1 /*
2 * Copyright © 2019 Raspberry Pi
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 #ifndef V3DV_CL_H
25 #define V3DV_CL_H
26
27 #include "broadcom/cle/v3d_packet_helpers.h"
28
29 #include "list.h"
30
31 struct v3dv_bo;
32 struct v3dv_job;
33 struct v3dv_cl;
34
35 void v3dv_job_add_bo(struct v3dv_job *job, struct v3dv_bo *bo);
36
37 /**
38 * Undefined structure, used for typechecking that you're passing the pointers
39 * to these functions correctly.
40 */
41 struct v3dv_cl_out;
42
43 /** A reference to a BO used in the CL packing functions */
44 struct v3dv_cl_reloc {
45 struct v3dv_bo *bo;
46 uint32_t offset;
47 };
48
49 struct v3dv_cl {
50 void *base;
51 struct v3dv_job *job;
52 struct v3dv_cl_out *next;
53 struct v3dv_bo *bo;
54 uint32_t size;
55 struct list_head bo_list;
56 };
57
58 static inline struct v3dv_cl_reloc
__unpack_address(const uint8_t * cl,uint32_t s,uint32_t e)59 __unpack_address(const uint8_t *cl, uint32_t s, uint32_t e)
60 {
61 struct v3dv_cl_reloc reloc =
62 { NULL, __gen_unpack_uint(cl, s, e) << (31 - (e - s)) };
63 return reloc;
64 }
65
66 static inline uint32_t
v3dv_cl_offset(struct v3dv_cl * cl)67 v3dv_cl_offset(struct v3dv_cl *cl)
68 {
69 return (char *)cl->next - (char *)cl->base;
70 }
71
72 static inline struct v3dv_cl_reloc
v3dv_cl_address(struct v3dv_bo * bo,uint32_t offset)73 v3dv_cl_address(struct v3dv_bo *bo, uint32_t offset)
74 {
75 struct v3dv_cl_reloc reloc = {
76 .bo = bo,
77 .offset = offset,
78 };
79 return reloc;
80 }
81
82 static inline struct v3dv_cl_reloc
v3dv_cl_get_address(struct v3dv_cl * cl)83 v3dv_cl_get_address(struct v3dv_cl *cl)
84 {
85 return (struct v3dv_cl_reloc){ .bo = cl->bo, .offset = v3dv_cl_offset(cl) };
86 }
87
88 void v3dv_cl_init(struct v3dv_job *job, struct v3dv_cl *cl);
89 void v3dv_cl_destroy(struct v3dv_cl *cl);
90
91 static inline struct v3dv_cl_out *
cl_start(struct v3dv_cl * cl)92 cl_start(struct v3dv_cl *cl)
93 {
94 return cl->next;
95 }
96
97 static inline void
cl_end(struct v3dv_cl * cl,struct v3dv_cl_out * next)98 cl_end(struct v3dv_cl *cl, struct v3dv_cl_out *next)
99 {
100 cl->next = next;
101 assert(v3dv_cl_offset(cl) <= cl->size);
102 }
103
104 static inline void
cl_advance(struct v3dv_cl_out ** cl,uint32_t n)105 cl_advance(struct v3dv_cl_out **cl, uint32_t n)
106 {
107 (*cl) = (struct v3dv_cl_out *)((char *)(*cl) + n);
108 }
109
110 static inline void
cl_aligned_u32(struct v3dv_cl_out ** cl,uint32_t n)111 cl_aligned_u32(struct v3dv_cl_out **cl, uint32_t n)
112 {
113 *(uint32_t *)(*cl) = n;
114 cl_advance(cl, 4);
115 }
116
117 static inline void
cl_aligned_f(struct v3dv_cl_out ** cl,float f)118 cl_aligned_f(struct v3dv_cl_out **cl, float f)
119 {
120 cl_aligned_u32(cl, fui(f));
121 }
122
123 static inline void
cl_aligned_reloc(struct v3dv_cl * cl,struct v3dv_cl_out ** cl_out,struct v3dv_bo * bo,uint32_t offset)124 cl_aligned_reloc(struct v3dv_cl *cl,
125 struct v3dv_cl_out **cl_out,
126 struct v3dv_bo *bo,
127 uint32_t offset)
128 {
129 cl_aligned_u32(cl_out, bo->offset + offset);
130 v3dv_job_add_bo(cl->job, bo);
131 }
132
133 uint32_t v3dv_cl_ensure_space(struct v3dv_cl *cl, uint32_t space, uint32_t alignment);
134 void v3dv_cl_ensure_space_with_branch(struct v3dv_cl *cl, uint32_t space);
135
136 /* We redefine ALIGN as a macro as we want to use cl_aligned_packet_length for
137 * struct fields
138 */
139 #define ALIGN(value, alignment) \
140 (((value) + (alignment) - 1) & ~((alignment) - 1))
141
142 #define cl_packet_header(packet) V3DX(packet ## _header)
143 #define cl_packet_length(packet) V3DX(packet ## _length)
144 #define cl_aligned_packet_length(packet, alignment) ALIGN(cl_packet_length(packet), alignment)
145 #define cl_packet_pack(packet) V3DX(packet ## _pack)
146 #define cl_packet_struct(packet) V3DX(packet)
147
148 /* Macro for setting up an emit of a CL struct. A temporary unpacked struct
149 * is created, which you get to set fields in of the form:
150 *
151 * cl_emit(bcl, FLAT_SHADE_FLAGS, flags) {
152 * .flags.flat_shade_flags = 1 << 2,
153 * }
154 *
155 * or default values only can be emitted with just:
156 *
157 * cl_emit(bcl, FLAT_SHADE_FLAGS, flags);
158 *
159 * The trick here is that we make a for loop that will execute the body
160 * (either the block or the ';' after the macro invocation) exactly once.
