fix HW WA

Put the WA at the end of the HW comformity, because HW comformity itself may generate the code pattern which need be updated.

Author: bcheng0127

visa/HWConformity.cpp

@@ -5989,7 +5989,7 @@ void HWConformity::conformBB(G4_BB* bb)
         }
         // do this early since otherwise the moves inserted by other passes may still
         // inherit bad regions from the original inst
-        fixSrcRegion(i, bb);
+        fixSrcRegion(inst);
 
         bool changed = fixMov(i, bb);
         if (changed)
@@ -6249,6 +6249,20 @@ void HWConformity::conformBB(G4_BB* bb)
 #endif
     }
 
+    if (builder.hasBDstWSrc1EvenAlignIssue())
+    {
+        i = bb->begin(), iEnd = bb->end();
+        next_iter = i;
+        for (; i != iEnd; i = next_iter)
+        {
+            ++next_iter;
+            fixOddAlignSrc1Region(i, bb);
+#ifdef _DEBUG
+            verifyG4Kernel(kernel, Optimizer::PI_HWConformityChk, false);
+#endif
+        }
+    }
+
     if (VISA_WA_CHECK(builder.getPWaTable(), Wa_16012725276))
     {
         for (auto it = bb->begin(), itEnd = bb->end(); it != itEnd; ++it)
@@ -7298,8 +7312,8 @@ G4_INST* HWConformity::splitInstWithByteDst(G4_INST* expand_op)
 //  7. Dst.HorzStride must not be 0.        -- this needs not to be checked.
 //  8. VertStride must be used to cross GRF register boundaries. This rule implies that
 //      elements within a 'Width' cannot cross GRF boundaries.
-void HWConformity::fixSrcRegion(INST_LIST_ITER i, G4_BB *bb) {
-    G4_INST *inst = *i;
+void HWConformity::fixSrcRegion(G4_INST *inst) {
+
     bool comprInst = isCompressedInst(inst);
     for (int i = 0; i < G4_MAX_SRCS; i++)
     {
@@ -7411,25 +7425,32 @@ void HWConformity::fixSrcRegion(INST_LIST_ITER i, G4_BB *bb) {
             }
         }
     }
-
     if (inst->getDst() && !inst->hasNULLDst())
     {
         MUST_BE_TRUE(inst->getDst()->getHorzStride() != 0,
             "Bad source region: Width is greater than execution size.");
+    }
+}
 
-        if (builder.hasBDstWSrc1EvenAlignIssue()) {
-            G4_Operand *src1 = inst->getSrc(1);
-            if (src1 && src1->isSrcRegRegion()) {
-                G4_Operand *dst = inst->getDst();
-                if (dst && dst->isDstRegRegion() &&
-                    dst->asDstRegRegion()->getHorzStride() > 2 &&
-                    (dst->getType() == Type_B || dst->getType() == Type_UB) &&
-                    (src1->getType() == Type_W || src1->getType() == Type_UW) &&
-                    (src1->asSrcRegRegion()->getSubRegOff() % 2)) {
-                    G4_Operand *new_src1 =
-                        insertMovBefore(i, 1, src1->getType(), bb, Even_Word);
-                    inst->setSrc(new_src1, 1);
-                }
+void HWConformity::fixOddAlignSrc1Region(INST_LIST_ITER i, G4_BB *bb)
+{
+    G4_INST *inst = *i;
+
+    if (inst->getDst() && !inst->hasNULLDst())
+    {
+        G4_Operand *src1 = inst->getSrc(1);
+        if (src1 && src1->isSrcRegRegion())
+        {
+            G4_Operand *dst = inst->getDst();
+            if (dst && dst->isDstRegRegion() &&
+                dst->asDstRegRegion()->getHorzStride() > 2 &&
+                (dst->getType() == Type_B || dst->getType() == Type_UB) &&
+                (src1->getType() == Type_W || src1->getType() == Type_UW) &&
+                (src1->asSrcRegRegion()->getSubRegOff() % 2))
+            {
+                G4_Operand *new_src1 =
+                    insertMovBefore(i, 1, src1->getType(), bb, Even_Word);
+                inst->setSrc(new_src1, 1);
             }
         }
     }

visa/HWConformity.h

@@ -126,8 +126,9 @@ namespace vISA
         bool fixAcc(INST_LIST_ITER i, G4_BB* bb);
         void fixDstHstride(INST_LIST_ITER i, int extypesize);
         void fixMADInst(G4_BB* bb);
-        void fixSrcRegion(INST_LIST_ITER i, G4_BB *bb);
-        void conformBB(G4_BB* bb);
+        void fixSrcRegion(G4_INST *inst);
+        void fixOddAlignSrc1Region(INST_LIST_ITER i, G4_BB *bb);
+        void conformBB(G4_BB * bb);
         void fixSADA2Inst(G4_BB* bb);
         void fixMixedHFInst(G4_BB* bb);
         void fixSendInst(G4_BB* bb);