stan/jdk-24
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

8343944: C2: MinLNode::add_ring() computes _widen wrongly leading to an endless widening/compilation

Browse Source 
Reviewed-by: thartmann, rcastanedalo
This commit is contained in:
Christian Hagedorn 2024-11-13 08:39:17 +00:00
parent e9ede464b2
commit 2eeaa57b19
2 changed files with 81 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/hotspot/share/opto/addnode.cpp
View File

@ -1546,7 +1546,7 @@ const Type* MinLNode::add_ring(const Type* t0, const Type* t1) const {
const TypeLong* r0 = t0->is_long();
const TypeLong* r1 = t1->is_long();
return TypeLong::make(MIN2(r0->_lo, r1->_lo), MIN2(r0->_hi, r1->_hi), MIN2(r0->_widen, r1->_widen));
return TypeLong::make(MIN2(r0->_lo, r1->_lo), MIN2(r0->_hi, r1->_hi), MAX2(r0->_widen, r1->_widen));
}
Node* MinLNode::Identity(PhaseGVN* phase) {

80
src/hotspot/test/hotspot/jtreg/compiler/ccp/TestWrongMinLWiden.java Normal file
View File

@ -0,0 +1,80 @@
/*
* Copyright (c) 2024, 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.
*/
/*
* @test
* @bug 8343944
* @summary Test that _widen is set correctly in MinL::add_ring() to prevent an endless widening in CCP.
* @run main/othervm -Xcomp -XX:CompileCommand=compileonly,compiler.ccp.TestWrongMinLWiden::*
* compiler.ccp.TestWrongMinLWiden
*/
package compiler.ccp;
public class TestWrongMinLWiden {
static long lFld;
static short sFld;
public static void main(String[] strArr) {
Math.min(3,3); // Make sure Math class is loaded.
test();
testWithMathMin();
}
static long test() {
long x = 50398;
for (int i = 1; i < 100; i++) {
long xMinus1 = x - 1; // x is a phi with type #long
// Long comparison:
// ConvI2L(sFld) <= xMinus1
// First converted to
// CMoveL(sFld <= xMinus1, sFld, xMinus1)
// CMoveL(ConvI2L(sFld) <= long_phi, ConvI2L(sFld), long_phi)
// with types
// CMoveL(#short <= #long, #short, #long)
// And then converted in CMoveNode::Ideal() to
// MinL(sFld, xMinus1)
// MinL(ConvI2L(sFld), long_phi)
//
// We wrongly set the _widen of the new type for MinL in MinL::add_ring() to the minimum of both types which
// is always 0 because the _widen of sFld is 0. As a result, we will endlessly widen the involved types
// because the new type for MinL keeps resetting _widen to 0.
// Instead, we should choose the maximum of both _widen fields for the new type for MinL which will then
// select the _widen of xMinus1 which will grow each time we set a new type for the long_phi. Eventually,
// we will saturate the type of long_phi to min_long to avoid an endless widening.
x = sFld <= xMinus1 ? sFld : xMinus1;
}
return x;
}
// Same as test() but with Math.min() which internally uses x <= y ? x : y which allows the CMoveL pattern to be
// replaced with MinL.
static long testWithMathMin() {
long x = 50398;
for (int i = 1; i < 100; i++) {
x = Math.min(sFld, x - 1);
}
return x;
}
}