src/gallium/drivers/vc4/vc4_cl.h

   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 #include "kernel/vc4_packet.h"
  33
  34 struct vc4_bo;
  35 struct vc4_job;
  36
  37 /**
  38  * Undefined structure, used for typechecking that you're passing the pointers
  39  * to these functions correctly.
  40  */
  41 struct vc4_cl_out;
  42
  43 /** A reference to a BO used in the CL packing functions */
  44 struct vc4_cl_reloc {
  45         struct vc4_bo *bo;
  46         uint32_t offset;
  47 };
  48
  49 /* We don't call anything that packs a reloc yet, so don't implement it. */
  50 static inline void cl_pack_emit_reloc(void *cl, const struct vc4_cl_reloc *reloc)
  51 {
  52         abort();
  53 }
  54
  55 /* We don't use the data arg yet */
  56 #define __gen_user_data void
  57 #define __gen_address_type struct vc4_cl_reloc
  58 #define __gen_address_offset(reloc) ((reloc)->offset)
  59 #define __gen_emit_reloc cl_pack_emit_reloc
  60
  61 #include "kernel/vc4_packet.h"
  62 #include "broadcom/cle/v3d_packet_v21_pack.h"
  63
  64 struct vc4_cl {
  65         void *base;
  66         struct vc4_cl_out *next;
  67         struct vc4_cl_out *reloc_next;
  68         uint32_t size;
  69 #ifdef DEBUG
  70         uint32_t reloc_count;
  71 #endif
  72 };
  73
  74 void vc4_init_cl(void *mem_ctx, struct vc4_cl *cl);
  75 void vc4_reset_cl(struct vc4_cl *cl);
  76 void vc4_dump_cl(void *cl, uint32_t size, bool is_render);
  77 uint32_t vc4_gem_hindex(struct vc4_job *job, struct vc4_bo *bo);
  78
  79 struct PACKED unaligned_16 { uint16_t x; };
  80 struct PACKED unaligned_32 { uint32_t x; };
  81
  82 static inline uint32_t cl_offset(struct vc4_cl *cl)
  83 {
  84         return (char *)cl->next - (char *)cl->base;
  85 }
  86
  87 static inline void
  88 cl_advance(struct vc4_cl_out **cl, uint32_t n)
  89 {
  90         (*cl) = (struct vc4_cl_out *)((char *)(*cl) + n);
  91 }
  92
  93 static inline struct vc4_cl_out *
  94 cl_start(struct vc4_cl *cl)
  95 {
  96         return cl->next;
  97 }
  98
  99 static inline void
 100 cl_end(struct vc4_cl *cl, struct vc4_cl_out *next)
 101 {
 102         cl->next = next;
 103         assert(cl_offset(cl) <= cl->size);
 104 }
 105
 106
 107 static inline void
 108 put_unaligned_32(struct vc4_cl_out *ptr, uint32_t val)
 109 {
 110         struct unaligned_32 *p = (void *)ptr;
 111         p->x = val;
 112 }
 113
 114 static inline void
 115 put_unaligned_16(struct vc4_cl_out *ptr, uint16_t val)
 116 {
 117         struct unaligned_16 *p = (void *)ptr;
 118         p->x = val;
 119 }
 120
 121 static inline void
 122 cl_u8(struct vc4_cl_out **cl, uint8_t n)
 123 {
 124         *(uint8_t *)(*cl) = n;
 125         cl_advance(cl, 1);
 126 }
 127
 128 static inline void
 129 cl_u16(struct vc4_cl_out **cl, uint16_t n)
 130 {
 131         put_unaligned_16(*cl, n);
 132         cl_advance(cl, 2);
 133 }
 134
 135 static inline void
 136 cl_u32(struct vc4_cl_out **cl, uint32_t n)
 137 {
 138         put_unaligned_32(*cl, n);
 139         cl_advance(cl, 4);
 140 }
 141
 142 static inline void
 143 cl_aligned_u32(struct vc4_cl_out **cl, uint32_t n)
 144 {
 145         *(uint32_t *)(*cl) = n;
 146         cl_advance(cl, 4);
 147 }
 148
 149 static inline void
 150 cl_ptr(struct vc4_cl_out **cl, void *ptr)
 151 {
 152         *(struct vc4_cl_out **)(*cl) = ptr;
 153         cl_advance(cl, sizeof(void *));
 154 }
 155
 156 static inline void
 157 cl_f(struct vc4_cl_out **cl, float f)
 158 {
 159         cl_u32(cl, fui(f));
 160 }
 161
 162 static inline void
 163 cl_aligned_f(struct vc4_cl_out **cl, float f)
 164 {
 165         cl_aligned_u32(cl, fui(f));
 166 }
 167
 168 static inline void
 169 cl_start_reloc(struct vc4_cl *cl, struct vc4_cl_out **out, uint32_t n)
 170 {
