anv: Handle clamping of inverted depth ranges
[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 #include "common/gen_guardband.h"
35
36 #if GEN_GEN == 8
37 void
38 gen8_cmd_buffer_emit_viewport(struct anv_cmd_buffer *cmd_buffer)
39 {
40 struct anv_framebuffer *fb = cmd_buffer->state.framebuffer;
41 uint32_t count = cmd_buffer->state.gfx.dynamic.viewport.count;
42 const VkViewport *viewports =
43 cmd_buffer->state.gfx.dynamic.viewport.viewports;
44 struct anv_state sf_clip_state =
45 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, count * 64, 64);
46
47 for (uint32_t i = 0; i < count; i++) {
48 const VkViewport *vp = &viewports[i];
49
50 /* The gen7 state struct has just the matrix and guardband fields, the
51 * gen8 struct adds the min/max viewport fields. */
52 struct GENX(SF_CLIP_VIEWPORT) sfv = {
53 .ViewportMatrixElementm00 = vp->width / 2,
54 .ViewportMatrixElementm11 = vp->height / 2,
55 .ViewportMatrixElementm22 = vp->maxDepth - vp->minDepth,
56 .ViewportMatrixElementm30 = vp->x + vp->width / 2,
57 .ViewportMatrixElementm31 = vp->y + vp->height / 2,
58 .ViewportMatrixElementm32 = vp->minDepth,
59 .XMinClipGuardband = -1.0f,
60 .XMaxClipGuardband = 1.0f,
61 .YMinClipGuardband = -1.0f,
62 .YMaxClipGuardband = 1.0f,
63 .XMinViewPort = vp->x,
64 .XMaxViewPort = vp->x + vp->width - 1,
65 .YMinViewPort = MIN2(vp->y, vp->y + vp->height),
66 .YMaxViewPort = MAX2(vp->y, vp->y + vp->height) - 1,
67 };
68
69 if (fb) {
70 /* We can only calculate a "real" guardband clip if we know the
71 * framebuffer at the time we emit the packet. Otherwise, we have
72 * fall back to a worst-case guardband of [-1, 1].
73 */
74 gen_calculate_guardband_size(fb->width, fb->height,
75 sfv.ViewportMatrixElementm00,
76 sfv.ViewportMatrixElementm11,
77 sfv.ViewportMatrixElementm30,
78 sfv.ViewportMatrixElementm31,
79 &sfv.XMinClipGuardband,
80 &sfv.XMaxClipGuardband,
81 &sfv.YMinClipGuardband,
82 &sfv.YMaxClipGuardband);
83 }
84
85 GENX(SF_CLIP_VIEWPORT_pack)(NULL, sf_clip_state.map + i * 64, &sfv);
86 }
87
88 anv_batch_emit(&cmd_buffer->batch,
89 GENX(3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP), clip) {
90 clip.SFClipViewportPointer = sf_clip_state.offset;
91 }
92 }
93
94 void
95 gen8_cmd_buffer_emit_depth_viewport(struct anv_cmd_buffer *cmd_buffer,
96 bool depth_clamp_enable)
97 {
98 uint32_t count = cmd_buffer->state.gfx.dynamic.viewport.count;
99 const VkViewport *viewports =
100 cmd_buffer->state.gfx.dynamic.viewport.viewports;
101 struct anv_state cc_state =
102 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, count * 8, 32);
103
104 for (uint32_t i = 0; i < count; i++) {
105 const VkViewport *vp = &viewports[i];
106
107 /* From the Vulkan spec:
108 *
109 * "It is valid for minDepth to be greater than or equal to
110 * maxDepth."
