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_type.h"
  39 #include "gallivm/lp_bld_const.h"
  40 #include "gallivm/lp_bld_conv.h"
  41 #include "gallivm/lp_bld_debug.h"
  42 #include "lp_test.h"
  43
  44
  45 typedef void (*conv_test_ptr_t)(const void *src, const void *dst);
  46
  47
  48 void
  49 write_tsv_header(FILE *fp)
  50 {
  51    fprintf(fp,
  52            "result\t"
  53            "cycles_per_channel\t"
  54            "src_type\t"
  55            "dst_type\n");
  56
  57    fflush(fp);
  58 }
  59
  60
  61 static void
  62 write_tsv_row(FILE *fp,
  63               struct lp_type src_type,
  64               struct lp_type dst_type,
  65               double cycles,
  66               boolean success)
  67 {
  68    fprintf(fp, "%s\t", success ? "pass" : "fail");
  69
  70    fprintf(fp, "%.1f\t", cycles / MAX2(src_type.length, dst_type.length));
  71
  72    dump_type(fp, src_type);
  73    fprintf(fp, "\t");
  74
  75    dump_type(fp, dst_type);
  76    fprintf(fp, "\n");
  77
  78    fflush(fp);
  79 }
  80
  81
  82 static void
  83 dump_conv_types(FILE *fp,
  84                struct lp_type src_type,
  85                struct lp_type dst_type)
  86 {
  87    fprintf(fp, "src_type=");
  88    dump_type(fp, src_type);
  89
  90    fprintf(fp, " dst_type=");
  91    dump_type(fp, dst_type);
  92
  93    fprintf(fp, " ...\n");
  94    fflush(fp);
  95 }
  96
  97
  98 static LLVMValueRef
  99 add_conv_test(LLVMModuleRef module,
 100               struct lp_type src_type, unsigned num_srcs,
 101               struct lp_type dst_type, unsigned num_dsts)
 102 {
 103    LLVMTypeRef args[2];
 104    LLVMValueRef func;
 105    LLVMValueRef src_ptr;
 106    LLVMValueRef dst_ptr;
 107    LLVMBasicBlockRef block;
 108    LLVMBuilderRef builder;
 109    LLVMValueRef src[LP_MAX_VECTOR_LENGTH];
 110    LLVMValueRef dst[LP_MAX_VECTOR_LENGTH];
 111    unsigned i;
 112
 113    args[0] = LLVMPointerType(lp_build_vec_type(src_type), 0);
 114    args[1] = LLVMPointerType(lp_build_vec_type(dst_type), 0);
 115
 116    func = LLVMAddFunction(module, "test", LLVMFunctionType(LLVMVoidType(), args, 2, 0));
 117    LLVMSetFunctionCallConv(func, LLVMCCallConv);
 118    src_ptr = LLVMGetParam(func, 0);
 119    dst_ptr = LLVMGetParam(func, 1);
 120
 121    block = LLVMAppendBasicBlock(func, "entry");
 122    builder = LLVMCreateBuilder();
 123    LLVMPositionBuilderAtEnd(builder, block);
 124
 125    for(i = 0; i < num_srcs; ++i) {
 126       LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
 127       LLVMValueRef ptr = LLVMBuildGEP(builder, src_ptr, &index, 1, "");
 128       src[i] = LLVMBuildLoad(builder, ptr, "");
 129    }
 130
 131    lp_build_conv(builder, src_type, dst_type, src, num_srcs, dst, num_dsts);
 132
 133    for(i = 0; i < num_dsts; ++i) {
 134       LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
 135       LLVMValueRef ptr = LLVMBuildGEP(builder, dst_ptr, &index, 1, "");
 136       LLVMBuildStore(builder, dst[i], ptr);
 137    }
 138
 139    LLVMBuildRetVoid(builder);;
 140
 141    LLVMDisposeBuilder(builder);
 142    return func;
 143 }
 144
 145
 146 PIPE_ALIGN_STACK
 147 static boolean
 148 test_one(unsigned verbose,
 149          FILE *fp,
 150          struct lp_type src_type,
