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 /* When blitting from an XRGB source to a ARGB destination, we need to
287 * interpret the missing channel as 1.0. Blending can do that for us:
288 * we simply use the RGB values from the fragment shader ("source RGB"),
289 * but smash the alpha channel to 1.
291 if (_mesa_get_format_bits(params->dst.mt->format, GL_ALPHA_BITS) > 0 &&
292 _mesa_get_format_bits(params->src.mt->format, GL_ALPHA_BITS) == 0) {
293 blend->blend0.blend_enable = 1;
294 blend->blend0.ia_blend_enable = 1;
296 blend->blend0.blend_func = BRW_BLENDFUNCTION_ADD;
297 blend->blend0.ia_blend_func = BRW_BLENDFUNCTION_ADD;
299 blend->blend0.source_blend_factor = BRW_BLENDFACTOR_SRC_COLOR;
300 blend->blend0.dest_blend_factor = BRW_BLENDFACTOR_ZERO;
301 blend->blend0.ia_source_blend_factor = BRW_BLENDFACTOR_ONE;
302 blend->blend0.ia_dest_blend_factor = BRW_BLENDFACTOR_ZERO;
305 return cc_blend_state_offset;
311 gen6_blorp_emit_cc_state(struct brw_context *brw,
312 const brw_blorp_params *params)
314 uint32_t cc_state_offset;
316 struct gen6_color_calc_state *cc = (struct gen6_color_calc_state *)
317 brw_state_batch(brw, AUB_TRACE_CC_STATE,
318 sizeof(gen6_color_calc_state), 64,
320 memset(cc, 0, sizeof(*cc));
322 return cc_state_offset;
327 * \param out_offset is relative to
328 * CMD_STATE_BASE_ADDRESS.DynamicStateBaseAddress.
331 gen6_blorp_emit_depth_stencil_state(struct brw_context *brw,
332 const brw_blorp_params *params)
334 uint32_t depthstencil_offset;
336 struct gen6_depth_stencil_state *state;
337 state = (struct gen6_depth_stencil_state *)
338 brw_state_batch(brw, AUB_TRACE_DEPTH_STENCIL_STATE,
340 &depthstencil_offset);
341 memset(state, 0, sizeof(*state));
343 /* See the following sections of the Sandy Bridge PRM, Volume 1, Part2:
344 * - 7.5.3.1 Depth Buffer Clear
345 * - 7.5.3.2 Depth Buffer Resolve
346 * - 7.5.3.3 Hierarchical Depth Buffer Resolve
348 state->ds2.depth_write_enable = 1;
349 if (params->hiz_op == GEN6_HIZ_OP_DEPTH_RESOLVE) {
350 state->ds2.depth_test_enable = 1;
351 state->ds2.depth_test_func = COMPAREFUNC_NEVER;
354 return depthstencil_offset;
358 /* 3DSTATE_CC_STATE_POINTERS
360 * The pointer offsets are relative to
361 * CMD_STATE_BASE_ADDRESS.DynamicStateBaseAddress.
363 * The HiZ op doesn't use BLEND_STATE or COLOR_CALC_STATE.
366 gen6_blorp_emit_cc_state_pointers(struct brw_context *brw,
367 const brw_blorp_params *params,
368 uint32_t cc_blend_state_offset,
369 uint32_t depthstencil_offset,
370 uint32_t cc_state_offset)
372 struct intel_context *intel = &brw->intel;
375 OUT_BATCH(_3DSTATE_CC_STATE_POINTERS << 16 | (4 - 2));
376 OUT_BATCH(cc_blend_state_offset | 1); /* BLEND_STATE offset */
377 OUT_BATCH(depthstencil_offset | 1); /* DEPTH_STENCIL_STATE offset */
378 OUT_BATCH(cc_state_offset | 1); /* COLOR_CALC_STATE offset */
383 /* WM push constants */
385 gen6_blorp_emit_wm_constants(struct brw_context *brw,
386 const brw_blorp_params *params)
388 uint32_t wm_push_const_offset;
390 void *constants = brw_state_batch(brw, AUB_TRACE_WM_CONSTANTS,
391 sizeof(params->wm_push_consts),
392 32, &wm_push_const_offset);
393 memcpy(constants, ¶ms->wm_push_consts,
394 sizeof(params->wm_push_consts));
396 return wm_push_const_offset;
400 /* SURFACE_STATE for renderbuffer or texture surface (see
401 * brw_update_renderbuffer_surface and brw_update_texture_surface)
404 gen6_blorp_emit_surface_state(struct brw_context *brw,
405 const brw_blorp_params *params,
406 const brw_blorp_surface_info *surface,
407 uint32_t read_domains, uint32_t write_domain)
409 uint32_t wm_surf_offset;
410 uint32_t width = surface->width;
411 uint32_t height = surface->height;
412 if (surface->num_samples > 1) {
413 /* Since gen6 uses INTEL_MSAA_LAYOUT_IMS, width and height are measured
414 * in samples. But SURFACE_STATE wants them in pixels, so we need to
415 * divide them each by 2.