111 */
112 float min_depth = MIN2(vp->minDepth, vp->maxDepth);
113 float max_depth = MAX2(vp->minDepth, vp->maxDepth);
114
115 struct GENX(CC_VIEWPORT) cc_viewport = {
116 .MinimumDepth = depth_clamp_enable ? min_depth : 0.0f,
117 .MaximumDepth = depth_clamp_enable ? max_depth : 1.0f,
118 };
119
120 GENX(CC_VIEWPORT_pack)(NULL, cc_state.map + i * 8, &cc_viewport);
121 }
122
123 anv_batch_emit(&cmd_buffer->batch,
124 GENX(3DSTATE_VIEWPORT_STATE_POINTERS_CC), cc) {
125 cc.CCViewportPointer = cc_state.offset;
126 }
127 }
128 #endif
129
130 void
131 genX(cmd_buffer_enable_pma_fix)(struct anv_cmd_buffer *cmd_buffer, bool enable)
132 {
133 if (cmd_buffer->state.pma_fix_enabled == enable)
134 return;
135
136 cmd_buffer->state.pma_fix_enabled = enable;
137
138 /* According to the Broadwell PIPE_CONTROL documentation, software should
139 * emit a PIPE_CONTROL with the CS Stall and Depth Cache Flush bits set
140 * prior to the LRI. If stencil buffer writes are enabled, then a Render
141 * Cache Flush is also necessary.
142 *
143 * The Skylake docs say to use a depth stall rather than a command
144 * streamer stall. However, the hardware seems to violently disagree.
145 * A full command streamer stall seems to be needed in both cases.
146 */
147 anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
148 pc.DepthCacheFlushEnable = true;
149 pc.CommandStreamerStallEnable = true;
150 pc.RenderTargetCacheFlushEnable = true;
151 #if GEN_GEN >= 12
152 pc.TileCacheFlushEnable = true;
153
154 /* GEN:BUG:1409600907: "PIPE_CONTROL with Depth Stall Enable bit must
155 * be set with any PIPE_CONTROL with Depth Flush Enable bit set.
156 */
157 pc.DepthStallEnable = true;
158 #endif
159 }
160
161 #if GEN_GEN == 9
162
163 uint32_t cache_mode;
164 anv_pack_struct(&cache_mode, GENX(CACHE_MODE_0),
165 .STCPMAOptimizationEnable = enable,
166 .STCPMAOptimizationEnableMask = true);
167 anv_batch_emit(&cmd_buffer->batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
168 lri.RegisterOffset = GENX(CACHE_MODE_0_num);
169 lri.DataDWord = cache_mode;
170 }
171
172 #elif GEN_GEN == 8
173
174 uint32_t cache_mode;
175 anv_pack_struct(&cache_mode, GENX(CACHE_MODE_1),
176 .NPPMAFixEnable = enable,
177 .NPEarlyZFailsDisable = enable,
178 .NPPMAFixEnableMask = true,
179 .NPEarlyZFailsDisableMask = true);
180 anv_batch_emit(&cmd_buffer->batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
181 lri.RegisterOffset = GENX(CACHE_MODE_1_num);
182 lri.DataDWord = cache_mode;
183 }
184
185 #endif /* GEN_GEN == 8 */
186
187 /* After the LRI, a PIPE_CONTROL with both the Depth Stall and Depth Cache
188 * Flush bits is often necessary. We do it regardless because it's easier.
189 * The render cache flush is also necessary if stencil writes are enabled.
190 *
191 * Again, the Skylake docs give a different set of flushes but the BDW
192 * flushes seem to work just as well.
193 */
194 anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
195 pc.DepthStallEnable = true;
196 pc.DepthCacheFlushEnable = true;
197 pc.RenderTargetCacheFlushEnable = true;
198 #if GEN_GEN >= 12
199 pc.TileCacheFlushEnable = true;
200 #endif
201 }
202 }
203
204 UNUSED static bool
205 want_depth_pma_fix(struct anv_cmd_buffer *cmd_buffer)
206 {
207 assert(GEN_GEN == 8);
208
209 /* From the Broadwell PRM Vol. 2c CACHE_MODE_1::NP_PMA_FIX_ENABLE:
210 *
211 * SW must set this bit in order to enable this fix when following
212 * expression is TRUE.
