Merge remote-tracking branch 'origin/master' into cwi/GR-38909-substitutions-move

Author: olpaw

compiler/src/org.graalvm.compiler.hotspot/src/org/graalvm/compiler/hotspot/replacements/HotSpotAllocationSnippets.java

@@ -352,7 +352,7 @@ static boolean useNullAllocationStubs(@InjectedParameter GraalHotSpotVMConfig co
     }
 
     @Override
-    protected final Object callNewArrayStub(Word hub, int length, int fillStartOffset) {
+    protected final Object callNewArrayStub(Word hub, int length) {
         KlassPointer klassPtr = KlassPointer.fromWord(hub);
         if (useNullAllocationStubs(INJECTED_VMCONFIG)) {
             return nonNullOrDeopt(newArrayOrNull(NEW_ARRAY_OR_NULL, klassPtr, length));

compiler/src/org.graalvm.compiler.replacements/src/org/graalvm/compiler/replacements/AllocationSnippets.java

@@ -93,11 +93,11 @@ public Object allocateArrayImpl(Word hub,
         if (useTLAB() && probability(FAST_PATH_PROBABILITY, shouldAllocateInTLAB(allocationSize, true)) && probability(FAST_PATH_PROBABILITY, newTop.belowOrEqual(end))) {
             writeTlabTop(thread, newTop);
             emitPrefetchAllocate(newTop, true);
-            result = formatArray(hub, allocationSize, length, top, fillContents, fillStartOffset, emitMemoryBarrier, maybeUnroll, supportsBulkZeroing, supportsOptimizedFilling,
+            result = formatArray(hub, allocationSize, length, top, fillContents, emitMemoryBarrier, fillStartOffset, maybeUnroll, supportsBulkZeroing, supportsOptimizedFilling,
                             profilingData.snippetCounters);
         } else {
             profilingData.snippetCounters.stub.inc();
-            result = callNewArrayStub(hub, length, fillStartOffset);
+            result = callNewArrayStub(hub, length);
         }
         profileAllocation(profilingData, allocationSize);
         return verifyOop(result);
@@ -124,8 +124,7 @@ protected UnsignedWord arrayAllocationSize(int length, int arrayBaseOffset, int
     public static long arrayAllocationSize(int length, int arrayBaseOffset, int log2ElementSize, int alignment) {
         long size = ((length & 0xFFFFFFFFL) << log2ElementSize) + arrayBaseOffset + (alignment - 1);
         long mask = ~(alignment - 1);
-        long result = size & mask;
-        return result;
+        return size & mask;
     }
 
     /**
@@ -268,11 +267,7 @@ public Object formatObject(Word hub,
                     AllocationSnippetCounters snippetCounters) {
         initializeObjectHeader(memory, hub, false);
         int headerSize = instanceHeaderSize();
-        if (fillContents == FillContent.WITH_ZEROES) {
-            zeroMemory(memory, headerSize, size, constantSize, false, false, false, snippetCounters);
-        } else if (REPLACEMENTS_ASSERTIONS_ENABLED && fillContents == FillContent.WITH_GARBAGE_IF_ASSERTIONS_ENABLED) {
-            fillWithGarbage(memory, headerSize, size, constantSize, false, false, snippetCounters);
-        }
+        fillContents(memory, fillContents, headerSize, size, constantSize, false, false, false, snippetCounters);
         emitMemoryBarrierIf(emitMemoryBarrier);
         return memory.toObjectNonNull();
     }
@@ -285,23 +280,33 @@ public Object formatArray(Word hub,
                     int length,
                     Word memory,
                     FillContent fillContents,
-                    int fillStartOffset,
                     boolean emitMemoryBarrier,
+                    int fillStartOffset,
                     boolean maybeUnroll,
                     boolean supportsBulkZeroing,
                     boolean supportsOptimizedFilling,
                     AllocationSnippetCounters snippetCounters) {
-        memory.writeInt(arrayLengthOffset(), length, LocationIdentity.init());
-        // Store hub last as the concurrent garbage collectors assume length is valid if hub field
-        // is not null.
+        /*
+         * For TLAB allocations, the initialization order does not matter. Therefore, it is also not
+         * necessary to use STORE_RELEASE semantics when storing the hub into the newly allocated
+         * object. This is a major difference to the slow-path allocation where the initialization
+         * order and the STORE_RELEASE semantics are crucial for concurrent GCs (the slow-path
+         * allocation can directly allocate in the old generation).
+         */
         initializeObjectHeader(memory, hub, true);
+        memory.writeInt(arrayLengthOffset(), length, LocationIdentity.init());
+        fillContents(memory, fillContents, fillStartOffset, allocationSize, false, maybeUnroll, supportsBulkZeroing, supportsOptimizedFilling, snippetCounters);
+        emitMemoryBarrierIf(emitMemoryBarrier);
+        return memory.toObjectNonNull();
+    }
+
+    private void fillContents(Word memory, FillContent fillContents, int startOffset, UnsignedWord endOffset, boolean isEndOffsetConstant, boolean maybeUnroll, boolean supportsBulkZeroing,
+                    boolean supportsOptimizedFilling, AllocationSnippetCounters snippetCounters) {
         if (fillContents == FillContent.WITH_ZEROES) {
-            zeroMemory(memory, fillStartOffset, allocationSize, false, maybeUnroll, supportsBulkZeroing, supportsOptimizedFilling, snippetCounters);
+            zeroMemory(memory, startOffset, endOffset, isEndOffsetConstant, maybeUnroll, supportsBulkZeroing, supportsOptimizedFilling, snippetCounters);
         } else if (REPLACEMENTS_ASSERTIONS_ENABLED && fillContents == FillContent.WITH_GARBAGE_IF_ASSERTIONS_ENABLED) {
-            fillWithGarbage(memory, fillStartOffset, allocationSize, false, maybeUnroll, supportsOptimizedFilling, snippetCounters);
+            fillWithGarbage(memory, startOffset, endOffset, isEndOffsetConstant, maybeUnroll, supportsOptimizedFilling, snippetCounters);
         }
-        emitMemoryBarrierIf(emitMemoryBarrier);
-        return memory.toObjectNonNull();
     }
 
