src/panfrost/bifrost/test/bi_test_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  * Authors (Collabora):
  24  *      Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
  25  */
  26
  27 #include "bit.h"
  28 #include "bi_print.h"
  29 #include "util/half_float.h"
  30 #include "bifrost/disassemble.h"
  31
  32 /* Instruction packing tests */
  33
  34 static bool
  35 bit_test_single(struct panfrost_device *dev,
  36                 bi_instruction *ins,
  37                 uint32_t input[4],
  38                 bool fma, enum bit_debug debug)
  39 {
  40         /* First, simulate the instruction */
  41         struct bit_state s = { 0 };
  42         memcpy(s.r, input, 16);
  43         bit_step(&s, ins, fma);
  44
  45         /* Next, wrap it up and pack it */
  46
  47         bi_instruction ldubo = {
  48                 .type = BI_LOAD_UNIFORM,
  49                 .src = {
  50                         BIR_INDEX_CONSTANT,
  51                         BIR_INDEX_ZERO
  52                 },
  53                 .src_types = {
  54                         nir_type_uint32,
  55                         nir_type_uint32,
  56                 },
  57                 .dest = BIR_INDEX_REGISTER | 0,
  58                 .dest_type = nir_type_uint32,
  59                 .writemask = 0xFFFF
  60         };
  61
  62         bi_instruction ldva = {
  63                 .type = BI_LOAD_VAR_ADDRESS,
  64                 .writemask = (1 << 12) - 1,
  65                 .dest = BIR_INDEX_REGISTER | 32,
  66                 .dest_type = nir_type_uint32,
  67                 .src = {
  68                         BIR_INDEX_CONSTANT,
  69                         BIR_INDEX_REGISTER | 61,
  70                         BIR_INDEX_REGISTER | 62,
  71                         0,
  72                 },
  73                 .src_types = {
  74                         nir_type_uint32,
  75                         nir_type_uint32,
  76                         nir_type_uint32,
  77                         nir_type_uint32,
  78                 }
  79         };
  80
  81         bi_instruction st = {
  82                 .type = BI_STORE_VAR,
  83                 .src = {
  84                         BIR_INDEX_REGISTER | 0,
  85                         ldva.dest, ldva.dest + 1, ldva.dest + 2,
  86                 },
  87                 .src_types = {
  88                         nir_type_uint32,
  89                         nir_type_uint32, nir_type_uint32, nir_type_uint32,
  90                 },
  91                 .store_channels = 4
  92         };
  93
  94         bi_context *ctx = rzalloc(NULL, bi_context);
  95         ctx->stage = MESA_SHADER_VERTEX;
  96
  97         bi_block *blk = rzalloc(ctx, bi_block);
  98         blk->scheduled = true;
  99
 100         blk->base.predecessors = _mesa_set_create(blk,
 101                         _mesa_hash_pointer,
 102                         _mesa_key_pointer_equal);
 103
 104         list_inithead(&ctx->blocks);
 105         list_addtail(&blk->base.link, &ctx->blocks);
 106         list_inithead(&blk->clauses);
 107
 108         bi_clause *clauses[4] = {
 109                 rzalloc(ctx, bi_clause),
 110                 rzalloc(ctx, bi_clause),
 111                 rzalloc(ctx, bi_clause),
 112                 rzalloc(ctx, bi_clause)
 113         };
 114
 115         for (unsigned i = 0; i < 4; ++i) {
 116                 clauses[i]->bundle_count = 1;
 117                 list_addtail(&clauses[i]->link, &blk->clauses);
 118                 clauses[i]->scoreboard_id = (i & 1);
 119
 120                 if (i) {
 121                         clauses[i]->dependencies = 1 << (~i & 1);
 122                         clauses[i]->data_register_write_barrier = true;
 123                 }
 124         }
 125
