Implement truncated d-to-i as a stub on x86

- Added a general DoubleToIStub so that it's possible to extend to other platforms and non-truncating case.
- This version handles all cases of truncation (previous code deopted in some cases) and all source/destination register combinations without clobbering any temps.

R=yangguo@chromium.org

Review URL: https://codereview.chromium.org/18612005

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15645 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

src/code-stubs.h

@@ -72,6 +72,7 @@ namespace internal {
   V(ArgumentsAccess)                     \
   V(RegExpConstructResult)               \
   V(NumberToString)                      \
+  V(DoubleToI)                           \
   V(CEntry)                              \
   V(JSEntry)                             \
   V(KeyedLoadElement)                    \
@@ -1746,6 +1747,60 @@ class KeyedLoadDictionaryElementStub : public PlatformCodeStub {
 };
 
 
+class DoubleToIStub : public PlatformCodeStub {
+ public:
+  DoubleToIStub(Register source,
+                Register destination,
+                int offset,
+                bool is_truncating) : bit_field_(0) {
+    bit_field_ = SourceRegisterBits::encode(source.code_) |
+      DestinationRegisterBits::encode(destination.code_) |
+      OffsetBits::encode(offset) |
+      IsTruncatingBits::encode(is_truncating);
+  }
+
+  Register source() {
+    Register result = { SourceRegisterBits::decode(bit_field_) };
+    return result;
+  }
+
+  Register destination() {
+    Register result = { DestinationRegisterBits::decode(bit_field_) };
+    return result;
+  }
+
+  bool is_truncating() {
+    return IsTruncatingBits::decode(bit_field_);
+  }
+
+  int offset() {
+    return OffsetBits::decode(bit_field_);
+  }
+
+  void Generate(MacroAssembler* masm);
+
+ private:
+  static const int kBitsPerRegisterNumber = 6;
+  STATIC_ASSERT((1L << kBitsPerRegisterNumber) >= Register::kNumRegisters);
+  class SourceRegisterBits:
+      public BitField<int, 0, kBitsPerRegisterNumber> {};  // NOLINT
+  class DestinationRegisterBits:
+      public BitField<int, kBitsPerRegisterNumber,
+        kBitsPerRegisterNumber> {};  // NOLINT
+  class IsTruncatingBits:
+      public BitField<bool, 2 * kBitsPerRegisterNumber, 1> {};  // NOLINT
+  class OffsetBits:
+      public BitField<int, 2 * kBitsPerRegisterNumber + 1, 3> {};  // NOLINT
+
+  Major MajorKey() { return DoubleToI; }
+  int MinorKey() { return bit_field_; }
+
+  int bit_field_;
+
+  DISALLOW_COPY_AND_ASSIGN(DoubleToIStub);
+};
+
+
 class KeyedLoadFastElementStub : public HydrogenCodeStub {
  public:
   KeyedLoadFastElementStub(bool is_js_array, ElementsKind elements_kind) {

src/ia32/lithium-codegen-ia32.cc

@@ -5645,93 +5645,22 @@ void LCodeGen::DoDoubleToI(LDoubleToI* instr) {
   XMMRegister input_reg = ToDoubleRegister(input);
   Register result_reg = ToRegister(result);
 