 171         assert(n == 1 || n == 2);
 172 #ifdef DEBUG
 173         assert(cl->reloc_count == 0);
 174         cl->reloc_count = n;
 175 #endif
 176
 177         cl_u8(out, VC4_PACKET_GEM_HANDLES);
 178         cl->reloc_next = *out;
 179         cl_u32(out, 0); /* Space where hindex will be written. */
 180         cl_u32(out, 0); /* Space where hindex will be written. */
 181 }
 182
 183 static inline struct vc4_cl_out *
 184 cl_start_shader_reloc(struct vc4_cl *cl, uint32_t n)
 185 {
 186 #ifdef DEBUG
 187         assert(cl->reloc_count == 0);
 188         cl->reloc_count = n;
 189 #endif
 190         cl->reloc_next = cl->next;
 191
 192         /* Reserve the space where hindex will be written. */
 193         cl_advance(&cl->next, n * 4);
 194
 195         return cl->next;
 196 }
 197
 198 static inline void
 199 cl_reloc(struct vc4_job *job, struct vc4_cl *cl, struct vc4_cl_out **cl_out,
 200          struct vc4_bo *bo, uint32_t offset)
 201 {
 202         *(uint32_t *)cl->reloc_next = vc4_gem_hindex(job, bo);
 203         cl_advance(&cl->reloc_next, 4);
 204
 205 #ifdef DEBUG
 206         cl->reloc_count--;
 207 #endif
 208
 209         cl_u32(cl_out, offset);
 210 }
 211
 212 static inline void
 213 cl_aligned_reloc(struct vc4_job *job, struct vc4_cl *cl,
 214                  struct vc4_cl_out **cl_out,
 215                  struct vc4_bo *bo, uint32_t offset)
 216 {
 217         *(uint32_t *)cl->reloc_next = vc4_gem_hindex(job, bo);
 218         cl_advance(&cl->reloc_next, 4);
 219
 220 #ifdef DEBUG
 221         cl->reloc_count--;
 222 #endif
 223
 224         cl_aligned_u32(cl_out, offset);
 225 }
 226
 227 void cl_ensure_space(struct vc4_cl *cl, uint32_t size);
 228
 229 #define cl_packet_header(packet) V3D21_ ## packet ## _header
 230 #define cl_packet_length(packet) V3D21_ ## packet ## _length
 231 #define cl_packet_pack(packet)   V3D21_ ## packet ## _pack
 232 #define cl_packet_struct(packet)   V3D21_ ## packet
 233
 234 static inline void *
 235 cl_get_emit_space(struct vc4_cl_out **cl, size_t size)
 236 {
 237         void *addr = *cl;
 238         cl_advance(cl, size);
 239         return addr;
 240 }
 241
 242 /* Macro for setting up an emit of a CL struct.  A temporary unpacked struct
 243  * is created, which you get to set fields in of the form:
 244  *
 245  * cl_emit(bcl, FLAT_SHADE_FLAGS, flags) {
 246  *     .flags.flat_shade_flags = 1 << 2,
 247  * }
 248  *
 249  * or default values only can be emitted with just:
 250  *
 251  * cl_emit(bcl, FLAT_SHADE_FLAGS, flags);
 252  *
 253  * The trick here is that we make a for loop that will execute the body
 254  * (either the block or the ';' after the macro invocation) exactly once.
 255  * Also, *dst is actually of the wrong type, it's the
 256  * uint8_t[cl_packet_length()] in the CL, not a cl_packet_struct(packet).
 257  */
 258 #define cl_emit(cl_out, packet, name)                            \
 259         for (struct cl_packet_struct(packet) name = {            \
 260                 cl_packet_header(packet)                         \
 261         },                                                       \
 262         *_dst = cl_get_emit_space(cl_out, cl_packet_length(packet)); \
 263         __builtin_expect(_dst != NULL, 1);                       \
 264         ({                                                       \
 265                 cl_packet_pack(packet)(NULL, (uint8_t *)_dst, &name);  \
 266                 VG(VALGRIND_CHECK_MEM_IS_DEFINED(_dst,           \
 267                                                  cl_packet_length(packet))); \
 268                 _dst = NULL;                                     \
 269         }))                                                      \
 270
 271 #endif /* VC4_CL_H */