Merge remote-tracking branch 'mesa-public/master' into vulkan
[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
39 p = anv_batch_emitn(&pipeline->batch, num_dwords,
40 GEN8_3DSTATE_VERTEX_ELEMENTS);
41
42 for (uint32_t i = 0; i < info->attributeCount; i++) {
43 const VkVertexInputAttributeDescription *desc =
44 &info->pVertexAttributeDescriptions[i];
45 const struct anv_format *format = anv_format_for_vk_format(desc->format);
46
47 struct GEN8_VERTEX_ELEMENT_STATE element = {
48 .VertexBufferIndex = desc->binding,
49 .Valid = true,
50 .SourceElementFormat = format->surface_format,
51 .EdgeFlagEnable = false,
52 .SourceElementOffset = desc->offsetInBytes,
53 .Component0Control = VFCOMP_STORE_SRC,
54 .Component1Control = format->num_channels >= 2 ? VFCOMP_STORE_SRC : VFCOMP_STORE_0,
55 .Component2Control = format->num_channels >= 3 ? VFCOMP_STORE_SRC : VFCOMP_STORE_0,
56 .Component3Control = format->num_channels >= 4 ? VFCOMP_STORE_SRC : VFCOMP_STORE_1_FP
57 };
58 GEN8_VERTEX_ELEMENT_STATE_pack(NULL, &p[1 + i * 2], &element);
59
60 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_INSTANCING,
61 .InstancingEnable = pipeline->instancing_enable[desc->binding],
62 .VertexElementIndex = i,
63 /* Vulkan so far doesn't have an instance divisor, so
64 * this is always 1 (ignored if not instancing). */
65 .InstanceDataStepRate = 1);
66 }
67
68 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_SGVS,
69 .VertexIDEnable = pipeline->vs_prog_data.uses_vertexid,
70 .VertexIDComponentNumber = 2,
71 .VertexIDElementOffset = info->bindingCount,
72 .InstanceIDEnable = pipeline->vs_prog_data.uses_instanceid,
73 .InstanceIDComponentNumber = 3,
74 .InstanceIDElementOffset = info->bindingCount);
75 }
76
77 static void
78 emit_ia_state(struct anv_pipeline *pipeline,
79 const VkPipelineInputAssemblyStateCreateInfo *info,
80 const struct anv_graphics_pipeline_create_info *extra)
81 {
82 struct GEN8_3DSTATE_VF vf = {
83 GEN8_3DSTATE_VF_header,
84 .IndexedDrawCutIndexEnable = pipeline->primitive_restart
85 };
86 GEN8_3DSTATE_VF_pack(NULL, pipeline->gen8.vf, &vf);
87
88 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_TOPOLOGY,
89 .PrimitiveTopologyType = pipeline->topology);
90 }
91
92 static void
93 emit_rs_state(struct anv_pipeline *pipeline,
94 const VkPipelineRasterStateCreateInfo *info,
95 const struct anv_graphics_pipeline_create_info *extra)
96 {
97 static const uint32_t vk_to_gen_cullmode[] = {
98 [VK_CULL_MODE_NONE] = CULLMODE_NONE,
99 [VK_CULL_MODE_FRONT] = CULLMODE_FRONT,
100 [VK_CULL_MODE_BACK] = CULLMODE_BACK,
101 [VK_CULL_MODE_FRONT_AND_BACK] = CULLMODE_BOTH
102 };
103
104 static const uint32_t vk_to_gen_fillmode[] = {
105 [VK_FILL_MODE_POINTS] = RASTER_POINT,
106 [VK_FILL_MODE_WIREFRAME] = RASTER_WIREFRAME,
107 [VK_FILL_MODE_SOLID] = RASTER_SOLID
108 };
109
110 static const uint32_t vk_to_gen_front_face[] = {
111 [VK_FRONT_FACE_CCW] = CounterClockwise,
112 [VK_FRONT_FACE_CW] = Clockwise
113 };
114
115 struct GEN8_3DSTATE_SF sf = {
116 GEN8_3DSTATE_SF_header,