420 struct intel_region *region = surface->mt->region;
421 uint32_t tile_x, tile_y;
423 uint32_t *surf = (uint32_t *)
424 brw_state_batch(brw, AUB_TRACE_SURFACE_STATE, 6 * 4, 32,
427 surf[0] = (BRW_SURFACE_2D << BRW_SURFACE_TYPE_SHIFT |
428 BRW_SURFACE_MIPMAPLAYOUT_BELOW << BRW_SURFACE_MIPLAYOUT_SHIFT |
429 BRW_SURFACE_CUBEFACE_ENABLES |
430 surface->brw_surfaceformat << BRW_SURFACE_FORMAT_SHIFT);
433 surf[1] = (surface->compute_tile_offsets(&tile_x, &tile_y) +
436 surf[2] = (0 << BRW_SURFACE_LOD_SHIFT |
437 (width - 1) << BRW_SURFACE_WIDTH_SHIFT |
438 (height - 1) << BRW_SURFACE_HEIGHT_SHIFT);
440 uint32_t tiling = surface->map_stencil_as_y_tiled
441 ? BRW_SURFACE_TILED | BRW_SURFACE_TILED_Y
442 : brw_get_surface_tiling_bits(region->tiling);
443 uint32_t pitch_bytes = region->pitch;
444 if (surface->map_stencil_as_y_tiled)
447 0 << BRW_SURFACE_DEPTH_SHIFT |
448 (pitch_bytes - 1) << BRW_SURFACE_PITCH_SHIFT);
450 surf[4] = brw_get_surface_num_multisamples(surface->num_samples);
452 /* Note that the low bits of these fields are missing, so
453 * there's the possibility of getting in trouble.
455 assert(tile_x % 4 == 0);
456 assert(tile_y % 2 == 0);
457 surf[5] = ((tile_x / 4) << BRW_SURFACE_X_OFFSET_SHIFT |
458 (tile_y / 2) << BRW_SURFACE_Y_OFFSET_SHIFT |
459 (surface->mt->align_h == 4 ?
460 BRW_SURFACE_VERTICAL_ALIGN_ENABLE : 0));
462 /* Emit relocation to surface contents */
463 drm_intel_bo_emit_reloc(brw->intel.batch.bo,
466 surf[1] - region->bo->offset,
467 read_domains, write_domain);
469 return wm_surf_offset;
473 /* BINDING_TABLE. See brw_wm_binding_table(). */
475 gen6_blorp_emit_binding_table(struct brw_context *brw,
476 const brw_blorp_params *params,
477 uint32_t wm_surf_offset_renderbuffer,
478 uint32_t wm_surf_offset_texture)
480 uint32_t wm_bind_bo_offset;
481 uint32_t *bind = (uint32_t *)
482 brw_state_batch(brw, AUB_TRACE_BINDING_TABLE,
484 BRW_BLORP_NUM_BINDING_TABLE_ENTRIES,
487 bind[BRW_BLORP_RENDERBUFFER_BINDING_TABLE_INDEX] =
488 wm_surf_offset_renderbuffer;
489 bind[BRW_BLORP_TEXTURE_BINDING_TABLE_INDEX] = wm_surf_offset_texture;
491 return wm_bind_bo_offset;
496 * SAMPLER_STATE. See brw_update_sampler_state().
