2 * Copyright © 2015 Intel Corporation
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:
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
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
34 VkResult
anv_CreateShaderModule(
36 const VkShaderModuleCreateInfo
* pCreateInfo
,
37 VkShader
* pShaderModule
)
39 ANV_FROM_HANDLE(anv_device
, device
, _device
);
40 struct anv_shader_module
*module
;
42 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO
);
43 assert(pCreateInfo
->flags
== 0);
45 module
= anv_device_alloc(device
, sizeof(*module
) + pCreateInfo
->codeSize
, 8,
46 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
48 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
50 module
->size
= pCreateInfo
->codeSize
;
51 memcpy(module
->data
, pCreateInfo
->pCode
, module
->size
);
53 *pShaderModule
= (VkShaderModule
) module
;
58 VkResult
anv_CreateShader(
60 const VkShaderCreateInfo
* pCreateInfo
,
63 ANV_FROM_HANDLE(anv_device
, device
, _device
);
64 ANV_FROM_HANDLE(anv_shader_module
, module
, pCreateInfo
->module
);
65 struct anv_shader
*shader
;
67 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_SHADER_CREATE_INFO
);
68 assert(pCreateInfo
->flags
== 0);
70 size_t name_len
= strlen(pCreateInfo
->pName
);
72 if (strcmp(pCreateInfo
->pName
, "main") != 0) {
73 anv_finishme("Multiple shaders per module not really supported");
76 shader
= anv_device_alloc(device
, sizeof(*shader
) + name_len
+ 1, 8,
77 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
79 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
81 shader
->module
= module
;
82 memcpy(shader
->entrypoint
, pCreateInfo
->pName
, name_len
+ 1);
84 *pShader
= (VkShader
) shader
;
89 VkResult
anv_CreatePipelineCache(
91 const VkPipelineCacheCreateInfo
* pCreateInfo
,
92 VkPipelineCache
* pPipelineCache
)
96 stub_return(VK_SUCCESS
);
99 size_t anv_GetPipelineCacheSize(
101 VkPipelineCache pipelineCache
)
106 VkResult
anv_GetPipelineCacheData(
108 VkPipelineCache pipelineCache
,
111 stub_return(VK_UNSUPPORTED
);
114 VkResult
anv_MergePipelineCaches(
116 VkPipelineCache destCache
,
117 uint32_t srcCacheCount
,
118 const VkPipelineCache
* pSrcCaches
)
120 stub_return(VK_UNSUPPORTED
);
123 // Pipeline functions
126 emit_vertex_input(struct anv_pipeline
*pipeline
,
127 const VkPipelineVertexInputStateCreateInfo
*info
)
129 const uint32_t num_dwords
= 1 + info
->attributeCount
* 2;
131 bool instancing_enable
[32];
133 pipeline
->vb_used
= 0;
134 for (uint32_t i
= 0; i
< info
->bindingCount
; i
++) {
135 const VkVertexInputBindingDescription
*desc
=
136 &info
->pVertexBindingDescriptions
[i
];
138 pipeline
->vb_used
|= 1 << desc
->binding
;
139 pipeline
->binding_stride
[desc
->binding
] = desc
->strideInBytes
;
141 /* Step rate is programmed per vertex element (attribute), not
142 * binding. Set up a map of which bindings step per instance, for
143 * reference by vertex element setup. */
144 switch (desc
->stepRate
) {
146 case VK_VERTEX_INPUT_STEP_RATE_VERTEX
:
147 instancing_enable
[desc
->binding
] = false;
149 case VK_VERTEX_INPUT_STEP_RATE_INSTANCE
:
150 instancing_enable
[desc
->binding
] = true;
155 p
= anv_batch_emitn(&pipeline
->batch
, num_dwords
,
156 GEN8_3DSTATE_VERTEX_ELEMENTS
);
158 for (uint32_t i
= 0; i
< info
->attributeCount
; i
++) {
159 const VkVertexInputAttributeDescription
*desc
=
160 &info
->pVertexAttributeDescriptions
[i
];
161 const struct anv_format
*format
= anv_format_for_vk_format(desc
->format
);
163 struct GEN8_VERTEX_ELEMENT_STATE element
= {
164 .VertexBufferIndex
= desc
->binding
,
166 .SourceElementFormat
= format
->surface_format
,
167 .EdgeFlagEnable
= false,
168 .SourceElementOffset
= desc
->offsetInBytes
,
169 .Component0Control
= VFCOMP_STORE_SRC
,
170 .Component1Control
= format
->num_channels
>= 2 ? VFCOMP_STORE_SRC
: VFCOMP_STORE_0
,
171 .Component2Control
= format
->num_channels
>= 3 ? VFCOMP_STORE_SRC
: VFCOMP_STORE_0
,
