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
30 #include "anv_private.h"
32 #include "gen8_pack.h"
33 #include "gen9_pack.h"
36 emit_vertex_input(struct anv_pipeline
*pipeline
,
37 const VkPipelineVertexInputStateCreateInfo
*info
)
39 const uint32_t num_dwords
= 1 + info
->attributeCount
* 2;
42 static_assert(ANV_GEN
>= 8, "should be compiling this for gen < 8");
44 if (info
->attributeCount
> 0) {
45 p
= anv_batch_emitn(&pipeline
->batch
, num_dwords
,
46 GENX(3DSTATE_VERTEX_ELEMENTS
));
49 for (uint32_t i
= 0; i
< info
->attributeCount
; i
++) {
50 const VkVertexInputAttributeDescription
*desc
=
51 &info
->pVertexAttributeDescriptions
[i
];
52 const struct anv_format
*format
= anv_format_for_vk_format(desc
->format
);
54 struct GENX(VERTEX_ELEMENT_STATE
) element
= {
55 .VertexBufferIndex
= desc
->binding
,
57 .SourceElementFormat
= format
->surface_format
,
58 .EdgeFlagEnable
= false,
59 .SourceElementOffset
= desc
->offsetInBytes
,
60 .Component0Control
= VFCOMP_STORE_SRC
,
61 .Component1Control
= format
->num_channels
>= 2 ? VFCOMP_STORE_SRC
: VFCOMP_STORE_0
,
62 .Component2Control
= format
->num_channels
>= 3 ? VFCOMP_STORE_SRC
: VFCOMP_STORE_0
,
63 .Component3Control
= format
->num_channels
>= 4 ? VFCOMP_STORE_SRC
: VFCOMP_STORE_1_FP
65 GENX(VERTEX_ELEMENT_STATE_pack
)(NULL
, &p
[1 + i
* 2], &element
);
67 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_VF_INSTANCING
),
68 .InstancingEnable
= pipeline
->instancing_enable
[desc
->binding
],
69 .VertexElementIndex
= i
,
70 /* Vulkan so far doesn't have an instance divisor, so
71 * this is always 1 (ignored if not instancing). */
72 .InstanceDataStepRate
= 1);
75 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_VF_SGVS
),
76 .VertexIDEnable
= pipeline
->vs_prog_data
.uses_vertexid
,
77 .VertexIDComponentNumber
= 2,
78 .VertexIDElementOffset
= info
->bindingCount
,
79 .InstanceIDEnable
= pipeline
->vs_prog_data
.uses_instanceid
,
80 .InstanceIDComponentNumber
= 3,
81 .InstanceIDElementOffset
= info
->bindingCount
);
85 emit_ia_state(struct anv_pipeline
*pipeline
,
86 const VkPipelineInputAssemblyStateCreateInfo
*info
,
87 const struct anv_graphics_pipeline_create_info
*extra
)
89 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_VF_TOPOLOGY
),
90 .PrimitiveTopologyType
= pipeline
->topology
);
94 emit_rs_state(struct anv_pipeline
*pipeline
,
95 const VkPipelineRasterStateCreateInfo
*info
,
96 const struct anv_graphics_pipeline_create_info
*extra
)
98 static const uint32_t vk_to_gen_cullmode
[] = {
99 [VK_CULL_MODE_NONE
] = CULLMODE_NONE
,
100 [VK_CULL_MODE_FRONT_BIT
] = CULLMODE_FRONT
,
101 [VK_CULL_MODE_BACK_BIT
] = CULLMODE_BACK
,
102 [VK_CULL_MODE_FRONT_AND_BACK
] = CULLMODE_BOTH
105 static const uint32_t vk_to_gen_fillmode
[] = {
106 [VK_POLYGON_MODE_FILL
] = RASTER_SOLID
,
107 [VK_POLYGON_MODE_LINE
] = RASTER_WIREFRAME
,
108 [VK_POLYGON_MODE_POINT
] = RASTER_POINT
,
111 static const uint32_t vk_to_gen_front_face
[] = {