499 gen6_blorp_emit_sampler_state(struct brw_context *brw,
500 const brw_blorp_params *params)
502 uint32_t sampler_offset;
504 struct brw_sampler_state *sampler = (struct brw_sampler_state *)
505 brw_state_batch(brw, AUB_TRACE_SAMPLER_STATE,
506 sizeof(struct brw_sampler_state),
507 32, &sampler_offset);
508 memset(sampler, 0, sizeof(*sampler));
510 sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
511 sampler->ss0.mip_filter = BRW_MIPFILTER_NONE;
512 sampler->ss0.mag_filter = BRW_MAPFILTER_LINEAR;
514 sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
515 sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
516 sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
518 sampler->ss0.min_mag_neq = 1;
522 sampler->ss0.lod_bias = 0;
524 sampler->ss0.lod_preclamp = 1; /* OpenGL mode */
525 sampler->ss0.default_color_mode = 0; /* OpenGL/DX10 mode */
527 /* Set BaseMipLevel, MaxLOD, MinLOD:
529 * XXX: I don't think that using firstLevel, lastLevel works,
530 * because we always setup the surface state as if firstLevel ==
531 * level zero. Probably have to subtract firstLevel from each of
534 sampler->ss0.base_level = U_FIXED(0, 1);
536 sampler->ss1.max_lod = U_FIXED(0, 6);
537 sampler->ss1.min_lod = U_FIXED(0, 6);
539 sampler->ss3.non_normalized_coord = 1;
541 sampler->ss3.address_round |= BRW_ADDRESS_ROUNDING_ENABLE_U_MIN |
542 BRW_ADDRESS_ROUNDING_ENABLE_V_MIN |
543 BRW_ADDRESS_ROUNDING_ENABLE_R_MIN;
544 sampler->ss3.address_round |= BRW_ADDRESS_ROUNDING_ENABLE_U_MAG |
545 BRW_ADDRESS_ROUNDING_ENABLE_V_MAG |
546 BRW_ADDRESS_ROUNDING_ENABLE_R_MAG;
548 return sampler_offset;
553 * 3DSTATE_SAMPLER_STATE_POINTERS. See upload_sampler_state_pointers().
556 gen6_blorp_emit_sampler_state_pointers(struct brw_context *brw,
557 const brw_blorp_params *params,
558 uint32_t sampler_offset)
560 struct intel_context *intel = &brw->intel;
563 OUT_BATCH(_3DSTATE_SAMPLER_STATE_POINTERS << 16 |
564 VS_SAMPLER_STATE_CHANGE |
565 GS_SAMPLER_STATE_CHANGE |
566 PS_SAMPLER_STATE_CHANGE |
568 OUT_BATCH(0); /* VS */
569 OUT_BATCH(0); /* GS */
570 OUT_BATCH(sampler_offset);
577 * Disable vertex shader.
580 gen6_blorp_emit_vs_disable(struct brw_context *brw,
581 const brw_blorp_params *params)
583 struct intel_context *intel = &brw->intel;
585 if (intel->gen == 6) {
586 /* From the BSpec, Volume 2a, Part 3 "Vertex Shader", Section
587 * 3DSTATE_VS, Dword 5.0 "VS Function Enable":
589 * [DevSNB] A pipeline flush must be programmed prior to a
590 * 3DSTATE_VS command that causes the VS Function Enable to
591 * toggle. Pipeline flush can be executed by sending a PIPE_CONTROL
592 * command with CS stall bit set and a post sync operation.
594 intel_emit_post_sync_nonzero_flush(intel);
598 OUT_BATCH(_3DSTATE_VS << 16 | (6 - 2));
610 * Disable the geometry shader.
613 gen6_blorp_emit_gs_disable(struct brw_context *brw,
614 const brw_blorp_params *params)
616 struct intel_context *intel = &brw->intel;
619 OUT_BATCH(_3DSTATE_GS << 16 | (7 - 2));
632 * Disable the clipper.
634 * The BLORP op emits a rectangle primitive, which requires clipping to
635 * be disabled. From page 10 of the Sandy Bridge PRM Volume 2 Part 1
636 * Section 1.3 "3D Primitives Overview":
638 * Either the CLIP unit should be DISABLED, or the CLIP unit's Clip
639 * Mode should be set to a value other than CLIPMODE_NORMAL.
