anv/gen8: Stall when needed in Cmd(Set|Reset)Event
[mesa.git] / src / intel / vulkan / gen8_cmd_buffer.c
1 /*
2 * Copyright © 2015 Intel Corporation
3 *
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:
10 *
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
13 * Software.
14 *
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
21 * IN THE SOFTWARE.
22 */
23
24 #include <assert.h>
25 #include <stdbool.h>
26 #include <string.h>
27 #include <unistd.h>
28 #include <fcntl.h>
29
30 #include "anv_private.h"
31
32 #include "genxml/gen_macros.h"
33 #include "genxml/genX_pack.h"
34
35 #if GEN_GEN == 8
36 void
37 gen8_cmd_buffer_emit_viewport(struct anv_cmd_buffer *cmd_buffer)
38 {
39 uint32_t count = cmd_buffer->state.dynamic.viewport.count;
40 const VkViewport *viewports = cmd_buffer->state.dynamic.viewport.viewports;
41 struct anv_state sf_clip_state =
42 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, count * 64, 64);
43
44 for (uint32_t i = 0; i < count; i++) {
45 const VkViewport *vp = &viewports[i];
46
47 /* The gen7 state struct has just the matrix and guardband fields, the
48 * gen8 struct adds the min/max viewport fields. */
49 struct GENX(SF_CLIP_VIEWPORT) sf_clip_viewport = {
50 .ViewportMatrixElementm00 = vp->width / 2,
51 .ViewportMatrixElementm11 = vp->height / 2,
52 .ViewportMatrixElementm22 = 1.0,
53 .ViewportMatrixElementm30 = vp->x + vp->width / 2,
54 .ViewportMatrixElementm31 = vp->y + vp->height / 2,
55 .ViewportMatrixElementm32 = 0.0,
56 .XMinClipGuardband = -1.0f,
57 .XMaxClipGuardband = 1.0f,
58 .YMinClipGuardband = -1.0f,
59 .YMaxClipGuardband = 1.0f,
60 .XMinViewPort = vp->x,
61 .XMaxViewPort = vp->x + vp->width - 1,
62 .YMinViewPort = vp->y,
63 .YMaxViewPort = vp->y + vp->height - 1,
64 };
65
66 GENX(SF_CLIP_VIEWPORT_pack)(NULL, sf_clip_state.map + i * 64,
67 &sf_clip_viewport);
68 }
69
70 if (!cmd_buffer->device->info.has_llc)
71 anv_state_clflush(sf_clip_state);
72
73 anv_batch_emit(&cmd_buffer->batch,
74 GENX(3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP), clip) {
75 clip.SFClipViewportPointer = sf_clip_state.offset;
76 }
77 }
78
79 void
80 gen8_cmd_buffer_emit_depth_viewport(struct anv_cmd_buffer *cmd_buffer,
81 bool depth_clamp_enable)
82 {
83 uint32_t count = cmd_buffer->state.dynamic.viewport.count;
84 const VkViewport *viewports = cmd_buffer->state.dynamic.viewport.viewports;
85 struct anv_state cc_state =
86 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, count * 8, 32);
87
88 for (uint32_t i = 0; i < count; i++) {
89 const VkViewport *vp = &viewports[i];
90
91 struct GENX(CC_VIEWPORT) cc_viewport = {
92 .MinimumDepth = depth_clamp_enable ? vp->minDepth : 0.0f,
93 .MaximumDepth = depth_clamp_enable ? vp->maxDepth : 1.0f,
94 };
95
96 GENX(CC_VIEWPORT_pack)(NULL, cc_state.map + i * 8, &cc_viewport);
97 }
98
99 if (!cmd_buffer->device->info.has_llc)
100 anv_state_clflush(cc_state);
101
102 anv_batch_emit(&cmd_buffer->batch,
103 GENX(3DSTATE_VIEWPORT_STATE_POINTERS_CC), cc) {
104 cc.CCViewportPointer = cc_state.offset;
105 }
106 }
107 #endif
108
109 static void
110 __emit_genx_sf_state(struct anv_cmd_buffer *cmd_buffer)
111 {
112 uint32_t sf_dw[GENX(3DSTATE_SF_length)];
113 struct GENX(3DSTATE_SF) sf = {
114 GENX(3DSTATE_SF_header),
115 .LineWidth = cmd_buffer->state.dynamic.line_width,
116 };
117 GENX(3DSTATE_SF_pack)(NULL, sf_dw, &sf);
118 /* FIXME: gen9.fs */
119 anv_batch_emit_merge(&cmd_buffer->batch, sf_dw,
120 cmd_buffer->state.pipeline->gen8.sf);
121 }
122
123 #include "genxml/gen9_pack.h"
124 static void
125 __emit_gen9_sf_state(struct anv_cmd_buffer *cmd_buffer)
126 {
127 uint32_t sf_dw[GENX(3DSTATE_SF_length)];
128 struct GEN9_3DSTATE_SF sf = {