112 [VK_FRONT_FACE_COUNTER_CLOCKWISE
] = 1,
113 [VK_FRONT_FACE_CLOCKWISE
] = 0
116 struct GENX(3DSTATE_SF
) sf
= {
117 GENX(3DSTATE_SF_header
),
118 .ViewportTransformEnable
= !(extra
&& extra
->disable_viewport
),
119 .TriangleStripListProvokingVertexSelect
= 0,
120 .LineStripListProvokingVertexSelect
= 0,
121 .TriangleFanProvokingVertexSelect
= 0,
122 .PointWidthSource
= pipeline
->writes_point_size
? Vertex
: State
,
126 /* FINISHME: VkBool32 rasterizerDiscardEnable; */
128 GENX(3DSTATE_SF_pack
)(NULL
, pipeline
->gen8
.sf
, &sf
);
130 struct GENX(3DSTATE_RASTER
) raster
= {
131 GENX(3DSTATE_RASTER_header
),
132 .FrontWinding
= vk_to_gen_front_face
[info
->frontFace
],
133 .CullMode
= vk_to_gen_cullmode
[info
->cullMode
],
134 .FrontFaceFillMode
= vk_to_gen_fillmode
[info
->polygonMode
],
135 .BackFaceFillMode
= vk_to_gen_fillmode
[info
->polygonMode
],
136 .ScissorRectangleEnable
= !(extra
&& extra
->disable_scissor
),
138 .ViewportZClipTestEnable
= info
->depthClipEnable
140 /* GEN9+ splits ViewportZClipTestEnable into near and far enable bits */
141 .ViewportZFarClipTestEnable
= info
->depthClipEnable
,
142 .ViewportZNearClipTestEnable
= info
->depthClipEnable
,
146 GENX(3DSTATE_RASTER_pack
)(NULL
, pipeline
->gen8
.raster
, &raster
);
150 emit_cb_state(struct anv_pipeline
*pipeline
,
151 const VkPipelineColorBlendStateCreateInfo
*info
)
153 struct anv_device
*device
= pipeline
->device
;
155 static const uint32_t vk_to_gen_logic_op
[] = {
156 [VK_LOGIC_OP_COPY
] = LOGICOP_COPY
,
157 [VK_LOGIC_OP_CLEAR
] = LOGICOP_CLEAR
,
158 [VK_LOGIC_OP_AND
] = LOGICOP_AND
,
159 [VK_LOGIC_OP_AND_REVERSE
] = LOGICOP_AND_REVERSE
,
160 [VK_LOGIC_OP_AND_INVERTED
] = LOGICOP_AND_INVERTED
,
161 [VK_LOGIC_OP_NO_OP
] = LOGICOP_NOOP
,
162 [VK_LOGIC_OP_XOR
] = LOGICOP_XOR
,
163 [VK_LOGIC_OP_OR
] = LOGICOP_OR
,
164 [VK_LOGIC_OP_NOR
] = LOGICOP_NOR
,
165 [VK_LOGIC_OP_EQUIVALENT
] = LOGICOP_EQUIV
,
166 [VK_LOGIC_OP_INVERT
] = LOGICOP_INVERT
,
167 [VK_LOGIC_OP_OR_REVERSE
] = LOGICOP_OR_REVERSE
,
168 [VK_LOGIC_OP_COPY_INVERTED
] = LOGICOP_COPY_INVERTED
,
169 [VK_LOGIC_OP_OR_INVERTED
] = LOGICOP_OR_INVERTED
,
170 [VK_LOGIC_OP_NAND
] = LOGICOP_NAND
,
171 [VK_LOGIC_OP_SET
] = LOGICOP_SET
,
174 static const uint32_t vk_to_gen_blend
[] = {
175 [VK_BLEND_FACTOR_ZERO
] = BLENDFACTOR_ZERO
,
176 [VK_BLEND_FACTOR_ONE
] = BLENDFACTOR_ONE
,
177 [VK_BLEND_FACTOR_SRC_COLOR
] = BLENDFACTOR_SRC_COLOR
,
178 [VK_BLEND_FACTOR_ONE_MINUS_SRC_COLOR
] = BLENDFACTOR_INV_SRC_COLOR
,
179 [VK_BLEND_FACTOR_DST_COLOR
] = BLENDFACTOR_DST_COLOR
,
180 [VK_BLEND_FACTOR_ONE_MINUS_DST_COLOR
] = BLENDFACTOR_INV_DST_COLOR
,
181 [VK_BLEND_FACTOR_SRC_ALPHA
] = BLENDFACTOR_SRC_ALPHA
,
182 [VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA
] = BLENDFACTOR_INV_SRC_ALPHA
,
183 [VK_BLEND_FACTOR_DST_ALPHA
] = BLENDFACTOR_DST_ALPHA
,
184 [VK_BLEND_FACTOR_ONE_MINUS_DST_ALPHA
] = BLENDFACTOR_INV_DST_ALPHA
,
185 [VK_BLEND_FACTOR_CONSTANT_COLOR
] = BLENDFACTOR_CONST_COLOR
,
