anv/cmd_buffer: Use some pre-existing pipeline temporaries
[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 = vp->maxDepth - vp->minDepth,
53 .ViewportMatrixElementm30 = vp->x + vp->width / 2,
54 .ViewportMatrixElementm31 = vp->y + vp->height / 2,
55 .ViewportMatrixElementm32 = vp->minDepth,
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 = MIN2(vp->y, vp->y + vp->height),
63 .YMaxViewPort = MAX2(vp->y, 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 anv_state_flush(cmd_buffer->device, sf_clip_state);
71
72 anv_batch_emit(&cmd_buffer->batch,
73 GENX(3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP), clip) {
74 clip.SFClipViewportPointer = sf_clip_state.offset;
75 }
76 }
77
78 void
79 gen8_cmd_buffer_emit_depth_viewport(struct anv_cmd_buffer *cmd_buffer,
80 bool depth_clamp_enable)
81 {
82 uint32_t count = cmd_buffer->state.dynamic.viewport.count;
83 const VkViewport *viewports = cmd_buffer->state.dynamic.viewport.viewports;
84 struct anv_state cc_state =
85 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, count * 8, 32);
86
87 for (uint32_t i = 0; i < count; i++) {
88 const VkViewport *vp = &viewports[i];
89
90 struct GENX(CC_VIEWPORT) cc_viewport = {
91 .MinimumDepth = depth_clamp_enable ? vp->minDepth : 0.0f,
92 .MaximumDepth = depth_clamp_enable ? vp->maxDepth : 1.0f,
93 };
94
95 GENX(CC_VIEWPORT_pack)(NULL, cc_state.map + i * 8, &cc_viewport);
96 }
97
98 anv_state_flush(cmd_buffer->device, cc_state);
99
100 anv_batch_emit(&cmd_buffer->batch,
101 GENX(3DSTATE_VIEWPORT_STATE_POINTERS_CC), cc) {
102 cc.CCViewportPointer = cc_state.offset;
103 }
104 }
105 #endif
106
107 void
108 genX(cmd_buffer_enable_pma_fix)(struct anv_cmd_buffer *cmd_buffer, bool enable)
109 {
110 if (cmd_buffer->state.pma_fix_enabled == enable)
111 return;
112
113 cmd_buffer->state.pma_fix_enabled = enable;
114
115 /* According to the Broadwell PIPE_CONTROL documentation, software should
116 * emit a PIPE_CONTROL with the CS Stall and Depth Cache Flush bits set
117 * prior to the LRI. If stencil buffer writes are enabled, then a Render
118 * Cache Flush is also necessary.
119 *
120 * The Skylake docs say to use a depth stall rather than a command
121 * streamer stall. However, the hardware seems to violently disagree.
122 * A full command streamer stall seems to be needed in both cases.
123 */
124 anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
125 pc.DepthCacheFlushEnable = true;
126 pc.CommandStreamerStallEnable = true;
127 pc.RenderTargetCacheFlushEnable = true;
128 }
129
130 #if GEN_GEN == 9
131
132 uint32_t cache_mode;
133 anv_pack_struct(&cache_mode, GENX(CACHE_MODE_0),
134 .STCPMAOptimizationEnable = enable,
135 .STCPMAOptimizationEnableMask = true);
136 anv_batch_emit(&cmd_buffer->batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
137 lri.RegisterOffset = GENX(CACHE_MODE_0_num);
138 lri.DataDWord = cache_mode;
139 }
140
141 #elif GEN_GEN == 8
142
143 uint32_t cache_mode;
144 anv_pack_struct(&cache_mode, GENX(CACHE_MODE_1),
145 .NPPMAFixEnable = enable,
146 .NPEarlyZFailsDisable = enable,
147 .NPPMAFixEnableMask = true,
148 .NPEarlyZFailsDisableMask = true);
149 anv_batch_emit(&cmd_buffer->batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
150 lri.RegisterOffset = GENX(CACHE_MODE_1_num);
151 lri.DataDWord = cache_mode;
152 }
153
154 #endif /* GEN_GEN == 8 */
155
156 /* After the LRI, a PIPE_CONTROL with both the Depth Stall and Depth Cache
157 * Flush bits is often necessary. We do it regardless because it's easier.