 151          struct lp_type dst_type)
 152 {
 153    LLVMModuleRef module = NULL;
 154    LLVMValueRef func = NULL;
 155    LLVMExecutionEngineRef engine = NULL;
 156    LLVMModuleProviderRef provider = NULL;
 157    LLVMPassManagerRef pass = NULL;
 158    char *error = NULL;
 159    conv_test_ptr_t conv_test_ptr;
 160    boolean success;
 161    const unsigned n = LP_TEST_NUM_SAMPLES;
 162    int64_t cycles[LP_TEST_NUM_SAMPLES];
 163    double cycles_avg = 0.0;
 164    unsigned num_srcs;
 165    unsigned num_dsts;
 166    double eps;
 167    unsigned i, j;
 168    void *code;
 169
 170    if(verbose >= 1)
 171       dump_conv_types(stdout, src_type, dst_type);
 172
 173    if(src_type.length > dst_type.length) {
 174       num_srcs = 1;
 175       num_dsts = src_type.length/dst_type.length;
 176    }
 177    else  {
 178       num_dsts = 1;
 179       num_srcs = dst_type.length/src_type.length;
 180    }
 181
 182    assert(src_type.width * src_type.length == dst_type.width * dst_type.length);
 183
 184    /* We must not loose or gain channels. Only precision */
 185    assert(src_type.length * num_srcs == dst_type.length * num_dsts);
 186
 187    eps = MAX2(lp_const_eps(src_type), lp_const_eps(dst_type));
 188
 189    module = LLVMModuleCreateWithName("test");
 190
 191    func = add_conv_test(module, src_type, num_srcs, dst_type, num_dsts);
 192
 193    if(LLVMVerifyModule(module, LLVMPrintMessageAction, &error)) {
 194       LLVMDumpModule(module);
 195       abort();
 196    }
 197    LLVMDisposeMessage(error);
 198
 199    provider = LLVMCreateModuleProviderForExistingModule(module);
 200    if (LLVMCreateJITCompiler(&engine, provider, 1, &error)) {
 201       if(verbose < 1)
 202          dump_conv_types(stderr, src_type, dst_type);
 203       fprintf(stderr, "%s\n", error);
 204       LLVMDisposeMessage(error);
 205       abort();
 206    }
 207
 208 #if 0
 209    pass = LLVMCreatePassManager();
 210    LLVMAddTargetData(LLVMGetExecutionEngineTargetData(engine), pass);
 211    /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
 212     * but there are more on SVN. */
 213    LLVMAddConstantPropagationPass(pass);
 214    LLVMAddInstructionCombiningPass(pass);
 215    LLVMAddPromoteMemoryToRegisterPass(pass);
 216    LLVMAddGVNPass(pass);
 217    LLVMAddCFGSimplificationPass(pass);
 218    LLVMRunPassManager(pass, module);
 219 #else
 220    (void)pass;
 221 #endif
 222
 223    if(verbose >= 2)
 224       LLVMDumpModule(module);
 225
 226    code = LLVMGetPointerToGlobal(engine, func);
 227    conv_test_ptr = (conv_test_ptr_t)pointer_to_func(code);
 228
 229    if(verbose >= 2)
 230       lp_disassemble(code);
 231
 232    success = TRUE;
 233    for(i = 0; i < n && success; ++i) {
 234       unsigned src_stride = src_type.length*src_type.width/8;
 235       unsigned dst_stride = dst_type.length*dst_type.width/8;
 236       PIPE_ALIGN_VAR(16) uint8_t src[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
 237       PIPE_ALIGN_VAR(16) uint8_t dst[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
 238       double fref[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
 239       uint8_t ref[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
 240       int64_t start_counter = 0;
