aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorBill Schmidt <wschmidt@linux.vnet.ibm.com>2015-12-01 16:10:45 +0000
committerWilliam Schmidt <wschmidt@gcc.gnu.org>2015-12-01 16:10:45 +0000
commit3b47f4cc104bd8ab91eb1d72faf129f13f94b3c0 (patch)
tree2a306bf5c0894d0b129fc5c8cdc905005e5119f5
parent37d5ad46dde4e1cf71bd94f39f66e1fa98591222 (diff)
downloadgcc-3b47f4cc104bd8ab91eb1d72faf129f13f94b3c0.zip
gcc-3b47f4cc104bd8ab91eb1d72faf129f13f94b3c0.tar.gz
gcc-3b47f4cc104bd8ab91eb1d72faf129f13f94b3c0.tar.bz2
rs6000.c (const_load_sequence_p): Handle extra indirection for large and small code models.
[gcc] 2015-12-01 Bill Schmidt <wschmidt@linux.vnet.ibm.com> * config/rs6000/rs6000.c (const_load_sequence_p): Handle extra indirection for large and small code models. (adjust_vperm): Likewise. [gcc/testsuite] 2015-12-01 Bill Schmidt <wschmidt@linux.vnet.ibm.com> * gcc.target/powerpc/swaps-p8-22.c: New. From-SVN: r231119
-rw-r--r--gcc/ChangeLog6
-rw-r--r--gcc/config/rs6000/rs6000.c18
-rw-r--r--gcc/testsuite/ChangeLog4
-rw-r--r--gcc/testsuite/gcc.target/powerpc/swaps-p8-22.c29
4 files changed, 55 insertions, 2 deletions
diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index ce77e8b..e582c8a 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,9 @@
+2015-12-01 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
+
+ * config/rs6000/rs6000.c (const_load_sequence_p): Handle extra
+ indirection for large and small code models.
+ (adjust_vperm): Likewise.
+
2015-12-01 Julian Brown <julian@codesourcery.com>
Cesar Philippidis <cesar@codesourcery.com>
James Norris <James_Norris@mentor.com>
diff --git a/gcc/config/rs6000/rs6000.c b/gcc/config/rs6000/rs6000.c
index c4e3f2f..d451d76 100644
--- a/gcc/config/rs6000/rs6000.c
+++ b/gcc/config/rs6000/rs6000.c
@@ -36613,7 +36613,12 @@ const_load_sequence_p (swap_web_entry *insn_entry, rtx insn)
rtx base, offset;
if (GET_CODE (tocrel_body) != SET)
return false;
- if (!toc_relative_expr_p (SET_SRC (tocrel_body), false))
+ /* There is an extra level of indirection for small/large
+ code models. */
+ rtx tocrel_expr = SET_SRC (tocrel_body);
+ if (GET_CODE (tocrel_expr) == MEM)
+ tocrel_expr = XEXP (tocrel_expr, 0);
+ if (!toc_relative_expr_p (tocrel_expr, false))
return false;
split_const (XVECEXP (tocrel_base, 0, 0), &base, &offset);
if (GET_CODE (base) != SYMBOL_REF || !CONSTANT_POOL_ADDRESS_P (base))
@@ -37294,10 +37299,19 @@ adjust_vperm (rtx_insn *insn)
to set tocrel_base; otherwise it would be unnecessary as we've
already established it will return true. */
rtx base, offset;
- if (!toc_relative_expr_p (SET_SRC (PATTERN (tocrel_insn)), false))
+ rtx tocrel_expr = SET_SRC (PATTERN (tocrel_insn));
+ /* There is an extra level of indirection for small/large code models. */
+ if (GET_CODE (tocrel_expr) == MEM)
+ tocrel_expr = XEXP (tocrel_expr, 0);
+ if (!toc_relative_expr_p (tocrel_expr, false))
gcc_unreachable ();
split_const (XVECEXP (tocrel_base, 0, 0), &base, &offset);
rtx const_vector = get_pool_constant (base);
+ /* With the extra indirection, get_pool_constant will produce the
+ real constant from the reg_equal expression, so get the real
+ constant. */
+ if (GET_CODE (const_vector) == SYMBOL_REF)
+ const_vector = get_pool_constant (const_vector);
gcc_assert (GET_CODE (const_vector) == CONST_VECTOR);
/* Create an adjusted mask from the initial mask. */
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index e009034..44055d1 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,7 @@
+2015-12-01 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
+
+ * gcc.target/powerpc/swaps-p8-22.c: New.
+
2015-12-01 Marek Polacek <polacek@redhat.com>
PR tree-optimization/67916
diff --git a/gcc/testsuite/gcc.target/powerpc/swaps-p8-22.c b/gcc/testsuite/gcc.target/powerpc/swaps-p8-22.c
new file mode 100644
index 0000000..4be217b
--- /dev/null
+++ b/gcc/testsuite/gcc.target/powerpc/swaps-p8-22.c
@@ -0,0 +1,29 @@
+/* { dg-do compile { target { powerpc64le-*-* } } } */
+/* { dg-skip-if "do not override -mcpu" { powerpc*-*-* } { "-mcpu=*" } { "-mcpu=power8" } }
+/* { dg-options "-O2 -mcpu=power8 -maltivec -mcmodel=large" } */
+
+/* The expansion for vector character multiply introduces a vperm operation.
+ This tests that changing the vperm mask allows us to remove all swaps
+ from the generated code. It is a duplicate of swaps-p8-21.c, except
+ that it applies the large code model, which requires an extra indirection
+ in the load of the constant mask. */
+
+#include <altivec.h>
+
+void abort ();
+
+vector unsigned char r;
+vector unsigned char v =
+ { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 };
+vector unsigned char i =
+ { 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16 };
+
+int main ()
+{
+ int j;
+ r = v * i;
+ return 0;
+}
+
+/* { dg-final { scan-assembler-times "vperm" 1 } } */
+/* { dg-final { scan-assembler-not "xxpermdi" } } */