gen8/pipeline: Minor blending fixes
[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 #include "gen8_pack.h"
33
34 static void
35 emit_vertex_input(struct anv_pipeline *pipeline,
36 const VkPipelineVertexInputStateCreateInfo *info)
37 {
38 const uint32_t num_dwords = 1 + info->attributeCount * 2;
39 uint32_t *p;
40
41 if (info->attributeCount > 0) {
42 p = anv_batch_emitn(&pipeline->batch, num_dwords,
43 GEN8_3DSTATE_VERTEX_ELEMENTS);
44 }
45
46 for (uint32_t i = 0; i < info->attributeCount; i++) {
47 const VkVertexInputAttributeDescription *desc =
48 &info->pVertexAttributeDescriptions[i];
49 const struct anv_format *format = anv_format_for_vk_format(desc->format);
50
51 struct GEN8_VERTEX_ELEMENT_STATE element = {
52 .VertexBufferIndex = desc->binding,
53 .Valid = true,
54 .SourceElementFormat = format->surface_format,
55 .EdgeFlagEnable = false,
56 .SourceElementOffset = desc->offsetInBytes,
57 .Component0Control = VFCOMP_STORE_SRC,
58 .Component1Control = format->num_channels >= 2 ? VFCOMP_STORE_SRC : VFCOMP_STORE_0,
59 .Component2Control = format->num_channels >= 3 ? VFCOMP_STORE_SRC : VFCOMP_STORE_0,
60 .Component3Control = format->num_channels >= 4 ? VFCOMP_STORE_SRC : VFCOMP_STORE_1_FP
61 };
62 GEN8_VERTEX_ELEMENT_STATE_pack(NULL, &p[1 + i * 2], &element);
63
64 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_INSTANCING,
65 .InstancingEnable = pipeline->instancing_enable[desc->binding],
66 .VertexElementIndex = i,
67 /* Vulkan so far doesn't have an instance divisor, so
68 * this is always 1 (ignored if not instancing). */
69 .InstanceDataStepRate = 1);
70 }
71
72 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_SGVS,
73 .VertexIDEnable = pipeline->vs_prog_data.uses_vertexid,
74 .VertexIDComponentNumber = 2,
75 .VertexIDElementOffset = info->bindingCount,
76 .InstanceIDEnable = pipeline->vs_prog_data.uses_instanceid,
77 .InstanceIDComponentNumber = 3,
78 .InstanceIDElementOffset = info->bindingCount);
79 }
80
81 static void
82 emit_ia_state(struct anv_pipeline *pipeline,
83 const VkPipelineInputAssemblyStateCreateInfo *info,
84 const struct anv_graphics_pipeline_create_info *extra)
85 {
86 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_TOPOLOGY,
87 .PrimitiveTopologyType = pipeline->topology);
88 }
89
90 static void
91 emit_rs_state(struct anv_pipeline *pipeline,
92 const VkPipelineRasterStateCreateInfo *info,
93 const struct anv_graphics_pipeline_create_info *extra)
94 {
95 static const uint32_t vk_to_gen_cullmode[] = {
96 [VK_CULL_MODE_NONE] = CULLMODE_NONE,
97 [VK_CULL_MODE_FRONT] = CULLMODE_FRONT,
98 [VK_CULL_MODE_BACK] = CULLMODE_BACK,
99 [VK_CULL_MODE_FRONT_AND_BACK] = CULLMODE_BOTH
100 };
101
102 static const uint32_t vk_to_gen_fillmode[] = {
103 [VK_FILL_MODE_POINTS] = RASTER_POINT,
104 [VK_FILL_MODE_WIREFRAME] = RASTER_WIREFRAME,
105 [VK_FILL_MODE_SOLID] = RASTER_SOLID
106 };
107
108 static const uint32_t vk_to_gen_front_face[] = {
109 [VK_FRONT_FACE_CCW] = CounterClockwise,
110 [VK_FRONT_FACE_CW] = Clockwise
111 };
112
113 struct GEN8_3DSTATE_SF sf = {
114 GEN8_3DSTATE_SF_header,
115 .ViewportTransformEnable = !(extra && extra->disable_viewport),
116 .TriangleStripListProvokingVertexSelect = 0,
117 .LineStripListProvokingVertexSelect = 0,
118 .TriangleFanProvokingVertexSelect = 0,
119 .PointWidthSource = pipeline->writes_point_size ? Vertex : State,
120 .PointWidth = 1.0,
121 };
122