186 [VK_BLEND_FACTOR_ONE_MINUS_CONSTANT_COLOR
]= BLENDFACTOR_INV_CONST_COLOR
,
187 [VK_BLEND_FACTOR_CONSTANT_ALPHA
] = BLENDFACTOR_CONST_ALPHA
,
188 [VK_BLEND_FACTOR_ONE_MINUS_CONSTANT_ALPHA
]= BLENDFACTOR_INV_CONST_ALPHA
,
189 [VK_BLEND_FACTOR_SRC_ALPHA_SATURATE
] = BLENDFACTOR_SRC_ALPHA_SATURATE
,
190 [VK_BLEND_FACTOR_SRC1_COLOR
] = BLENDFACTOR_SRC1_COLOR
,
191 [VK_BLEND_FACTOR_ONE_MINUS_SRC1_COLOR
] = BLENDFACTOR_INV_SRC1_COLOR
,
192 [VK_BLEND_FACTOR_SRC1_ALPHA
] = BLENDFACTOR_SRC1_ALPHA
,
193 [VK_BLEND_FACTOR_ONE_MINUS_SRC1_ALPHA
] = BLENDFACTOR_INV_SRC1_ALPHA
,
196 static const uint32_t vk_to_gen_blend_op
[] = {
197 [VK_BLEND_OP_ADD
] = BLENDFUNCTION_ADD
,
198 [VK_BLEND_OP_SUBTRACT
] = BLENDFUNCTION_SUBTRACT
,
199 [VK_BLEND_OP_REVERSE_SUBTRACT
] = BLENDFUNCTION_REVERSE_SUBTRACT
,
200 [VK_BLEND_OP_MIN
] = BLENDFUNCTION_MIN
,
201 [VK_BLEND_OP_MAX
] = BLENDFUNCTION_MAX
,
204 uint32_t num_dwords
= GENX(BLEND_STATE_length
);
205 pipeline
->blend_state
=
206 anv_state_pool_alloc(&device
->dynamic_state_pool
, num_dwords
* 4, 64);
208 struct GENX(BLEND_STATE
) blend_state
= {
209 .AlphaToCoverageEnable
= info
->alphaToCoverageEnable
,
210 .AlphaToOneEnable
= info
->alphaToOneEnable
,
213 for (uint32_t i
= 0; i
< info
->attachmentCount
; i
++) {
214 const VkPipelineColorBlendAttachmentState
*a
= &info
->pAttachments
[i
];
216 if (a
->srcColorBlendFactor
!= a
->srcAlphaBlendFactor
||
217 a
->dstColorBlendFactor
!= a
->dstAlphaBlendFactor
||
218 a
->colorBlendOp
!= a
->alphaBlendOp
) {
219 blend_state
.IndependentAlphaBlendEnable
= true;
222 blend_state
.Entry
[i
] = (struct GENX(BLEND_STATE_ENTRY
)) {
223 .LogicOpEnable
= info
->logicOpEnable
,
224 .LogicOpFunction
= vk_to_gen_logic_op
[info
->logicOp
],
225 .ColorBufferBlendEnable
= a
->blendEnable
,
226 .PreBlendSourceOnlyClampEnable
= false,
227 .ColorClampRange
= COLORCLAMP_RTFORMAT
,
228 .PreBlendColorClampEnable
= true,
229 .PostBlendColorClampEnable
= true,
230 .SourceBlendFactor
= vk_to_gen_blend
[a
->srcColorBlendFactor
],
231 .DestinationBlendFactor
= vk_to_gen_blend
[a
->dstColorBlendFactor
],
232 .ColorBlendFunction
= vk_to_gen_blend_op
[a
->colorBlendOp
],
233 .SourceAlphaBlendFactor
= vk_to_gen_blend
[a
->srcAlphaBlendFactor
],
234 .DestinationAlphaBlendFactor
= vk_to_gen_blend
[a
->dstAlphaBlendFactor
],
235 .AlphaBlendFunction
= vk_to_gen_blend_op
[a
->alphaBlendOp
],
236 .WriteDisableAlpha
= !(a
->channelWriteMask
& VK_CHANNEL_A_BIT
),
237 .WriteDisableRed
= !(a
->channelWriteMask
& VK_CHANNEL_R_BIT
),
238 .WriteDisableGreen
= !(a
->channelWriteMask
& VK_CHANNEL_G_BIT
),
239 .WriteDisableBlue
= !(a
->channelWriteMask
& VK_CHANNEL_B_BIT
),
242 /* Our hardware applies the blend factor prior to the blend function
243 * regardless of what function is used. Technically, this means the
244 * hardware can do MORE than GL or Vulkan specify. However, it also
245 * means that, for MIN and MAX, we have to stomp the blend factor to
246 * ONE to make it a no-op.