129 GEN9_3DSTATE_SF_header,
130 .LineWidth = cmd_buffer->state.dynamic.line_width,
131 };
132 GEN9_3DSTATE_SF_pack(NULL, sf_dw, &sf);
133 /* FIXME: gen9.fs */
134 anv_batch_emit_merge(&cmd_buffer->batch, sf_dw,
135 cmd_buffer->state.pipeline->gen8.sf);
136 }
137
138 static void
139 __emit_sf_state(struct anv_cmd_buffer *cmd_buffer)
140 {
141 if (cmd_buffer->device->info.is_cherryview)
142 __emit_gen9_sf_state(cmd_buffer);
143 else
144 __emit_genx_sf_state(cmd_buffer);
145 }
146
147 void
148 genX(cmd_buffer_flush_dynamic_state)(struct anv_cmd_buffer *cmd_buffer)
149 {
150 struct anv_pipeline *pipeline = cmd_buffer->state.pipeline;
151
152 if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE |
153 ANV_CMD_DIRTY_DYNAMIC_LINE_WIDTH)) {
154 __emit_sf_state(cmd_buffer);
155 }
156
157 if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE |
158 ANV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS)){
159 uint32_t raster_dw[GENX(3DSTATE_RASTER_length)];
160 struct GENX(3DSTATE_RASTER) raster = {
161 GENX(3DSTATE_RASTER_header),
162 .GlobalDepthOffsetConstant = cmd_buffer->state.dynamic.depth_bias.bias,
163 .GlobalDepthOffsetScale = cmd_buffer->state.dynamic.depth_bias.slope,
164 .GlobalDepthOffsetClamp = cmd_buffer->state.dynamic.depth_bias.clamp
165 };
166 GENX(3DSTATE_RASTER_pack)(NULL, raster_dw, &raster);
167 anv_batch_emit_merge(&cmd_buffer->batch, raster_dw,
168 pipeline->gen8.raster);
169 }
170
171 /* Stencil reference values moved from COLOR_CALC_STATE in gen8 to
172 * 3DSTATE_WM_DEPTH_STENCIL in gen9. That means the dirty bits gets split
173 * across different state packets for gen8 and gen9. We handle that by
174 * using a big old #if switch here.
175 */
176 #if GEN_GEN == 8
177 if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS |
178 ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE)) {
179 struct anv_dynamic_state *d = &cmd_buffer->state.dynamic;
180 struct anv_state cc_state =
181 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer,
182 GENX(COLOR_CALC_STATE_length) * 4,
183 64);
184 struct GENX(COLOR_CALC_STATE) cc = {
185 .BlendConstantColorRed = cmd_buffer->state.dynamic.blend_constants[0],
186 .BlendConstantColorGreen = cmd_buffer->state.dynamic.blend_constants[1],
187 .BlendConstantColorBlue = cmd_buffer->state.dynamic.blend_constants[2],
188 .BlendConstantColorAlpha = cmd_buffer->state.dynamic.blend_constants[3],
189 .StencilReferenceValue = d->stencil_reference.front & 0xff,
190 .BackFaceStencilReferenceValue = d->stencil_reference.back & 0xff,
191 };
192 GENX(COLOR_CALC_STATE_pack)(NULL, cc_state.map, &cc);
193
194 if (!cmd_buffer->device->info.has_llc)
195 anv_state_clflush(cc_state);
196
197 anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_CC_STATE_POINTERS), ccp) {
198 ccp.ColorCalcStatePointer = cc_state.offset;
199 ccp.ColorCalcStatePointerValid = true;
200 }
201 }
202
203 if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE |
204 ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK |
205 ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK)) {
206 uint32_t wm_depth_stencil_dw[GENX(3DSTATE_WM_DEPTH_STENCIL_length)];
207 struct anv_dynamic_state *d = &cmd_buffer->state.dynamic;
208
209 struct GENX(3DSTATE_WM_DEPTH_STENCIL wm_depth_stencil) = {
210 GENX(3DSTATE_WM_DEPTH_STENCIL_header),
211
212 .StencilTestMask = d->stencil_compare_mask.front & 0xff,
213 .StencilWriteMask = d->stencil_write_mask.front & 0xff,
214
215 .BackfaceStencilTestMask = d->stencil_compare_mask.back & 0xff,
216 .BackfaceStencilWriteMask = d->stencil_write_mask.back & 0xff,
217 };
218 GENX(3DSTATE_WM_DEPTH_STENCIL_pack)(NULL, wm_depth_stencil_dw,
