8287697: Limit auto vectorization to 32-byte vector on Cascade Lake

Reviewed-by: kvn, jbhateja

Author: Sandhya Viswanathan

src/hotspot/cpu/x86/vm_version_x86.cpp

@@ -1295,6 +1295,25 @@ void VM_Version::get_processor_features() {
     FLAG_SET_DEFAULT(MaxVectorSize, max_vector_size);
   }
 
+  if (FLAG_IS_DEFAULT(SuperWordMaxVectorSize)) {
+    if (FLAG_IS_DEFAULT(UseAVX) && UseAVX > 2 &&
+        is_intel_skylake() && _stepping >= 5) {
+      // Limit auto vectorization to 256 bit (32 byte) by default on Cascade Lake
+      FLAG_SET_DEFAULT(SuperWordMaxVectorSize, MIN2(MaxVectorSize, (intx)32));
+    } else {
+      FLAG_SET_DEFAULT(SuperWordMaxVectorSize, MaxVectorSize);
+    }
+  } else {
+    if (SuperWordMaxVectorSize > MaxVectorSize) {
+      warning("SuperWordMaxVectorSize cannot be greater than MaxVectorSize %i", (int) MaxVectorSize);
+      FLAG_SET_DEFAULT(SuperWordMaxVectorSize, MaxVectorSize);
+    }
+    if (!is_power_of_2(SuperWordMaxVectorSize)) {
+      warning("SuperWordMaxVectorSize must be a power of 2, setting to MaxVectorSize: %i", (int) MaxVectorSize);
+      FLAG_SET_DEFAULT(SuperWordMaxVectorSize, MaxVectorSize);
+    }
+  }
+
 #if defined(COMPILER2) && defined(ASSERT)
   if (MaxVectorSize > 0) {
     if (supports_avx() && PrintMiscellaneous && Verbose && TraceNewVectors) {

src/hotspot/share/opto/c2_globals.hpp

@@ -82,6 +82,11 @@
           "actual size could be less depending on elements type")           \
           range(0, max_jint)                                                \
                                                                             \
+  product(intx, SuperWordMaxVectorSize, 64, DIAGNOSTIC,                     \
+          "Vector size limit in bytes for superword, "                      \
+          "superword vector size limit in bytes")                           \
+          range(0, max_jint)                                                \
+                                                                            \
   product(intx, ArrayOperationPartialInlineSize, 0, DIAGNOSTIC,             \
           "Partial inline size used for small array operations"             \
           "(e.g. copy,cmp) acceleration.")                                  \

src/hotspot/share/opto/superword.cpp

@@ -199,6 +199,16 @@ bool SuperWord::transform_loop(IdealLoopTree* lpt, bool do_optimization) {
   return success;
 }
 
+//------------------------------max vector size------------------------------
+int SuperWord::max_vector_size(BasicType bt) {
+  int max_vector = Matcher::max_vector_size(bt);
+  int sw_max_vector_limit = SuperWordMaxVectorSize / type2aelembytes(bt);
+  if (max_vector > sw_max_vector_limit) {
+    max_vector = sw_max_vector_limit;
+  }
+  return max_vector;
+}
+
 //------------------------------early unrolling analysis------------------------------
 void SuperWord::unrolling_analysis(int &local_loop_unroll_factor) {
   bool is_slp = true;
@@ -217,7 +227,7 @@ void SuperWord::unrolling_analysis(int &local_loop_unroll_factor) {
     ignored_loop_nodes[i] = -1;
   }
 
-  int max_vector = Matcher::max_vector_size(T_BYTE);
+  int max_vector = max_vector_size(T_BYTE);
 
   // Process the loop, some/all of the stack entries will not be in order, ergo
   // need to preprocess the ignored initial state before we process the loop
@@ -352,7 +362,7 @@ void SuperWord::unrolling_analysis(int &local_loop_unroll_factor) {
 
       if (is_java_primitive(bt) == false) continue;
 
-      int cur_max_vector = Matcher::max_vector_size(bt);
+      int cur_max_vector = max_vector_size(bt);
 
       // If a max vector exists which is not larger than _local_loop_unroll_factor
       // stop looking, we already have the max vector to map to.
@@ -991,7 +1001,7 @@ int SuperWord::get_vw_bytes_special(MemNode* s) {
       }
     }
     if (should_combine_adjacent) {
-      vw = MIN2(Matcher::max_vector_size(btype)*type2aelembytes(btype), vw * 2);
+      vw = MIN2(max_vector_size(btype)*type2aelembytes(btype), vw * 2);
     }
   }
 
@@ -1689,7 +1699,7 @@ void SuperWord::combine_packs() {
     Node_List* p1 = _packset.at(i);
     if (p1 != NULL) {
       BasicType bt = velt_basic_type(p1->at(0));
-      uint max_vlen = Matcher::max_vector_size(bt); // Max elements in vector
+      uint max_vlen = max_vector_size(bt); // Max elements in vector
       assert(is_power_of_2(max_vlen), "sanity");
       uint psize = p1->size();
       if (!is_power_of_2(psize)) {

src/hotspot/share/opto/superword.hpp

@@ -327,6 +327,8 @@ class SuperWord : public ResourceObj {
 
   bool transform_loop(IdealLoopTree* lpt, bool do_optimization);
 
+  int max_vector_size(BasicType bt);
+
   void unrolling_analysis(int &local_loop_unroll_factor);
 
   // Accessors for SWPointer

src/hotspot/share/opto/vectornode.cpp

@@ -294,12 +294,17 @@ int VectorNode::replicate_opcode(BasicType bt) {
   }
 }
 
+bool VectorNode::vector_size_supported(BasicType bt, uint vlen) {
+  return (Matcher::vector_size_supported(bt, vlen) &&
+          (vlen * type2aelembytes(bt) <= (uint)SuperWordMaxVectorSize));
+}
+
 // Also used to check if the code generator
 // supports the vector operation.
 bool VectorNode::implemented(int opc, uint vlen, BasicType bt) {
   if (is_java_primitive(bt) &&
       (vlen > 1) && is_power_of_2(vlen) &&
-      Matcher::vector_size_supported(bt, vlen)) {
+      vector_size_supported(bt, vlen)) {
     int vopc = VectorNode::opcode(opc, bt);
     // For rotate operation we will do a lazy de-generation into
     // OrV/LShiftV/URShiftV pattern if the target does not support
@@ -1320,7 +1325,7 @@ Node* ReductionNode::make_reduction_input(PhaseGVN& gvn, int opc, BasicType bt)
 bool ReductionNode::implemented(int opc, uint vlen, BasicType bt) {
   if (is_java_primitive(bt) &&
       (vlen > 1) && is_power_of_2(vlen) &&
-      Matcher::vector_size_supported(bt, vlen)) {
+      VectorNode::vector_size_supported(bt, vlen)) {
     int vopc = ReductionNode::opcode(opc, bt);
     return vopc != opc && Matcher::match_rule_supported_vector(vopc, vlen, bt);
   }

src/hotspot/share/opto/vectornode.hpp

@@ -88,6 +88,7 @@ class VectorNode : public TypeNode {
 
   static int  opcode(int opc, BasicType bt);
   static int replicate_opcode(BasicType bt);
+  static bool vector_size_supported(BasicType bt, uint vlen);
   static bool implemented(int opc, uint vlen, BasicType bt);
   static bool is_shift(Node* n);
   static bool is_vshift_cnt(Node* n);