248 if (a
->colorBlendOp
== VK_BLEND_OP_MIN
||
249 a
->colorBlendOp
== VK_BLEND_OP_MAX
) {
250 blend_state
.Entry
[i
].SourceBlendFactor
= BLENDFACTOR_ONE
;
251 blend_state
.Entry
[i
].DestinationBlendFactor
= BLENDFACTOR_ONE
;
253 if (a
->alphaBlendOp
== VK_BLEND_OP_MIN
||
254 a
->alphaBlendOp
== VK_BLEND_OP_MAX
) {
255 blend_state
.Entry
[i
].SourceAlphaBlendFactor
= BLENDFACTOR_ONE
;
256 blend_state
.Entry
[i
].DestinationAlphaBlendFactor
= BLENDFACTOR_ONE
;
260 GENX(BLEND_STATE_pack
)(NULL
, pipeline
->blend_state
.map
, &blend_state
);
262 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_BLEND_STATE_POINTERS
),
263 .BlendStatePointer
= pipeline
->blend_state
.offset
,
264 .BlendStatePointerValid
= true);
267 static const uint32_t vk_to_gen_compare_op
[] = {
268 [VK_COMPARE_OP_NEVER
] = PREFILTEROPNEVER
,
269 [VK_COMPARE_OP_LESS
] = PREFILTEROPLESS
,
270 [VK_COMPARE_OP_EQUAL
] = PREFILTEROPEQUAL
,
271 [VK_COMPARE_OP_LESS_OR_EQUAL
] = PREFILTEROPLEQUAL
,
272 [VK_COMPARE_OP_GREATER
] = PREFILTEROPGREATER
,
273 [VK_COMPARE_OP_NOT_EQUAL
] = PREFILTEROPNOTEQUAL
,
274 [VK_COMPARE_OP_GREATER_OR_EQUAL
] = PREFILTEROPGEQUAL
,
275 [VK_COMPARE_OP_ALWAYS
] = PREFILTEROPALWAYS
,
278 static const uint32_t vk_to_gen_stencil_op
[] = {
279 [VK_STENCIL_OP_KEEP
] = STENCILOP_KEEP
,
280 [VK_STENCIL_OP_ZERO
] = STENCILOP_ZERO
,
281 [VK_STENCIL_OP_REPLACE
] = STENCILOP_REPLACE
,
282 [VK_STENCIL_OP_INCREMENT_AND_CLAMP
] = STENCILOP_INCRSAT
,
283 [VK_STENCIL_OP_DECREMENT_AND_CLAMP
] = STENCILOP_DECRSAT
,
284 [VK_STENCIL_OP_INVERT
] = STENCILOP_INVERT
,
285 [VK_STENCIL_OP_INCREMENT_AND_WRAP
] = STENCILOP_INCR
,
286 [VK_STENCIL_OP_DECREMENT_AND_WRAP
] = STENCILOP_DECR
,
290 emit_ds_state(struct anv_pipeline
*pipeline
,
291 const VkPipelineDepthStencilStateCreateInfo
*info
)
293 uint32_t *dw
= ANV_GEN
== 8 ?
294 pipeline
->gen8
.wm_depth_stencil
: pipeline
->gen9
.wm_depth_stencil
;
297 /* We're going to OR this together with the dynamic state. We need
298 * to make sure it's initialized to something useful.
300 memset(pipeline
->gen8
.wm_depth_stencil
, 0,
301 sizeof(pipeline
->gen8
.wm_depth_stencil
));
302 memset(pipeline
->gen9
.wm_depth_stencil
, 0,
303 sizeof(pipeline
->gen9
.wm_depth_stencil
));
307 /* VkBool32 depthBoundsTestEnable; // optional (depth_bounds_test) */
309 struct GENX(3DSTATE_WM_DEPTH_STENCIL
) wm_depth_stencil
= {
310 .DepthTestEnable
= info
->depthTestEnable
,
311 .DepthBufferWriteEnable
= info
->depthWriteEnable
,
312 .DepthTestFunction
= vk_to_gen_compare_op
[info
->depthCompareOp
],
313 .DoubleSidedStencilEnable
= true,
315 .StencilTestEnable
= info
->stencilTestEnable
,
316 .StencilFailOp
= vk_to_gen_stencil_op
[info
->front
.stencilFailOp
],
317 .StencilPassDepthPassOp
= vk_to_gen_stencil_op
[info
->front
.stencilPassOp
],
318 .StencilPassDepthFailOp
= vk_to_gen_stencil_op
[info
->front
.stencilDepthFailOp
],
319 .StencilTestFunction
= vk_to_gen_compare_op
[info
->front
.stencilCompareOp
],
320 .BackfaceStencilFailOp
= vk_to_gen_stencil_op
[info
->back
.stencilFailOp
],
321 .BackfaceStencilPassDepthPassOp
= vk_to_gen_stencil_op
[info
->back
.stencilPassOp
],
322 .BackfaceStencilPassDepthFailOp
=vk_to_gen_stencil_op
[info
->back
.stencilDepthFailOp
],
323 .BackfaceStencilTestFunction
= vk_to_gen_compare_op
[info
->back
.stencilCompareOp
],
326 GENX(3DSTATE_WM_DEPTH_STENCIL_pack
)(NULL
, dw
, &wm_depth_stencil
);
330 genX(graphics_pipeline_create
)(
332 const VkGraphicsPipelineCreateInfo
* pCreateInfo
,
333 const struct anv_graphics_pipeline_create_info
*extra
,
334 VkPipeline
* pPipeline
)
336 ANV_FROM_HANDLE(anv_device
, device
, _device
);
337 struct anv_pipeline
*pipeline
;
339 uint32_t offset
, length
;
341 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO
);
343 pipeline
= anv_device_alloc(device
, sizeof(*pipeline
), 8,
344 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
345 if (pipeline
== NULL
)
346 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
348 result
= anv_pipeline_init(pipeline
, device
, pCreateInfo
, extra
);
349 if (result
!= VK_SUCCESS
)
352 /* FIXME: The compiler dead-codes FS inputs when we don't have a VS, so we
353 * hard code this to num_attributes - 2. This is because the attributes
354 * include VUE header and position, which aren't counted as varying
356 if (pipeline
->vs_simd8
== NO_KERNEL
) {
357 pipeline
->wm_prog_data
.num_varying_inputs
=
358 pCreateInfo
->pVertexInputState
->attributeCount
- 2;
361 assert(pCreateInfo
->pVertexInputState
);
362 emit_vertex_input(pipeline
, pCreateInfo
->pVertexInputState
);
363 assert(pCreateInfo
->pInputAssemblyState
);
364 emit_ia_state(pipeline
, pCreateInfo
->pInputAssemblyState
, extra
);
365 assert(pCreateInfo
->pRasterState
);
366 emit_rs_state(pipeline
