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 \
50 * Compute masks to determine how much of draw_x and draw_y should be
51 * performed using the fine adjustment of "depth coordinate offset X/Y"
52 * (dw5 of 3DSTATE_DEPTH_BUFFER). See the emit_depthbuffer() function for
56 gen6_blorp_compute_tile_masks(const brw_blorp_params *params,
57 uint32_t *tile_mask_x, uint32_t *tile_mask_y)
59 uint32_t depth_mask_x, depth_mask_y, hiz_mask_x, hiz_mask_y;
60 intel_region_get_tile_masks(params->depth.mt->region,
61 &depth_mask_x, &depth_mask_y);
62 intel_region_get_tile_masks(params->depth.mt->hiz_mt->region,
63 &hiz_mask_x, &hiz_mask_y);
65 /* Each HiZ row represents 2 rows of pixels */
66 hiz_mask_y = hiz_mask_y << 1 | 1;
68 *tile_mask_x = depth_mask_x | hiz_mask_x;
69 *tile_mask_y = depth_mask_y | hiz_mask_y;
74 gen6_blorp_emit_batch_head(struct brw_context *brw,
75 const brw_blorp_params *params)
77 struct gl_context *ctx = &brw->intel.ctx;
78 struct intel_context *intel = &brw->intel;
80 /* To ensure that the batch contains only the resolve, flush the batch
81 * before beginning and after finishing emitting the resolve packets.
83 * Ideally, we would not need to flush for the resolve op. But, I suspect
84 * that it's unsafe for CMD_PIPELINE_SELECT to occur multiple times in
85 * a single batch, and there is no safe way to ensure that other than by
86 * fencing the resolve with flushes. Ideally, we would just detect if
87 * a batch is in progress and do the right thing, but that would require
88 * the ability to *safely* access brw_context::state::dirty::brw
89 * outside of the brw_upload_state() codepath.
93 /* CMD_PIPELINE_SELECT
95 * Select the 3D pipeline, as opposed to the media pipeline.
99 OUT_BATCH(brw->CMD_PIPELINE_SELECT << 16);
103 /* 3DSTATE_MULTISAMPLE */
105 int length = intel->gen == 7 ? 4 : 3;
108 OUT_BATCH(_3DSTATE_MULTISAMPLE << 16 | (length - 2));
109 OUT_BATCH(MS_PIXEL_LOCATION_CENTER |
118 /* 3DSTATE_SAMPLE_MASK */
121 OUT_BATCH(_3DSTATE_SAMPLE_MASK << 16 | (2 - 2));
126 /* CMD_STATE_BASE_ADDRESS
128 * From the Sandy Bridge PRM, Volume 1, Part 1, Table STATE_BASE_ADDRESS:
129 * The following commands must be reissued following any change to the
131 * 3DSTATE_CC_POINTERS
132 * 3DSTATE_BINDING_TABLE_POINTERS
133 * 3DSTATE_SAMPLER_STATE_POINTERS
134 * 3DSTATE_VIEWPORT_STATE_POINTERS
135 * MEDIA_STATE_POINTERS
139 OUT_BATCH(CMD_STATE_BASE_ADDRESS << 16 | (10 - 2));
140 OUT_BATCH(1); /* GeneralStateBaseAddressModifyEnable */
141 /* SurfaceStateBaseAddress */
142 OUT_RELOC(intel->batch.bo, I915_GEM_DOMAIN_SAMPLER, 0, 1);
143 /* DynamicStateBaseAddress */
144 OUT_RELOC(intel->batch.bo, (I915_GEM_DOMAIN_RENDER |
145 I915_GEM_DOMAIN_INSTRUCTION), 0, 1);
146 OUT_BATCH(1); /* IndirectObjectBaseAddress */
147 if (params->use_wm_prog) {
148 OUT_RELOC(brw->cache.bo, I915_GEM_DOMAIN_INSTRUCTION, 0,
149 1); /* Instruction base address: shader kernels */
151 OUT_BATCH(1); /* InstructionBaseAddress */
153 OUT_BATCH(1); /* GeneralStateUpperBound */
154 OUT_BATCH(1); /* DynamicStateUpperBound */
155 OUT_BATCH(1); /* IndirectObjectUpperBound*/
156 OUT_BATCH(1); /* InstructionAccessUpperBound */
162 gen6_blorp_emit_vertices(struct brw_context *brw,
163 const brw_blorp_params *params)
165 struct intel_context *intel = &brw->intel;
166 uint32_t vertex_offset;
168 /* Setup VBO for the rectangle primitive..
170 * A rectangle primitive (3DPRIM_RECTLIST) consists of only three
171 * vertices. The vertices reside in screen space with DirectX coordinates
172 * (that is, (0, 0) is the upper left corner).
