Skip to content
21 changes: 21 additions & 0 deletions src/hotspot/share/opto/callGenerator.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -729,6 +729,27 @@ void CallGenerator::do_late_inline_helper() {
}
C->set_inlining_progress(true);
C->set_do_cleanup(kit.stopped()); // path is dead; needs cleanup

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I suggest to extract the code into a helper method (e.g., needs_cleanup) and call it from here (C->set_do_cleanup(needs_cleanup(...));)

// If the specific field accessed by a memory operation is discovered after
// inlining a method incrementally, request a cleanup so that the
// corresponding IGVN-recorded address type is updated. This is not just an
// optimization: failure to update the address type can lead to a slice
// mismatch when parsing subsequent accesses to the address, because the
// memory slice corresponding to *any* field of a class K is not the same as
// the slice corresponding to a specific field of K. This mismatch can in
// its turn lead to e.g. incorrect memory graphs.
if (C->inlining_incrementally() &&

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is C->inlining_incrementally() == true? It's CallGenerator::do_late_inline_helper() after all.

!result->is_top() && result->is_Con() && result->bottom_type()->isa_intptr_t()) {

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

isa_intptr_t() looks fishy. I don't see how it depends on bitness. It originates from byte codes where sizes are explicit and hard-coded (Java int vs long).

What about cases involving int->long casts? Does it catch them?

Node* result_proj = call->proj_out_or_null(TypeFunc::Parms);
if (result_proj != nullptr) {
for (DUIterator_Fast imax, i = result_proj->fast_outs(imax); i < imax; i++) {
Node* use = result_proj->fast_out(i);
if (use->is_AddP() && use->in(AddPNode::Offset) == result_proj) {
C->set_do_cleanup(true);
break;
}
}
}
}
kit.replace_call(call, result, true, do_asserts);
}
}
Expand Down
Original file line number Diff line number Diff line change
@@ -0,0 +1,225 @@
/*
* Copyright (c) 2026, 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
* under the terms of the GNU General Public License version 2 only, as
* published by the Free Software Foundation.
*
* This code is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
* version 2 for more details (a copy is included in the LICENSE file that
* accompanied this code).
*
* You should have received a copy of the GNU General Public License version
* 2 along with this work; if not, write to the Free Software Foundation,
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
*
* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
* or visit www.oracle.com if you need additional information or have any
* questions.
*/

package compiler.inlining;

import java.lang.reflect.Field;
import jdk.internal.misc.Unsafe;
import jdk.test.lib.Asserts;

/**
* @test
* @bug 8374783
* @summary Test that address type refinements after an incremental inlining
* step are propagated by IGVN before the next step. Failing to
* propagate such refinements could lead to slice mismatches between
* field-derived and IGVN-recorded address types when parsing bytecode
* in subsequent inlining steps.
* @library /test/lib
* @modules java.base/jdk.internal.misc
* @run main ${test.main.class}
* @run main/othervm -Xbatch
-XX:CompileCommand=compileonly,${test.main.class}::test*
-XX:CompileCommand=dontinline,${test.main.class}::notInlined*
-XX:CompileCommand=delayinline,${test.main.class}::late*
${test.main.class}
*/

class A {
int f;
}

public class TestLateInliningWithSliceNarrowing {

private static Unsafe UNSAFE = Unsafe.getUnsafe();
private static final long F_OFFSET;
private static final long INT_ARRAY_OFFSET;

static {
try {
Field fField = A.class.getDeclaredField("f");
F_OFFSET = UNSAFE.objectFieldOffset(fField);
} catch (Exception e) {
throw new RuntimeException(e);
}
INT_ARRAY_OFFSET = UNSAFE.arrayBaseOffset(int[].class);
}

static A notInlinedId(A a) {
return a;
}

static long lateOffset() {
return F_OFFSET;
}

static long lateOffsetMinusFour() {
return F_OFFSET - 4;
}

static long lateOffsetDividedByTwo() {
return F_OFFSET / 2;
}

static long lateArrayOffset() {
return INT_ARRAY_OFFSET;
}

static void lateStore(A a) {
a.f = 42;
}

static void lateArrayStore(int[] a) {
a[0] = 42;
}

static int lateLoad(A a) {
return a.f;
}

static Object lateBase(A a) {
return a;
}

// Test that when lateStore() is inlined, the IGVN-recorded type of the
// accessed memory address (captured by an AddP) has been updated to reflect
// the compiler-known offset discovered by inlining lateOffset(). Failure to
// do so leads to a slice mismatch when parsing the inlined store.
static int testLoadFromLateDiscoveredOffsetThenStoreAtConstOffset(A a) {
long o = lateOffset();
int val = UNSAFE.getInt(a, o);
lateStore(a);
return val;
}

