add tests for faster version to Travis; change some tests to accomodate for MP_8BIT; update documentation

Author: czurnieden

.travis.yml

@@ -24,8 +24,12 @@ env:
     BUILDOPTIONS="--test-vs-mtest=333333 --mtest-real-rand"
   - |
     BUILDOPTIONS="--with-low-mp"
+  - |
+    BUILDOPTIONS="--with-low-mp --cflags=-DLTM_USE_FASTER_VERSIONS"
   - |
     BUILDOPTIONS="--with-m64 --with-m32 --with-mx32"
+  - |
+    BUILDOPTIONS="--with-m64 --with-m32 --with-mx32 --cflags=-DLTM_USE_FASTER_VERSIONS"
 
 after_failure:
   - cat test_*.log

bn_mp_n_root_ex.c

@@ -18,7 +18,7 @@
  * Result found such that (c)**b <= a and (c+1)**b > a
  *
  */
-#if ( (defined LTM_USE_FASTER_VERSIONS) || (defined LTM_USE_FASTER_NTH_ROOT))
+#if (( (defined LTM_USE_FASTER_VERSIONS) || (defined LTM_USE_FASTER_NTH_ROOT))  )
 /* TODO: needs a benchmark script and be put in bncore.c */
 static const int NTHROOT_NEWTON_HALLEY_CUTOFF = 100000;
 
@@ -421,7 +421,7 @@ int mp_n_root_ex(const mp_int *a, mp_digit b, mp_int *c, int fast)
    ilog2 = mp_count_bits(a);
 
    if (ilog2 < (int)(b)) {
-      mp_set(c, 1);
+      mp_set(c, 1uL);
       c->sign = neg;
       return MP_OKAY;
    }
@@ -462,7 +462,7 @@ int mp_n_root_ex(const mp_int *a, mp_digit b, mp_int *c, int fast)
    }
    cmp = mp_cmp(c, &A);
    if (cmp == MP_GT) {
-      if ((e = mp_sub_d(c, 1, c)) != MP_OKAY) {
+      if ((e = mp_sub_d(c, 1u, c)) != MP_OKAY) {
          goto LTM_ERR;
       }
       for (;;) {
@@ -473,12 +473,12 @@ int mp_n_root_ex(const mp_int *a, mp_digit b, mp_int *c, int fast)
          if (cmp != MP_GT) {
             break;
          }
-         if ((e = mp_sub_d(c, 1, c)) != MP_OKAY) {
+         if ((e = mp_sub_d(c, 1u, c)) != MP_OKAY) {
             goto LTM_ERR;
          }
       }
    } else if (cmp == MP_LT) {
-      if ((e = mp_add_d(c, 1, c)) != MP_OKAY) {
+      if ((e = mp_add_d(c, 1u, c)) != MP_OKAY) {
          goto LTM_ERR;
       }
       for (;;) {
@@ -489,13 +489,13 @@ int mp_n_root_ex(const mp_int *a, mp_digit b, mp_int *c, int fast)
          if (cmp != MP_LT) {
             break;
          }
-         if ((e = mp_add_d(c, 1, c)) != MP_OKAY) {
+         if ((e = mp_add_d(c, 1u, c)) != MP_OKAY) {
             goto LTM_ERR;
          }
       }
       /* Does overshoot in contrast to the other branch above */
       if (cmp != MP_EQ) {
-         if ((e = mp_sub_d(c, 1, c)) != MP_OKAY) {
+         if ((e = mp_sub_d(c, 1u, c)) != MP_OKAY) {
             goto LTM_ERR;
          }
       }
@@ -512,6 +512,7 @@ int mp_n_root_ex(const mp_int *a, mp_digit b, mp_int *c, int fast)
 {
    mp_int  t1, t2, t3, a_;
    int     res;
+   int ilog2;
 
    /* input must be positive if b is even */
    if (((b & 1u) == 0u) && (a->sign == MP_NEG)) {
@@ -534,8 +535,29 @@ int mp_n_root_ex(const mp_int *a, mp_digit b, mp_int *c, int fast)
    a_ = *a;
    a_.sign = MP_ZPOS;
 
+#if 0
    /* t2 = 2 */
    mp_set(&t2, 2uL);
+#endif
+   ilog2 = mp_count_bits(a);
+   if (ilog2 < (int)b) {
+      mp_set(c, 1uL);
+      c->sign = a->sign;
+      res = MP_OKAY;
+      goto LBL_T3;
+   }
+
+   ilog2 = (int)( ( (mp_digit) ilog2 ) / b );
+   if (ilog2 == 0) {
+      mp_set(c, 1uL);
+      c->sign = a->sign;
+      res = MP_OKAY;
+      goto LBL_T3;
+   }
+   /* Start value must be larger than root */
+   if ((  res = mp_2expt(&t2,ilog2 + 2)) != MP_OKAY) {
+      goto LBL_T3;
+   }
 
    do {
       /* t1 = t2 */

demo/test.c

@@ -1230,7 +1230,6 @@ static int test_mp_radix_size(void) {
    mp_init(&a);
 
    mp_read_radix(&a, rand_a,64);
-
    for (i = 2; i < 65; i++) {
       mp_radix_size(&a,i,&len);
       if (len < length_rand[i]) {
@@ -1252,25 +1251,39 @@ static int test_mp_radix_size(void) {
    return EXIT_FAILURE;
 }
 