213 *
214 * 3DSTATE_WM::ForceThreadDispatch != 1 &&
215 * !(3DSTATE_RASTER::ForceSampleCount != NUMRASTSAMPLES_0) &&
216 * (3DSTATE_DEPTH_BUFFER::SURFACE_TYPE != NULL) &&
217 * (3DSTATE_DEPTH_BUFFER::HIZ Enable) &&
218 * !(3DSTATE_WM::EDSC_Mode == EDSC_PREPS) &&
219 * (3DSTATE_PS_EXTRA::PixelShaderValid) &&
220 * !(3DSTATE_WM_HZ_OP::DepthBufferClear ||
221 * 3DSTATE_WM_HZ_OP::DepthBufferResolve ||
222 * 3DSTATE_WM_HZ_OP::Hierarchical Depth Buffer Resolve Enable ||
223 * 3DSTATE_WM_HZ_OP::StencilBufferClear) &&
224 * (3DSTATE_WM_DEPTH_STENCIL::DepthTestEnable) &&
225 * (((3DSTATE_PS_EXTRA::PixelShaderKillsPixels ||
226 * 3DSTATE_PS_EXTRA::oMask Present to RenderTarget ||
227 * 3DSTATE_PS_BLEND::AlphaToCoverageEnable ||
228 * 3DSTATE_PS_BLEND::AlphaTestEnable ||
229 * 3DSTATE_WM_CHROMAKEY::ChromaKeyKillEnable) &&
230 * 3DSTATE_WM::ForceKillPix != ForceOff &&
231 * ((3DSTATE_WM_DEPTH_STENCIL::DepthWriteEnable &&
232 * 3DSTATE_DEPTH_BUFFER::DEPTH_WRITE_ENABLE) ||
233 * (3DSTATE_WM_DEPTH_STENCIL::Stencil Buffer Write Enable &&
234 * 3DSTATE_DEPTH_BUFFER::STENCIL_WRITE_ENABLE &&
235 * 3DSTATE_STENCIL_BUFFER::STENCIL_BUFFER_ENABLE))) ||
236 * (3DSTATE_PS_EXTRA:: Pixel Shader Computed Depth mode != PSCDEPTH_OFF))
237 */
238
239 /* These are always true:
240 * 3DSTATE_WM::ForceThreadDispatch != 1 &&
241 * !(3DSTATE_RASTER::ForceSampleCount != NUMRASTSAMPLES_0)
242 */
243
244 /* We only enable the PMA fix if we know for certain that HiZ is enabled.
245 * If we don't know whether HiZ is enabled or not, we disable the PMA fix
246 * and there is no harm.
247 *
248 * (3DSTATE_DEPTH_BUFFER::SURFACE_TYPE != NULL) &&
249 * 3DSTATE_DEPTH_BUFFER::HIZ Enable
250 */
251 if (!cmd_buffer->state.hiz_enabled)
252 return false;
253
254 /* 3DSTATE_PS_EXTRA::PixelShaderValid */
255 struct anv_graphics_pipeline *pipeline = cmd_buffer->state.gfx.pipeline;
256 if (!anv_pipeline_has_stage(pipeline, MESA_SHADER_FRAGMENT))
257 return false;
258
259 /* !(3DSTATE_WM::EDSC_Mode == EDSC_PREPS) */
260 const struct brw_wm_prog_data *wm_prog_data = get_wm_prog_data(pipeline);
261 if (wm_prog_data->early_fragment_tests)
262 return false;
263
264 /* We never use anv_pipeline for HiZ ops so this is trivially true:
265 * !(3DSTATE_WM_HZ_OP::DepthBufferClear ||
266 * 3DSTATE_WM_HZ_OP::DepthBufferResolve ||
267 * 3DSTATE_WM_HZ_OP::Hierarchical Depth Buffer Resolve Enable ||
268 * 3DSTATE_WM_HZ_OP::StencilBufferClear)
269 */
270
271 /* 3DSTATE_WM_DEPTH_STENCIL::DepthTestEnable */
272 if (!pipeline->depth_test_enable)
273 return false;
274
275 /* (((3DSTATE_PS_EXTRA::PixelShaderKillsPixels ||
276 * 3DSTATE_PS_EXTRA::oMask Present to RenderTarget ||
277 * 3DSTATE_PS_BLEND::AlphaToCoverageEnable ||
278 * 3DSTATE_PS_BLEND::AlphaTestEnable ||
279 * 3DSTATE_WM_CHROMAKEY::ChromaKeyKillEnable) &&
