bigdecimal 3.1.8 → 4.1.2

data/ext/bigdecimal/bigdecimal.h

@@ -17,24 +17,15 @@
 # include <float.h>
 #endif
 
-#ifdef HAVE_INT64_T
-# define DECDIG uint32_t
-# define DECDIG_DBL uint64_t
-# define DECDIG_DBL_SIGNED int64_t
-# define SIZEOF_DECDIG 4
-# define PRI_DECDIG_PREFIX ""
-# ifdef PRI_LL_PREFIX
-#  define PRI_DECDIG_DBL_PREFIX PRI_LL_PREFIX
-# else
-#  define PRI_DECDIG_DBL_PREFIX "l"
-# endif
+#define DECDIG uint32_t
+#define DECDIG_DBL uint64_t
+#define DECDIG_DBL_SIGNED int64_t
+#define SIZEOF_DECDIG 4
+#define PRI_DECDIG_PREFIX ""
+#ifdef PRI_LL_PREFIX
+# define PRI_DECDIG_DBL_PREFIX PRI_LL_PREFIX
 #else
-# define DECDIG uint16_t
-# define DECDIG_DBL uint32_t
-# define DECDIG_DBL_SIGNED int32_t
-# define SIZEOF_DECDIG 2
-# define PRI_DECDIG_PREFIX "h"
-# define PRI_DECDIG_DBL_PREFIX ""
+# define PRI_DECDIG_DBL_PREFIX "l"
 #endif
 
 #define PRIdDECDIG PRI_DECDIG_PREFIX"d"
@@ -51,31 +42,15 @@
 #define PRIxDECDIG_DBL PRI_DECDIG_DBL_PREFIX"x"
 #define PRIXDECDIG_DBL PRI_DECDIG_DBL_PREFIX"X"
 
-#if SIZEOF_DECDIG == 4
-# define BIGDECIMAL_BASE ((DECDIG)1000000000U)
-# define BIGDECIMAL_COMPONENT_FIGURES 9
+#define BIGDECIMAL_BASE ((DECDIG)1000000000U)
+#define BIGDECIMAL_COMPONENT_FIGURES 9
 /*
  * The number of components required for a 64-bit integer.
  *
  *   INT64_MAX:   9_223372036_854775807
  *   UINT64_MAX: 18_446744073_709551615
  */
-# define BIGDECIMAL_INT64_MAX_LENGTH 3
-
-#elif SIZEOF_DECDIG == 2
-# define BIGDECIMAL_BASE ((DECDIG)10000U)
-# define BIGDECIMAL_COMPONENT_FIGURES 4
-/*
- * The number of components required for a 64-bit integer.
- *
- *   INT64_MAX:   922_3372_0368_5477_5807
- *   UINT64_MAX: 1844_6744_0737_0955_1615
- */
-# define BIGDECIMAL_INT64_MAX_LENGTH 5
-
-#else
-# error Unknown size of DECDIG
-#endif
+#define BIGDECIMAL_INT64_MAX_LENGTH 3
 
 #define BIGDECIMAL_DOUBLE_FIGURES (1+DBL_DIG)
 
@@ -167,7 +142,6 @@ enum rbd_rounding_mode {
  *  r = 0.xxxxxxxxx *BASE**exponent
  */
 typedef struct {
-    VALUE  obj;     /* Back pointer(VALUE) for Ruby object.     */
     size_t MaxPrec; /* Maximum precision size                   */
                     /* This is the actual size of frac[]        */
                     /*(frac[0] to frac[MaxPrec] are available). */
@@ -189,19 +163,23 @@ typedef struct {
     DECDIG frac[FLEXIBLE_ARRAY_SIZE]; /* Array of fraction part. */
 } Real;
 
+typedef struct {
+    VALUE bigdecimal;
+    Real *real;
+} BDVALUE;
+
+typedef struct {
+    VALUE bigdecimal_or_nil;
+    Real *real_or_null;
+} NULLABLE_BDVALUE;
+
 /*
  *  ------------------
  *   EXPORTables.
  *  ------------------
  */
 
