8286197: C2: Optimize MemorySegment shape in int loop
Reviewed-by: kvn, thartmann
This commit is contained in:
parent
94b473e464
commit
dae4c493e4
@ -371,6 +371,34 @@ void CastIINode::dump_spec(outputStream* st) const {
|
||||
}
|
||||
#endif
|
||||
|
||||
Node* CastLLNode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
||||
Node* progress = ConstraintCastNode::Ideal(phase, can_reshape);
|
||||
if (progress != NULL) {
|
||||
return progress;
|
||||
}
|
||||
// transform (CastLL (ConvI2L ..)) into (ConvI2L (CastII ..)) if the type of the CastLL is narrower than the type of
|
||||
// the ConvI2L.
|
||||
Node* in1 = in(1);
|
||||
if (in1 != NULL && in1->Opcode() == Op_ConvI2L) {
|
||||
const Type* t = Value(phase);
|
||||
const Type* t_in = phase->type(in1);
|
||||
if (t != Type::TOP && t_in != Type::TOP) {
|
||||
const TypeLong* tl = t->is_long();
|
||||
const TypeLong* t_in_l = t_in->is_long();
|
||||
assert(tl->_lo >= t_in_l->_lo && tl->_hi <= t_in_l->_hi, "CastLL type should be narrower than or equal to the type of its input");
|
||||
assert((tl != t_in_l) == (tl->_lo > t_in_l->_lo || tl->_hi < t_in_l->_hi), "if type differs then this nodes's type must be narrower");
|
||||
if (tl != t_in_l) {
|
||||
const TypeInt* ti = TypeInt::make(checked_cast<jint>(tl->_lo), checked_cast<jint>(tl->_hi), tl->_widen);
|
||||
Node* castii = phase->transform(new CastIINode(in(0), in1->in(1), ti));
|
||||
Node* convi2l = in1->clone();
|
||||
convi2l->set_req(1, castii);
|
||||
return convi2l;
|
||||
}
|
||||
}
|
||||
}
|
||||
return NULL;
|
||||
}
|
||||
|
||||
//=============================================================================
|
||||
//------------------------------Identity---------------------------------------
|
||||
// If input is already higher or equal to cast type, then this is an identity.
|
||||
|
@ -118,6 +118,7 @@ public:
|
||||
init_class_id(Class_CastLL);
|
||||
}
|
||||
|
||||
virtual Node* Ideal(PhaseGVN* phase, bool can_reshape);
|
||||
virtual int Opcode() const;
|
||||
virtual uint ideal_reg() const { return Op_RegL; }
|
||||
};
|
||||
|
@ -1077,6 +1077,11 @@ Node *PhaseIdealLoop::split_if_with_blocks_pre( Node *n ) {
|
||||
(n_blk->is_LongCountedLoop() && n->Opcode() == Op_AddL)) {
|
||||
return n;
|
||||
}
|
||||
// Pushing a shift through the iv Phi can get in the way of addressing optimizations or range check elimination
|
||||
if (n_blk->is_BaseCountedLoop() && n->Opcode() == Op_LShift(n_blk->as_BaseCountedLoop()->bt()) &&
|
||||
n->in(1) == n_blk->as_BaseCountedLoop()->phi()) {
|
||||
return n;
|
||||
}
|
||||
|
||||
// Check for having no control input; not pinned. Allow
|
||||
// dominating control.
|
||||
|
@ -0,0 +1,130 @@
|
||||
/*
|
||||
* Copyright (c) 2022, Red Hat, Inc. 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.c2.irTests;
|
||||
|
||||
import compiler.lib.ir_framework.*;
|
||||
import jdk.test.lib.Utils;
|
||||
import jdk.internal.misc.Unsafe;
|
||||
import java.util.Objects;
|
||||
import java.util.Random;
|
||||
|
||||
/*
|
||||
* @test
|
||||
* @bug 8286197
|
||||
* @key randomness
|
||||
* @summary C2: Optimize MemorySegment shape in int loop
|
||||
* @modules java.base/jdk.internal.misc
|
||||
* @library /test/lib /
|
||||
* @run driver compiler.c2.irTests.TestConvI2LCastLongLoop
|
||||
*/
|
||||
|
||||
public class TestConvI2LCastLongLoop {
|
||||
private static final Unsafe UNSAFE = Unsafe.getUnsafe();
|
||||
private static final Random RANDOM = Utils.getRandomInstance();
|
||||
|
||||
public static void main(String[] args) {
|
||||
TestFramework.runWithFlags("--add-modules", "java.base", "--add-exports", "java.base/jdk.internal.misc=ALL-UNNAMED", "-XX:LoopMaxUnroll=0", "-XX:-UseCountedLoopSafepoints");
|
||||
}
|
||||
|
||||
static int size = 1024;
|
||||
static long base = UNSAFE.allocateMemory(size * 4);
|
||||
|
||||
@Test
|
||||
@IR(failOn = { IRNode.CAST_LL })
|
||||
public static int test1() {
|
||||
// Make sure enough round of loop opts are executed
|
||||
for (int i = 0; i < 10; i++) {
|
||||
for (int j = 0; j < 10; j++) {
|
||||
for (int k = 0; k < 10; k++) {
|
||||
}
|
||||
}
|
||||
}
|
||||
int v = 0;
|
||||
// In order to optimize the range check the loop is
|
||||
// transformed to:
|
||||
//
|
||||
// for (int i1;;) {
|
||||
// for (int i2;;) {
|
||||
// long j = (i1 + i2) * UNSAFE.ARRAY_INT_INDEX_SCALE; // (i1 + i2) << 2
|
||||
// v += UNSAFE.getInt(base + j);
|
||||
// }
|
||||
// }
|
||||
//
|
||||
// (i1 + i2) << 2 is transformed to (i1 << 2) + (i2 << 2)
|
||||
// because i1 is loop invariant in the inner loop.
