Add OverrideHWGenerateLID

Add OverrideHWGenerateLID to configure the value.

Author: iwwu

IGC/Compiler/CISACodeGen/CSWalkOrder.cpp

@@ -236,6 +236,20 @@ void IGC::overrideWalkOrderKeysInPass(
     {
         threadIDLayout = (ThreadIDLayout)IGC_IS_FLAG_ENABLED(OverrideCsTileLayout);
     }
+
+    const IGC::TriboolFlag overrideHWGenerateLID =
+        static_cast<TriboolFlag>(IGC_GET_FLAG_VALUE(OverrideHWGenerateLID));
+    switch (overrideHWGenerateLID)
+    {
+    case TriboolFlag::Enabled:
+        enableHWGenerateLID = true;
+        break;
+    case TriboolFlag::Disabled:
+        enableHWGenerateLID = false;
+        break;
+    default:
+        break;
+    }
 }
 
 bool IGC::enableHWGenerateLIDInPass(

IGC/Compiler/CodeGenPublicEnums.h

@@ -327,12 +327,12 @@ namespace IGC
 
 
     enum class CS_WALK_ORDER : unsigned char {
-        WO_XYZ = 0,
-        WO_XZY = 1,
-        WO_YXZ = 2,
-        WO_ZXY = 3,
-        WO_YZX = 4,
-        WO_ZYX = 5
+        WO_XYZ = 0, //012
+        WO_XZY = 1, //021
+        WO_YXZ = 2, //102
+        WO_ZXY = 3, //201 - incorrect enum value, need to swap with next line
+        WO_YZX = 4, //120 - incorrect enum value, need to swap with the line above
+        WO_ZYX = 5  //210
     };
 
     enum EMIT_LOCAL_MASK {

IGC/common/igc_flags.h

@@ -920,6 +920,11 @@ DECLARE_IGC_REGKEY(bool, OverrideCsWalkOrderEnable,  false, "Enable overriding c
 DECLARE_IGC_REGKEY(int,  OverrideCsWalkOrder,        0,     "Override compute walker walk order", true)
 DECLARE_IGC_REGKEY(bool, OverrideCsTileLayoutEnable, false, "Enable overriding compute walker tile layout", true)
 DECLARE_IGC_REGKEY(bool, OverrideCsTileLayout,       0,     "Override compute walker tile layout. False is linear. True is TileY", true)
+DECLARE_IGC_REGKEY_ENUM(OverrideHWGenerateLID, -1, "Override HW Generate Local ID setting" \
+    "-1 - default behavior," \
+    " 0 - force disabled," \
+    " 1 - force enabled", \
+    TRIBOOL_OPTIONS, true)
 DECLARE_IGC_REGKEY(DWORD, MemCpyLoweringUnrollThreshold, 12, "Min number of mem instructions that require non-unrolled loop when lowering memcpy", false)
 DECLARE_IGC_REGKEY(DWORD, EnablePrivMemNewSOATranspose,            1, "0 : disable new algo; 1 and up : enable new algo. " \
                                                                       "1 : enable new algo just for array of struct; " \