use long as return type of timestamp intrinsic

Using long as timestamp intrinsic's return type can save a few
instructions on platforms that supports QW.

Author: jgu222

IGC/BiFModule/Implementation/IGCBiF_Intrinsics.cl

@@ -133,7 +133,7 @@ half      __builtin_IB_ftoh_rtz(float)  __attribute__((const));
 #endif // defined(cl_khr_fp64)
 
 // Debug/Testing Built-In Functions
-uint2     __builtin_IB_read_cycle_counter(void) __attribute__((const));
+ulong     __builtin_IB_read_cycle_counter(void) __attribute__((const));
 void      __builtin_IB_source_value(uint reg);
 uint      __builtin_IB_set_dbg_register(uint dgb0_0);
 uint      __builtin_IB_movreg(uint reg) __attribute__((const));

IGC/Compiler/CISACodeGen/EmitVISAPass.cpp

@@ -4765,10 +4765,11 @@ void EmitPass::emitcycleCounter(llvm::Instruction* inst)
 {
     // Must move 2 DW at the same time, so use "(W) mov (2|M0)"
     CVariable* dst = m_destination;
+    CVariable* dst2DW = m_currShader->GetNewAlias(dst, ISA_TYPE_UD, 0, 0);
     m_encoder->SetNoMask();
     m_encoder->SetSrcRegion(0, 1, 1, 0);
     m_encoder->SetUniformSIMDSize(SIMDMode::SIMD2);
-    m_encoder->Copy(dst, m_currShader->GetTSC());
+    m_encoder->Copy(dst2DW, m_currShader->GetTSC());
     m_encoder->Push();
 }
 

IGC/GenISAIntrinsics/Intrinsic_definitions.py

@@ -807,7 +807,7 @@
     "None"]],
 ####################################################################################################
 "GenISA_cycleCounter": ["",
-    [("int2",                          ""),
+    [("long",                          "result"),
     [],
     "None"]],
 ####################################################################################################