vk: Move gen8 specific state into gen8 sub-structs
[mesa.git] / src / vulkan / gen8_pipeline.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 static void
33 emit_vertex_input(struct anv_pipeline *pipeline,
34 const VkPipelineVertexInputStateCreateInfo *info)
35 {
36 const uint32_t num_dwords = 1 + info->attributeCount * 2;
37 uint32_t *p;
38 bool instancing_enable[32];
39
40 pipeline->vb_used = 0;
41 for (uint32_t i = 0; i < info->bindingCount; i++) {
42 const VkVertexInputBindingDescription *desc =
43 &info->pVertexBindingDescriptions[i];
44
45 pipeline->vb_used |= 1 << desc->binding;
46 pipeline->binding_stride[desc->binding] = desc->strideInBytes;
47
48 /* Step rate is programmed per vertex element (attribute), not
49 * binding. Set up a map of which bindings step per instance, for
50 * reference by vertex element setup. */
51 switch (desc->stepRate) {
52 default:
53 case VK_VERTEX_INPUT_STEP_RATE_VERTEX:
54 instancing_enable[desc->binding] = false;
55 break;
56 case VK_VERTEX_INPUT_STEP_RATE_INSTANCE:
57 instancing_enable[desc->binding] = true;
58 break;
59 }
60 }
61
62 p = anv_batch_emitn(&pipeline->batch, num_dwords,
63 GEN8_3DSTATE_VERTEX_ELEMENTS);
64
65 for (uint32_t i = 0; i < info->attributeCount; i++) {
66 const VkVertexInputAttributeDescription *desc =
67 &info->pVertexAttributeDescriptions[i];
68 const struct anv_format *format = anv_format_for_vk_format(desc->format);
69
70 struct GEN8_VERTEX_ELEMENT_STATE element = {
71 .VertexBufferIndex = desc->binding,
72 .Valid = true,
73 .SourceElementFormat = format->surface_format,
74 .EdgeFlagEnable = false,
75 .SourceElementOffset = desc->offsetInBytes,
76 .Component0Control = VFCOMP_STORE_SRC,
77 .Component1Control = format->num_channels >= 2 ? VFCOMP_STORE_SRC : VFCOMP_STORE_0,
78 .Component2Control = format->num_channels >= 3 ? VFCOMP_STORE_SRC : VFCOMP_STORE_0,
79 .Component3Control = format->num_channels >= 4 ? VFCOMP_STORE_SRC : VFCOMP_STORE_1_FP
80 };
81 GEN8_VERTEX_ELEMENT_STATE_pack(NULL, &p[1 + i * 2], &element);
82
83 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_INSTANCING,
84 .InstancingEnable = instancing_enable[desc->binding],
85 .VertexElementIndex = i,
86 /* Vulkan so far doesn't have an instance divisor, so
87 * this is always 1 (ignored if not instancing). */
88 .InstanceDataStepRate = 1);
89 }
90
91 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_SGVS,
92 .VertexIDEnable = pipeline->vs_prog_data.uses_vertexid,
93 .VertexIDComponentNumber = 2,
94 .VertexIDElementOffset = info->bindingCount,
95 .InstanceIDEnable = pipeline->vs_prog_data.uses_instanceid,
96 .InstanceIDComponentNumber = 3,
97 .InstanceIDElementOffset = info->bindingCount);
98 }
99
100 static void
101 emit_ia_state(struct anv_pipeline *pipeline,
102 const VkPipelineInputAssemblyStateCreateInfo *info,
103 const struct anv_graphics_pipeline_create_info *extra)