123 /* FINISHME: VkBool32 rasterizerDiscardEnable; */
124
125 GEN8_3DSTATE_SF_pack(NULL, pipeline->gen8.sf, &sf);
126
127 struct GEN8_3DSTATE_RASTER raster = {
128 GEN8_3DSTATE_RASTER_header,
129 .FrontWinding = vk_to_gen_front_face[info->frontFace],
130 .CullMode = vk_to_gen_cullmode[info->cullMode],
131 .FrontFaceFillMode = vk_to_gen_fillmode[info->fillMode],
132 .BackFaceFillMode = vk_to_gen_fillmode[info->fillMode],
133 .ScissorRectangleEnable = !(extra && extra->disable_scissor),
134 .ViewportZClipTestEnable = info->depthClipEnable
135 };
136
137 GEN8_3DSTATE_RASTER_pack(NULL, pipeline->gen8.raster, &raster);
138 }
139
140 static void
141 emit_cb_state(struct anv_pipeline *pipeline,
142 const VkPipelineColorBlendStateCreateInfo *info)
143 {
144 struct anv_device *device = pipeline->device;
145
146 static const uint32_t vk_to_gen_logic_op[] = {
147 [VK_LOGIC_OP_COPY] = LOGICOP_COPY,
148 [VK_LOGIC_OP_CLEAR] = LOGICOP_CLEAR,
149 [VK_LOGIC_OP_AND] = LOGICOP_AND,
150 [VK_LOGIC_OP_AND_REVERSE] = LOGICOP_AND_REVERSE,
151 [VK_LOGIC_OP_AND_INVERTED] = LOGICOP_AND_INVERTED,
152 [VK_LOGIC_OP_NOOP] = LOGICOP_NOOP,
153 [VK_LOGIC_OP_XOR] = LOGICOP_XOR,
154 [VK_LOGIC_OP_OR] = LOGICOP_OR,
155 [VK_LOGIC_OP_NOR] = LOGICOP_NOR,
156 [VK_LOGIC_OP_EQUIV] = LOGICOP_EQUIV,
157 [VK_LOGIC_OP_INVERT] = LOGICOP_INVERT,
158 [VK_LOGIC_OP_OR_REVERSE] = LOGICOP_OR_REVERSE,
159 [VK_LOGIC_OP_COPY_INVERTED] = LOGICOP_COPY_INVERTED,
160 [VK_LOGIC_OP_OR_INVERTED] = LOGICOP_OR_INVERTED,
161 [VK_LOGIC_OP_NAND] = LOGICOP_NAND,
162 [VK_LOGIC_OP_SET] = LOGICOP_SET,
163 };
164
165 static const uint32_t vk_to_gen_blend[] = {
166 [VK_BLEND_ZERO] = BLENDFACTOR_ZERO,
167 [VK_BLEND_ONE] = BLENDFACTOR_ONE,
168 [VK_BLEND_SRC_COLOR] = BLENDFACTOR_SRC_COLOR,
169 [VK_BLEND_ONE_MINUS_SRC_COLOR] = BLENDFACTOR_INV_SRC_COLOR,
170 [VK_BLEND_DEST_COLOR] = BLENDFACTOR_DST_COLOR,
171 [VK_BLEND_ONE_MINUS_DEST_COLOR] = BLENDFACTOR_INV_DST_COLOR,
172 [VK_BLEND_SRC_ALPHA] = BLENDFACTOR_SRC_ALPHA,
173 [VK_BLEND_ONE_MINUS_SRC_ALPHA] = BLENDFACTOR_INV_SRC_ALPHA,
174 [VK_BLEND_DEST_ALPHA] = BLENDFACTOR_DST_ALPHA,
175 [VK_BLEND_ONE_MINUS_DEST_ALPHA] = BLENDFACTOR_INV_DST_ALPHA,
176 [VK_BLEND_CONSTANT_COLOR] = BLENDFACTOR_CONST_COLOR,
177 [VK_BLEND_ONE_MINUS_CONSTANT_COLOR] = BLENDFACTOR_INV_CONST_COLOR,
178 [VK_BLEND_CONSTANT_ALPHA] = BLENDFACTOR_CONST_ALPHA,
179 [VK_BLEND_ONE_MINUS_CONSTANT_ALPHA] = BLENDFACTOR_INV_CONST_ALPHA,
180 [VK_BLEND_SRC_ALPHA_SATURATE] = BLENDFACTOR_SRC_ALPHA_SATURATE,
181 [VK_BLEND_SRC1_COLOR] = BLENDFACTOR_SRC1_COLOR,
182 [VK_BLEND_ONE_MINUS_SRC1_COLOR] = BLENDFACTOR_INV_SRC1_COLOR,
183 [VK_BLEND_SRC1_ALPHA] = BLENDFACTOR_SRC1_ALPHA,
184 [VK_BLEND_ONE_MINUS_SRC1_ALPHA] = BLENDFACTOR_INV_SRC1_ALPHA,
185 };
186
187 static const uint32_t vk_to_gen_blend_op[] = {
188 [VK_BLEND_OP_ADD] = BLENDFUNCTION_ADD,
189 [VK_BLEND_OP_SUBTRACT] = BLENDFUNCTION_SUBTRACT,
190 [VK_BLEND_OP_REVERSE_SUBTRACT] = BLENDFUNCTION_REVERSE_SUBTRACT,
191 [VK_BLEND_OP_MIN] = BLENDFUNCTION_MIN,
192 [VK_BLEND_OP_MAX] = BLENDFUNCTION_MAX,
193 };
194
195 uint32_t num_dwords = GEN8_BLEND_STATE_length;
196 pipeline->blend_state =
197 anv_state_pool_alloc(&device->dynamic_state_pool, num_dwords * 4, 64);
198
199 struct GEN8_BLEND_STATE blend_state = {
200 .AlphaToCoverageEnable = info->alphaToCoverageEnable,
201 .AlphaToOneEnable = info->alphaToOneEnable,
202 };
203
204 for (uint32_t i = 0; i < info->attachmentCount; i++) {