117 .ViewportTransformEnable = !(extra && extra->disable_viewport),
118 .TriangleStripListProvokingVertexSelect = 0,
119 .LineStripListProvokingVertexSelect = 0,
120 .TriangleFanProvokingVertexSelect = 0,
121 .PointWidthSource = pipeline->writes_point_size ? Vertex : State,
122 .PointWidth = 1.0,
123 };
124
125 /* FINISHME: VkBool32 rasterizerDiscardEnable; */
126
127 GEN8_3DSTATE_SF_pack(NULL, pipeline->gen8.sf, &sf);
128
129 struct GEN8_3DSTATE_RASTER raster = {
130 GEN8_3DSTATE_RASTER_header,
131 .FrontWinding = vk_to_gen_front_face[info->frontFace],
132 .CullMode = vk_to_gen_cullmode[info->cullMode],
133 .FrontFaceFillMode = vk_to_gen_fillmode[info->fillMode],
134 .BackFaceFillMode = vk_to_gen_fillmode[info->fillMode],
135 .ScissorRectangleEnable = !(extra && extra->disable_scissor),
136 .ViewportZClipTestEnable = info->depthClipEnable
137 };
138
139 GEN8_3DSTATE_RASTER_pack(NULL, pipeline->gen8.raster, &raster);
140 }
141
142 static void
143 emit_cb_state(struct anv_pipeline *pipeline,
144 const VkPipelineColorBlendStateCreateInfo *info)
145 {
146 struct anv_device *device = pipeline->device;
147
148 static const uint32_t vk_to_gen_logic_op[] = {
149 [VK_LOGIC_OP_COPY] = LOGICOP_COPY,
150 [VK_LOGIC_OP_CLEAR] = LOGICOP_CLEAR,
151 [VK_LOGIC_OP_AND] = LOGICOP_AND,
152 [VK_LOGIC_OP_AND_REVERSE] = LOGICOP_AND_REVERSE,
153 [VK_LOGIC_OP_AND_INVERTED] = LOGICOP_AND_INVERTED,
154 [VK_LOGIC_OP_NOOP] = LOGICOP_NOOP,
155 [VK_LOGIC_OP_XOR] = LOGICOP_XOR,
156 [VK_LOGIC_OP_OR] = LOGICOP_OR,
157 [VK_LOGIC_OP_NOR] = LOGICOP_NOR,
158 [VK_LOGIC_OP_EQUIV] = LOGICOP_EQUIV,
159 [VK_LOGIC_OP_INVERT] = LOGICOP_INVERT,
160 [VK_LOGIC_OP_OR_REVERSE] = LOGICOP_OR_REVERSE,
161 [VK_LOGIC_OP_COPY_INVERTED] = LOGICOP_COPY_INVERTED,
162 [VK_LOGIC_OP_OR_INVERTED] = LOGICOP_OR_INVERTED,
163 [VK_LOGIC_OP_NAND] = LOGICOP_NAND,
164 [VK_LOGIC_OP_SET] = LOGICOP_SET,
165 };
166
167 static const uint32_t vk_to_gen_blend[] = {
168 [VK_BLEND_ZERO] = BLENDFACTOR_ZERO,
169 [VK_BLEND_ONE] = BLENDFACTOR_ONE,
170 [VK_BLEND_SRC_COLOR] = BLENDFACTOR_SRC_COLOR,
171 [VK_BLEND_ONE_MINUS_SRC_COLOR] = BLENDFACTOR_INV_SRC_COLOR,
172 [VK_BLEND_DEST_COLOR] = BLENDFACTOR_DST_COLOR,
173 [VK_BLEND_ONE_MINUS_DEST_COLOR] = BLENDFACTOR_INV_DST_COLOR,
174 [VK_BLEND_SRC_ALPHA] = BLENDFACTOR_SRC_ALPHA,
175 [VK_BLEND_ONE_MINUS_SRC_ALPHA] = BLENDFACTOR_INV_SRC_ALPHA,
176 [VK_BLEND_DEST_ALPHA] = BLENDFACTOR_DST_ALPHA,
177 [VK_BLEND_ONE_MINUS_DEST_ALPHA] = BLENDFACTOR_INV_DST_ALPHA,
178 [VK_BLEND_CONSTANT_COLOR] = BLENDFACTOR_CONST_COLOR,
179 [VK_BLEND_ONE_MINUS_CONSTANT_COLOR] = BLENDFACTOR_INV_CONST_COLOR,
180 [VK_BLEND_CONSTANT_ALPHA] = BLENDFACTOR_CONST_ALPHA,
181 [VK_BLEND_ONE_MINUS_CONSTANT_ALPHA] = BLENDFACTOR_INV_CONST_ALPHA,
182 [VK_BLEND_SRC_ALPHA_SATURATE] = BLENDFACTOR_SRC_ALPHA_SATURATE,
183 [VK_BLEND_SRC1_COLOR] = BLENDFACTOR_SRC1_COLOR,
184 [VK_BLEND_ONE_MINUS_SRC1_COLOR] = BLENDFACTOR_INV_SRC1_COLOR,
185 [VK_BLEND_SRC1_ALPHA] = BLENDFACTOR_SRC1_ALPHA,
186 [VK_BLEND_ONE_MINUS_SRC1_ALPHA] = BLENDFACTOR_INV_SRC1_ALPHA,
187 };
188
189 static const uint32_t vk_to_gen_blend_op[] = {