104 {
105 static const uint32_t vk_to_gen_primitive_type[] = {
106 [VK_PRIMITIVE_TOPOLOGY_POINT_LIST] = _3DPRIM_POINTLIST,
107 [VK_PRIMITIVE_TOPOLOGY_LINE_LIST] = _3DPRIM_LINELIST,
108 [VK_PRIMITIVE_TOPOLOGY_LINE_STRIP] = _3DPRIM_LINESTRIP,
109 [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST] = _3DPRIM_TRILIST,
110 [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP] = _3DPRIM_TRISTRIP,
111 [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN] = _3DPRIM_TRIFAN,
112 [VK_PRIMITIVE_TOPOLOGY_LINE_LIST_ADJ] = _3DPRIM_LINELIST_ADJ,
113 [VK_PRIMITIVE_TOPOLOGY_LINE_STRIP_ADJ] = _3DPRIM_LINESTRIP_ADJ,
114 [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_ADJ] = _3DPRIM_TRILIST_ADJ,
115 [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP_ADJ] = _3DPRIM_TRISTRIP_ADJ,
116 [VK_PRIMITIVE_TOPOLOGY_PATCH] = _3DPRIM_PATCHLIST_1
117 };
118 uint32_t topology = vk_to_gen_primitive_type[info->topology];
119
120 if (extra && extra->use_rectlist)
121 topology = _3DPRIM_RECTLIST;
122
123 struct GEN8_3DSTATE_VF vf = {
124 GEN8_3DSTATE_VF_header,
125 .IndexedDrawCutIndexEnable = info->primitiveRestartEnable,
126 };
127 GEN8_3DSTATE_VF_pack(NULL, pipeline->gen8.vf, &vf);
128
129 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_TOPOLOGY,
130 .PrimitiveTopologyType = topology);
131 }
132
133 static void
134 emit_rs_state(struct anv_pipeline *pipeline,
135 const VkPipelineRasterStateCreateInfo *info,
136 const struct anv_graphics_pipeline_create_info *extra)
137 {
138 static const uint32_t vk_to_gen_cullmode[] = {
139 [VK_CULL_MODE_NONE] = CULLMODE_NONE,
140 [VK_CULL_MODE_FRONT] = CULLMODE_FRONT,
141 [VK_CULL_MODE_BACK] = CULLMODE_BACK,
142 [VK_CULL_MODE_FRONT_AND_BACK] = CULLMODE_BOTH
143 };
144
145 static const uint32_t vk_to_gen_fillmode[] = {
146 [VK_FILL_MODE_POINTS] = RASTER_POINT,
147 [VK_FILL_MODE_WIREFRAME] = RASTER_WIREFRAME,
148 [VK_FILL_MODE_SOLID] = RASTER_SOLID
149 };
150
151 static const uint32_t vk_to_gen_front_face[] = {
152 [VK_FRONT_FACE_CCW] = CounterClockwise,
153 [VK_FRONT_FACE_CW] = Clockwise
154 };
155
156 struct GEN8_3DSTATE_SF sf = {
157 GEN8_3DSTATE_SF_header,
158 .ViewportTransformEnable = !(extra && extra->disable_viewport),
159 .TriangleStripListProvokingVertexSelect = 0,
160 .LineStripListProvokingVertexSelect = 0,
161 .TriangleFanProvokingVertexSelect = 0,
162 .PointWidthSource = pipeline->writes_point_size ? Vertex : State,
163 .PointWidth = 1.0,
164 };
165
166 /* FINISHME: VkBool32 rasterizerDiscardEnable; */
167
168 GEN8_3DSTATE_SF_pack(NULL, pipeline->gen8.sf, &sf);
169
170 struct GEN8_3DSTATE_RASTER raster = {
171 GEN8_3DSTATE_RASTER_header,
172 .FrontWinding = vk_to_gen_front_face[info->frontFace],
173 .CullMode = vk_to_gen_cullmode[info->cullMode],
174 .FrontFaceFillMode = vk_to_gen_fillmode[info->fillMode],
175 .BackFaceFillMode = vk_to_gen_fillmode[info->fillMode],