179 * Since the VS is disabled, the clipper loads each VUE directly from
180 * the URB. This is controlled by the 3DSTATE_VERTEX_BUFFERS and
181 * 3DSTATE_VERTEX_ELEMENTS packets below. The VUE contents are as follows:
182 * dw0: Reserved, MBZ.
183 * dw1: Render Target Array Index. The HiZ op does not use indexed
184 * vertices, so set the dword to 0.
185 * dw2: Viewport Index. The HiZ op disables viewport mapping and
186 * scissoring, so set the dword to 0.
187 * dw3: Point Width: The HiZ op does not emit the POINTLIST primitive, so
188 * set the dword to 0.
189 * dw4: Vertex Position X.
190 * dw5: Vertex Position Y.
191 * dw6: Vertex Position Z.
192 * dw7: Vertex Position W.
194 * For details, see the Sandybridge PRM, Volume 2, Part 1, Section 1.5.1
195 * "Vertex URB Entry (VUE) Formats".
200 const float vertices[GEN6_BLORP_VBO_SIZE] = {
201 /* v0 */ 0, 0, 0, 0, params->x0, params->y1, 0, 1,
202 /* v1 */ 0, 0, 0, 0, params->x1, params->y1, 0, 1,
203 /* v2 */ 0, 0, 0, 0, params->x0, params->y0, 0, 1,
206 vertex_data = (float *) brw_state_batch(brw, AUB_TRACE_NO_TYPE,
207 GEN6_BLORP_VBO_SIZE, 32,
209 memcpy(vertex_data, vertices, GEN6_BLORP_VBO_SIZE);
212 /* 3DSTATE_VERTEX_BUFFERS */
214 const int num_buffers = 1;
215 const int batch_length = 1 + 4 * num_buffers;
217 uint32_t dw0 = GEN6_VB0_ACCESS_VERTEXDATA |
218 (GEN6_BLORP_NUM_VUE_ELEMS * sizeof(float)) << BRW_VB0_PITCH_SHIFT;
221 dw0 |= GEN7_VB0_ADDRESS_MODIFYENABLE;
223 BEGIN_BATCH(batch_length);
224 OUT_BATCH((_3DSTATE_VERTEX_BUFFERS << 16) | (batch_length - 2));
227 OUT_RELOC(intel->batch.bo, I915_GEM_DOMAIN_VERTEX, 0,
230 OUT_RELOC(intel->batch.bo, I915_GEM_DOMAIN_VERTEX, 0,
231 vertex_offset + GEN6_BLORP_VBO_SIZE - 1);
236 /* 3DSTATE_VERTEX_ELEMENTS
238 * Fetch dwords 0 - 7 from each VUE. See the comments above where
239 * the vertex_bo is filled with data.
242 const int num_elements = 2;
243 const int batch_length = 1 + 2 * num_elements;
245 BEGIN_BATCH(batch_length);
246 OUT_BATCH((_3DSTATE_VERTEX_ELEMENTS << 16) | (batch_length - 2));
248 OUT_BATCH(GEN6_VE0_VALID |
249 BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT |
250 0 << BRW_VE0_SRC_OFFSET_SHIFT);
251 OUT_BATCH(BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_0_SHIFT |
252 BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_1_SHIFT |
253 BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_2_SHIFT |
254 BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_3_SHIFT);
256 OUT_BATCH(GEN6_VE0_VALID |
257 BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT |
258 16 << BRW_VE0_SRC_OFFSET_SHIFT);
259 OUT_BATCH(BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_0_SHIFT |
260 BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_1_SHIFT |
261 BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_2_SHIFT |
262 BRW_VE1_COMPONENT_STORE_SRC << BRW_VE1_COMPONENT_3_SHIFT);
270 * Assign the entire URB to the VS. Even though the VS disabled, URB space
271 * is still needed because the clipper loads the VUE's from the URB. From
272 * the Sandybridge PRM, Volume 2, Part 1, Section 3DSTATE,
273 * Dword 1.15:0 "VS Number of URB Entries":
274 * This field is always used (even if VS Function Enable is DISABLED).
276 * The warning below appears in the PRM (Section 3DSTATE_URB), but we can
277 * safely ignore it because this batch contains only one draw call.
