From c173d416f749348bee42e1a9436a999700d0f0e8 Mon Sep 17 00:00:00 2001
From: Boris Ulasevich <bulasevich@openjdk.org>
Date: Thu, 6 Nov 2025 12:56:37 +0000
Subject: [PATCH] 8359256: AArch64: Use SHA3 GPR intrinsic where it's faster

Reviewed-by: eastigeevich, phh
---
 src/hotspot/cpu/aarch64/globals_aarch64.hpp   |  2 +-
 .../cpu/aarch64/vm_version_aarch64.cpp        | 30 +++++++++++--------
 2 files changed, 19 insertions(+), 13 deletions(-)

diff --git a/src/hotspot/cpu/aarch64/globals_aarch64.hpp b/src/hotspot/cpu/aarch64/globals_aarch64.hpp
index 8e520314c8b..107919b2cbd 100644
--- a/src/hotspot/cpu/aarch64/globals_aarch64.hpp
+++ b/src/hotspot/cpu/aarch64/globals_aarch64.hpp
@@ -95,7 +95,7 @@ define_pd_global(intx, InlineSmallCode,          1000);
           "Use simplest and shortest implementation for array equals")  \
   product(bool, UseSIMDForBigIntegerShiftIntrinsics, true,              \
           "Use SIMD instructions for left/right shift of BigInteger")   \
-  product(bool, UseSIMDForSHA3Intrinsic, true,                          \
+  product(bool, UseSIMDForSHA3Intrinsic, false,                         \
           "Use SIMD SHA3 instructions for SHA3 intrinsic")              \
   product(bool, AvoidUnalignedAccesses, false,                          \
           "Avoid generating unaligned memory accesses")                 \
diff --git a/src/hotspot/cpu/aarch64/vm_version_aarch64.cpp b/src/hotspot/cpu/aarch64/vm_version_aarch64.cpp
index a04e9defa4b..b47bfe6a89f 100644
--- a/src/hotspot/cpu/aarch64/vm_version_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/vm_version_aarch64.cpp
@@ -375,18 +375,24 @@ void VM_Version::initialize() {
     FLAG_SET_DEFAULT(UseSHA256Intrinsics, false);
   }
 
-  if (UseSHA && VM_Version::supports_sha3()) {
-    // Auto-enable UseSHA3Intrinsics on hardware with performance benefit.
-    // Note that the evaluation of UseSHA3Intrinsics shows better performance
-    // on Apple silicon but worse performance on Neoverse V1 and N2.
-    if (_cpu == CPU_APPLE) {  // Apple silicon
-      if (FLAG_IS_DEFAULT(UseSHA3Intrinsics)) {
-        FLAG_SET_DEFAULT(UseSHA3Intrinsics, true);
-      }
-    }
-  } else if (UseSHA3Intrinsics && UseSIMDForSHA3Intrinsic) {
-    warning("Intrinsics for SHA3-224, SHA3-256, SHA3-384 and SHA3-512 crypto hash functions not available on this CPU.");
-    FLAG_SET_DEFAULT(UseSHA3Intrinsics, false);
+  if (UseSHA && VM_Version::supports_sha3() && _cpu == CPU_APPLE && FLAG_IS_DEFAULT(UseSIMDForSHA3Intrinsic)) {
+    // Note: SIMD faster on Apple, worse on Neoverse V1, V2 and N2.
+    FLAG_SET_DEFAULT(UseSIMDForSHA3Intrinsic, true);
+  }
+
+  // Enable SHA-3 intrinsics (SIMD or GPR). The GPR path does not require SHA instructions.
+  if (FLAG_IS_DEFAULT(UseSHA3Intrinsics)) {
+    FLAG_SET_DEFAULT(UseSHA3Intrinsics, true);
+  }
+
+  if (!UseSHA3Intrinsics && UseSIMDForSHA3Intrinsic) {
+    // Keep flags consistent: if SHA3 intrinsics are off, disable the SHA3 SIMD variant.
+    FLAG_SET_DEFAULT(UseSIMDForSHA3Intrinsic, false);
+  }
+
+  if (!VM_Version::supports_sha3() && UseSIMDForSHA3Intrinsic) {
+    warning("SHA3 instructions are not available on this CPU");
+    FLAG_SET_DEFAULT(UseSIMDForSHA3Intrinsic, false);
   }
 
   if (UseSHA && VM_Version::supports_sha512()) {