1 /*
2 * Copyright © 2014 Broadcom
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 VC4_CL_H
25 #define VC4_CL_H
26
27 #include <stdint.h>
28
29 #include "util/u_math.h"
30 #include "util/macros.h"
31
32 struct vc4_bo;
33 struct vc4_job;
34 struct vc4_cl;
35
36 /**
37 * Undefined structure, used for typechecking that you're passing the pointers
38 * to these functions correctly.
39 */
40 struct vc4_cl_out;
41
42 /** A reference to a BO used in the CL packing functions */
43 struct vc4_cl_reloc {
44 struct vc4_bo *bo;
45 uint32_t offset;
46 };
47
48 static inline void cl_pack_emit_reloc(struct vc4_cl *cl, const struct vc4_cl_reloc *);
49
50 #define __gen_user_data struct vc4_cl
51 #define __gen_address_type struct vc4_cl_reloc
52 #define __gen_address_offset(reloc) ((reloc)->offset)
53 #define __gen_emit_reloc cl_pack_emit_reloc
54
55 #include "kernel/vc4_packet.h"
56 #include "broadcom/cle/v3d_packet_v21_pack.h"
57
58 struct vc4_cl {
59 void *base;
60 struct vc4_job *job;
61 struct vc4_cl_out *next;
62 struct vc4_cl_out *reloc_next;
63 uint32_t size;
64 #ifndef NDEBUG
65 uint32_t reloc_count;
66 #endif
67 };
68
69 void vc4_init_cl(struct vc4_job *job, struct vc4_cl *cl);
70 void vc4_reset_cl(struct vc4_cl *cl);
71 uint32_t vc4_gem_hindex(struct vc4_job *job, struct vc4_bo *bo);
72
73 struct PACKED unaligned_16 { uint16_t x; };
74 struct PACKED unaligned_32 { uint32_t x; };
75
cl_offset(struct vc4_cl * cl)76 static inline uint32_t cl_offset(struct vc4_cl *cl)
77 {
78 return (char *)cl->next - (char *)cl->base;
79 }
80
81 static inline void
cl_advance(struct vc4_cl_out ** cl,uint32_t n)82 cl_advance(struct vc4_cl_out **cl, uint32_t n)
83 {
84 (*cl) = (struct vc4_cl_out *)((char *)(*cl) + n);
85 }
86
87 static inline struct vc4_cl_out *
cl_start(struct vc4_cl * cl)88 cl_start(struct vc4_cl *cl)
89 {
90 return cl->next;
91 }
92
93 static inline void
cl_end(struct vc4_cl * cl,struct vc4_cl_out * next)94 cl_end(struct vc4_cl *cl, struct vc4_cl_out *next)
95 {
96 cl->next = next;
97 assert(cl_offset(cl) <= cl->size);
98 }
99
100
101 static inline void
put_unaligned_32(struct vc4_cl_out * ptr,uint32_t val)102 put_unaligned_32(struct vc4_cl_out *ptr, uint32_t val)
103 {
104 struct unaligned_32 *p = (void *)ptr;
105 p->x = val;
106 }
107
108 static inline void
put_unaligned_16(struct vc4_cl_out * ptr,uint16_t val)109 put_unaligned_16(struct vc4_cl_out *ptr, uint16_t val)
110 {
111 struct unaligned_16 *p = (void *)ptr;
112 p->x = val;
113 }
114
115 static inline void
cl_u8(struct vc4_cl_out ** cl,uint8_t n)116 cl_u8(struct vc4_cl_out **cl, uint8_t n)
117 {
118 *(uint8_t *)(*cl) = n;
119 cl_advance(cl, 1);
120 }
121
122 static inline void
cl_u16(struct vc4_cl_out ** cl,uint16_t n)123 cl_u16(struct vc4_cl_out **cl, uint16_t n)
124 {
125 put_unaligned_16(*cl, n);
126 cl_advance(cl, 2);
127 }
128
129 static inline void
cl_u32(struct vc4_cl_out ** cl,uint32_t n)130 cl_u32(struct vc4_cl_out **cl, uint32_t n)
131 {
132 put_unaligned_32(*cl, n);
133 cl_advance(cl, 4);
134 }
135
136 static inline void
cl_aligned_u32(struct vc4_cl_out ** cl,uint32_t n)137 cl_aligned_u32(struct vc4_cl_out **cl, uint32_t n)
138 {
139 *(uint32_t *)(*cl) = n;
140 cl_advance(cl, 4);
141 }
142
143 static inline void
cl_ptr(struct vc4_cl_out ** cl,void * ptr)144 cl_ptr(struct vc4_cl_out **cl, void *ptr)
145 {
146 *(struct vc4_cl_out **)(*cl) = ptr;
147 cl_advance(cl, sizeof(void *));
148 }
149
150 static inline void
cl_f(struct vc4_cl_out ** cl,float f)151 cl_f(struct vc4_cl_out **cl, float f)
152 {
153 cl_u32(cl, fui(f));
154 }
155
156 static inline void
cl_aligned_f(struct vc4_cl_out ** cl,float f)157 cl_aligned_f(struct vc4_cl_out **cl, float f)
158 {
159 cl_aligned_u32(cl, fui(f));
160 }
161
162 static inline struct vc4_cl_out *
cl_start_shader_reloc(struct vc4_cl * cl,uint32_t n)163 cl_start_shader_reloc(struct vc4_cl *cl, uint32_t n)
164 {
165 assert(cl->reloc_count == 0);
166 #ifndef NDEBUG
167 cl->reloc_count = n;
168 #endif
169 cl->reloc_next = cl->next;
170
