2 * Copyright © 2016 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
24 #include "anv_private.h"
26 #include "genxml/gen_macros.h"
27 #include "genxml/genX_pack.h"
29 #include "common/gen_l3_config.h"
32 * This file implements some lightweight memcpy/memset operations on the GPU
33 * using a vertex buffer and streamout.
37 * Returns the greatest common divisor of a and b that is a power of two.
40 gcd_pow2_u64(uint64_t a
, uint64_t b
)
42 assert(a
> 0 || b
> 0);
44 unsigned a_log2
= ffsll(a
) - 1;
45 unsigned b_log2
= ffsll(b
) - 1;
47 /* If either a or b is 0, then a_log2 or b_log2 will be UINT_MAX in which
48 * case, the MIN2() will take the other one. If both are 0 then we will
49 * hit the assert above.
51 return 1 << MIN2(a_log2
, b_log2
);
55 genX(cmd_buffer_mi_memcpy
)(struct anv_cmd_buffer
*cmd_buffer
,
56 struct anv_bo
*dst
, uint32_t dst_offset
,
57 struct anv_bo
*src
, uint32_t src_offset
,
60 /* This memcpy operates in units of dwords. */
61 assert(size
% 4 == 0);
62 assert(dst_offset
% 4 == 0);
63 assert(src_offset
% 4 == 0);
65 for (uint32_t i
= 0; i
< size
; i
+= 4) {
66 const struct anv_address src_addr
=
67 (struct anv_address
) { src
, src_offset
+ i
};
68 const struct anv_address dst_addr
=
69 (struct anv_address
) { dst
, dst_offset
+ i
};
71 anv_batch_emit(&cmd_buffer
->batch
, GENX(MI_COPY_MEM_MEM
), cp
) {
72 cp
.DestinationMemoryAddress
= dst_addr
;
73 cp
.SourceMemoryAddress
= src_addr
;
76 /* IVB does not have a general purpose register for command streamer
77 * commands. Therefore, we use an alternate temporary register.
79 #define TEMP_REG 0x2440 /* GEN7_3DPRIM_BASE_VERTEX */
80 anv_batch_emit(&cmd_buffer
->batch
, GENX(MI_LOAD_REGISTER_MEM
), load
) {
81 load
.RegisterAddress
= TEMP_REG
;
82 load
.MemoryAddress
= src_addr
;
84 anv_batch_emit(&cmd_buffer
->batch
, GENX(MI_STORE_REGISTER_MEM
), store
) {
85 store
.RegisterAddress
= TEMP_REG
;
86 store
.MemoryAddress
= dst_addr
;
95 genX(cmd_buffer_so_memcpy
)(struct anv_cmd_buffer
*cmd_buffer
,
96 struct anv_bo
*dst
, uint32_t dst_offset
,
97 struct anv_bo
*src
, uint32_t src_offset
,
103 assert(dst_offset
+ size
<= dst
->size
);
104 assert(src_offset
+ size
<= src
->size
);
106 /* The maximum copy block size is 4 32-bit components at a time. */
108 bs
= gcd_pow2_u64(bs
, src_offset
);
109 bs
= gcd_pow2_u64(bs
, dst_offset
);
110 bs
= gcd_pow2_u64(bs
, size
);
112 enum isl_format format
;
114 case 4: format
= ISL_FORMAT_R32_UINT
; break;
115 case 8: format
= ISL_FORMAT_R32G32_UINT
; break;
116 case 16: format
= ISL_FORMAT_R32G32B32A32_UINT
; break;
118 unreachable("Invalid size");
121 if (!cmd_buffer
->state
.current_l3_config
) {
122 const struct gen_l3_config
*cfg
=
123 gen_get_default_l3_config(&cmd_buffer
->device
->info
);
124 genX(cmd_buffer_config_l3
)(cmd_buffer
, cfg
);
127 genX(cmd_buffer_apply_pipe_flushes
)(cmd_buffer
);
129 genX(flush_pipeline_select_3d
)(cmd_buffer
);
132 dw
= anv_batch_emitn(&cmd_buffer
->batch
, 5, GENX(3DSTATE_VERTEX_BUFFERS
));
133 GENX(VERTEX_BUFFER_STATE_pack
)(&cmd_buffer
->batch
, dw
+ 1,
134 &(struct GENX(VERTEX_BUFFER_STATE
)) {
135 .VertexBufferIndex
= 32, /* Reserved for this */
136 .AddressModifyEnable
= true,
137 .BufferStartingAddress
= { src
, src_offset
},
140 .MemoryObjectControlState
= GENX(MOCS
),
143 .VertexBufferMemoryObjectControlState
= GENX(MOCS
),
144 .EndAddress
= { src
, src_offset
+ size
- 1 },
148 dw
= anv_batch_emitn(&cmd_buffer
->batch
, 3, GENX(3DSTATE_VERTEX_ELEMENTS
));
149 GENX(VERTEX_ELEMENT_STATE_pack
)(&cmd_buffer
->batch
, dw
+ 1,
150 &(struct GENX(VERTEX_ELEMENT_STATE
)) {
151 .VertexBufferIndex
= 32,
153 .SourceElementFormat
= (enum GENX(SURFACE_FORMAT
)) format
,
154 .SourceElementOffset
= 0,
155 .Component0Control
