tests/tcg/test-i386-fprem.c

   1 /*
   2  *  x86 FPREM test - executes the FPREM and FPREM1 instructions with corner case
   3  *  operands and prints the operands, result and FPU status word.
   4  *
   5  *  Run this on real hardware, then under QEMU, and diff the outputs, to compare
   6  *  QEMU's implementation to your hardware. The 'run-test-i386-fprem' make
   7  *  target does this.
   8  *
   9  *  Copyright (c) 2003 Fabrice Bellard
  10  *  Copyright (c) 2012 Catalin Patulea
  11  *
  12  *  This program is free software; you can redistribute it and/or modify
  13  *  it under the terms of the GNU General Public License as published by
  14  *  the Free Software Foundation; either version 2 of the License, or
  15  *  (at your option) any later version.
  16  *
  17  *  This program is distributed in the hope that it will be useful,
  18  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  19  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  20  *  GNU General Public License for more details.
  21  *
  22  *  You should have received a copy of the GNU General Public License
  23  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
  24  */
  25 #include "compiler.h"
  26 #include "osdep.h"
  27 #include <stdio.h>
  28 #include <inttypes.h>
  29
  30 /*
  31  * Inspired by <ieee754.h>'s union ieee854_long_double, but with single
  32  * long long mantissa fields and assuming little-endianness for simplicity.
  33  */
  34 union float80u {
  35     long double d;
  36
  37     /* This is the IEEE 854 double-extended-precision format.  */
  38     struct {
  39         unsigned long long mantissa:63;
  40         unsigned int one:1;
  41         unsigned int exponent:15;
  42         unsigned int negative:1;
  43         unsigned int empty:16;
  44     } QEMU_PACKED ieee;
  45
  46     /* This is for NaNs in the IEEE 854 double-extended-precision format.  */
  47     struct {
  48         unsigned long long mantissa:62;
  49         unsigned int quiet_nan:1;
  50         unsigned int one:1;
  51         unsigned int exponent:15;
  52         unsigned int negative:1;
  53         unsigned int empty:16;
  54     } QEMU_PACKED ieee_nan;
  55 };
  56
  57 #define IEEE854_LONG_DOUBLE_BIAS 0x3fff
  58
  59 static const union float80u q_nan = {
  60     .ieee_nan.negative = 0,  /* X */
  61     .ieee_nan.exponent = 0x7fff,
  62     .ieee_nan.one = 1,
  63     .ieee_nan.quiet_nan = 1,
  64     .ieee_nan.mantissa = 0,
  65 };
  66
  67 static const union float80u s_nan = {
  68     .ieee_nan.negative = 0,  /* X */
  69     .ieee_nan.exponent = 0x7fff,
  70     .ieee_nan.one = 1,
  71     .ieee_nan.quiet_nan = 0,
  72     .ieee_nan.mantissa = 1,  /* nonzero */
  73 };
  74
  75 static const union float80u pos_inf = {
  76     .ieee.negative = 0,
  77     .ieee.exponent = 0x7fff,
  78     .ieee.one = 1,
  79     .ieee.mantissa = 0,
  80 };
  81
  82 static const union float80u pseudo_pos_inf = {  /* "unsupported" */
  83     .ieee.negative = 0,
  84     .ieee.exponent = 0x7fff,
  85     .ieee.one = 0,
  86     .ieee.mantissa = 0,
  87 };
  88
  89 static const union float80u pos_denorm = {
  90     .ieee.negative = 0,
  91     .ieee.exponent = 0,
  92     .ieee.one = 0,
  93     .ieee.mantissa = 1,
  94 };
  95
  96 static const union float80u smallest_positive_norm = {
  97     .ieee.negative = 0,
  98     .ieee.exponent = 1,
  99     .ieee.one = 1,
 100     .ieee.mantissa = 0,
 101 };
 102
 103 static void fninit()
 104 {
 105     asm volatile ("fninit\n");
 106 }
 107
 108 static long double fprem(long double a, long double b, uint16_t *sw)
 109 {
 110     long double result;
 111     asm volatile ("fprem\n"
 112                   "fnstsw %1\n"
