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 "lp_bld_type.h"
  38 #include "lp_bld_const.h"
  39 #include "lp_bld_conv.h"
  40 #include "lp_bld_debug.h"
  41 #include "lp_test.h"
  42
  43
  44 typedef void (*conv_test_ptr_t)(const void *src, const void *dst);
  45
  46
  47 void
  48 write_tsv_header(FILE *fp)
  49 {
  50    fprintf(fp,
  51            "result\t"
  52            "cycles_per_channel\t"
  53            "src_type\t"
  54            "dst_type\n");
  55
  56    fflush(fp);
  57 }
  58
  59
  60 static void
  61 write_tsv_row(FILE *fp,
  62               struct lp_type src_type,
  63               struct lp_type dst_type,
  64               double cycles,
  65               boolean success)
  66 {
  67    fprintf(fp, "%s\t", success ? "pass" : "fail");
  68
  69    fprintf(fp, "%.1f\t", cycles / MAX2(src_type.length, dst_type.length));
  70
  71    dump_type(fp, src_type);
  72    fprintf(fp, "\t");
  73
  74    dump_type(fp, dst_type);
  75    fprintf(fp, "\n");
  76
  77    fflush(fp);
  78 }
  79
  80
  81 static void
  82 dump_conv_types(FILE *fp,
  83                struct lp_type src_type,
  84                struct lp_type dst_type)
  85 {
  86    fprintf(fp, "src_type=");
  87    dump_type(fp, src_type);
  88
  89    fprintf(fp, " dst_type=");
  90    dump_type(fp, dst_type);
  91
  92    fprintf(fp, " ...\n");
  93    fflush(fp);
  94 }
  95
  96
  97 static LLVMValueRef
  98 add_conv_test(LLVMModuleRef module,
  99               struct lp_type src_type, unsigned num_srcs,
 100               struct lp_type dst_type, unsigned num_dsts)
 101 {
 102    LLVMTypeRef args[2];
 103    LLVMValueRef func;
 104    LLVMValueRef src_ptr;
 105    LLVMValueRef dst_ptr;
 106    LLVMBasicBlockRef block;
 107    LLVMBuilderRef builder;
 108    LLVMValueRef src[LP_MAX_VECTOR_LENGTH];
 109    LLVMValueRef dst[LP_MAX_VECTOR_LENGTH];
 110    unsigned i;
 111
 112    args[0] = LLVMPointerType(lp_build_vec_type(src_type), 0);
 113    args[1] = LLVMPointerType(lp_build_vec_type(dst_type), 0);
 114
 115    func = LLVMAddFunction(module, "test", LLVMFunctionType(LLVMVoidType(), args, 2, 0));
 116    LLVMSetFunctionCallConv(func, LLVMCCallConv);
 117    src_ptr = LLVMGetParam(func, 0);
 118    dst_ptr = LLVMGetParam(func, 1);
 119
 120    block = LLVMAppendBasicBlock(func, "entry");
 121    builder = LLVMCreateBuilder();
 122    LLVMPositionBuilderAtEnd(builder, block);
 123
 124    for(i = 0; i < num_srcs; ++i) {
 125       LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
 126       LLVMValueRef ptr = LLVMBuildGEP(builder, src_ptr, &index, 1, "");
 127       src[i] = LLVMBuildLoad(builder, ptr, "");
 128    }
 129
 130    lp_build_conv(builder, src_type, dst_type, src, num_srcs, dst, num_dsts);
 131
 132    for(i = 0; i < num_dsts; ++i) {
 133       LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
 134       LLVMValueRef ptr = LLVMBuildGEP(builder, dst_ptr, &index, 1, "");
 135       LLVMBuildStore(builder, dst[i], ptr);
 136    }
 137
 138    LLVMBuildRetVoid(builder);;
 139
 140    LLVMDisposeBuilder(builder);
 141    return func;
 142 }
 143
 144
 145 static boolean
 146 test_one(unsigned verbose,
 147          FILE *fp,
 148          struct lp_type src_type,
 149          struct lp_type dst_type)
 150 {
 151    LLVMModuleRef module = NULL;
 152    LLVMValueRef func = NULL;
 153    LLVMExecutionEngineRef engine = NULL;
 154    LLVMModuleProviderRef provider = NULL;
 155    LLVMPassManagerRef pass = NULL;
 156    char *error = NULL;
 157    conv_test_ptr_t conv_test_ptr;
 158    boolean success;
 159    const unsigned n = 32;
 160    int64_t cycles[n];
 161    double cycles_avg = 0.0;
 162    unsigned num_srcs;
 163    unsigned num_dsts;
 164    double eps;
 165    unsigned i, j;
 166
 167    if(verbose >= 1)
 168       dump_conv_types(stdout, src_type, dst_type);
 169
 170    if(src_type.length > dst_type.length) {
 171       num_srcs = 1;
 172       num_dsts = src_type.length/dst_type.length;
 173    }
 174    else  {
 175       num_dsts = 1;
 176       num_srcs = dst_type.length/src_type.length;
 177    }
 178
 179    assert(src_type.width * src_type.length == dst_type.width * dst_type.length);