190 [VK_BLEND_OP_ADD] = BLENDFUNCTION_ADD,
191 [VK_BLEND_OP_SUBTRACT] = BLENDFUNCTION_SUBTRACT,
192 [VK_BLEND_OP_REVERSE_SUBTRACT] = BLENDFUNCTION_REVERSE_SUBTRACT,
193 [VK_BLEND_OP_MIN] = BLENDFUNCTION_MIN,
194 [VK_BLEND_OP_MAX] = BLENDFUNCTION_MAX,
195 };
196
197 uint32_t num_dwords = GEN8_BLEND_STATE_length;
198 pipeline->blend_state =
199 anv_state_pool_alloc(&device->dynamic_state_pool, num_dwords * 4, 64);
200
201 struct GEN8_BLEND_STATE blend_state = {
202 .AlphaToCoverageEnable = info->alphaToCoverageEnable,
203 };
204
205 for (uint32_t i = 0; i < info->attachmentCount; i++) {
206 const VkPipelineColorBlendAttachmentState *a = &info->pAttachments[i];
207
208 blend_state.Entry[i] = (struct GEN8_BLEND_STATE_ENTRY) {
209 .LogicOpEnable = info->logicOpEnable,
210 .LogicOpFunction = vk_to_gen_logic_op[info->logicOp],
211 .ColorBufferBlendEnable = a->blendEnable,
212 .PreBlendSourceOnlyClampEnable = false,
213 .PreBlendColorClampEnable = false,
214 .PostBlendColorClampEnable = false,
215 .SourceBlendFactor = vk_to_gen_blend[a->srcBlendColor],
216 .DestinationBlendFactor = vk_to_gen_blend[a->destBlendColor],
217 .ColorBlendFunction = vk_to_gen_blend_op[a->blendOpColor],
218 .SourceAlphaBlendFactor = vk_to_gen_blend[a->srcBlendAlpha],
219 .DestinationAlphaBlendFactor = vk_to_gen_blend[a->destBlendAlpha],
220 .AlphaBlendFunction = vk_to_gen_blend_op[a->blendOpAlpha],
221 .WriteDisableAlpha = !(a->channelWriteMask & VK_CHANNEL_A_BIT),
222 .WriteDisableRed = !(a->channelWriteMask & VK_CHANNEL_R_BIT),
223 .WriteDisableGreen = !(a->channelWriteMask & VK_CHANNEL_G_BIT),
224 .WriteDisableBlue = !(a->channelWriteMask & VK_CHANNEL_B_BIT),
225 };
226 }
227
228 GEN8_BLEND_STATE_pack(NULL, pipeline->blend_state.map, &blend_state);
229
230 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_BLEND_STATE_POINTERS,
231 .BlendStatePointer = pipeline->blend_state.offset,
232 .BlendStatePointerValid = true);
233 }
234
235 static const uint32_t vk_to_gen_compare_op[] = {
236 [VK_COMPARE_OP_NEVER] = COMPAREFUNCTION_NEVER,
237 [VK_COMPARE_OP_LESS] = COMPAREFUNCTION_LESS,
238 [VK_COMPARE_OP_EQUAL] = COMPAREFUNCTION_EQUAL,
239 [VK_COMPARE_OP_LESS_EQUAL] = COMPAREFUNCTION_LEQUAL,
240 [VK_COMPARE_OP_GREATER] = COMPAREFUNCTION_GREATER,
241 [VK_COMPARE_OP_NOT_EQUAL] = COMPAREFUNCTION_NOTEQUAL,
242 [VK_COMPARE_OP_GREATER_EQUAL] = COMPAREFUNCTION_GEQUAL,
243 [VK_COMPARE_OP_ALWAYS] = COMPAREFUNCTION_ALWAYS,
244 };
245
246 static const uint32_t vk_to_gen_stencil_op[] = {
247 [VK_STENCIL_OP_KEEP] = 0,
248 [VK_STENCIL_OP_ZERO] = 0,
249 [VK_STENCIL_OP_REPLACE] = 0,
250 [VK_STENCIL_OP_INC_CLAMP] = 0,
251 [VK_STENCIL_OP_DEC_CLAMP] = 0,
252 [VK_STENCIL_OP_INVERT] = 0,
253 [VK_STENCIL_OP_INC_WRAP] = 0,
254 [VK_STENCIL_OP_DEC_WRAP] = 0
255 };
256
257 static void
258 emit_ds_state(struct anv_pipeline *pipeline,
259 const VkPipelineDepthStencilStateCreateInfo *info)
260 {
261 if (info == NULL) {
262 /* We're going to OR this together with the dynamic state. We need
263 * to make sure it's initialized to something useful.