278 * Because of URB corruption caused by allocating a previous GS unit
279 * URB entry to the VS unit, software is required to send a “GS NULL
280 * Fence” (Send URB fence with VS URB size == 1 and GS URB size == 0)
281 * plus a dummy DRAW call before any case where VS will be taking over
285 gen6_blorp_emit_urb_config(struct brw_context *brw,
286 const brw_blorp_params *params)
288 struct intel_context *intel = &brw->intel;
291 OUT_BATCH(_3DSTATE_URB << 16 | (3 - 2));
292 OUT_BATCH(brw->urb.max_vs_entries << GEN6_URB_VS_ENTRIES_SHIFT);
300 gen6_blorp_emit_blend_state(struct brw_context *brw,
301 const brw_blorp_params *params)
303 uint32_t cc_blend_state_offset;
305 struct gen6_blend_state *blend = (struct gen6_blend_state *)
306 brw_state_batch(brw, AUB_TRACE_BLEND_STATE,
307 sizeof(struct gen6_blend_state), 64,
308 &cc_blend_state_offset);
310 memset(blend, 0, sizeof(*blend));
312 // TODO: handle other formats.
313 blend->blend1.pre_blend_clamp_enable = 1;
314 blend->blend1.post_blend_clamp_enable = 1;
315 blend->blend1.clamp_range = BRW_RENDERTARGET_CLAMPRANGE_FORMAT;
317 blend->blend1.write_disable_r = false;
318 blend->blend1.write_disable_g = false;
319 blend->blend1.write_disable_b = false;
320 blend->blend1.write_disable_a = false;
322 return cc_blend_state_offset;
328 gen6_blorp_emit_cc_state(struct brw_context *brw,
329 const brw_blorp_params *params)
331 uint32_t cc_state_offset;
333 struct gen6_color_calc_state *cc = (struct gen6_color_calc_state *)
334 brw_state_batch(brw, AUB_TRACE_CC_STATE,
335 sizeof(gen6_color_calc_state), 64,
337 memset(cc, 0, sizeof(*cc));
339 return cc_state_offset;
344 * \param out_offset is relative to
345 * CMD_STATE_BASE_ADDRESS.DynamicStateBaseAddress.
348 gen6_blorp_emit_depth_stencil_state(struct brw_context *brw,
349 const brw_blorp_params *params)
351 uint32_t depthstencil_offset;
353 struct gen6_depth_stencil_state *state;
354 state = (struct gen6_depth_stencil_state *)
355 brw_state_batch(brw, AUB_TRACE_DEPTH_STENCIL_STATE,
357 &depthstencil_offset);
358 memset(state, 0, sizeof(*state));
360 /* See the following sections of the Sandy Bridge PRM, Volume 1, Part2:
361 * - 7.5.3.1 Depth Buffer Clear
362 * - 7.5.3.2 Depth Buffer Resolve
363 * - 7.5.3.3 Hierarchical Depth Buffer Resolve
365 state->ds2.depth_write_enable = 1;
366 if (params->hiz_op == GEN6_HIZ_OP_DEPTH_RESOLVE) {
367 state->ds2.depth_test_enable = 1;
368 state->ds2.depth_test_func = COMPAREFUNC_NEVER;
371 return depthstencil_offset;
375 /* 3DSTATE_CC_STATE_POINTERS
377 * The pointer offsets are relative to
378 * CMD_STATE_BASE_ADDRESS.DynamicStateBaseAddress.
380 * The HiZ op doesn't use BLEND_STATE or COLOR_CALC_STATE.
383 gen6_blorp_emit_cc_state_pointers(struct brw_context *brw,
384 const brw_blorp_params *params,
385 uint32_t cc_blend_state_offset,
386 uint32_t depthstencil_offset,
387 uint32_t cc_state_offset)
389 struct intel_context *intel = &brw->intel;
392 OUT_BATCH(_3DSTATE_CC_STATE_POINTERS << 16 | (4 - 2));
393 OUT_BATCH(cc_blend_state_offset | 1); /* BLEND_STATE offset */
394 OUT_BATCH(depthstencil_offset | 1); /* DEPTH_STENCIL_STATE offset */
395 OUT_BATCH(cc_state_offset | 1); /* COLOR_CALC_STATE offset */
400 /* WM push constants */
402 gen6_blorp_emit_wm_constants(struct brw_context *brw,
403 const brw_blorp_params *params)
405 uint32_t wm_push_const_offset;
407 void *constants = brw_state_batch(brw, AUB_TRACE_WM_CONSTANTS,
408 sizeof(params->wm_push_consts),
409 32, &wm_push_const_offset);
410 memcpy(constants, ¶ms->wm_push_consts,
411 sizeof(params->wm_push_consts));
413 return wm_push_const_offset;
417 /* SURFACE_STATE for renderbuffer or texture surface (see
418 * brw_update_renderbuffer_surface and brw_update_texture_surface)
421 gen6_blorp_emit_surface_state(struct brw_context *brw,
422 const brw_blorp_params *params,
423 const brw_blorp_surface_info *surface,
424 uint32_t read_domains, uint32_t write_domain)
