src/gallium/drivers/llvmpipe/lp_test_conv.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2009 VMware, Inc.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28
  29 /**
  30  * @file
  31  * Unit tests for type conversion.
  32  *
  33  * @author Jose Fonseca <jfonseca@vmware.com>
  34  */
  35
  36
  37 #include "util/u_pointer.h"
  38 #include "gallivm/lp_bld_init.h"
  39 #include "gallivm/lp_bld_type.h"
  40 #include "gallivm/lp_bld_const.h"
  41 #include "gallivm/lp_bld_conv.h"
  42 #include "gallivm/lp_bld_debug.h"
  43 #include "lp_test.h"
  44
  45
  46 typedef void (*conv_test_ptr_t)(const void *src, const void *dst);
  47
  48
  49 void
  50 write_tsv_header(FILE *fp)
  51 {
  52    fprintf(fp,
  53            "result\t"
  54            "cycles_per_channel\t"
  55            "src_type\t"
  56            "dst_type\n");
  57
  58    fflush(fp);
  59 }
  60
  61
  62 static void
  63 write_tsv_row(FILE *fp,
  64               struct lp_type src_type,
  65               struct lp_type dst_type,
  66               double cycles,
  67               boolean success)
  68 {
  69    fprintf(fp, "%s\t", success ? "pass" : "fail");
  70
  71    fprintf(fp, "%.1f\t", cycles / MAX2(src_type.length, dst_type.length));
  72
  73    dump_type(fp, src_type);
  74    fprintf(fp, "\t");
  75
  76    dump_type(fp, dst_type);
  77    fprintf(fp, "\n");
  78
  79    fflush(fp);
  80 }
  81
  82
  83 static void
  84 dump_conv_types(FILE *fp,
  85                struct lp_type src_type,
  86                struct lp_type dst_type)
  87 {
  88    fprintf(fp, "src_type=");
  89    dump_type(fp, src_type);
  90
  91    fprintf(fp, " dst_type=");
  92    dump_type(fp, dst_type);
  93
  94    fprintf(fp, " ...\n");
  95    fflush(fp);
  96 }
  97
  98
  99 static LLVMValueRef
 100 add_conv_test(struct gallivm_state *gallivm,
 101               struct lp_type src_type, unsigned num_srcs,
 102               struct lp_type dst_type, unsigned num_dsts)
 103 {
 104    LLVMModuleRef module = gallivm->module;
 105    LLVMContextRef context = gallivm->context;
 106    LLVMBuilderRef builder = gallivm->builder;
 107    LLVMTypeRef args[2];
 108    LLVMValueRef func;
 109    LLVMValueRef src_ptr;
 110    LLVMValueRef dst_ptr;
 111    LLVMBasicBlockRef block;
 112    LLVMValueRef src[LP_MAX_VECTOR_LENGTH];
 113    LLVMValueRef dst[LP_MAX_VECTOR_LENGTH];
 114    unsigned i;
 115
 116    args[0] = LLVMPointerType(lp_build_vec_type(gallivm, src_type), 0);
 117    args[1] = LLVMPointerType(lp_build_vec_type(gallivm, dst_type), 0);
 118
 119    func = LLVMAddFunction(module, "test",
 120                           LLVMFunctionType(LLVMVoidTypeInContext(context),
 121                                            args, 2, 0));
 122    LLVMSetFunctionCallConv(func, LLVMCCallConv);
 123    src_ptr = LLVMGetParam(func, 0);
 124    dst_ptr = LLVMGetParam(func, 1);
 125
 126    block = LLVMAppendBasicBlockInContext(context, func, "entry");
 127    LLVMPositionBuilderAtEnd(builder, block);
 128
 129    for(i = 0; i < num_srcs; ++i) {
 130       LLVMValueRef index = LLVMConstInt(LLVMInt32TypeInContext(context), i, 0);
 131       LLVMValueRef ptr = LLVMBuildGEP(builder, src_ptr, &index, 1, "");
 132       src[i] = LLVMBuildLoad(builder, ptr, "");
 133    }
