Add support for vector permute cost since various permutes can expand

into a complex sequence of instructions.  This fixes major performance
regressions due to recent changes in SLP vectorizer (which now vectorizes
more aggressively and emits many complex permutes).  Set the cost to > 1
for all microarchitectures so that the number of permutes is usually zero
and regressions disappear.

2016-01-28  Wilco Dijkstra  <wdijkstr@arm.com>

	* config/aarch64/aarch64.c (generic_vector_cost):
	Set vec_permute_cost.
	(cortexa57_vector_cost): Likewise.
	(exynosm1_vector_cost): Likewise.
	(xgene1_vector_cost): Likewise.
	(aarch64_builtin_vectorization_cost): Use vec_permute_cost.
	* config/aarch64/aarch64-protos.h (cpu_vector_cost):
	Add vec_permute_cost entry.


git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@232922 138bc75d-0d04-0410-961f-82ee72b054a4

3 files changed, 19 insertions, 1 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 99f2bdb32d7..1967e922798 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,5 +1,16 @@
 2016-01-28  Wilco Dijkstra  <wdijkstr@arm.com>
 
+	* config/aarch64/aarch64.c (generic_vector_cost):
+	Set vec_permute_cost.
+	(cortexa57_vector_cost): Likewise.
+	(exynosm1_vector_cost): Likewise.
+	(xgene1_vector_cost): Likewise.
+	(aarch64_builtin_vectorization_cost): Use vec_permute_cost.
+	* config/aarch64/aarch64-protos.h (cpu_vector_cost):
+	Add vec_permute_cost entry.
+
+2016-01-28  Wilco Dijkstra  <wdijkstr@arm.com>
+
 	* config/aarch64/aarch64.md (ccmp<mode>): Disassemble
 	immediate as %1.
 	(add<mode>3_compare0): Likewise.
diff --git a/gcc/config/aarch64/aarch64-protos.h b/gcc/config/aarch64/aarch64-protos.h
index 15fc37deb9a..bd900c6269f 100644
--- a/gcc/config/aarch64/aarch64-protos.h
+++ b/gcc/config/aarch64/aarch64-protos.h
@@ -156,9 +156,10 @@ struct cpu_vector_cost
   const int scalar_load_cost;		 /* Cost of scalar load.  */
   const int scalar_store_cost;		 /* Cost of scalar store.  */
   const int vec_stmt_cost;		 /* Cost of any vector operation,
-					    excluding load, store,
+					    excluding load, store, permute,
 					    vector-to-scalar and
 					    scalar-to-vector operation.  */
+  const int vec_permute_cost;		 /* Cost of permute operation.  */
   const int vec_to_scalar_cost;		 /* Cost of vec-to-scalar operation.  */
   const int scalar_to_vec_cost;		 /* Cost of scalar-to-vector
 					    operation.  */
diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c
index df3dec0a72b..5b3771eca67 100644
--- a/gcc/config/aarch64/aarch64.c
+++ b/gcc/config/aarch64/aarch64.c
@@ -315,6 +315,7 @@ static const struct cpu_vector_cost generic_vector_cost =
   1, /* scalar_load_cost  */
   1, /* scalar_store_cost  */
   1, /* vec_stmt_cost  */
+  2, /* vec_permute_cost  */
   1, /* vec_to_scalar_cost  */
   1, /* scalar_to_vec_cost  */
   1, /* vec_align_load_cost  */
@@ -332,6 +333,7 @@ static const struct cpu_vector_cost cortexa57_vector_cost =
   4, /* scalar_load_cost  */
   1, /* scalar_store_cost  */
   3, /* vec_stmt_cost  */
+  3, /* vec_permute_cost  */
   8, /* vec_to_scalar_cost  */
   8, /* scalar_to_vec_cost  */
   5, /* vec_align_load_cost  */
@@ -348,6 +350,7 @@ static const struct cpu_vector_cost exynosm1_vector_cost =
   5, /* scalar_load_cost  */
   1, /* scalar_store_cost  */
   3, /* vec_stmt_cost  */
+  3, /* vec_permute_cost  */
   3, /* vec_to_scalar_cost  */
   3, /* scalar_to_vec_cost  */
   5, /* vec_align_load_cost  */
@@ -365,6 +368,7 @@ static const struct cpu_vector_cost xgene1_vector_cost =
   5, /* scalar_load_cost  */
   1, /* scalar_store_cost  */
   2, /* vec_stmt_cost  */
+  2, /* vec_permute_cost  */
   4, /* vec_to_scalar_cost  */
   4, /* scalar_to_vec_cost  */
   10, /* vec_align_load_cost  */
@@ -7574,6 +7578,8 @@ aarch64_builtin_vectorization_cost (enum vect_cost_for_stmt type_of_cost,
 	return aarch64_tune_params.vec_costs->cond_not_taken_branch_cost;
 
       case vec_perm:
+	return aarch64_tune_params.vec_costs->vec_permute_cost;
+
       case vec_promote_demote:
 	return aarch64_tune_params.vec_costs->vec_stmt_cost;