158 * The render cache flush is also necessary if stencil writes are enabled.
159 *
160 * Again, the Skylake docs give a different set of flushes but the BDW
161 * flushes seem to work just as well.
162 */
163 anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
164 pc.DepthStallEnable = true;
165 pc.DepthCacheFlushEnable = true;
166 pc.RenderTargetCacheFlushEnable = true;
167 }
168 }
169
170 UNUSED static bool
171 want_depth_pma_fix(struct anv_cmd_buffer *cmd_buffer)
172 {
173 assert(GEN_GEN == 8);
174
175 /* From the Broadwell PRM Vol. 2c CACHE_MODE_1::NP_PMA_FIX_ENABLE:
176 *
177 * SW must set this bit in order to enable this fix when following
178 * expression is TRUE.
179 *
180 * 3DSTATE_WM::ForceThreadDispatch != 1 &&
181 * !(3DSTATE_RASTER::ForceSampleCount != NUMRASTSAMPLES_0) &&
182 * (3DSTATE_DEPTH_BUFFER::SURFACE_TYPE != NULL) &&
183 * (3DSTATE_DEPTH_BUFFER::HIZ Enable) &&
184 * !(3DSTATE_WM::EDSC_Mode == EDSC_PREPS) &&
185 * (3DSTATE_PS_EXTRA::PixelShaderValid) &&
186 * !(3DSTATE_WM_HZ_OP::DepthBufferClear ||
187 * 3DSTATE_WM_HZ_OP::DepthBufferResolve ||
188 * 3DSTATE_WM_HZ_OP::Hierarchical Depth Buffer Resolve Enable ||
189 * 3DSTATE_WM_HZ_OP::StencilBufferClear) &&
190 * (3DSTATE_WM_DEPTH_STENCIL::DepthTestEnable) &&
191 * (((3DSTATE_PS_EXTRA::PixelShaderKillsPixels ||
192 * 3DSTATE_PS_EXTRA::oMask Present to RenderTarget ||
193 * 3DSTATE_PS_BLEND::AlphaToCoverageEnable ||
194 * 3DSTATE_PS_BLEND::AlphaTestEnable ||
195 * 3DSTATE_WM_CHROMAKEY::ChromaKeyKillEnable) &&
196 * 3DSTATE_WM::ForceKillPix != ForceOff &&
197 * ((3DSTATE_WM_DEPTH_STENCIL::DepthWriteEnable &&
198 * 3DSTATE_DEPTH_BUFFER::DEPTH_WRITE_ENABLE) ||
199 * (3DSTATE_WM_DEPTH_STENCIL::Stencil Buffer Write Enable &&
200 * 3DSTATE_DEPTH_BUFFER::STENCIL_WRITE_ENABLE &&
201 * 3DSTATE_STENCIL_BUFFER::STENCIL_BUFFER_ENABLE))) ||
202 * (3DSTATE_PS_EXTRA:: Pixel Shader Computed Depth mode != PSCDEPTH_OFF))
203 */
204
205 /* These are always true:
206 * 3DSTATE_WM::ForceThreadDispatch != 1 &&
207 * !(3DSTATE_RASTER::ForceSampleCount != NUMRASTSAMPLES_0)
208 */
209
210 /* We only enable the PMA fix if we know for certain that HiZ is enabled.
211 * If we don't know whether HiZ is enabled or not, we disable the PMA fix
212 * and there is no harm.