 134
 135    lp_build_conv(gallivm, src_type, dst_type, src, num_srcs, dst, num_dsts);
 136
 137    for(i = 0; i < num_dsts; ++i) {
 138       LLVMValueRef index = LLVMConstInt(LLVMInt32TypeInContext(context), i, 0);
 139       LLVMValueRef ptr = LLVMBuildGEP(builder, dst_ptr, &index, 1, "");
 140       LLVMBuildStore(builder, dst[i], ptr);
 141    }
 142
 143    LLVMBuildRetVoid(builder);;
 144
 145    return func;
 146 }
 147
 148
 149 PIPE_ALIGN_STACK
 150 static boolean
 151 test_one(struct gallivm_state *gallivm, unsigned verbose,
 152          FILE *fp,
 153          struct lp_type src_type,
 154          struct lp_type dst_type)
 155 {
 156    LLVMModuleRef module = gallivm->module;
 157    LLVMExecutionEngineRef engine = gallivm->engine;
 158    LLVMValueRef func = NULL;
 159    char *error = NULL;
 160    conv_test_ptr_t conv_test_ptr;
 161    boolean success;
 162    const unsigned n = LP_TEST_NUM_SAMPLES;
 163    int64_t cycles[LP_TEST_NUM_SAMPLES];
 164    double cycles_avg = 0.0;
 165    unsigned num_srcs;
 166    unsigned num_dsts;
 167    double eps;
 168    unsigned i, j;
 169    void *code;
 170
 171    if (src_type.width * src_type.length != dst_type.width * dst_type.length &&
 172        src_type.length != dst_type.length) {
 173       return TRUE;
 174    }
 175
 176    if(verbose >= 1)
 177       dump_conv_types(stdout, src_type, dst_type);
 178
 179    if (src_type.length > dst_type.length) {
 180       num_srcs = 1;
 181       num_dsts = src_type.length/dst_type.length;
 182    }
 183    else if (src_type.length < dst_type.length) {
 184       num_dsts = 1;
 185       num_srcs = dst_type.length/src_type.length;
 186    }
 187    else  {
 188       num_dsts = 1;
 189       num_srcs = 1;
 190    }
 191
 192    /* We must not loose or gain channels. Only precision */
 193    assert(src_type.length * num_srcs == dst_type.length * num_dsts);
 194
 195    eps = MAX2(lp_const_eps(src_type), lp_const_eps(dst_type));
 196
 197    func = add_conv_test(gallivm, src_type, num_srcs, dst_type, num_dsts);
 198
 199    if(LLVMVerifyModule(module, LLVMPrintMessageAction, &error)) {
 200       LLVMDumpModule(module);
 201       abort();
 202    }
 203    LLVMDisposeMessage(error);
 204
 205    if(verbose >= 2)
 206       LLVMDumpModule(module);
 207
 208    code = LLVMGetPointerToGlobal(engine, func);
 209    conv_test_ptr = (conv_test_ptr_t)pointer_to_func(code);
 210
 211    if(verbose >= 2)
 212       lp_disassemble(code);
 213
 214    success = TRUE;
 215    for(i = 0; i < n && success; ++i) {
 216       unsigned src_stride = src_type.length*src_type.width/8;
 217       unsigned dst_stride = dst_type.length*dst_type.width/8;
 218       PIPE_ALIGN_VAR(16) uint8_t src[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
 219       PIPE_ALIGN_VAR(16) uint8_t dst[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
 220       double fref[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
 221       uint8_t ref[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
 222       int64_t start_counter = 0;
 223       int64_t end_counter = 0;
 224
 225       for(j = 0; j < num_srcs; ++j) {
 226          random_vec(src_type, src + j*src_stride);
 227          read_vec(src_type, src + j*src_stride, fref + j*src_type.length);