280 * 3DSTATE_WM::ForceKillPix != ForceOff &&
281 * ((3DSTATE_WM_DEPTH_STENCIL::DepthWriteEnable &&
282 * 3DSTATE_DEPTH_BUFFER::DEPTH_WRITE_ENABLE) ||
283 * (3DSTATE_WM_DEPTH_STENCIL::Stencil Buffer Write Enable &&
284 * 3DSTATE_DEPTH_BUFFER::STENCIL_WRITE_ENABLE &&
285 * 3DSTATE_STENCIL_BUFFER::STENCIL_BUFFER_ENABLE))) ||
286 * (3DSTATE_PS_EXTRA:: Pixel Shader Computed Depth mode != PSCDEPTH_OFF))
287 */
288 return (pipeline->kill_pixel && (pipeline->writes_depth ||
289 pipeline->writes_stencil)) ||
290 wm_prog_data->computed_depth_mode != PSCDEPTH_OFF;
291 }
292
293 UNUSED static bool
294 want_stencil_pma_fix(struct anv_cmd_buffer *cmd_buffer)
295 {
296 if (GEN_GEN > 9)
297 return false;
298 assert(GEN_GEN == 9);
299
300 /* From the Skylake PRM Vol. 2c CACHE_MODE_1::STC PMA Optimization Enable:
301 *
302 * Clearing this bit will force the STC cache to wait for pending
303 * retirement of pixels at the HZ-read stage and do the STC-test for
304 * Non-promoted, R-computed and Computed depth modes instead of
305 * postponing the STC-test to RCPFE.
306 *
307 * STC_TEST_EN = 3DSTATE_STENCIL_BUFFER::STENCIL_BUFFER_ENABLE &&
308 * 3DSTATE_WM_DEPTH_STENCIL::StencilTestEnable
309 *
310 * STC_WRITE_EN = 3DSTATE_STENCIL_BUFFER::STENCIL_BUFFER_ENABLE &&
311 * (3DSTATE_WM_DEPTH_STENCIL::Stencil Buffer Write Enable &&
312 * 3DSTATE_DEPTH_BUFFER::STENCIL_WRITE_ENABLE)
313 *
314 * COMP_STC_EN = STC_TEST_EN &&
315 * 3DSTATE_PS_EXTRA::PixelShaderComputesStencil
316 *
317 * SW parses the pipeline states to generate the following logical
318 * signal indicating if PMA FIX can be enabled.
319 *
320 * STC_PMA_OPT =
321 * 3DSTATE_WM::ForceThreadDispatch != 1 &&
322 * !(3DSTATE_RASTER::ForceSampleCount != NUMRASTSAMPLES_0) &&
323 * 3DSTATE_DEPTH_BUFFER::SURFACE_TYPE != NULL &&
324 * 3DSTATE_DEPTH_BUFFER::HIZ Enable &&
325 * !(3DSTATE_WM::EDSC_Mode == 2) &&
326 * 3DSTATE_PS_EXTRA::PixelShaderValid &&
327 * !(3DSTATE_WM_HZ_OP::DepthBufferClear ||
328 * 3DSTATE_WM_HZ_OP::DepthBufferResolve ||
329 * 3DSTATE_WM_HZ_OP::Hierarchical Depth Buffer Resolve Enable ||
330 * 3DSTATE_WM_HZ_OP::StencilBufferClear) &&
331 * (COMP_STC_EN || STC_WRITE_EN) &&
332 * ((3DSTATE_PS_EXTRA::PixelShaderKillsPixels ||
333 * 3DSTATE_WM::ForceKillPix == ON ||
334 * 3DSTATE_PS_EXTRA::oMask Present to RenderTarget ||
335 * 3DSTATE_PS_BLEND::AlphaToCoverageEnable ||
336 * 3DSTATE_PS_BLEND::AlphaTestEnable ||
337 * 3DSTATE_WM_CHROMAKEY::ChromaKeyKillEnable) ||
338 * (3DSTATE_PS_EXTRA::Pixel Shader Computed Depth mode != PSCDEPTH_OFF))
339 */
340
341 /* These are always true:
342 * 3DSTATE_WM::ForceThreadDispatch != 1 &&
343 * !(3DSTATE_RASTER::ForceSampleCount != NUMRASTSAMPLES_0)
344 */
345
346 /* We only enable the PMA fix if we know for certain that HiZ is enabled.