219 &wm_depth_stencil);
220
221 anv_batch_emit_merge(&cmd_buffer->batch, wm_depth_stencil_dw,
222 pipeline->gen8.wm_depth_stencil);
223 }
224 #else
225 if (cmd_buffer->state.dirty & ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS) {
226 struct anv_state cc_state =
227 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer,
228 GEN9_COLOR_CALC_STATE_length * 4,
229 64);
230 struct GEN9_COLOR_CALC_STATE cc = {
231 .BlendConstantColorRed = cmd_buffer->state.dynamic.blend_constants[0],
232 .BlendConstantColorGreen = cmd_buffer->state.dynamic.blend_constants[1],
233 .BlendConstantColorBlue = cmd_buffer->state.dynamic.blend_constants[2],
234 .BlendConstantColorAlpha = cmd_buffer->state.dynamic.blend_constants[3],
235 };
236 GEN9_COLOR_CALC_STATE_pack(NULL, cc_state.map, &cc);
237
238 if (!cmd_buffer->device->info.has_llc)
239 anv_state_clflush(cc_state);
240
241 anv_batch_emit(&cmd_buffer->batch, GEN9_3DSTATE_CC_STATE_POINTERS, ccp) {
242 ccp.ColorCalcStatePointer = cc_state.offset;
243 ccp.ColorCalcStatePointerValid = true;
244 }
245 }
246
247 if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE |
248 ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK |
249 ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK |
250 ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE)) {
251 uint32_t dwords[GEN9_3DSTATE_WM_DEPTH_STENCIL_length];
252 struct anv_dynamic_state *d = &cmd_buffer->state.dynamic;
253 struct GEN9_3DSTATE_WM_DEPTH_STENCIL wm_depth_stencil = {
254 GEN9_3DSTATE_WM_DEPTH_STENCIL_header,
255
256 .StencilTestMask = d->stencil_compare_mask.front & 0xff,
257 .StencilWriteMask = d->stencil_write_mask.front & 0xff,
258
259 .BackfaceStencilTestMask = d->stencil_compare_mask.back & 0xff,
260 .BackfaceStencilWriteMask = d->stencil_write_mask.back & 0xff,
261
262 .StencilReferenceValue = d->stencil_reference.front & 0xff,
263 .BackfaceStencilReferenceValue = d->stencil_reference.back & 0xff,
264 };
265 GEN9_3DSTATE_WM_DEPTH_STENCIL_pack(NULL, dwords, &wm_depth_stencil);
266
267 anv_batch_emit_merge(&cmd_buffer->batch, dwords,
268 pipeline->gen9.wm_depth_stencil);
269 }
270 #endif
271
272 if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE |
273 ANV_CMD_DIRTY_INDEX_BUFFER)) {
274 anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_VF), vf) {
275 vf.IndexedDrawCutIndexEnable = pipeline->primitive_restart;
276 vf.CutIndex = cmd_buffer->state.restart_index;
277 }
278 }
279
280 cmd_buffer->state.dirty = 0;
281 }
282
283 void genX(CmdBindIndexBuffer)(
284 VkCommandBuffer commandBuffer,
285 VkBuffer _buffer,
286 VkDeviceSize offset,
287 VkIndexType indexType)
288 {
289 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
290 ANV_FROM_HANDLE(anv_buffer, buffer, _buffer);
291
292 static const uint32_t vk_to_gen_index_type[] = {
293 [VK_INDEX_TYPE_UINT16] = INDEX_WORD,
294 [VK_INDEX_TYPE_UINT32] = INDEX_DWORD,
295 };
296
297 static const uint32_t restart_index_for_type[] = {
298 [VK_INDEX_TYPE_UINT16] = UINT16_MAX,
299 [VK_INDEX_TYPE_UINT32] = UINT32_MAX,
300 };
301
302 cmd_buffer->state.restart_index = restart_index_for_type[indexType];
303
304 anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_INDEX_BUFFER), ib) {
305 ib.IndexFormat = vk_to_gen_index_type[indexType];
306 ib.MemoryObjectControlState = GENX(MOCS);
307 ib.BufferStartingAddress =
308 (struct anv_address) { buffer->bo, buffer->offset + offset };
309 ib.BufferSize = buffer->size - offset;
310 }
311
312 cmd_buffer->state.dirty |= ANV_CMD_DIRTY_INDEX_BUFFER;
313 }
314
315
316 /**
317 * Emit the HZ_OP packet in the sequence specified by the BDW PRM section
318 * entitled: "Optimized Depth Buffer Clear and/or Stencil Buffer Clear."