, pCreateInfo
->pRasterState
, extra
);
367 emit_ds_state(pipeline
, pCreateInfo
->pDepthStencilState
);
368 emit_cb_state(pipeline
, pCreateInfo
->pColorBlendState
);
370 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_VF_STATISTICS
),
371 .StatisticsEnable
= true);
372 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_HS
), .Enable
= false);
373 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_TE
), .TEEnable
= false);
374 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_DS
), .FunctionEnable
= false);
375 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_STREAMOUT
), .SOFunctionEnable
= false);
377 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_PUSH_CONSTANT_ALLOC_VS
),
378 .ConstantBufferOffset
= 0,
379 .ConstantBufferSize
= 4);
380 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_PUSH_CONSTANT_ALLOC_GS
),
381 .ConstantBufferOffset
= 4,
382 .ConstantBufferSize
= 4);
383 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_PUSH_CONSTANT_ALLOC_PS
),
384 .ConstantBufferOffset
= 8,
385 .ConstantBufferSize
= 4);
387 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_WM_CHROMAKEY
),
388 .ChromaKeyKillEnable
= false);
389 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_AA_LINE_PARAMETERS
));
391 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_CLIP
),
393 .ViewportXYClipTestEnable
= !(extra
&& extra
->disable_viewport
),
394 .MinimumPointWidth
= 0.125,
395 .MaximumPointWidth
= 255.875);
397 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_WM
),
398 .StatisticsEnable
= true,
399 .LineEndCapAntialiasingRegionWidth
= _05pixels
,
400 .LineAntialiasingRegionWidth
= _10pixels
,
401 .EarlyDepthStencilControl
= NORMAL
,
402 .ForceThreadDispatchEnable
= NORMAL
,
403 .PointRasterizationRule
= RASTRULE_UPPER_RIGHT
,
404 .BarycentricInterpolationMode
=
405 pipeline
->wm_prog_data
.barycentric_interp_modes
);
407 uint32_t samples
= 1;
408 uint32_t log2_samples
= __builtin_ffs(samples
) - 1;
409 bool enable_sampling
= samples
> 1 ? true : false;
411 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_MULTISAMPLE
),
412 .PixelPositionOffsetEnable
= enable_sampling
,
413 .PixelLocation
= CENTER
,
414 .NumberofMultisamples
= log2_samples
);
416 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_SAMPLE_MASK
),
417 .SampleMask
= 0xffff);
419 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_URB_VS
),
420 .VSURBStartingAddress
= pipeline
->urb
.vs_start
,
421 .VSURBEntryAllocationSize
= pipeline
->urb
.vs_size
- 1,
422 .VSNumberofURBEntries
= pipeline
->urb
.nr_vs_entries
);
424 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_URB_GS
),
425 .GSURBStartingAddress
= pipeline
->urb
.gs_start
,
426 .GSURBEntryAllocationSize
= pipeline
->urb
.gs_size
- 1,
427 .GSNumberofURBEntries
= pipeline
->urb
.nr_gs_entries
);
429 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_URB_HS
),
430 .HSURBStartingAddress
= pipeline
->urb
.vs_start
,
431 .HSURBEntryAllocationSize
= 0,
432 .HSNumberofURBEntries
= 0);
434 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_URB_DS
),
435 .DSURBStartingAddress
= pipeline
->urb
.vs_start
,
436 .DSURBEntryAllocationSize
= 0,
437 .DSNumberofURBEntries
= 0);
439 const struct brw_gs_prog_data
*gs_prog_data
= &pipeline
->gs_prog_data
;
441 length
= (gs_prog_data
->base
.vue_map
.num_slots
+ 1) / 2 - offset
;
443 if (pipeline
->gs_vec4
== NO_KERNEL
)
444 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_GS
), .Enable
= false);
446 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_GS
),
447 .SingleProgramFlow
= false,
448 .KernelStartPointer
= pipeline
->gs_vec4
,
449 .VectorMaskEnable
= Dmask
,
451 .BindingTableEntryCount
= 0,
452 .ExpectedVertexCount
= pipeline
->gs_vertex_count
,
454 .ScratchSpaceBasePointer
= pipeline
->scratch_start
[VK_SHADER_STAGE_GEOMETRY
],
455 .PerThreadScratchSpace
= ffs(gs_prog_data
->base
.base
.total_scratch
/ 2048),
457 .OutputVertexSize
= gs_prog_data
->output_vertex_size_hwords
* 2 - 1,
458 .OutputTopology
= gs_prog_data
->output_topology
,
459 .VertexURBEntryReadLength
= gs_prog_data
->base
.urb_read_length
,
460 .DispatchGRFStartRegisterForURBData
=
461 gs_prog_data
->base
.base
.dispatch_grf_start_reg
,
463 .MaximumNumberofThreads
= device
->info
.max_gs_threads
/ 2 - 1,
464 .ControlDataHeaderSize
= gs_prog_data
->control_data_header_size_hwords
,
465 .DispatchMode
= gs_prog_data
->base
.dispatch_mode
,
466 .StatisticsEnable
= true,
467 .IncludePrimitiveID
= gs_prog_data
->include_primitive_id
,
468 .ReorderMode
= TRAILING
,
471 .ControlDataFormat
= gs_prog_data
->control_data_format
,
473 .StaticOutput
= gs_prog_data
->static_vertex_count
>= 0,
474 .StaticOutputVertexCount
=
475 gs_prog_data
->static_vertex_count
>= 0 ?