347 * If we don't know whether HiZ is enabled or not, we disable the PMA fix
348 * and there is no harm.
349 *
350 * (3DSTATE_DEPTH_BUFFER::SURFACE_TYPE != NULL) &&
351 * 3DSTATE_DEPTH_BUFFER::HIZ Enable
352 */
353 if (!cmd_buffer->state.hiz_enabled)
354 return false;
355
356 /* We can't possibly know if HiZ is enabled without the framebuffer */
357 assert(cmd_buffer->state.framebuffer);
358
359 /* HiZ is enabled so we had better have a depth buffer with HiZ */
360 const struct anv_image_view *ds_iview =
361 anv_cmd_buffer_get_depth_stencil_view(cmd_buffer);
362 assert(ds_iview && ds_iview->image->planes[0].aux_usage == ISL_AUX_USAGE_HIZ);
363
364 /* 3DSTATE_PS_EXTRA::PixelShaderValid */
365 struct anv_graphics_pipeline *pipeline = cmd_buffer->state.gfx.pipeline;
366 if (!anv_pipeline_has_stage(pipeline, MESA_SHADER_FRAGMENT))
367 return false;
368
369 /* !(3DSTATE_WM::EDSC_Mode == 2) */
370 const struct brw_wm_prog_data *wm_prog_data = get_wm_prog_data(pipeline);
371 if (wm_prog_data->early_fragment_tests)
372 return false;
373
374 /* We never use anv_pipeline for HiZ ops so this is trivially true:
375 * !(3DSTATE_WM_HZ_OP::DepthBufferClear ||
376 * 3DSTATE_WM_HZ_OP::DepthBufferResolve ||
377 * 3DSTATE_WM_HZ_OP::Hierarchical Depth Buffer Resolve Enable ||
378 * 3DSTATE_WM_HZ_OP::StencilBufferClear)
379 */
380
381 /* 3DSTATE_STENCIL_BUFFER::STENCIL_BUFFER_ENABLE &&
382 * 3DSTATE_WM_DEPTH_STENCIL::StencilTestEnable
383 */
384 const bool stc_test_en =
385 (ds_iview->image->aspects & VK_IMAGE_ASPECT_STENCIL_BIT) &&
386 pipeline->stencil_test_enable;
387
388 /* 3DSTATE_STENCIL_BUFFER::STENCIL_BUFFER_ENABLE &&
389 * (3DSTATE_WM_DEPTH_STENCIL::Stencil Buffer Write Enable &&
390 * 3DSTATE_DEPTH_BUFFER::STENCIL_WRITE_ENABLE)
391 */
392 const bool stc_write_en =
393 (ds_iview->image->aspects & VK_IMAGE_ASPECT_STENCIL_BIT) &&
394 (cmd_buffer->state.gfx.dynamic.stencil_write_mask.front ||
395 cmd_buffer->state.gfx.dynamic.stencil_write_mask.back) &&
396 pipeline->writes_stencil;
397
398 /* STC_TEST_EN && 3DSTATE_PS_EXTRA::PixelShaderComputesStencil */
399 const bool comp_stc_en = stc_test_en && wm_prog_data->computed_stencil;
400
401 /* COMP_STC_EN || STC_WRITE_EN */