213 *
214 * (3DSTATE_DEPTH_BUFFER::SURFACE_TYPE != NULL) &&
215 * 3DSTATE_DEPTH_BUFFER::HIZ Enable
216 */
217 if (!cmd_buffer->state.hiz_enabled)
218 return false;
219
220 /* 3DSTATE_PS_EXTRA::PixelShaderValid */
221 struct anv_pipeline *pipeline = cmd_buffer->state.pipeline;
222 if (!anv_pipeline_has_stage(pipeline, MESA_SHADER_FRAGMENT))
223 return false;
224
225 /* !(3DSTATE_WM::EDSC_Mode == EDSC_PREPS) */
226 const struct brw_wm_prog_data *wm_prog_data = get_wm_prog_data(pipeline);
227 if (wm_prog_data->early_fragment_tests)
228 return false;
229
230 /* We never use anv_pipeline for HiZ ops so this is trivially true:
231 * !(3DSTATE_WM_HZ_OP::DepthBufferClear ||
232 * 3DSTATE_WM_HZ_OP::DepthBufferResolve ||
233 * 3DSTATE_WM_HZ_OP::Hierarchical Depth Buffer Resolve Enable ||
234 * 3DSTATE_WM_HZ_OP::StencilBufferClear)
235 */
236
237 /* 3DSTATE_WM_DEPTH_STENCIL::DepthTestEnable */
238 if (!pipeline->depth_test_enable)
239 return false;
240
241 /* (((3DSTATE_PS_EXTRA::PixelShaderKillsPixels ||
242 * 3DSTATE_PS_EXTRA::oMask Present to RenderTarget ||
243 * 3DSTATE_PS_BLEND::AlphaToCoverageEnable ||
244 * 3DSTATE_PS_BLEND::AlphaTestEnable ||
245 * 3DSTATE_WM_CHROMAKEY::ChromaKeyKillEnable) &&
246 * 3DSTATE_WM::ForceKillPix != ForceOff &&
247 * ((3DSTATE_WM_DEPTH_STENCIL::DepthWriteEnable &&
248 * 3DSTATE_DEPTH_BUFFER::DEPTH_WRITE_ENABLE) ||
249 * (3DSTATE_WM_DEPTH_STENCIL::Stencil Buffer Write Enable &&
250 * 3DSTATE_DEPTH_BUFFER::STENCIL_WRITE_ENABLE &&
251 * 3DSTATE_STENCIL_BUFFER::STENCIL_BUFFER_ENABLE))) ||
252 * (3DSTATE_PS_EXTRA:: Pixel Shader Computed Depth mode != PSCDEPTH_OFF))
253 */
254 return (pipeline->kill_pixel && (pipeline->writes_depth ||
255 pipeline->writes_stencil)) ||
256 wm_prog_data->computed_depth_mode != PSCDEPTH_OFF;
257 }
258
259 UNUSED static bool
260 want_stencil_pma_fix(struct anv_cmd_buffer *cmd_buffer)
261 {
262 if (GEN_GEN > 9)
263 return false;
264 assert(GEN_GEN == 9);
265
266 /* From the Skylake PRM Vol. 2c CACHE_MODE_1::STC PMA Optimization Enable:
267 *
268 * Clearing this bit will force the STC cache to wait for pending
269 * retirement of pixels at the HZ-read stage and do the STC-test for
270 * Non-promoted, R-computed and Computed depth modes instead of
271 * postponing the STC-test to RCPFE.
272 *
273 * STC_TEST_EN = 3DSTATE_STENCIL_BUFFER::STENCIL_BUFFER_ENABLE &&
274 * 3DSTATE_WM_DEPTH_STENCIL::StencilTestEnable
275 *
276 * STC_WRITE_EN = 3DSTATE_STENCIL_BUFFER::STENCIL_BUFFER_ENABLE &&
277 * (3DSTATE_WM_DEPTH_STENCIL::Stencil Buffer Write Enable &&
278 * 3DSTATE_DEPTH_BUFFER::STENCIL_WRITE_ENABLE)
279 *
280 * COMP_STC_EN = STC_TEST_EN &&
281 * 3DSTATE_PS_EXTRA::PixelShaderComputesStencil
282 *
283 * SW parses the pipeline states to generate the following logical
284 * signal indicating if PMA FIX can be enabled.