264 */
265 memset(pipeline->gen8.wm_depth_stencil, 0,
266 sizeof(pipeline->gen8.wm_depth_stencil));
267 return;
268 }
269
270 /* VkBool32 depthBoundsEnable; // optional (depth_bounds_test) */
271
272 struct GEN8_3DSTATE_WM_DEPTH_STENCIL wm_depth_stencil = {
273 .DepthTestEnable = info->depthTestEnable,
274 .DepthBufferWriteEnable = info->depthWriteEnable,
275 .DepthTestFunction = vk_to_gen_compare_op[info->depthCompareOp],
276 .DoubleSidedStencilEnable = true,
277
278 .StencilTestEnable = info->stencilTestEnable,
279 .StencilFailOp = vk_to_gen_stencil_op[info->front.stencilFailOp],
280 .StencilPassDepthPassOp = vk_to_gen_stencil_op[info->front.stencilPassOp],
281 .StencilPassDepthFailOp = vk_to_gen_stencil_op[info->front.stencilDepthFailOp],
282 .StencilTestFunction = vk_to_gen_compare_op[info->front.stencilCompareOp],
283 .BackfaceStencilFailOp = vk_to_gen_stencil_op[info->back.stencilFailOp],
284 .BackfaceStencilPassDepthPassOp = vk_to_gen_stencil_op[info->back.stencilPassOp],
285 .BackfaceStencilPassDepthFailOp =vk_to_gen_stencil_op[info->back.stencilDepthFailOp],
286 .BackfaceStencilTestFunction = vk_to_gen_compare_op[info->back.stencilCompareOp],
287 };
288
289 GEN8_3DSTATE_WM_DEPTH_STENCIL_pack(NULL, pipeline->gen8.wm_depth_stencil, &wm_depth_stencil);
290 }
291
292 VkResult
293 gen8_graphics_pipeline_create(
294 VkDevice _device,
295 const VkGraphicsPipelineCreateInfo* pCreateInfo,
296 const struct anv_graphics_pipeline_create_info *extra,
297 VkPipeline* pPipeline)
298 {
299 ANV_FROM_HANDLE(anv_device, device, _device);
300 struct anv_pipeline *pipeline;
301 VkResult result;
302 uint32_t offset, length;
303
304 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO);
305
306 pipeline = anv_device_alloc(device, sizeof(*pipeline), 8,
307 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
308 if (pipeline == NULL)
309 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
310
311 result = anv_pipeline_init(pipeline, device, pCreateInfo, extra);
312 if (result != VK_SUCCESS)
313 return result;
314
315 /* FIXME: The compiler dead-codes FS inputs when we don't have a VS, so we
316 * hard code this to num_attributes - 2. This is because the attributes
317 * include VUE header and position, which aren't counted as varying
318 * inputs. */
319 if (pipeline->vs_simd8 == NO_KERNEL) {
320 pipeline->wm_prog_data.num_varying_inputs =
321 pCreateInfo->pVertexInputState->attributeCount - 2;
322 }
323
324 assert(pCreateInfo->pVertexInputState);
325 emit_vertex_input(pipeline, pCreateInfo->pVertexInputState);
326 assert(pCreateInfo->pInputAssemblyState);
327 emit_ia_state(pipeline, pCreateInfo->pInputAssemblyState, extra);
328 assert(pCreateInfo->pRasterState);
329 emit_rs_state(pipeline, pCreateInfo->pRasterState, extra);
330 emit_ds_state(pipeline, pCreateInfo->pDepthStencilState);
331 emit_cb_state(pipeline, pCreateInfo->pColorBlendState);
332
333 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_STATISTICS,
334 .StatisticsEnable = true);
335 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_HS, .Enable = false);
336 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_TE, .TEEnable = false);
337 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_DS, .FunctionEnable = false);
338 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_STREAMOUT, .SOFunctionEnable = false);
339
340 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_VS,