 241       int64_t end_counter = 0;
 242
 243       for(j = 0; j < num_srcs; ++j) {
 244          random_vec(src_type, src + j*src_stride);
 245          read_vec(src_type, src + j*src_stride, fref + j*src_type.length);
 246       }
 247
 248       for(j = 0; j < num_dsts; ++j) {
 249          write_vec(dst_type, ref + j*dst_stride, fref + j*dst_type.length);
 250       }
 251
 252       start_counter = rdtsc();
 253       conv_test_ptr(src, dst);
 254       end_counter = rdtsc();
 255
 256       cycles[i] = end_counter - start_counter;
 257
 258       for(j = 0; j < num_dsts; ++j) {
 259          if(!compare_vec_with_eps(dst_type, dst + j*dst_stride, ref + j*dst_stride, eps))
 260             success = FALSE;
 261       }
 262
 263       if (!success || verbose >= 3) {
 264          if(verbose < 1)
 265             dump_conv_types(stderr, src_type, dst_type);
 266          if (success) {
 267             fprintf(stderr, "PASS\n");
 268          }
 269          else {
 270             fprintf(stderr, "MISMATCH\n");
 271          }
 272
 273          for(j = 0; j < num_srcs; ++j) {
 274             fprintf(stderr, "  Src%u: ", j);
 275             dump_vec(stderr, src_type, src + j*src_stride);
 276             fprintf(stderr, "\n");
 277          }
 278
 279 #if 1
 280          fprintf(stderr, "  Ref: ");
 281          for(j = 0; j < src_type.length*num_srcs; ++j)
 282             fprintf(stderr, " %f", fref[j]);
 283          fprintf(stderr, "\n");
 284 #endif
 285
 286          for(j = 0; j < num_dsts; ++j) {
 287             fprintf(stderr, "  Dst%u: ", j);
 288             dump_vec(stderr, dst_type, dst + j*dst_stride);
 289             fprintf(stderr, "\n");
 290
 291             fprintf(stderr, "  Ref%u: ", j);
 292             dump_vec(stderr, dst_type, ref + j*dst_stride);
 293             fprintf(stderr, "\n");
 294          }
 295       }
 296    }
 297
 298    /*
 299     * Unfortunately the output of cycle counter is not very reliable as it comes
 300     * -- sometimes we get outliers (due IRQs perhaps?) which are
 301     * better removed to avoid random or biased data.
 302     */
 303    {
 304       double sum = 0.0, sum2 = 0.0;
 305       double avg, std;
 306       unsigned m;
 307
 308       for(i = 0; i < n; ++i) {
 309          sum += cycles[i];
 310          sum2 += cycles[i]*cycles[i];
 311       }
 312
 313       avg = sum/n;
 314       std = sqrtf((sum2 - n*avg*avg)/n);
 315
 316       m = 0;
 317       sum = 0.0;
 318       for(i = 0; i < n; ++i) {
 319          if(fabs(cycles[i] - avg) <= 4.0*std) {
 320             sum += cycles[i];
 321             ++m;
 322          }
 323       }
 324
 325       cycles_avg = sum/m;
 326
 327    }
 328
 329    if(fp)
 330       write_tsv_row(fp, src_type, dst_type, cycles_avg, success);
 331
 332    if (!success) {
 333       static boolean firsttime = TRUE;
 334       if(firsttime) {
 335          if(verbose < 2)
 336             LLVMDumpModule(module);
 337          LLVMWriteBitcodeToFile(module, "conv.bc");
 338          fprintf(stderr, "conv.bc written\n");
 339          fprintf(stderr, "Invoke as \"llc -o - conv.bc\"\n");
 340          firsttime = FALSE;
 341          /* abort(); */
 342       }
 343    }
 344
