Update DPAS latency

The latency will be used in the instruction scheduling and SWSB
dependence tracking.

Author: bcheng0127

visa/G4_IR.cpp

@@ -7702,6 +7702,15 @@ bool G4_InstDpas::isInt() const {
   return false;
 }
 
+bool G4_InstDpas::isInt8() const {
+  if (Src1Precision == GenPrecision::S8 || Src1Precision == GenPrecision::U8 ||
+      Src2Precision == GenPrecision::S8 || Src2Precision == GenPrecision::U8) {
+    return true;
+  }
+
+  return false;
+}
+
 bool G4_InstDpas::is2xInt8() const {
   if ((Src1Precision == GenPrecision::S4 || Src1Precision == GenPrecision::U4 ||
        Src1Precision == GenPrecision::S2 ||

visa/G4_IR.hpp

@@ -1006,6 +1006,7 @@ class G4_InstDpas : public G4_INST {
   bool isFP16() const { return Src1Precision == GenPrecision::FP16; }
   bool isTF32() const { return Src1Precision == GenPrecision::TF32; }
   bool isInt() const;
+  bool isInt8() const;
   bool is2xInt8() const; // true if it is 2xint8 dpas
 
   uint8_t getOpsPerChan() const;

visa/LocalScheduler/LatencyTable.cpp

@@ -306,9 +306,10 @@ LatencyTableXe<PlatformGen::XE>::getMathLatency(const G4_INST *Inst) const {
   return value_of(LI::MATH) + value_of(LI::DELTA_MATH) * Scale;
 }
 
-template<>
-uint16_t LatencyTableXe<PlatformGen::XE>::getDPASLatency(
-    const G4_InstDpas *dpas) const {
+template <>
+uint16_t
+LatencyTableXe<PlatformGen::XE>::getDPASLatency(const G4_InstDpas *dpas) const {
+
   return getDPASLatency(dpas->getRepeatCount());
 }