426 uint32_t wm_surf_offset;
427 uint32_t width, height;
428 surface->get_miplevel_dims(&width, &height);
429 if (surface->map_stencil_as_y_tiled) {
433 struct intel_region *region = surface->mt->region;
435 /* TODO: handle other formats */
436 uint32_t format = surface->map_stencil_as_y_tiled
437 ? BRW_SURFACEFORMAT_R8_UNORM : BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
439 uint32_t *surf = (uint32_t *)
440 brw_state_batch(brw, AUB_TRACE_SURFACE_STATE, 6 * 4, 32,
443 surf[0] = (BRW_SURFACE_2D << BRW_SURFACE_TYPE_SHIFT |
444 BRW_SURFACE_MIPMAPLAYOUT_BELOW << BRW_SURFACE_MIPLAYOUT_SHIFT |
445 BRW_SURFACE_CUBEFACE_ENABLES |
446 format << BRW_SURFACE_FORMAT_SHIFT);
449 surf[1] = region->bo->offset; /* No tile offsets needed */
451 surf[2] = (0 << BRW_SURFACE_LOD_SHIFT |
452 (width - 1) << BRW_SURFACE_WIDTH_SHIFT |
453 (height - 1) << BRW_SURFACE_HEIGHT_SHIFT);
455 uint32_t tiling = surface->map_stencil_as_y_tiled
456 ? BRW_SURFACE_TILED | BRW_SURFACE_TILED_Y
457 : brw_get_surface_tiling_bits(region->tiling);
458 uint32_t pitch_bytes = region->pitch * region->cpp;
459 if (surface->map_stencil_as_y_tiled)
462 0 << BRW_SURFACE_DEPTH_SHIFT |
463 (pitch_bytes - 1) << BRW_SURFACE_PITCH_SHIFT);
467 surf[5] = (0 << BRW_SURFACE_X_OFFSET_SHIFT |
468 0 << BRW_SURFACE_Y_OFFSET_SHIFT |
469 (surface->mt->align_h == 4 ?
470 BRW_SURFACE_VERTICAL_ALIGN_ENABLE : 0));
472 /* Emit relocation to surface contents */
473 drm_intel_bo_emit_reloc(brw->intel.batch.bo,
476 surf[1] - region->bo->offset,
477 read_domains, write_domain);
479 return wm_surf_offset;
483 /* BINDING_TABLE. See brw_wm_binding_table(). */
485 gen6_blorp_emit_binding_table(struct brw_context *brw,
486 const brw_blorp_params *params,
487 uint32_t wm_surf_offset_renderbuffer,
488 uint32_t wm_surf_offset_texture)
490 uint32_t wm_bind_bo_offset;
491 uint32_t *bind = (uint32_t *)
492 brw_state_batch(brw, AUB_TRACE_BINDING_TABLE,
494 BRW_BLORP_NUM_BINDING_TABLE_ENTRIES,
497 bind[BRW_BLORP_RENDERBUFFER_BINDING_TABLE_INDEX] =
498 wm_surf_offset_renderbuffer;
499 bind[BRW_BLORP_TEXTURE_BINDING_TABLE_INDEX] = wm_surf_offset_texture;
501 return wm_bind_bo_offset;
506 * SAMPLER_STATE. See brw_update_sampler_state().
509 gen6_blorp_emit_sampler_state(struct brw_context *brw,
510 const brw_blorp_params *params)
512 uint32_t sampler_offset;
514 struct brw_sampler_state *sampler = (struct brw_sampler_state *)
515 brw_state_batch(brw, AUB_TRACE_SAMPLER_STATE,
516 sizeof(struct brw_sampler_state),
517 32, &sampler_offset);
518 memset(sampler, 0, sizeof(*sampler));
520 sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
521 sampler->ss0.mip_filter = BRW_MIPFILTER_NONE;
522 sampler->ss0.mag_filter = BRW_MAPFILTER_LINEAR;
524 sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
525 sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
526 sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
528 sampler->ss0.min_mag_neq = 1;
532 sampler->ss0.lod_bias = 0;
534 sampler->ss0.lod_preclamp = 1; /* OpenGL mode */
535 sampler->ss0.default_color_mode = 0; /* OpenGL/DX10 mode */
537 /* Set BaseMipLevel, MaxLOD, MinLOD:
539 * XXX: I don't think that using firstLevel, lastLevel works,
540 * because we always setup the surface state as if firstLevel ==
541 * level zero. Probably have to subtract firstLevel from each of
544 sampler->ss0.base_level = U_FIXED(0, 1);
546 sampler->ss1.max_lod = U_FIXED(0, 6);
547 sampler->ss1.min_lod = U_FIXED(0, 6);
549 sampler->ss3.non_normalized_coord = 1;
551 sampler->ss3.address_round |= BRW_ADDRESS_ROUNDING_ENABLE_U_MIN |
552 BRW_ADDRESS_ROUNDING_ENABLE_V_MIN |
553 BRW_ADDRESS_ROUNDING_ENABLE_R_MIN;
554 sampler->ss3.address_round |= BRW_ADDRESS_ROUNDING_ENABLE_U_MAG |
555 BRW_ADDRESS_ROUNDING_ENABLE_V_MAG |
556 BRW_ADDRESS_ROUNDING_ENABLE_R_MAG;
558 return sampler_offset;
563 * 3DSTATE_SAMPLER_STATE_POINTERS. See upload_sampler_state_pointers().