|
||||
//
|
||||
// long j = ... really is (CastLL (Convi2L ...))
|
||||
//
|
||||
// With that transformed into (ConvI2L (CastII ...)), The AddL
|
||||
// (i1 << 2) + (i2 << 2) can be pushed through the CastII and
|
||||
// ConvI2L.
|
||||
// The address of the getInt is then:
|
||||
// (AddP base (AddL I V)) with I, loop invariant and V loop invariant
|
||||
// which can be transformed into:
|
||||
// (AddP (AddP base I) V)
|
||||
// The inner AddP is computed out of loop
|
||||
for (int i = 0; i < size; i++) {
|
||||
long j = i * UNSAFE.ARRAY_INT_INDEX_SCALE;
|
||||
|
||||
j = Objects.checkIndex(j, size * 4);
|
||||
|
||||
if (((base + j) & 3) != 0) {
|
||||
throw new RuntimeException();
|
||||
}
|
||||
|
||||
v += UNSAFE.getInt(base + j);
|
||||
}
|
||||
return v;
|
||||
}
|
||||
|
||||
@Test
|
||||
@IR(counts = { IRNode.CAST_II, ">=1", IRNode.CONV_I2L, ">=1" })
|
||||
@IR(failOn = { IRNode.CAST_LL })
|
||||
public static long test2(int i) {
|
||||
// Convert (CastLL (ConvI2L ...)) into (ConvI2L (CastII ...))
|
||||
long j = i * UNSAFE.ARRAY_INT_INDEX_SCALE;
|
||||
j = Objects.checkIndex(j, size * 4);
|
||||
return j;
|
||||
}
|
||||
|
||||
@Run(test = "test2")
|
||||
public static void test2_runner() {
|
||||
int i = RANDOM.nextInt(size);
|
||||
long res = test2(i);
|
||||
if (res != i * UNSAFE.ARRAY_INT_INDEX_SCALE) {
|
||||
throw new RuntimeException("incorrect result: " + res);
|
||||
}
|
||||
}
|
||||
|
||||
@Test
|
||||
@IR(counts = { IRNode.PHI, "2" })
|
||||
public static int test3() {
|
||||
int v = 0;
|
||||
// splif if should not push LshiftI through the iv Phi
|
||||
for (int i = 0; i < 1024; i++) {
|
||||
v += i * UNSAFE.ARRAY_INT_INDEX_SCALE;
|
||||
}
|
||||
return v;
|
||||
}
|
||||
}
|
@ -181,7 +181,10 @@ public class IRNode {
|
||||
public static final String CON_L = START + "ConL" + MID + END;
|
||||
public static final String CONV_I2L = START + "ConvI2L" + MID + END;
|
||||
public static final String CONV_L2I = START + "ConvL2I" + MID + END;
|
||||
public static final String CAST_II = START + "CastII" + MID + END;
|
||||
public static final String CAST_LL = START + "CastLL" + MID + END;
|
||||
public static final String POPCOUNT_L = START + "PopCountL" + MID + END;
|
||||
public static final String PHI = START + "Phi" + MID + END;
|
||||
|
||||
public static final String VECTOR_CAST_B2X = START + "VectorCastB2X" + MID + END;
|
||||
public static final String VECTOR_CAST_S2X = START + "VectorCastS2X" + MID + END;
|
||||
|
Loading…
x
Reference in New Issue
Block a user