anv/dump: Fix vkCmdPipelineBarrier flags
[mesa.git] / src / intel / vulkan / genX_pipeline_util.h
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 "vk_format_info.h"
25
26 static uint32_t
27 vertex_element_comp_control(enum isl_format format, unsigned comp)
28 {
29 uint8_t bits;
30 switch (comp) {
31 case 0: bits = isl_format_layouts[format].channels.r.bits; break;
32 case 1: bits = isl_format_layouts[format].channels.g.bits; break;
33 case 2: bits = isl_format_layouts[format].channels.b.bits; break;
34 case 3: bits = isl_format_layouts[format].channels.a.bits; break;
35 default: unreachable("Invalid component");
36 }
37
38 if (bits) {
39 return VFCOMP_STORE_SRC;
40 } else if (comp < 3) {
41 return VFCOMP_STORE_0;
42 } else if (isl_format_layouts[format].channels.r.type == ISL_UINT ||
43 isl_format_layouts[format].channels.r.type == ISL_SINT) {
44 assert(comp == 3);
45 return VFCOMP_STORE_1_INT;
46 } else {
47 assert(comp == 3);
48 return VFCOMP_STORE_1_FP;
49 }
50 }
51
52 static void
53 emit_vertex_input(struct anv_pipeline *pipeline,
54 const VkPipelineVertexInputStateCreateInfo *info,
55 const struct anv_graphics_pipeline_create_info *extra)
56 {
57 const struct brw_vs_prog_data *vs_prog_data = get_vs_prog_data(pipeline);
58
59 uint32_t elements;
60 if (extra && extra->disable_vs) {
61 /* If the VS is disabled, just assume the user knows what they're
62 * doing and apply the layout blindly. This can only come from
63 * meta, so this *should* be safe.
64 */
65 elements = 0;
66 for (uint32_t i = 0; i < info->vertexAttributeDescriptionCount; i++)
67 elements |= (1 << info->pVertexAttributeDescriptions[i].location);
68 } else {
69 /* Pull inputs_read out of the VS prog data */
70 uint64_t inputs_read = vs_prog_data->inputs_read;
71 assert((inputs_read & ((1 << VERT_ATTRIB_GENERIC0) - 1)) == 0);
72 elements = inputs_read >> VERT_ATTRIB_GENERIC0;
73 }
74
75 #if GEN_GEN >= 8
76 /* On BDW+, we only need to allocate space for base ids. Setting up
77 * the actual vertex and instance id is a separate packet.
78 */
79 const bool needs_svgs_elem = vs_prog_data->uses_basevertex ||
80 vs_prog_data->uses_baseinstance;
81 #else
82 /* On Haswell and prior, vertex and instance id are created by using the
83 * ComponentControl fields, so we need an element for any of them.
84 */
85 const bool needs_svgs_elem = vs_prog_data->uses_vertexid ||
86 vs_prog_data->uses_instanceid ||
87 vs_prog_data->uses_basevertex ||
88 vs_prog_data->uses_baseinstance;
89 #endif
90
91 uint32_t elem_count = __builtin_popcount(elements) + needs_svgs_elem;
92 if (elem_count == 0)
93 return;
94
95 uint32_t *p;
96
97 const uint32_t num_dwords = 1 + elem_count * 2;
98 p = anv_batch_emitn(&pipeline->batch, num_dwords,
99 GENX(3DSTATE_VERTEX_ELEMENTS));
100 memset(p + 1, 0, (num_dwords - 1) * 4);
101
102 for (uint32_t i = 0; i < info->vertexAttributeDescriptionCount; i++) {
103 const VkVertexInputAttributeDescription *desc =
104 &info->pVertexAttributeDescriptions[i];
105 enum isl_format format = anv_get_isl_format(&pipeline->device->info,
106 desc->format,
107 VK_IMAGE_ASPECT_COLOR_BIT,
108 VK_IMAGE_TILING_LINEAR);
109
110 assert(desc->binding < 32);
111
112 if ((elements & (1 << desc->location)) == 0)
113 continue; /* Binding unused */
114
115 uint32_t slot = __builtin_popcount(elements & ((1 << desc->location) - 1));
116
117 struct GENX(VERTEX_ELEMENT_STATE) element = {
118 .VertexBufferIndex = desc->binding,
119 .Valid = true,
120 .SourceElementFormat = format,
121 .EdgeFlagEnable = false,
122 .SourceElementOffset = desc->offset,
123 .Component0Control = vertex_element_comp_control(format, 0),
124 .Component1Control = vertex_element_comp_control(format, 1),
125 .Component2Control = vertex_element_comp_control(format, 2),
126 .Component3Control = vertex_element_comp_control(format, 3),
127 };
128 GENX(VERTEX_ELEMENT_STATE_pack)(NULL, &p[1 + slot * 2], &element);
129
130 #if GEN_GEN >= 8
131 /* On Broadwell and later, we have a separate VF_INSTANCING packet
132 * that controls instancing. On Haswell and prior, that's part of
133 * VERTEX_BUFFER_STATE which we emit later.
