Subzero: Use the memset inline threshold for memset.

src/IceTargetLoweringX8632Traits.h

@@ -711,7 +711,7 @@ public:
   /// @{
   static constexpr uint32_t MEMCPY_UNROLL_LIMIT = 8;
   static constexpr uint32_t MEMMOVE_UNROLL_LIMIT = 8;
-  static constexpr uint32_t MEMSET_UNROLL_LIMIT = 16;
+  static constexpr uint32_t MEMSET_UNROLL_LIMIT = 8;
   /// @}
 
   /// Value is in bytes. Return Value adjusted to the next highest multiple of

src/IceTargetLoweringX8664Traits.h

@@ -762,7 +762,7 @@ public:
   /// @{
   static constexpr uint32_t MEMCPY_UNROLL_LIMIT = 8;
   static constexpr uint32_t MEMMOVE_UNROLL_LIMIT = 8;
-  static constexpr uint32_t MEMSET_UNROLL_LIMIT = 16;
+  static constexpr uint32_t MEMSET_UNROLL_LIMIT = 8;
   /// @}
 
   /// Value is in bytes. Return Value adjusted to the next highest multiple of

src/IceTargetLoweringX86BaseImpl.h

@@ -5015,13 +5015,13 @@ void TargetX86Base<TraitsType>::lowerMemset(Operand *Dest, Operand *Val,
     // memory unit as the access to the same memory are far apart.
     Type Ty;
     if (ValValue == 0 && CountValue >= BytesPerStoreq &&
-        CountValue <= BytesPerStorep * Traits::MEMCPY_UNROLL_LIMIT) {
+        CountValue <= BytesPerStorep * Traits::MEMSET_UNROLL_LIMIT) {
       // When the value is zero it can be loaded into a vector register cheaply
       // using the xor trick.
       Base = legalizeToReg(Dest);
       VecReg = makeVectorOfZeros(IceType_v16i8);
       Ty = largestTypeInSize(CountValue);
-    } else if (CountValue <= BytesPerStorei32 * Traits::MEMCPY_UNROLL_LIMIT) {
+    } else if (CountValue <= BytesPerStorei32 * Traits::MEMSET_UNROLL_LIMIT) {
       // When the value is non-zero or the count is small we can't use vector
       // instructions so are limited to 32-bit stores.
       Base = legalizeToReg(Dest);