ARM improvements to constant div, mod and mul.

* Fast runtime calls for div and mod.
* Fix assembly and disassembly of multiply instructions.
* Strength reduce and inline multiplications to shift-add.
* Strength reduce and inline mod by power of 2.
* Strength reduce mod by other small integers to mul.
* Strength reduce div by 2 and 3.
Review URL: http://codereview.chromium.org/155047

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@2355 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

src/arm/assembler-arm.cc

@@ -837,6 +837,7 @@ void Assembler::mla(Register dst, Register src1, Register src2, Register srcA,
 void Assembler::mul(Register dst, Register src1, Register src2,
                     SBit s, Condition cond) {
   ASSERT(!dst.is(pc) && !src1.is(pc) && !src2.is(pc));
+  // dst goes in bits 16-19 for this instruction!
   emit(cond | s | dst.code()*B16 | src2.code()*B8 | B7 | B4 | src1.code());
 }
 
@@ -888,7 +889,7 @@ void Assembler::umull(Register dstL,
                       Condition cond) {
   ASSERT(!dstL.is(pc) && !dstH.is(pc) && !src1.is(pc) && !src2.is(pc));
   ASSERT(!dstL.is(dstH));
-  emit(cond | B23 | B22 | s | dstH.code()*B16 | dstL.code()*B12 |
+  emit(cond | B23 | s | dstH.code()*B16 | dstL.code()*B12 |
        src2.code()*B8 | B7 | B4 | src1.code());
 }
 

src/arm/codegen-arm.h

@@ -186,6 +186,8 @@ class CodeGenerator: public AstVisitor {
   bool in_spilled_code() const { return in_spilled_code_; }
   void set_in_spilled_code(bool flag) { in_spilled_code_ = flag; }
 
+  static const int kUnknownIntValue = -1;
+
  private:
   // Construction/Destruction
   CodeGenerator(int buffer_size, Handle<Script> script, bool is_eval);
@@ -291,7 +293,9 @@ class CodeGenerator: public AstVisitor {
 
   void ToBoolean(JumpTarget* true_target, JumpTarget* false_target);
 
-  void GenericBinaryOperation(Token::Value op, OverwriteMode overwrite_mode);
+  void GenericBinaryOperation(Token::Value op,
+                              OverwriteMode overwrite_mode,
+                              int known_rhs = kUnknownIntValue);
   void Comparison(Condition cc,
                   Expression* left,
                   Expression* right,

src/arm/disasm-arm.cc

@@ -438,7 +438,19 @@ int Decoder::FormatOption(Instr* instr, const char* format) {
       return 6;
     }
     case 'u': {  // 'u: signed or unsigned multiplies
-      if (instr->Bit(22) == 1) {
+      // The manual gets the meaning of bit 22 backwards in the multiply
+      // instruction overview on page A3.16.2.  The instructions that
+      // exist in u and s variants are the following:
+      // smull A4.1.87
+      // umull A4.1.129
+      // umlal A4.1.128
+      // smlal A4.1.76
+      // For these 0 means u and 1 means s.  As can be seen on their individual
+      // pages.  The other 18 mul instructions have the bit set or unset in
+      // arbitrary ways that are unrelated to the signedness of the instruction.
+      // None of these 18 instructions exist in both a 'u' and an 's' variant.
+
+      if (instr->Bit(22) == 0) {
         Print("u");
       } else {
         Print("s");
@@ -494,11 +506,16 @@ void Decoder::DecodeType01(Instr* instr) {
         // multiply instructions
         if (instr->Bit(23) == 0) {
           if (instr->Bit(21) == 0) {
-            Format(instr, "mul'cond's 'rd, 'rm, 'rs");
+            // Mul calls it Rd.  Everyone else calls it Rn.
+            Format(instr, "mul'cond's 'rn, 'rm, 'rs");
           } else {
-            Format(instr, "mla'cond's 'rd, 'rm, 'rs, 'rn");
+            // In the manual the order is rd, rm, rs, rn.  But mla swaps the
+            // positions of rn and rd in the encoding.
+            Format(instr, "mla'cond's 'rn, 'rm, 'rs, 'rd");
           }
         } else {
+          // In the manual the order is RdHi, RdLo, Rm, Rs.
+          // RdHi is what other instructions call Rn and RdLo is Rd.
           Format(instr, "'um'al'cond's 'rn, 'rd, 'rm, 'rs");
         }
       } else {