285 *
286 * STC_PMA_OPT =
287 * 3DSTATE_WM::ForceThreadDispatch != 1 &&
288 * !(3DSTATE_RASTER::ForceSampleCount != NUMRASTSAMPLES_0) &&
289 * 3DSTATE_DEPTH_BUFFER::SURFACE_TYPE != NULL &&
290 * 3DSTATE_DEPTH_BUFFER::HIZ Enable &&
291 * !(3DSTATE_WM::EDSC_Mode == 2) &&
292 * 3DSTATE_PS_EXTRA::PixelShaderValid &&
293 * !(3DSTATE_WM_HZ_OP::DepthBufferClear ||
294 * 3DSTATE_WM_HZ_OP::DepthBufferResolve ||
295 * 3DSTATE_WM_HZ_OP::Hierarchical Depth Buffer Resolve Enable ||
296 * 3DSTATE_WM_HZ_OP::StencilBufferClear) &&
297 * (COMP_STC_EN || STC_WRITE_EN) &&
298 * ((3DSTATE_PS_EXTRA::PixelShaderKillsPixels ||
299 * 3DSTATE_WM::ForceKillPix == ON ||
300 * 3DSTATE_PS_EXTRA::oMask Present to RenderTarget ||
301 * 3DSTATE_PS_BLEND::AlphaToCoverageEnable ||
302 * 3DSTATE_PS_BLEND::AlphaTestEnable ||
303 * 3DSTATE_WM_CHROMAKEY::ChromaKeyKillEnable) ||
304 * (3DSTATE_PS_EXTRA::Pixel Shader Computed Depth mode != PSCDEPTH_OFF))
305 */
306
307 /* These are always true:
308 * 3DSTATE_WM::ForceThreadDispatch != 1 &&
309 * !(3DSTATE_RASTER::ForceSampleCount != NUMRASTSAMPLES_0)
310 */
311
312 /* We only enable the PMA fix if we know for certain that HiZ is enabled.
313 * If we don't know whether HiZ is enabled or not, we disable the PMA fix
314 * and there is no harm.
315 *
316 * (3DSTATE_DEPTH_BUFFER::SURFACE_TYPE != NULL) &&
317 * 3DSTATE_DEPTH_BUFFER::HIZ Enable
318 */
319 if (!cmd_buffer->state.hiz_enabled)
320 return false;
321
322 /* We can't possibly know if HiZ is enabled without the framebuffer */
323 assert(cmd_buffer->state.framebuffer);
324
325 /* HiZ is enabled so we had better have a depth buffer with HiZ */
326 const struct anv_image_view *ds_iview =
327 anv_cmd_buffer_get_depth_stencil_view(cmd_buffer);
328 assert(ds_iview && ds_iview->image->planes[0].aux_usage == ISL_AUX_USAGE_HIZ);
329
330 /* 3DSTATE_PS_EXTRA::PixelShaderValid */
331 struct anv_pipeline *pipeline = cmd_buffer->state.pipeline;
332 if (!anv_pipeline_has_stage(pipeline, MESA_SHADER_FRAGMENT))
333 return false;
334
335 /* !(3DSTATE_WM::EDSC_Mode == 2) */
336 const struct brw_wm_prog_data *wm_prog_data = get_wm_prog_data(pipeline);
337 if (wm_prog_data->early_fragment_tests)
338 return false;
339
340 /* We never use anv_pipeline for HiZ ops so this is trivially true:
341 * !(3DSTATE_WM_HZ_OP::DepthBufferClear ||
342 * 3DSTATE_WM_HZ_OP::DepthBufferResolve ||
343 * 3DSTATE_WM_HZ_OP::Hierarchical Depth Buffer Resolve Enable ||
344 * 3DSTATE_WM_HZ_OP::StencilBufferClear)
345 */
346
347 /* 3DSTATE_STENCIL_BUFFER::STENCIL_BUFFER_ENABLE &&
348 * 3DSTATE_WM_DEPTH_STENCIL::StencilTestEnable
349 */
350 const bool stc_test_en =
351 (ds_iview->image->aspects & VK_IMAGE_ASPECT_STENCIL_BIT) &&
352 pipeline->stencil_test_enable;