-VP_EXPORT Real *VpNewRbClass(size_t mx, char const *str, VALUE klass, bool strict_p, bool raise_exception);
-
-VP_EXPORT Real *VpCreateRbObject(size_t mx, const char *str, bool raise_exception);
-
 #define VpBaseFig() BIGDECIMAL_COMPONENT_FIGURES
-#define VpDblFig() BIGDECIMAL_DOUBLE_FIGURES
-#define VpBaseVal() BIGDECIMAL_BASE
 
 /* Zero,Inf,NaN (isinf(),isnan() used to check) */
 VP_EXPORT double VpGetDoubleNaN(void);
@@ -211,7 +189,7 @@ VP_EXPORT double VpGetDoubleNegZero(void);
 
 /* These 2 functions added at v1.1.7 */
 VP_EXPORT size_t VpGetPrecLimit(void);
-VP_EXPORT size_t VpSetPrecLimit(size_t n);
+VP_EXPORT void VpSetPrecLimit(size_t n);
 
 /* Round mode */
 VP_EXPORT int            VpIsRoundMode(unsigned short n);
@@ -219,16 +197,14 @@ VP_EXPORT unsigned short VpGetRoundMode(void);
 VP_EXPORT unsigned short VpSetRoundMode(unsigned short n);
 
 VP_EXPORT int VpException(unsigned short f,const char *str,int always);
-#if 0  /* unused */
-VP_EXPORT int VpIsNegDoubleZero(double v);
-#endif
 VP_EXPORT size_t VpNumOfChars(Real *vp,const char *pszFmt);
 VP_EXPORT size_t VpInit(DECDIG BaseVal);
-VP_EXPORT Real *VpAlloc(size_t mx, const char *szVal, int strict_p, int exc);
+VP_EXPORT NULLABLE_BDVALUE VpAlloc(const char *szVal, int strict_p, int exc);
 VP_EXPORT size_t VpAsgn(Real *c, Real *a, int isw);
 VP_EXPORT size_t VpAddSub(Real *c,Real *a,Real *b,int operation);
 VP_EXPORT size_t VpMult(Real *c,Real *a,Real *b);
 VP_EXPORT size_t VpDivd(Real *c,Real *r,Real *a,Real *b);
+VP_EXPORT int VpNmlz(Real *a);
 VP_EXPORT int VpComp(Real *a,Real *b);
 VP_EXPORT ssize_t VpExponent10(Real *a);
 VP_EXPORT void VpSzMantissa(Real *a, char *buf, size_t bufsize);
@@ -237,21 +213,35 @@ VP_EXPORT void VpToString(Real *a, char *buf, size_t bufsize, size_t fFmt, int f
 VP_EXPORT void VpToFString(Real *a, char *buf, size_t bufsize, size_t fFmt, int fPlus);
 VP_EXPORT int VpCtoV(Real *a, const char *int_chr, size_t ni, const char *frac, size_t nf, const char *exp_chr, size_t ne);
 VP_EXPORT int VpVtoD(double *d, SIGNED_VALUE *e, Real *m);
-VP_EXPORT void VpDtoV(Real *m,double d);
-#if 0  /* unused */
-VP_EXPORT void VpItoV(Real *m,S_INT ival);
-#endif
-VP_EXPORT int VpSqrt(Real *y,Real *x);
 VP_EXPORT int VpActiveRound(Real *y, Real *x, unsigned short f, ssize_t il);
 VP_EXPORT int VpMidRound(Real *y, unsigned short f, ssize_t nf);
 VP_EXPORT int VpLeftRound(Real *y, unsigned short f, ssize_t nf);
 VP_EXPORT void VpFrac(Real *y, Real *x);
-VP_EXPORT int VpPowerByInt(Real *y, Real *x, SIGNED_VALUE n);
-#define VpPower VpPowerByInt
+VP_EXPORT int AddExponent(Real *a, SIGNED_VALUE n);
 
 /* VP constants */
 VP_EXPORT Real *VpOne(void);
 