176 .ScissorRectangleEnable = !(extra && extra->disable_scissor),
177 .ViewportZClipTestEnable = info->depthClipEnable
178 };
179
180 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_SBE,
181 .ForceVertexURBEntryReadLength = false,
182 .ForceVertexURBEntryReadOffset = false,
183 .PointSpriteTextureCoordinateOrigin = UPPERLEFT,
184 .NumberofSFOutputAttributes =
185 pipeline->wm_prog_data.num_varying_inputs);
186
187 GEN8_3DSTATE_RASTER_pack(NULL, pipeline->gen8.raster, &raster);
188 }
189
190 static void
191 emit_cb_state(struct anv_pipeline *pipeline,
192 const VkPipelineColorBlendStateCreateInfo *info)
193 {
194 struct anv_device *device = pipeline->device;
195
196 static const uint32_t vk_to_gen_logic_op[] = {
197 [VK_LOGIC_OP_COPY] = LOGICOP_COPY,
198 [VK_LOGIC_OP_CLEAR] = LOGICOP_CLEAR,
199 [VK_LOGIC_OP_AND] = LOGICOP_AND,
200 [VK_LOGIC_OP_AND_REVERSE] = LOGICOP_AND_REVERSE,
201 [VK_LOGIC_OP_AND_INVERTED] = LOGICOP_AND_INVERTED,
202 [VK_LOGIC_OP_NOOP] = LOGICOP_NOOP,
203 [VK_LOGIC_OP_XOR] = LOGICOP_XOR,
204 [VK_LOGIC_OP_OR] = LOGICOP_OR,
205 [VK_LOGIC_OP_NOR] = LOGICOP_NOR,
206 [VK_LOGIC_OP_EQUIV] = LOGICOP_EQUIV,
207 [VK_LOGIC_OP_INVERT] = LOGICOP_INVERT,
208 [VK_LOGIC_OP_OR_REVERSE] = LOGICOP_OR_REVERSE,
209 [VK_LOGIC_OP_COPY_INVERTED] = LOGICOP_COPY_INVERTED,
210 [VK_LOGIC_OP_OR_INVERTED] = LOGICOP_OR_INVERTED,
211 [VK_LOGIC_OP_NAND] = LOGICOP_NAND,
212 [VK_LOGIC_OP_SET] = LOGICOP_SET,
213 };
214
215 static const uint32_t vk_to_gen_blend[] = {
216 [VK_BLEND_ZERO] = BLENDFACTOR_ZERO,
217 [VK_BLEND_ONE] = BLENDFACTOR_ONE,
218 [VK_BLEND_SRC_COLOR] = BLENDFACTOR_SRC_COLOR,
219 [VK_BLEND_ONE_MINUS_SRC_COLOR] = BLENDFACTOR_INV_SRC_COLOR,
220 [VK_BLEND_DEST_COLOR] = BLENDFACTOR_DST_COLOR,
221 [VK_BLEND_ONE_MINUS_DEST_COLOR] = BLENDFACTOR_INV_DST_COLOR,
222 [VK_BLEND_SRC_ALPHA] = BLENDFACTOR_SRC_ALPHA,
223 [VK_BLEND_ONE_MINUS_SRC_ALPHA] = BLENDFACTOR_INV_SRC_ALPHA,
224 [VK_BLEND_DEST_ALPHA] = BLENDFACTOR_DST_ALPHA,
225 [VK_BLEND_ONE_MINUS_DEST_ALPHA] = BLENDFACTOR_INV_DST_ALPHA,
226 [VK_BLEND_CONSTANT_COLOR] = BLENDFACTOR_CONST_COLOR,
227 [VK_BLEND_ONE_MINUS_CONSTANT_COLOR] = BLENDFACTOR_INV_CONST_COLOR,
228 [VK_BLEND_CONSTANT_ALPHA] = BLENDFACTOR_CONST_ALPHA,
229 [VK_BLEND_ONE_MINUS_CONSTANT_ALPHA] = BLENDFACTOR_INV_CONST_ALPHA,
230 [VK_BLEND_SRC_ALPHA_SATURATE] = BLENDFACTOR_SRC_ALPHA_SATURATE,
231 [VK_BLEND_SRC1_COLOR] = BLENDFACTOR_SRC1_COLOR,
232 [VK_BLEND_ONE_MINUS_SRC1_COLOR] = BLENDFACTOR_INV_SRC1_COLOR,
233 [VK_BLEND_SRC1_ALPHA] = BLENDFACTOR_SRC1_ALPHA,
234 [VK_BLEND_ONE_MINUS_SRC1_ALPHA] = BLENDFACTOR_INV_SRC1_ALPHA,
235 };
236
237 static const uint32_t vk_to_gen_blend_op[] = {
238 [VK_BLEND_OP_ADD] = BLENDFUNCTION_ADD,
239 [VK_BLEND_OP_SUBTRACT] = BLENDFUNCTION_SUBTRACT,
240 [VK_BLEND_OP_REVERSE_SUBTRACT] = BLENDFUNCTION_REVERSE_SUBTRACT,