353
354 /* 3DSTATE_STENCIL_BUFFER::STENCIL_BUFFER_ENABLE &&
355 * (3DSTATE_WM_DEPTH_STENCIL::Stencil Buffer Write Enable &&
356 * 3DSTATE_DEPTH_BUFFER::STENCIL_WRITE_ENABLE)
357 */
358 const bool stc_write_en =
359 (ds_iview->image->aspects & VK_IMAGE_ASPECT_STENCIL_BIT) &&
360 pipeline->writes_stencil;
361
362 /* STC_TEST_EN && 3DSTATE_PS_EXTRA::PixelShaderComputesStencil */
363 const bool comp_stc_en = stc_test_en && wm_prog_data->computed_stencil;
364
365 /* COMP_STC_EN || STC_WRITE_EN */
366 if (!(comp_stc_en || stc_write_en))
367 return false;
368
369 /* (3DSTATE_PS_EXTRA::PixelShaderKillsPixels ||
370 * 3DSTATE_WM::ForceKillPix == ON ||
371 * 3DSTATE_PS_EXTRA::oMask Present to RenderTarget ||
372 * 3DSTATE_PS_BLEND::AlphaToCoverageEnable ||
373 * 3DSTATE_PS_BLEND::AlphaTestEnable ||
374 * 3DSTATE_WM_CHROMAKEY::ChromaKeyKillEnable) ||
375 * (3DSTATE_PS_EXTRA::Pixel Shader Computed Depth mode != PSCDEPTH_OFF)
376 */
377 return pipeline->kill_pixel ||
378 wm_prog_data->computed_depth_mode != PSCDEPTH_OFF;
379 }
380
381 void
382 genX(cmd_buffer_flush_dynamic_state)(struct anv_cmd_buffer *cmd_buffer)
383 {
384 struct anv_pipeline *pipeline = cmd_buffer->state.pipeline;
385
386 if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE |
387 ANV_CMD_DIRTY_DYNAMIC_LINE_WIDTH)) {
388 uint32_t sf_dw[GENX(3DSTATE_SF_length)];
389 struct GENX(3DSTATE_SF) sf = {
390 GENX(3DSTATE_SF_header),
391 };
392 #if GEN_GEN == 8
393 if (cmd_buffer->device->info.is_cherryview) {
394 sf.CHVLineWidth = cmd_buffer->state.dynamic.line_width;
395 } else {
396 sf.LineWidth = cmd_buffer->state.dynamic.line_width;
397 }
398 #else
399 sf.LineWidth = cmd_buffer->state.dynamic.line_width,
400 #endif
401 GENX(3DSTATE_SF_pack)(NULL, sf_dw, &sf);
402 anv_batch_emit_merge(&cmd_buffer->batch, sf_dw, pipeline->gen8.sf);
403 }
404
405 if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE |
406 ANV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS)){
407 uint32_t raster_dw[GENX(3DSTATE_RASTER_length)];
408 struct GENX(3DSTATE_RASTER) raster = {
409 GENX(3DSTATE_RASTER_header),
410 .GlobalDepthOffsetConstant = cmd_buffer->state.dynamic.depth_bias.bias,
411 .GlobalDepthOffsetScale = cmd_buffer->state.dynamic.depth_bias.slope,
412 .GlobalDepthOffsetClamp = cmd_buffer->state.dynamic.depth_bias.clamp
413 };
414 GENX(3DSTATE_RASTER_pack)(NULL, raster_dw, &raster);
415 anv_batch_emit_merge(&cmd_buffer->batch, raster_dw,
416 pipeline->gen8.raster);
417 }
418
419 /* Stencil reference values moved from COLOR_CALC_STATE in gen8 to
420 * 3DSTATE_WM_DEPTH_STENCIL in gen9. That means the dirty bits gets split
421 * across different state packets for gen8 and gen9. We handle that by
422 * using a big old #if switch here.