+/*
+ *  **** BigDecimal part ****
+ */
+VP_EXPORT VALUE BigDecimal_lt(VALUE self, VALUE r);
+VP_EXPORT VALUE BigDecimal_ge(VALUE self, VALUE r);
+VP_EXPORT VALUE BigDecimal_exponent(VALUE self);
+VP_EXPORT VALUE BigDecimal_fix(VALUE self);
+VP_EXPORT VALUE BigDecimal_frac(VALUE self);
+VP_EXPORT VALUE BigDecimal_add(VALUE self, VALUE b);
+VP_EXPORT VALUE BigDecimal_sub(VALUE self, VALUE b);
+VP_EXPORT VALUE BigDecimal_mult(VALUE self, VALUE b);
+VP_EXPORT VALUE BigDecimal_add2(VALUE self, VALUE b, VALUE n);
+VP_EXPORT VALUE BigDecimal_sub2(VALUE self, VALUE b, VALUE n);
+VP_EXPORT VALUE BigDecimal_mult2(VALUE self, VALUE b, VALUE n);
+VP_EXPORT VALUE BigDecimal_split(VALUE self);
+VP_EXPORT VALUE BigDecimal_decimal_shift(VALUE self, VALUE v);
+VP_EXPORT inline BDVALUE GetBDValueMust(VALUE v);
+VP_EXPORT inline BDVALUE rbd_allocate_struct_zero_wrap(int sign, size_t const digits);
+#define NewZeroWrap rbd_allocate_struct_zero_wrap
+
 /*
  *  ------------------
  *  MACRO definitions.
@@ -261,10 +251,6 @@ VP_EXPORT Real *VpOne(void);
 #define Max(a, b)  (((a)>(b))?(a):(b))
 #define Min(a, b)  (((a)>(b))?(b):(a))
 
-#define VpMaxPrec(a)   ((a)->MaxPrec)
-#define VpPrec(a)      ((a)->Prec)
-#define VpGetFlag(a)   ((a)->flag)
-
 /* Sign */
 
 /* VpGetSign(a) returns 1,-1 if a>0,a<0 respectively */
@@ -299,7 +285,6 @@ VP_EXPORT Real *VpOne(void);
 #define VpSetInf(a,s)   (void)(((s)>0)?VpSetPosInf(a):VpSetNegInf(a))
 #define VpHasVal(a)     (a->frac[0])
 #define VpIsOne(a)      ((a->Prec==1)&&(a->frac[0]==1)&&(a->exponent==1))
-#define VpExponent(a)   (a->exponent)
 #ifdef BIGDECIMAL_DEBUG
 int VpVarCheck(Real * v);
 #endif /* BIGDECIMAL_DEBUG */

data/ext/bigdecimal/bits.h

@@ -26,6 +26,9 @@
       ((b) > 0 ? (max) / (a) < (b) : (min) / (a) > (b)) : \
       ((b) > 0 ? (min) / (a) < (b) : (max) / (a) > (b)))
 
+#define ADD_OVERFLOW_SIGNED_INTEGER_P(a, b, min, max) ( \
+    ((a) > 0) == ((b) > 0) && ((a) > 0 ? (max) - (a) < (b) : (min) - (a) > (b)))
+
 #ifdef HAVE_UINT128_T
 # define bit_length(x) \
     (unsigned int) \