241 [VK_BLEND_OP_MIN] = BLENDFUNCTION_MIN,
242 [VK_BLEND_OP_MAX] = BLENDFUNCTION_MAX,
243 };
244
245 uint32_t num_dwords = GEN8_BLEND_STATE_length;
246 pipeline->blend_state =
247 anv_state_pool_alloc(&device->dynamic_state_pool, num_dwords * 4, 64);
248
249 struct GEN8_BLEND_STATE blend_state = {
250 .AlphaToCoverageEnable = info->alphaToCoverageEnable,
251 };
252
253 for (uint32_t i = 0; i < info->attachmentCount; i++) {
254 const VkPipelineColorBlendAttachmentState *a = &info->pAttachments[i];
255
256 blend_state.Entry[i] = (struct GEN8_BLEND_STATE_ENTRY) {
257 .LogicOpEnable = info->logicOpEnable,
258 .LogicOpFunction = vk_to_gen_logic_op[info->logicOp],
259 .ColorBufferBlendEnable = a->blendEnable,
260 .PreBlendSourceOnlyClampEnable = false,
261 .PreBlendColorClampEnable = false,
262 .PostBlendColorClampEnable = false,
263 .SourceBlendFactor = vk_to_gen_blend[a->srcBlendColor],
264 .DestinationBlendFactor = vk_to_gen_blend[a->destBlendColor],
265 .ColorBlendFunction = vk_to_gen_blend_op[a->blendOpColor],
266 .SourceAlphaBlendFactor = vk_to_gen_blend[a->srcBlendAlpha],
267 .DestinationAlphaBlendFactor = vk_to_gen_blend[a->destBlendAlpha],
268 .AlphaBlendFunction = vk_to_gen_blend_op[a->blendOpAlpha],
269 .WriteDisableAlpha = !(a->channelWriteMask & VK_CHANNEL_A_BIT),
270 .WriteDisableRed = !(a->channelWriteMask & VK_CHANNEL_R_BIT),
271 .WriteDisableGreen = !(a->channelWriteMask & VK_CHANNEL_G_BIT),
272 .WriteDisableBlue = !(a->channelWriteMask & VK_CHANNEL_B_BIT),
273 };
274 }
275
276 GEN8_BLEND_STATE_pack(NULL, pipeline->blend_state.map, &blend_state);
277
278 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_BLEND_STATE_POINTERS,
279 .BlendStatePointer = pipeline->blend_state.offset,
280 .BlendStatePointerValid = true);
281 }
282
283 static const uint32_t vk_to_gen_compare_op[] = {
284 [VK_COMPARE_OP_NEVER] = COMPAREFUNCTION_NEVER,
285 [VK_COMPARE_OP_LESS] = COMPAREFUNCTION_LESS,
286 [VK_COMPARE_OP_EQUAL] = COMPAREFUNCTION_EQUAL,
287 [VK_COMPARE_OP_LESS_EQUAL] = COMPAREFUNCTION_LEQUAL,
288 [VK_COMPARE_OP_GREATER] = COMPAREFUNCTION_GREATER,
289 [VK_COMPARE_OP_NOT_EQUAL] = COMPAREFUNCTION_NOTEQUAL,
290 [VK_COMPARE_OP_GREATER_EQUAL] = COMPAREFUNCTION_GEQUAL,
291 [VK_COMPARE_OP_ALWAYS] = COMPAREFUNCTION_ALWAYS,
292 };
293
294 static const uint32_t vk_to_gen_stencil_op[] = {
295 [VK_STENCIL_OP_KEEP] = 0,
296 [VK_STENCIL_OP_ZERO] = 0,
297 [VK_STENCIL_OP_REPLACE] = 0,
298 [VK_STENCIL_OP_INC_CLAMP] = 0,
299 [VK_STENCIL_OP_DEC_CLAMP] = 0,
300 [VK_STENCIL_OP_INVERT] = 0,
301 [VK_STENCIL_OP_INC_WRAP] = 0,
302 [VK_STENCIL_OP_DEC_WRAP] = 0
303 };
304
305 static void
306 emit_ds_state(struct anv_pipeline *pipeline,
307 const VkPipelineDepthStencilStateCreateInfo *info)
308 {
309 if (info == NULL) {
310 /* We're going to OR this together with the dynamic state. We need
311 * to make sure it's initialized to something useful.