423 */
424 #if GEN_GEN == 8
425 if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS |
426 ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE)) {
427 struct anv_dynamic_state *d = &cmd_buffer->state.dynamic;
428 struct anv_state cc_state =
429 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer,
430 GENX(COLOR_CALC_STATE_length) * 4,
431 64);
432 struct GENX(COLOR_CALC_STATE) cc = {
433 .BlendConstantColorRed = cmd_buffer->state.dynamic.blend_constants[0],
434 .BlendConstantColorGreen = cmd_buffer->state.dynamic.blend_constants[1],
435 .BlendConstantColorBlue = cmd_buffer->state.dynamic.blend_constants[2],
436 .BlendConstantColorAlpha = cmd_buffer->state.dynamic.blend_constants[3],
437 .StencilReferenceValue = d->stencil_reference.front & 0xff,
438 .BackfaceStencilReferenceValue = d->stencil_reference.back & 0xff,
439 };
440 GENX(COLOR_CALC_STATE_pack)(NULL, cc_state.map, &cc);
441
442 anv_state_flush(cmd_buffer->device, cc_state);
443
444 anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_CC_STATE_POINTERS), ccp) {
445 ccp.ColorCalcStatePointer = cc_state.offset;
446 ccp.ColorCalcStatePointerValid = true;
447 }
448 }
449
450 if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE |
451 ANV_CMD_DIRTY_RENDER_TARGETS |
452 ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK |
453 ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK)) {
454 uint32_t wm_depth_stencil_dw[GENX(3DSTATE_WM_DEPTH_STENCIL_length)];
455 struct anv_dynamic_state *d = &cmd_buffer->state.dynamic;
456
457 struct GENX(3DSTATE_WM_DEPTH_STENCIL wm_depth_stencil) = {
458 GENX(3DSTATE_WM_DEPTH_STENCIL_header),
459
460 .StencilTestMask = d->stencil_compare_mask.front & 0xff,
461 .StencilWriteMask = d->stencil_write_mask.front & 0xff,
462
463 .BackfaceStencilTestMask = d->stencil_compare_mask.back & 0xff,
464 .BackfaceStencilWriteMask = d->stencil_write_mask.back & 0xff,
465
466 .StencilBufferWriteEnable =
467 (d->stencil_write_mask.front || d->stencil_write_mask.back) &&
468 pipeline->writes_stencil,
469 };
470 GENX(3DSTATE_WM_DEPTH_STENCIL_pack)(NULL, wm_depth_stencil_dw,
471 &wm_depth_stencil);
472
473 anv_batch_emit_merge(&cmd_buffer->batch, wm_depth_stencil_dw,
474 pipeline->gen8.wm_depth_stencil);
475
476 genX(cmd_buffer_enable_pma_fix)(cmd_buffer,
477 want_depth_pma_fix(cmd_buffer));
478 }
479 #else
480 if (cmd_buffer->state.dirty & ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS) {
481 struct anv_state cc_state =
482 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer,
483 GENX(COLOR_CALC_STATE_length) * 4,
484 64);
485 struct GENX(COLOR_CALC_STATE) cc = {
486 .BlendConstantColorRed = cmd_buffer->state.dynamic.blend_constants[0],
487 .BlendConstantColorGreen = cmd_buffer->state.dynamic.blend_constants[1],
488 .BlendConstantColorBlue = cmd_buffer->state.dynamic.blend_constants[2],
489 .BlendConstantColorAlpha = cmd_buffer->state.dynamic.blend_constants[3],
490 };
491 GENX(COLOR_CALC_STATE_pack)(NULL, cc_state.map, &cc);
492
493 anv_state_flush(cmd_buffer->device, cc_state);