172 .Component3Control
= format
->num_channels
>= 4 ? VFCOMP_STORE_SRC
: VFCOMP_STORE_1_FP
174 GEN8_VERTEX_ELEMENT_STATE_pack(NULL
, &p
[1 + i
* 2], &element
);
176 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_VF_INSTANCING
,
177 .InstancingEnable
= instancing_enable
[desc
->binding
],
178 .VertexElementIndex
= i
,
179 /* Vulkan so far doesn't have an instance divisor, so
180 * this is always 1 (ignored if not instancing). */
181 .InstanceDataStepRate
= 1);
184 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_VF_SGVS
,
185 .VertexIDEnable
= pipeline
->vs_prog_data
.uses_vertexid
,
186 .VertexIDComponentNumber
= 2,
187 .VertexIDElementOffset
= info
->bindingCount
,
188 .InstanceIDEnable
= pipeline
->vs_prog_data
.uses_instanceid
,
189 .InstanceIDComponentNumber
= 3,
190 .InstanceIDElementOffset
= info
->bindingCount
);
194 emit_ia_state(struct anv_pipeline
*pipeline
,
195 const VkPipelineIaStateCreateInfo
*info
,
196 const struct anv_pipeline_create_info
*extra
)
198 static const uint32_t vk_to_gen_primitive_type
[] = {
199 [VK_PRIMITIVE_TOPOLOGY_POINT_LIST
] = _3DPRIM_POINTLIST
,
200 [VK_PRIMITIVE_TOPOLOGY_LINE_LIST
] = _3DPRIM_LINELIST
,
201 [VK_PRIMITIVE_TOPOLOGY_LINE_STRIP
] = _3DPRIM_LINESTRIP
,
202 [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST
] = _3DPRIM_TRILIST
,
203 [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP
] = _3DPRIM_TRISTRIP
,
204 [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN
] = _3DPRIM_TRIFAN
,
205 [VK_PRIMITIVE_TOPOLOGY_LINE_LIST_ADJ
] = _3DPRIM_LINELIST_ADJ
,
206 [VK_PRIMITIVE_TOPOLOGY_LINE_STRIP_ADJ
] = _3DPRIM_LINESTRIP_ADJ
,
207 [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_ADJ
] = _3DPRIM_TRILIST_ADJ
,
208 [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP_ADJ
] = _3DPRIM_TRISTRIP_ADJ
,
209 [VK_PRIMITIVE_TOPOLOGY_PATCH
] = _3DPRIM_PATCHLIST_1
211 uint32_t topology
= vk_to_gen_primitive_type
[info
->topology
];
213 if (extra
&& extra
->use_rectlist
)
214 topology
= _3DPRIM_RECTLIST
;
216 struct GEN8_3DSTATE_VF vf
= {
217 GEN8_3DSTATE_VF_header
,
218 .IndexedDrawCutIndexEnable
= info
->primitiveRestartEnable
,
220 GEN8_3DSTATE_VF_pack(NULL
, pipeline
->state_vf
, &vf
);
222 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_VF_TOPOLOGY
,
223 .PrimitiveTopologyType
= topology
);
227 emit_rs_state(struct anv_pipeline
*pipeline
,
228 const VkPipelineRsStateCreateInfo
*info
,
229 const struct anv_pipeline_create_info
*extra
)
231 static const uint32_t vk_to_gen_cullmode
[] = {
232 [VK_CULL_MODE_NONE
] = CULLMODE_NONE
,
233 [VK_CULL_MODE_FRONT
] = CULLMODE_FRONT
,
234 [VK_CULL_MODE_BACK
] = CULLMODE_BACK
,
235 [VK_CULL_MODE_FRONT_AND_BACK
] = CULLMODE_BOTH
238 static const uint32_t vk_to_gen_fillmode
[] = {
239 [VK_FILL_MODE_POINTS
] = RASTER_POINT
,
240 [VK_FILL_MODE_WIREFRAME
] = RASTER_WIREFRAME
,
241 [VK_FILL_MODE_SOLID
] = RASTER_SOLID
244 static const uint32_t vk_to_gen_front_face
[] = {
245 [VK_FRONT_FACE_CCW
] = CounterClockwise
,
246 [VK_FRONT_FACE_CW
] = Clockwise
249 struct GEN8_3DSTATE_SF sf
= {
250 GEN8_3DSTATE_SF_header
,
251 .ViewportTransformEnable
= !(extra
&& extra
->disable_viewport
),
252 .TriangleStripListProvokingVertexSelect
= 0,
253 .LineStripListProvokingVertexSelect
= 0,
254 .TriangleFanProvokingVertexSelect
= 0,
255 .PointWidthSource
= pipeline
->writes_point_size
? Vertex
: State
,
259 /* FINISHME: bool32_t rasterizerDiscardEnable; */
261 GEN8_3DSTATE_SF_pack(NULL
, pipeline
->state_sf
, &sf
);
263 struct GEN8_3DSTATE_RASTER raster
= {
264 GEN8_3DSTATE_RASTER_header
,
265 .FrontWinding
= vk_to_gen_front_face
[info
->frontFace
],
266 .CullMode
= vk_to_gen_cullmode
[info
->cullMode
],
267 .FrontFaceFillMode
= vk_to_gen_fillmode
[info
->fillMode
],
268 .BackFaceFillMode
= vk_to_gen_fillmode
[info
->fillMode
],
269 .ScissorRectangleEnable
= !(extra
&& extra
->disable_scissor
),
270 .ViewportZClipTestEnable
= info
->depthClipEnable
273 GEN8_3DSTATE_RASTER_pack(NULL