 228       }
 229
 230       for(j = 0; j < num_dsts; ++j) {
 231          write_vec(dst_type, ref + j*dst_stride, fref + j*dst_type.length);
 232       }
 233
 234       start_counter = rdtsc();
 235       conv_test_ptr(src, dst);
 236       end_counter = rdtsc();
 237
 238       cycles[i] = end_counter - start_counter;
 239
 240       for(j = 0; j < num_dsts; ++j) {
 241          if(!compare_vec_with_eps(dst_type, dst + j*dst_stride, ref + j*dst_stride, eps))
 242             success = FALSE;
 243       }
 244
 245       if (!success || verbose >= 3) {
 246          if(verbose < 1)
 247             dump_conv_types(stderr, src_type, dst_type);
 248          if (success) {
 249             fprintf(stderr, "PASS\n");
 250          }
 251          else {
 252             fprintf(stderr, "MISMATCH\n");
 253          }
 254
 255          for(j = 0; j < num_srcs; ++j) {
 256             fprintf(stderr, "  Src%u: ", j);
 257             dump_vec(stderr, src_type, src + j*src_stride);
 258             fprintf(stderr, "\n");
 259          }
 260
 261 #if 1
 262          fprintf(stderr, "  Ref: ");
 263          for(j = 0; j < src_type.length*num_srcs; ++j)
 264             fprintf(stderr, " %f", fref[j]);
 265          fprintf(stderr, "\n");
 266 #endif
 267
 268          for(j = 0; j < num_dsts; ++j) {
 269             fprintf(stderr, "  Dst%u: ", j);
 270             dump_vec(stderr, dst_type, dst + j*dst_stride);
 271             fprintf(stderr, "\n");
 272
 273             fprintf(stderr, "  Ref%u: ", j);
 274             dump_vec(stderr, dst_type, ref + j*dst_stride);
 275             fprintf(stderr, "\n");
 276          }
 277       }
 278    }
 279
 280    /*
 281     * Unfortunately the output of cycle counter is not very reliable as it comes
 282     * -- sometimes we get outliers (due IRQs perhaps?) which are
 283     * better removed to avoid random or biased data.
 284     */
 285    {
 286       double sum = 0.0, sum2 = 0.0;
 287       double avg, std;
 288       unsigned m;
 289
 290       for(i = 0; i < n; ++i) {
 291          sum += cycles[i];
 292          sum2 += cycles[i]*cycles[i];
 293       }
 294
 295       avg = sum/n;
 296       std = sqrtf((sum2 - n*avg*avg)/n);
 297
 298       m = 0;
 299       sum = 0.0;
 300       for(i = 0; i < n; ++i) {
 301          if(fabs(cycles[i] - avg) <= 4.0*std) {
 302             sum += cycles[i];
 303             ++m;
 304          }
 305       }
 306
 307       cycles_avg = sum/m;
 308
 309    }
 310
 311    if(fp)
 312       write_tsv_row(fp, src_type, dst_type, cycles_avg, success);
 313
 314    if (!success) {
 315       static boolean firsttime = TRUE;
 316       if(firsttime) {
 317          if(verbose < 2)
 318             LLVMDumpModule(module);
 319          LLVMWriteBitcodeToFile(module, "conv.bc");
 320          fprintf(stderr, "conv.bc written\n");
 321          fprintf(stderr, "Invoke as \"llc -o - conv.bc\"\n");
 322          firsttime = FALSE;
 323          /* abort(); */
 324       }
 325    }
 326
 327    LLVMFreeMachineCodeForFunction(engine, func);
 328
 329    return success;
 330 }
 331
 332
 333 const struct lp_type conv_types[] = {
 334    /* float, fixed,  sign,  norm, width, len */
 335
 336    {   TRUE, FALSE,  TRUE,  TRUE,    32,   4 },
