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 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 #ifdef DEBUG
  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 void vc4_dump_cl(void *cl, uint32_t size, bool is_render);
  72 uint32_t vc4_gem_hindex(struct vc4_job *job, struct vc4_bo *bo);
  73
  74 struct PACKED unaligned_16 { uint16_t x; };
  75 struct PACKED unaligned_32 { uint32_t x; };
  76
  77 static inline uint32_t cl_offset(struct vc4_cl *cl)
  78 {
  79         return (char *)cl->next - (char *)cl->base;
  80 }
  81
  82 static inline void
  83 cl_advance(struct vc4_cl_out **cl, uint32_t n)
  84 {
  85         (*cl) = (struct vc4_cl_out *)((char *)(*cl) + n);
  86 }
  87
  88 static inline struct vc4_cl_out *
  89 cl_start(struct vc4_cl *cl)
  90 {
  91         return cl->next;
  92 }
  93
  94 static inline void
  95 cl_end(struct vc4_cl *cl, struct vc4_cl_out *next)
  96 {
  97         cl->next = next;
  98         assert(cl_offset(cl) <= cl->size);
  99 }
 100
 101
 102 static inline void
 103 put_unaligned_32(struct vc4_cl_out *ptr, uint32_t val)
 104 {
 105         struct unaligned_32 *p = (void *)ptr;
 106         p->x = val;
 107 }
 108
 109 static inline void
 110 put_unaligned_16(struct vc4_cl_out *ptr, uint16_t val)
 111 {
 112         struct unaligned_16 *p = (void *)ptr;
 113         p->x = val;
 114 }
 115
 116 static inline void
 117 cl_u8(struct vc4_cl_out **cl, uint8_t n)
 118 {
 119         *(uint8_t *)(*cl) = n;
 120         cl_advance(cl, 1);
 121 }
 122
 123 static inline void
 124 cl_u16(struct vc4_cl_out **cl, uint16_t n)
 125 {
 126         put_unaligned_16(*cl, n);
 127         cl_advance(cl, 2);
 128 }
 129
 130 static inline void
 131 cl_u32(struct vc4_cl_out **cl, uint32_t n)
 132 {
 133         put_unaligned_32(*cl, n);
 134         cl_advance(cl, 4);
 135 }
 136
 137 static inline void
 138 cl_aligned_u32(struct vc4_cl_out **cl, uint32_t n)
 139 {
 140         *(uint32_t *)(*cl) = n;
 141         cl_advance(cl, 4);
 142 }
 143
 144 static inline void
 145 cl_ptr(struct vc4_cl_out **cl, void *ptr)
 146 {
 147         *(struct vc4_cl_out **)(*cl) = ptr;
 148         cl_advance(cl, sizeof(void *));
 149 }
 150
 151 static inline void
 152 cl_f(struct vc4_cl_out **cl, float f)
 153 {
 154         cl_u32(cl, fui(f));
 155 }
 156
 157 static inline void
 158 cl_aligned_f(struct vc4_cl_out **cl, float f)
 159 {
 160         cl_aligned_u32(cl, fui(f));
 161 }
 162
 163 static inline void
 164 cl_start_reloc(struct vc4_cl *cl, struct vc4_cl_out **out, uint32_t n)
 165 {
 166         assert(n == 1 || n == 2);
 167 #ifdef DEBUG
 168         assert(cl->reloc_count == 0);
 169         cl->reloc_count = n;
 170 #endif
 171
 172         cl_u8(out, VC4_PACKET_GEM_HANDLES);
 173         cl->reloc_next = *out;
 174         cl_u32(out, 0); /* Space where hindex will be written. */
 175         cl_u32(out, 0); /* Space where hindex will be written. */
 176 }
 177
 178 static inline struct vc4_cl_out *
 179 cl_start_shader_reloc(struct vc4_cl *cl, uint32_t n)
 180 {
 181 #ifdef DEBUG
 182         assert(cl->reloc_count == 0);
 183         cl->reloc_count = n;
 184 #endif
 185         cl->reloc_next = cl->next;
 186
 187         /* Reserve the space where hindex will be written. */
 188         cl_advance(&cl->next, n * 4);
 189
 190         return cl->next;
 191 }
 192
 193 static inline void
 194 cl_reloc(struct vc4_job *job, struct vc4_cl *cl, struct vc4_cl_out **cl_out,
 195          struct vc4_bo *bo, uint32_t offset)
 196 {
 197         *(uint32_t *)cl->reloc_next = vc4_gem_hindex(job, bo);
 198         cl_advance(&cl->reloc_next, 4);
 199
 200 #ifdef DEBUG
 201         cl->reloc_count--;
 202 #endif
 203
 204         cl_u32(cl_out, offset);
 205 }
 206
 207 static inline void
 208 cl_aligned_reloc(struct vc4_job *job, struct vc4_cl *cl,
 209                  struct vc4_cl_out **cl_out,
 210                  struct vc4_bo *bo, uint32_t offset)
 211 {
 212         *(uint32_t *)cl->reloc_next = vc4_gem_hindex(job, bo);
 213         cl_advance(&cl->reloc_next, 4);
 214
 215 #ifdef DEBUG
 216         cl->reloc_count--;
 217 #endif
 218
 219         cl_aligned_u32(cl_out, offset);
 220 }
 221
 222 /**
 223  * Reference to a BO with its associated offset, used in the pack process.