566 gen6_blorp_emit_sampler_state_pointers(struct brw_context *brw,
567 const brw_blorp_params *params,
568 uint32_t sampler_offset)
570 struct intel_context *intel = &brw->intel;
573 OUT_BATCH(_3DSTATE_SAMPLER_STATE_POINTERS << 16 |
574 VS_SAMPLER_STATE_CHANGE |
575 GS_SAMPLER_STATE_CHANGE |
576 PS_SAMPLER_STATE_CHANGE |
578 OUT_BATCH(0); /* VS */
579 OUT_BATCH(0); /* GS */
580 OUT_BATCH(sampler_offset);
587 * Disable vertex shader.
590 gen6_blorp_emit_vs_disable(struct brw_context *brw,
591 const brw_blorp_params *params)
593 struct intel_context *intel = &brw->intel;
595 if (intel->gen == 6) {
596 /* From the BSpec, Volume 2a, Part 3 "Vertex Shader", Section
597 * 3DSTATE_VS, Dword 5.0 "VS Function Enable":
599 * [DevSNB] A pipeline flush must be programmed prior to a
600 * 3DSTATE_VS command that causes the VS Function Enable to
601 * toggle. Pipeline flush can be executed by sending a PIPE_CONTROL
602 * command with CS stall bit set and a post sync operation.
604 intel_emit_post_sync_nonzero_flush(intel);
608 OUT_BATCH(_3DSTATE_VS << 16 | (6 - 2));
620 * Disable the geometry shader.
623 gen6_blorp_emit_gs_disable(struct brw_context *brw,
624 const brw_blorp_params *params)
626 struct intel_context *intel = &brw->intel;
629 OUT_BATCH(_3DSTATE_GS << 16 | (7 - 2));
642 * Disable the clipper.
644 * The BLORP op emits a rectangle primitive, which requires clipping to
645 * be disabled. From page 10 of the Sandy Bridge PRM Volume 2 Part 1
646 * Section 1.3 "3D Primitives Overview":
648 * Either the CLIP unit should be DISABLED, or the CLIP unit's Clip
649 * Mode should be set to a value other than CLIPMODE_NORMAL.
651 * Also disable perspective divide. This doesn't change the clipper's
652 * output, but does spare a few electrons.
655 gen6_blorp_emit_clip_disable(struct brw_context *brw,
656 const brw_blorp_params *params)
658 struct intel_context *intel = &brw->intel;
661 OUT_BATCH(_3DSTATE_CLIP << 16 | (4 - 2));
663 OUT_BATCH(GEN6_CLIP_PERSPECTIVE_DIVIDE_DISABLE);
671 * Disable ViewportTransformEnable (dw2.1)
673 * From the SandyBridge PRM, Volume 2, Part 1, Section 1.3, "3D
674 * Primitives Overview":
675 * RECTLIST: Viewport Mapping must be DISABLED (as is typical with the
676 * use of screen- space coordinates).
678 * A solid rectangle must be rendered, so set FrontFaceFillMode (dw2.4:3)
679 * and BackFaceFillMode (dw2.5:6) to SOLID(0).
681 * From the Sandy Bridge PRM, Volume 2, Part 1, Section
682 * 6.4.1.1 3DSTATE_SF, Field FrontFaceFillMode:
683 * SOLID: Any triangle or rectangle object found to be front-facing
684 * is rendered as a solid object. This setting is required when
685 * (rendering rectangle (RECTLIST) objects.
688 gen6_blorp_emit_sf_config(struct brw_context *brw,
689 const brw_blorp_params *params)
691 struct intel_context *intel = &brw->intel;
694 OUT_BATCH(_3DSTATE_SF << 16 | (20 - 2));
695 OUT_BATCH((1 - 1) << GEN6_SF_NUM_OUTPUTS_SHIFT | /* only position */
696 1 << GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT |
697 0 << GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT);
698 for (int i = 0; i < 18; ++i)
705 * Enable or disable thread dispatch and set the HiZ op appropriately.