641 * Also disable perspective divide. This doesn't change the clipper's
642 * output, but does spare a few electrons.
645 gen6_blorp_emit_clip_disable(struct brw_context *brw,
646 const brw_blorp_params *params)
648 struct intel_context *intel = &brw->intel;
651 OUT_BATCH(_3DSTATE_CLIP << 16 | (4 - 2));
653 OUT_BATCH(GEN6_CLIP_PERSPECTIVE_DIVIDE_DISABLE);
661 * Disable ViewportTransformEnable (dw2.1)
663 * From the SandyBridge PRM, Volume 2, Part 1, Section 1.3, "3D
664 * Primitives Overview":
665 * RECTLIST: Viewport Mapping must be DISABLED (as is typical with the
666 * use of screen- space coordinates).
668 * A solid rectangle must be rendered, so set FrontFaceFillMode (dw2.4:3)
669 * and BackFaceFillMode (dw2.5:6) to SOLID(0).
671 * From the Sandy Bridge PRM, Volume 2, Part 1, Section
672 * 6.4.1.1 3DSTATE_SF, Field FrontFaceFillMode:
673 * SOLID: Any triangle or rectangle object found to be front-facing
674 * is rendered as a solid object. This setting is required when
675 * (rendering rectangle (RECTLIST) objects.
678 gen6_blorp_emit_sf_config(struct brw_context *brw,
679 const brw_blorp_params *params)
681 struct intel_context *intel = &brw->intel;
684 OUT_BATCH(_3DSTATE_SF << 16 | (20 - 2));
685 OUT_BATCH((1 - 1) << GEN6_SF_NUM_OUTPUTS_SHIFT | /* only position */
686 1 << GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT |
687 0 << GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT);
688 OUT_BATCH(0); /* dw2 */
689 OUT_BATCH(params->num_samples > 1 ? GEN6_SF_MSRAST_ON_PATTERN : 0);
690 for (int i = 0; i < 16; ++i)
697 * Enable or disable thread dispatch and set the HiZ op appropriately.
700 gen6_blorp_emit_wm_config(struct brw_context *brw,
701 const brw_blorp_params *params,
702 uint32_t prog_offset,
703 brw_blorp_prog_data *prog_data)
705 struct intel_context *intel = &brw->intel;
706 uint32_t dw2, dw4, dw5, dw6;
708 /* Even when thread dispatch is disabled, max threads (dw5.25:31) must be
709 * nonzero to prevent the GPU from hanging. See the valid ranges in the
710 * BSpec, Volume 2a.11 Windower, Section 3DSTATE_WM, Dword 5.25:31
711 * "Maximum Number Of Threads".
713 * To be safe (and to minimize extraneous code) we go ahead and fully
714 * configure the WM state whether or not there is a WM program.
717 dw2 = dw4 = dw5 = dw6 = 0;
718 switch (params->hiz_op) {
719 case GEN6_HIZ_OP_DEPTH_CLEAR:
720 dw4 |= GEN6_WM_DEPTH_CLEAR;
722 case GEN6_HIZ_OP_DEPTH_RESOLVE:
723 dw4 |= GEN6_WM_DEPTH_RESOLVE;
725 case GEN6_HIZ_OP_HIZ_RESOLVE:
726 dw4 |= GEN6_WM_HIERARCHICAL_DEPTH_RESOLVE;
728 case GEN6_HIZ_OP_NONE:
734 dw4 |= GEN6_WM_STATISTICS_ENABLE;
735 dw5 |= GEN6_WM_LINE_AA_WIDTH_1_0;
