS_SHIFT | /* only position */
667 1 << GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT |
668 0 << GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT);
669 OUT_BATCH(0); /* dw2 */
670 OUT_BATCH(params->num_samples > 1 ? GEN6_SF_MSRAST_ON_PATTERN : 0);
671 for (int i = 0; i < 16; ++i)
678 * Enable or disable thread dispatch and set the HiZ op appropriately.
681 gen6_blorp_emit_wm_config(struct brw_context *brw,
682 const brw_blorp_params *params,
683 uint32_t prog_offset,
684 brw_blorp_prog_data *prog_data)
686 struct intel_context *intel = &brw->intel;
687 uint32_t dw2, dw4, dw5, dw6;
689 /* Even when thread dispatch is disabled, max threads (dw5.25:31) must be
690 * nonzero to prevent the GPU from hanging. See the valid ranges in the
691 * BSpec, Volume 2a.11 Windower, Section 3DSTATE_WM, Dword 5.25:31
692 * "Maximum Number Of Threads".
694 * To be safe (and to minimize extraneous code) we go ahead and fully
695 * configure the WM state whether or not there is a WM program.
698 dw2 = dw4 = dw5 = dw6 = 0;
699 switch (params->hiz_op) {
700 case GEN6_HIZ_OP_DEPTH_CLEAR:
701 dw4 |= GEN6_WM_DEPTH_CLEAR;
703 case GEN6_HIZ_OP_DEPTH_RESOLVE:
704 dw4 |= GEN6_WM_DEPTH_RESOLVE;
706 case GEN6_HIZ_OP_HIZ_RESOLVE:
707 dw4 |= GEN6_WM_HIERARCHICAL_DEPTH_RESOLVE;
709 case GEN6_HIZ_OP_NONE:
715 dw4 |= GEN6_WM_STATISTICS_ENABLE;
716 dw5 |= GEN6_WM_LINE_AA_WIDTH_1_0;
717 dw5 |= GEN6_WM_LINE_END_CAP_AA_WIDTH_0_5;
718 dw5 |= (brw->max_wm_threads - 1) << GEN6_WM_MAX_THREADS_SHIFT;
719 dw6 |= 0 << GEN6_WM_BARYCENTRIC_INTERPOLATION_MODE_SHIFT; /* No interp */
720 dw6 |= 0 << GEN6_WM_NUM_SF_OUTPUTS_SHIFT; /* No inputs from SF */
721 if (params->use_wm_prog) {
722 dw2 |= 1 << GEN6_WM_SAMPLER_COUNT_SHIFT; /* Up to 4 samplers */
723 dw4 |= prog_data->first_curbe_grf << GEN6_WM_DISPATCH_START_GRF_SHIFT_0;
724 dw5 |= GEN6_WM_16_DISPATCH_ENABLE;
725 dw5 |= GEN6_WM_KILL_ENABLE; /* TODO: temporarily smash on */
726 dw5 |= GEN6_WM_DISPATCH_ENABLE; /* We are rendering */
729 if (params->num_samples > 1) {
730 dw6 |= GEN6_WM_MSRAST_ON_PATTERN;
731 if (prog_data && prog_data->persample_msaa_dispatch)
732 dw6 |= GEN6_WM_MSDISPMODE_PERSAMPLE;
734 dw6 |= GEN6_WM_MSDISPMODE_PERPIXEL;
736 dw6 |= GEN6_WM_MSRAST_OFF_PIXEL;
737 dw6 |= GEN6_WM_MSDISPMODE_PERSAMPLE;
741 OUT_BATCH(_3DSTATE_WM << 16 | (9 - 2));
742 OUT_BATCH(params->use_wm_prog ? prog_offset : 0);
