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    gallivm_verify_function(gallivm, func);
 146
 147    return func;
 148 }
 149
 150
 151 PIPE_ALIGN_STACK
 152 static boolean
 153 test_one(unsigned verbose,
 154          FILE *fp,
 155          struct lp_type src_type,
 156          struct lp_type dst_type)
 157 {
 158    LLVMContextRef context;
 159    struct gallivm_state *gallivm;
 160    LLVMValueRef func = NULL;
 161    conv_test_ptr_t conv_test_ptr;
 162    boolean success;
 163    const unsigned n = LP_TEST_NUM_SAMPLES;
 164    int64_t cycles[LP_TEST_NUM_SAMPLES];
 165    double cycles_avg = 0.0;
 166    unsigned num_srcs;
 167    unsigned num_dsts;
 168    double eps;
 169    unsigned i, j;
 170
 171    if ((src_type.width >= dst_type.width && src_type.length > dst_type.length) ||
 172        (src_type.width <= dst_type.width && src_type.length < dst_type.length)) {
 173       return TRUE;
 174    }
 175
 176    /* Known failures
 177     * - fixed point 32 -> float 32
 178     * - float 32 -> signed normalised integer 32
 179     */
 180    if ((src_type.floating && !dst_type.floating && dst_type.sign && dst_type.norm && src_type.width == dst_type.width) ||
 181        (!src_type.floating && dst_type.floating && src_type.fixed && src_type.width == dst_type.width)) {
 182       return TRUE;
 183    }
 184
 185    /* Known failures
 186     * - fixed point 32 -> float 32
 187     * - float 32 -> signed normalised integer 32
 188     */
 189    if ((src_type.floating && !dst_type.floating && dst_type.sign && dst_type.norm && src_type.width == dst_type.width) ||
 190        (!src_type.floating && dst_type.floating && src_type.fixed && src_type.width == dst_type.width)) {
 191       return TRUE;
 192    }
 193
 194    if(verbose >= 1)
 195       dump_conv_types(stderr, src_type, dst_type);
 196
 197    if (src_type.length > dst_type.length) {
 198       num_srcs = 1;
 199       num_dsts = src_type.length/dst_type.length;
 200    }
 201    else if (src_type.length < dst_type.length) {
 202       num_dsts = 1;
 203       num_srcs = dst_type.length/src_type.length;
 204    }
 205    else  {
 206       num_dsts = 1;
 207       num_srcs = 1;
 208    }
 209
 210    /* We must not loose or gain channels. Only precision */
 211    assert(src_type.length * num_srcs == dst_type.length * num_dsts);
 212
 213    eps = MAX2(lp_const_eps(src_type), lp_const_eps(dst_type));
 214
 215    context = LLVMContextCreate();
 216    gallivm = gallivm_create("test_module", context);
 217
 218    func = add_conv_test(gallivm, src_type, num_srcs, dst_type, num_dsts);
 219
 220    gallivm_compile_module(gallivm);
 221
 222    conv_test_ptr = (conv_test_ptr_t)gallivm_jit_function(gallivm, func);
 223
 224    gallivm_free_ir(gallivm);
 225
 226    success = TRUE;
 227    for(i = 0; i < n && success; ++i) {
 228       unsigned src_stride = src_type.length*src_type.width/8;
 229       unsigned dst_stride = dst_type.length*dst_type.width/8;
 230       PIPE_ALIGN_VAR(LP_MIN_VECTOR_ALIGN) uint8_t src[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
 231       PIPE_ALIGN_VAR(LP_MIN_VECTOR_ALIGN) uint8_t dst[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
 232       double fref[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
 233       uint8_t ref[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
 234       int64_t start_counter = 0;
 235       int64_t end_counter = 0;
 236
 237       for(j = 0; j < num_srcs; ++j) {