708 gen6_blorp_emit_wm_config(struct brw_context *brw,
709 const brw_blorp_params *params,
710 uint32_t prog_offset,
711 brw_blorp_prog_data *prog_data)
713 struct intel_context *intel = &brw->intel;
714 uint32_t dw2, dw4, dw5, dw6;
716 /* Even when thread dispatch is disabled, max threads (dw5.25:31) must be
717 * nonzero to prevent the GPU from hanging. See the valid ranges in the
718 * BSpec, Volume 2a.11 Windower, Section 3DSTATE_WM, Dword 5.25:31
719 * "Maximum Number Of Threads".
721 * To be safe (and to minimize extraneous code) we go ahead and fully
722 * configure the WM state whether or not there is a WM program.
725 dw2 = dw4 = dw5 = dw6 = 0;
726 switch (params->hiz_op) {
727 case GEN6_HIZ_OP_DEPTH_CLEAR:
728 assert(!"not implemented");
729 dw4 |= GEN6_WM_DEPTH_CLEAR;
731 case GEN6_HIZ_OP_DEPTH_RESOLVE:
732 dw4 |= GEN6_WM_DEPTH_RESOLVE;
734 case GEN6_HIZ_OP_HIZ_RESOLVE:
735 dw4 |= GEN6_WM_HIERARCHICAL_DEPTH_RESOLVE;
737 case GEN6_HIZ_OP_NONE:
743 dw4 |= GEN6_WM_STATISTICS_ENABLE;
744 dw5 |= GEN6_WM_LINE_AA_WIDTH_1_0;
745 dw5 |= GEN6_WM_LINE_END_CAP_AA_WIDTH_0_5;
746 dw5 |= (brw->max_wm_threads - 1) << GEN6_WM_MAX_THREADS_SHIFT;
747 dw6 |= 0 << GEN6_WM_BARYCENTRIC_INTERPOLATION_MODE_SHIFT; /* No interp */
748 dw6 |= 0 << GEN6_WM_NUM_SF_OUTPUTS_SHIFT; /* No inputs from SF */
749 if (params->use_wm_prog) {
750 dw2 |= 1 << GEN6_WM_SAMPLER_COUNT_SHIFT; /* Up to 4 samplers */
751 dw4 |= prog_data->first_curbe_grf << GEN6_WM_DISPATCH_START_GRF_SHIFT_0;
752 dw5 |= GEN6_WM_16_DISPATCH_ENABLE;
753 dw5 |= GEN6_WM_KILL_ENABLE; /* TODO: temporarily smash on */
754 dw5 |= GEN6_WM_DISPATCH_ENABLE; /* We are rendering */
758 OUT_BATCH(_3DSTATE_WM << 16 | (9 - 2));
759 OUT_BATCH(params->use_wm_prog ? prog_offset : 0);
761 OUT_BATCH(0); /* No scratch needed */
764 OUT_BATCH(dw6); /* only position */
765 OUT_BATCH(0); /* No other programs */
766 OUT_BATCH(0); /* No other programs */
772 gen6_blorp_emit_constant_ps(struct brw_context *brw,
773 const brw_blorp_params *params,
774 uint32_t wm_push_const_offset)
776 struct intel_context *intel = &brw->intel;
778 /* Make sure the push constants fill an exact integer number of
781 assert(sizeof(brw_blorp_wm_push_constants) % 32 == 0);
783 /* There must be at least one register worth of push constant data. */
784 assert(BRW_BLORP_NUM_PUSH_CONST_REGS > 0);
786 /* Enable push constant buffer 0. */
788 OUT_BATCH(_3DSTATE_CONSTANT_PS << 16 |
789 GEN6_CONSTANT_BUFFER_0_ENABLE |
791 OUT_BATCH(wm_push_const_offset + (BRW_BLORP_NUM_PUSH_CONST_REGS - 1));
800 * 3DSTATE_BINDING_TABLE_POINTERS
803 gen6_blorp_emit_binding_table_pointers(struct brw_context *brw,
804 const brw_blorp_params *params,
805 uint32_t wm_bind_bo_offset)
807 struct intel_context *intel = &brw->intel;
810 OUT_BATCH(_3DSTATE_BINDING_TABLE_POINTERS << 16 |
811 GEN6_BINDING_TABLE_MODIFY_PS |
813 OUT_BATCH(0); /* vs -- ignored */
814 OUT_BATCH(0); /* gs -- ignored */
815 OUT_BATCH(wm_bind_bo_offset); /* wm/ps */
821 gen6_blorp_emit_depth_stencil_config(struct brw_context *brw,
822 const brw_blorp_params *params)
824 struct intel_context *intel = &brw->intel;
825 uint32_t draw_x, draw_y;
826 uint32_t tile_mask_x, tile_mask_y;
828 gen6_blorp_compute_tile_masks(params, &tile_mask_x, &tile_mask_y);
829 params->depth.get_draw_offsets(&draw_x, &draw_y);
831 /* 3DSTATE_DEPTH_BUFFER */
833 uint32_t width, height;
834 params->depth.get_miplevel_dims(&width, &height);
836 uint32_t tile_x = draw_x & tile_mask_x;
837 uint32_t tile_y = draw_y & tile_mask_y;
839 intel_region_get_aligned_offset(params->depth.mt->region,
840 draw_x & ~tile_mask_x,
841 draw_y & ~tile_mask_y);
843 /* According to the Sandy Bridge PRM, volume 2 part 1, pp326-327
844 * (3DSTATE_DEPTH_BUFFER dw5), in the documentation for "Depth
845 * Coordinate Offset X/Y":
847 * "The 3 LSBs of both offsets must be zero to ensure correct
850 * We have no guarantee that tile_x and tile_y are correctly aligned,
851 * since they are determined by the mipmap layout, which is only aligned
854 * So, to avoid hanging the GPU, just smash the low order 3 bits of
855 * tile_x and tile_y to 0. This is a temporary workaround until we come
856 * up with a better solution.