736 dw5 |= GEN6_WM_LINE_END_CAP_AA_WIDTH_0_5;
737 dw5 |= (brw->max_wm_threads - 1) << GEN6_WM_MAX_THREADS_SHIFT;
738 dw6 |= 0 << GEN6_WM_BARYCENTRIC_INTERPOLATION_MODE_SHIFT; /* No interp */
739 dw6 |= 0 << GEN6_WM_NUM_SF_OUTPUTS_SHIFT; /* No inputs from SF */
740 if (params->use_wm_prog) {
741 dw2 |= 1 << GEN6_WM_SAMPLER_COUNT_SHIFT; /* Up to 4 samplers */
742 dw4 |= prog_data->first_curbe_grf << GEN6_WM_DISPATCH_START_GRF_SHIFT_0;
743 dw5 |= GEN6_WM_16_DISPATCH_ENABLE;
744 dw5 |= GEN6_WM_KILL_ENABLE; /* TODO: temporarily smash on */
745 dw5 |= GEN6_WM_DISPATCH_ENABLE; /* We are rendering */
748 if (params->num_samples > 1) {
749 dw6 |= GEN6_WM_MSRAST_ON_PATTERN;
750 if (prog_data && prog_data->persample_msaa_dispatch)
751 dw6 |= GEN6_WM_MSDISPMODE_PERSAMPLE;
753 dw6 |= GEN6_WM_MSDISPMODE_PERPIXEL;
755 dw6 |= GEN6_WM_MSRAST_OFF_PIXEL;
756 dw6 |= GEN6_WM_MSDISPMODE_PERSAMPLE;
760 OUT_BATCH(_3DSTATE_WM << 16 | (9 - 2));
761 OUT_BATCH(params->use_wm_prog ? prog_offset : 0);
763 OUT_BATCH(0); /* No scratch needed */
767 OUT_BATCH(0); /* No other programs */
768 OUT_BATCH(0); /* No other programs */
774 gen6_blorp_emit_constant_ps(struct brw_context *brw,
775 const brw_blorp_params *params,
776 uint32_t wm_push_const_offset)
778 struct intel_context *intel = &brw->intel;
780 /* Make sure the push constants fill an exact integer number of
783 assert(sizeof(brw_blorp_wm_push_constants) % 32 == 0);
785 /* There must be at least one register worth of push constant data. */
786 assert(BRW_BLORP_NUM_PUSH_CONST_REGS > 0);
788 /* Enable push constant buffer 0. */
790 OUT_BATCH(_3DSTATE_CONSTANT_PS << 16 |
791 GEN6_CONSTANT_BUFFER_0_ENABLE |
793 OUT_BATCH(wm_push_const_offset + (BRW_BLORP_NUM_PUSH_CONST_REGS - 1));
802 * 3DSTATE_BINDING_TABLE_POINTERS
805 gen6_blorp_emit_binding_table_pointers(struct brw_context *brw,
806 const brw_blorp_params *params,
807 uint32_t wm_bind_bo_offset)
809 struct intel_context *intel = &brw->intel;
812 OUT_BATCH(_3DSTATE_BINDING_TABLE_POINTERS << 16 |
813 GEN6_BINDING_TABLE_MODIFY_PS |
815 OUT_BATCH(0); /* vs -- ignored */
816 OUT_BATCH(0); /* gs -- ignored */
817 OUT_BATCH(wm_bind_bo_offset); /* wm/ps */
823 gen6_blorp_emit_depth_stencil_config(struct brw_context *brw,
824 const brw_blorp_params *params)
826 struct intel_context *intel = &brw->intel;
827 uint32_t draw_x = params->depth.x_offset;
828 uint32_t draw_y = params->depth.y_offset;
829 uint32_t tile_mask_x, tile_mask_y;
831 brw_get_depthstencil_tile_masks(params->depth.mt, NULL,
832 &tile_mask_x, &tile_mask_y);
834 /* 3DSTATE_DEPTH_BUFFER */
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, false);
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.