 238          random_vec(src_type, src + j*src_stride);
 239          read_vec(src_type, src + j*src_stride, fref + j*src_type.length);
 240       }
 241
 242       for(j = 0; j < num_dsts; ++j) {
 243          write_vec(dst_type, ref + j*dst_stride, fref + j*dst_type.length);
 244       }
 245
 246       start_counter = rdtsc();
 247       conv_test_ptr(src, dst);
 248       end_counter = rdtsc();
 249
 250       cycles[i] = end_counter - start_counter;
 251
 252       for(j = 0; j < num_dsts; ++j) {
 253          if(!compare_vec_with_eps(dst_type, dst + j*dst_stride, ref + j*dst_stride, eps))
 254             success = FALSE;
 255       }
 256
 257       if (!success || verbose >= 3) {
 258          if(verbose < 1)
 259             dump_conv_types(stderr, src_type, dst_type);
 260          if (success) {
 261             fprintf(stderr, "PASS\n");
 262          }
 263          else {
 264             fprintf(stderr, "MISMATCH\n");
 265          }
 266
 267          for(j = 0; j < num_srcs; ++j) {
 268             fprintf(stderr, "  Src%u: ", j);
 269             dump_vec(stderr, src_type, src + j*src_stride);
 270             fprintf(stderr, "\n");
 271          }
 272
 273 #if 1
 274          fprintf(stderr, "  Ref: ");
 275          for(j = 0; j < src_type.length*num_srcs; ++j)
 276             fprintf(stderr, " %f", fref[j]);
 277          fprintf(stderr, "\n");
 278 #endif
 279
 280          for(j = 0; j < num_dsts; ++j) {
 281             fprintf(stderr, "  Dst%u: ", j);
 282             dump_vec(stderr, dst_type, dst + j*dst_stride);
 283             fprintf(stderr, "\n");
 284
 285             fprintf(stderr, "  Ref%u: ", j);
 286             dump_vec(stderr, dst_type, ref + j*dst_stride);
 287             fprintf(stderr, "\n");
 288          }
 289       }
 290    }
 291
 292    /*
 293     * Unfortunately the output of cycle counter is not very reliable as it comes
 294     * -- sometimes we get outliers (due IRQs perhaps?) which are
 295     * better removed to avoid random or biased data.
 296     */
 297    {
 298       double sum = 0.0, sum2 = 0.0;
 299       double avg, std;
 300       unsigned m;
 301
 302       for(i = 0; i < n; ++i) {
 303          sum += cycles[i];
 304          sum2 += cycles[i]*cycles[i];
 305       }
 306
 307       avg = sum/n;
 308       std = sqrtf((sum2 - n*avg*avg)/n);
 309
 310       m = 0;
 311       sum = 0.0;
 312       for(i = 0; i < n; ++i) {
 313          if(fabs(cycles[i] - avg) <= 4.0*std) {
 314             sum += cycles[i];
 315             ++m;
 316          }
 317       }
 318
 319       cycles_avg = sum/m;
 320
 321    }
 322
 323    if(fp)
 324       write_tsv_row(fp, src_type, dst_type, cycles_avg, success);
 325
 326    gallivm_destroy(gallivm);
 327    LLVMContextDispose(context);
 328
 329    return success;
 330 }
 331
 332
 333 const struct lp_type conv_types[] = {
 334    /* float, fixed,  sign,  norm, width, len */
 335
 336    /* Float */
 337    {   TRUE, FALSE,  TRUE,  TRUE,    32,   4 },
 338    {   TRUE, FALSE,  TRUE, FALSE,    32,   4 },
 339    {   TRUE, FALSE, FALSE,  TRUE,    32,   4 },
 340    {   TRUE, FALSE, FALSE, FALSE,    32,   4 },
 341
 342    {   TRUE, FALSE,  TRUE,  TRUE,    32,   8 },
 343    {   TRUE, FALSE,  TRUE, FALSE,    32,   8 },
 344    {   TRUE, FALSE, FALSE,  TRUE,    32,   8 },
 345    {   TRUE, FALSE, FALSE, FALSE,    32,   8 },
 346
 347    /* Fixed */