861 intel_emit_post_sync_nonzero_flush(intel);
862 intel_emit_depth_stall_flushes(intel);
865 OUT_BATCH(_3DSTATE_DEPTH_BUFFER << 16 | (7 - 2));
866 uint32_t pitch_bytes =
867 params->depth.mt->region->pitch * params->depth.mt->region->cpp;
868 OUT_BATCH((pitch_bytes - 1) |
869 params->depth_format << 18 |
870 1 << 21 | /* separate stencil enable */
871 1 << 22 | /* hiz enable */
872 BRW_TILEWALK_YMAJOR << 26 |
873 1 << 27 | /* y-tiled */
874 BRW_SURFACE_2D << 29);
875 OUT_RELOC(params->depth.mt->region->bo,
876 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
878 OUT_BATCH(BRW_SURFACE_MIPMAPLAYOUT_BELOW << 1 |
879 (width + tile_x - 1) << 6 |
880 (height + tile_y - 1) << 19);
888 /* 3DSTATE_HIER_DEPTH_BUFFER */
890 struct intel_region *hiz_region = params->depth.mt->hiz_mt->region;
891 uint32_t hiz_offset =
892 intel_region_get_aligned_offset(hiz_region,
893 draw_x & ~tile_mask_x,
894 (draw_y & ~tile_mask_y) / 2);
897 OUT_BATCH((_3DSTATE_HIER_DEPTH_BUFFER << 16) | (3 - 2));
898 OUT_BATCH(hiz_region->pitch * hiz_region->cpp - 1);
899 OUT_RELOC(hiz_region->bo,
900 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
905 /* 3DSTATE_STENCIL_BUFFER */
908 OUT_BATCH((_3DSTATE_STENCIL_BUFFER << 16) | (3 - 2));
917 gen6_blorp_emit_depth_disable(struct brw_context *brw,
918 const brw_blorp_params *params)
920 struct intel_context *intel = &brw->intel;
923 OUT_BATCH(_3DSTATE_DEPTH_BUFFER << 16 | (7 - 2));
924 OUT_BATCH((BRW_DEPTHFORMAT_D32_FLOAT << 18) |
925 (BRW_SURFACE_NULL << 29));
935 /* 3DSTATE_CLEAR_PARAMS
937 * From the Sandybridge PRM, Volume 2, Part 1, Section 3DSTATE_CLEAR_PARAMS:
938 * [DevSNB] 3DSTATE_CLEAR_PARAMS packet must follow the DEPTH_BUFFER_STATE
939 * packet when HiZ is enabled and the DEPTH_BUFFER_STATE changes.
942 gen6_blorp_emit_clear_params(struct brw_context *brw,
943 const brw_blorp_params *params)
945 struct intel_context *intel = &brw->intel;
948 OUT_BATCH(_3DSTATE_CLEAR_PARAMS << 16 | (2 - 2));
954 /* 3DSTATE_DRAWING_RECTANGLE */
956 gen6_blorp_emit_drawing_rectangle(struct brw_context *brw,
957 const brw_blorp_params *params)
959 struct intel_context *intel = &brw->intel;
962 OUT_BATCH(_3DSTATE_DRAWING_RECTANGLE << 16 | (4 - 2));
964 OUT_BATCH(((params->x1 - 1) & 0xffff) |
965 ((params->y1 - 1) << 16));
973 gen6_blorp_emit_primitive(struct brw_context *brw,
974 const brw_blorp_params *params)
976 struct intel_context *intel = &brw->intel;
979 OUT_BATCH(CMD_3D_PRIM << 16 | (6 - 2) |
980 _3DPRIM_RECTLIST << GEN4_3DPRIM_TOPOLOGY_TYPE_SHIFT |
981 GEN4_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL);
982 OUT_BATCH(3); /* vertex count per instance */
984 OUT_BATCH(1); /* instance count */
992 * \brief Execute a blit or render pass operation.