src/arm/simulator-arm.cc

@@ -1080,41 +1080,44 @@ void Simulator::DecodeType01(Instr* instr) {
     // multiply instruction or extra loads and stores
     if (instr->Bits(7, 4) == 9) {
       if (instr->Bit(24) == 0) {
-        // multiply instructions
-        int rd = instr->RdField();
+        // Multiply instructions have Rd in a funny place.
+        int rd = instr->RnField();
         int rm = instr->RmField();
         int rs = instr->RsField();
         int32_t rs_val = get_register(rs);
         int32_t rm_val = get_register(rm);
         if (instr->Bit(23) == 0) {
           if (instr->Bit(21) == 0) {
-            // Format(instr, "mul'cond's 'rd, 'rm, 'rs");
+            // Format(instr, "mul'cond's 'rn, 'rm, 'rs");
             int32_t alu_out = rm_val * rs_val;
             set_register(rd, alu_out);
             if (instr->HasS()) {
               SetNZFlags(alu_out);
             }
           } else {
-            Format(instr, "mla'cond's 'rd, 'rm, 'rs, 'rn");
+            UNIMPLEMENTED();  // mla is not used by V8.
           }
         } else {
           // Format(instr, "'um'al'cond's 'rn, 'rd, 'rs, 'rm");
-          int rn = instr->RnField();
+          int rd_lo = instr->RdField();
           int32_t hi_res = 0;
           int32_t lo_res = 0;
-          if (instr->Bit(22) == 0) {
-            // signed multiply
-            UNIMPLEMENTED();
+          if (instr->Bit(22) == 1) {
+            int64_t left_op  = static_cast<int32_t>(rm_val);
+            int64_t right_op = static_cast<int32_t>(rs_val);
+            uint64_t result = left_op * right_op;
+            hi_res = static_cast<int32_t>(result >> 32);
+            lo_res = static_cast<int32_t>(result & 0xffffffff);
           } else {
             // unsigned multiply
-            uint64_t left_op  = rm_val;
-            uint64_t right_op = rs_val;
+            uint64_t left_op  = static_cast<uint32_t>(rm_val);
+            uint64_t right_op = static_cast<uint32_t>(rs_val);
             uint64_t result = left_op * right_op;
             hi_res = static_cast<int32_t>(result >> 32);
             lo_res = static_cast<int32_t>(result & 0xffffffff);
           }
-          set_register(rn, hi_res);
-          set_register(rd, lo_res);
+          set_register(rd_lo, lo_res);
+          set_register(rd, hi_res);
           if (instr->HasS()) {
             UNIMPLEMENTED();
           }

src/assembler.cc

@@ -608,6 +608,16 @@ static double mul_two_doubles(double x, double y) {
 }
 
 
+static double div_two_doubles(double x, double y) {
+  return x / y;
+}
+
+
+static double mod_two_doubles(double x, double y) {
+  return fmod(x, y);
+}
+
+
 static int native_compare_doubles(double x, double y) {
   if (x == y) return 0;
   return x < y ? 1 : -1;
@@ -628,6 +638,12 @@ ExternalReference ExternalReference::double_fp_operation(
     case Token::MUL:
       function = &mul_two_doubles;
       break;
+    case Token::DIV:
+      function = &div_two_doubles;
+      break;
+    case Token::MOD:
+      function = &mod_two_doubles;
+      break;
     default:
       UNREACHABLE();
   }

src/serialize.cc

@@ -712,9 +712,17 @@ void ExternalReferenceTable::PopulateTable() {
       UNCLASSIFIED,
       13,
       "mul_two_doubles");
-  Add(ExternalReference::compare_doubles().address(),
+  Add(ExternalReference::double_fp_operation(Token::DIV).address(),
       UNCLASSIFIED,
       14,
+      "div_two_doubles");
+  Add(ExternalReference::double_fp_operation(Token::MOD).address(),
+      UNCLASSIFIED,
+      15,
+      "mod_two_doubles");
+  Add(ExternalReference::compare_doubles().address(),
+      UNCLASSIFIED,
+      16,
       "compare_doubles");
 }
 

