From: Matheus Ferst <matheus.ferst@eldorado.org.br>
The extract64 arguments are not endian dependent as they are only used
for bitwise operations. The current behavior in little-endian hosts is
correct; since the indexes in VRB are in PowerISA-ordering, we should
always invert the value before calling extract64. Also, using the VsrD
macro, we can have a single EXTRACT_BIT definition for big and
little-endian with the correct behavior.
Signed-off-by: Matheus Ferst <matheus.ferst@eldorado.org.br>
---
Found this bug while refactoring VECTOR_FOR_INORDER_I uses. The
complete patch series will also use Vsr[DB] instead of VBPERM[DQ]_INDEX,
but it will need more testing. For now, we're just changing what is
necessary to fix the instruction.
---
target/ppc/int_helper.c | 5 ++---
1 file changed, 2 insertions(+), 3 deletions(-)
diff --git a/target/ppc/int_helper.c b/target/ppc/int_helper.c
index 105b626d1b..4c5d3f03f8 100644
--- a/target/ppc/int_helper.c
+++ b/target/ppc/int_helper.c
@@ -1307,14 +1307,13 @@ XXGENPCV(XXGENPCVDM, 8)
#define VBPERMQ_INDEX(avr, i) ((avr)->u8[(i)])
#define VBPERMD_INDEX(i) (i)
#define VBPERMQ_DW(index) (((index) & 0x40) != 0)
-#define EXTRACT_BIT(avr, i, index) (extract64((avr)->u64[i], index, 1))
#else
#define VBPERMQ_INDEX(avr, i) ((avr)->u8[15 - (i)])
#define VBPERMD_INDEX(i) (1 - i)
#define VBPERMQ_DW(index) (((index) & 0x40) == 0)
-#define EXTRACT_BIT(avr, i, index) \
- (extract64((avr)->u64[1 - i], 63 - index, 1))
#endif
+#define EXTRACT_BIT(avr, i, index) \
+ (extract64((avr)->VsrD(i), 63 - index, 1))
void helper_vbpermd(ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b)
{