Updating xarch to utilize EVEX compares and blending where profitable

dotnet · Dec 14, 2024 · 928fea3 · 928fea3
1 parent 2182002
commit 928fea3
Show file tree

Hide file tree

Showing 6 changed files with 840 additions and 210 deletions.
diff --git a/src/coreclr/jit/gentree.cpp b/src/coreclr/jit/gentree.cpp
@@ -29540,7 +29540,7 @@ var_types GenTreeHWIntrinsic::GetLookupTypeForCmpOp(
     {
         case GT_EQ:
         {
-            if (simdSize == 64)
+            if ((simdSize == 64) || (comp->opts.OptimizationEnabled() && comp->canUseEvexEncoding()))
             {
                 lookupType = TYP_MASK;
             }
@@ -29551,7 +29551,8 @@ var_types GenTreeHWIntrinsic::GetLookupTypeForCmpOp(
         case GT_LE:
         case GT_NE:
         {
-            if ((simdSize == 64) || (varTypeIsIntegral(simdBaseType) && comp->canUseEvexEncoding()))
+            if ((simdSize == 64) ||
+                ((comp->opts.OptimizationEnabled() || varTypeIsIntegral(simdBaseType)) && comp->canUseEvexEncoding()))
             {
                 lookupType = TYP_MASK;
             }
@@ -29561,7 +29562,8 @@ var_types GenTreeHWIntrinsic::GetLookupTypeForCmpOp(
         case GT_GT:
         case GT_LT:
         {
-            if ((simdSize == 64) || (varTypeIsUnsigned(simdBaseType) && comp->canUseEvexEncoding()))
+            if ((simdSize == 64) ||
+                ((comp->opts.OptimizationEnabled() || varTypeIsUnsigned(simdBaseType)) && comp->canUseEvexEncoding()))
             {
                 lookupType = TYP_MASK;
             }

diff --git a/src/coreclr/jit/hwintrinsic.h b/src/coreclr/jit/hwintrinsic.h
@@ -546,12 +546,11 @@ struct HWIntrinsicInfo
     static bool isScalarIsa(CORINFO_InstructionSet isa);
 
 #ifdef TARGET_XARCH
-    static bool                isAVX2GatherIntrinsic(NamedIntrinsic id);
-    static FloatComparisonMode lookupFloatComparisonModeForSwappedArgs(FloatComparisonMode comparison);
-    static NamedIntrinsic      lookupIdForFloatComparisonMode(NamedIntrinsic      intrinsic,
-                                                              FloatComparisonMode comparison,
-                                                              var_types           simdBaseType,
-                                                              unsigned            simdSize);
+    static bool           isAVX2GatherIntrinsic(NamedIntrinsic id);
+    static NamedIntrinsic lookupIdForFloatComparisonMode(NamedIntrinsic      intrinsic,
+                                                         FloatComparisonMode comparison,
+                                                         var_types           simdBaseType,
+                                                         unsigned            simdSize);
 #endif
 
     // Member lookup