 113                   : "=t" (result), "=m" (*sw)
 114                   : "0" (a), "u" (b)
 115                   : "st(1)");
 116     return result;
 117 }
 118
 119 static long double fprem1(long double a, long double b, uint16_t *sw)
 120 {
 121     long double result;
 122     asm volatile ("fprem1\n"
 123                   "fnstsw %1\n"
 124                   : "=t" (result), "=m" (*sw)
 125                   : "0" (a), "u" (b)
 126                   : "st(1)");
 127     return result;
 128 }
 129
 130 #define FPUS_IE (1 << 0)
 131 #define FPUS_DE (1 << 1)
 132 #define FPUS_ZE (1 << 2)
 133 #define FPUS_OE (1 << 3)
 134 #define FPUS_UE (1 << 4)
 135 #define FPUS_PE (1 << 5)
 136 #define FPUS_SF (1 << 6)
 137 #define FPUS_SE (1 << 7)
 138 #define FPUS_C0 (1 << 8)
 139 #define FPUS_C1 (1 << 9)
 140 #define FPUS_C2 (1 << 10)
 141 #define FPUS_TOP 0x3800
 142 #define FPUS_C3 (1 << 14)
 143 #define FPUS_B  (1 << 15)
 144
 145 #define FPUS_EMASK 0x007f
 146
 147 #define FPUC_EM 0x3f
 148
 149 static void psw(uint16_t sw)
 150 {
 151     printf("SW:  C3 TopC2C1C0\n");
 152     printf("SW: %c %d %3d %d %d %d %c %c %c %c %c %c %c %c\n",
 153            sw & FPUS_B ? 'B' : 'b',
 154            !!(sw & FPUS_C3),
 155            (sw & FPUS_TOP) >> 11,
 156            !!(sw & FPUS_C2),
 157            !!(sw & FPUS_C1),
 158            !!(sw & FPUS_C0),
 159            (sw & FPUS_SE) ? 'S' : 's',
 160            (sw & FPUS_SF) ? 'F' : 'f',
 161            (sw & FPUS_PE) ? 'P' : 'p',
 162            (sw & FPUS_UE) ? 'U' : 'u',
 163            (sw & FPUS_OE) ? 'O' : 'o',
 164            (sw & FPUS_ZE) ? 'Z' : 'z',
 165            (sw & FPUS_DE) ? 'D' : 'd',
 166            (sw & FPUS_IE) ? 'I' : 'i');
 167 }
 168
 169 static void do_fprem(long double a, long double b)
 170 {
 171     const union float80u au = {.d = a};
 172     const union float80u bu = {.d = b};
 173     union float80u ru;
 174     uint16_t sw;
 175
 176     printf("A: S=%d Exp=%04x Int=%d (QNaN=%d) Sig=%016llx (%.06Le)\n",
 177            au.ieee.negative, au.ieee.exponent, au.ieee.one,
 178            au.ieee_nan.quiet_nan, (unsigned long long)au.ieee.mantissa,
 179            a);
 180     printf("B: S=%d Exp=%04x Int=%d (QNaN=%d) Sig=%016llx (%.06Le)\n",
 181            bu.ieee.negative, bu.ieee.exponent, bu.ieee.one,
 182            bu.ieee_nan.quiet_nan, (unsigned long long)bu.ieee.mantissa,
 183            b);
 184     fflush(stdout);
 185
 186     fninit();
 187     ru.d = fprem(a, b, &sw);
 188     psw(sw);
 189
 190     printf("R : S=%d Exp=%04x Int=%d (QNaN=%d) Sig=%016llx (%.06Le)\n",
 191            ru.ieee.negative, ru.ieee.exponent, ru.ieee.one,
 192            ru.ieee_nan.quiet_nan, (unsigned long long)ru.ieee.mantissa,
 193            ru.d);
 194
 195     fninit();
 196     ru.d = fprem1(a, b, &sw);
 197     psw(sw);
 198
 199     printf("R1: S=%d Exp=%04x Int=%d (QNaN=%d) Sig=%016llx (%.06Le)\n",
 200            ru.ieee.negative, ru.ieee.exponent, ru.ieee.one,
 201            ru.ieee_nan.quiet_nan, (unsigned long long)ru.ieee.mantissa,
 202            ru.d);
 203
 204     printf("\n");
 205 }
 206
 207 static void do_fprem_stack_underflow(void)
 208 {
 209     const long double a = 1.0;
 210     union float80u ru;
 211     uint16_t sw;
 212
 213     fninit();
 214     asm volatile ("fprem\n"
 215                   "fnstsw %1\n"
 216                   : "=t" (ru.d), "=m" (sw)
 217                   : "0" (a)
 218                   : "st(1)");
 219     psw(sw);
 220
 221     printf("R: S=%d Exp=%04x Int=%d (QNaN=%d) Sig=%016llx (%.06Le)\n",
 222            ru.ieee.negative, ru.ieee.exponent, ru.ieee.one,
 223            ru.ieee_nan.quiet_nan, (unsigned long long)ru.ieee.mantissa,
 224            ru.d);
 225     printf("\n");
 226 }