402 if (!(comp_stc_en || stc_write_en))
403 return false;
404
405 /* (3DSTATE_PS_EXTRA::PixelShaderKillsPixels ||
406 * 3DSTATE_WM::ForceKillPix == ON ||
407 * 3DSTATE_PS_EXTRA::oMask Present to RenderTarget ||
408 * 3DSTATE_PS_BLEND::AlphaToCoverageEnable ||
409 * 3DSTATE_PS_BLEND::AlphaTestEnable ||
410 * 3DSTATE_WM_CHROMAKEY::ChromaKeyKillEnable) ||
411 * (3DSTATE_PS_EXTRA::Pixel Shader Computed Depth mode != PSCDEPTH_OFF)
412 */
413 return pipeline->kill_pixel ||
414 wm_prog_data->computed_depth_mode != PSCDEPTH_OFF;
415 }
416
417 void
418 genX(cmd_buffer_flush_dynamic_state)(struct anv_cmd_buffer *cmd_buffer)
419 {
420 struct anv_graphics_pipeline *pipeline = cmd_buffer->state.gfx.pipeline;
421 struct anv_dynamic_state *d = &cmd_buffer->state.gfx.dynamic;
422
423 if (cmd_buffer->state.gfx.dirty & (ANV_CMD_DIRTY_PIPELINE |
424 ANV_CMD_DIRTY_DYNAMIC_LINE_WIDTH)) {
425 uint32_t sf_dw[GENX(3DSTATE_SF_length)];
426 struct GENX(3DSTATE_SF) sf = {
427 GENX(3DSTATE_SF_header),
428 };
429 #if GEN_GEN == 8
430 if (cmd_buffer->device->info.is_cherryview) {
431 sf.CHVLineWidth = d->line_width;
432 } else {
433 sf.LineWidth = d->line_width;
434 }
435 #else
436 sf.LineWidth = d->line_width,
437 #endif
438 GENX(3DSTATE_SF_pack)(NULL, sf_dw, &sf);
439 anv_batch_emit_merge(&cmd_buffer->batch, sf_dw, pipeline->gen8.sf);
440 }
441
442 if (cmd_buffer->state.gfx.dirty & (ANV_CMD_DIRTY_PIPELINE |
443 ANV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS)){
444 uint32_t raster_dw[GENX(3DSTATE_RASTER_length)];
445 struct GENX(3DSTATE_RASTER) raster = {
446 GENX(3DSTATE_RASTER_header),
447 .GlobalDepthOffsetConstant = d->depth_bias.bias,
448 .GlobalDepthOffsetScale = d->depth_bias.slope,
449 .GlobalDepthOffsetClamp = d->depth_bias.clamp
450 };
451 GENX(3DSTATE_RASTER_pack)(NULL, raster_dw, &raster);
452 anv_batch_emit_merge(&cmd_buffer->batch, raster_dw,
453 pipeline->gen8.raster);
454 }
455
456 /* Stencil reference values moved from COLOR_CALC_STATE in gen8 to
457 * 3DSTATE_WM_DEPTH_STENCIL in gen9. That means the dirty bits gets split
458 * across different state packets for gen8 and gen9. We handle that by
459 * using a big old #if switch here.