, pipeline
->state_raster
, &raster
);
275 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_SBE
,
276 .ForceVertexURBEntryReadLength
= false,
277 .ForceVertexURBEntryReadOffset
= false,
278 .PointSpriteTextureCoordinateOrigin
= UPPERLEFT
,
279 .NumberofSFOutputAttributes
=
280 pipeline
->wm_prog_data
.num_varying_inputs
);
285 emit_cb_state(struct anv_pipeline
*pipeline
,
286 const VkPipelineCbStateCreateInfo
*info
)
288 struct anv_device
*device
= pipeline
->device
;
290 static const uint32_t vk_to_gen_logic_op
[] = {
291 [VK_LOGIC_OP_COPY
] = LOGICOP_COPY
,
292 [VK_LOGIC_OP_CLEAR
] = LOGICOP_CLEAR
,
293 [VK_LOGIC_OP_AND
] = LOGICOP_AND
,
294 [VK_LOGIC_OP_AND_REVERSE
] = LOGICOP_AND_REVERSE
,
295 [VK_LOGIC_OP_AND_INVERTED
] = LOGICOP_AND_INVERTED
,
296 [VK_LOGIC_OP_NOOP
] = LOGICOP_NOOP
,
297 [VK_LOGIC_OP_XOR
] = LOGICOP_XOR
,
298 [VK_LOGIC_OP_OR
] = LOGICOP_OR
,
299 [VK_LOGIC_OP_NOR
] = LOGICOP_NOR
,
300 [VK_LOGIC_OP_EQUIV
] = LOGICOP_EQUIV
,
301 [VK_LOGIC_OP_INVERT
] = LOGICOP_INVERT
,
302 [VK_LOGIC_OP_OR_REVERSE
] = LOGICOP_OR_REVERSE
,
303 [VK_LOGIC_OP_COPY_INVERTED
] = LOGICOP_COPY_INVERTED
,
304 [VK_LOGIC_OP_OR_INVERTED
] = LOGICOP_OR_INVERTED
,
305 [VK_LOGIC_OP_NAND
] = LOGICOP_NAND
,
306 [VK_LOGIC_OP_SET
] = LOGICOP_SET
,
309 static const uint32_t vk_to_gen_blend
[] = {
310 [VK_BLEND_ZERO
] = BLENDFACTOR_ZERO
,
311 [VK_BLEND_ONE
] = BLENDFACTOR_ONE
,
312 [VK_BLEND_SRC_COLOR
] = BLENDFACTOR_SRC_COLOR
,
313 [VK_BLEND_ONE_MINUS_SRC_COLOR
] = BLENDFACTOR_INV_SRC_COLOR
,
314 [VK_BLEND_DEST_COLOR
] = BLENDFACTOR_DST_COLOR
,
315 [VK_BLEND_ONE_MINUS_DEST_COLOR
] = BLENDFACTOR_INV_DST_COLOR
,
316 [VK_BLEND_SRC_ALPHA
] = BLENDFACTOR_SRC_ALPHA
,
317 [VK_BLEND_ONE_MINUS_SRC_ALPHA
] = BLENDFACTOR_INV_SRC_ALPHA
,
318 [VK_BLEND_DEST_ALPHA
] = BLENDFACTOR_DST_ALPHA
,
319 [VK_BLEND_ONE_MINUS_DEST_ALPHA
] = BLENDFACTOR_INV_DST_ALPHA
,
320 [VK_BLEND_CONSTANT_COLOR
] = BLENDFACTOR_CONST_COLOR
,
321 [VK_BLEND_ONE_MINUS_CONSTANT_COLOR
] = BLENDFACTOR_INV_CONST_COLOR
,
322 [VK_BLEND_CONSTANT_ALPHA
] = BLENDFACTOR_CONST_ALPHA
,
323 [VK_BLEND_ONE_MINUS_CONSTANT_ALPHA
] = BLENDFACTOR_INV_CONST_ALPHA
,
324 [VK_BLEND_SRC_ALPHA_SATURATE
] = BLENDFACTOR_SRC_ALPHA_SATURATE
,
325 [VK_BLEND_SRC1_COLOR
] = BLENDFACTOR_SRC1_COLOR
,
326 [VK_BLEND_ONE_MINUS_SRC1_COLOR
] = BLENDFACTOR_INV_SRC1_COLOR
,
327 [VK_BLEND_SRC1_ALPHA
] = BLENDFACTOR_SRC1_ALPHA
,
328 [VK_BLEND_ONE_MINUS_SRC1_ALPHA
] = BLENDFACTOR_INV_SRC1_ALPHA
,
331 static const uint32_t vk_to_gen_blend_op
[] = {
332 [VK_BLEND_OP_ADD
] = BLENDFUNCTION_ADD
,
333 [VK_BLEND_OP_SUBTRACT
] = BLENDFUNCTION_SUBTRACT
,
334 [VK_BLEND_OP_REVERSE_SUBTRACT
] = BLENDFUNCTION_REVERSE_SUBTRACT
,
335 [VK_BLEND_OP_MIN
] = BLENDFUNCTION_MIN
,
336 [VK_BLEND_OP_MAX
] = BLENDFUNCTION_MAX
,
339 uint32_t num_dwords
= 1 + info
->attachmentCount
* 2;
340 pipeline
->blend_state
=
341 anv_state_pool_alloc(&device
->dynamic_state_pool
, num_dwords
* 4, 64);
343 struct GEN8_BLEND_STATE blend_state
= {
344 .AlphaToCoverageEnable
= info
->alphaToCoverageEnable
,
347 uint32_t *state
= pipeline
->blend_state
.map
;
348 GEN8_BLEND_STATE_pack(NULL
, state
, &blend_state
);
350 for (uint32_t i
= 0; i
< info
->attachmentCount
; i
++) {
351 const VkPipelineCbAttachmentState
*a
= &info
->pAttachments
[i
];
353 struct GEN8_BLEND_STATE_ENTRY entry
= {
354 .LogicOpEnable
= info
->logicOpEnable
,
355 .LogicOpFunction
= vk_to_gen_logic_op
[info
->logicOp
],
356 .ColorBufferBlendEnable
= a
->blendEnable
,
357 .PreBlendSourceOnlyClampEnable
= false,
358 .PreBlendColorClampEnable
= false,
359 .PostBlendColorClampEnable
= false,
360 .SourceBlendFactor
= vk_to_gen_blend
[a
->srcBlendColor
],
361 .DestinationBlendFactor
= vk_to_gen_blend
[a
->destBlendColor
],
362 .ColorBlendFunction
= vk_to_gen_blend_op
[a
->blendOpColor
],
363 .SourceAlphaBlendFactor
= vk_to_gen_blend
[a
->srcBlendAlpha
],
364 .DestinationAlphaBlendFactor
= vk_to_gen_blend
[a
->destBlendAlpha
],
365 .AlphaBlendFunction