134 */
135 anv_batch_emit(&pipeline->batch, GENX(3DSTATE_VF_INSTANCING), vfi) {
136 vfi.InstancingEnable = pipeline->instancing_enable[desc->binding],
137 vfi.VertexElementIndex = slot,
138 /* Vulkan so far doesn't have an instance divisor, so
139 * this is always 1 (ignored if not instancing). */
140 vfi.InstanceDataStepRate = 1;
141 }
142 #endif
143 }
144
145 const uint32_t id_slot = __builtin_popcount(elements);
146 if (needs_svgs_elem) {
147 /* From the Broadwell PRM for the 3D_Vertex_Component_Control enum:
148 * "Within a VERTEX_ELEMENT_STATE structure, if a Component
149 * Control field is set to something other than VFCOMP_STORE_SRC,
150 * no higher-numbered Component Control fields may be set to
151 * VFCOMP_STORE_SRC"
152 *
153 * This means, that if we have BaseInstance, we need BaseVertex as
154 * well. Just do all or nothing.
155 */
156 uint32_t base_ctrl = (vs_prog_data->uses_basevertex ||
157 vs_prog_data->uses_baseinstance) ?
158 VFCOMP_STORE_SRC : VFCOMP_STORE_0;
159
160 struct GENX(VERTEX_ELEMENT_STATE) element = {
161 .VertexBufferIndex = 32, /* Reserved for this */
162 .Valid = true,
163 .SourceElementFormat = ISL_FORMAT_R32G32_UINT,
164 .Component0Control = base_ctrl,
165 .Component1Control = base_ctrl,
166 #if GEN_GEN >= 8
167 .Component2Control = VFCOMP_STORE_0,
168 .Component3Control = VFCOMP_STORE_0,
169 #else
170 .Component2Control = VFCOMP_STORE_VID,
171 .Component3Control = VFCOMP_STORE_IID,
172 #endif
173 };
174 GENX(VERTEX_ELEMENT_STATE_pack)(NULL, &p[1 + id_slot * 2], &element);
175 }
176
177 #if GEN_GEN >= 8
178 anv_batch_emit(&pipeline->batch, GENX(3DSTATE_VF_SGVS), sgvs) {
179 sgvs.VertexIDEnable = vs_prog_data->uses_vertexid;
180 sgvs.VertexIDComponentNumber = 2;
181 sgvs.VertexIDElementOffset = id_slot;
182 sgvs.InstanceIDEnable = vs_prog_data->uses_instanceid;
183 sgvs.InstanceIDComponentNumber = 3;
184 sgvs.InstanceIDElementOffset = id_slot;
185 }
186 #endif
187 }
188
189 static inline void
190 emit_urb_setup(struct anv_pipeline *pipeline)
191 {
192 #if GEN_GEN == 7 && !GEN_IS_HASWELL
193 struct anv_device *device = pipeline->device;
194
195 /* From the IVB PRM Vol. 2, Part 1, Section 3.2.1:
196 *
197 * "A PIPE_CONTROL with Post-Sync Operation set to 1h and a depth stall
198 * needs to be sent just prior to any 3DSTATE_VS, 3DSTATE_URB_VS,
199 * 3DSTATE_CONSTANT_VS, 3DSTATE_BINDING_TABLE_POINTER_VS,
200 * 3DSTATE_SAMPLER_STATE_POINTER_VS command. Only one PIPE_CONTROL
201 * needs to be sent before any combination of VS associated 3DSTATE."