858 WARN_ONCE((tile_x & 7) || (tile_y & 7),
859 "Depth/stencil buffer needs alignment to 8-pixel boundaries.\n"
860 "Truncating offset, bad rendering may occur.\n");
864 intel_emit_post_sync_nonzero_flush(intel);
865 intel_emit_depth_stall_flushes(intel);
868 OUT_BATCH(_3DSTATE_DEPTH_BUFFER << 16 | (7 - 2));
869 OUT_BATCH((params->depth.mt->region->pitch - 1) |
870 params->depth_format << 18 |
871 1 << 21 | /* separate stencil enable */
872 1 << 22 | /* hiz enable */
873 BRW_TILEWALK_YMAJOR << 26 |
874 1 << 27 | /* y-tiled */
875 BRW_SURFACE_2D << 29);
876 OUT_RELOC(params->depth.mt->region->bo,
877 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
879 OUT_BATCH(BRW_SURFACE_MIPMAPLAYOUT_BELOW << 1 |
880 (params->depth.width + tile_x - 1) << 6 |
881 (params->depth.height + tile_y - 1) << 19);
889 /* 3DSTATE_HIER_DEPTH_BUFFER */
891 struct intel_region *hiz_region = params->depth.mt->hiz_mt->region;
892 uint32_t hiz_offset =
893 intel_region_get_aligned_offset(hiz_region,
894 draw_x & ~tile_mask_x,
895 (draw_y & ~tile_mask_y) / 2, false);
898 OUT_BATCH((_3DSTATE_HIER_DEPTH_BUFFER << 16) | (3 - 2));
899 OUT_BATCH(hiz_region->pitch - 1);
900 OUT_RELOC(hiz_region->bo,
901 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
906 /* 3DSTATE_STENCIL_BUFFER */
909 OUT_BATCH((_3DSTATE_STENCIL_BUFFER << 16) | (3 - 2));
918 gen6_blorp_emit_depth_disable(struct brw_context *brw,
919 const brw_blorp_params *params)
921 struct intel_context *intel = &brw->intel;
924 OUT_BATCH(_3DSTATE_DEPTH_BUFFER << 16 | (7 - 2));
925 OUT_BATCH((BRW_DEPTHFORMAT_D32_FLOAT << 18) |
926 (BRW_SURFACE_NULL << 29));
936 /* 3DSTATE_CLEAR_PARAMS
938 * From the Sandybridge PRM, Volume 2, Part 1, Section 3DSTATE_CLEAR_PARAMS:
939 * [DevSNB] 3DSTATE_CLEAR_PARAMS packet must follow the DEPTH_BUFFER_STATE
940 * packet when HiZ is enabled and the DEPTH_BUFFER_STATE changes.
943 gen6_blorp_emit_clear_params(struct brw_context *brw,
944 const brw_blorp_params *params)
946 struct intel_context *intel = &brw->intel;
949 OUT_BATCH(_3DSTATE_CLEAR_PARAMS << 16 |
950 GEN5_DEPTH_CLEAR_VALID |
952 OUT_BATCH(params->depth.mt ? params->depth.mt->depth_clear_value : 0);
957 /* 3DSTATE_DRAWING_RECTANGLE */
959 gen6_blorp_emit_drawing_rectangle(struct brw_context *brw,
960 const brw_blorp_params *params)
962 struct intel_context *intel = &brw->intel;
965 OUT_BATCH(_3DSTATE_DRAWING_RECTANGLE << 16 | (4 - 2));
967 OUT_BATCH(((params->x1 - 1) & 0xffff) |
968 ((params->y1 - 1) << 16));
973 /* 3DSTATE_VIEWPORT_STATE_POINTERS */
975 gen6_blorp_emit_viewport_state(struct brw_context *brw,
976 const brw_blorp_params *params)
978 struct intel_context *intel = &brw->intel;
979 struct brw_cc_viewport *ccv;
980 uint32_t cc_vp_offset;
982 ccv = (struct brw_cc_viewport *)brw_state_batch(brw, AUB_TRACE_CC_VP_STATE,
986 ccv->min_depth = 0.0;
987 ccv->max_depth = 1.0;
990 OUT_BATCH(_3DSTATE_VIEWPORT_STATE_POINTERS << 16 | (4 - 2) |
991 GEN6_CC_VIEWPORT_MODIFY);
992 OUT_BATCH(0); /* clip VP */
993 OUT_BATCH(0); /* SF VP */
994 OUT_BATCH(cc_vp_offset);
1001 gen6_blorp_emit_primitive(struct brw_context *brw,
1002 const brw_blorp_params *params)
1004 struct intel_context *intel = &brw->intel;
1007 OUT_BATCH(CMD_3D_PRIM << 16 | (6 - 2) |
1008 _3DPRIM_RECTLIST << GEN4_3DPRIM_TOPOLOGY_TYPE_SHIFT |
1009 GEN4_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL);
1010 OUT_BATCH(3); /* vertex count per instance */
1012 OUT_BATCH(1); /* instance count */
1020 * \brief Execute a blit or render pass operation.
