Add UseZtso to optimize memory fences on TSO-enabled platforms (openj…

…dk#40)
robehn · Oct 21, 2022 · 54f5aa1 · 54f5aa1
1 parent ba90b88
commit 54f5aa1
Show file tree

Hide file tree

Showing 2 changed files with 19 additions and 9 deletions.
diff --git a/src/hotspot/cpu/riscv/globals_riscv.hpp b/src/hotspot/cpu/riscv/globals_riscv.hpp
@@ -100,6 +100,7 @@ define_pd_global(intx, InlineSmallCode,          1000);
   product(bool, UseZicbom, false, EXPERIMENTAL, "Use Zicbom instructions")       \
   product(bool, UseZicbop, false, EXPERIMENTAL, "Use Zicbop instructions")       \
   product(bool, UseZicboz, false, EXPERIMENTAL, "Use Zicboz instructions")       \
+  product(bool, UseZtso, false, EXPERIMENTAL, "Assume Ztso memory model")        \
   product(bool, UseRVVForBigIntegerShiftIntrinsics, true,                        \
           "Use RVV instructions for left/right shift of BigInteger")
 

diff --git a/src/hotspot/cpu/riscv/macroAssembler_riscv.hpp b/src/hotspot/cpu/riscv/macroAssembler_riscv.hpp
@@ -347,21 +347,30 @@ class MacroAssembler: public Assembler {
   static void membar_mask_to_pred_succ(uint32_t order_constraint, uint32_t& predecessor, uint32_t& successor) {
     predecessor = (order_constraint >> 2) & 0x3;
     successor = order_constraint & 0x3;
-
-    // extend rw -> iorw:
-    // 01(w) -> 0101(ow)
-    // 10(r) -> 1010(ir)
-    // 11(rw)-> 1111(iorw)
-    if (UseConservativeFence) {
-      predecessor |= predecessor << 2;
-      successor |= successor << 2;
-    }
   }
 
   static int pred_succ_to_membar_mask(uint32_t predecessor, uint32_t successor) {
     return ((predecessor & 0x3) << 2) | (successor & 0x3);
   }
 
+  void fence(uint32_t predecessor, uint32_t successor) {
+    if (UseZtso) {
+      // do not emit fence if it's not at least a StoreLoad fence
+      if (!((predecessor & w) && (successor & r))) {
+        return;
+      }
+    }
+    if (UseConservativeFence) {
+      // extend rw -> iorw:
+      // 01(w) -> 0101(ow)
+      // 10(r) -> 1010(ir)
+      // 11(rw)-> 1111(iorw)
+      predecessor |= (predecessor & 0b11) << 2;
+      successor |= (successor & 0b11) << 2;
+    }
+    Assembler::fence(predecessor, successor);
+  }
+
   // prints msg, dumps registers and stops execution
   void stop(const char* msg);