202 */
203 anv_batch_emit(&pipeline->batch, GEN7_PIPE_CONTROL, pc) {
204 pc.DepthStallEnable = true;
205 pc.PostSyncOperation = WriteImmediateData;
206 pc.Address = (struct anv_address) { &device->workaround_bo, 0 };
207 }
208 #endif
209
210 for (int i = MESA_SHADER_VERTEX; i <= MESA_SHADER_GEOMETRY; i++) {
211 anv_batch_emit(&pipeline->batch, GENX(3DSTATE_URB_VS), urb) {
212 urb._3DCommandSubOpcode = 48 + i;
213 urb.VSURBStartingAddress = pipeline->urb.start[i];
214 urb.VSURBEntryAllocationSize = pipeline->urb.size[i] - 1;
215 urb.VSNumberofURBEntries = pipeline->urb.entries[i];
216 }
217 }
218 }
219
220 static void
221 emit_3dstate_sbe(struct anv_pipeline *pipeline)
222 {
223 const struct brw_vs_prog_data *vs_prog_data = get_vs_prog_data(pipeline);
224 const struct brw_gs_prog_data *gs_prog_data = get_gs_prog_data(pipeline);
225 const struct brw_wm_prog_data *wm_prog_data = get_wm_prog_data(pipeline);
226 const struct brw_vue_map *fs_input_map;
227
228 if (pipeline->gs_kernel == NO_KERNEL)
229 fs_input_map = &vs_prog_data->base.vue_map;
230 else
231 fs_input_map = &gs_prog_data->base.vue_map;
232
233 struct GENX(3DSTATE_SBE) sbe = {
234 GENX(3DSTATE_SBE_header),
235 .AttributeSwizzleEnable = true,
236 .PointSpriteTextureCoordinateOrigin = UPPERLEFT,
237 .NumberofSFOutputAttributes = wm_prog_data->num_varying_inputs,
238 .ConstantInterpolationEnable = wm_prog_data->flat_inputs,
239
240 #if GEN_GEN >= 9
241 .Attribute0ActiveComponentFormat = ACF_XYZW,
242 .Attribute1ActiveComponentFormat = ACF_XYZW,
243 .Attribute2ActiveComponentFormat = ACF_XYZW,
244 .Attribute3ActiveComponentFormat = ACF_XYZW,
245 .Attribute4ActiveComponentFormat = ACF_XYZW,
246 .Attribute5ActiveComponentFormat = ACF_XYZW,
247 .Attribute6ActiveComponentFormat = ACF_XYZW,
248 .Attribute7ActiveComponentFormat = ACF_XYZW,
249 .Attribute8ActiveComponentFormat = ACF_XYZW,
250 .Attribute9ActiveComponentFormat = ACF_XYZW,
251 .Attribute10ActiveComponentFormat = ACF_XYZW,
252 .Attribute11ActiveComponentFormat = ACF_XYZW,
253 .Attribute12ActiveComponentFormat = ACF_XYZW,
254 .Attribute13ActiveComponentFormat = ACF_XYZW,
255 .Attribute14ActiveComponentFormat = ACF_XYZW,
256 .Attribute15ActiveComponentFormat = ACF_XYZW,
257 /* wow, much field, very attribute */