= vk_to_gen_blend_op
[a
->blendOpAlpha
],
366 .WriteDisableAlpha
= !(a
->channelWriteMask
& VK_CHANNEL_A_BIT
),
367 .WriteDisableRed
= !(a
->channelWriteMask
& VK_CHANNEL_R_BIT
),
368 .WriteDisableGreen
= !(a
->channelWriteMask
& VK_CHANNEL_G_BIT
),
369 .WriteDisableBlue
= !(a
->channelWriteMask
& VK_CHANNEL_B_BIT
),
372 GEN8_BLEND_STATE_ENTRY_pack(NULL
, state
+ i
* 2 + 1, &entry
);
375 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_BLEND_STATE_POINTERS
,
376 .BlendStatePointer
= pipeline
->blend_state
.offset
,
377 .BlendStatePointerValid
= true);
380 static const uint32_t vk_to_gen_compare_op
[] = {
381 [VK_COMPARE_OP_NEVER
] = COMPAREFUNCTION_NEVER
,
382 [VK_COMPARE_OP_LESS
] = COMPAREFUNCTION_LESS
,
383 [VK_COMPARE_OP_EQUAL
] = COMPAREFUNCTION_EQUAL
,
384 [VK_COMPARE_OP_LESS_EQUAL
] = COMPAREFUNCTION_LEQUAL
,
385 [VK_COMPARE_OP_GREATER
] = COMPAREFUNCTION_GREATER
,
386 [VK_COMPARE_OP_NOT_EQUAL
] = COMPAREFUNCTION_NOTEQUAL
,
387 [VK_COMPARE_OP_GREATER_EQUAL
] = COMPAREFUNCTION_GEQUAL
,
388 [VK_COMPARE_OP_ALWAYS
] = COMPAREFUNCTION_ALWAYS
,
391 static const uint32_t vk_to_gen_stencil_op
[] = {
392 [VK_STENCIL_OP_KEEP
] = 0,
393 [VK_STENCIL_OP_ZERO
] = 0,
394 [VK_STENCIL_OP_REPLACE
] = 0,
395 [VK_STENCIL_OP_INC_CLAMP
] = 0,
396 [VK_STENCIL_OP_DEC_CLAMP
] = 0,
397 [VK_STENCIL_OP_INVERT
] = 0,
398 [VK_STENCIL_OP_INC_WRAP
] = 0,
399 [VK_STENCIL_OP_DEC_WRAP
] = 0
403 emit_ds_state(struct anv_pipeline
*pipeline
,
404 const VkPipelineDsStateCreateInfo
*info
)
407 /* We're going to OR this together with the dynamic state. We need
408 * to make sure it's initialized to something useful.
410 memset(pipeline
->state_wm_depth_stencil
, 0,
411 sizeof(pipeline
->state_wm_depth_stencil
));
415 /* bool32_t depthBoundsEnable; // optional (depth_bounds_test) */
417 struct GEN8_3DSTATE_WM_DEPTH_STENCIL wm_depth_stencil
= {
418 .DepthTestEnable
= info
->depthTestEnable
,
419 .DepthBufferWriteEnable
= info
->depthWriteEnable
,
420 .DepthTestFunction
= vk_to_gen_compare_op
[info
->depthCompareOp
],
421 .DoubleSidedStencilEnable
= true,
423 .StencilTestEnable
= info
->stencilTestEnable
,
424 .StencilFailOp
= vk_to_gen_stencil_op
[info
->front
.stencilFailOp
],
425 .StencilPassDepthPassOp
= vk_to_gen_stencil_op
[info
->front
.stencilPassOp
],
426 .StencilPassDepthFailOp
= vk_to_gen_stencil_op
[info
->front
.stencilDepthFailOp
],
427 .StencilTestFunction
= vk_to_gen_compare_op
[info
->front
.stencilCompareOp
],
428 .BackfaceStencilFailOp
= vk_to_gen_stencil_op
[info
->back
.stencilFailOp
],
429 .BackfaceStencilPassDepthPassOp
= vk_to_gen_stencil_op
[info
->back
.stencilPassOp
],
430 .BackfaceStencilPassDepthFailOp
=vk_to_gen_stencil_op
[info
->back
.stencilDepthFailOp
],
431 .BackfaceStencilTestFunction
= vk_to_gen_compare_op
[info
->back
.stencilCompareOp
],
434 GEN8_3DSTATE_WM_DEPTH_STENCIL_pack(NULL
, pipeline
->state_wm_depth_stencil
, &wm_depth_stencil
);
438 anv_pipeline_destroy(struct anv_device
*device
,
439 struct anv_object
*object
,
440 VkObjectType obj_type
)
442 struct anv_pipeline
*pipeline
= (struct anv_pipeline
*) object
;
444 assert(obj_type
== VK_OBJECT_TYPE_PIPELINE
);
446 anv_compiler_free(pipeline
);
447 anv_reloc_list_finish(&pipeline
->batch
.relocs
, pipeline
->device
);
448 anv_state_stream_finish(&pipeline
->program_stream
);
449 anv_state_pool_free(&device
->dynamic_state_pool
, pipeline
->blend_state
);
450 anv_device_free(pipeline
->device
, pipeline
);
456 const VkGraphicsPipelineCreateInfo
* pCreateInfo
,
457 const struct anv_pipeline_create_info
* extra
,
458 VkPipeline
* pPipeline
)
460 struct anv_device
*device
= (struct anv_device
*) _device
;
461 struct anv_pipeline
*pipeline
;
463 uint32_t offset
, length
;
465 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO
);
467 pipeline
= anv_device_alloc(device
, sizeof(*pipeline
), 8,
468 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
469 if (pipeline
== NULL
)
