commit cfa410b0ea561318f74a76c5323f0f6cd8eaaa50
author Mark Mendell <mark.p.mendell@intel.com> Mon May 25 16:02:44 2015 -0400
committer Mark Mendell <mark.p.mendell@intel.com> Tue Aug 11 10:32:47 2015 -0400
tree 80d989b8b26e3fd1afc232c5ecb9a0919823d15b
parent 1a5625be743a4a84329930ac1c7e96425e24ca8d

[optimizing] More x86_64 code improvements

Use the constant area some more, use 32-bit immediates in movq
instructions when possible, and other small tweaks.

Remove the commented out code for Math.Abs(float/double) as it would
fail for baseline compiler due to the output being the same as the
input.

Change-Id: Ifa39f1865b94cec2e1c0a99af3066a645e9d3618
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>

compiler/optimizing/intrinsics_x86_64.cc

diff --git a/compiler/optimizing/intrinsics_x86_64.cc b/compiler/optimizing/intrinsics_x86_64.cc
index 62cdb4c..85d40d7 100644
--- a/compiler/optimizing/intrinsics_x86_64.cc
+++ b/compiler/optimizing/intrinsics_x86_64.cc
@@ -282,8 +282,6 @@
                                                            LocationSummary::kNoCall,
                                                            kIntrinsified);
   locations->SetInAt(0, Location::RequiresFpuRegister());
-  // TODO: Allow x86 to work with memory. This requires assembler support, see below.
-  // locations->SetInAt(0, Location::Any());               // X86 can work on memory directly.
   locations->SetOut(Location::SameAsFirstInput());
   locations->AddTemp(Location::RequiresFpuRegister());  // FP reg to hold mask.
 }
@@ -294,34 +292,18 @@
                       CodeGeneratorX86_64* codegen) {
   Location output = locations->Out();
 
-  if (output.IsFpuRegister()) {
-    // In-register
-    XmmRegister xmm_temp = locations->GetTemp(0).AsFpuRegister<XmmRegister>();
+  DCHECK(output.IsFpuRegister());
+  XmmRegister xmm_temp = locations->GetTemp(0).AsFpuRegister<XmmRegister>();
 
-    // TODO: Can mask directly with constant area using pand if we can guarantee
-    // that the literal is aligned on a 16 byte boundary.  This will avoid a
-    // temporary.
-    if (is64bit) {
-      __ movsd(xmm_temp, codegen->LiteralInt64Address(INT64_C(0x7FFFFFFFFFFFFFFF)));
-      __ andpd(output.AsFpuRegister<XmmRegister>(), xmm_temp);
-    } else {
-      __ movss(xmm_temp, codegen->LiteralInt32Address(INT32_C(0x7FFFFFFF)));
-      __ andps(output.AsFpuRegister<XmmRegister>(), xmm_temp);
-    }
+  // TODO: Can mask directly with constant area using pand if we can guarantee
+  // that the literal is aligned on a 16 byte boundary.  This will avoid a
+  // temporary.
+  if (is64bit) {
+    __ movsd(xmm_temp, codegen->LiteralInt64Address(INT64_C(0x7FFFFFFFFFFFFFFF)));
+    __ andpd(output.AsFpuRegister<XmmRegister>(), xmm_temp);
   } else {
-    // TODO: update when assember support is available.
-    UNIMPLEMENTED(FATAL) << "Needs assembler support.";
-//  Once assembler support is available, in-memory operations look like this:
-//    if (is64bit) {
-//      DCHECK(output.IsDoubleStackSlot());
-//      // No 64b and with literal.
-//      __ movq(cpu_temp, Immediate(INT64_C(0x7FFFFFFFFFFFFFFF)));
-//      __ andq(Address(CpuRegister(RSP), output.GetStackIndex()), cpu_temp);
-//    } else {
-//      DCHECK(output.IsStackSlot());
-//      // Can use and with a literal directly.
-//      __ andl(Address(CpuRegister(RSP), output.GetStackIndex()), Immediate(INT64_C(0x7FFFFFFF)));
-//    }
+    __ movss(xmm_temp, codegen->LiteralInt32Address(INT32_C(0x7FFFFFFF)));
+    __ andps(output.AsFpuRegister<XmmRegister>(), xmm_temp);
   }
 }
 
@@ -736,6 +718,7 @@
   codegen_->Load64BitValue(out, kPrimIntMax);
 
   // if inPlusPointFive >= maxInt goto done
+  __ movl(out, Immediate(kPrimIntMax));
   __ comiss(inPlusPointFive, codegen_->LiteralFloatAddress(static_cast<float>(kPrimIntMax)));
   __ j(kAboveEqual, &done);
 
@@ -783,6 +766,7 @@
   codegen_->Load64BitValue(out, kPrimLongMax);
 
   // if inPlusPointFive >= maxLong goto done
+  __ movq(out, Immediate(kPrimLongMax));
   __ comisd(inPlusPointFive, codegen_->LiteralDoubleAddress(static_cast<double>(kPrimLongMax)));
   __ j(kAboveEqual, &done);