341 .ConstantBufferOffset = 0,
342 .ConstantBufferSize = 4);
343 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_GS,
344 .ConstantBufferOffset = 4,
345 .ConstantBufferSize = 4);
346 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_PS,
347 .ConstantBufferOffset = 8,
348 .ConstantBufferSize = 4);
349
350 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_WM_CHROMAKEY,
351 .ChromaKeyKillEnable = false);
352 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_SBE_SWIZ);
353 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_AA_LINE_PARAMETERS);
354
355 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_CLIP,
356 .ClipEnable = true,
357 .ViewportXYClipTestEnable = !(extra && extra->disable_viewport),
358 .MinimumPointWidth = 0.125,
359 .MaximumPointWidth = 255.875);
360
361 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_WM,
362 .StatisticsEnable = true,
363 .LineEndCapAntialiasingRegionWidth = _05pixels,
364 .LineAntialiasingRegionWidth = _10pixels,
365 .EarlyDepthStencilControl = NORMAL,
366 .ForceThreadDispatchEnable = NORMAL,
367 .PointRasterizationRule = RASTRULE_UPPER_RIGHT,
368 .BarycentricInterpolationMode =
369 pipeline->wm_prog_data.barycentric_interp_modes);
370
371 uint32_t samples = 1;
372 uint32_t log2_samples = __builtin_ffs(samples) - 1;
373 bool enable_sampling = samples > 1 ? true : false;
374
375 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_MULTISAMPLE,
376 .PixelPositionOffsetEnable = enable_sampling,
377 .PixelLocation = CENTER,
378 .NumberofMultisamples = log2_samples);
379
380 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_SAMPLE_MASK,
381 .SampleMask = 0xffff);
382
383 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_VS,
384 .VSURBStartingAddress = pipeline->urb.vs_start,
385 .VSURBEntryAllocationSize = pipeline->urb.vs_size - 1,
386 .VSNumberofURBEntries = pipeline->urb.nr_vs_entries);
387
388 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_GS,
389 .GSURBStartingAddress = pipeline->urb.gs_start,
390 .GSURBEntryAllocationSize = pipeline->urb.gs_size - 1,
391 .GSNumberofURBEntries = pipeline->urb.nr_gs_entries);
392
393 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_HS,
394 .HSURBStartingAddress = pipeline->urb.vs_start,
395 .HSURBEntryAllocationSize = 0,
396 .HSNumberofURBEntries = 0);
397
398 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_DS,
399 .DSURBStartingAddress = pipeline->urb.vs_start,
400 .DSURBEntryAllocationSize = 0,
401 .DSNumberofURBEntries = 0);
402
403 const struct brw_gs_prog_data *gs_prog_data = &pipeline->gs_prog_data;
404 offset = 1;
405 length = (gs_prog_data->base.vue_map.num_slots + 1) / 2 - offset;
406
407 if (pipeline->gs_vec4 == NO_KERNEL)
408 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_GS, .Enable = false);
409 else
410 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_GS,
411 .SingleProgramFlow = false,
412 .KernelStartPointer = pipeline->gs_vec4,
413 .VectorMaskEnable = Vmask,
414 .SamplerCount = 0,
415 .BindingTableEntryCount = 0,
416 .ExpectedVertexCount = pipeline->gs_vertex_count,
417
418 .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_GEOMETRY],
419 .PerThreadScratchSpace = ffs(gs_prog_data->base.base.total_scratch / 2048),
420
421 .OutputVertexSize = gs_prog_data->output_vertex_size_hwords * 2 - 1,
422 .OutputTopology = gs_prog_data->output_topology,