494
495 anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_CC_STATE_POINTERS), ccp) {
496 ccp.ColorCalcStatePointer = cc_state.offset;
497 ccp.ColorCalcStatePointerValid = true;
498 }
499 }
500
501 if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE |
502 ANV_CMD_DIRTY_RENDER_TARGETS |
503 ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK |
504 ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK |
505 ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE)) {
506 uint32_t dwords[GENX(3DSTATE_WM_DEPTH_STENCIL_length)];
507 struct anv_dynamic_state *d = &cmd_buffer->state.dynamic;
508 struct GENX(3DSTATE_WM_DEPTH_STENCIL) wm_depth_stencil = {
509 GENX(3DSTATE_WM_DEPTH_STENCIL_header),
510
511 .StencilTestMask = d->stencil_compare_mask.front & 0xff,
512 .StencilWriteMask = d->stencil_write_mask.front & 0xff,
513
514 .BackfaceStencilTestMask = d->stencil_compare_mask.back & 0xff,
515 .BackfaceStencilWriteMask = d->stencil_write_mask.back & 0xff,
516
517 .StencilReferenceValue = d->stencil_reference.front & 0xff,
518 .BackfaceStencilReferenceValue = d->stencil_reference.back & 0xff,
519
520 .StencilBufferWriteEnable =
521 (d->stencil_write_mask.front || d->stencil_write_mask.back) &&
522 pipeline->writes_stencil,
523 };
524 GENX(3DSTATE_WM_DEPTH_STENCIL_pack)(NULL, dwords, &wm_depth_stencil);
525
526 anv_batch_emit_merge(&cmd_buffer->batch, dwords,
527 pipeline->gen9.wm_depth_stencil);
528
529 genX(cmd_buffer_enable_pma_fix)(cmd_buffer,
530 want_stencil_pma_fix(cmd_buffer));
531 }
532 #endif
533
534 if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE |
535 ANV_CMD_DIRTY_INDEX_BUFFER)) {
536 anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_VF), vf) {
537 vf.IndexedDrawCutIndexEnable = pipeline->primitive_restart;
538 vf.CutIndex = cmd_buffer->state.restart_index;
539 }
540 }
541
542 cmd_buffer->state.dirty = 0;
543 }
544
545 void genX(CmdBindIndexBuffer)(
546 VkCommandBuffer commandBuffer,
547 VkBuffer _buffer,
548 VkDeviceSize offset,
549 VkIndexType indexType)
550 {
551 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
552 ANV_FROM_HANDLE(anv_buffer, buffer, _buffer);
553
554 static const uint32_t vk_to_gen_index_type[] = {
555 [VK_INDEX_TYPE_UINT16] = INDEX_WORD,
556 [VK_INDEX_TYPE_UINT32] = INDEX_DWORD,
557 };
558
559 static const uint32_t restart_index_for_type[] = {
560 [VK_INDEX_TYPE_UINT16] = UINT16_MAX,
561 [VK_INDEX_TYPE_UINT32] = UINT32_MAX,
562 };
563
564 cmd_buffer->state.restart_index = restart_index_for_type[indexType];
565
566 anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_INDEX_BUFFER), ib) {
567 ib.IndexFormat = vk_to_gen_index_type[indexType];
568 ib.MemoryObjectControlState = GENX(MOCS);
569 ib.BufferStartingAddress =
570 (struct anv_address) { buffer->bo, buffer->offset + offset };
571 ib.BufferSize = buffer->size - offset;
572 }
573
574 cmd_buffer->state.dirty |= ANV_CMD_DIRTY_INDEX_BUFFER;
575 }
576
577 /* Set of stage bits for which are pipelined, i.e. they get queued by the
578 * command streamer for later execution.