470 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
472 pipeline
->base
.destructor
= anv_pipeline_destroy
;
473 pipeline
->device
= device
;
474 pipeline
->layout
= (struct anv_pipeline_layout
*) pCreateInfo
->layout
;
475 memset(pipeline
->shaders
, 0, sizeof(pipeline
->shaders
));
477 result
= anv_reloc_list_init(&pipeline
->batch
.relocs
, device
);
478 if (result
!= VK_SUCCESS
) {
479 anv_device_free(device
, pipeline
);
482 pipeline
->batch
.next
= pipeline
->batch
.start
= pipeline
->batch_data
;
483 pipeline
->batch
.end
= pipeline
->batch
.start
+ sizeof(pipeline
->batch_data
);
485 anv_state_stream_init(&pipeline
->program_stream
,
486 &device
->instruction_block_pool
);
488 for (uint32_t i
= 0; i
< pCreateInfo
->stageCount
; i
++) {
489 pipeline
->shaders
[pCreateInfo
->pStages
[i
].stage
] =
490 (struct anv_shader
*) pCreateInfo
->pStages
[i
].shader
;
493 if (pCreateInfo
->pTessState
)
494 anv_finishme("VK_STRUCTURE_TYPE_PIPELINE_TESS_STATE_CREATE_INFO");
495 if (pCreateInfo
->pVpState
)
496 anv_finishme("VK_STRUCTURE_TYPE_PIPELINE_VP_STATE_CREATE_INFO");
497 if (pCreateInfo
->pMsState
)
498 anv_finishme("VK_STRUCTURE_TYPE_PIPELINE_MS_STATE_CREATE_INFO");
500 pipeline
->use_repclear
= extra
&& extra
->use_repclear
;
502 anv_compiler_run(device
->compiler
, pipeline
);
504 /* FIXME: The compiler dead-codes FS inputs when we don't have a VS, so we
505 * hard code this to num_attributes - 2. This is because the attributes
506 * include VUE header and position, which aren't counted as varying
508 if (pipeline
->vs_simd8
== NO_KERNEL
) {
509 pipeline
->wm_prog_data
.num_varying_inputs
=
510 pCreateInfo
->pVertexInputState
->attributeCount
- 2;
513 assert(pCreateInfo
->pVertexInputState
);
514 emit_vertex_input(pipeline
, pCreateInfo
->pVertexInputState
);
515 assert(pCreateInfo
->pIaState
);
516 emit_ia_state(pipeline
, pCreateInfo
->pIaState
, extra
);
517 assert(pCreateInfo
->pRsState
);
518 emit_rs_state(pipeline
, pCreateInfo
->pRsState
, extra
);
519 emit_ds_state(pipeline
, pCreateInfo
->pDsState
);
520 emit_cb_state(pipeline
, pCreateInfo
->pCbState
);
522 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_VF_STATISTICS
,
523 .StatisticsEnable
= true);
524 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_HS
, .Enable
= false);
525 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_TE
, .TEEnable
= false);
526 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_DS
, .FunctionEnable
= false);
527 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_STREAMOUT
, .SOFunctionEnable
= false);
529 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_VS
,
530 .ConstantBufferOffset
= 0,
531 .ConstantBufferSize
= 4);
532 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_GS
,
533 .ConstantBufferOffset
= 4,
534 .ConstantBufferSize
= 4);
535 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_PS
,
536 .ConstantBufferOffset
= 8,
537 .ConstantBufferSize
= 4);
539 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_WM_CHROMAKEY
,
540 .ChromaKeyKillEnable
= false);
541 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_SBE_SWIZ
);
542 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_AA_LINE_PARAMETERS
);
544 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_CLIP
,
546 .ViewportXYClipTestEnable
= !(extra
&& extra
->disable_viewport
),
547 .MinimumPointWidth
= 0.125,
548 .MaximumPointWidth
= 255.875);
550 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_WM
,
551 .StatisticsEnable
= true,
552 .LineEndCapAntialiasingRegionWidth
= _05pixels
,
553 .LineAntialiasingRegionWidth
= _10pixels
,
554 .EarlyDepthStencilControl
= NORMAL
,
555 .ForceThreadDispatchEnable
= NORMAL
,
556 .PointRasterizationRule
= RASTRULE_UPPER_RIGHT
,
557 .BarycentricInterpolationMode
=
558 pipeline
->wm_prog_data
.barycentric_interp_modes
);
560 uint32_t samples
= 1;
561 uint32_t log2_samples
= __builtin_ffs(samples
) - 1;
562 bool enable_sampling
= samples
