src/gallium/drivers/vc4/vc4_qir.c

   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 #include "util/u_memory.h"
  25 #include "util/u_simple_list.h"
  26 #include "util/ralloc.h"
  27
  28 #include "vc4_qir.h"
  29 #include "vc4_qpu.h"
  30
  31 struct qir_op_info {
  32         const char *name;
  33         uint8_t ndst, nsrc;
  34         bool has_side_effects;
  35 };
  36
  37 static const struct qir_op_info qir_op_info[] = {
  38         [QOP_MOV] = { "mov", 1, 1 },
  39         [QOP_FADD] = { "fadd", 1, 2 },
  40         [QOP_FSUB] = { "fsub", 1, 2 },
  41         [QOP_FMUL] = { "fmul", 1, 2 },
  42         [QOP_MUL24] = { "mul24", 1, 2 },
  43         [QOP_FMIN] = { "fmin", 1, 2 },
  44         [QOP_FMAX] = { "fmax", 1, 2 },
  45         [QOP_FMINABS] = { "fminabs", 1, 2 },
  46         [QOP_FMAXABS] = { "fmaxabs", 1, 2 },
  47         [QOP_FTOI] = { "ftoi", 1, 1 },
  48         [QOP_ITOF] = { "itof", 1, 1 },
  49         [QOP_ADD] = { "add", 1, 2 },
  50         [QOP_SUB] = { "sub", 1, 2 },
  51         [QOP_SHR] = { "shr", 1, 2 },
  52         [QOP_ASR] = { "asr", 1, 2 },
  53         [QOP_SHL] = { "shl", 1, 2 },
  54         [QOP_MIN] = { "min", 1, 2 },
  55         [QOP_MAX] = { "max", 1, 2 },
  56         [QOP_AND] = { "and", 1, 2 },
  57         [QOP_OR] = { "or", 1, 2 },
  58         [QOP_XOR] = { "xor", 1, 2 },
  59         [QOP_NOT] = { "not", 1, 1 },
  60
  61         [QOP_SF] = { "sf", 0, 1 },
  62         [QOP_SEL_X_0_NS] = { "fsel_x_0_ns", 1, 1 },
  63         [QOP_SEL_X_0_NC] = { "fsel_x_0_nc", 1, 1 },
  64         [QOP_SEL_X_0_ZS] = { "fsel_x_0_zs", 1, 1 },
  65         [QOP_SEL_X_0_ZC] = { "fsel_x_0_zc", 1, 1 },
  66         [QOP_SEL_X_Y_NS] = { "fsel_x_y_ns", 1, 2 },
  67         [QOP_SEL_X_Y_NC] = { "fsel_x_y_nc", 1, 2 },
  68         [QOP_SEL_X_Y_ZS] = { "fsel_x_y_zs", 1, 2 },
  69         [QOP_SEL_X_Y_ZC] = { "fsel_x_y_zc", 1, 2 },
  70
  71         [QOP_RCP] = { "rcp", 1, 1 },
  72         [QOP_RSQ] = { "rsq", 1, 1 },
  73         [QOP_EXP2] = { "exp2", 1, 2 },
  74         [QOP_LOG2] = { "log2", 1, 2 },
  75         [QOP_PACK_COLORS] = { "pack_colors", 1, 4 },
  76         [QOP_PACK_SCALED] = { "pack_scaled", 1, 2 },
  77         [QOP_VPM_WRITE] = { "vpm_write", 0, 1, true },
  78         [QOP_VPM_READ] = { "vpm_read", 0, 1, true },
  79         [QOP_TLB_DISCARD_SETUP] = { "discard", 0, 1, true },
  80         [QOP_TLB_STENCIL_SETUP] = { "tlb_stencil_setup", 0, 1, true },
  81         [QOP_TLB_Z_WRITE] = { "tlb_z", 0, 1, true },
  82         [QOP_TLB_COLOR_WRITE] = { "tlb_color", 0, 1, true },
  83         [QOP_TLB_COLOR_READ] = { "tlb_color_read", 1, 0, true },
  84         [QOP_VARY_ADD_C] = { "vary_add_c", 1, 1 },
  85
  86         [QOP_FRAG_X] = { "frag_x", 1, 0 },
  87         [QOP_FRAG_Y] = { "frag_y", 1, 0 },
  88         [QOP_FRAG_Z] = { "frag_z", 1, 0 },
  89         [QOP_FRAG_W] = { "frag_w", 1, 0 },
  90
  91         [QOP_TEX_S] = { "tex_s", 0, 2 },
  92         [QOP_TEX_T] = { "tex_t", 0, 2 },
  93         [QOP_TEX_R] = { "tex_r", 0, 2 },
  94         [QOP_TEX_B] = { "tex_b", 0, 2 },
  95         [QOP_TEX_RESULT] = { "tex_result", 1, 0, true },
  96         [QOP_R4_UNPACK_A] = { "r4_unpack_a", 1, 1 },
  97         [QOP_R4_UNPACK_B] = { "r4_unpack_b", 1, 1 },
  98         [QOP_R4_UNPACK_C] = { "r4_unpack_c", 1, 1 },
  99         [QOP_R4_UNPACK_D] = { "r4_unpack_d", 1, 1 },
 100 };
 101
 102 static const char *
 103 qir_get_op_name(enum qop qop)
 104 {