 224  */
 225 static inline struct vc4_cl_reloc
 226 cl_address(struct vc4_bo *bo, uint32_t offset)
 227 {
 228         struct vc4_cl_reloc reloc = {
 229                 .bo = bo,
 230                 .offset = offset,
 231         };
 232         return reloc;
 233 }
 234
 235 void cl_ensure_space(struct vc4_cl *cl, uint32_t size);
 236
 237 #define cl_packet_header(packet) V3D21_ ## packet ## _header
 238 #define cl_packet_length(packet) V3D21_ ## packet ## _length
 239 #define cl_packet_pack(packet)   V3D21_ ## packet ## _pack
 240 #define cl_packet_struct(packet)   V3D21_ ## packet
 241
 242 static inline void *
 243 cl_get_emit_space(struct vc4_cl_out **cl, size_t size)
 244 {
 245         void *addr = *cl;
 246         cl_advance(cl, size);
 247         return addr;
 248 }
 249
 250 /* Macro for setting up an emit of a CL struct.  A temporary unpacked struct
 251  * is created, which you get to set fields in of the form:
 252  *
 253  * cl_emit(bcl, FLAT_SHADE_FLAGS, flags) {
 254  *     .flags.flat_shade_flags = 1 << 2,
 255  * }
 256  *
 257  * or default values only can be emitted with just:
 258  *
 259  * cl_emit(bcl, FLAT_SHADE_FLAGS, flags);
 260  *
 261  * The trick here is that we make a for loop that will execute the body
 262  * (either the block or the ';' after the macro invocation) exactly once.
 263  * Also, *dst is actually of the wrong type, it's the
 264  * uint8_t[cl_packet_length()] in the CL, not a cl_packet_struct(packet).
 265  */
 266 #define cl_emit(cl, packet, name)                                \
 267         for (struct cl_packet_struct(packet) name = {            \
 268                 cl_packet_header(packet)                         \
 269         },                                                       \
 270         *_loop_terminate = &name;                                \
 271         __builtin_expect(_loop_terminate != NULL, 1);            \
 272         ({                                                       \
 273                 struct vc4_cl_out *cl_out = cl_start(cl);        \
 274                 cl_packet_pack(packet)(cl, (uint8_t *)cl_out, &name); \
 275                 VG(VALGRIND_CHECK_MEM_IS_DEFINED(cl_out,         \
 276                                                  cl_packet_length(packet))); \
 277                 cl_advance(&cl_out, cl_packet_length(packet));   \
 278                 cl_end(cl, cl_out);                              \
 279                 _loop_terminate = NULL;                          \
 280         }))                                                      \
 281
 282 /**
 283  * Helper function called by the XML-generated pack functions for filling in
 284  * an address field in shader records.
 285  *
 286  * Relocations for shader recs and texturing involve the packet (or uniforms
 287  * stream) being preceded by the handles to the BOs, and the offset within the
 288  * BO being in the stream (the output of this function).
 289  */
 290 static inline void
 291 cl_pack_emit_reloc(struct vc4_cl *cl, const struct vc4_cl_reloc *reloc)
 292 {
 293         *(uint32_t *)cl->reloc_next = vc4_gem_hindex(cl->job, reloc->bo);
 294         cl_advance(&cl->reloc_next, 4);
 295
 296 #ifdef DEBUG
 297         cl->reloc_count--;
 298 #endif
 299 }
 300
 301 #endif /* VC4_CL_H */