> 1 ? true : false;
564 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_MULTISAMPLE
,
565 .PixelPositionOffsetEnable
= enable_sampling
,
566 .PixelLocation
= CENTER
,
567 .NumberofMultisamples
= log2_samples
);
569 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_SAMPLE_MASK
,
570 .SampleMask
= 0xffff);
572 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_URB_VS
,
573 .VSURBStartingAddress
= pipeline
->urb
.vs_start
,
574 .VSURBEntryAllocationSize
= pipeline
->urb
.vs_size
- 1,
575 .VSNumberofURBEntries
= pipeline
->urb
.nr_vs_entries
);
577 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_URB_GS
,
578 .GSURBStartingAddress
= pipeline
->urb
.gs_start
,
579 .GSURBEntryAllocationSize
= pipeline
->urb
.gs_size
- 1,
580 .GSNumberofURBEntries
= pipeline
->urb
.nr_gs_entries
);
582 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_URB_HS
,
583 .HSURBStartingAddress
= pipeline
->urb
.vs_start
,
584 .HSURBEntryAllocationSize
= 0,
585 .HSNumberofURBEntries
= 0);
587 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_URB_DS
,
588 .DSURBStartingAddress
= pipeline
->urb
.vs_start
,
589 .DSURBEntryAllocationSize
= 0,
590 .DSNumberofURBEntries
= 0);
592 const struct brw_gs_prog_data
*gs_prog_data
= &pipeline
->gs_prog_data
;
594 length
= (gs_prog_data
->base
.vue_map
.num_slots
+ 1) / 2 - offset
;
596 if (pipeline
->gs_vec4
== NO_KERNEL
)
597 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_GS
, .Enable
= false);
599 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_GS
,
600 .SingleProgramFlow
= false,
601 .KernelStartPointer
= pipeline
->gs_vec4
,
602 .VectorMaskEnable
= Vmask
,
604 .BindingTableEntryCount
= 0,
605 .ExpectedVertexCount
= pipeline
->gs_vertex_count
,
607 .ScratchSpaceBasePointer
= pipeline
->scratch_start
[VK_SHADER_STAGE_GEOMETRY
],
608 .PerThreadScratchSpace
= ffs(gs_prog_data
->base
.base
.total_scratch
/ 2048),
610 .OutputVertexSize
= gs_prog_data
->output_vertex_size_hwords
* 2 - 1,
611 .OutputTopology
= gs_prog_data
->output_topology
,
612 .VertexURBEntryReadLength
= gs_prog_data
->base
.urb_read_length
,
613 .DispatchGRFStartRegisterForURBData
=
614 gs_prog_data
->base
.base
.dispatch_grf_start_reg
,
616 .MaximumNumberofThreads
= device
->info
.max_gs_threads
,
617 .ControlDataHeaderSize
= gs_prog_data
->control_data_header_size_hwords
,
618 //pipeline->gs_prog_data.dispatch_mode |
619 .StatisticsEnable
= true,
620 .IncludePrimitiveID
= gs_prog_data
->include_primitive_id
,
621 .ReorderMode
= TRAILING
,
624 .ControlDataFormat
= gs_prog_data
->control_data_format
,
626 /* FIXME: mesa sets this based on ctx->Transform.ClipPlanesEnabled:
627 * UserClipDistanceClipTestEnableBitmask_3DSTATE_GS(v)
628 * UserClipDistanceCullTestEnableBitmask(v)
631 .VertexURBEntryOutputReadOffset
= offset
,
632 .VertexURBEntryOutputLength
= length
);
634 const struct brw_vue_prog_data
*vue_prog_data
= &pipeline
->vs_prog_data
.base
;
635 /* Skip the VUE header and position slots */
637 length
= (vue_prog_data
->vue_map
.num_slots
+ 1) / 2 - offset
;
639 if (pipeline
->vs_simd8
== NO_KERNEL
|| (extra
&& extra
->disable_vs
))
640 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_VS
,
641 .FunctionEnable
= false,
642 .VertexURBEntryOutputReadOffset
= 1,
643 /* Even if VS is disabled, SBE still gets the amount of
644 * vertex data to read from this field. We use attribute
645 * count - 1, as we don't count the VUE header here. */
646 .VertexURBEntryOutputLength
=
647 DIV_ROUND_UP(pCreateInfo
->pVertexInputState
->attributeCount
- 1, 2));
649 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_VS
,
650 .KernelStartPointer
= pipeline
->vs_simd8
,
651 .SingleVertexDispatch
= Multiple
,
652 .VectorMaskEnable
= Dmask
,
654 .BindingTableEntryCount
=
655 vue_prog_data
->base
.binding_table
.size_bytes
/ 4,
656 .ThreadDispatchPriority
= Normal
,
657 .FloatingPointMode
= IEEE754
,
658 .IllegalOpcodeExceptionEnable
= false,
659 .AccessesUAV
= false,