744 OUT_BATCH(0); /* No scratch needed */
748 OUT_BATCH(0); /* No other programs */
749 OUT_BATCH(0); /* No other programs */
755 gen6_blorp_emit_constant_ps(struct brw_context *brw,
756 const brw_blorp_params *params,
757 uint32_t wm_push_const_offset)
759 struct intel_context *intel = &brw->intel;
761 /* Make sure the push constants fill an exact integer number of
764 assert(sizeof(brw_blorp_wm_push_constants) % 32 == 0);
766 /* There must be at least one register worth of push constant data. */
767 assert(BRW_BLORP_NUM_PUSH_CONST_REGS > 0);
769 /* Enable push constant buffer 0. */
771 OUT_BATCH(_3DSTATE_CONSTANT_PS << 16 |
772 GEN6_CONSTANT_BUFFER_0_ENABLE |
774 OUT_BATCH(wm_push_const_offset + (BRW_BLORP_NUM_PUSH_CONST_REGS - 1));
783 * 3DSTATE_BINDING_TABLE_POINTERS
786 gen6_blorp_emit_binding_table_pointers(struct brw_context *brw,
787 const brw_blorp_params *params,
788 uint32_t wm_bind_bo_offset)
790 struct intel_context *intel = &brw->intel;
793 OUT_BATCH(_3DSTATE_BINDING_TABLE_POINTERS << 16 |
794 GEN6_BINDING_TABLE_MODIFY_PS |
796 OUT_BATCH(0); /* vs -- ignored */
797 OUT_BATCH(0); /* gs -- ignored */
798 OUT_BATCH(wm_bind_bo_offset); /* wm/ps */
804 gen6_blorp_emit_depth_stencil_config(struct brw_context *brw,
805 const brw_blorp_params *params)
807 struct intel_context *intel = &brw->intel;
808 uint32_t draw_x = params->depth.x_offset;
809 uint32_t draw_y = params->depth.y_offset;
810 uint32_t tile_mask_x, tile_mask_y;
812 brw_get_depthstencil_tile_masks(params->depth.mt, NULL,
813 &tile_mask_x, &tile_mask_y);
815 /* 3DSTATE_DEPTH_BUFFER */
817 uint32_t tile_x = draw_x & tile_mask_x;
818 uint32_t tile_y = draw_y & tile_mask_y;
820 intel_region_get_aligned_offset(params->depth.mt->region,
821 draw_x & ~tile_mask_x,
822 draw_y & ~tile_mask_y, false);
824 /* According to the Sandy Bridge PRM, volume 2 part 1, pp326-327
825 * (3DSTATE_DEPTH_BUFFER dw5), in the documentation for "Depth
826 * Coordinate Offset X/Y":
828 * "The 3 LSBs of both offsets must be zero to ensure correct
831 * We have no guarantee that tile_x and tile_y are correctly aligned,
832 * since they are determined by the mipmap layout, which is only aligned
835 * So, to avoid hanging the GPU, just smash the low order 3 bits of
836 * tile_x and tile_y to 0. This is a temporary workaround until we come
837 * up with a better solution.