 345    LLVMFreeMachineCodeForFunction(engine, func);
 346
 347    LLVMDisposeExecutionEngine(engine);
 348    if(pass)
 349       LLVMDisposePassManager(pass);
 350
 351    return success;
 352 }
 353
 354
 355 const struct lp_type conv_types[] = {
 356    /* float, fixed,  sign,  norm, width, len */
 357
 358    {   TRUE, FALSE,  TRUE,  TRUE,    32,   4 },
 359    {   TRUE, FALSE,  TRUE, FALSE,    32,   4 },
 360    {   TRUE, FALSE, FALSE,  TRUE,    32,   4 },
 361    {   TRUE, FALSE, FALSE, FALSE,    32,   4 },
 362
 363    /* TODO: test fixed formats too */
 364
 365    {  FALSE, FALSE,  TRUE,  TRUE,    16,   8 },
 366    {  FALSE, FALSE,  TRUE, FALSE,    16,   8 },
 367    {  FALSE, FALSE, FALSE,  TRUE,    16,   8 },
 368    {  FALSE, FALSE, FALSE, FALSE,    16,   8 },
 369
 370    {  FALSE, FALSE,  TRUE,  TRUE,    32,   4 },
 371    {  FALSE, FALSE,  TRUE, FALSE,    32,   4 },
 372    {  FALSE, FALSE, FALSE,  TRUE,    32,   4 },
 373    {  FALSE, FALSE, FALSE, FALSE,    32,   4 },
 374
 375    {  FALSE, FALSE,  TRUE,  TRUE,    16,   8 },
 376    {  FALSE, FALSE,  TRUE, FALSE,    16,   8 },
 377    {  FALSE, FALSE, FALSE,  TRUE,    16,   8 },
 378    {  FALSE, FALSE, FALSE, FALSE,    16,   8 },
 379
 380    {  FALSE, FALSE,  TRUE,  TRUE,     8,  16 },
 381    {  FALSE, FALSE,  TRUE, FALSE,     8,  16 },
 382    {  FALSE, FALSE, FALSE,  TRUE,     8,  16 },
 383    {  FALSE, FALSE, FALSE, FALSE,     8,  16 },
 384 };
 385
 386
 387 const unsigned num_types = sizeof(conv_types)/sizeof(conv_types[0]);
 388
 389
 390 boolean
 391 test_all(unsigned verbose, FILE *fp)
 392 {
 393    const struct lp_type *src_type;
 394    const struct lp_type *dst_type;
 395    boolean success = TRUE;
 396
 397    for(src_type = conv_types; src_type < &conv_types[num_types]; ++src_type) {
 398       for(dst_type = conv_types; dst_type < &conv_types[num_types]; ++dst_type) {
 399
 400          if(src_type == dst_type)
 401             continue;
 402
 403          if(src_type->norm != dst_type->norm)
 404             continue;
 405
 406          if(!test_one(verbose, fp, *src_type, *dst_type))
 407            success = FALSE;
 408
 409       }
 410    }
 411
 412    return success;
 413 }
 414
 415
 416 boolean
 417 test_some(unsigned verbose, FILE *fp, unsigned long n)
 418 {
 419    const struct lp_type *src_type;
 420    const struct lp_type *dst_type;
 421    unsigned long i;
 422    boolean success = TRUE;
 423
 424    for(i = 0; i < n; ++i) {
 425       src_type = &conv_types[rand() % num_types];
 426
 427       do {
 428          dst_type = &conv_types[rand() % num_types];
 429       } while (src_type == dst_type || src_type->norm != dst_type->norm);
 430
 431       if(!test_one(verbose, fp, *src_type, *dst_type))
 432         success = FALSE;
 433    }
 434
 435    return success;
 436 }
 437
 438
 439 boolean
 440 test_single(unsigned verbose, FILE *fp)
 441 {
 442    /*    float, fixed,  sign,  norm, width, len */
 443    struct lp_type f32x4_type =
 444       {   TRUE, FALSE,  TRUE,  TRUE,    32,   4 };
 445    struct lp_type ub8x4_type =
 446       {  FALSE, FALSE, FALSE,  TRUE,     8,  16 };
 447
 448    boolean success;
 449
 450    success = test_one(verbose, fp, f32x4_type, ub8x4_type);
 451
 452    return success;
 453 }