test/mjsunit/div-mod.js

+// Copyright 2009 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.
+
+// Test fast div and mod.
+
+function divmod(div_func, mod_func, x, y) {
+  var div_answer = (div_func)(x);
+  assertEquals(x / y, div_answer, x + "/" + y);
+  var mod_answer = (mod_func)(x);
+  assertEquals(x % y, mod_answer, x + "%" + y);
+  var minus_div_answer = (div_func)(-x);
+  assertEquals(-x / y, minus_div_answer, "-" + x + "/" + y);
+  var minus_mod_answer = (mod_func)(-x);
+  assertEquals(-x % y, minus_mod_answer, "-" + x + "%" + y);
+}
+
+
+function run_tests_for(divisor) {
+  print("(function(left) { return left / " + divisor + "; })");
+  var div_func = this.eval("(function(left) { return left / " + divisor + "; })");
+  var mod_func = this.eval("(function(left) { return left % " + divisor + "; })");
+  var exp;
+  // Strange number test.
+  divmod(div_func, mod_func, 0, divisor);
+  divmod(div_func, mod_func, 1 / 0, divisor);
+  // Floating point number test.
+  for (exp = -1024; exp <= 1024; exp += 4) {
+    divmod(div_func, mod_func, Math.pow(2, exp), divisor);
+    divmod(div_func, mod_func, 0.9999999 * Math.pow(2, exp), divisor);
+    divmod(div_func, mod_func, 1.0000001 * Math.pow(2, exp), divisor);
+  }
+  // Integer number test.
+  for (exp = 0; exp <= 32; exp++) {
+    divmod(div_func, mod_func, 1 << exp, divisor);
+    divmod(div_func, mod_func, (1 << exp) + 1, divisor);
+    divmod(div_func, mod_func, (1 << exp) - 1, divisor);
+  }
+  divmod(div_func, mod_func, Math.floor(0x1fffffff / 3), divisor);
+  divmod(div_func, mod_func, Math.floor(-0x20000000 / 3), divisor);
+}
+
+
+var divisors = [
+  0,
+  1,
+  2,
+  3,
+  4,
+  5,
+  6,
+  7,
+  8,
+  9,
+  10,
+  // These ones in the middle don't add much apart from slowness to the test.
+  0x1000000,
+  0x2000000,
+  0x4000000,
+  0x8000000,
+  0x10000000,
+  0x20000000,
+  0x40000000,
+  12,
+  60,
+  100,
+  1000 * 60 * 60 * 24];
+
+for (var i = 0; i < divisors.length; i++) {
+  run_tests_for(divisors[i]);
+}
+

test/mjsunit/smi-negative-zero.js

@@ -37,64 +37,64 @@ var minus_four = -4;
 
 // variable op variable
 
-assertEquals(one / (-zero), -Infinity);
+assertEquals(one / (-zero), -Infinity, "one / -0 I");
 
-assertEquals(one / (zero * minus_one), -Infinity);
-assertEquals(one / (minus_one * zero), -Infinity);
-assertEquals(one / (zero * zero), Infinity);
-assertEquals(one / (minus_one * minus_one), 1);
+assertEquals(one / (zero * minus_one), -Infinity, "one / -1");
+assertEquals(one / (minus_one * zero), -Infinity, "one / -0 II");
+assertEquals(one / (zero * zero), Infinity, "one / 0 I");
+assertEquals(one / (minus_one * minus_one), 1, "one / 1");
 
-assertEquals(one / (zero / minus_one), -Infinity);
-assertEquals(one / (zero / one), Infinity);
+assertEquals(one / (zero / minus_one), -Infinity, "one / -0 III");
+assertEquals(one / (zero / one), Infinity, "one / 0 II");
 
-assertEquals(one / (minus_four % two), -Infinity);
-assertEquals(one / (minus_four % minus_two), -Infinity);
-assertEquals(one / (four % two), Infinity);
-assertEquals(one / (four % minus_two), Infinity);
+assertEquals(one / (minus_four % two), -Infinity, "foo");
+assertEquals(one / (minus_four % minus_two), -Infinity, "foo");
+assertEquals(one / (four % two), Infinity, "foo");
+assertEquals(one / (four % minus_two), Infinity, "foo");
 