 180
 181    /* We must not loose or gain channels. Only precision */
 182    assert(src_type.length * num_srcs == dst_type.length * num_dsts);
 183
 184    eps = MAX2(lp_const_eps(src_type), lp_const_eps(dst_type));
 185
 186    module = LLVMModuleCreateWithName("test");
 187
 188    func = add_conv_test(module, src_type, num_srcs, dst_type, num_dsts);
 189
 190    if(LLVMVerifyModule(module, LLVMPrintMessageAction, &error)) {
 191       LLVMDumpModule(module);
 192       abort();
 193    }
 194    LLVMDisposeMessage(error);
 195
 196    provider = LLVMCreateModuleProviderForExistingModule(module);
 197    if (LLVMCreateJITCompiler(&engine, provider, 1, &error)) {
 198       if(verbose < 1)
 199          dump_conv_types(stderr, src_type, dst_type);
 200       fprintf(stderr, "%s\n", error);
 201       LLVMDisposeMessage(error);
 202       abort();
 203    }
 204
 205 #if 0
 206    pass = LLVMCreatePassManager();
 207    LLVMAddTargetData(LLVMGetExecutionEngineTargetData(engine), pass);
 208    /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
 209     * but there are more on SVN. */
 210    LLVMAddConstantPropagationPass(pass);
 211    LLVMAddInstructionCombiningPass(pass);
 212    LLVMAddPromoteMemoryToRegisterPass(pass);
 213    LLVMAddGVNPass(pass);
 214    LLVMAddCFGSimplificationPass(pass);
 215    LLVMRunPassManager(pass, module);
 216 #else
 217    (void)pass;
 218 #endif
 219
 220    if(verbose >= 2)
 221       LLVMDumpModule(module);
 222
 223    conv_test_ptr = (conv_test_ptr_t)LLVMGetPointerToGlobal(engine, func);
 224
 225    if(verbose >= 2)
 226       lp_disassemble(conv_test_ptr);
 227
 228    success = TRUE;
 229    for(i = 0; i < n && success; ++i) {
 230       unsigned src_stride = src_type.length*src_type.width/8;
 231       unsigned dst_stride = dst_type.length*dst_type.width/8;
 232       uint8_t src[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
 233       uint8_t dst[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
 234       double fref[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
 235       uint8_t ref[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
 236       int64_t start_counter = 0;
 237       int64_t end_counter = 0;
 238
 239       for(j = 0; j < num_srcs; ++j) {
 240          random_vec(src_type, src + j*src_stride);
 241          read_vec(src_type, src + j*src_stride, fref + j*src_type.length);
 242       }
 243
 244       for(j = 0; j < num_dsts; ++j) {
 245          write_vec(dst_type, ref + j*dst_stride, fref + j*dst_type.length);
 246       }
 247
 248       start_counter = rdtsc();
 249       conv_test_ptr(src, dst);
 250       end_counter = rdtsc();
 251
 252       cycles[i] = end_counter - start_counter;
 253
 254       for(j = 0; j < num_dsts; ++j) {
 255          if(!compare_vec_with_eps(dst_type, dst + j*dst_stride, ref + j*dst_stride, eps))
 256             success = FALSE;
 257       }
 258
 259       if (!success) {
 260          if(verbose < 1)
 261             dump_conv_types(stderr, src_type, dst_type);
 262          fprintf(stderr, "MISMATCH\n");
 263
 264          for(j = 0; j < num_srcs; ++j) {
 265             fprintf(stderr, "  Src%u: ", j);
 266             dump_vec(stderr, src_type, src + j*src_stride);
 267             fprintf(stderr, "\n");
 268          }
 269
 270 #if 1
 271          fprintf(stderr, "  Ref: ");
 272          for(j = 0; j < src_type.length*num_srcs; ++j)
 273             fprintf(stderr, " %f", fref[j]);
 274          fprintf(stderr, "\n");
 275 #endif
 276
 277          for(j = 0; j < num_dsts; ++j) {
 278             fprintf(stderr, "  Dst%u: ", j);
 279             dump_vec(stderr, dst_type, dst + j*dst_stride);
 280             fprintf(stderr, "\n");
 281
 282             fprintf(stderr, "  Ref%u: ", j);
 283             dump_vec(stderr, dst_type, ref + j*dst_stride);
 284             fprintf(stderr, "\n");
 285          }
 286       }
 287    }
 288
 289    /*
 290     * Unfortunately the output of cycle counter is not very reliable as it comes
 291     * -- sometimes we get outliers (due IRQs perhaps?) which are
 292     * better removed to avoid random or biased data.