319 *
320 * \todo Enable Stencil Buffer-only clears
321 */
322 void
323 genX(cmd_buffer_emit_hz_op)(struct anv_cmd_buffer *cmd_buffer,
324 enum blorp_hiz_op op)
325 {
326 struct anv_cmd_state *cmd_state = &cmd_buffer->state;
327 const struct anv_image_view *iview =
328 anv_cmd_buffer_get_depth_stencil_view(cmd_buffer);
329
330 if (iview == NULL || !anv_image_has_hiz(iview->image))
331 return;
332
333 /* FINISHME: Implement multi-subpass HiZ */
334 if (cmd_buffer->state.pass->subpass_count > 1)
335 return;
336
337 const uint32_t ds = cmd_state->subpass->depth_stencil_attachment;
338
339 /* Section 7.4. of the Vulkan 1.0.27 spec states:
340 *
341 * "The render area must be contained within the framebuffer dimensions."
342 *
343 * Therefore, the only way the extent of the render area can match that of
344 * the image view is if the render area offset equals (0, 0).
345 */
346 const bool full_surface_op =
347 cmd_state->render_area.extent.width == iview->extent.width &&
348 cmd_state->render_area.extent.height == iview->extent.height;
349 if (full_surface_op)
350 assert(cmd_state->render_area.offset.x == 0 &&
351 cmd_state->render_area.offset.y == 0);
352
353 /* This variable corresponds to the Pixel Dim column in the table below */
354 struct isl_extent2d px_dim;
355
356 /* Validate that we can perform the HZ operation and that it's necessary. */
357 switch (op) {
358 case BLORP_HIZ_OP_DEPTH_CLEAR:
359 if (cmd_buffer->state.pass->attachments[ds].load_op !=
360 VK_ATTACHMENT_LOAD_OP_CLEAR)
361 return;
362
363 /* Apply alignment restrictions. Despite the BDW PRM mentioning this is
364 * only needed for a depth buffer surface type of D16_UNORM, testing
365 * showed it to be necessary for other depth formats as well
366 * (e.g., D32_FLOAT).
367 */
368 #if GEN_GEN == 8
369 /* Pre-SKL, HiZ has an 8x4 sample block. As the number of samples
370 * increases, the number of pixels representable by this block
371 * decreases by a factor of the sample dimensions. Sample dimensions
372 * scale following the MSAA interleaved pattern.
373 *
374 * Sample|Sample|Pixel
375 * Count |Dim |Dim
376 * ===================
377 * 1 | 1x1 | 8x4
378 * 2 | 2x1 | 4x4
379 * 4 | 2x2 | 4x2
380 * 8 | 4x2 | 2x2
381 * 16 | 4x4 | 2x1
382 *
383 * Table: Pixel Dimensions in a HiZ Sample Block Pre-SKL
384 */
385 /* This variable corresponds to the Sample Dim column in the table
386 * above.
387 */
388 const struct isl_extent2d sa_dim =
389 isl_get_interleaved_msaa_px_size_sa(iview->image->samples);
390 px_dim.w = 8 / sa_dim.w;
391 px_dim.h = 4 / sa_dim.h;
392 #elif GEN_GEN >= 9
393 /* SKL+, the sample block becomes a "pixel block" so the expected
394 * pixel dimension is a constant 8x4 px for all sample counts.
395 */
396 px_dim = (struct isl_extent2d) { .w = 8, .h = 4};
397 #endif
398
399 if (!full_surface_op) {
400 /* Fast depth clears clear an entire sample block at a time. As a
401 * result, the rectangle must be aligned to the pixel dimensions of
402 * a sample block for a successful operation.
403 *
404 * Fast clears can still work if the offset is aligned and the render
405 * area offset + extent touches the edge of a depth buffer whose extent
406 * is unaligned. This is because each physical HiZ miplevel is padded
407 * by the px_dim. In this case, the size of the clear rectangle will be
408 * padded later on in this function.