579 */
580 #define ANV_PIPELINE_STAGE_PIPELINED_BITS \
581 (VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | \
582 VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | \
583 VK_PIPELINE_STAGE_TESSELLATION_CONTROL_SHADER_BIT | \
584 VK_PIPELINE_STAGE_TESSELLATION_EVALUATION_SHADER_BIT | \
585 VK_PIPELINE_STAGE_GEOMETRY_SHADER_BIT | \
586 VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | \
587 VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | \
588 VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT | \
589 VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | \
590 VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | \
591 VK_PIPELINE_STAGE_TRANSFER_BIT | \
592 VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT | \
593 VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT | \
594 VK_PIPELINE_STAGE_ALL_COMMANDS_BIT)
595
596 void genX(CmdSetEvent)(
597 VkCommandBuffer commandBuffer,
598 VkEvent _event,
599 VkPipelineStageFlags stageMask)
600 {
601 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
602 ANV_FROM_HANDLE(anv_event, event, _event);
603
604 anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
605 if (stageMask & ANV_PIPELINE_STAGE_PIPELINED_BITS) {
606 pc.StallAtPixelScoreboard = true;
607 pc.CommandStreamerStallEnable = true;
608 }
609
610 pc.DestinationAddressType = DAT_PPGTT,
611 pc.PostSyncOperation = WriteImmediateData,
612 pc.Address = (struct anv_address) {
613 &cmd_buffer->device->dynamic_state_pool.block_pool.bo,
614 event->state.offset
615 };
616 pc.ImmediateData = VK_EVENT_SET;
617 }
618 }
619
620 void genX(CmdResetEvent)(
621 VkCommandBuffer commandBuffer,
622 VkEvent _event,
623 VkPipelineStageFlags stageMask)
624 {
625 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
626 ANV_FROM_HANDLE(anv_event, event, _event);
627
628 anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
629 if (stageMask & ANV_PIPELINE_STAGE_PIPELINED_BITS) {
630 pc.StallAtPixelScoreboard = true;
631 pc.CommandStreamerStallEnable = true;
632 }
633
634 pc.DestinationAddressType = DAT_PPGTT;
635 pc.PostSyncOperation = WriteImmediateData;
636 pc.Address = (struct anv_address) {
637 &cmd_buffer->device->dynamic_state_pool.block_pool.bo,
638 event->state.offset
639 };
640 pc.ImmediateData = VK_EVENT_RESET;
641 }
642 }
643
644 void genX(CmdWaitEvents)(
645 VkCommandBuffer commandBuffer,
646 uint32_t eventCount,
647 const VkEvent* pEvents,
648 VkPipelineStageFlags srcStageMask,
649 VkPipelineStageFlags destStageMask,
650 uint32_t memoryBarrierCount,
651 const VkMemoryBarrier* pMemoryBarriers,
652 uint32_t bufferMemoryBarrierCount,
653 const VkBufferMemoryBarrier* pBufferMemoryBarriers,
654 uint32_t imageMemoryBarrierCount,
655 const VkImageMemoryBarrier* pImageMemoryBarriers)
656 {
657 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
658 for (uint32_t i = 0; i < eventCount; i++) {
659 ANV_FROM_HANDLE(anv_event, event, pEvents[i]);
660
661 anv_batch_emit(&cmd_buffer->batch, GENX(MI_SEMAPHORE_WAIT), sem) {
662 sem.WaitMode = PollingMode,
663 sem.CompareOperation = COMPARE_SAD_EQUAL_SDD,
664 sem.SemaphoreDataDword = VK_EVENT_SET,
665 sem.SemaphoreAddress = (struct anv_address) {
666 &cmd_buffer->device->dynamic_state_pool.block_pool.bo,
667 event->state.offset
668 };
669 }
670 }
671
672 genX(CmdPipelineBarrier)(commandBuffer, srcStageMask, destStageMask,
673 false, /* byRegion */
674 memoryBarrierCount, pMemoryBarriers,
675 bufferMemoryBarrierCount, pBufferMemoryBarriers,
676 imageMemoryBarrierCount, pImageMemoryBarriers);
677 }