312 */
313 memset(pipeline->gen8.wm_depth_stencil, 0,
314 sizeof(pipeline->gen8.wm_depth_stencil));
315 return;
316 }
317
318 /* VkBool32 depthBoundsEnable; // optional (depth_bounds_test) */
319
320 struct GEN8_3DSTATE_WM_DEPTH_STENCIL wm_depth_stencil = {
321 .DepthTestEnable = info->depthTestEnable,
322 .DepthBufferWriteEnable = info->depthWriteEnable,
323 .DepthTestFunction = vk_to_gen_compare_op[info->depthCompareOp],
324 .DoubleSidedStencilEnable = true,
325
326 .StencilTestEnable = info->stencilTestEnable,
327 .StencilFailOp = vk_to_gen_stencil_op[info->front.stencilFailOp],
328 .StencilPassDepthPassOp = vk_to_gen_stencil_op[info->front.stencilPassOp],
329 .StencilPassDepthFailOp = vk_to_gen_stencil_op[info->front.stencilDepthFailOp],
330 .StencilTestFunction = vk_to_gen_compare_op[info->front.stencilCompareOp],
331 .BackfaceStencilFailOp = vk_to_gen_stencil_op[info->back.stencilFailOp],
332 .BackfaceStencilPassDepthPassOp = vk_to_gen_stencil_op[info->back.stencilPassOp],
333 .BackfaceStencilPassDepthFailOp =vk_to_gen_stencil_op[info->back.stencilDepthFailOp],
334 .BackfaceStencilTestFunction = vk_to_gen_compare_op[info->back.stencilCompareOp],
335 };
336
337 GEN8_3DSTATE_WM_DEPTH_STENCIL_pack(NULL, pipeline->gen8.wm_depth_stencil, &wm_depth_stencil);
338 }
339
340 VkResult
341 gen8_graphics_pipeline_create(
342 VkDevice _device,
343 const VkGraphicsPipelineCreateInfo* pCreateInfo,
344 const struct anv_graphics_pipeline_create_info *extra,
345 VkPipeline* pPipeline)
346 {
347 ANV_FROM_HANDLE(anv_device, device, _device);
348 struct anv_pipeline *pipeline;
349 VkResult result;
350 uint32_t offset, length;
351
352 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO);
353
354 pipeline = anv_device_alloc(device, sizeof(*pipeline), 8,
355 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
356 if (pipeline == NULL)
357 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
358
359 pipeline->device = device;
360 pipeline->layout = anv_pipeline_layout_from_handle(pCreateInfo->layout);
361 memset(pipeline->shaders, 0, sizeof(pipeline->shaders));
362
363 result = anv_reloc_list_init(&pipeline->batch_relocs, device);
364 if (result != VK_SUCCESS) {
365 anv_device_free(device, pipeline);
366 return result;
367 }
368 pipeline->batch.next = pipeline->batch.start = pipeline->batch_data;
369 pipeline->batch.end = pipeline->batch.start + sizeof(pipeline->batch_data);
370 pipeline->batch.relocs = &pipeline->batch_relocs;
371
372 anv_state_stream_init(&pipeline->program_stream,
373 &device->instruction_block_pool);
374
375 for (uint32_t i = 0; i < pCreateInfo->stageCount; i++) {
376 pipeline->shaders[pCreateInfo->pStages[i].stage] =
377 anv_shader_from_handle(pCreateInfo->pStages[i].shader);
378 }
379
380 if (pCreateInfo->pTessellationState)
381 anv_finishme("VK_STRUCTURE_TYPE_PIPELINE_TESSELLATION_STATE_CREATE_INFO");
382 if (pCreateInfo->pViewportState)
383 anv_finishme("VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO");
384 if (pCreateInfo->pMultisampleState)
385 anv_finishme("VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO");
386
387 pipeline->use_repclear = extra && extra->use_repclear;
388
389 anv_compiler_run(device->compiler, pipeline);
390
391 /* FIXME: The compiler dead-codes FS inputs when we don't have a VS, so we
392 * hard code this to num_attributes - 2. This is because the attributes
393 * include VUE header and position, which aren't counted as varying
394 * inputs. */
395 if (pipeline->vs_simd8 == NO_KERNEL) {
396 pipeline->wm_prog_data.num_varying_inputs =
397 pCreateInfo->pVertexInputState->attributeCount - 2;
398 }
399
400 assert(pCreateInfo->pVertexInputState);
401 emit_vertex_input(pipeline, pCreateInfo->pVertexInputState);
402 assert(pCreateInfo->pInputAssemblyState);
403 emit_ia_state(pipeline, pCreateInfo->pInputAssemblyState, extra);
404 assert(pCreateInfo->pRasterState);