409 */
410 if (cmd_state->render_area.offset.x % px_dim.w ||
411 cmd_state->render_area.offset.y % px_dim.h)
412 return;
413 if (cmd_state->render_area.offset.x +
414 cmd_state->render_area.extent.width != iview->extent.width &&
415 cmd_state->render_area.extent.width % px_dim.w)
416 return;
417 if (cmd_state->render_area.offset.y +
418 cmd_state->render_area.extent.height != iview->extent.height &&
419 cmd_state->render_area.extent.height % px_dim.h)
420 return;
421 }
422 break;
423 case BLORP_HIZ_OP_DEPTH_RESOLVE:
424 if (cmd_buffer->state.pass->attachments[ds].store_op !=
425 VK_ATTACHMENT_STORE_OP_STORE)
426 return;
427 break;
428 case BLORP_HIZ_OP_HIZ_RESOLVE:
429 /* If the render area covers the entire surface *and* load_op is either
430 * CLEAR or DONT_CARE then the previous contents of the depth buffer
431 * will be entirely discarded. In this case, we can skip the HiZ
432 * resolve.
433 *
434 * If the render area is not the full surface, we need to do
435 * the resolve because otherwise data outside the render area may get
436 * garbled by the resolve at the end of the render pass.
437 */
438 if (full_surface_op &&
439 cmd_buffer->state.pass->attachments[ds].load_op !=
440 VK_ATTACHMENT_LOAD_OP_LOAD)
441 return;
442 break;
443 case BLORP_HIZ_OP_NONE:
444 unreachable("Invalid HiZ OP");
445 break;
446 }
447
448 anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_WM_HZ_OP), hzp) {
449 switch (op) {
450 case BLORP_HIZ_OP_DEPTH_CLEAR:
451 hzp.StencilBufferClearEnable = VK_IMAGE_ASPECT_STENCIL_BIT &
452 cmd_state->attachments[ds].pending_clear_aspects;
453 hzp.DepthBufferClearEnable = VK_IMAGE_ASPECT_DEPTH_BIT &
454 cmd_state->attachments[ds].pending_clear_aspects;
455 hzp.FullSurfaceDepthandStencilClear = full_surface_op;
456 hzp.StencilClearValue =
457 cmd_state->attachments[ds].clear_value.depthStencil.stencil & 0xff;
458
459 /* Mark aspects as cleared */
460 cmd_state->attachments[ds].pending_clear_aspects = 0;
461 break;
462 case BLORP_HIZ_OP_DEPTH_RESOLVE:
463 hzp.DepthBufferResolveEnable = true;
464 break;
465 case BLORP_HIZ_OP_HIZ_RESOLVE:
466 hzp.HierarchicalDepthBufferResolveEnable = true;
467 break;
468 case BLORP_HIZ_OP_NONE:
469 unreachable("Invalid HiZ OP");
470 break;
471 }
472
473 if (op != BLORP_HIZ_OP_DEPTH_CLEAR) {
474 /* The Optimized HiZ resolve rectangle must be the size of the full RT
475 * and aligned to 8x4. The non-optimized Depth resolve rectangle must
476 * be the size of the full RT. The same alignment is assumed to be
477 * required.
478 */
479 hzp.ClearRectangleXMin = 0;
480 hzp.ClearRectangleYMin = 0;
481 hzp.ClearRectangleXMax = align_u32(iview->extent.width, 8);
482 hzp.ClearRectangleYMax = align_u32(iview->extent.height, 4);
483 } else {
484 /* This clear rectangle is aligned */
485 hzp.ClearRectangleXMin = cmd_state->render_area.offset.x;
486 hzp.ClearRectangleYMin = cmd_state->render_area.offset.y;
487 hzp.ClearRectangleXMax = cmd_state->render_area.offset.x +
488 align_u32(cmd_state->render_area.extent.width, px_dim.width);
489 hzp.ClearRectangleYMax = cmd_state->render_area.offset.y +
490 align_u32(cmd_state->render_area.extent.height, px_dim.height);
491 }
492
493
494 /* Due to a hardware issue, this bit MBZ */
495 hzp.ScissorRectangleEnable = false;
496 hzp.NumberofMultisamples = ffs(iview->image->samples) - 1;
497 hzp.SampleMask = 0xFFFF;
498 }
499
500 anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
501 pc.PostSyncOperation = WriteImmediateData;
502 pc.Address =
503 (struct anv_address){ &cmd_buffer->device->workaround_bo, 0 };
504 }
505
506 anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_WM_HZ_OP), hzp);
507
508 if (!full_surface_op && op == BLORP_HIZ_OP_DEPTH_CLEAR) {
509 anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
510 pc.DepthStallEnable = true;
511 pc.DepthCacheFlushEnable = true;
512 }
513 }
514 }
515
516 /* Set of stage bits for which are pipelined, i.e. they get queued by the
517 * command streamer for later execution.
