src/hotspot/cpu/x86/vm_version_x86.hpp

@@ -216,11 +216,14 @@
                         : 4,
                 avx512f : 1,
                avx512dq : 1,
                         : 1,
                     adx : 1,
-                        : 6,
+                        : 3,
+             clflushopt : 1,
+                   clwb : 1,
+                        : 1,
                avx512pf : 1,
                avx512er : 1,
                avx512cd : 1,
                     sha : 1,
                avx512bw : 1,

@@ -335,11 +338,14 @@
 #define CPU_VZEROUPPER ((uint64_t)UCONST64(0x1000000000))       // Vzeroupper instruction
 #define CPU_AVX512_VPOPCNTDQ ((uint64_t)UCONST64(0x2000000000)) // Vector popcount
 #define CPU_VPCLMULQDQ ((uint64_t)UCONST64(0x4000000000)) //Vector carryless multiplication
 #define CPU_VAES ((uint64_t)UCONST64(0x8000000000))    // Vector AES instructions
 
-  enum Extended_Family {
+#define CPU_FLUSHOPT ((uint64_t)UCONST64(0x1000000000)) // flushopt instruction
+#define CPU_CLWB ((uint64_t)UCONST64(0x2000000000)) // clwb instruction
+
+enum Extended_Family {
     // AMD
     CPU_FAMILY_AMD_11H       = 0x11,
     // ZX
     CPU_FAMILY_ZX_CORE_F6    = 6,
     CPU_FAMILY_ZX_CORE_F7    = 7,

@@ -570,10 +576,12 @@
       result |= CPU_BMI2;
     if (_cpuid_info.sef_cpuid7_ebx.bits.sha != 0)
       result |= CPU_SHA;
     if (_cpuid_info.std_cpuid1_ecx.bits.fma != 0)
       result |= CPU_FMA;
+    if (_cpuid_info.sef_cpuid7_ebx.bits.clflushopt != 0)
+      result |= CPU_FLUSHOPT;
 
     // AMD features.
     if (is_amd()) {
       if ((_cpuid_info.ext_cpuid1_edx.bits.tdnow != 0) ||
           (_cpuid_info.ext_cpuid1_ecx.bits.prefetchw != 0))

@@ -589,10 +597,13 @@
         result |= CPU_LZCNT;
       // for Intel, ecx.bits.misalignsse bit (bit 8) indicates support for prefetchw
       if (_cpuid_info.ext_cpuid1_ecx.bits.misalignsse != 0) {
         result |= CPU_3DNOW_PREFETCH;
       }
+      if (_cpuid_info.sef_cpuid7_ebx.bits.clwb != 0) {
+        result |= CPU_CLWB;
+      }
     }
 
     // ZX features.
     if (is_zx()) {
       if (_cpuid_info.ext_cpuid1_ecx.bits.lzcnt_intel != 0)

@@ -923,8 +934,34 @@
 
   // SSE2 and later processors implement a 'pause' instruction
   // that can be used for efficient implementation of
   // the intrinsic for java.lang.Thread.onSpinWait()
   static bool supports_on_spin_wait() { return supports_sse2(); }
+
+  // there are several insns to force cache line sync to memory which
+  // we can use to ensure mapped persistent memory is up to date with
+  // pending in-cache changes.
+  //
+  // 64 bit cpus always support clflush which writes back and evicts
+  //
+  // clflushopt is optional and acts like clflush except it does
+  // not synchronize with other memory ops. it needs a preceding
+  // and trailing StoreStore fence
+  //
+  // clwb is an optional, intel-specific instruction optional which
+  // writes back without evicting the line. it also does not
+  // synchronize with other memory ops. so, it also needs a preceding
+  // and trailing StoreStore fence.
+
+#ifdef _LP64
+  static bool supports_clflush() { return true; }
+  static bool supports_clflushopt() { return (_features & CPU_FLUSHOPT == 0); }
+  static bool supports_clwb() { return (_features & CPU_CLWB == 0); }
+#else
+  static bool supports_clflush() { return true; }
+  static bool supports_clflushopt() { return false; }
+  static bool supports_clwb() { return false; }
+#endif // _LP64
+
 };
 
 #endif // CPU_X86_VM_VM_VERSION_X86_HPP