src/panfrost/bifrost/bi_pack.c

   1 /*
   2  * Copyright (C) 2020 Collabora, Ltd.
   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 FROM,
  20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  21  * SOFTWARE.
  22  */
  23
  24 #include "compiler.h"
  25
  26 /* This file contains the final passes of the compiler. Running after
  27  * scheduling and RA, the IR is now finalized, so we need to emit it to actual
  28  * bits on the wire (as well as fixup branches) */
  29
  30 static uint64_t
  31 bi_pack_header(bi_clause *clause, bi_clause *next)
  32 {
  33         struct bifrost_header header = {
  34                 /* stub */
  35                 .no_end_of_shader = (next != NULL),
  36         };
  37
  38         uint64_t u = 0;
  39         memcpy(&u, &header, sizeof(header));
  40         return u;
  41 }
  42
  43 /* Represents the assignment of ports for a given bundle */
  44
  45 struct bi_registers {
  46         /* Register to assign to each port */
  47         unsigned port[4];
  48
  49         /* Read ports can be disabled */
  50         bool enabled[2];
  51
  52         /* Should we write FMA? what about ADD? If only a single port is
  53          * enabled it is in port 2, else ADD/FMA is 2/3 respectively */
  54         bool write_fma, write_add;
  55
  56         /* Should we read with port 3? */
  57         bool read_port3;
  58
  59         /* Packed uniform/constant */
  60         unsigned uniform_constant;
  61
  62         /* Whether writes are actually for the last instruction */
  63         bool first_instruction;
  64 };
  65
  66 /* Assigns a port for reading, before anything is written */
  67
  68 static void
  69 bi_assign_port_read(struct bi_registers *regs, unsigned src)
  70 {
  71         /* We only assign for registers */
  72         if (!(src & BIR_INDEX_REGISTER))
  73                 return;
  74
  75         unsigned reg = src & ~BIR_INDEX_REGISTER;
  76
  77         /* Check if we already assigned the port */
  78         for (unsigned i = 0; i <= 1; ++i) {
  79                 if (regs->port[i] == reg && regs->enabled[i])
  80                         return;
  81         }
  82
  83         if (regs->port[3] == reg && regs->read_port3)
  84                 return;
  85
  86         /* Assign it now */
  87
  88         for (unsigned i = 0; i <= 1; ++i) {
  89                 if (!regs->enabled[i]) {
  90                         regs->port[i] = reg;
  91                         regs->enabled[i] = true;
  92                         return;
  93                 }
  94         }
  95
  96         if (!regs->read_port3) {
  97                 regs->port[3] = reg;
  98                 regs->read_port3 = true;
  99         }
 100 }
 101
 102 static struct bi_registers
 103 bi_assign_ports(bi_bundle now, bi_bundle prev)
 104 {
 105         struct bi_registers regs = { 0 };
 106
 107         /* First, assign reads */
 108
 109         if (now.fma)
 110                 bi_foreach_src(now.fma, src)
 111                         bi_assign_port_read(&regs, now.fma->src[src]);
 112
 113         if (now.add)
 114                 bi_foreach_src(now.add, src)
 115                         bi_assign_port_read(&regs, now.add->src[src]);
 116
 117         /* Next, assign writes */
 118
 119         if (prev.fma && prev.fma->dest & BIR_INDEX_REGISTER) {
 120                 regs.port[2] = prev.fma->dest & ~BIR_INDEX_REGISTER;
 121                 regs.write_fma = true;
 122         }
 123
 124         if (prev.add && prev.add->dest & BIR_INDEX_REGISTER) {
 125                 unsigned r = prev.add->dest & ~BIR_INDEX_REGISTER;
 126
 127                 if (regs.write_fma) {
 128                         /* Scheduler constraint: cannot read 3 and write 2 */
 129                         assert(!regs.read_port3);
 130                         regs.port[3] = r;
 131                 } else {
 132                         regs.port[2] = r;
 133                 }
 134
 135                 regs.write_add = true;
 136         }
 137
 138         /* Finally, ensure port 1 > port 0 for the 63-x trick to function */
 139
 140         if (regs.enabled[0] && regs.enabled[1] && regs.port[1] < regs.port[0]) {
 141                 unsigned temp = regs.port[0];
 142                 regs.port[0] = regs.port[1];
 143                 regs.port[1] = temp;
 144         }
 145
 146         return regs;
 147 }
 148
 149 /* Determines the register control field, ignoring the first? flag */
 150
 151 static enum bifrost_reg_control
 152 bi_pack_register_ctrl_lo(struct bi_registers r)
 153 {
 154         if (r.write_fma) {
 155                 if (r.write_add) {
 156                         assert(!r.read_port3);
 157                         return BIFROST_WRITE_ADD_P2_FMA_P3;
 158                 } else {
 159                         if (r.read_port3)
 160                                 return BIFROST_WRITE_FMA_P2_READ_P3;
 161                         else
 162                                 return BIFROST_WRITE_FMA_P2;
 163                 }
 164         } else if (r.write_add) {
 165                 if (r.read_port3)
 166                         return BIFROST_WRITE_ADD_P2_READ_P3;
 167                 else
 168                         return BIFROST_WRITE_ADD_P2;
 169         } else if (r.read_port3)
 170                 return BIFROST_READ_P3;
 171         else