423 .VertexURBEntryReadLength = gs_prog_data->base.urb_read_length,
424 .DispatchGRFStartRegisterForURBData =
425 gs_prog_data->base.base.dispatch_grf_start_reg,
426
427 .MaximumNumberofThreads = device->info.max_gs_threads,
428 .ControlDataHeaderSize = gs_prog_data->control_data_header_size_hwords,
429 //pipeline->gs_prog_data.dispatch_mode |
430 .StatisticsEnable = true,
431 .IncludePrimitiveID = gs_prog_data->include_primitive_id,
432 .ReorderMode = TRAILING,
433 .Enable = true,
434
435 .ControlDataFormat = gs_prog_data->control_data_format,
436
437 /* FIXME: mesa sets this based on ctx->Transform.ClipPlanesEnabled:
438 * UserClipDistanceClipTestEnableBitmask_3DSTATE_GS(v)
439 * UserClipDistanceCullTestEnableBitmask(v)
440 */
441
442 .VertexURBEntryOutputReadOffset = offset,
443 .VertexURBEntryOutputLength = length);
444
445 const struct brw_vue_prog_data *vue_prog_data = &pipeline->vs_prog_data.base;
446 /* Skip the VUE header and position slots */
447 offset = 1;
448 length = (vue_prog_data->vue_map.num_slots + 1) / 2 - offset;
449
450 if (pipeline->vs_simd8 == NO_KERNEL || (extra && extra->disable_vs))
451 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VS,
452 .FunctionEnable = false,
453 /* Even if VS is disabled, SBE still gets the amount of
454 * vertex data to read from this field. */
455 .VertexURBEntryOutputReadOffset = offset,
456 .VertexURBEntryOutputLength = length);
457 else
458 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VS,
459 .KernelStartPointer = pipeline->vs_simd8,
460 .SingleVertexDispatch = Multiple,
461 .VectorMaskEnable = Dmask,
462 .SamplerCount = 0,
463 .BindingTableEntryCount =
464 vue_prog_data->base.binding_table.size_bytes / 4,
465 .ThreadDispatchPriority = Normal,
466 .FloatingPointMode = IEEE754,
467 .IllegalOpcodeExceptionEnable = false,
468 .AccessesUAV = false,
469 .SoftwareExceptionEnable = false,
470
471 .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_VERTEX],
472 .PerThreadScratchSpace = ffs(vue_prog_data->base.total_scratch / 2048),
473
474 .DispatchGRFStartRegisterForURBData =
475 vue_prog_data->base.dispatch_grf_start_reg,
476 .VertexURBEntryReadLength = vue_prog_data->urb_read_length,
477 .VertexURBEntryReadOffset = 0,
478
479 .MaximumNumberofThreads = device->info.max_vs_threads - 1,
480 .StatisticsEnable = false,
481 .SIMD8DispatchEnable = true,
482 .VertexCacheDisable = false,
483 .FunctionEnable = true,
484
485 .VertexURBEntryOutputReadOffset = offset,
486 .VertexURBEntryOutputLength = length,
487 .UserClipDistanceClipTestEnableBitmask = 0,
488 .UserClipDistanceCullTestEnableBitmask = 0);
489
490 const struct brw_wm_prog_data *wm_prog_data = &pipeline->wm_prog_data;
491
492 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_SBE,
493 .ForceVertexURBEntryReadLength = false,
494 .ForceVertexURBEntryReadOffset = false,
495 .PointSpriteTextureCoordinateOrigin = UPPERLEFT,
496 .NumberofSFOutputAttributes =
497 wm_prog_data->num_varying_inputs);
498
499 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PS,
500 .KernelStartPointer0 = pipeline->ps_ksp0,
501
502 .SingleProgramFlow = false,
503 .VectorMaskEnable = true,
504 .SamplerCount = 1,
505
506 .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_FRAGMENT],
507 .PerThreadScratchSpace = ffs(wm_prog_data->base.total_scratch / 2048),
508
509 .MaximumNumberofThreadsPerPSD = 64 - 2,
510 .PositionXYOffsetSelect = wm_prog_data->uses_pos_offset ?