258 .Attribute16ActiveComponentFormat = ACF_XYZW,
259 .Attribute17ActiveComponentFormat = ACF_XYZW,
260 .Attribute18ActiveComponentFormat = ACF_XYZW,
261 .Attribute19ActiveComponentFormat = ACF_XYZW,
262 .Attribute20ActiveComponentFormat = ACF_XYZW,
263 .Attribute21ActiveComponentFormat = ACF_XYZW,
264 .Attribute22ActiveComponentFormat = ACF_XYZW,
265 .Attribute23ActiveComponentFormat = ACF_XYZW,
266 .Attribute24ActiveComponentFormat = ACF_XYZW,
267 .Attribute25ActiveComponentFormat = ACF_XYZW,
268 .Attribute26ActiveComponentFormat = ACF_XYZW,
269 .Attribute27ActiveComponentFormat = ACF_XYZW,
270 .Attribute28ActiveComponentFormat = ACF_XYZW,
271 .Attribute29ActiveComponentFormat = ACF_XYZW,
272 .Attribute28ActiveComponentFormat = ACF_XYZW,
273 .Attribute29ActiveComponentFormat = ACF_XYZW,
274 .Attribute30ActiveComponentFormat = ACF_XYZW,
275 #endif
276 };
277
278 #if GEN_GEN >= 8
279 /* On Broadwell, they broke 3DSTATE_SBE into two packets */
280 struct GENX(3DSTATE_SBE_SWIZ) swiz = {
281 GENX(3DSTATE_SBE_SWIZ_header),
282 };
283 #else
284 # define swiz sbe
285 #endif
286
287 int max_source_attr = 0;
288 for (int attr = 0; attr < VARYING_SLOT_MAX; attr++) {
289 int input_index = wm_prog_data->urb_setup[attr];
290
291 if (input_index < 0)
292 continue;
293
294 const int slot = fs_input_map->varying_to_slot[attr];
295
296 if (input_index >= 16)
297 continue;
298
299 if (slot == -1) {
300 /* This attribute does not exist in the VUE--that means that the
301 * vertex shader did not write to it. It could be that it's a
302 * regular varying read by the fragment shader but not written by
303 * the vertex shader or it's gl_PrimitiveID. In the first case the
304 * value is undefined, in the second it needs to be
305 * gl_PrimitiveID.
306 */
307 swiz.Attribute[input_index].ConstantSource = PRIM_ID;
308 swiz.Attribute[input_index].ComponentOverrideX = true;
309 swiz.Attribute[input_index].ComponentOverrideY = true;
310 swiz.Attribute[input_index].ComponentOverrideZ = true;
311 swiz.Attribute[input_index].ComponentOverrideW = true;
312 } else {
313 assert(slot >= 2);
314 const int source_attr = slot - 2;
315 max_source_attr = MAX2(max_source_attr, source_attr);
316 /* We have to subtract two slots to accout for the URB entry output
317 * read offset in the VS and GS stages.
