Do integer mod via sum-of-digits technique. This benefits the date

code.
Review URL: http://codereview.chromium.org/2876011

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@4964 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

src/arm/codegen-arm.h

@@ -881,6 +881,102 @@ class StringCompareStub: public CodeStub {
 };
 
 
+// This stub can do a fast mod operation without using fp.
+// It is tail called from the GenericBinaryOpStub and it always
+// returns an answer.  It never causes GC so it doesn't need a real frame.
+//
+// The inputs are always positive Smis.  This is never called
+// where the denominator is a power of 2.  We handle that separately.
+//
+// If we consider the denominator as an odd number multiplied by a power of 2,
+// then:
+// * The exponent (power of 2) is in the shift_distance register.
+// * The odd number is in the odd_number register.  It is always in the range
+//   of 3 to 25.
+// * The bits from the numerator that are to be copied to the answer (there are
+//   shift_distance of them) are in the mask_bits register.
+// * The other bits of the numerator have been shifted down and are in the lhs
+//   register.
+class IntegerModStub : public CodeStub {
+ public:
+  IntegerModStub(Register result,
+                 Register shift_distance,
+                 Register odd_number,
+                 Register mask_bits,
+                 Register lhs,
+                 Register scratch)
+      : result_(result),
+        shift_distance_(shift_distance),
+        odd_number_(odd_number),
+        mask_bits_(mask_bits),
+        lhs_(lhs),
+        scratch_(scratch) {
+    // We don't code these in the minor key, so they should always be the same.
+    // We don't really want to fix that since this stub is rather large and we
+    // don't want many copies of it.
+    ASSERT(shift_distance_.is(r9));
+    ASSERT(odd_number_.is(r4));
+    ASSERT(mask_bits_.is(r3));
+    ASSERT(scratch_.is(r5));
+  }
+
+ private:
+  Register result_;
+  Register shift_distance_;
+  Register odd_number_;
+  Register mask_bits_;
+  Register lhs_;
+  Register scratch_;
+
+  // Minor key encoding in 16 bits.
+  class ResultRegisterBits: public BitField<int, 0, 4> {};
+  class LhsRegisterBits: public BitField<int, 4, 4> {};
+
+  Major MajorKey() { return IntegerMod; }
+  int MinorKey() {
+    // Encode the parameters in a unique 16 bit value.
+    return ResultRegisterBits::encode(result_.code())
+           | LhsRegisterBits::encode(lhs_.code());
+  }
+
+  void Generate(MacroAssembler* masm);
+
+  const char* GetName() { return "IntegerModStub"; }
+
+  // Utility functions.
+  void DigitSum(MacroAssembler* masm,
+                Register lhs,
+                int mask,
+                int shift,
+                Label* entry);
+  void DigitSum(MacroAssembler* masm,
+                Register lhs,
+                Register scratch,
+                int mask,
+                int shift1,
+                int shift2,
+                Label* entry);
+  void ModGetInRangeBySubtraction(MacroAssembler* masm,
+                                  Register lhs,
+                                  int shift,
+                                  int rhs);
+  void ModReduce(MacroAssembler* masm,
+                 Register lhs,
+                 int max,
+                 int denominator);
+  void ModAnswer(MacroAssembler* masm,
+                 Register result,
+                 Register shift_distance,
+                 Register mask_bits,
+                 Register sum_of_digits);
+
+
+#ifdef DEBUG
+  void Print() { PrintF("IntegerModStub\n"); }
+#endif
+};
+
+
 // This stub can convert a signed int32 to a heap number (double).  It does
 // not work for int32s that are in Smi range!  No GC occurs during this stub
 // so you don't have to set up the frame.

src/arm/ic-arm.cc

@@ -1722,7 +1722,7 @@ static void ConvertIntToFloat(MacroAssembler* masm,
     // Count leading zeros.
     // Gets the wrong answer for 0, but we already checked for that case above.
     Register zeros = scratch2;
-    __ CountLeadingZeros(ival, scratch1, zeros);
+    __ CountLeadingZeros(zeros, ival, scratch1);
 
     // Compute exponent and or it into the exponent register.
     __ rsb(scratch1,

src/arm/macro-assembler-arm.cc

@@ -1686,14 +1686,19 @@ void MacroAssembler::AllocateHeapNumber(Register result,
 }
 
 
-void MacroAssembler::CountLeadingZeros(Register source,
-                                       Register scratch,
-                                       Register zeros) {
+void MacroAssembler::CountLeadingZeros(Register zeros,   // Answer.
+                                       Register source,  // Input.
+                                       Register scratch) {
+  ASSERT(!zeros.is(source) || !source.is(zeros));
+  ASSERT(!zeros.is(scratch));
+  ASSERT(!scratch.is(ip));
+  ASSERT(!source.is(ip));
+  ASSERT(!zeros.is(ip));
 #ifdef CAN_USE_ARMV5_INSTRUCTIONS
   clz(zeros, source);  // This instruction is only supported after ARM5.
 #else
   mov(zeros, Operand(0));
-  mov(scratch, source);
+  Move(scratch, source);
   // Top 16.
   tst(scratch, Operand(0xffff0000));
   add(zeros, zeros, Operand(16), LeaveCC, eq);

src/arm/macro-assembler-arm.h

@@ -471,10 +471,12 @@ class MacroAssembler: public Assembler {
 
   // Count leading zeros in a 32 bit word.  On ARM5 and later it uses the clz
   // instruction.  On pre-ARM5 hardware this routine gives the wrong answer
-  // for 0 (31 instead of 32).
-  void CountLeadingZeros(Register source,
-                         Register scratch,
-                         Register zeros);
+  // for 0 (31 instead of 32).  Source and scratch can be the same in which case
+  // the source is clobbered.  Source and zeros can also be the same in which
+  // case scratch should be a different register.
+  void CountLeadingZeros(Register zeros,
+                         Register source,
+                         Register scratch);
 
   // ---------------------------------------------------------------------------
   // Runtime calls

src/code-stubs.h

@@ -46,6 +46,7 @@ namespace internal {
   V(RecordWrite)                         \
   V(ConvertToDouble)                     \
   V(WriteInt32ToHeapNumber)              \
+  V(IntegerMod)                          \
   V(StackCheck)                          \
   V(FastNewClosure)                      \
   V(FastNewContext)                      \

test/mjsunit/mod.js

+// Copyright 2010 the V8 project authors. All rights reserved.
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+//       notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+//       copyright notice, this list of conditions and the following
+//       disclaimer in the documentation and/or other materials provided
+//       with the distribution.
+//     * Neither the name of Google Inc. nor the names of its
+//       contributors may be used to endorse or promote products derived
+//       from this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+function foo() {
+  for (var i = 1; i < 100; i++) {
+    var answer = 1;
+    for (var j = 1; j < 100; j++) {
+      if (answer == i) answer = 0;
+      // Positive case.
+      print(j + " % " + i + " = " + answer);
+      m = j % i;
+      assertEquals(answer, m, j + " % " + i);
+      m = j % (-i);
+      assertEquals(answer, m, j + " % -" + i);
+      // Negative case.
+      m = (-j) % i;
+      assertEquals(-answer, m, j + " % " + i);
+      // Check for negative zero.
+      if (answer == 0) assertEquals(-Infinity, 1/m);
+      m = (-j) % (-i);
+      assertEquals(-answer, m, j + " % -" + i);
+      // Check for negative zero.
+      if (answer == 0) assertEquals(-Infinity, 1/m);
+      answer++;
+    }
+  }
+}
+
+foo();