205 const VkPipelineColorBlendAttachmentState *a = &info->pAttachments[i];
206
207 blend_state.Entry[i] = (struct GEN8_BLEND_STATE_ENTRY) {
208 .LogicOpEnable = info->logicOpEnable,
209 .LogicOpFunction = vk_to_gen_logic_op[info->logicOp],
210 .ColorBufferBlendEnable = a->blendEnable,
211 .PreBlendSourceOnlyClampEnable = false,
212 .ColorClampRange = COLORCLAMP_RTFORMAT,
213 .PreBlendColorClampEnable = true,
214 .PostBlendColorClampEnable = true,
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] = STENCILOP_KEEP,
248 [VK_STENCIL_OP_ZERO] = STENCILOP_ZERO,
249 [VK_STENCIL_OP_REPLACE] = STENCILOP_REPLACE,
250 [VK_STENCIL_OP_INC_CLAMP] = STENCILOP_INCRSAT,
251 [VK_STENCIL_OP_DEC_CLAMP] = STENCILOP_DECRSAT,
252 [VK_STENCIL_OP_INVERT] = STENCILOP_INVERT,
253 [VK_STENCIL_OP_INC_WRAP] = STENCILOP_INCR,
254 [VK_STENCIL_OP_DEC_WRAP] = STENCILOP_DECR,
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 depthBoundsTestEnable; // 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_AA_LINE_PARAMETERS);
353
354 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_CLIP,
355 .ClipEnable = true,
356 .ViewportXYClipTestEnable = !(extra && extra->disable_viewport),
357 .MinimumPointWidth = 0.125,
358 .MaximumPointWidth = 255.875);
359
360 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_WM,
361 .StatisticsEnable = true,
362 .LineEndCapAntialiasingRegionWidth = _05pixels,
363 .LineAntialiasingRegionWidth = _10pixels,
364 .EarlyDepthStencilControl = NORMAL,
365 .ForceThreadDispatchEnable = NORMAL,
366 .PointRasterizationRule = RASTRULE_UPPER_RIGHT,
367 .BarycentricInterpolationMode =
368 pipeline->wm_prog_data.barycentric_interp_modes);
369
370 uint32_t samples = 1;
371 uint32_t log2_samples = __builtin_ffs(samples) - 1;
372 bool enable_sampling = samples > 1 ? true : false;
373
374 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_MULTISAMPLE,
375 .PixelPositionOffsetEnable = enable_sampling,
376 .PixelLocation = CENTER,
377 .NumberofMultisamples = log2_samples);
378
379 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_SAMPLE_MASK,
380 .SampleMask = 0xffff);
381
382 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_VS,
383 .VSURBStartingAddress = pipeline->urb.vs_start,
384 .VSURBEntryAllocationSize = pipeline->urb.vs_size - 1,
385 .VSNumberofURBEntries = pipeline->urb.nr_vs_entries);
386
387 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_GS,
388 .GSURBStartingAddress = pipeline->urb.gs_start,
389 .GSURBEntryAllocationSize = pipeline->urb.gs_size - 1,
390 .GSNumberofURBEntries = pipeline->urb.nr_gs_entries);
391
392 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_HS,
393 .HSURBStartingAddress = pipeline->urb.vs_start,
394 .HSURBEntryAllocationSize = 0,
395 .HSNumberofURBEntries = 0);
396
397 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_DS,
398 .DSURBStartingAddress = pipeline->urb.vs_start,
399 .DSURBEntryAllocationSize = 0,
400 .DSNumberofURBEntries = 0);
401
402 const struct brw_gs_prog_data *gs_prog_data = &pipeline->gs_prog_data;
403 offset = 1;
404 length = (gs_prog_data->base.vue_map.num_slots + 1) / 2 - offset;
405
406 if (pipeline->gs_vec4 == NO_KERNEL)
407 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_GS, .Enable = false);
408 else
409 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_GS,
410 .SingleProgramFlow = false,
411 .KernelStartPointer = pipeline->gs_vec4,
412 .VectorMaskEnable = Dmask,