161 */
162 #define cl_emit(cl, packet, name) \
163 for (struct cl_packet_struct(packet) name = { \
164 cl_packet_header(packet) \
165 }, \
166 *_loop_terminate = &name; \
167 __builtin_expect(_loop_terminate != NULL, 1); \
168 ({ \
169 struct v3dv_cl_out *cl_out = cl_start(cl); \
170 cl_packet_pack(packet)(cl, (uint8_t *)cl_out, &name); \
171 cl_advance(&cl_out, cl_packet_length(packet)); \
172 cl_end(cl, cl_out); \
173 _loop_terminate = NULL; \
174 })) \
175
176 #define cl_emit_with_prepacked(cl, packet, prepacked, name) \
177 for (struct cl_packet_struct(packet) name = { \
178 cl_packet_header(packet) \
179 }, \
180 *_loop_terminate = &name; \
181 __builtin_expect(_loop_terminate != NULL, 1); \
182 ({ \
183 struct v3dv_cl_out *cl_out = cl_start(cl); \
184 uint8_t packed[cl_packet_length(packet)]; \
185 cl_packet_pack(packet)(cl, packed, &name); \
186 for (int _i = 0; _i < cl_packet_length(packet); _i++) \
187 ((uint8_t *)cl_out)[_i] = packed[_i] | (prepacked)[_i]; \
188 cl_advance(&cl_out, cl_packet_length(packet)); \
189 cl_end(cl, cl_out); \
190 _loop_terminate = NULL; \
191 })) \
192
193 /**
194 * Helper function called by the XML-generated pack functions for filling in
195 * an address field in shader records.
196 *
197 * Since we have a private address space as of VC5, our BOs can have lifelong
198 * offsets, and all the kernel needs to know is which BOs need to be paged in
199 * for this exec.
200 */
201 static inline void
cl_pack_emit_reloc(struct v3dv_cl * cl,const struct v3dv_cl_reloc * reloc)202 cl_pack_emit_reloc(struct v3dv_cl *cl, const struct v3dv_cl_reloc *reloc)
203 {
204 if (reloc->bo)
205 v3dv_job_add_bo(cl->job, reloc->bo);
206 }
207
208 #define cl_emit_prepacked_sized(cl, packet, size) do { \
209 memcpy((cl)->next, packet, size); \
210 cl_advance(&(cl)->next, size); \
211 } while (0)
212
213 #define cl_emit_prepacked(cl, packet) \
214 cl_emit_prepacked_sized(cl, packet, sizeof(*(packet)))
215
216 #define v3dv_pack(packed, packet, name) \
217 for (struct cl_packet_struct(packet) name = { \
218 cl_packet_header(packet) \
219 }, \
220 *_loop_terminate = &name; \
221 __builtin_expect(_loop_terminate != NULL, 1); \
222 ({ \
223 cl_packet_pack(packet)(NULL, (uint8_t *)packed, &name); \
224 VG(VALGRIND_CHECK_MEM_IS_DEFINED((uint8_t *)packed, \
225 cl_packet_length(packet))); \
226 _loop_terminate = NULL; \
227 })) \
228
229 #endif /* V3DV_CL_H */
230