 172                 return BIFROST_REG_NONE;
 173 }
 174
 175 /* Ditto but account for the first? flag this time */
 176
 177 static enum bifrost_reg_control
 178 bi_pack_register_ctrl(struct bi_registers r)
 179 {
 180         enum bifrost_reg_control ctrl = bi_pack_register_ctrl_lo(r);
 181
 182         if (r.first_instruction) {
 183                 if (ctrl == BIFROST_REG_NONE)
 184                         ctrl = BIFROST_FIRST_NONE;
 185                 else
 186                         ctrl |= BIFROST_FIRST_NONE;
 187         }
 188
 189         return ctrl;
 190 }
 191
 192 static uint64_t
 193 bi_pack_registers(struct bi_registers regs)
 194 {
 195         enum bifrost_reg_control ctrl = bi_pack_register_ctrl(regs);
 196         struct bifrost_regs s;
 197         uint64_t packed = 0;
 198
 199         if (regs.enabled[1]) {
 200                 /* Gotta save that bit!~ Required by the 63-x trick */
 201                 assert(regs.port[1] > regs.port[0]);
 202                 assert(regs.enabled[0]);
 203
 204                 /* Do the 63-x trick, see docs/disasm */
 205                 if (regs.port[0] > 31) {
 206                         regs.port[0] = 63 - regs.port[0];
 207                         regs.port[1] = 63 - regs.port[1];
 208                 }
 209
 210                 assert(regs.port[0] <= 31);
 211                 assert(regs.port[1] <= 63);
 212
 213                 s.ctrl = ctrl;
 214                 s.reg1 = regs.port[1];
 215                 s.reg0 = regs.port[0];
 216         } else {
 217                 /* Port 1 disabled, so set to zero and use port 1 for ctrl */
 218                 s.reg1 = ctrl << 2;
 219
 220                 if (regs.enabled[0]) {
 221                         /* Bit 0 upper bit of port 0 */
 222                         s.reg1 |= (regs.port[0] >> 5);
 223
 224                         /* Rest of port 0 in usual spot */
 225                         s.reg0 = (regs.port[0] & 0b11111);
 226                 } else {
 227                         /* Bit 1 set if port 0 also disabled */
 228                         s.reg1 |= (1 << 1);
 229                 }
 230         }
 231
 232         s.reg3 = regs.port[3];
 233         s.reg2 = regs.port[2];
 234         s.uniform_const = regs.uniform_constant;
 235
 236         memcpy(&packed, &s, sizeof(s));
 237         return packed;
 238 }
 239
 240 static unsigned
 241 bi_pack_fma(bi_clause *clause, bi_bundle bundle)
 242 {
 243         /* TODO */
 244         return BIFROST_FMA_NOP;
 245 }
 246
 247 static unsigned
 248 bi_pack_add(bi_clause *clause, bi_bundle bundle)
 249 {
 250         /* TODO */
 251         return BIFROST_ADD_NOP;
 252 }
 253
 254 struct bi_packed_bundle {
 255         uint64_t lo;
 256         uint64_t hi;
 257 };
 258
 259 static struct bi_packed_bundle
 260 bi_pack_bundle(bi_clause *clause, bi_bundle bundle, bi_bundle prev)
 261 {
 262         struct bi_registers regs = bi_assign_ports(bundle, prev);
 263
 264         uint64_t reg = bi_pack_registers(regs);
 265         uint64_t fma = bi_pack_fma(clause, bundle);
 266         uint64_t add = bi_pack_add(clause, bundle);
 267
 268         struct bi_packed_bundle packed = {
 269                 .lo = reg | (fma << 35) | ((add & 0b111111) << 58),
 270                 .hi = add >> 6
 271         };
 272
 273         return packed;
 274 }
 275
 276 static void
 277 bi_pack_clause(bi_context *ctx, bi_clause *clause, bi_clause *next,
 278                 struct util_dynarray *emission)
 279 {
 280         struct bi_packed_bundle ins_1 = bi_pack_bundle(clause, clause->bundles[0], clause->bundles[0]);
 281         assert(clause->bundle_count == 1);
 282
 283         struct bifrost_fmt1 quad_1 = {
 284                 .tag = BIFROST_FMT1_FINAL,
 285                 .header = bi_pack_header(clause, next),
 286                 .ins_1 = ins_1.lo,
 287                 .ins_2 = ins_1.hi & ((1 << 11) - 1),
 288                 .ins_0 = (ins_1.hi >> 11) & 0b111,
 289         };
 290
 291         util_dynarray_append(emission, struct bifrost_fmt1, quad_1);
 292 }
 293
 294 static bi_clause *
 295 bi_next_clause(bi_context *ctx, pan_block *block, bi_clause *clause)
 296 {
 297         /* Try the next clause in this block */
 298         if (clause->link.next != &((bi_block *) block)->clauses)
 299                 return list_first_entry(&(clause->link), bi_clause, link);
 300
 301         /* Try the next block, or the one after that if it's empty, etc .*/
 302         pan_block *next_block = pan_next_block(block);
 303
 304         bi_foreach_block_from(ctx, next_block, block) {
 305                 bi_block *blk = (bi_block *) block;
 306
 307                 if (!list_is_empty(&blk->clauses))
 308                         return list_first_entry(&(blk->clauses), bi_clause, link);
 309         }
 310
 311         return NULL;
 312 }
 313
 314 void
 315 bi_pack(bi_context *ctx, struct util_dynarray *emission)
 316 {
 317         util_dynarray_init(emission, NULL);
 318
 319         bi_foreach_block(ctx, _block) {
 320                 bi_block *block = (bi_block *) _block;
 321
 322                 bi_foreach_clause_in_block(block, clause) {
 323                         bi_clause *next = bi_next_clause(ctx, _block, clause);
 324                         bi_pack_clause(ctx, clause, next, emission);
 325                 }
 326         }
 327 }