660 .SoftwareExceptionEnable
= false,
662 .ScratchSpaceBasePointer
= pipeline
->scratch_start
[VK_SHADER_STAGE_VERTEX
],
663 .PerThreadScratchSpace
= ffs(vue_prog_data
->base
.total_scratch
/ 2048),
665 .DispatchGRFStartRegisterForURBData
=
666 vue_prog_data
->base
.dispatch_grf_start_reg
,
667 .VertexURBEntryReadLength
= vue_prog_data
->urb_read_length
,
668 .VertexURBEntryReadOffset
= 0,
670 .MaximumNumberofThreads
= device
->info
.max_vs_threads
- 1,
671 .StatisticsEnable
= false,
672 .SIMD8DispatchEnable
= true,
673 .VertexCacheDisable
= false,
674 .FunctionEnable
= true,
676 .VertexURBEntryOutputReadOffset
= offset
,
677 .VertexURBEntryOutputLength
= length
,
678 .UserClipDistanceClipTestEnableBitmask
= 0,
679 .UserClipDistanceCullTestEnableBitmask
= 0);
681 const struct brw_wm_prog_data
*wm_prog_data
= &pipeline
->wm_prog_data
;
682 uint32_t ksp0
, ksp2
, grf_start0
, grf_start2
;
686 if (pipeline
->ps_simd8
!= NO_KERNEL
) {
687 ksp0
= pipeline
->ps_simd8
;
688 grf_start0
= wm_prog_data
->base
.dispatch_grf_start_reg
;
689 if (pipeline
->ps_simd16
!= NO_KERNEL
) {
690 ksp2
= pipeline
->ps_simd16
;
691 grf_start2
= wm_prog_data
->dispatch_grf_start_reg_16
;
693 } else if (pipeline
->ps_simd16
!= NO_KERNEL
) {
694 ksp0
= pipeline
->ps_simd16
;
695 grf_start0
= wm_prog_data
->dispatch_grf_start_reg_16
;
697 unreachable("no ps shader");
700 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_PS
,
701 .KernelStartPointer0
= ksp0
,
703 .SingleProgramFlow
= false,
704 .VectorMaskEnable
= true,
707 .ScratchSpaceBasePointer
= pipeline
->scratch_start
[VK_SHADER_STAGE_FRAGMENT
],
708 .PerThreadScratchSpace
= ffs(wm_prog_data
->base
.total_scratch
/ 2048),
710 .MaximumNumberofThreadsPerPSD
= 64 - 2,
711 .PositionXYOffsetSelect
= wm_prog_data
->uses_pos_offset
?
712 POSOFFSET_SAMPLE
: POSOFFSET_NONE
,
713 .PushConstantEnable
= wm_prog_data
->base
.nr_params
> 0,
714 ._8PixelDispatchEnable
= pipeline
->ps_simd8
!= NO_KERNEL
,
715 ._16PixelDispatchEnable
= pipeline
->ps_simd16
!= NO_KERNEL
,
716 ._32PixelDispatchEnable
= false,
718 .DispatchGRFStartRegisterForConstantSetupData0
= grf_start0
,
719 .DispatchGRFStartRegisterForConstantSetupData1
= 0,
720 .DispatchGRFStartRegisterForConstantSetupData2
= grf_start2
,
722 .KernelStartPointer1
= 0,
723 .KernelStartPointer2
= ksp2
);
725 bool per_sample_ps
= false;
726 anv_batch_emit(&pipeline
->batch
, GEN8_3DSTATE_PS_EXTRA
,
727 .PixelShaderValid
= true,
728 .PixelShaderKillsPixel
= wm_prog_data
->uses_kill
,
729 .PixelShaderComputedDepthMode
= wm_prog_data
->computed_depth_mode
,
730 .AttributeEnable
= wm_prog_data
->num_varying_inputs
> 0,
731 .oMaskPresenttoRenderTarget
= wm_prog_data
->uses_omask
,
732 .PixelShaderIsPerSample
= per_sample_ps
);
734 *pPipeline
= (VkPipeline
) pipeline
;
739 VkResult
anv_CreateGraphicsPipelines(
741 VkPipelineCache pipelineCache
,
743 const VkGraphicsPipelineCreateInfo
* pCreateInfos
,
744 VkPipeline
* pPipelines
)
746 ANV_FROM_HANDLE(anv_device
, device
, _device
);
747 VkResult result
= VK_SUCCESS
;
750 for (; i
< count
; i
++) {
751 result
= anv_pipeline_create(_device
, &pCreateInfos
[i
],
752 NULL
, &pPipelines
[i
]);
753 if (result
!= VK_SUCCESS
) {
754 for (unsigned j
= 0; j
< i
; j
++) {
755 anv_pipeline_destroy(device
, (struct anv_object
*)pPipelines
[j
],
756 VK_OBJECT_TYPE_PIPELINE
);
766 static VkResult
anv_compute_pipeline_create(
768 const VkComputePipelineCreateInfo
* pCreateInfo
,
769 VkPipeline
* pPipeline
)
771 struct anv_device
*device
= (struct anv_device
*) _device
;
772 struct anv_pipeline
*pipeline
;
775 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO
);
777 pipeline
= anv_device_alloc(device
, sizeof(*pipeline
), 8,
778 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
779 if (pipeline
== NULL
)
780 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
782 pipeline
->base
.destructor
= anv_pipeline_destroy
;
783 pipeline
->device
= device
;
784 pipeline