842 intel_emit_post_sync_nonzero_flush(intel);
843 intel_emit_depth_stall_flushes(intel);
846 OUT_BATCH(_3DSTATE_DEPTH_BUFFER << 16 | (7 - 2));
847 uint32_t pitch_bytes =
848 params->depth.mt->region->pitch * params->depth.mt->region->cpp;
849 OUT_BATCH((pitch_bytes - 1) |
850 params->depth_format << 18 |
851 1 << 21 | /* separate stencil enable */
852 1 << 22 | /* hiz enable */
853 BRW_TILEWALK_YMAJOR << 26 |
854 1 << 27 | /* y-tiled */
855 BRW_SURFACE_2D << 29);
856 OUT_RELOC(params->depth.mt->region->bo,
857 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
859 OUT_BATCH(BRW_SURFACE_MIPMAPLAYOUT_BELOW << 1 |
860 (params->depth.width + tile_x - 1) << 6 |
861 (params->depth.height + tile_y - 1) << 19);
869 /* 3DSTATE_HIER_DEPTH_BUFFER */
871 struct intel_region *hiz_region = params->depth.mt->hiz_mt->region;
872 uint32_t hiz_offset =
873 intel_region_get_aligned_offset(hiz_region,
874 draw_x & ~tile_mask_x,
875 (draw_y & ~tile_mask_y) / 2, false);
878 OUT_BATCH((_3DSTATE_HIER_DEPTH_BUFFER << 16) | (3 - 2));
879 OUT_BATCH(hiz_region->pitch * hiz_region->cpp - 1);
880 OUT_RELOC(hiz_region->bo,
881 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
886 /* 3DSTATE_STENCIL_BUFFER */
889 OUT_BATCH((_3DSTATE_STENCIL_BUFFER << 16) | (3 - 2));
898 gen6_blorp_emit_depth_disable(struct brw_context *brw,
899 const brw_blorp_params *params)
901 struct intel_context *intel = &brw->intel;
904 OUT_BATCH(_3DSTATE_DEPTH_BUFFER << 16 | (7 - 2));
905 OUT_BATCH((BRW_DEPTHFORMAT_D32_FLOAT << 18) |
906 (BRW_SURFACE_NULL << 29));
916 /* 3DSTATE_CLEAR_PARAMS
918 * From the Sandybridge PRM, Volume 2, Part 1, Section 3DSTATE_CLEAR_PARAMS:
919 * [DevSNB] 3DSTATE_CLEAR_PARAMS packet must follow the DEPTH_BUFFER_STATE
920 * packet when HiZ is enabled and the DEPTH_BUFFER_STATE changes.
923 gen6_blorp_emit_clear_params(struct brw_context *brw,
924 const brw_blorp_params *params)
926 struct intel_context *intel = &brw->intel;
929 OUT_BATCH(_3DSTATE_CLEAR_PARAMS << 16 |
930 GEN5_DEPTH_CLEAR_VALID |
932 OUT_BATCH(params->depth.mt ? params->depth.mt->depth_clear_value : 0);
937 /* 3DSTATE_DRAWING_RECTANGLE */
939 gen6_blorp_emit_drawing_rectangle(struct brw_context *brw,
940 const brw_blorp_params *params)
942 struct intel_context *intel = &brw->intel;
945 OUT_BATCH(_3DSTATE_DRAWING_RECTANGLE << 16 | (4 - 2));
947 OUT_BATCH(((params->x1 - 1) & 0xffff) |
948 ((params->y1 - 1) << 16));
953 /* 3DSTATE_VIEWPORT_STATE_POINTERS */
955 gen6_blorp_emit_viewport_state(struct brw_context *brw,
956 const brw_blorp_params *params)
958 struct intel_context *intel = &brw->intel;
959 struct brw_cc_viewport *ccv;
960 uint32_t cc_vp_offset;
962 ccv = (struct brw_cc_viewport *)brw_state_batch(brw, AUB_TRACE_CC_VP_STATE,
966 ccv->min_depth = 0.0;
967 ccv->max_depth = 1.0;
970 OUT_BATCH(_3DSTATE_VIEWPORT_STATE_POINTERS << 16 | (4 - 2) |
971 GEN6_CC_VIEWPORT_MODIFY);
972 OUT_BATCH(0); /* clip VP */
973 OUT_BATCH(0); /* SF VP */
974 OUT_BATCH(cc_vp_offset);
981 gen6_blorp_emit_primitive(struct brw_context *brw,
982 const brw_blorp_params *params)
984 struct intel_context *intel = &brw->intel;
987 OUT_BATCH(CMD_3D_PRIM << 16 | (6 - 2) |
988 _3DPRIM_RECTLIST << GEN4_3DPRIM_TOPOLOGY_TYPE_SHIFT |
989 GEN4_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL);
990 OUT_BATCH(3); /* vertex count per instance */
992 OUT_BATCH(1); /* instance count */
1000 * \brief Execute a blit or render pass operation.
1002 * To execute the operation, this function manually constructs and emits a
1003 * batch to draw a rectangle primitive. The batchbuffer is flushed before
1004 * constructing and after emitting the batch.