413 .SamplerCount = 0,
414 .BindingTableEntryCount = 0,
415 .ExpectedVertexCount = pipeline->gs_vertex_count,
416
417 .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_GEOMETRY],
418 .PerThreadScratchSpace = ffs(gs_prog_data->base.base.total_scratch / 2048),
419
420 .OutputVertexSize = gs_prog_data->output_vertex_size_hwords * 2 - 1,
421 .OutputTopology = gs_prog_data->output_topology,
422 .VertexURBEntryReadLength = gs_prog_data->base.urb_read_length,
423 .DispatchGRFStartRegisterForURBData =
424 gs_prog_data->base.base.dispatch_grf_start_reg,
425
426 .MaximumNumberofThreads = device->info.max_gs_threads / 2 - 1,
427 .ControlDataHeaderSize = gs_prog_data->control_data_header_size_hwords,
428 .DispatchMode = gs_prog_data->base.dispatch_mode,
429 .StatisticsEnable = true,
430 .IncludePrimitiveID = gs_prog_data->include_primitive_id,
431 .ReorderMode = TRAILING,
432 .Enable = true,
433
434 .ControlDataFormat = gs_prog_data->control_data_format,
435
436 .StaticOutput = gs_prog_data->static_vertex_count >= 0,
437 .StaticOutputVertexCount =
438 gs_prog_data->static_vertex_count >= 0 ?
439 gs_prog_data->static_vertex_count : 0,
440
441 /* FIXME: mesa sets this based on ctx->Transform.ClipPlanesEnabled:
442 * UserClipDistanceClipTestEnableBitmask_3DSTATE_GS(v)
443 * UserClipDistanceCullTestEnableBitmask(v)
444 */
445
446 .VertexURBEntryOutputReadOffset = offset,
447 .VertexURBEntryOutputLength = length);
448
449 const struct brw_vue_prog_data *vue_prog_data = &pipeline->vs_prog_data.base;
450 /* Skip the VUE header and position slots */
451 offset = 1;
452 length = (vue_prog_data->vue_map.num_slots + 1) / 2 - offset;
453
454 if (pipeline->vs_simd8 == NO_KERNEL || (extra && extra->disable_vs))
455 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VS,
456 .FunctionEnable = false,
457 /* Even if VS is disabled, SBE still gets the amount of
458 * vertex data to read from this field. */
459 .VertexURBEntryOutputReadOffset = offset,
460 .VertexURBEntryOutputLength = length);
461 else
462 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VS,
463 .KernelStartPointer = pipeline->vs_simd8,
464 .SingleVertexDispatch = Multiple,
465 .VectorMaskEnable = Dmask,
466 .SamplerCount = 0,
467 .BindingTableEntryCount =
468 vue_prog_data->base.binding_table.size_bytes / 4,
469 .ThreadDispatchPriority = Normal,
470 .FloatingPointMode = IEEE754,
471 .IllegalOpcodeExceptionEnable = false,
472 .AccessesUAV = false,
473 .SoftwareExceptionEnable = false,
474
475 .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_VERTEX],
476 .PerThreadScratchSpace = ffs(vue_prog_data->base.total_scratch / 2048),
477
478 .DispatchGRFStartRegisterForURBData =
479 vue_prog_data->base.dispatch_grf_start_reg,
480 .VertexURBEntryReadLength = vue_prog_data->urb_read_length,
481 .VertexURBEntryReadOffset = 0,
482
483 .MaximumNumberofThreads = device->info.max_vs_threads - 1,
484 .StatisticsEnable = false,
485 .SIMD8DispatchEnable = true,
486 .VertexCacheDisable = false,
487 .FunctionEnable = true,
488
489 .VertexURBEntryOutputReadOffset = offset,
490 .VertexURBEntryOutputLength = length,
491 .UserClipDistanceClipTestEnableBitmask = 0,
492 .UserClipDistanceCullTestEnableBitmask = 0);
493
494 const struct brw_wm_prog_data *wm_prog_data = &pipeline->wm_prog_data;
495
496 /* TODO: We should clean this up. Among other things, this is mostly
497 * shared with other gens.