460 */
461 #if GEN_GEN == 8
462 if (cmd_buffer->state.gfx.dirty & (ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS |
463 ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE)) {
464 struct anv_state cc_state =
465 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer,
466 GENX(COLOR_CALC_STATE_length) * 4,
467 64);
468 struct GENX(COLOR_CALC_STATE) cc = {
469 .BlendConstantColorRed = d->blend_constants[0],
470 .BlendConstantColorGreen = d->blend_constants[1],
471 .BlendConstantColorBlue = d->blend_constants[2],
472 .BlendConstantColorAlpha = d->blend_constants[3],
473 .StencilReferenceValue = d->stencil_reference.front & 0xff,
474 .BackfaceStencilReferenceValue = d->stencil_reference.back & 0xff,
475 };
476 GENX(COLOR_CALC_STATE_pack)(NULL, cc_state.map, &cc);
477
478 anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_CC_STATE_POINTERS), ccp) {
479 ccp.ColorCalcStatePointer = cc_state.offset;
480 ccp.ColorCalcStatePointerValid = true;
481 }
482 }
483
484 if (cmd_buffer->state.gfx.dirty & (ANV_CMD_DIRTY_PIPELINE |
485 ANV_CMD_DIRTY_RENDER_TARGETS |
486 ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK |
487 ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK)) {
488 uint32_t wm_depth_stencil_dw[GENX(3DSTATE_WM_DEPTH_STENCIL_length)];
489
490 struct GENX(3DSTATE_WM_DEPTH_STENCIL wm_depth_stencil) = {
491 GENX(3DSTATE_WM_DEPTH_STENCIL_header),
492
493 .StencilTestMask = d->stencil_compare_mask.front & 0xff,
494 .StencilWriteMask = d->stencil_write_mask.front & 0xff,
495
496 .BackfaceStencilTestMask = d->stencil_compare_mask.back & 0xff,
497 .BackfaceStencilWriteMask = d->stencil_write_mask.back & 0xff,
498
499 .StencilBufferWriteEnable =
500 (d->stencil_write_mask.front || d->stencil_write_mask.back) &&
501 pipeline->writes_stencil,
502 };
503 GENX(3DSTATE_WM_DEPTH_STENCIL_pack)(NULL, wm_depth_stencil_dw,
504 &wm_depth_stencil);
505
506 anv_batch_emit_merge(&cmd_buffer->batch, wm_depth_stencil_dw,
507 pipeline->gen8.wm_depth_stencil);
508
509 genX(cmd_buffer_enable_pma_fix)(cmd_buffer,
510 want_depth_pma_fix(cmd_buffer));
511 }
512 #else
513 if (cmd_buffer->state.gfx.dirty & ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS) {
514 struct anv_state cc_state =
515 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer,
516 GENX(COLOR_CALC_STATE_length) * 4,
517 64);
518 struct GENX(COLOR_CALC_STATE) cc = {
519 .BlendConstantColorRed = d->blend_constants[0],
520 .BlendConstantColorGreen = d->blend_constants[1],
521 .BlendConstantColorBlue = d->blend_constants[2],
522 .BlendConstantColorAlpha = d->blend_constants[3],
523 };
524 GENX(COLOR_CALC_STATE_pack)(NULL, cc_state.map, &cc);
525
526 anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_CC_STATE_POINTERS), ccp) {
527 ccp.ColorCalcStatePointer = cc_state.offset;
528 ccp.ColorCalcStatePointerValid = true;
529 }
530 }
531
532 if (cmd_buffer->state.gfx.dirty & (ANV_CMD_DIRTY_PIPELINE |
533 ANV_CMD_DIRTY_RENDER_TARGETS |
534 ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK |
535 ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK |
536 ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE)) {
537 uint32_t dwords[GENX(3DSTATE_WM_DEPTH_STENCIL_length)];
538 struct GENX(3DSTATE_WM_DEPTH_STENCIL) wm_depth_stencil = {
539 GENX(3DSTATE_WM_DEPTH_STENCIL_header),
540
541 .StencilTestMask = d->stencil_compare_mask.front & 0xff,