 293     */
 294    {
 295       double sum = 0.0, sum2 = 0.0;
 296       double avg, std;
 297       unsigned m;
 298
 299       for(i = 0; i < n; ++i) {
 300          sum += cycles[i];
 301          sum2 += cycles[i]*cycles[i];
 302       }
 303
 304       avg = sum/n;
 305       std = sqrtf((sum2 - n*avg*avg)/n);
 306
 307       m = 0;
 308       sum = 0.0;
 309       for(i = 0; i < n; ++i) {
 310          if(fabs(cycles[i] - avg) <= 4.0*std) {
 311             sum += cycles[i];
 312             ++m;
 313          }
 314       }
 315
 316       cycles_avg = sum/m;
 317
 318    }
 319
 320    if(fp)
 321       write_tsv_row(fp, src_type, dst_type, cycles_avg, success);
 322
 323    if (!success) {
 324       static boolean firsttime = TRUE;
 325       if(firsttime) {
 326          if(verbose < 2)
 327             LLVMDumpModule(module);
 328          LLVMWriteBitcodeToFile(module, "conv.bc");
 329          fprintf(stderr, "conv.bc written\n");
 330          fprintf(stderr, "Invoke as \"llc -o - conv.bc\"\n");
 331          firsttime = FALSE;
 332          //abort();
 333       }
 334    }
 335
 336    LLVMFreeMachineCodeForFunction(engine, func);
 337
 338    LLVMDisposeExecutionEngine(engine);
 339    if(pass)
 340       LLVMDisposePassManager(pass);
 341
 342    return success;
 343 }
 344
 345
 346 const struct lp_type conv_types[] = {
 347    /* float, fixed,  sign,  norm, width, len */
 348
 349    {   TRUE, FALSE,  TRUE,  TRUE,    32,   4 },
 350    {   TRUE, FALSE,  TRUE, FALSE,    32,   4 },
 351    {   TRUE, FALSE, FALSE,  TRUE,    32,   4 },
 352    {   TRUE, FALSE, FALSE, FALSE,    32,   4 },
 353
 354    /* TODO: test fixed formats too */
 355
 356    {  FALSE, FALSE,  TRUE,  TRUE,    16,   8 },
 357    {  FALSE, FALSE,  TRUE, FALSE,    16,   8 },
 358    {  FALSE, FALSE, FALSE,  TRUE,    16,   8 },
 359    {  FALSE, FALSE, FALSE, FALSE,    16,   8 },
 360
 361    {  FALSE, FALSE,  TRUE,  TRUE,    32,   4 },
 362    {  FALSE, FALSE,  TRUE, FALSE,    32,   4 },
 363    {  FALSE, FALSE, FALSE,  TRUE,    32,   4 },
 364    {  FALSE, FALSE, FALSE, FALSE,    32,   4 },
 365
 366    {  FALSE, FALSE,  TRUE,  TRUE,    16,   8 },
 367    {  FALSE, FALSE,  TRUE, FALSE,    16,   8 },
 368    {  FALSE, FALSE, FALSE,  TRUE,    16,   8 },
 369    {  FALSE, FALSE, FALSE, FALSE,    16,   8 },
 370
 371    {  FALSE, FALSE,  TRUE,  TRUE,     8,  16 },
 372    {  FALSE, FALSE,  TRUE, FALSE,     8,  16 },
 373    {  FALSE, FALSE, FALSE,  TRUE,     8,  16 },
 374    {  FALSE, FALSE, FALSE, FALSE,     8,  16 },
 375 };
 376
 377
 378 const unsigned num_types = sizeof(conv_types)/sizeof(conv_types[0]);
 379
 380
 381 boolean
 382 test_all(unsigned verbose, FILE *fp)
 383 {
 384    const struct lp_type *src_type;
 385    const struct lp_type *dst_type;
 386    bool success = TRUE;
 387
 388    for(src_type = conv_types; src_type < &conv_types[num_types]; ++src_type) {
 389       for(dst_type = conv_types; dst_type < &conv_types[num_types]; ++dst_type) {
 390
 391          if(src_type == dst_type)
 392             continue;
 393
 394          if(src_type->norm != dst_type->norm)
 395             continue;
 396
 397          if(!test_one(verbose, fp, *src_type, *dst_type))
 398            success = FALSE;
 399
 400       }
 401    }
 402
 403    return success;
 404 }
 405
 406
 407 boolean
 408 test_some(unsigned verbose, FILE *fp, unsigned long n)
 409 {
 410    const struct lp_type *src_type;
 411    const struct lp_type *dst_type;
 412    unsigned long i;
 413    bool success = TRUE;
 414
 415    for(i = 0; i < n; ++i) {
 416       src_type = &conv_types[rand() % num_types];
 417
 418       do {
 419          dst_type = &conv_types[rand() % num_types];
 420       } while (src_type == dst_type || src_type->norm != dst_type->norm);
 421
 422       if(!test_one(verbose, fp, *src_type, *dst_type))
 423         success = FALSE;
 424    }
 425
 426    return success;
 427 }