Support CMOV for x86_64 Select
If possible, generate CMOV to implement HSelect. Tricky cases are
an FP condition (no single CC generated), FP inputs (no FP CMOV)
and when the condition is a boolean or not emitted at the use site.
In these cases, keep using the existing HSelect code.
Added Load32BitValue for int and FP and used that to remove code
duplication. Added minimal checker test for int/long CMOV generation.
Change-Id: Id71e515f0afa5a30f53c5de3a5244de1ea429aae
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>
diff --git a/compiler/optimizing/code_generator_x86_64.h b/compiler/optimizing/code_generator_x86_64.h
index 318087e..6aa0814 100644
--- a/compiler/optimizing/code_generator_x86_64.h
+++ b/compiler/optimizing/code_generator_x86_64.h
@@ -264,6 +264,7 @@
void GenerateExplicitNullCheck(HNullCheck* instruction);
void PushOntoFPStack(Location source, uint32_t temp_offset,
uint32_t stack_adjustment, bool is_float);
+ void GenerateCompareTest(HCondition* condition);
template<class LabelType>
void GenerateTestAndBranch(HInstruction* instruction,
size_t condition_input_index,
@@ -481,6 +482,10 @@
// Load a 32/64 bit value into a register in the most efficient manner.
void Load32BitValue(CpuRegister dest, int32_t value);
void Load64BitValue(CpuRegister dest, int64_t value);
+ void Load32BitValue(XmmRegister dest, int32_t value);
+ void Load64BitValue(XmmRegister dest, int64_t value);
+ void Load32BitValue(XmmRegister dest, float value);
+ void Load64BitValue(XmmRegister dest, double value);
Address LiteralCaseTable(HPackedSwitch* switch_instr);