542 .StencilWriteMask = d->stencil_write_mask.front & 0xff,
543
544 .BackfaceStencilTestMask = d->stencil_compare_mask.back & 0xff,
545 .BackfaceStencilWriteMask = d->stencil_write_mask.back & 0xff,
546
547 .StencilReferenceValue = d->stencil_reference.front & 0xff,
548 .BackfaceStencilReferenceValue = d->stencil_reference.back & 0xff,
549
550 .StencilBufferWriteEnable =
551 (d->stencil_write_mask.front || d->stencil_write_mask.back) &&
552 pipeline->writes_stencil,
553 };
554 GENX(3DSTATE_WM_DEPTH_STENCIL_pack)(NULL, dwords, &wm_depth_stencil);
555
556 anv_batch_emit_merge(&cmd_buffer->batch, dwords,
557 pipeline->gen9.wm_depth_stencil);
558
559 genX(cmd_buffer_enable_pma_fix)(cmd_buffer,
560 want_stencil_pma_fix(cmd_buffer));
561 }
562 #endif
563
564 #if GEN_GEN >= 12
565 if(cmd_buffer->state.gfx.dirty & (ANV_CMD_DIRTY_PIPELINE |
566 ANV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS)) {
567 anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_DEPTH_BOUNDS), db) {
568 db.DepthBoundsTestValueModifyDisable = false;
569 db.DepthBoundsTestEnableModifyDisable = false;
570 db.DepthBoundsTestEnable = pipeline->depth_bounds_test_enable;
571 db.DepthBoundsTestMinValue = d->depth_bounds.min;
572 db.DepthBoundsTestMaxValue = d->depth_bounds.max;
573 }
574 }
575 #endif
576
577 if (cmd_buffer->state.gfx.dirty & ANV_CMD_DIRTY_DYNAMIC_LINE_STIPPLE) {
578 anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_LINE_STIPPLE), ls) {
579 ls.LineStipplePattern = d->line_stipple.pattern;
580 ls.LineStippleInverseRepeatCount =
581 1.0f / MAX2(1, d->line_stipple.factor);
582 ls.LineStippleRepeatCount = d->line_stipple.factor;
583 }
584 }
585
586 if (cmd_buffer->state.gfx.dirty & (ANV_CMD_DIRTY_PIPELINE |
587 ANV_CMD_DIRTY_INDEX_BUFFER)) {
588 anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_VF), vf) {
589 vf.IndexedDrawCutIndexEnable = pipeline->primitive_restart;
590 vf.CutIndex = cmd_buffer->state.restart_index;
591 }
592 }
593
594 cmd_buffer->state.gfx.dirty = 0;
595 }
596
597 static uint32_t vk_to_gen_index_type(VkIndexType type)
598 {
599 switch (type) {
600 case VK_INDEX_TYPE_UINT8_EXT:
601 return INDEX_BYTE;
602 case VK_INDEX_TYPE_UINT16:
603 return INDEX_WORD;
604 case VK_INDEX_TYPE_UINT32:
605 return INDEX_DWORD;
606 default:
607 unreachable("invalid index type");
608 }
609 }
610
611 static uint32_t restart_index_for_type(VkIndexType type)
612 {
613 switch (type) {
614 case VK_INDEX_TYPE_UINT8_EXT:
615 return UINT8_MAX;
616 case VK_INDEX_TYPE_UINT16:
617 return UINT16_MAX;
618 case VK_INDEX_TYPE_UINT32:
619 return UINT32_MAX;
620 default:
621 unreachable("invalid index type");
622 }
623 }
624
625 void genX(CmdBindIndexBuffer)(
626 VkCommandBuffer commandBuffer,
627 VkBuffer _buffer,
628 VkDeviceSize offset,
629 VkIndexType indexType)
630 {
631 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
632 ANV_FROM_HANDLE(anv_buffer, buffer, _buffer);
633
634 cmd_buffer->state.restart_index = restart_index_for_type(indexType);
635
636 anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_INDEX_BUFFER), ib) {
637 ib.IndexFormat = vk_to_gen_index_type(indexType);
638 ib.MOCS = anv_mocs_for_bo(cmd_buffer->device,
639 buffer->address.bo);
640 ib.BufferStartingAddress = anv_address_add(buffer->address, offset);
641 ib.BufferSize = buffer->size - offset;
642 }
643
644 cmd_buffer->state.gfx.dirty |= ANV_CMD_DIRTY_INDEX_BUFFER;
645 }