498 */
499 const struct brw_vue_map *fs_input_map;
500 if (pipeline->gs_vec4 == NO_KERNEL)
501 fs_input_map = &vue_prog_data->vue_map;
502 else
503 fs_input_map = &gs_prog_data->base.vue_map;
504
505 struct GEN8_3DSTATE_SBE_SWIZ swiz = {
506 GEN8_3DSTATE_SBE_SWIZ_header,
507 };
508
509 int max_source_attr = 0;
510 for (int attr = 0; attr < VARYING_SLOT_MAX; attr++) {
511 int input_index = wm_prog_data->urb_setup[attr];
512
513 if (input_index < 0)
514 continue;
515
516 /* We have to subtract two slots to accout for the URB entry output
517 * read offset in the VS and GS stages.
518 */
519 int source_attr = fs_input_map->varying_to_slot[attr] - 2;
520 max_source_attr = MAX2(max_source_attr, source_attr);
521
522 if (input_index >= 16)
523 continue;
524
525 swiz.Attribute[input_index].SourceAttribute = source_attr;
526 }
527
528 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_SBE,
529 .AttributeSwizzleEnable = true,
530 .ForceVertexURBEntryReadLength = false,
531 .ForceVertexURBEntryReadOffset = false,
532 .VertexURBEntryReadLength = DIV_ROUND_UP(max_source_attr + 1, 2),
533 .PointSpriteTextureCoordinateOrigin = UPPERLEFT,
534 .NumberofSFOutputAttributes =
535 wm_prog_data->num_varying_inputs);
536
537 uint32_t *dw = anv_batch_emit_dwords(&pipeline->batch,
538 GEN8_3DSTATE_SBE_SWIZ_length);
539 GEN8_3DSTATE_SBE_SWIZ_pack(&pipeline->batch, dw, &swiz);
540
541 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PS,
542 .KernelStartPointer0 = pipeline->ps_ksp0,
543
544 .SingleProgramFlow = false,
545 .VectorMaskEnable = true,
546 .SamplerCount = 1,
547
548 .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_FRAGMENT],
549 .PerThreadScratchSpace = ffs(wm_prog_data->base.total_scratch / 2048),
550
551 .MaximumNumberofThreadsPerPSD = 64 - 2,
552 .PositionXYOffsetSelect = wm_prog_data->uses_pos_offset ?