405 emit_rs_state(pipeline, pCreateInfo->pRasterState, extra);
406 emit_ds_state(pipeline, pCreateInfo->pDepthStencilState);
407 emit_cb_state(pipeline, pCreateInfo->pColorBlendState);
408
409 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_STATISTICS,
410 .StatisticsEnable = true);
411 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_HS, .Enable = false);
412 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_TE, .TEEnable = false);
413 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_DS, .FunctionEnable = false);
414 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_STREAMOUT, .SOFunctionEnable = false);
415
416 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_VS,
417 .ConstantBufferOffset = 0,
418 .ConstantBufferSize = 4);
419 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_GS,
420 .ConstantBufferOffset = 4,
421 .ConstantBufferSize = 4);
422 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_PS,
423 .ConstantBufferOffset = 8,
424 .ConstantBufferSize = 4);
425
426 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_WM_CHROMAKEY,
427 .ChromaKeyKillEnable = false);
428 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_SBE_SWIZ);
429 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_AA_LINE_PARAMETERS);
430
431 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_CLIP,
432 .ClipEnable = true,
433 .ViewportXYClipTestEnable = !(extra && extra->disable_viewport),
434 .MinimumPointWidth = 0.125,
435 .MaximumPointWidth = 255.875);
436
437 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_WM,
438 .StatisticsEnable = true,
439 .LineEndCapAntialiasingRegionWidth = _05pixels,
440 .LineAntialiasingRegionWidth = _10pixels,
441 .EarlyDepthStencilControl = NORMAL,
442 .ForceThreadDispatchEnable = NORMAL,
443 .PointRasterizationRule = RASTRULE_UPPER_RIGHT,
444 .BarycentricInterpolationMode =
445 pipeline->wm_prog_data.barycentric_interp_modes);
446
447 uint32_t samples = 1;
448 uint32_t log2_samples = __builtin_ffs(samples) - 1;
449 bool enable_sampling = samples > 1 ? true : false;
450
451 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_MULTISAMPLE,
452 .PixelPositionOffsetEnable = enable_sampling,
453 .PixelLocation = CENTER,
454 .NumberofMultisamples = log2_samples);
455
456 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_SAMPLE_MASK,
457 .SampleMask = 0xffff);
458
459 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_VS,
460 .VSURBStartingAddress = pipeline->urb.vs_start,
461 .VSURBEntryAllocationSize = pipeline->urb.vs_size - 1,
462 .VSNumberofURBEntries = pipeline->urb.nr_vs_entries);
463
464 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_GS,
465 .GSURBStartingAddress = pipeline->urb.gs_start,
466 .GSURBEntryAllocationSize = pipeline->urb.gs_size - 1,
467 .GSNumberofURBEntries = pipeline->urb.nr_gs_entries);
468
469 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_HS,
470 .HSURBStartingAddress = pipeline->urb.vs_start,
471 .HSURBEntryAllocationSize = 0,
472 .HSNumberofURBEntries = 0);
473
474 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_DS,
475 .DSURBStartingAddress = pipeline->urb.vs_start,
476 .DSURBEntryAllocationSize = 0,
477 .DSNumberofURBEntries = 0);
478
479 const struct brw_gs_prog_data *gs_prog_data = &pipeline->gs_prog_data;
480 offset = 1;
481 length = (gs_prog_data->base.vue_map.num_slots + 1) / 2 - offset;
482
483 if (pipeline->gs_vec4 == NO_KERNEL)
484 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_GS, .Enable = false);
485 else
486 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_GS,
487 .SingleProgramFlow = false,
488 .KernelStartPointer = pipeline->gs_vec4,
489 .VectorMaskEnable = Vmask,
490 .SamplerCount = 0,
491 .BindingTableEntryCount = 0,
492 .ExpectedVertexCount = pipeline->gs_vertex_count,
493
494 .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_GEOMETRY],