511 POSOFFSET_SAMPLE: POSOFFSET_NONE,
512 .PushConstantEnable = wm_prog_data->base.nr_params > 0,
513 ._8PixelDispatchEnable = pipeline->ps_simd8 != NO_KERNEL,
514 ._16PixelDispatchEnable = pipeline->ps_simd16 != NO_KERNEL,
515 ._32PixelDispatchEnable = false,
516
517 .DispatchGRFStartRegisterForConstantSetupData0 = pipeline->ps_grf_start0,
518 .DispatchGRFStartRegisterForConstantSetupData1 = 0,
519 .DispatchGRFStartRegisterForConstantSetupData2 = pipeline->ps_grf_start2,
520
521 .KernelStartPointer1 = 0,
522 .KernelStartPointer2 = pipeline->ps_ksp2);
523
524 bool per_sample_ps = false;
525 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PS_EXTRA,
526 .PixelShaderValid = true,
527 .PixelShaderKillsPixel = wm_prog_data->uses_kill,
528 .PixelShaderComputedDepthMode = wm_prog_data->computed_depth_mode,
529 .AttributeEnable = wm_prog_data->num_varying_inputs > 0,
530 .oMaskPresenttoRenderTarget = wm_prog_data->uses_omask,
531 .PixelShaderIsPerSample = per_sample_ps);
532
533 *pPipeline = anv_pipeline_to_handle(pipeline);
534
535 return VK_SUCCESS;
536 }
537
538 VkResult gen8_compute_pipeline_create(
539 VkDevice _device,
540 const VkComputePipelineCreateInfo* pCreateInfo,
541 VkPipeline* pPipeline)
542 {
543 ANV_FROM_HANDLE(anv_device, device, _device);
544 struct anv_pipeline *pipeline;
545 VkResult result;
546
547 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO);
548
549 pipeline = anv_device_alloc(device, sizeof(*pipeline), 8,
550 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
551 if (pipeline == NULL)
552 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
553
554 pipeline->device = device;
555 pipeline->layout = anv_pipeline_layout_from_handle(pCreateInfo->layout);
556
557 result = anv_reloc_list_init(&pipeline->batch_relocs, device);
558 if (result != VK_SUCCESS) {
559 anv_device_free(device, pipeline);
560 return result;
561 }
562 pipeline->batch.next = pipeline->batch.start = pipeline->batch_data;
563 pipeline->batch.end = pipeline->batch.start + sizeof(pipeline->batch_data);
564 pipeline->batch.relocs = &pipeline->batch_relocs;
565
566 anv_state_stream_init(&pipeline->program_stream,
567 &device->instruction_block_pool);
568
569 memset(pipeline->shaders, 0, sizeof(pipeline->shaders));
570
571 pipeline->shaders[VK_SHADER_STAGE_COMPUTE] =
572 anv_shader_from_handle(pCreateInfo->cs.shader);
573
574 pipeline->use_repclear = false;
575
576 anv_compiler_run(device->compiler, pipeline);
577
578 const struct brw_cs_prog_data *cs_prog_data = &pipeline->cs_prog_data;
579
580 anv_batch_emit(&pipeline->batch, GEN8_MEDIA_VFE_STATE,
581 .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_FRAGMENT],
582 .PerThreadScratchSpace = ffs(cs_prog_data->base.total_scratch / 2048),
583 .ScratchSpaceBasePointerHigh = 0,
584 .StackSize = 0,
585
586 .MaximumNumberofThreads = device->info.max_cs_threads - 1,
587 .NumberofURBEntries = 2,
588 .ResetGatewayTimer = true,
589 .BypassGatewayControl = true,
590 .URBEntryAllocationSize = 2,
591 .CURBEAllocationSize = 0);
592
593 struct brw_cs_prog_data *prog_data = &pipeline->cs_prog_data;
594 uint32_t group_size = prog_data->local_size[0] *
595 prog_data->local_size[1] * prog_data->local_size[2];
596 pipeline->cs_thread_width_max = DIV_ROUND_UP(group_size, prog_data->simd_size);
597 uint32_t remainder = group_size & (prog_data->simd_size - 1);
598
599 if (remainder > 0)
600 pipeline->cs_right_mask = ~0u >> (32 - remainder);
601 else
602 pipeline->cs_right_mask = ~0u >> (32 - prog_data->simd_size);
603
604
605 *pPipeline = anv_pipeline_to_handle(pipeline);
606
607 return VK_SUCCESS;
608 }