data/ext/bigdecimal/div.h

@@ -0,0 +1,192 @@
+// Calculate the inverse of x using the Newton-Raphson method.
+static VALUE
+newton_raphson_inverse(VALUE x, size_t prec) {
+    BDVALUE bdone = NewZeroWrap(1, 1);
+    VpSetOne(bdone.real);
+    VALUE one = bdone.bigdecimal;
+
+    // Initial approximation in 2 digits
+    BDVALUE bdx = GetBDValueMust(x);
+    BDVALUE inv0 = NewZeroWrap(1, 2 * BIGDECIMAL_COMPONENT_FIGURES);
+    VpSetOne(inv0.real);
+    DECDIG_DBL numerator = (DECDIG_DBL)BIGDECIMAL_BASE * 100;
+    DECDIG_DBL denominator = (DECDIG_DBL)bdx.real->frac[0] * 100 + (DECDIG_DBL)(bdx.real->Prec >= 2 ? bdx.real->frac[1] : 0) * 100 / BIGDECIMAL_BASE;
+    inv0.real->frac[0] = (DECDIG)(numerator / denominator);
+    inv0.real->frac[1] = (DECDIG)((numerator % denominator) * (BIGDECIMAL_BASE / 100) / denominator * 100);
+    inv0.real->Prec = 2;
+    inv0.real->exponent = 1 - bdx.real->exponent;
+    VpNmlz(inv0.real);
+    RB_GC_GUARD(bdx.bigdecimal);
+    VALUE inv = inv0.bigdecimal;
+
+    int bl = 1;
+    while (((size_t)1 << bl) < prec) bl++;
+
+    for (int i = bl; i >= 0; i--) {
+        size_t n = (prec >> i) + 2;
+        if (n > prec) n = prec;
+        // Newton-Raphson iteration: inv_next = inv + inv * (1 - x * inv)
+        VALUE one_minus_x_inv = BigDecimal_sub2(
+            one,
+            BigDecimal_mult(BigDecimal_mult2(x, one, SIZET2NUM(n + 1)), inv),
+            SIZET2NUM(SIZET2NUM(n / 2))
+        );
+        inv = BigDecimal_add2(
+            inv,
+            BigDecimal_mult(inv, one_minus_x_inv),
+            SIZET2NUM(n)
+        );
+    }
+    return inv;
+}
+
+// Calculates divmod by multiplying approximate reciprocal of y
+static void
+divmod_by_inv_mul(VALUE x, VALUE y, VALUE inv, VALUE *res_div, VALUE *res_mod) {
+    VALUE div = BigDecimal_fix(BigDecimal_mult(x, inv));
+    VALUE mod = BigDecimal_sub(x, BigDecimal_mult(div, y));
+    while (RTEST(BigDecimal_lt(mod, INT2FIX(0)))) {
+        mod = BigDecimal_add(mod, y);
+        div = BigDecimal_sub(div, INT2FIX(1));
+    }
+    while (RTEST(BigDecimal_ge(mod, y))) {
+        mod = BigDecimal_sub(mod, y);
+        div = BigDecimal_add(div, INT2FIX(1));
+    }
+    *res_div = div;
+    *res_mod = mod;
+}
+
+static void
+slice_copy(DECDIG *dest, Real *src, size_t rshift, size_t length) {
+    ssize_t start = src->exponent - (ssize_t)rshift - (ssize_t)length;
+    if (start >= (ssize_t)src->Prec) return;
+    if (start < 0) {
+        dest -= start;
+        length -= (size_t)(-start);
+        start = 0;
+    }
+    size_t max_length = (size_t)((ssize_t)src->Prec - start);
+    memcpy(dest, src->frac + start, Min(length, max_length) * sizeof(DECDIG));
+}
+
+/* Calculates divmod using Newton-Raphson method.
+ * x and y must be a BigDecimal representing an integer value.
+ *
+ * To calculate with low cost, we need to split x into blocks and perform divmod for each block.
+ * x_digits = remaining_digits(<= y_digits) + block_digits * num_blocks
+ *
+ * Example:
+ * xxx_xxxxx_xxxxx_xxxxx(18 digits) / yyyyy(5 digits)
+ * remaining_digits = 3, block_digits = 5, num_blocks = 3
+ * repeating xxxxx_xxxxxx.divmod(yyyyy) calculation 3 times.
+ *
+ * In each divmod step, dividend is at most (y_digits + block_digits) digits and divisor is y_digits digits.
+ * Reciprocal of y needs block_digits + 1 precision.
+ */
+static void
+divmod_newton(VALUE x, VALUE y, VALUE *div_out, VALUE *mod_out) {
+    size_t x_digits = NUM2SIZET(BigDecimal_exponent(x));
+    size_t y_digits = NUM2SIZET(BigDecimal_exponent(y));
+    if (x_digits <= y_digits) x_digits = y_digits + 1;
+
+    size_t n = x_digits / y_digits;
+    size_t block_figs = (x_digits - y_digits) / n / BIGDECIMAL_COMPONENT_FIGURES + 1;
+    size_t block_digits = block_figs * BIGDECIMAL_COMPONENT_FIGURES;