994 * To execute the operation, this function manually constructs and emits a
995 * batch to draw a rectangle primitive. The batchbuffer is flushed before
996 * constructing and after emitting the batch.
998 * This function alters no GL state.
1001 gen6_blorp_exec(struct intel_context *intel,
1002 const brw_blorp_params *params)
1004 struct gl_context *ctx = &intel->ctx;
1005 struct brw_context *brw = brw_context(ctx);
1006 brw_blorp_prog_data *prog_data = NULL;
1007 uint32_t cc_blend_state_offset = 0;
1008 uint32_t cc_state_offset = 0;
1009 uint32_t depthstencil_offset;
1010 uint32_t wm_push_const_offset = 0;
1011 uint32_t wm_bind_bo_offset = 0;
1013 uint32_t prog_offset = params->get_wm_prog(brw, &prog_data);
1014 gen6_blorp_emit_batch_head(brw, params);
1015 gen6_blorp_emit_vertices(brw, params);
1016 gen6_blorp_emit_urb_config(brw, params);
1017 if (params->use_wm_prog) {
1018 cc_blend_state_offset = gen6_blorp_emit_blend_state(brw, params);
1019 cc_state_offset = gen6_blorp_emit_cc_state(brw, params);
1021 depthstencil_offset = gen6_blorp_emit_depth_stencil_state(brw, params);
1022 gen6_blorp_emit_cc_state_pointers(brw, params, cc_blend_state_offset,
1023 depthstencil_offset, cc_state_offset);
1024 if (params->use_wm_prog) {
1025 uint32_t wm_surf_offset_renderbuffer;
1026 uint32_t wm_surf_offset_texture;
1027 uint32_t sampler_offset;
1028 wm_push_const_offset = gen6_blorp_emit_wm_constants(brw, params);
1029 wm_surf_offset_renderbuffer =
1030 gen6_blorp_emit_surface_state(brw, params, ¶ms->dst,
1031 I915_GEM_DOMAIN_RENDER,
1032 I915_GEM_DOMAIN_RENDER);
1033 wm_surf_offset_texture =
1034 gen6_blorp_emit_surface_state(brw, params, ¶ms->src,
1035 I915_GEM_DOMAIN_SAMPLER, 0);
1037 gen6_blorp_emit_binding_table(brw, params,
1038 wm_surf_offset_renderbuffer,
1039 wm_surf_offset_texture);
1040 sampler_offset = gen6_blorp_emit_sampler_state(brw, params);
1041 gen6_blorp_emit_sampler_state_pointers(brw, params, sampler_offset);
1043 gen6_blorp_emit_vs_disable(brw, params);
1044 gen6_blorp_emit_gs_disable(brw, params);
1045 gen6_blorp_emit_clip_disable(brw, params);
1046 gen6_blorp_emit_sf_config(brw, params);
1047 if (params->use_wm_prog)
1048 gen6_blorp_emit_constant_ps(brw, params, wm_push_const_offset);
1049 gen6_blorp_emit_wm_config(brw, params, prog_offset, prog_data);
1050 if (params->use_wm_prog)
1051 gen6_blorp_emit_binding_table_pointers(brw, params, wm_bind_bo_offset);
1053 if (params->depth.mt)
1054 gen6_blorp_emit_depth_stencil_config(brw, params);
1056 gen6_blorp_emit_depth_disable(brw, params);
1057 gen6_blorp_emit_clear_params(brw, params);
1058 gen6_blorp_emit_drawing_rectangle(brw, params);
1059 gen6_blorp_emit_primitive(brw, params);
1061 /* See comments above at first invocation of intel_flush() in
1062 * gen6_blorp_emit_batch_head().
1067 brw->state.dirty.brw = ~0;
1068 brw->state.dirty.cache = ~0;
1071 /** \see intel_context::vtbl::resolve_hiz_slice */
1073 gen6_resolve_hiz_slice(struct intel_context *intel,
1074 struct intel_mipmap_tree *mt,
1078 brw_hiz_op_params params(mt, level, layer, GEN6_HIZ_OP_HIZ_RESOLVE);
1079 gen6_blorp_exec(intel, ¶ms);
1082 /** \see intel_context::vtbl::resolve_depth_slice */
1084 gen6_resolve_depth_slice(struct intel_context *intel,
1085 struct intel_mipmap_tree *mt,
1089 brw_hiz_op_params params(mt, level, layer, GEN6_HIZ_OP_DEPTH_RESOLVE);
1090 gen6_blorp_exec(intel, ¶ms);