318 */
319 swiz.Attribute[input_index].SourceAttribute = source_attr;
320 }
321 }
322
323 sbe.VertexURBEntryReadOffset = 1; /* Skip the VUE header and position slots */
324 sbe.VertexURBEntryReadLength = DIV_ROUND_UP(max_source_attr + 1, 2);
325
326 uint32_t *dw = anv_batch_emit_dwords(&pipeline->batch,
327 GENX(3DSTATE_SBE_length));
328 GENX(3DSTATE_SBE_pack)(&pipeline->batch, dw, &sbe);
329
330 #if GEN_GEN >= 8
331 dw = anv_batch_emit_dwords(&pipeline->batch, GENX(3DSTATE_SBE_SWIZ_length));
332 GENX(3DSTATE_SBE_SWIZ_pack)(&pipeline->batch, dw, &swiz);
333 #endif
334 }
335
336 static inline uint32_t
337 scratch_space(const struct brw_stage_prog_data *prog_data)
338 {
339 return ffs(prog_data->total_scratch / 2048);
340 }
341
342 static const uint32_t vk_to_gen_cullmode[] = {
343 [VK_CULL_MODE_NONE] = CULLMODE_NONE,
344 [VK_CULL_MODE_FRONT_BIT] = CULLMODE_FRONT,
345 [VK_CULL_MODE_BACK_BIT] = CULLMODE_BACK,
346 [VK_CULL_MODE_FRONT_AND_BACK] = CULLMODE_BOTH
347 };
348
349 static const uint32_t vk_to_gen_fillmode[] = {
350 [VK_POLYGON_MODE_FILL] = FILL_MODE_SOLID,
351 [VK_POLYGON_MODE_LINE] = FILL_MODE_WIREFRAME,
352 [VK_POLYGON_MODE_POINT] = FILL_MODE_POINT,
353 };
354
355 static const uint32_t vk_to_gen_front_face[] = {
356 [VK_FRONT_FACE_COUNTER_CLOCKWISE] = 1,
357 [VK_FRONT_FACE_CLOCKWISE] = 0
358 };
359
360 static const uint32_t vk_to_gen_logic_op[] = {
361 [VK_LOGIC_OP_COPY] = LOGICOP_COPY,
362 [VK_LOGIC_OP_CLEAR] = LOGICOP_CLEAR,
363 [VK_LOGIC_OP_AND] = LOGICOP_AND,
364 [VK_LOGIC_OP_AND_REVERSE] = LOGICOP_AND_REVERSE,
365 [VK_LOGIC_OP_AND_INVERTED] = LOGICOP_AND_INVERTED,
366 [VK_LOGIC_OP_NO_OP] = LOGICOP_NOOP,
367 [VK_LOGIC_OP_XOR] = LOGICOP_XOR,
368 [VK_LOGIC_OP_OR] = LOGICOP_OR,
369 [VK_LOGIC_OP_NOR] = LOGICOP_NOR,
370 [VK_LOGIC_OP_EQUIVALENT] = LOGICOP_EQUIV,
371 [VK_LOGIC_OP_INVERT] = LOGICOP_INVERT,
372 [VK_LOGIC_OP_OR_REVERSE] = LOGICOP_OR_REVERSE,
373 [VK_LOGIC_OP_COPY_INVERTED] = LOGICOP_COPY_INVERTED,
374 [VK_LOGIC_OP_OR_INVERTED] = LOGICOP_OR_INVERTED,
375 [VK_LOGIC_OP_NAND] = LOGICOP_NAND,
376 [VK_LOGIC_OP_SET] = LOGICOP_SET,
377 };
378
379 static const uint32_t vk_to_gen_blend[] = {
380 [VK_BLEND_FACTOR_ZERO] = BLENDFACTOR_ZERO,
381 [VK_BLEND_FACTOR_ONE] = BLENDFACTOR_ONE,
382 [VK_BLEND_FACTOR_SRC_COLOR] = BLENDFACTOR_SRC_COLOR,
383 [VK_BLEND_FACTOR_ONE_MINUS_SRC_COLOR] = BLENDFACTOR_INV_SRC_COLOR,
384 [VK_BLEND_FACTOR_DST_COLOR] = BLENDFACTOR_DST_COLOR,
385 [VK_BLEND_FACTOR_ONE_MINUS_DST_COLOR] = BLENDFACTOR_INV_DST_COLOR,
386 [VK_BLEND_FACTOR_SRC_ALPHA] = BLENDFACTOR_SRC_ALPHA,
387 [VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA] = BLENDFACTOR_INV_SRC_ALPHA,