     protected void emitMemoryBarrierIf(boolean emitMemoryBarrier) {
@@ -351,7 +356,7 @@ public void emitPrefetchAllocate(Word address, boolean isArray) {
 
     protected abstract Object callNewInstanceStub(Word hub);
 
-    protected abstract Object callNewArrayStub(Word hub, int length, int fillStartOffset);
+    protected abstract Object callNewArrayStub(Word hub, int length);
 
     protected abstract Object callNewMultiArrayStub(Word hub, int rank, Word dims);
 

compiler/src/org.graalvm.compiler.truffle.runtime.serviceprovider/src/org/graalvm/compiler/truffle/runtime/serviceprovider/TruffleRuntimeServices.java

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2018, 2022, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -37,6 +37,17 @@ public final class TruffleRuntimeServices {
      * @param service the service whose provider is being requested
      */
     public static <S> Iterable<S> load(Class<S> service) {
-        return ServiceLoader.load(service);
+        Class<?> lookupClass = TruffleRuntimeServices.class;
+        ModuleLayer moduleLayer = lookupClass.getModule().getLayer();
+        Iterable<S> services;
+        if (moduleLayer != null) {
+            services = ServiceLoader.load(moduleLayer, service);
+        } else {
+            services = ServiceLoader.load(service, lookupClass.getClassLoader());
+        }
+        if (!services.iterator().hasNext()) {
+            services = ServiceLoader.load(service);
+        }
+        return services;
     }
 }

sdk/llvm-patches/README.md

@@ -4,4 +4,4 @@ LLVM Upstream Patches
 This directory contains patches which were used to build this
 LLVM distribution but are not yet upstream or have been backported
 to LLVM 14. To build this LLVM distribution yourself, apply the patches
-on top of an LLVM [14.0.3](https://github.com/llvm/llvm-project/tree/llvmorg-14.0.3) source tree.
+on top of an LLVM [14.0.6](https://github.com/llvm/llvm-project/tree/llvmorg-14.0.6) source tree.

sdk/llvm-patches/native-image/0001-GR-17692-Statepoints-Support-for-compressed-pointers.patch

@@ -1,7 +1,7 @@
-From 0122431d63266f033e4fd8abf78e950c124f027f Mon Sep 17 00:00:00 2001
+From b9902a9d9f57ea78d4b11bdfd7295738ef3547fb Mon Sep 17 00:00:00 2001
 From: Loic Ottet <loic.ottet@oracle.com>
 Date: Mon, 23 Sep 2019 16:55:33 +0200
-Subject: [PATCH 1/2] [GR-17692] [Statepoints] Support for compressed pointers
+Subject: [PATCH 1/3] [GR-17692] [Statepoints] Support for compressed pointers
  in the statepoint emission pass
 
 ---
@@ -424,5 +424,5 @@ index b795ad3899bc..5a5d1d67e5f4 100644
    Out.insert(LiveOut.begin(), LiveOut.end());
  }
 -- 
-2.33.1
+2.36.0
 

sdk/llvm-patches/native-image/0002-GR-23578-AArch64-Introduce-option-to-force-placement.patch

@@ -1,7 +1,7 @@
-From 772a7a659e7f2f231920cff828cff00fab198af0 Mon Sep 17 00:00:00 2001
+From ca6facf3f34d646783d082776378da40d06e9035 Mon Sep 17 00:00:00 2001
 From: Loic Ottet <loic.ottet@oracle.com>
 Date: Tue, 8 Sep 2020 13:03:06 +0200
-Subject: [PATCH 2/2] [GR-23578][AArch64] Introduce option to force placement
+Subject: [PATCH 2/3] [GR-23578][AArch64] Introduce option to force placement
  of the frame record on top of the stack frame
 
 ---
@@ -34,5 +34,5 @@ index d1b901e58d27..96038fbb8f28 100644
  }
 
 -- 
-2.33.1
+2.36.0