 348    {  FALSE,  TRUE,  TRUE,  TRUE,    32,   4 },
 349    {  FALSE,  TRUE,  TRUE, FALSE,    32,   4 },
 350    {  FALSE,  TRUE, FALSE,  TRUE,    32,   4 },
 351    {  FALSE,  TRUE, FALSE, FALSE,    32,   4 },
 352
 353    {  FALSE,  TRUE,  TRUE,  TRUE,    32,   8 },
 354    {  FALSE,  TRUE,  TRUE, FALSE,    32,   8 },
 355    {  FALSE,  TRUE, FALSE,  TRUE,    32,   8 },
 356    {  FALSE,  TRUE, FALSE, FALSE,    32,   8 },
 357
 358    /* Integer */
 359    {  FALSE, FALSE,  TRUE,  TRUE,    32,   4 },
 360    {  FALSE, FALSE,  TRUE, FALSE,    32,   4 },
 361    {  FALSE, FALSE, FALSE,  TRUE,    32,   4 },
 362    {  FALSE, FALSE, FALSE, FALSE,    32,   4 },
 363
 364    {  FALSE, FALSE,  TRUE,  TRUE,    32,   8 },
 365    {  FALSE, FALSE,  TRUE, FALSE,    32,   8 },
 366    {  FALSE, FALSE, FALSE,  TRUE,    32,   8 },
 367    {  FALSE, FALSE, FALSE, FALSE,    32,   8 },
 368
 369    {  FALSE, FALSE,  TRUE,  TRUE,    16,   8 },
 370    {  FALSE, FALSE,  TRUE, FALSE,    16,   8 },
 371    {  FALSE, FALSE, FALSE,  TRUE,    16,   8 },
 372    {  FALSE, FALSE, FALSE, FALSE,    16,   8 },
 373
 374    {  FALSE, FALSE,  TRUE,  TRUE,     8,  16 },
 375    {  FALSE, FALSE,  TRUE, FALSE,     8,  16 },
 376    {  FALSE, FALSE, FALSE,  TRUE,     8,  16 },
 377    {  FALSE, FALSE, FALSE, FALSE,     8,  16 },
 378
 379    {  FALSE, FALSE,  TRUE,  TRUE,     8,   4 },
 380    {  FALSE, FALSE,  TRUE, FALSE,     8,   4 },
 381    {  FALSE, FALSE, FALSE,  TRUE,     8,   4 },
 382    {  FALSE, FALSE, FALSE, FALSE,     8,   4 },
 383
 384    {  FALSE, FALSE,  FALSE,  TRUE,    8,   8 },
 385 };
 386
 387
 388 const unsigned num_types = ARRAY_SIZE(conv_types);
 389
 390
 391 boolean
 392 test_all(unsigned verbose, FILE *fp)
 393 {
 394    const struct lp_type *src_type;
 395    const struct lp_type *dst_type;
 396    boolean success = TRUE;
 397    int error_count = 0;
 398
 399    for(src_type = conv_types; src_type < &conv_types[num_types]; ++src_type) {
 400       for(dst_type = conv_types; dst_type < &conv_types[num_types]; ++dst_type) {
 401
 402          if(src_type == dst_type)
 403             continue;
 404
 405          if(!test_one(verbose, fp, *src_type, *dst_type)){
 406             success = FALSE;
 407             ++error_count;
 408          }
 409       }
 410    }
 411
 412    fprintf(stderr, "%d failures\n", error_count);
 413
 414    return success;
 415 }
 416
 417
 418 boolean
 419 test_some(unsigned verbose, FILE *fp,
 420           unsigned long n)
 421 {
 422    const struct lp_type *src_type;
 423    const struct lp_type *dst_type;
 424    unsigned long i;
 425    boolean success = TRUE;
 426
 427    for(i = 0; i < n; ++i) {
 428       src_type = &conv_types[rand() % num_types];
 429
 430       do {
 431          dst_type = &conv_types[rand() % num_types];
 432       } while (src_type == dst_type || src_type->norm != dst_type->norm);
 433
 434       if(!test_one(verbose, fp, *src_type, *dst_type))
 435         success = FALSE;
 436    }
 437
 438    return success;
 439 }
 440
 441
 442 boolean
 443 test_single(unsigned verbose, FILE *fp)
 444 {
 445    /*    float, fixed,  sign,  norm, width, len */
 446    struct lp_type f32x4_type =
 447       {   TRUE, FALSE,  TRUE,  TRUE,    32,   4 };
 448    struct lp_type ub8x4_type =
 449       {  FALSE, FALSE, FALSE,  TRUE,     8,  16 };
 450
 451    boolean success;
 452
 453    success = test_one(verbose, fp, f32x4_type, ub8x4_type);
 454
 455    return success;
 456 }