388 [VK_BLEND_FACTOR_DST_ALPHA] = BLENDFACTOR_DST_ALPHA,
389 [VK_BLEND_FACTOR_ONE_MINUS_DST_ALPHA] = BLENDFACTOR_INV_DST_ALPHA,
390 [VK_BLEND_FACTOR_CONSTANT_COLOR] = BLENDFACTOR_CONST_COLOR,
391 [VK_BLEND_FACTOR_ONE_MINUS_CONSTANT_COLOR]= BLENDFACTOR_INV_CONST_COLOR,
392 [VK_BLEND_FACTOR_CONSTANT_ALPHA] = BLENDFACTOR_CONST_ALPHA,
393 [VK_BLEND_FACTOR_ONE_MINUS_CONSTANT_ALPHA]= BLENDFACTOR_INV_CONST_ALPHA,
394 [VK_BLEND_FACTOR_SRC_ALPHA_SATURATE] = BLENDFACTOR_SRC_ALPHA_SATURATE,
395 [VK_BLEND_FACTOR_SRC1_COLOR] = BLENDFACTOR_SRC1_COLOR,
396 [VK_BLEND_FACTOR_ONE_MINUS_SRC1_COLOR] = BLENDFACTOR_INV_SRC1_COLOR,
397 [VK_BLEND_FACTOR_SRC1_ALPHA] = BLENDFACTOR_SRC1_ALPHA,
398 [VK_BLEND_FACTOR_ONE_MINUS_SRC1_ALPHA] = BLENDFACTOR_INV_SRC1_ALPHA,
399 };
400
401 static const uint32_t vk_to_gen_blend_op[] = {
402 [VK_BLEND_OP_ADD] = BLENDFUNCTION_ADD,
403 [VK_BLEND_OP_SUBTRACT] = BLENDFUNCTION_SUBTRACT,
404 [VK_BLEND_OP_REVERSE_SUBTRACT] = BLENDFUNCTION_REVERSE_SUBTRACT,
405 [VK_BLEND_OP_MIN] = BLENDFUNCTION_MIN,
406 [VK_BLEND_OP_MAX] = BLENDFUNCTION_MAX,
407 };
408
409 static const uint32_t vk_to_gen_compare_op[] = {
410 [VK_COMPARE_OP_NEVER] = PREFILTEROPNEVER,
411 [VK_COMPARE_OP_LESS] = PREFILTEROPLESS,
412 [VK_COMPARE_OP_EQUAL] = PREFILTEROPEQUAL,
413 [VK_COMPARE_OP_LESS_OR_EQUAL] = PREFILTEROPLEQUAL,
414 [VK_COMPARE_OP_GREATER] = PREFILTEROPGREATER,
415 [VK_COMPARE_OP_NOT_EQUAL] = PREFILTEROPNOTEQUAL,
416 [VK_COMPARE_OP_GREATER_OR_EQUAL] = PREFILTEROPGEQUAL,
417 [VK_COMPARE_OP_ALWAYS] = PREFILTEROPALWAYS,
418 };
419
420 static const uint32_t vk_to_gen_stencil_op[] = {
421 [VK_STENCIL_OP_KEEP] = STENCILOP_KEEP,
422 [VK_STENCIL_OP_ZERO] = STENCILOP_ZERO,
423 [VK_STENCIL_OP_REPLACE] = STENCILOP_REPLACE,
424 [VK_STENCIL_OP_INCREMENT_AND_CLAMP] = STENCILOP_INCRSAT,
425 [VK_STENCIL_OP_DECREMENT_AND_CLAMP] = STENCILOP_DECRSAT,
426 [VK_STENCIL_OP_INVERT] = STENCILOP_INVERT,
427 [VK_STENCIL_OP_INCREMENT_AND_WRAP] = STENCILOP_INCR,
428 [VK_STENCIL_OP_DECREMENT_AND_WRAP] = STENCILOP_DECR,
429 };
430
431 static void
432 emit_ds_state(struct anv_pipeline *pipeline,
433 const VkPipelineDepthStencilStateCreateInfo *info,
434 const struct anv_render_pass *pass,
435 const struct anv_subpass *subpass)
436 {
437 #if GEN_GEN == 7
438 # define depth_stencil_dw pipeline->gen7.depth_stencil_state
439 #elif GEN_GEN == 8
440 # define depth_stencil_dw pipeline->gen8.wm_depth_stencil
441 #else
442 # define depth_stencil_dw pipeline->gen9.wm_depth_stencil
443 #endif
444
445 if (info == NULL) {
446 /* We're going to OR this together with the dynamic state. We need
447 * to make sure it's initialized to something useful.