476 gs_prog_data
->static_vertex_count
: 0,
478 /* FIXME: mesa sets this based on ctx->Transform.ClipPlanesEnabled:
479 * UserClipDistanceClipTestEnableBitmask_3DSTATE_GS(v)
480 * UserClipDistanceCullTestEnableBitmask(v)
483 .VertexURBEntryOutputReadOffset
= offset
,
484 .VertexURBEntryOutputLength
= length
);
486 const struct brw_vue_prog_data
*vue_prog_data
= &pipeline
->vs_prog_data
.base
;
487 /* Skip the VUE header and position slots */
489 length
= (vue_prog_data
->vue_map
.num_slots
+ 1) / 2 - offset
;
491 if (pipeline
->vs_simd8
== NO_KERNEL
|| (extra
&& extra
->disable_vs
))
492 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_VS
),
493 .FunctionEnable
= false,
494 /* Even if VS is disabled, SBE still gets the amount of
495 * vertex data to read from this field. */
496 .VertexURBEntryOutputReadOffset
= offset
,
497 .VertexURBEntryOutputLength
= length
);
499 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_VS
),
500 .KernelStartPointer
= pipeline
->vs_simd8
,
501 .SingleVertexDispatch
= Multiple
,
502 .VectorMaskEnable
= Dmask
,
504 .BindingTableEntryCount
=
505 vue_prog_data
->base
.binding_table
.size_bytes
/ 4,
506 .ThreadDispatchPriority
= Normal
,
507 .FloatingPointMode
= IEEE754
,
508 .IllegalOpcodeExceptionEnable
= false,
509 .AccessesUAV
= false,
510 .SoftwareExceptionEnable
= false,
512 .ScratchSpaceBasePointer
= pipeline
->scratch_start
[VK_SHADER_STAGE_VERTEX
],
513 .PerThreadScratchSpace
= ffs(vue_prog_data
->base
.total_scratch
/ 2048),
515 .DispatchGRFStartRegisterForURBData
=
516 vue_prog_data
->base
.dispatch_grf_start_reg
,
517 .VertexURBEntryReadLength
= vue_prog_data
->urb_read_length
,
518 .VertexURBEntryReadOffset
= 0,
520 .MaximumNumberofThreads
= device
->info
.max_vs_threads
- 1,
521 .StatisticsEnable
= false,
522 .SIMD8DispatchEnable
= true,
523 .VertexCacheDisable
= false,
524 .FunctionEnable
= true,
526 .VertexURBEntryOutputReadOffset
= offset
,
527 .VertexURBEntryOutputLength
= length
,
528 .UserClipDistanceClipTestEnableBitmask
= 0,
529 .UserClipDistanceCullTestEnableBitmask
= 0);
531 const struct brw_wm_prog_data
*wm_prog_data
= &pipeline
->wm_prog_data
;
533 /* TODO: We should clean this up. Among other things, this is mostly
534 * shared with other gens.
536 const struct brw_vue_map
*fs_input_map
;
537 if (pipeline
->gs_vec4
== NO_KERNEL
)
538 fs_input_map
= &vue_prog_data
->vue_map
;
540 fs_input_map
= &gs_prog_data
->base
.vue_map
;
542 struct GENX(3DSTATE_SBE_SWIZ
) swiz
= {
543 GENX(3DSTATE_SBE_SWIZ_header
),
546 int max_source_attr
= 0;
547 for (int attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
548 int input_index
= wm_prog_data
->urb_setup
[attr
];
553 /* We have to subtract two slots to accout for the URB entry output
554 * read offset in the VS and GS stages.
