8339063: [aarch64] Skip verify_sve_vector_length after native calls if SVE supports 128 bits VL only

Reviewed-by: adinn, fgao

src/hotspot/cpu/aarch64/aarch64.ad

@@ -2334,7 +2334,7 @@ bool Matcher::is_short_branch_offset(int rule, int br_size, int offset) {
 // Vector width in bytes.
 int Matcher::vector_width_in_bytes(BasicType bt) {
   // The MaxVectorSize should have been set by detecting SVE max vector register size.
-  int size = MIN2((UseSVE > 0) ? 256 : 16, (int)MaxVectorSize);
+  int size = MIN2((UseSVE > 0) ? (int)FloatRegister::sve_vl_max : (int)FloatRegister::neon_vl, (int)MaxVectorSize);
   // Minimum 2 values in vector
   if (size < 2*type2aelembytes(bt)) size = 0;
   // But never < 4
@@ -2373,7 +2373,7 @@ int Matcher::scalable_vector_reg_size(const BasicType bt) {
 
 // Vector ideal reg.
 uint Matcher::vector_ideal_reg(int len) {
-  if (UseSVE > 0 && 16 < len && len <= 256) {
+  if (UseSVE > 0 && FloatRegister::neon_vl < len && len <= FloatRegister::sve_vl_max) {
     return Op_VecA;
   }
   switch(len) {

src/hotspot/cpu/aarch64/aarch64_vector.ad

@@ -155,7 +155,7 @@ source %{
     }
 
     int length_in_bytes = vlen * type2aelembytes(bt);
-    if (UseSVE == 0 && length_in_bytes > 16) {
+    if (UseSVE == 0 && length_in_bytes > FloatRegister::neon_vl) {
       return false;
     }
 

src/hotspot/cpu/aarch64/aarch64_vector_ad.m4

@@ -145,7 +145,7 @@ source %{
     }
 
     int length_in_bytes = vlen * type2aelembytes(bt);
-    if (UseSVE == 0 && length_in_bytes > 16) {
+    if (UseSVE == 0 && length_in_bytes > FloatRegister::neon_vl) {
       return false;
     }
 

src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp

@@ -6413,8 +6413,10 @@ void MacroAssembler::cache_wbsync(bool is_pre) {
 }
 
 void MacroAssembler::verify_sve_vector_length(Register tmp) {
+  if (!UseSVE || VM_Version::get_max_supported_sve_vector_length() == FloatRegister::sve_vl_min) {
+    return;
+  }
   // Make sure that native code does not change SVE vector length.
-  if (!UseSVE) return;
   Label verify_ok;
   movw(tmp, zr);
   sve_inc(tmp, B);

src/hotspot/cpu/aarch64/register_aarch64.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2000, 2023, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2000, 2024, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2014, 2021, Red Hat Inc. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -166,7 +166,13 @@ class FloatRegister {
     max_slots_per_register  =  4,
     save_slots_per_register =  2,
     slots_per_neon_register =  4,
-    extra_save_slots_per_neon_register = slots_per_neon_register - save_slots_per_register
+    extra_save_slots_per_neon_register = slots_per_neon_register - save_slots_per_register,
+    neon_vl = 16,
+    // VLmax: The maximum sve vector length is determined by the hardware
+    // sve_vl_min <= VLmax <= sve_vl_max.
+    sve_vl_min = 16,
+    // Maximum supported vector length across all CPUs
+    sve_vl_max = 256
   };
 
   class FloatRegisterImpl: public AbstractRegisterImpl {

src/hotspot/cpu/aarch64/vm_version_aarch64.cpp

@@ -25,6 +25,7 @@
 
 #include "precompiled.hpp"
 #include "pauth_aarch64.hpp"
+#include "register_aarch64.hpp"
 #include "runtime/arguments.hpp"
 #include "runtime/globals_extension.hpp"
 #include "runtime/java.hpp"
@@ -44,6 +45,7 @@ int VM_Version::_zva_length;
 int VM_Version::_dcache_line_size;
 int VM_Version::_icache_line_size;
 int VM_Version::_initial_sve_vector_length;
+int VM_Version::_max_supported_sve_vector_length;
 bool VM_Version::_rop_protection;
 uintptr_t VM_Version::_pac_mask;
 