448 */
449 memset(depth_stencil_dw, 0, sizeof(depth_stencil_dw));
450 return;
451 }
452
453 /* VkBool32 depthBoundsTestEnable; // optional (depth_bounds_test) */
454
455 #if GEN_GEN <= 7
456 struct GENX(DEPTH_STENCIL_STATE) depth_stencil = {
457 #else
458 struct GENX(3DSTATE_WM_DEPTH_STENCIL) depth_stencil = {
459 #endif
460 .DepthTestEnable = info->depthTestEnable,
461 .DepthBufferWriteEnable = info->depthWriteEnable,
462 .DepthTestFunction = vk_to_gen_compare_op[info->depthCompareOp],
463 .DoubleSidedStencilEnable = true,
464
465 .StencilTestEnable = info->stencilTestEnable,
466 .StencilBufferWriteEnable = info->stencilTestEnable,
467 .StencilFailOp = vk_to_gen_stencil_op[info->front.failOp],
468 .StencilPassDepthPassOp = vk_to_gen_stencil_op[info->front.passOp],
469 .StencilPassDepthFailOp = vk_to_gen_stencil_op[info->front.depthFailOp],
470 .StencilTestFunction = vk_to_gen_compare_op[info->front.compareOp],
471 .BackfaceStencilFailOp = vk_to_gen_stencil_op[info->back.failOp],
472 .BackfaceStencilPassDepthPassOp = vk_to_gen_stencil_op[info->back.passOp],
473 .BackfaceStencilPassDepthFailOp =vk_to_gen_stencil_op[info->back.depthFailOp],
474 .BackfaceStencilTestFunction = vk_to_gen_compare_op[info->back.compareOp],
475 };
476
477 VkImageAspectFlags aspects = 0;
478 if (pass->attachments == NULL) {
479 /* This comes from meta. Assume we have verything. */
480 aspects = VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT;
481 } else if (subpass->depth_stencil_attachment != VK_ATTACHMENT_UNUSED) {
482 VkFormat depth_stencil_format =
483 pass->attachments[subpass->depth_stencil_attachment].format;
484 aspects = vk_format_aspects(depth_stencil_format);
485 }
486
487 /* The Vulkan spec requires that if either depth or stencil is not present,
488 * the pipeline is to act as if the test silently passes.
489 */
490 if (!(aspects & VK_IMAGE_ASPECT_DEPTH_BIT)) {
491 depth_stencil.DepthBufferWriteEnable = false;
492 depth_stencil.DepthTestFunction = PREFILTEROPALWAYS;
493 }
494
495 if (!(aspects & VK_IMAGE_ASPECT_STENCIL_BIT)) {
496 depth_stencil.StencilBufferWriteEnable = false;
497 depth_stencil.StencilTestFunction = PREFILTEROPALWAYS;
498 depth_stencil.BackfaceStencilTestFunction = PREFILTEROPALWAYS;
499 }
500
501 /* From the Broadwell PRM:
502 *
503 * "If Depth_Test_Enable = 1 AND Depth_Test_func = EQUAL, the
504 * Depth_Write_Enable must be set to 0."
505 */
506 if (info->depthTestEnable && info->depthCompareOp == VK_COMPARE_OP_EQUAL)
507 depth_stencil.DepthBufferWriteEnable = false;
508
509 #if GEN_GEN <= 7
510 GENX(DEPTH_STENCIL_STATE_pack)(NULL, depth_stencil_dw, &depth_stencil);
511 #else
512 GENX(3DSTATE_WM_DEPTH_STENCIL_pack)(NULL, depth_stencil_dw, &depth_stencil);
513 #endif
514 }