556 int source_attr
= fs_input_map
->varying_to_slot
[attr
] - 2;
557 max_source_attr
= MAX2(max_source_attr
, source_attr
);
559 if (input_index
>= 16)
562 swiz
.Attribute
[input_index
].SourceAttribute
= source_attr
;
565 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_SBE
),
566 .AttributeSwizzleEnable
= true,
567 .ForceVertexURBEntryReadLength
= false,
568 .ForceVertexURBEntryReadOffset
= false,
569 .VertexURBEntryReadLength
= DIV_ROUND_UP(max_source_attr
+ 1, 2),
570 .PointSpriteTextureCoordinateOrigin
= UPPERLEFT
,
571 .NumberofSFOutputAttributes
=
572 wm_prog_data
->num_varying_inputs
,
575 .Attribute0ActiveComponentFormat
= ACF_XYZW
,
576 .Attribute1ActiveComponentFormat
= ACF_XYZW
,
577 .Attribute2ActiveComponentFormat
= ACF_XYZW
,
578 .Attribute3ActiveComponentFormat
= ACF_XYZW
,
579 .Attribute4ActiveComponentFormat
= ACF_XYZW
,
580 .Attribute5ActiveComponentFormat
= ACF_XYZW
,
581 .Attribute6ActiveComponentFormat
= ACF_XYZW
,
582 .Attribute7ActiveComponentFormat
= ACF_XYZW
,
583 .Attribute8ActiveComponentFormat
= ACF_XYZW
,
584 .Attribute9ActiveComponentFormat
= ACF_XYZW
,
585 .Attribute10ActiveComponentFormat
= ACF_XYZW
,
586 .Attribute11ActiveComponentFormat
= ACF_XYZW
,
587 .Attribute12ActiveComponentFormat
= ACF_XYZW
,
588 .Attribute13ActiveComponentFormat
= ACF_XYZW
,
589 .Attribute14ActiveComponentFormat
= ACF_XYZW
,
590 .Attribute15ActiveComponentFormat
= ACF_XYZW
,
591 /* wow, much field, very attribute */
592 .Attribute16ActiveComponentFormat
= ACF_XYZW
,
593 .Attribute17ActiveComponentFormat
= ACF_XYZW
,
594 .Attribute18ActiveComponentFormat
= ACF_XYZW
,
595 .Attribute19ActiveComponentFormat
= ACF_XYZW
,
596 .Attribute20ActiveComponentFormat
= ACF_XYZW
,
597 .Attribute21ActiveComponentFormat
= ACF_XYZW
,
598 .Attribute22ActiveComponentFormat
= ACF_XYZW
,
599 .Attribute23ActiveComponentFormat
= ACF_XYZW
,
600 .Attribute24ActiveComponentFormat
= ACF_XYZW
,
601 .Attribute25ActiveComponentFormat
= ACF_XYZW
,
602 .Attribute26ActiveComponentFormat
= ACF_XYZW
,
603 .Attribute27ActiveComponentFormat
= ACF_XYZW
,
604 .Attribute28ActiveComponentFormat
= ACF_XYZW
,
605 .Attribute29ActiveComponentFormat
= ACF_XYZW
,
606 .Attribute28ActiveComponentFormat
= ACF_XYZW
,
607 .Attribute29ActiveComponentFormat
= ACF_XYZW
,
608 .Attribute30ActiveComponentFormat
= ACF_XYZW
,
612 uint32_t *dw
= anv_batch_emit_dwords(&pipeline
->batch
,
613 GENX(3DSTATE_SBE_SWIZ_length
));
614 GENX(3DSTATE_SBE_SWIZ_pack
)(&pipeline
->batch
, dw
, &swiz
);
616 const int num_thread_bias
= ANV_GEN
== 8 ? 2 : 1;
617 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_PS
),
618 .KernelStartPointer0
= pipeline
->ps_ksp0
,
620 .SingleProgramFlow
= false,
621 .VectorMaskEnable
= true,
624 .ScratchSpaceBasePointer
= pipeline
->scratch_start
[VK_SHADER_STAGE_FRAGMENT
],
625 .PerThreadScratchSpace
= ffs(wm_prog_data
->base
.total_scratch
/ 2048),
627 .MaximumNumberofThreadsPerPSD
= 64 - num_thread_bias
,
628 .PositionXYOffsetSelect
= wm_prog_data
->uses_pos_offset
?