 105         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 106                 return qir_op_info[qop].name;
 107         else
 108                 return "???";
 109 }
 110
 111 int
 112 qir_get_op_nsrc(enum qop qop)
 113 {
 114         if (qop < ARRAY_SIZE(qir_op_info) && qir_op_info[qop].name)
 115                 return qir_op_info[qop].nsrc;
 116         else
 117                 abort();
 118 }
 119
 120 bool
 121 qir_has_side_effects(struct qinst *inst)
 122 {
 123         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 124                 if (inst->src[i].file == QFILE_VARY)
 125                         return true;
 126         }
 127
 128         return qir_op_info[inst->op].has_side_effects;
 129 }
 130
 131 bool
 132 qir_depends_on_flags(struct qinst *inst)
 133 {
 134         switch (inst->op) {
 135         case QOP_SEL_X_0_NS:
 136         case QOP_SEL_X_0_NC:
 137         case QOP_SEL_X_0_ZS:
 138         case QOP_SEL_X_0_ZC:
 139         case QOP_SEL_X_Y_NS:
 140         case QOP_SEL_X_Y_NC:
 141         case QOP_SEL_X_Y_ZS:
 142         case QOP_SEL_X_Y_ZC:
 143                 return true;
 144         default:
 145                 return false;
 146         }
 147 }
 148
 149 bool
 150 qir_writes_r4(struct qinst *inst)
 151 {
 152         switch (inst->op) {
 153         case QOP_TEX_RESULT:
 154         case QOP_TLB_COLOR_READ:
 155         case QOP_RCP:
 156         case QOP_RSQ:
 157         case QOP_EXP2:
 158         case QOP_LOG2:
 159                 return true;
 160         default:
 161                 return false;
 162         }
 163 }
 164
 165 bool
 166 qir_reads_r4(struct qinst *inst)
 167 {
 168         switch (inst->op) {
 169         case QOP_R4_UNPACK_A:
 170         case QOP_R4_UNPACK_B:
 171         case QOP_R4_UNPACK_C:
 172         case QOP_R4_UNPACK_D:
 173                 return true;
 174         default:
 175                 return false;
 176         }
 177 }
 178
 179 static void
 180 qir_print_reg(struct qreg reg)
 181 {
 182         const char *files[] = {
 183                 [QFILE_TEMP] = "t",
 184                 [QFILE_VARY] = "v",
 185                 [QFILE_UNIF] = "u",
 186         };
 187
 188         if (reg.file == QFILE_NULL)
 189                 fprintf(stderr, "null");
 190         else
 191                 fprintf(stderr, "%s%d", files[reg.file], reg.index);
 192 }
 193
 194 void
 195 qir_dump_inst(struct qinst *inst)
 196 {
 197         fprintf(stderr, "%s ", qir_get_op_name(inst->op));
 198
 199         qir_print_reg(inst->dst);
 200         for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
 201                 fprintf(stderr, ", ");
 202                 qir_print_reg(inst->src[i]);
 203         }
 204 }
 205
 206 void
 207 qir_dump(struct vc4_compile *c)
 208 {
 209         struct simple_node *node;
 210
 211         foreach(node, &c->instructions) {
 212                 struct qinst *inst = (struct qinst *)node;
 213                 qir_dump_inst(inst);
 214                 fprintf(stderr, "\n");
 215         }
 216 }
 217
 218 struct qreg
 219 qir_get_temp(struct vc4_compile *c)
 220 {
 221         struct qreg reg;
 222
 223         reg.file = QFILE_TEMP;
 224         reg.index = c->num_temps++;
 225
 226         return reg;
 227 }
 228
 229 struct qinst *
 230 qir_inst(enum qop op, struct qreg dst, struct qreg src0, struct qreg src1)
 231 {
 232         struct qinst *inst = CALLOC_STRUCT(qinst);
 233
 234         inst->op = op;
 235         inst->dst = dst;
 236         inst->src = calloc(2, sizeof(inst->src[0]));