1006 * This function alters no GL state.
1009 gen6_blorp_exec(struct intel_context *intel,
1010 const brw_blorp_params *params)
1012 struct gl_context *ctx = &intel->ctx;
1013 struct brw_context *brw = brw_context(ctx);
1014 brw_blorp_prog_data *prog_data = NULL;
1015 uint32_t cc_blend_state_offset = 0;
1016 uint32_t cc_state_offset = 0;
1017 uint32_t depthstencil_offset;
1018 uint32_t wm_push_const_offset = 0;
1019 uint32_t wm_bind_bo_offset = 0;
1021 uint32_t prog_offset = params->get_wm_prog(brw, &prog_data);
1022 gen6_blorp_emit_batch_head(brw, params);
1023 gen6_emit_3dstate_multisample(brw, params->num_samples);
1024 gen6_emit_3dstate_sample_mask(brw, params->num_samples, 1.0, false);
1025 gen6_blorp_emit_state_base_address(brw, params);
1026 gen6_blorp_emit_vertices(brw, params);
1027 gen6_blorp_emit_urb_config(brw, params);
1028 if (params->use_wm_prog) {
1029 cc_blend_state_offset = gen6_blorp_emit_blend_state(brw, params);
1030 cc_state_offset = gen6_blorp_emit_cc_state(brw, params);
1032 depthstencil_offset = gen6_blorp_emit_depth_stencil_state(brw, params);
1033 gen6_blorp_emit_cc_state_pointers(brw, params, cc_blend_state_offset,
1034 depthstencil_offset, cc_state_offset);
1035 if (params->use_wm_prog) {
1036 uint32_t wm_surf_offset_renderbuffer;
1037 uint32_t wm_surf_offset_texture;
1038 uint32_t sampler_offset;
1039 wm_push_const_offset = gen6_blorp_emit_wm_constants(brw, params);
1040 wm_surf_offset_renderbuffer =
1041 gen6_blorp_emit_surface_state(brw, params, ¶ms->dst,
1042 I915_GEM_DOMAIN_RENDER,
1043 I915_GEM_DOMAIN_RENDER);
1044 wm_surf_offset_texture =
1045 gen6_blorp_emit_surface_state(brw, params, ¶ms->src,
1046 I915_GEM_DOMAIN_SAMPLER, 0);
1048 gen6_blorp_emit_binding_table(brw, params,
1049 wm_surf_offset_renderbuffer,
1050 wm_surf_offset_texture);
1051 sampler_offset = gen6_blorp_emit_sampler_state(brw, params);
1052 gen6_blorp_emit_sampler_state_pointers(brw, params, sampler_offset);
1054 gen6_blorp_emit_vs_disable(brw, params);
1055 gen6_blorp_emit_gs_disable(brw, params);
1056 gen6_blorp_emit_clip_disable(brw, params);
1057 gen6_blorp_emit_sf_config(brw, params);
1058 if (params->use_wm_prog)
1059 gen6_blorp_emit_constant_ps(brw, params, wm_push_const_offset);
1060 gen6_blorp_emit_wm_config(brw, params, prog_offset, prog_data);
1061 if (params->use_wm_prog)
1062 gen6_blorp_emit_binding_table_pointers(brw, params, wm_bind_bo_offset);
1063 gen6_blorp_emit_viewport_state(brw, params);
1065 if (params->depth.mt)
1066 gen6_blorp_emit_depth_stencil_config(brw, params);
1068 gen6_blorp_emit_depth_disable(brw, params);
1069 gen6_blorp_emit_clear_params(brw, params);
1070 gen6_blorp_emit_drawing_rectangle(brw, params);
1071 gen6_blorp_emit_primitive(brw, params);
1073 /* See comments above at first invocation of intel_flush() in
1074 * gen6_blorp_emit_batch_head().
1079 brw->state.dirty.brw = ~0;
1080 brw->state.dirty.cache = ~0;