 // literal op variable
 
-assertEquals(one / (0 * minus_one), -Infinity);
-assertEquals(one / (-1 * zero), -Infinity);
-assertEquals(one / (0 * zero), Infinity);
-assertEquals(one / (-1 * minus_one), 1);
+assertEquals(one / (0 * minus_one), -Infinity, "bar");
+assertEquals(one / (-1 * zero), -Infinity, "bar");
+assertEquals(one / (0 * zero), Infinity, "bar");
+assertEquals(one / (-1 * minus_one), 1, "bar");
 
-assertEquals(one / (0 / minus_one), -Infinity);
-assertEquals(one / (0 / one), Infinity);
+assertEquals(one / (0 / minus_one), -Infinity, "baz");
+assertEquals(one / (0 / one), Infinity, "baz");
 
-assertEquals(one / (-4 % two), -Infinity);
-assertEquals(one / (-4 % minus_two), -Infinity);
-assertEquals(one / (4 % two), Infinity);
-assertEquals(one / (4 % minus_two), Infinity);
+assertEquals(one / (-4 % two), -Infinity, "baz");
+assertEquals(one / (-4 % minus_two), -Infinity, "baz");
+assertEquals(one / (4 % two), Infinity, "baz");
+assertEquals(one / (4 % minus_two), Infinity, "baz");
 
 // variable op literal
 
-assertEquals(one / (zero * -1), -Infinity);
-assertEquals(one / (minus_one * 0), -Infinity);
-assertEquals(one / (zero * 0), Infinity);
-assertEquals(one / (minus_one * -1), 1);
+assertEquals(one / (zero * -1), -Infinity, "fizz");
+assertEquals(one / (minus_one * 0), -Infinity, "fizz");
+assertEquals(one / (zero * 0), Infinity, "fizz");
+assertEquals(one / (minus_one * -1), 1, "fizz");
 
-assertEquals(one / (zero / -1), -Infinity);
-assertEquals(one / (zero / 1), Infinity);
+assertEquals(one / (zero / -1), -Infinity, "buzz");
+assertEquals(one / (zero / 1), Infinity, "buzz");
 
-assertEquals(one / (minus_four % 2), -Infinity);
-assertEquals(one / (minus_four % -2), -Infinity);
-assertEquals(one / (four % 2), Infinity);
-assertEquals(one / (four % -2), Infinity);
+assertEquals(one / (minus_four % 2), -Infinity, "buzz");
+assertEquals(one / (minus_four % -2), -Infinity, "buzz");
+assertEquals(one / (four % 2), Infinity, "buzz");
+assertEquals(one / (four % -2), Infinity, "buzz");
 
 // literal op literal
 
-assertEquals(one / (-0), -Infinity);
+assertEquals(one / (-0), -Infinity, "fisk1");
 
-assertEquals(one / (0 * -1), -Infinity);
-assertEquals(one / (-1 * 0), -Infinity);
-assertEquals(one / (0 * 0), Infinity);
-assertEquals(one / (-1 * -1), 1);
+assertEquals(one / (0 * -1), -Infinity, "fisk2");
+assertEquals(one / (-1 * 0), -Infinity, "fisk3");
+assertEquals(one / (0 * 0), Infinity, "fisk4");
+assertEquals(one / (-1 * -1), 1, "fisk5");
 
-assertEquals(one / (0 / -1), -Infinity);
-assertEquals(one / (0 / 1), Infinity);
+assertEquals(one / (0 / -1), -Infinity, "hest");
+assertEquals(one / (0 / 1), Infinity, "hest");
 
-assertEquals(one / (-4 % 2), -Infinity);
-assertEquals(one / (-4 % -2), -Infinity);
-assertEquals(one / (4 % 2), Infinity);
-assertEquals(one / (4 % -2), Infinity);
+assertEquals(one / (-4 % 2), -Infinity, "fiskhest");
+assertEquals(one / (-4 % -2), -Infinity, "fiskhest");
+assertEquals(one / (4 % 2), Infinity, "fiskhest");
+assertEquals(one / (4 % -2), Infinity, "fiskhest");