 237         inst->src[0] = src0;
 238         inst->src[1] = src1;
 239
 240         return inst;
 241 }
 242
 243 struct qinst *
 244 qir_inst4(enum qop op, struct qreg dst,
 245           struct qreg a,
 246           struct qreg b,
 247           struct qreg c,
 248           struct qreg d)
 249 {
 250         struct qinst *inst = CALLOC_STRUCT(qinst);
 251
 252         inst->op = op;
 253         inst->dst = dst;
 254         inst->src = calloc(4, sizeof(*inst->src));
 255         inst->src[0] = a;
 256         inst->src[1] = b;
 257         inst->src[2] = c;
 258         inst->src[3] = d;
 259
 260         return inst;
 261 }
 262
 263 void
 264 qir_emit(struct vc4_compile *c, struct qinst *inst)
 265 {
 266         insert_at_tail(&c->instructions, &inst->link);
 267 }
 268
 269 bool
 270 qir_reg_equals(struct qreg a, struct qreg b)
 271 {
 272         return a.file == b.file && a.index == b.index;
 273 }
 274
 275 struct vc4_compile *
 276 qir_compile_init(void)
 277 {
 278         struct vc4_compile *c = rzalloc(NULL, struct vc4_compile);
 279
 280         make_empty_list(&c->instructions);
 281
 282         c->output_position_index = -1;
 283         c->output_color_index = -1;
 284
 285         return c;
 286 }
 287
 288 void
 289 qir_remove_instruction(struct qinst *qinst)
 290 {
 291         remove_from_list(&qinst->link);
 292         free(qinst->src);
 293         free(qinst);
 294 }
 295
 296 void
 297 qir_compile_destroy(struct vc4_compile *c)
 298 {
 299         while (!is_empty_list(&c->instructions)) {
 300                 struct qinst *qinst =
 301                         (struct qinst *)first_elem(&c->instructions);
 302                 qir_remove_instruction(qinst);
 303         }
 304
 305         ralloc_free(c);
 306 }
 307
 308 const char *
 309 qir_get_stage_name(enum qstage stage)
 310 {
 311         static const char *names[] = {
 312                 [QSTAGE_FRAG] = "FS",
 313                 [QSTAGE_VERT] = "VS",
 314                 [QSTAGE_COORD] = "CS",
 315         };
 316
 317         return names[stage];
 318 }
 319
 320 #define OPTPASS(func)                                                   \
 321         do {                                                            \
 322                 bool stage_progress = func(c);                          \
 323                 if (stage_progress) {                                   \
 324                         progress = true;                                \
 325                         if (print_opt_debug) {                          \
 326                                 fprintf(stderr,                         \
 327                                         "QIR opt pass %2d: %s progress\n", \
 328                                         pass, #func);                   \
 329                         }                                               \
 330                 }                                                       \
 331         } while (0)
 332
 333 void
 334 qir_optimize(struct vc4_compile *c)
 335 {
 336         bool print_opt_debug = false;
 337         int pass = 1;
 338
 339         while (true) {
 340                 bool progress = false;
 341
 342                 OPTPASS(qir_opt_algebraic);
 343                 OPTPASS(qir_opt_cse);
 344                 OPTPASS(qir_opt_copy_propagation);
 345                 OPTPASS(qir_opt_dead_code);
 346
 347                 if (!progress)
 348                         break;
 349
 350                 pass++;
 351         }
 352 }