518 */
519 #define ANV_PIPELINE_STAGE_PIPELINED_BITS \
520 (VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | \
521 VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | \
522 VK_PIPELINE_STAGE_TESSELLATION_CONTROL_SHADER_BIT | \
523 VK_PIPELINE_STAGE_TESSELLATION_EVALUATION_SHADER_BIT | \
524 VK_PIPELINE_STAGE_GEOMETRY_SHADER_BIT | \
525 VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | \
526 VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | \
527 VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT | \
528 VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | \
529 VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | \
530 VK_PIPELINE_STAGE_TRANSFER_BIT | \
531 VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT | \
532 VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT | \
533 VK_PIPELINE_STAGE_ALL_COMMANDS_BIT)
534
535 void genX(CmdSetEvent)(
536 VkCommandBuffer commandBuffer,
537 VkEvent _event,
538 VkPipelineStageFlags stageMask)
539 {
540 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
541 ANV_FROM_HANDLE(anv_event, event, _event);
542
543 anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
544 if (stageMask & ANV_PIPELINE_STAGE_PIPELINED_BITS) {
545 pc.StallAtPixelScoreboard = true;
546 pc.CommandStreamerStallEnable = true;
547 }
548
549 pc.DestinationAddressType = DAT_PPGTT,
550 pc.PostSyncOperation = WriteImmediateData,
551 pc.Address = (struct anv_address) {
552 &cmd_buffer->device->dynamic_state_block_pool.bo,
553 event->state.offset
554 };
555 pc.ImmediateData = VK_EVENT_SET;
556 }
557 }
558
559 void genX(CmdResetEvent)(
560 VkCommandBuffer commandBuffer,
561 VkEvent _event,
562 VkPipelineStageFlags stageMask)
563 {
564 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
565 ANV_FROM_HANDLE(anv_event, event, _event);
566
567 anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
568 if (stageMask & ANV_PIPELINE_STAGE_PIPELINED_BITS) {
569 pc.StallAtPixelScoreboard = true;
570 pc.CommandStreamerStallEnable = true;
571 }
572
573 pc.DestinationAddressType = DAT_PPGTT;
574 pc.PostSyncOperation = WriteImmediateData;
575 pc.Address = (struct anv_address) {
576 &cmd_buffer->device->dynamic_state_block_pool.bo,
577 event->state.offset
578 };
579 pc.ImmediateData = VK_EVENT_RESET;
580 }
581 }
582
583 void genX(CmdWaitEvents)(
584 VkCommandBuffer commandBuffer,
585 uint32_t eventCount,
586 const VkEvent* pEvents,
587 VkPipelineStageFlags srcStageMask,
588 VkPipelineStageFlags destStageMask,
589 uint32_t memoryBarrierCount,
590 const VkMemoryBarrier* pMemoryBarriers,
591 uint32_t bufferMemoryBarrierCount,
592 const VkBufferMemoryBarrier* pBufferMemoryBarriers,
593 uint32_t imageMemoryBarrierCount,
594 const VkImageMemoryBarrier* pImageMemoryBarriers)
595 {
596 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
597 for (uint32_t i = 0; i < eventCount; i++) {
598 ANV_FROM_HANDLE(anv_event, event, pEvents[i]);
599
600 anv_batch_emit(&cmd_buffer->batch, GENX(MI_SEMAPHORE_WAIT), sem) {
601 sem.WaitMode = PollingMode,
602 sem.CompareOperation = COMPARE_SAD_EQUAL_SDD,
603 sem.SemaphoreDataDword = VK_EVENT_SET,
604 sem.SemaphoreAddress = (struct anv_address) {
605 &cmd_buffer->device->dynamic_state_block_pool.bo,
606 event->state.offset
607 };
608 }
609 }
610
611 genX(CmdPipelineBarrier)(commandBuffer, srcStageMask, destStageMask,
612 false, /* byRegion */
613 memoryBarrierCount, pMemoryBarriers,
614 bufferMemoryBarrierCount, pBufferMemoryBarriers,
615 imageMemoryBarrierCount, pImageMemoryBarriers);
616 }