 337    {   TRUE, FALSE,  TRUE, FALSE,    32,   4 },
 338    {   TRUE, FALSE, FALSE,  TRUE,    32,   4 },
 339    {   TRUE, FALSE, FALSE, FALSE,    32,   4 },
 340
 341    /* TODO: test fixed formats too */
 342
 343    {  FALSE, FALSE,  TRUE,  TRUE,    16,   8 },
 344    {  FALSE, FALSE,  TRUE, FALSE,    16,   8 },
 345    {  FALSE, FALSE, FALSE,  TRUE,    16,   8 },
 346    {  FALSE, FALSE, FALSE, FALSE,    16,   8 },
 347
 348    {  FALSE, FALSE,  TRUE,  TRUE,    32,   4 },
 349    {  FALSE, FALSE,  TRUE, FALSE,    32,   4 },
 350    {  FALSE, FALSE, FALSE,  TRUE,    32,   4 },
 351    {  FALSE, FALSE, FALSE, FALSE,    32,   4 },
 352
 353    {  FALSE, FALSE,  TRUE,  TRUE,    16,   8 },
 354    {  FALSE, FALSE,  TRUE, FALSE,    16,   8 },
 355    {  FALSE, FALSE, FALSE,  TRUE,    16,   8 },
 356    {  FALSE, FALSE, FALSE, FALSE,    16,   8 },
 357
 358    {  FALSE, FALSE,  TRUE,  TRUE,     8,  16 },
 359    {  FALSE, FALSE,  TRUE, FALSE,     8,  16 },
 360    {  FALSE, FALSE, FALSE,  TRUE,     8,  16 },
 361    {  FALSE, FALSE, FALSE, FALSE,     8,  16 },
 362
 363    {  FALSE, FALSE,  TRUE,  TRUE,     8,   4 },
 364    {  FALSE, FALSE,  TRUE, FALSE,     8,   4 },
 365    {  FALSE, FALSE, FALSE,  TRUE,     8,   4 },
 366    {  FALSE, FALSE, FALSE, FALSE,     8,   4 },
 367 };
 368
 369
 370 const unsigned num_types = sizeof(conv_types)/sizeof(conv_types[0]);
 371
 372
 373 boolean
 374 test_all(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 375 {
 376    const struct lp_type *src_type;
 377    const struct lp_type *dst_type;
 378    boolean success = TRUE;
 379
 380    for(src_type = conv_types; src_type < &conv_types[num_types]; ++src_type) {
 381       for(dst_type = conv_types; dst_type < &conv_types[num_types]; ++dst_type) {
 382
 383          if(src_type == dst_type)
 384             continue;
 385
 386          if(src_type->norm != dst_type->norm)
 387             continue;
 388
 389          if(!test_one(gallivm, verbose, fp, *src_type, *dst_type))
 390            success = FALSE;
 391
 392       }
 393    }
 394
 395    return success;
 396 }
 397
 398
 399 boolean
 400 test_some(struct gallivm_state *gallivm, unsigned verbose, FILE *fp,
 401           unsigned long n)
 402 {
 403    const struct lp_type *src_type;
 404    const struct lp_type *dst_type;
 405    unsigned long i;
 406    boolean success = TRUE;
 407
 408    for(i = 0; i < n; ++i) {
 409       src_type = &conv_types[rand() % num_types];
 410
 411       do {
 412          dst_type = &conv_types[rand() % num_types];
 413       } while (src_type == dst_type || src_type->norm != dst_type->norm);
 414
 415       if(!test_one(gallivm, verbose, fp, *src_type, *dst_type))
 416         success = FALSE;
 417    }
 418
 419    return success;
 420 }
 421
 422
 423 boolean
 424 test_single(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 425 {
 426    /*    float, fixed,  sign,  norm, width, len */
 427    struct lp_type f32x4_type =
 428       {   TRUE, FALSE,  TRUE,  TRUE,    32,   4 };
 429    struct lp_type ub8x4_type =
 430       {  FALSE, FALSE, FALSE,  TRUE,     8,  16 };
 431
 432    boolean success;
 433
 434    success = test_one(gallivm, verbose, fp, f32x4_type, ub8x4_type);
 435
 436    return success;
 437 }