171 /* Reserve the space where hindex will be written. */
172 cl_advance(&cl->next, n * 4);
173
174 return cl->next;
175 }
176
177 static inline void
cl_reloc(struct vc4_job * job,struct vc4_cl * cl,struct vc4_cl_out ** cl_out,struct vc4_bo * bo,uint32_t offset)178 cl_reloc(struct vc4_job *job, struct vc4_cl *cl, struct vc4_cl_out **cl_out,
179 struct vc4_bo *bo, uint32_t offset)
180 {
181 *(uint32_t *)cl->reloc_next = vc4_gem_hindex(job, bo);
182 cl_advance(&cl->reloc_next, 4);
183
184 #ifndef NDEBUG
185 cl->reloc_count--;
186 #endif
187
188 cl_u32(cl_out, offset);
189 }
190
191 static inline void
cl_aligned_reloc(struct vc4_job * job,struct vc4_cl * cl,struct vc4_cl_out ** cl_out,struct vc4_bo * bo,uint32_t offset)192 cl_aligned_reloc(struct vc4_job *job, struct vc4_cl *cl,
193 struct vc4_cl_out **cl_out,
194 struct vc4_bo *bo, uint32_t offset)
195 {
196 *(uint32_t *)cl->reloc_next = vc4_gem_hindex(job, bo);
197 cl_advance(&cl->reloc_next, 4);
198
199 #ifndef NDEBUG
200 cl->reloc_count--;
201 #endif
202
203 cl_aligned_u32(cl_out, offset);
204 }
205
206 /**
207 * Reference to a BO with its associated offset, used in the pack process.
208 */
209 static inline struct vc4_cl_reloc
cl_address(struct vc4_bo * bo,uint32_t offset)210 cl_address(struct vc4_bo *bo, uint32_t offset)
211 {
212 struct vc4_cl_reloc reloc = {
213 .bo = bo,
214 .offset = offset,
215 };
216 return reloc;
217 }
218
219 void cl_ensure_space(struct vc4_cl *cl, uint32_t size);
220
221 #define cl_packet_header(packet) V3D21_ ## packet ## _header
222 #define cl_packet_length(packet) V3D21_ ## packet ## _length
223 #define cl_packet_pack(packet) V3D21_ ## packet ## _pack
224 #define cl_packet_struct(packet) V3D21_ ## packet
225
226 static inline void *
cl_get_emit_space(struct vc4_cl_out ** cl,size_t size)227 cl_get_emit_space(struct vc4_cl_out **cl, size_t size)
228 {
229 void *addr = *cl;
230 cl_advance(cl, size);
231 return addr;
232 }
233
234 /* Macro for setting up an emit of a CL struct. A temporary unpacked struct
235 * is created, which you get to set fields in of the form:
236 *
237 * cl_emit(bcl, FLAT_SHADE_FLAGS, flags) {
238 * .flags.flat_shade_flags = 1 << 2,
239 * }
240 *
241 * or default values only can be emitted with just:
242 *
243 * cl_emit(bcl, FLAT_SHADE_FLAGS, flags);
244 *
245 * The trick here is that we make a for loop that will execute the body
246 * (either the block or the ';' after the macro invocation) exactly once.
247 * Also, *dst is actually of the wrong type, it's the
248 * uint8_t[cl_packet_length()] in the CL, not a cl_packet_struct(packet).
249 */
250 #define cl_emit(cl, packet, name) \
251 for (struct cl_packet_struct(packet) name = { \
252 cl_packet_header(packet) \
253 }, \
254 *_loop_terminate = &name; \
255 __builtin_expect(_loop_terminate != NULL, 1); \
256 ({ \
257 struct vc4_cl_out *cl_out = cl_start(cl); \
258 cl_packet_pack(packet)(cl, (uint8_t *)cl_out, &name); \
259 VG(VALGRIND_CHECK_MEM_IS_DEFINED(cl_out, \
260 cl_packet_length(packet))); \
261 cl_advance(&cl_out, cl_packet_length(packet)); \
262 cl_end(cl, cl_out); \
263 _loop_terminate = NULL; \
264 })) \
265
266 #define cl_emit_prepacked(cl, packet) do { \
267 memcpy((cl)->next, packet, sizeof(*packet)); \
268 cl_advance(&(cl)->next, sizeof(*packet)); \
269 } while (0)
270
271 /**
272 * Helper function called by the XML-generated pack functions for filling in
273 * an address field in shader records.
274 *
275 * Relocations for shader recs and texturing involve the packet (or uniforms
276 * stream) being preceded by the handles to the BOs, and the offset within the
277 * BO being in the stream (the output of this function).
278 */
279 static inline void
cl_pack_emit_reloc(struct vc4_cl * cl,const struct vc4_cl_reloc * reloc)280 cl_pack_emit_reloc(struct vc4_cl *cl, const struct vc4_cl_reloc *reloc)
281 {
282 *(uint32_t *)cl->reloc_next = vc4_gem_hindex(cl->job, reloc->bo);
283 cl_advance(&cl->reloc_next, 4);
284
285 #ifndef NDEBUG
286 cl->reloc_count--;
287 #endif
288 }
289
290 #endif /* VC4_CL_H */
291