should not raise an exception, so implement vm_fast_mod(). fix #1471

The modulo between the Fixnum/Float is improved the performance by this patch.

* before
      user     system      total        real
  1.000000   0.000000   1.000000 (  1.000743)
  1.030000   0.000000   1.030000 (  1.030479)
  1.090000   0.000000   1.090000 (  1.097021)

* after
      user     system      total        real
  0.570000   0.000000   0.570000 (  0.577238)
  1.040000   0.000000   1.040000 (  1.031634)
  0.540000   0.000000   0.540000 (  0.539968)

Test Script:
----
require 'benchmark'

Benchmark.bm do |x|
  x.report do
    10_000_000.times do
      42 % 1
    end
  end

  x.report do
    10_000_000.times do
      42 % 1.0
    end
  end

  x.report do
    10_000_000.times do
      42.0 % 1.0
    end
  end
end

compiler.cpp

@@ -172,6 +172,7 @@ RoxorCompiler::RoxorCompiler(bool _debug_mode)
     fastMinusFunc = get_function("vm_fast_minus");
     fastMultFunc = get_function("vm_fast_mult");
     fastDivFunc = get_function("vm_fast_div");
+    fastModFunc = get_function("vm_fast_mod");
     fastLtFunc = get_function("vm_fast_lt");
     fastLeFunc = get_function("vm_fast_le");
     fastGtFunc = get_function("vm_fast_gt");
@@ -2298,7 +2299,7 @@ RoxorCompiler::compile_optimized_dispatch_call(SEL sel, int argc,
 	return pn;
     }
     // Pure arithmetic operations.
-    else if (sel == selPLUS || sel == selMINUS || sel == selDIV 
+    else if (sel == selPLUS || sel == selMINUS || sel == selDIV || sel == selMOD
 	     || sel == selMULT || sel == selLT || sel == selLE 
 	     || sel == selGT || sel == selGE || sel == selEq
 	     || sel == selNeq || sel == selEqq) {
@@ -2320,6 +2321,9 @@ RoxorCompiler::compile_optimized_dispatch_call(SEL sel, int argc,
 	else if (sel == selDIV) {
 	    func = fastDivFunc;
 	}
+	else if (sel == selMOD) {
+	    func = fastModFunc;
+	}
 	else if (sel == selMULT) {
 	    func = fastMultFunc;
 	}

compiler.h

@@ -159,6 +159,7 @@ class RoxorCompiler {
 	Function *fastMinusFunc;
 	Function *fastMultFunc;
 	Function *fastDivFunc;
+	Function *fastModFunc;
 	Function *fastLtFunc;
 	Function *fastLeFunc;
 	Function *fastGtFunc;

exported_symbols_list

@@ -14,6 +14,7 @@ _rstr_concat
 _selPLUS
 _selMINUS
 _selDIV
+_selMOD
 _selMULT
 _selLT
 _selLE

kernel.c

@@ -438,6 +438,64 @@ vm_fast_div(VALUE left, VALUE right, unsigned char overriden)
     return vm_dispatch(0, left, selDIV, NULL, 0, 1, &right);
 }
 
+PRIMITIVE VALUE
+vm_fast_mod(VALUE left, VALUE right, unsigned char overriden)
+{
+    if (overriden == 0 && NUMERIC_IMM_P(left) && NUMERIC_IMM_P(right)) {
+	if (FIXNUM_P(left) && FIXNUM_P(right)) {
+	    const long x = FIX2LONG(left);
+	    const long y = FIX2LONG(right);
+	    long div, mod;
+
+	    if (y == 0) {
+		rb_num_zerodiv();
+	    }
+	    if (y < 0) {
+		if (x < 0) {
+		    div = -x / -y;
+		}
+		else {
+		    div = - (x / -y);
+		}
+	    }
+	    else {
+		if (x < 0) {
+		    div = - (-x / y);
+		}
+		else {
+		    div = x / y;
+		}
+	    }
+	    mod = x - div*y;
+	    if ((mod < 0 && y > 0) || (mod > 0 && y < 0)) {
+		mod += y;
+		div -= 1;
+	    }
+	    return LONG2FIX(mod);
+	}
+	else if (FIXFLOAT_P(left) && FIXFLOAT_P(right)) {
+	    const double x = IMM2DBL(left);
+	    const double y = IMM2DBL(right);
+	    double div, mod;
+
+#ifdef HAVE_FMOD
+	    mod = fmod(x, y);
+#else
+	    double z;
+	    modf(x/y, &z);
+	    mod = x - z * y;
+#endif
+	    div = (x - mod) / y;
+	    if (y * mod < 0) {
+		mod += y;
+		div -= 1.0;
+	    }
+	    return DBL2FIXFLOAT(mod);
+	}
+    }
+    return vm_dispatch(0, left, selMOD, NULL, 0, 1, &right);
+}
+
 PRIMITIVE VALUE
 vm_fast_lt(VALUE left, VALUE right, unsigned char overriden)
 {