// Test that when lateLoad() is inlined, the IGVN-recorded type of the
// accessed memory address (captured by an AddP) has been updated to reflect
// the compiler-known offset discovered by inlining lateOffset(). Failure to
// do so leads to a slice mismatch when parsing the inlined load.
static int testLoadFromLateDiscoveredOffsetThenLoadFromConstOffset(A a) {
long o = lateOffset();
int val = UNSAFE.getInt(a, o);
lateLoad(a);
return val;
}

// Test a variation of the above where lateOffsetMinusFour() is not used
// directly by an AddP node. This test does not require updating the
// IGVN-recorded type of the accessed memory address for correctness,
// because lateStore() does not reuse the corresponding AddP node.
static int testLoadFromLateDiscoveredOffsetPlusFourThenStoreAtConstOffset(A a) {
long o = lateOffsetMinusFour();
int val = UNSAFE.getInt(a, o + 4);
lateStore(a);
return val;
}

// Test a variation of the above using a different arithmetic operation,
// with the same expectations.
static int testLoadFromLateDiscoveredOffsetTimesTwoThenStoreAtConstOffset(A a) {
long o = lateOffsetDividedByTwo();
int val = UNSAFE.getInt(a, o * 2);
lateStore(a);
return val;
}

// Test a variation of the first test where failing to update the
// IGVN-recorded type of the accessed memory address would result in a slice
// mismatch that will lead to an incorrect memory graph (the memory input of
// the last load would bypass the memory output of the store).
static int testLoadFromLateDiscoveredOffsetThenStoreAtConstOffsetThenReloadFromConstOffset(A a) {
A a2 = notInlinedId(a);
long o = lateOffset();
int val = UNSAFE.getInt(a, o);
lateStore(a);
return a2.f + val;
}

// Test a variation of the first test where the offset is compiler-known
// from the beginning, but the unsafe base address is only discovered by
// inlining lateBase(). This variation does not require a cleanup between
// the late inlining of lateBase() and lateLoad() for correctness: a slice
// mismatch cannot occur because the memory access within lateLoad() does
// not reuse the same address node (AddP) as the unsafe load. The unsafe
// load address node is not reusable by the lateLoad() access because it is
// obscured by casts by the time lateLoad() is late inlined. Making the
// address node reusable by both loads would require a cleanup round, which
// would prevent the mismatch from happening in the first place.
static int testLoadFromLateDiscoveredBaseThenLoadFromKnownBase(A a) {
Object obj = lateBase(a);
int val = UNSAFE.getInt(obj, F_OFFSET);
lateLoad(a);
return val;
}

// Test a variation of the first test using an array instead of a class
// instance. No slice mismatch occurs because the address types for both
// memory accesses lead to the same slice, regardless of whether the offset
// is compiler-known.
static int testArrayLoadFromLateDiscoveredOffsetThenStoreAtConstOffset(int[] a) {
long o = lateArrayOffset();
int val = UNSAFE.getInt(a, o);
lateArrayStore(a);
return val;
}

public static void main(String[] args) {
for (int i = 0; i < 10_000; i++) {
{
A a = new A();
int result = testLoadFromLateDiscoveredOffsetThenStoreAtConstOffset(a);
Asserts.assertEquals(0, result);
}
{
A a = new A();
int result = testLoadFromLateDiscoveredOffsetThenLoadFromConstOffset(a);
Asserts.assertEquals(0, result);
}
{
A a = new A();
int result = testLoadFromLateDiscoveredOffsetPlusFourThenStoreAtConstOffset(a);
Asserts.assertEquals(0, result);
}
{
A a = new A();
int result = testLoadFromLateDiscoveredOffsetTimesTwoThenStoreAtConstOffset(a);
Asserts.assertEquals(0, result);
}
{
A a = new A();
int result = testLoadFromLateDiscoveredOffsetThenStoreAtConstOffsetThenReloadFromConstOffset(a);
Asserts.assertEquals(42, result);
}
{
A a = new A();
int result = testLoadFromLateDiscoveredBaseThenLoadFromKnownBase(a);
Asserts.assertEquals(0, result);
}
{
int[] a = new int[1];
int result = testArrayLoadFromLateDiscoveredOffsetThenStoreAtConstOffset(a);
Asserts.assertEquals(0, result);
}
}
}
}