->layout
= (struct anv_pipeline_layout
*) pCreateInfo
->layout
;
786 result
= anv_reloc_list_init(&pipeline
->batch
.relocs
, device
);
787 if (result
!= VK_SUCCESS
) {
788 anv_device_free(device
, pipeline
);
791 pipeline
->batch
.next
= pipeline
->batch
.start
= pipeline
->batch_data
;
792 pipeline
->batch
.end
= pipeline
->batch
.start
+ sizeof(pipeline
->batch_data
);
794 anv_state_stream_init(&pipeline
->program_stream
,
795 &device
->instruction_block_pool
);
797 memset(pipeline
->shaders
, 0, sizeof(pipeline
->shaders
));
799 pipeline
->shaders
[VK_SHADER_STAGE_COMPUTE
] =
800 (struct anv_shader
*) pCreateInfo
->cs
.shader
;
802 pipeline
->use_repclear
= false;
804 anv_compiler_run(device
->compiler
, pipeline
);
806 const struct brw_cs_prog_data
*cs_prog_data
= &pipeline
->cs_prog_data
;
808 anv_batch_emit(&pipeline
->batch
, GEN8_MEDIA_VFE_STATE
,
809 .ScratchSpaceBasePointer
= pipeline
->scratch_start
[VK_SHADER_STAGE_FRAGMENT
],
810 .PerThreadScratchSpace
= ffs(cs_prog_data
->base
.total_scratch
/ 2048),
811 .ScratchSpaceBasePointerHigh
= 0,
814 .MaximumNumberofThreads
= device
->info
.max_cs_threads
- 1,
815 .NumberofURBEntries
= 2,
816 .ResetGatewayTimer
= true,
817 .BypassGatewayControl
= true,
818 .URBEntryAllocationSize
= 2,
819 .CURBEAllocationSize
= 0);
821 struct brw_cs_prog_data
*prog_data
= &pipeline
->cs_prog_data
;
822 uint32_t group_size
= prog_data
->local_size
[0] *
823 prog_data
->local_size
[1] * prog_data
->local_size
[2];
824 pipeline
->cs_thread_width_max
= DIV_ROUND_UP(group_size
, prog_data
->simd_size
);
825 uint32_t remainder
= group_size
& (prog_data
->simd_size
- 1);
828 pipeline
->cs_right_mask
= ~0u >> (32 - remainder
);
830 pipeline
->cs_right_mask
= ~0u >> (32 - prog_data
->simd_size
);
833 *pPipeline
= (VkPipeline
) pipeline
;
838 VkResult
anv_CreateComputePipelines(
840 VkPipelineCache pipelineCache
,
842 const VkComputePipelineCreateInfo
* pCreateInfos
,
843 VkPipeline
* pPipelines
)
845 ANV_FROM_HANDLE(anv_device
, device
, _device
);
846 VkResult result
= VK_SUCCESS
;
849 for (; i
< count
; i
++) {
850 result
= anv_compute_pipeline_create(_device
, &pCreateInfos
[i
],
852 if (result
!= VK_SUCCESS
) {
853 for (unsigned j
= 0; j
< i
; j
++) {
854 anv_pipeline_destroy(device
, (struct anv_object
*)pPipelines
[j
],
855 VK_OBJECT_TYPE_PIPELINE
);
865 // Pipeline layout functions
867 VkResult
anv_CreatePipelineLayout(
869 const VkPipelineLayoutCreateInfo
* pCreateInfo
,
870 VkPipelineLayout
* pPipelineLayout
)
872 struct anv_device
*device
= (struct anv_device
*) _device
;
873 struct anv_pipeline_layout
*layout
;
875 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO
);
877 layout
= anv_device_alloc(device
, sizeof(*layout
), 8,
878 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
880 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
882 layout
->num_sets
= pCreateInfo
->descriptorSetCount
;
884 uint32_t surface_start
[VK_SHADER_STAGE_NUM
] = { 0, };
885 uint32_t sampler_start
[VK_SHADER_STAGE_NUM
] = { 0, };
887 for (uint32_t s
= 0; s
< VK_SHADER_STAGE_NUM
; s
++) {
888 layout
->stage
[s
].surface_count
= 0;
889 layout
->stage
[s
].sampler_count
= 0;
892 for (uint32_t i
= 0; i
< pCreateInfo
->descriptorSetCount
; i
++) {
893 struct anv_descriptor_set_layout
*set_layout
=
894 (struct anv_descriptor_set_layout
*) pCreateInfo
->pSetLayouts
[i
];
896 layout
->set
[i
].layout
= set_layout
;
897 for (uint32_t s
= 0; s
< VK_SHADER_STAGE_NUM
; s
++) {
898 layout
->set
[i
].surface_start
[s
] = surface_start
[s
];
899 surface_start
[s
] += set_layout
->stage
[s
].surface_count
;
900 layout
->set
[i
].sampler_start
[s
] = sampler_start
[s
];
901 sampler_start
[s
] += set_layout
->stage
[s
].sampler_count
;
903 layout
->stage
[s
].surface_count
+= set_layout
->stage
[s
].surface_count
;
904 layout
->stage
[s
].sampler_count
+= set_layout
->stage
[s
].sampler_count
;
908 *pPipelineLayout
= (VkPipelineLayout
) layout
;