+  __ cvttsd2si(result_reg, Operand(input_reg));
+
   if (instr->truncating()) {
     // Performs a truncating conversion of a floating point number as used by
     // the JS bitwise operations.
-    __ cvttsd2si(result_reg, Operand(input_reg));
+    Label fast_case_succeeded;
     __ cmp(result_reg, 0x80000000u);
-    if (CpuFeatures::IsSupported(SSE3)) {
-      // This will deoptimize if the exponent of the input in out of range.
-      CpuFeatureScope scope(masm(), SSE3);
-      Label convert, done;
-      __ j(not_equal, &done, Label::kNear);
-      __ sub(Operand(esp), Immediate(kDoubleSize));
-      __ movdbl(Operand(esp, 0), input_reg);
-      // Get exponent alone and check for too-big exponent.
-      __ mov(result_reg, Operand(esp, sizeof(int32_t)));
-      __ and_(result_reg, HeapNumber::kExponentMask);
-      const uint32_t kTooBigExponent =
-          (HeapNumber::kExponentBias + 63) << HeapNumber::kExponentShift;
-      __ cmp(Operand(result_reg), Immediate(kTooBigExponent));
-      __ j(less, &convert, Label::kNear);
-      __ add(Operand(esp), Immediate(kDoubleSize));
-      DeoptimizeIf(no_condition, instr->environment());
-      __ bind(&convert);
-      // Do conversion, which cannot fail because we checked the exponent.
-      __ fld_d(Operand(esp, 0));
-      __ fisttp_d(Operand(esp, 0));
-      __ mov(result_reg, Operand(esp, 0));  // Low word of answer is the result.
-      __ add(Operand(esp), Immediate(kDoubleSize));
-      __ bind(&done);
-    } else {
-      Label done;
-      Register temp_reg = ToRegister(instr->temp());
-      XMMRegister xmm_scratch = xmm0;
-
-      // If cvttsd2si succeeded, we're done. Otherwise, we attempt
-      // manual conversion.
-      __ j(not_equal, &done, Label::kNear);
-
-      // Get high 32 bits of the input in result_reg and temp_reg.
-      __ pshufd(xmm_scratch, input_reg, 1);
-      __ movd(Operand(temp_reg), xmm_scratch);
-      __ mov(result_reg, temp_reg);
-
-      // Prepare negation mask in temp_reg.
-      __ sar(temp_reg, kBitsPerInt - 1);
-
-      // Extract the exponent from result_reg and subtract adjusted
-      // bias from it. The adjustment is selected in a way such that
-      // when the difference is zero, the answer is in the low 32 bits
-      // of the input, otherwise a shift has to be performed.
-      __ shr(result_reg, HeapNumber::kExponentShift);
-      __ and_(result_reg,
-              HeapNumber::kExponentMask >> HeapNumber::kExponentShift);
-      __ sub(Operand(result_reg),
-             Immediate(HeapNumber::kExponentBias +
-                       HeapNumber::kExponentBits +
-                       HeapNumber::kMantissaBits));
-      // Don't handle big (> kMantissaBits + kExponentBits == 63) or
-      // special exponents.
-      DeoptimizeIf(greater, instr->environment());
-
-      // Zero out the sign and the exponent in the input (by shifting
-      // it to the left) and restore the implicit mantissa bit,
-      // i.e. convert the input to unsigned int64 shifted left by
-      // kExponentBits.
-      ExternalReference minus_zero = ExternalReference::address_of_minus_zero();
-      // Minus zero has the most significant bit set and the other
-      // bits cleared.
-      __ movdbl(xmm_scratch, Operand::StaticVariable(minus_zero));
-      __ psllq(input_reg, HeapNumber::kExponentBits);
-      __ por(input_reg, xmm_scratch);
-
-      // Get the amount to shift the input right in xmm_scratch.
-      __ neg(result_reg);
-      __ movd(xmm_scratch, Operand(result_reg));
-
-      // Shift the input right and extract low 32 bits.
-      __ psrlq(input_reg, xmm_scratch);
-      __ movd(Operand(result_reg), input_reg);
-
-      // Use the prepared mask in temp_reg to negate the result if necessary.
-      __ xor_(result_reg, Operand(temp_reg));
-      __ sub(result_reg, Operand(temp_reg));
-      __ bind(&done);
-    }
+    __ j(not_equal, &fast_case_succeeded);
+    __ sub(esp, Immediate(kDoubleSize));
+    __ movdbl(MemOperand(esp, 0), input_reg);
+    DoubleToIStub stub(esp, result_reg, 0, true);
+    __ call(stub.GetCode(isolate()), RelocInfo::CODE_TARGET);
+    __ add(esp, Immediate(kDoubleSize));
+    __ bind(&fast_case_succeeded);
   } else {
     Label done;
-    __ cvttsd2si(result_reg, Operand(input_reg));
     __ cvtsi2sd(xmm0, Operand(result_reg));
     __ ucomisd(xmm0, input_reg);
     DeoptimizeIf(not_equal, instr->environment());

test/cctest/cctest.gyp

@@ -110,6 +110,7 @@
         ['v8_target_arch=="ia32"', {
           'sources': [
             'test-assembler-ia32.cc',
+            'test-code-stubs-ia32.cc',
             'test-disasm-ia32.cc',
             'test-log-stack-tracer.cc'
           ],

test/mjsunit/double-truncation.js

+// Copyright 2012 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.
+
+// Flags: --allow-natives-syntax
+
+function RunOneTruncationTest(a, b) {
+  var temp = a | 0;
+  assertEquals(b, temp);
+}
+
+function RunAllTruncationTests() {
+  RunOneTruncationTest(0, 0);
+  RunOneTruncationTest(0.5, 0);
+  RunOneTruncationTest(-0.5, 0);
+  RunOneTruncationTest(1.5, 1);
+  RunOneTruncationTest(-1.5, -1);
+  RunOneTruncationTest(5.5, 5);
+  RunOneTruncationTest(-5.0, -5);
+  RunOneTruncationTest(NaN, 0);
+  RunOneTruncationTest(Infinity, 0);
+  RunOneTruncationTest(-NaN, 0);
+  RunOneTruncationTest(-Infinity, 0);
+
+  RunOneTruncationTest(4.5036e+15, 0x1635E000);
+  RunOneTruncationTest(-4.5036e+15, -372629504);
+
+  RunOneTruncationTest(4503603922337791.0, -1);
+  RunOneTruncationTest(-4503603922337791.0, 1);
+  RunOneTruncationTest(4503601774854143.0, 2147483647);
+  RunOneTruncationTest(-4503601774854143.0, -2147483647);
+  RunOneTruncationTest(9007207844675582.0, -2);
+  RunOneTruncationTest(-9007207844675582.0, 2);
+
+  RunOneTruncationTest(2.4178527921507624e+24, -536870912);
+  RunOneTruncationTest(-2.4178527921507624e+24, 536870912);
+  RunOneTruncationTest(2.417853945072267e+24, -536870912);
+  RunOneTruncationTest(-2.417853945072267e+24, 536870912);
+
+  RunOneTruncationTest(4.8357055843015248e+24, -1073741824);
+  RunOneTruncationTest(-4.8357055843015248e+24, 1073741824);
+  RunOneTruncationTest(4.8357078901445341e+24, -1073741824);
+  RunOneTruncationTest(-4.8357078901445341e+24, 1073741824);
+
+  RunOneTruncationTest(9.6714111686030497e+24, -2147483648);
+  RunOneTruncationTest(-9.6714111686030497e+24, -2147483648);
+  RunOneTruncationTest(9.6714157802890681e+24, -2147483648);
+  RunOneTruncationTest(-9.6714157802890681e+24, -2147483648);
+}
+
+RunAllTruncationTests();
+RunAllTruncationTests();
+%OptimizeFunctionOnNextCall(RunOneTruncationTest);
+RunAllTruncationTests();
+RunAllTruncationTests();