@@ -507,30 +509,37 @@ void VM_Version::initialize() {
   if (UseSVE > 0) {
     if (FLAG_IS_DEFAULT(MaxVectorSize)) {
       MaxVectorSize = _initial_sve_vector_length;
-    } else if (MaxVectorSize < 16) {
+    } else if (MaxVectorSize < FloatRegister::sve_vl_min) {
-      warning("SVE does not support vector length less than 16 bytes. Disabling SVE.");
+      warning("SVE does not support vector length less than %d bytes. Disabling SVE.",
+              FloatRegister::sve_vl_min);
       UseSVE = 0;
-    } else if ((MaxVectorSize % 16) == 0 && is_power_of_2(MaxVectorSize)) {
+    } else if (!((MaxVectorSize % FloatRegister::sve_vl_min) == 0 && is_power_of_2(MaxVectorSize))) {
-      int new_vl = set_and_get_current_sve_vector_length(MaxVectorSize);
-      _initial_sve_vector_length = new_vl;
-      // Update MaxVectorSize to the largest supported value.
-      if (new_vl < 0) {
-        vm_exit_during_initialization(
-          err_msg("Current system does not support SVE vector length for MaxVectorSize: %d",
-                  (int)MaxVectorSize));
-      } else if (new_vl != MaxVectorSize) {
-        warning("Current system only supports max SVE vector length %d. Set MaxVectorSize to %d",
-                new_vl, new_vl);
-      }
-      MaxVectorSize = new_vl;
-    } else {
       vm_exit_during_initialization(err_msg("Unsupported MaxVectorSize: %d", (int)MaxVectorSize));
     }
+
+    if (UseSVE > 0) {
+      // Acquire the largest supported vector length of this machine
+      _max_supported_sve_vector_length = set_and_get_current_sve_vector_length(FloatRegister::sve_vl_max);
+
+      if (MaxVectorSize != _max_supported_sve_vector_length) {
+        int new_vl = set_and_get_current_sve_vector_length(MaxVectorSize);
+        if (new_vl < 0) {
+          vm_exit_during_initialization(
+            err_msg("Current system does not support SVE vector length for MaxVectorSize: %d",
+                    (int)MaxVectorSize));
+        } else if (new_vl != MaxVectorSize) {
+          warning("Current system only supports max SVE vector length %d. Set MaxVectorSize to %d",
+                  new_vl, new_vl);
+        }
+        MaxVectorSize = new_vl;
+      }
+      _initial_sve_vector_length = MaxVectorSize;
+    }
   }
 
   if (UseSVE == 0) {  // NEON
     int min_vector_size = 8;
-    int max_vector_size = 16;
+    int max_vector_size = FloatRegister::neon_vl;
     if (!FLAG_IS_DEFAULT(MaxVectorSize)) {
       if (!is_power_of_2(MaxVectorSize)) {
         vm_exit_during_initialization(err_msg("Unsupported MaxVectorSize: %d", (int)MaxVectorSize));
@@ -542,11 +551,11 @@ void VM_Version::initialize() {
         FLAG_SET_DEFAULT(MaxVectorSize, max_vector_size);
       }
     } else {
-      FLAG_SET_DEFAULT(MaxVectorSize, 16);
+      FLAG_SET_DEFAULT(MaxVectorSize, FloatRegister::neon_vl);
     }
   }
 
-  int inline_size = (UseSVE > 0 && MaxVectorSize >= 16) ? MaxVectorSize : 0;
+  int inline_size = (UseSVE > 0 && MaxVectorSize >= FloatRegister::sve_vl_min) ? MaxVectorSize : 0;
   if (FLAG_IS_DEFAULT(ArrayOperationPartialInlineSize)) {
     FLAG_SET_DEFAULT(ArrayOperationPartialInlineSize, inline_size);
   } else if (ArrayOperationPartialInlineSize != 0 && ArrayOperationPartialInlineSize != inline_size) {

src/hotspot/cpu/aarch64/vm_version_aarch64.hpp

@@ -46,6 +46,7 @@ protected:
   static int _dcache_line_size;
   static int _icache_line_size;
   static int _initial_sve_vector_length;
+  static int _max_supported_sve_vector_length;
   static bool _rop_protection;
   static uintptr_t _pac_mask;
 
@@ -164,7 +165,8 @@ enum Ampere_CPU_Model {
 
   static int icache_line_size() { return _icache_line_size; }
   static int dcache_line_size() { return _dcache_line_size; }
-  static int get_initial_sve_vector_length()  { return _initial_sve_vector_length; };
+  static int get_initial_sve_vector_length()        { return _initial_sve_vector_length; };
+  static int get_max_supported_sve_vector_length()  { return _max_supported_sve_vector_length; };
 
   // Aarch64 supports fast class initialization checks
   static bool supports_fast_class_init_checks() { return true; }