1022 * To execute the operation, this function manually constructs and emits a
1023 * batch to draw a rectangle primitive. The batchbuffer is flushed before
1024 * constructing and after emitting the batch.
1026 * This function alters no GL state.
1029 gen6_blorp_exec(struct intel_context *intel,
1030 const brw_blorp_params *params)
1032 struct gl_context *ctx = &intel->ctx;
1033 struct brw_context *brw = brw_context(ctx);
1034 brw_blorp_prog_data *prog_data = NULL;
1035 uint32_t cc_blend_state_offset = 0;
1036 uint32_t cc_state_offset = 0;
1037 uint32_t depthstencil_offset;
1038 uint32_t wm_push_const_offset = 0;
1039 uint32_t wm_bind_bo_offset = 0;
1041 uint32_t prog_offset = params->get_wm_prog(brw, &prog_data);
1042 gen6_blorp_emit_batch_head(brw, params);
1043 gen6_emit_3dstate_multisample(brw, params->num_samples);
1044 gen6_emit_3dstate_sample_mask(brw, params->num_samples, 1.0, false);
1045 gen6_blorp_emit_state_base_address(brw, params);
1046 gen6_blorp_emit_vertices(brw, params);
1047 gen6_blorp_emit_urb_config(brw, params);
1048 if (params->use_wm_prog) {
1049 cc_blend_state_offset = gen6_blorp_emit_blend_state(brw, params);
1050 cc_state_offset = gen6_blorp_emit_cc_state(brw, params);
1052 depthstencil_offset = gen6_blorp_emit_depth_stencil_state(brw, params);
1053 gen6_blorp_emit_cc_state_pointers(brw, params, cc_blend_state_offset,
1054 depthstencil_offset, cc_state_offset);
1055 if (params->use_wm_prog) {
1056 uint32_t wm_surf_offset_renderbuffer;
1057 uint32_t wm_surf_offset_texture;
1058 uint32_t sampler_offset;
1059 wm_push_const_offset = gen6_blorp_emit_wm_constants(brw, params);
1060 wm_surf_offset_renderbuffer =
1061 gen6_blorp_emit_surface_state(brw, params, ¶ms->dst,
1062 I915_GEM_DOMAIN_RENDER,
1063 I915_GEM_DOMAIN_RENDER);
1064 wm_surf_offset_texture =
1065 gen6_blorp_emit_surface_state(brw, params, ¶ms->src,
1066 I915_GEM_DOMAIN_SAMPLER, 0);
1068 gen6_blorp_emit_binding_table(brw, params,
1069 wm_surf_offset_renderbuffer,
1070 wm_surf_offset_texture);
1071 sampler_offset = gen6_blorp_emit_sampler_state(brw, params);
1072 gen6_blorp_emit_sampler_state_pointers(brw, params, sampler_offset);
1074 gen6_blorp_emit_vs_disable(brw, params);
1075 gen6_blorp_emit_gs_disable(brw, params);
1076 gen6_blorp_emit_clip_disable(brw, params);
1077 gen6_blorp_emit_sf_config(brw, params);
1078 if (params->use_wm_prog)
1079 gen6_blorp_emit_constant_ps(brw, params, wm_push_const_offset);
1080 gen6_blorp_emit_wm_config(brw, params, prog_offset, prog_data);
1081 if (params->use_wm_prog)
1082 gen6_blorp_emit_binding_table_pointers(brw, params, wm_bind_bo_offset);
1083 gen6_blorp_emit_viewport_state(brw, params);
1085 if (params->depth.mt)
1086 gen6_blorp_emit_depth_stencil_config(brw, params);
1088 gen6_blorp_emit_depth_disable(brw, params);
1089 gen6_blorp_emit_clear_params(brw, params);
1090 gen6_blorp_emit_drawing_rectangle(brw, params);
1091 gen6_blorp_emit_primitive(brw, params);
1093 /* See comments above at first invocation of intel_flush() in
1094 * gen6_blorp_emit_batch_head().
1099 brw->state.dirty.brw = ~0;
1100 brw->state.dirty.cache = ~0;