+    size_t num_blocks = (x_digits - y_digits + block_digits - 1) / block_digits;
+    size_t y_figs = (y_digits - 1) / BIGDECIMAL_COMPONENT_FIGURES + 1;
+    VALUE yinv = newton_raphson_inverse(y, block_digits + 1);
+
+    BDVALUE divident = NewZeroWrap(1, BIGDECIMAL_COMPONENT_FIGURES * (y_figs + block_figs));
+    BDVALUE div_result = NewZeroWrap(1, BIGDECIMAL_COMPONENT_FIGURES * (num_blocks * block_figs + 1));
+    BDVALUE bdx = GetBDValueMust(x);
+
+    VALUE mod = BigDecimal_fix(BigDecimal_decimal_shift(x, SSIZET2NUM(-(ssize_t)(num_blocks * block_digits))));
+    for (ssize_t i = (ssize_t)(num_blocks - 1); i >= 0; i--) {
+        memset(divident.real->frac, 0, (y_figs + block_figs) * sizeof(DECDIG));
+
+        BDVALUE bdmod = GetBDValueMust(mod);
+        slice_copy(divident.real->frac, bdmod.real, 0, y_figs);
+        slice_copy(divident.real->frac + y_figs, bdx.real, (size_t)i * block_figs, block_figs);
+        RB_GC_GUARD(bdmod.bigdecimal);
+
+        VpSetSign(divident.real, 1);
+        divident.real->exponent = (ssize_t)(y_figs + block_figs);
+        divident.real->Prec = y_figs + block_figs;
+        VpNmlz(divident.real);
+
+        VALUE div;
+        divmod_by_inv_mul(divident.bigdecimal, y, yinv, &div, &mod);
+        BDVALUE bddiv = GetBDValueMust(div);
+        slice_copy(div_result.real->frac + (num_blocks - (size_t)i - 1) * block_figs, bddiv.real, 0, block_figs + 1);
+        RB_GC_GUARD(bddiv.bigdecimal);
+    }
+    VpSetSign(div_result.real, 1);
+    div_result.real->exponent = (ssize_t)(num_blocks * block_figs + 1);
+    div_result.real->Prec = num_blocks * block_figs + 1;
+    VpNmlz(div_result.real);
+    RB_GC_GUARD(bdx.bigdecimal);
+    RB_GC_GUARD(divident.bigdecimal);
+    RB_GC_GUARD(div_result.bigdecimal);
+    *div_out = div_result.bigdecimal;
+    *mod_out = mod;
+}
+
+static VALUE
+VpDivdNewtonInner(VALUE args_ptr)
+{
+    Real **args = (Real**)args_ptr;
+    Real *c = args[0], *r = args[1], *a = args[2], *b = args[3];
+    BDVALUE a2, b2, c2, r2;
+    VALUE div, mod, a2_frac = Qnil;
+    size_t div_prec = c->MaxPrec - 1;
+    size_t base_prec = b->Prec;
+
+    a2 = NewZeroWrap(1, a->Prec * BIGDECIMAL_COMPONENT_FIGURES);
+    b2 = NewZeroWrap(1, b->Prec * BIGDECIMAL_COMPONENT_FIGURES);
+    VpAsgn(a2.real, a, 1);
+    VpAsgn(b2.real, b, 1);
+    VpSetSign(a2.real, 1);
+    VpSetSign(b2.real, 1);
+    a2.real->exponent = (ssize_t)(base_prec + div_prec);
+    b2.real->exponent = (ssize_t)base_prec;
+
+    if ((ssize_t)a2.real->Prec > a2.real->exponent) {
+        a2_frac = BigDecimal_frac(a2.bigdecimal);
+        VpMidRound(a2.real, VP_ROUND_DOWN, 0);
+    }
+    divmod_newton(a2.bigdecimal, b2.bigdecimal, &div, &mod);
+    if (a2_frac != Qnil) mod = BigDecimal_add(mod, a2_frac);
+
+    c2 = GetBDValueMust(div);
+    r2 = GetBDValueMust(mod);
+    VpAsgn(c, c2.real, VpGetSign(a) * VpGetSign(b));
+    VpAsgn(r, r2.real, VpGetSign(a));
+    AddExponent(c, a->exponent);
+    AddExponent(c, -b->exponent);
+    AddExponent(c, -(ssize_t)div_prec);
+    AddExponent(r, a->exponent);
+    AddExponent(r, -(ssize_t)(base_prec + div_prec));
+    RB_GC_GUARD(a2.bigdecimal);
+    RB_GC_GUARD(a2.bigdecimal);
+    RB_GC_GUARD(c2.bigdecimal);
+    RB_GC_GUARD(r2.bigdecimal);
+    return Qnil;
+}
+
+static VALUE
+ensure_restore_prec_limit(VALUE limit)
+{
+    VpSetPrecLimit(NUM2SIZET(limit));
+    return Qnil;
+}
+
+static void
+VpDivdNewton(Real *c, Real *r, Real *a, Real *b)
+{
+    Real *args[4] = {c, r, a, b};
+    size_t pl = VpGetPrecLimit();
+    VpSetPrecLimit(0);
+    // Ensure restoring prec limit because some methods used in VpDivdNewtonInner may raise an exception
+    rb_ensure(VpDivdNewtonInner, (VALUE)args, ensure_restore_prec_limit, SIZET2NUM(pl));
+}