= (bs
>= 4) ? VFCOMP_STORE_SRC
: VFCOMP_STORE_0
,
156 .Component1Control
= (bs
>= 8) ? VFCOMP_STORE_SRC
: VFCOMP_STORE_0
,
157 .Component2Control
= (bs
>= 12) ? VFCOMP_STORE_SRC
: VFCOMP_STORE_0
,
158 .Component3Control
= (bs
>= 16) ? VFCOMP_STORE_SRC
: VFCOMP_STORE_0
,
162 anv_batch_emit(&cmd_buffer
->batch
, GENX(3DSTATE_VF_SGVS
), sgvs
);
165 /* Disable all shader stages */
166 anv_batch_emit(&cmd_buffer
->batch
, GENX(3DSTATE_VS
), vs
);
167 anv_batch_emit(&cmd_buffer
->batch
, GENX(3DSTATE_HS
), hs
);
168 anv_batch_emit(&cmd_buffer
->batch
, GENX(3DSTATE_TE
), te
);
169 anv_batch_emit(&cmd_buffer
->batch
, GENX(3DSTATE_DS
), DS
);
170 anv_batch_emit(&cmd_buffer
->batch
, GENX(3DSTATE_GS
), gs
);
171 anv_batch_emit(&cmd_buffer
->batch
, GENX(3DSTATE_PS
), gs
);
173 anv_batch_emit(&cmd_buffer
->batch
, GENX(3DSTATE_SBE
), sbe
) {
174 sbe
.VertexURBEntryReadOffset
= 1;
175 sbe
.NumberofSFOutputAttributes
= 1;
176 sbe
.VertexURBEntryReadLength
= 1;
178 sbe
.ForceVertexURBEntryReadLength
= true;
179 sbe
.ForceVertexURBEntryReadOffset
= true;
183 for (unsigned i
= 0; i
< 32; i
++)
184 sbe
.AttributeActiveComponentFormat
[i
] = ACF_XYZW
;
188 /* Emit URB setup. We tell it that the VS is active because we want it to
189 * allocate space for the VS. Even though one isn't run, we need VUEs to
190 * store the data that VF is going to pass to SOL.
192 const unsigned entry_size
[4] = { DIV_ROUND_UP(32, 64), 1, 1, 1 };
194 genX(emit_urb_setup
)(cmd_buffer
->device
, &cmd_buffer
->batch
,
195 cmd_buffer
->state
.current_l3_config
,
196 VK_SHADER_STAGE_VERTEX_BIT
, entry_size
);
198 anv_batch_emit(&cmd_buffer
->batch
, GENX(3DSTATE_SO_BUFFER
), sob
) {
199 sob
.SOBufferIndex
= 0;
200 sob
.SOBufferObjectControlState
= GENX(MOCS
);
201 sob
.SurfaceBaseAddress
= (struct anv_address
) { dst
, dst_offset
};
204 sob
.SOBufferEnable
= true;
205 sob
.SurfaceSize
= size
- 1;
207 sob
.SurfacePitch
= bs
;
208 sob
.SurfaceEndAddress
= sob
.SurfaceBaseAddress
;
209 sob
.SurfaceEndAddress
.offset
+= size
;
213 /* As SOL writes out data, it updates the SO_WRITE_OFFSET registers with
214 * the end position of the stream. We need to reset this value to 0 at
215 * the beginning of the run or else SOL will start at the offset from
218 sob
.StreamOffsetWriteEnable
= true;
219 sob
.StreamOffset
= 0;
224 /* The hardware can do this for us on BDW+ (see above) */
225 anv_batch_emit(&cmd_buffer
->batch
, GENX(MI_LOAD_REGISTER_IMM
), load
) {
226 load
.RegisterOffset
= GENX(SO_WRITE_OFFSET0_num
);
231 dw
= anv_batch_emitn(&cmd_buffer
->batch
, 5, GENX(3DSTATE_SO_DECL_LIST
),
232 .StreamtoBufferSelects0
= (1 << 0),
234 GENX(SO_DECL_ENTRY_pack
)(&cmd_buffer
->batch
, dw
+ 3,
235 &(struct GENX(SO_DECL_ENTRY
)) {
237 .OutputBufferSlot
= 0,
239 .ComponentMask
= (1 << (bs
/ 4)) - 1,
243 anv_batch_emit(&cmd_buffer
->batch
, GENX(3DSTATE_STREAMOUT
), so
) {
244 so
.SOFunctionEnable
= true;
245 so
.RenderingDisable
= true;
246 so
.Stream0VertexReadOffset
= 0;
247 so
.Stream0VertexReadLength
= DIV_ROUND_UP(32, 64);
249 so
.Buffer0SurfacePitch
= bs
;
251 so
.SOBufferEnable0
= true;
256 anv_batch_emit(&cmd_buffer
->batch
, GENX(3DSTATE_VF_TOPOLOGY
), topo
) {
257 topo
.PrimitiveTopologyType
= _3DPRIM_POINTLIST
;
261 anv_batch_emit(&cmd_buffer
->batch
, GENX(3DSTATE_VF_STATISTICS
), vf
) {
262 vf
.StatisticsEnable
= false;
265 anv_batch_emit(&cmd_buffer
->batch
, GENX(3DPRIMITIVE
), prim
) {
266 prim
.VertexAccessType
= SEQUENTIAL
;
267 prim
.PrimitiveTopologyType
= _3DPRIM_POINTLIST
;
268 prim
.VertexCountPerInstance
= size
/ bs
;
269 prim
.StartVertexLocation
= 0;
270 prim
.InstanceCount
= 1;
271 prim
.StartInstanceLocation
= 0;
272 prim
.BaseVertexLocation
= 0;
275 cmd_buffer
->state
.dirty
|= ANV_CMD_DIRTY_PIPELINE
;