ART: Release inputs in Long.reverse intrinsic in x86
In the worst case we are using two temps each for input and output.
Then we do not have a temp left over for the swap operations. The
input is dead, however, after the first swap. So try to release it
(a no-op if it isn't actually a temp).
Bug: 22324327
Change-Id: I1fc50159afdad14160e34abeaf4670958171d6b2
diff --git a/test/082-inline-execute/src/Main.java b/test/082-inline-execute/src/Main.java
index 4dfa73c..177c5a4 100644
--- a/test/082-inline-execute/src/Main.java
+++ b/test/082-inline-execute/src/Main.java
@@ -1000,6 +1000,45 @@
Assert.assertEquals(Long.reverse(0x8765432187654321L), 0x84c2a6e184c2a6e1L);
Assert.assertEquals(Long.reverse(Long.MAX_VALUE), 0xfffffffffffffffeL);
Assert.assertEquals(Long.reverse(Long.MIN_VALUE), 1L);
+
+ Assert.assertEquals(test_Long_reverse_b22324327(0xaaaaaaaaaaaaaaaaL, 0x5555555555555555L),
+ 157472205507277347L);
+ }
+
+ // A bit more complicated than the above. Use local variables to stress register allocation.
+ private static long test_Long_reverse_b22324327(long l1, long l2) {
+ // A couple of local integers. Use them in a loop, so they get promoted.
+ int i1 = 0, i2 = 1, i3 = 2, i4 = 3, i5 = 4, i6 = 5, i7 = 6, i8 = 7;
+ for (int k = 0; k < 10; k++) {
+ i1 += 1;
+ i2 += 2;
+ i3 += 3;
+ i4 += 4;
+ i5 += 5;
+ i6 += 6;
+ i7 += 7;
+ i8 += 8;
+ }
+
+ // Do the Long.reverse() calls, save the results.
+ long r1 = Long.reverse(l1);
+ long r2 = Long.reverse(l2);
+
+ // Some more looping with the ints.
+ for (int k = 0; k < 10; k++) {
+ i1 += 1;
+ i2 += 2;
+ i3 += 3;
+ i4 += 4;
+ i5 += 5;
+ i6 += 6;
+ i7 += 7;
+ i8 += 8;
+ }
+
+ // Include everything in the result, so things are kept live. Try to be a little bit clever to
+ // avoid things being folded somewhere.
+ return (r1 / i1) + (r2 / i2) + i3 + i4 + i5 + i6 + i7 + i8;
}
static Object runtime;