495 .PerThreadScratchSpace = ffs(gs_prog_data->base.base.total_scratch / 2048),
496
497 .OutputVertexSize = gs_prog_data->output_vertex_size_hwords * 2 - 1,
498 .OutputTopology = gs_prog_data->output_topology,
499 .VertexURBEntryReadLength = gs_prog_data->base.urb_read_length,
500 .DispatchGRFStartRegisterForURBData =
501 gs_prog_data->base.base.dispatch_grf_start_reg,
502
503 .MaximumNumberofThreads = device->info.max_gs_threads,
504 .ControlDataHeaderSize = gs_prog_data->control_data_header_size_hwords,
505 //pipeline->gs_prog_data.dispatch_mode |
506 .StatisticsEnable = true,
507 .IncludePrimitiveID = gs_prog_data->include_primitive_id,
508 .ReorderMode = TRAILING,
509 .Enable = true,
510
511 .ControlDataFormat = gs_prog_data->control_data_format,
512
513 /* FIXME: mesa sets this based on ctx->Transform.ClipPlanesEnabled:
514 * UserClipDistanceClipTestEnableBitmask_3DSTATE_GS(v)
515 * UserClipDistanceCullTestEnableBitmask(v)
516 */
517
518 .VertexURBEntryOutputReadOffset = offset,
519 .VertexURBEntryOutputLength = length);
520
521 const struct brw_vue_prog_data *vue_prog_data = &pipeline->vs_prog_data.base;
522 /* Skip the VUE header and position slots */
523 offset = 1;
524 length = (vue_prog_data->vue_map.num_slots + 1) / 2 - offset;
525
526 if (pipeline->vs_simd8 == NO_KERNEL || (extra && extra->disable_vs))
527 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VS,
528 .FunctionEnable = false,
529 /* Even if VS is disabled, SBE still gets the amount of
530 * vertex data to read from this field. */
531 .VertexURBEntryOutputReadOffset = offset,
532 .VertexURBEntryOutputLength = length);
533 else
534 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VS,
535 .KernelStartPointer = pipeline->vs_simd8,
536 .SingleVertexDispatch = Multiple,
537 .VectorMaskEnable = Dmask,
538 .SamplerCount = 0,
539 .BindingTableEntryCount =
540 vue_prog_data->base.binding_table.size_bytes / 4,
541 .ThreadDispatchPriority = Normal,
542 .FloatingPointMode = IEEE754,
543 .IllegalOpcodeExceptionEnable = false,
544 .AccessesUAV = false,
545 .SoftwareExceptionEnable = false,
546
547 .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_VERTEX],
548 .PerThreadScratchSpace = ffs(vue_prog_data->base.total_scratch / 2048),
549
550 .DispatchGRFStartRegisterForURBData =
551 vue_prog_data->base.dispatch_grf_start_reg,
552 .VertexURBEntryReadLength = vue_prog_data->urb_read_length,
553 .VertexURBEntryReadOffset = 0,
554
555 .MaximumNumberofThreads = device->info.max_vs_threads - 1,
556 .StatisticsEnable = false,
557 .SIMD8DispatchEnable = true,
558 .VertexCacheDisable = false,
559 .FunctionEnable = true,
560
561 .VertexURBEntryOutputReadOffset = offset,
562 .VertexURBEntryOutputLength = length,
563 .UserClipDistanceClipTestEnableBitmask = 0,
564 .UserClipDistanceCullTestEnableBitmask = 0);
565
566 const struct brw_wm_prog_data *wm_prog_data = &pipeline->wm_prog_data;
567 uint32_t ksp0, ksp2, grf_start0, grf_start2;
568
569 ksp2 = 0;
570 grf_start2 = 0;
571 if (pipeline->ps_simd8 != NO_KERNEL) {
572 ksp0 = pipeline->ps_simd8;
573 grf_start0 = wm_prog_data->base.dispatch_grf_start_reg;
574 if (pipeline->ps_simd16 != NO_KERNEL) {
575 ksp2 = pipeline->ps_simd16;
576 grf_start2 = wm_prog_data->dispatch_grf_start_reg_16;
577 }
578 } else if (pipeline->ps_simd16 != NO_KERNEL) {
579 ksp0 = pipeline->ps_simd16;
580 grf_start0 = wm_prog_data->dispatch_grf_start_reg_16;
581 } else {
582 unreachable("no ps shader");
583 }
584
585 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PS,
586 .KernelStartPointer0 = ksp0,
587
588 .SingleProgramFlow = false,
589 .VectorMaskEnable = true,
590 .SamplerCount = 1,
591
592 .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_FRAGMENT],
593 .PerThreadScratchSpace = ffs(wm_prog_data->base.total_scratch / 2048),
594
595 .MaximumNumberofThreadsPerPSD = 64 - 2,
596 .PositionXYOffsetSelect = wm_prog_data->uses_pos_offset ?