data/ext/bigdecimal/extconf.rb

@@ -1,4 +1,4 @@
-# frozen_string_literal: false
+# frozen_string_literal: true
 require 'mkmf'
 
 def have_builtin_func(name, check_expr, opt = "", &b)
@@ -24,39 +24,40 @@ have_header("math.h")
 have_header("stdbool.h")
 have_header("stdlib.h")
 
-have_header("x86intrin.h")
-have_func("_lzcnt_u32", "x86intrin.h")
-have_func("_lzcnt_u64", "x86intrin.h")
-
-have_header("intrin.h")
-have_func("__lzcnt", "intrin.h")
-have_func("__lzcnt64", "intrin.h")
-have_func("_BitScanReverse", "intrin.h")
-have_func("_BitScanReverse64", "intrin.h")
+if have_header("x86intrin.h")
+  have_func("_lzcnt_u32", "x86intrin.h")
+  have_func("_lzcnt_u64", "x86intrin.h")
+end
 
-have_func("labs", "stdlib.h")
-have_func("llabs", "stdlib.h")
-have_func("finite", "math.h")
-have_func("isfinite", "math.h")
+if have_header("intrin.h")
+  have_func("__lzcnt", "intrin.h")
+  have_func("__lzcnt64", "intrin.h")
+  have_func("_BitScanReverse", "intrin.h")
+  have_func("_BitScanReverse64", "intrin.h")
+end
 
 have_header("ruby/atomic.h")
 have_header("ruby/internal/has/builtin.h")
 have_header("ruby/internal/static_assert.h")
 
-have_func("rb_rational_num", "ruby.h")
-have_func("rb_rational_den", "ruby.h")
 have_func("rb_complex_real", "ruby.h")
 have_func("rb_complex_imag", "ruby.h")
 have_func("rb_opts_exception_p", "ruby.h")
 have_func("rb_category_warn", "ruby.h")
 have_const("RB_WARN_CATEGORY_DEPRECATED", "ruby.h")
 
+if RUBY_ENGINE == "ruby"
+  have_const("RUBY_TYPED_EMBEDDABLE", "ruby.h") # RUBY_VERSION >= 3.3
+end
+
 if File.file?(File.expand_path('../lib/bigdecimal.rb', __FILE__))
   bigdecimal_rb = "$(srcdir)/lib/bigdecimal.rb"
 else
   bigdecimal_rb = "$(srcdir)/../../lib/bigdecimal.rb"
 end
 
+$defs.push '-DBIGDECIMAL_USE_VP_TEST_METHODS' if ENV['BIGDECIMAL_USE_VP_TEST_METHODS'] == 'true'
+
 create_makefile('bigdecimal') {|mf|
   mf << "BIGDECIMAL_RB = #{bigdecimal_rb}\n"
 }