629 POSOFFSET_SAMPLE
: POSOFFSET_NONE
,
630 .PushConstantEnable
= wm_prog_data
->base
.nr_params
> 0,
631 ._8PixelDispatchEnable
= pipeline
->ps_simd8
!= NO_KERNEL
,
632 ._16PixelDispatchEnable
= pipeline
->ps_simd16
!= NO_KERNEL
,
633 ._32PixelDispatchEnable
= false,
635 .DispatchGRFStartRegisterForConstantSetupData0
= pipeline
->ps_grf_start0
,
636 .DispatchGRFStartRegisterForConstantSetupData1
= 0,
637 .DispatchGRFStartRegisterForConstantSetupData2
= pipeline
->ps_grf_start2
,
639 .KernelStartPointer1
= 0,
640 .KernelStartPointer2
= pipeline
->ps_ksp2
);
642 bool per_sample_ps
= false;
643 anv_batch_emit(&pipeline
->batch
, GENX(3DSTATE_PS_EXTRA
),
644 .PixelShaderValid
= true,
645 .PixelShaderKillsPixel
= wm_prog_data
->uses_kill
,
646 .PixelShaderComputedDepthMode
= wm_prog_data
->computed_depth_mode
,
647 .AttributeEnable
= wm_prog_data
->num_varying_inputs
> 0,
648 .oMaskPresenttoRenderTarget
= wm_prog_data
->uses_omask
,
649 .PixelShaderIsPerSample
= per_sample_ps
,
651 .PixelShaderPullsBary
= wm_prog_data
->pulls_bary
,
652 .InputCoverageMaskState
= ICMS_NONE
656 *pPipeline
= anv_pipeline_to_handle(pipeline
);
661 VkResult
genX(compute_pipeline_create
)(
663 const VkComputePipelineCreateInfo
* pCreateInfo
,
664 VkPipeline
* pPipeline
)
666 ANV_FROM_HANDLE(anv_device
, device
, _device
);
667 struct anv_pipeline
*pipeline
;
670 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO
);
672 pipeline
= anv_device_alloc(device
, sizeof(*pipeline
), 8,
673 VK_SYSTEM_ALLOC_TYPE_API_OBJECT
);
674 if (pipeline
== NULL
)
675 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
677 pipeline
->device
= device
;
678 pipeline
->layout
= anv_pipeline_layout_from_handle(pCreateInfo
->layout
);
680 pipeline
->blend_state
.map
= NULL
;
682 result
= anv_reloc_list_init(&pipeline
->batch_relocs
, device
);
683 if (result
!= VK_SUCCESS
) {
684 anv_device_free(device
, pipeline
);
687 pipeline
->batch
.next
= pipeline
->batch
.start
= pipeline
->batch_data
;
688 pipeline
->batch
.end
= pipeline
->batch
.start
+ sizeof(pipeline
->batch_data
);
689 pipeline
->batch
.relocs
= &pipeline
->batch_relocs
;
691 anv_state_stream_init(&pipeline
->program_stream
,
692 &device
->instruction_block_pool
);
694 /* When we free the pipeline, we detect stages based on the NULL status
695 * of various prog_data pointers. Make them NULL by default.
697 memset(pipeline
->prog_data
, 0, sizeof(pipeline
->prog_data
));
698 memset(pipeline
->scratch_start
, 0, sizeof(pipeline
->scratch_start
));
700 pipeline
->vs_simd8
= NO_KERNEL
;
701 pipeline
->vs_vec4
= NO_KERNEL
;
702 pipeline
->gs_vec4
= NO_KERNEL
;
704 pipeline
->active_stages
= 0;
705 pipeline
->total_scratch
= 0;
707 assert(pCreateInfo
->stage
.stage
== VK_SHADER_STAGE_COMPUTE
);
708 ANV_FROM_HANDLE(anv_shader
, shader
, pCreateInfo
->stage
.shader
);
709 anv_pipeline_compile_cs(pipeline
, pCreateInfo
, shader
);
711 pipeline
->use_repclear
= false;
713 const struct brw_cs_prog_data
*cs_prog_data
= &pipeline
->cs_prog_data
;
715 anv_batch_emit(&pipeline
->batch
, GENX(MEDIA_VFE_STATE
),
716 .ScratchSpaceBasePointer
= pipeline
->scratch_start
[VK_SHADER_STAGE_COMPUTE
],
717 .PerThreadScratchSpace
= ffs(cs_prog_data
->base
.total_scratch
/ 2048),
718 .ScratchSpaceBasePointerHigh
= 0,
721 .MaximumNumberofThreads
= device
->info
.max_cs_threads
- 1,
722 .NumberofURBEntries
= 2,
723 .ResetGatewayTimer
= true,
725 .BypassGatewayControl
= true,
727 .URBEntryAllocationSize
= 2,
728 .CURBEAllocationSize
= 0);
730 struct brw_cs_prog_data
*prog_data
= &pipeline
->cs_prog_data
;
731 uint32_t group_size
= prog_data
->local_size
[0] *
732 prog_data
->local_size
[1] * prog_data
->local_size
[2];
733 pipeline
->cs_thread_width_max
= DIV_ROUND_UP(group_size
, prog_data
->simd_size
);
734 uint32_t remainder
= group_size
& (prog_data
->simd_size
- 1);
737 pipeline
->cs_right_mask
= ~0u >> (32 - remainder
);
739 pipeline
->cs_right_mask
= ~0u >> (32 - prog_data
->simd_size
);
742 *pPipeline
= anv_pipeline_to_handle(pipeline
);