597 POSOFFSET_SAMPLE: POSOFFSET_NONE,
598 .PushConstantEnable = wm_prog_data->base.nr_params > 0,
599 ._8PixelDispatchEnable = pipeline->ps_simd8 != NO_KERNEL,
600 ._16PixelDispatchEnable = pipeline->ps_simd16 != NO_KERNEL,
601 ._32PixelDispatchEnable = false,
602
603 .DispatchGRFStartRegisterForConstantSetupData0 = grf_start0,
604 .DispatchGRFStartRegisterForConstantSetupData1 = 0,
605 .DispatchGRFStartRegisterForConstantSetupData2 = grf_start2,
606
607 .KernelStartPointer1 = 0,
608 .KernelStartPointer2 = ksp2);
609
610 bool per_sample_ps = false;
611 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PS_EXTRA,
612 .PixelShaderValid = true,
613 .PixelShaderKillsPixel = wm_prog_data->uses_kill,
614 .PixelShaderComputedDepthMode = wm_prog_data->computed_depth_mode,
615 .AttributeEnable = wm_prog_data->num_varying_inputs > 0,
616 .oMaskPresenttoRenderTarget = wm_prog_data->uses_omask,
617 .PixelShaderIsPerSample = per_sample_ps);
618
619 *pPipeline = anv_pipeline_to_handle(pipeline);
620
621 return VK_SUCCESS;
622 }
623
624 VkResult gen8_compute_pipeline_create(
625 VkDevice _device,
626 const VkComputePipelineCreateInfo* pCreateInfo,
627 VkPipeline* pPipeline)
628 {
629 ANV_FROM_HANDLE(anv_device, device, _device);
630 struct anv_pipeline *pipeline;
631 VkResult result;
632
633 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO);
634
635 pipeline = anv_device_alloc(device, sizeof(*pipeline), 8,
636 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
637 if (pipeline == NULL)
638 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
639
640 pipeline->device = device;
641 pipeline->layout = anv_pipeline_layout_from_handle(pCreateInfo->layout);
642
643 result = anv_reloc_list_init(&pipeline->batch_relocs, device);
644 if (result != VK_SUCCESS) {
645 anv_device_free(device, pipeline);
646 return result;
647 }
648 pipeline->batch.next = pipeline->batch.start = pipeline->batch_data;
649 pipeline->batch.end = pipeline->batch.start + sizeof(pipeline->batch_data);
650 pipeline->batch.relocs = &pipeline->batch_relocs;
651
652 anv_state_stream_init(&pipeline->program_stream,
653 &device->instruction_block_pool);
654
655 memset(pipeline->shaders, 0, sizeof(pipeline->shaders));
656
657 pipeline->shaders[VK_SHADER_STAGE_COMPUTE] =
658 anv_shader_from_handle(pCreateInfo->cs.shader);
659
660 pipeline->use_repclear = false;
661
662 anv_compiler_run(device->compiler, pipeline);
663
664 const struct brw_cs_prog_data *cs_prog_data = &pipeline->cs_prog_data;
665
666 anv_batch_emit(&pipeline->batch, GEN8_MEDIA_VFE_STATE,
667 .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_FRAGMENT],
668 .PerThreadScratchSpace = ffs(cs_prog_data->base.total_scratch / 2048),
669 .ScratchSpaceBasePointerHigh = 0,
670 .StackSize = 0,
671
672 .MaximumNumberofThreads = device->info.max_cs_threads - 1,
673 .NumberofURBEntries = 2,
674 .ResetGatewayTimer = true,
675 .BypassGatewayControl = true,
676 .URBEntryAllocationSize = 2,
677 .CURBEAllocationSize = 0);
678
679 struct brw_cs_prog_data *prog_data = &pipeline->cs_prog_data;
680 uint32_t group_size = prog_data->local_size[0] *
681 prog_data->local_size[1] * prog_data->local_size[2];
682 pipeline->cs_thread_width_max = DIV_ROUND_UP(group_size, prog_data->simd_size);
683 uint32_t remainder = group_size & (prog_data->simd_size - 1);
684
685 if (remainder > 0)
686 pipeline->cs_right_mask = ~0u >> (32 - remainder);
687 else
688 pipeline->cs_right_mask = ~0u >> (32 - prog_data->simd_size);
689
690
691 *pPipeline = anv_pipeline_to_handle(pipeline);
692
693 return VK_SUCCESS;
694 }