-#if ( (defined LTM_USE_FASTER_VERSIONS) || (defined LTM_USE_FASTER_NTH_ROOT))
+#if ( (defined LTM_USE_FASTER_VERSIONS) || (defined LTM_USE_FASTER_NTH_ROOT) ) 
 static int test_mp_n_root(void) {
    mp_int a, tmp, root;
    mp_digit b;
 
    mp_init_multi(&a, &tmp, &root, NULL);
-   /* a = 12345^{255} + 12345^{222} + 12345^{123} + 1 */
+
+#ifdef MP_8BIT
+   /* a bit smaller, MP_MASK = 127 for MP_8BIT */
+   mp_set_int(&a, 12345ul);
+   mp_expt_d(&a, 126u, &tmp);
+   mp_expt_d(&a, 126u, &root);
+   mp_mul(&tmp, &root, &tmp);
+   mp_mul(&a, &tmp, &tmp);
+   mp_expt_d(&a, (mp_digit)123u, &root);
+   mp_add(&tmp, &root, &a);
+   mp_add_d(&a, 1u, &a);
+#else
+   /* a = 12345^{255} + 12345^{222} + 12345^{123} + 1, 3465 bits */
    mp_set_int(&a, 12345ul);
    mp_expt_d(&a, (mp_digit)255u, &tmp);
    mp_expt_d(&a, (mp_digit)222u, &root);
    mp_add(&tmp, &root, &tmp);
    mp_expt_d(&a, (mp_digit)123u, &root);
    mp_add(&tmp, &root, &a);
    mp_add_d(&a, 1u, &a);
+#endif
 
-#if (defined MP_8BIT)
-   for (b = 2u; b < 255u, b++) {
+#ifdef MP_8BIT
+   for (b = 2u; b < MP_MASK; b++) {
 #else
    for (b = 2u; b < 3466u; b++) {
+#endif
       printf("Testing nth-root, large input with index %5u \r",(unsigned)b);
       fflush(stdout);
       mp_n_root(&a, b, &root);
@@ -1286,8 +1299,6 @@ static int test_mp_n_root(void) {
          goto LBL_ERR;
       }
    }
-#endif
-
    mp_clear_multi(&a, &tmp, &root, NULL);
    return EXIT_SUCCESS;
 LBL_ERR:
@@ -1300,13 +1311,16 @@ static int test_mp_n_root(void) {
    mp_digit b;
 
    mp_init_multi(&a, &tmp, &root, NULL);
-   /* a = 123^{123}} + 1 */
+   /* a = 123^{123}} + 1, 853 bits */
    mp_set_int(&a, 123ul);
    mp_expt_d(&a, (mp_digit)123u, &a);
    mp_add_d(&a, 1u, &a);
-
-   /* even 21 is much! */
-   for (b = 2u; b < 21u; b++) {
+   /* MP_8BIT is really slow here! */
+#ifdef MP_8BIT
+   for (b = 2u; b < 10u; b++) {
+#else
+   for (b = 2u; b < 834u; b++) {
+#endif
       printf("Testing nth-root, small input with index %2u \r",(unsigned)b);
       fflush(stdout);
       mp_n_root(&a, b, &root);
@@ -1363,7 +1377,7 @@ int unit_tests(void) {
       T(mp_tc_or),
       T(mp_tc_xor),
       T(mp_radix_size),
-      T(mp_n_root),
+      T(mp_n_root)
 #undef T
    };
    unsigned long i;

doc/bn.tex

@@ -311,6 +311,7 @@ \subsection{Build Tweaks}\label{ssec:buildtweaks}
 \addtocounter{footnote}{1}
 \footnotetext{6248 bytes instead of 2736 bytes with GCC 4.8 on a 64-bit AMD A8-6600K}
 
+Note that not all of the faster versions are compatible with \texttt{MP\_8BIT}!
 
 \subsection{Build Trims}
 A trim is a manner of removing functionality from a function that is not required.  For instance, to perform

testme.sh

@@ -90,10 +90,16 @@ _make()
 _runtest()
 {
   make clean > /dev/null
-  _make "$1" "$2" "test_standalone"
+  if (echo $CFLAGS | grep -q FASTER ); then
+     echo -e "\rRun test $1 $2 faster_test"
+    _make "$1" "$2" "faster_test"
+  else
+    echo -e "\rRun test $1 $2 test_standalone"
+    _make "$1" "$2" "test_standalone"
+  fi
   local _timeout=""
   which timeout >/dev/null && _timeout="timeout --foreground 90"
-  echo -e "\rRun test $1 $2"
+
   $_timeout ./test > test_${suffix}.log || _die "running tests" $?
 }
 
@@ -107,9 +113,9 @@ _exit()
 {
   if [ "$ret" == "0" ]
   then
-    echo "Tests successful"
+    echo -e "\nTests successful"
   else
-    echo "$ret tests timed out"
+    echo -e "\n$ret tests timed out"
   fi
 
   exit $ret
@@ -173,7 +179,7 @@ then
 elif [[ "$COMPILERS" == "" ]]
 then
   _banner gcc
-  _runtest "gcc" ""
+  _runtest "gcc" "$CFLAGS"
   _exit
 fi