 126         clauses[0]->bundles[0].add = &ldubo;
 127         clauses[0]->clause_type = BIFROST_CLAUSE_UBO;
 128
 129         if (fma)
 130                 clauses[1]->bundles[0].fma = ins;
 131         else
 132                 clauses[1]->bundles[0].add = ins;
 133
 134         clauses[0]->constant_count = 1;
 135         clauses[1]->constant_count = 1;
 136         clauses[1]->constants[0] = ins->constant.u64;
 137
 138         clauses[2]->bundles[0].add = &ldva;
 139         clauses[3]->bundles[0].add = &st;
 140
 141         clauses[2]->clause_type = BIFROST_CLAUSE_UBO;
 142         clauses[3]->clause_type = BIFROST_CLAUSE_SSBO_STORE;
 143
 144         panfrost_program prog;
 145         bi_pack(ctx, &prog.compiled);
 146
 147         bool succ = bit_vertex(dev, prog, input, 16, NULL, 0,
 148                         s.r, 16, debug);
 149
 150         if (debug >= BIT_DEBUG_ALL || (!succ && debug >= BIT_DEBUG_FAIL)) {
 151                 bi_print_shader(ctx, stderr);
 152                 disassemble_bifrost(stderr, prog.compiled.data, prog.compiled.size, true);
 153         }
 154
 155         return succ;
 156 }
 157
 158 /* Utilities for generating tests */
 159
 160 static void
 161 bit_generate_vector(uint32_t *mem)
 162 {
 163         for (unsigned i = 0; i < 4; ++i)
 164                 mem[i] = rand();
 165 }
 166
 167 static bi_instruction
 168 bit_ins(enum bi_class C, unsigned argc, nir_alu_type base, unsigned size)
 169 {
 170         nir_alu_type T = base | size;
 171
 172         bi_instruction ins = {
 173                 .type = C,
 174                 .dest = BIR_INDEX_REGISTER | 0,
 175                 .dest_type = T,
 176         };
 177
 178         for (unsigned i = 0; i < argc; ++i) {
 179                 ins.src[i] = BIR_INDEX_REGISTER | i;
 180                 ins.src_types[i] = T;
 181         }
 182
 183         return ins;
 184 }
 185
 186 /* Tests all 64 combinations of floating point modifiers for a given
 187  * instruction / floating-type / test type */
 188
 189 static void
 190 bit_fmod_helper(struct panfrost_device *dev,
 191                 enum bi_class c, unsigned size, bool fma,
 192                 uint32_t *input, enum bit_debug debug)
 193 {
 194         bi_instruction ins = bit_ins(c, 2, nir_type_float, size);
 195
 196         for (unsigned outmod = 0; outmod < 4; ++outmod) {
 197                 for (unsigned inmod = 0; inmod < 16; ++inmod) {
 198                         ins.outmod = outmod;
 199                         ins.src_abs[0] = (inmod & 0x1);
 200                         ins.src_abs[1] = (inmod & 0x2);
 201                         ins.src_neg[0] = (inmod & 0x4);
 202                         ins.src_neg[1] = (inmod & 0x8);
 203
 204                         /* Skip over tests that cannot run on FMA */
 205                         if (fma && (size == 16) && ins.src_abs[0] && ins.src_abs[1])
 206                                 continue;
 207
 208                         if (!bit_test_single(dev, &ins, input, fma, debug)) {
 209                                 fprintf(stderr, "FAIL: fmod.%s%u.%s%s.%u\n",
 210                                                 bi_class_name(c),
 211                                                 size,
 212                                                 fma ? "fma" : "add",
 213                                                 outmod ? bi_output_mod_name(outmod) : ".none",
 214                                                 inmod);
 215                         }
 216                 }
 217         }
 218 }
 219
 220 static void
 221 bit_fma_helper(struct panfrost_device *dev,
 222                 unsigned size, uint32_t *input, enum bit_debug debug)
 223 {
 224         bi_instruction ins = bit_ins(BI_FMA, 3, nir_type_float, size);
 225
 226         for (unsigned outmod = 0; outmod < 4; ++outmod) {