 227
 228 static void test_fprem_cases(void)
 229 {
 230     printf("= stack underflow =\n");
 231     do_fprem_stack_underflow();
 232
 233     printf("= invalid operation =\n");
 234     do_fprem(s_nan.d, 1.0);
 235     do_fprem(1.0, 0.0);
 236     do_fprem(pos_inf.d, 1.0);
 237     do_fprem(pseudo_pos_inf.d, 1.0);
 238
 239     printf("= denormal =\n");
 240     do_fprem(pos_denorm.d, 1.0);
 241     do_fprem(1.0, pos_denorm.d);
 242
 243     /* printf("= underflow =\n"); */
 244     /* TODO: Is there a case where FPREM raises underflow? */
 245 }
 246
 247 static void test_fprem_pairs(void)
 248 {
 249     unsigned long long count;
 250
 251     unsigned int negative_index_a = 0;
 252     unsigned int negative_index_b = 0;
 253     static const unsigned int negative_values[] = {
 254         0,
 255         1,
 256     };
 257
 258     unsigned int exponent_index_a = 0;
 259     unsigned int exponent_index_b = 0;
 260     static const unsigned int exponent_values[] = {
 261         0,
 262         1,
 263         2,
 264         IEEE854_LONG_DOUBLE_BIAS - 1,
 265         IEEE854_LONG_DOUBLE_BIAS,
 266         IEEE854_LONG_DOUBLE_BIAS + 1,
 267         0x7ffd,
 268         0x7ffe,
 269         0x7fff,
 270     };
 271
 272     unsigned int one_index_a = 0;
 273     unsigned int one_index_b = 0;
 274     static const unsigned int one_values[] = {
 275         0,
 276         1,
 277     };
 278
 279     unsigned int quiet_nan_index_a = 0;
 280     unsigned int quiet_nan_index_b = 0;
 281     static const unsigned int quiet_nan_values[] = {
 282         0,
 283         1,
 284     };
 285
 286     unsigned int mantissa_index_a = 0;
 287     unsigned int mantissa_index_b = 0;
 288     static const unsigned long long mantissa_values[] = {
 289         0,
 290         1,
 291         2,
 292         0x3ffffffffffffffdULL,
 293         0x3ffffffffffffffeULL,
 294         0x3fffffffffffffffULL,
 295     };
 296
 297     for (count = 0; ; ++count) {
 298 #define INIT_FIELD(var, field) \
 299             .ieee_nan.field = field##_values[field##_index_##var]
 300         const union float80u a = {
 301             INIT_FIELD(a, negative),
 302             INIT_FIELD(a, exponent),
 303             INIT_FIELD(a, one),
 304             INIT_FIELD(a, quiet_nan),
 305             INIT_FIELD(a, mantissa),
 306         };
 307         const union float80u b = {
 308             INIT_FIELD(b, negative),
 309             INIT_FIELD(b, exponent),
 310             INIT_FIELD(b, one),
 311             INIT_FIELD(b, quiet_nan),
 312             INIT_FIELD(b, mantissa),
 313         };
 314 #undef INIT_FIELD
 315
 316         do_fprem(a.d, b.d);
 317
 318         int carry = 1;
 319 #define CARRY_INTO(var, field) do { \
 320             if (carry) { \
 321                 if (++field##_index_##var == ARRAY_SIZE(field##_values)) { \
 322                     field##_index_##var = 0; \
 323                 } else { \
 324                     carry = 0; \
 325                 } \
 326             } \
 327         } while (0)
 328         CARRY_INTO(b, mantissa);
 329         CARRY_INTO(b, quiet_nan);
 330         CARRY_INTO(b, one);
 331         CARRY_INTO(b, exponent);
 332         CARRY_INTO(b, negative);
 333         CARRY_INTO(a, mantissa);
 334         CARRY_INTO(a, quiet_nan);
 335         CARRY_INTO(a, one);
 336         CARRY_INTO(a, exponent);
 337         CARRY_INTO(a, negative);
 338 #undef CARRY_INTO
 339
 340         if (carry) {
 341             break;
 342         }
 343     }
 344
 345     fprintf(stderr, "test-i386-fprem: tested %llu cases\n", count);
 346 }
 347
 348 int main(int argc, char **argv)
 349 {
 350     test_fprem_cases();
 351     test_fprem_pairs();
 352     return 0;
 353 }