553 POSOFFSET_SAMPLE: POSOFFSET_NONE,
554 .PushConstantEnable = wm_prog_data->base.nr_params > 0,
555 ._8PixelDispatchEnable = pipeline->ps_simd8 != NO_KERNEL,
556 ._16PixelDispatchEnable = pipeline->ps_simd16 != NO_KERNEL,
557 ._32PixelDispatchEnable = false,
558
559 .DispatchGRFStartRegisterForConstantSetupData0 = pipeline->ps_grf_start0,
560 .DispatchGRFStartRegisterForConstantSetupData1 = 0,
561 .DispatchGRFStartRegisterForConstantSetupData2 = pipeline->ps_grf_start2,
562
563 .KernelStartPointer1 = 0,
564 .KernelStartPointer2 = pipeline->ps_ksp2);
565
566 bool per_sample_ps = false;
567 anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PS_EXTRA,
568 .PixelShaderValid = true,
569 .PixelShaderKillsPixel = wm_prog_data->uses_kill,
570 .PixelShaderComputedDepthMode = wm_prog_data->computed_depth_mode,
571 .AttributeEnable = wm_prog_data->num_varying_inputs > 0,
572 .oMaskPresenttoRenderTarget = wm_prog_data->uses_omask,
573 .PixelShaderIsPerSample = per_sample_ps);
574
575 *pPipeline = anv_pipeline_to_handle(pipeline);
576
577 return VK_SUCCESS;
578 }
579
580 VkResult gen8_compute_pipeline_create(
581 VkDevice _device,
582 const VkComputePipelineCreateInfo* pCreateInfo,
583 VkPipeline* pPipeline)
584 {
585 ANV_FROM_HANDLE(anv_device, device, _device);
586 struct anv_pipeline *pipeline;
587 VkResult result;
588
589 assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO);
590
591 pipeline = anv_device_alloc(device, sizeof(*pipeline), 8,
592 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
593 if (pipeline == NULL)
594 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
595
596 pipeline->device = device;
597 pipeline->layout = anv_pipeline_layout_from_handle(pCreateInfo->layout);
598
599 pipeline->blend_state.map = NULL;
600
601 result = anv_reloc_list_init(&pipeline->batch_relocs, device);
602 if (result != VK_SUCCESS) {
603 anv_device_free(device, pipeline);
604 return result;
605 }
606 pipeline->batch.next = pipeline->batch.start = pipeline->batch_data;
607 pipeline->batch.end = pipeline->batch.start + sizeof(pipeline->batch_data);
608 pipeline->batch.relocs = &pipeline->batch_relocs;
609
610 anv_state_stream_init(&pipeline->program_stream,
611 &device->instruction_block_pool);
612
613 /* When we free the pipeline, we detect stages based on the NULL status
614 * of various prog_data pointers. Make them NULL by default.
615 */
616 memset(pipeline->prog_data, 0, sizeof(pipeline->prog_data));
617 memset(pipeline->scratch_start, 0, sizeof(pipeline->scratch_start));
618
619 pipeline->vs_simd8 = NO_KERNEL;
620 pipeline->vs_vec4 = NO_KERNEL;
621 pipeline->gs_vec4 = NO_KERNEL;
622
623 pipeline->active_stages = 0;
624 pipeline->total_scratch = 0;
625
626 assert(pCreateInfo->stage.stage == VK_SHADER_STAGE_COMPUTE);
627 ANV_FROM_HANDLE(anv_shader, shader, pCreateInfo->stage.shader);
628 anv_pipeline_compile_cs(pipeline, pCreateInfo, shader);
629
630 pipeline->use_repclear = false;
631
632 const struct brw_cs_prog_data *cs_prog_data = &pipeline->cs_prog_data;
633
634 anv_batch_emit(&pipeline->batch, GEN8_MEDIA_VFE_STATE,
635 .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_COMPUTE],
636 .PerThreadScratchSpace = ffs(cs_prog_data->base.total_scratch / 2048),
637 .ScratchSpaceBasePointerHigh = 0,
638 .StackSize = 0,
639
640 .MaximumNumberofThreads = device->info.max_cs_threads - 1,
641 .NumberofURBEntries = 2,
642 .ResetGatewayTimer = true,
643 .BypassGatewayControl = true,
644 .URBEntryAllocationSize = 2,
645 .CURBEAllocationSize = 0);
646
647 struct brw_cs_prog_data *prog_data = &pipeline->cs_prog_data;
648 uint32_t group_size = prog_data->local_size[0] *
649 prog_data->local_size[1] * prog_data->local_size[2];
650 pipeline->cs_thread_width_max = DIV_ROUND_UP(group_size, prog_data->simd_size);
651 uint32_t remainder = group_size & (prog_data->simd_size - 1);
652
653 if (remainder > 0)
654 pipeline->cs_right_mask = ~0u >> (32 - remainder);
655 else
656 pipeline->cs_right_mask = ~0u >> (32 - prog_data->simd_size);
657
658
659 *pPipeline = anv_pipeline_to_handle(pipeline);
660
661 return VK_SUCCESS;
662 }