 227                 for (unsigned inmod = 0; inmod < 8; ++inmod) {
 228                         ins.outmod = outmod;
 229                         ins.src_neg[0] = (inmod & 0x1);
 230                         ins.src_neg[1] = (inmod & 0x2);
 231                         ins.src_neg[2] = (inmod & 0x4);
 232
 233                         if (!bit_test_single(dev, &ins, input, true, debug)) {
 234                                 fprintf(stderr, "FAIL: fma%u%s.%u\n",
 235                                                 size,
 236                                                 outmod ? bi_output_mod_name(outmod) : ".none",
 237                                                 inmod);
 238                         }
 239                 }
 240         }
 241 }
 242
 243 static void
 244 bit_csel_helper(struct panfrost_device *dev,
 245                 unsigned size, uint32_t *input, enum bit_debug debug)
 246 {
 247         bi_instruction ins = bit_ins(BI_CSEL, 4, nir_type_uint, size);
 248
 249         /* SCHEDULER: We can only read 3 registers at once. */
 250         ins.src[2] = ins.src[0];
 251
 252         for (enum bi_cond cond = BI_COND_LT; cond <= BI_COND_NE; ++cond) {
 253                 ins.csel_cond = cond;
 254
 255                 if (!bit_test_single(dev, &ins, input, true, debug)) {
 256                         fprintf(stderr, "FAIL: csel%u.%s\n",
 257                                         size, bi_cond_name(cond));
 258                 }
 259         }
 260 }
 261
 262 void
 263 bit_fmod(struct panfrost_device *dev, enum bit_debug debug)
 264 {
 265         float input32[4] = { 0.8, 1.7, 0.0, 0.0 };
 266
 267         uint32_t input16[4] = {
 268                 _mesa_float_to_half(input32[0]) | (_mesa_float_to_half(-1.2) << 16),
 269                 _mesa_float_to_half(input32[1]) | (_mesa_float_to_half(0.9) << 16),
 270                 0, 0
 271         };
 272
 273         for (unsigned sz = 16; sz <= 32; sz *= 2) {
 274                 uint32_t *input =
 275                         (sz == 16) ? input16 :
 276                         (uint32_t *) input32;
 277
 278                 bit_fmod_helper(dev, BI_ADD, sz, true, input, debug);
 279         }
 280 }
 281
 282 void
 283 bit_fma(struct panfrost_device *dev, enum bit_debug debug)
 284 {
 285         float input32[4] = { 0.2, 1.6, -3.5, 0.0 };
 286
 287         uint32_t input16[4] = {
 288                 _mesa_float_to_half(input32[0]) | (_mesa_float_to_half(-1.8) << 16),
 289                 _mesa_float_to_half(input32[1]) | (_mesa_float_to_half(0.6) << 16),
 290                 _mesa_float_to_half(input32[1]) | (_mesa_float_to_half(16.2) << 16),
 291                 0
 292         };
 293
 294         for (unsigned sz = 16; sz <= 32; sz *= 2) {
 295                 uint32_t *input =
 296                         (sz == 16) ? input16 :
 297                         (uint32_t *) input32;
 298
 299                 bit_fma_helper(dev, sz, input, debug);
 300         }
 301 }
 302
 303 void
 304 bit_csel(struct panfrost_device *dev, enum bit_debug debug)
 305 {
 306         float input32[4] = { 0.2, 1.6, -3.5, 3.0 };
 307
 308         uint32_t input16[4] = {
 309                 _mesa_float_to_half(input32[0]) | (_mesa_float_to_half(-1.8) << 16),
 310                 _mesa_float_to_half(input32[1]) | (_mesa_float_to_half(0.6) << 16),
 311                 _mesa_float_to_half(input32[1]) | (_mesa_float_to_half(16.2) << 16),
 312                 _mesa_float_to_half(input32[2]) | (_mesa_float_to_half(4.9) << 16),
 313         };
 314
 315         for (unsigned sz = 32; sz <= 32; sz *= 2) {
 316                 uint32_t *input =
 317                         (sz == 16) ? input16 :
 318                         (uint32_t *) input32;
 319
 320                 bit_csel_helper(dev, sz, input, debug);
 321         }
 322 }