[PATCH] rs6000: Move rs6000_vect_nonmem into target cost_data

[PATCH] rs6000: Move rs6000_vect_nonmem into target cost_data

[Kewen.Lin]

[-- Attachment #1: Type: text/plain, Size: 657 bytes --]

Hi,

This patch is to move rs6000_vect_nonmem (target cost_data
related information) into target cost_data struct.
Following Richi's comments in the thread[1], we can gather
data from add_stmt_cost invocations.  This is one pre-step
to centralize target cost_data related stuffs. 

Is it ok for trunk?

BR,
Kewen

[1] https://gcc.gnu.org/pipermail/gcc-patches/2021-May/569836.html
-----
gcc/ChangeLog:

	* config/rs6000/rs6000.c (rs6000_vect_nonmem): Renamed to
	vect_nonmem and moved into...
	(struct rs6000_cost_data): ...here.
	(rs6000_init_cost): Use vect_nonmem of cost_data instead.
	(rs6000_add_stmt_cost): Likewise.
	(rs6000_finish_cost): Likewise.

[-- Attachment #2: 0001-rs6000-Move-rs6000_vect_nonmem-into-target-cost_data.patch --]
[-- Type: text/plain, Size: 1933 bytes --]

 gcc/config/rs6000/rs6000.c | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)

diff --git a/gcc/config/rs6000/rs6000.c b/gcc/config/rs6000/rs6000.c
index 48b8efd732b..e2ff00145c5 100644
--- a/gcc/config/rs6000/rs6000.c
+++ b/gcc/config/rs6000/rs6000.c
@@ -5231,6 +5231,9 @@ typedef struct _rs6000_cost_data
 {
   struct loop *loop_info;
   unsigned cost[3];
+  /* For each vectorized loop, this var holds TRUE iff a non-memory vector
+     instruction is needed by the vectorization.  */
+  bool vect_nonmem;
 } rs6000_cost_data;
 
 /* Test for likely overcommitment of vector hardware resources.  If a
@@ -5288,10 +5291,6 @@ rs6000_density_test (rs6000_cost_data *data)
 
 /* Implement targetm.vectorize.init_cost.  */
 
-/* For each vectorized loop, this var holds TRUE iff a non-memory vector
-   instruction is needed by the vectorization.  */
-static bool rs6000_vect_nonmem;
-
 static void *
 rs6000_init_cost (struct loop *loop_info)
 {
@@ -5300,7 +5299,7 @@ rs6000_init_cost (struct loop *loop_info)
   data->cost[vect_prologue] = 0;
   data->cost[vect_body]     = 0;
   data->cost[vect_epilogue] = 0;
-  rs6000_vect_nonmem = false;
+  data->vect_nonmem = false;
   return data;
 }
 
@@ -5360,7 +5359,7 @@ rs6000_add_stmt_cost (class vec_info *vinfo, void *data, int count,
 	   || kind == vec_promote_demote || kind == vec_construct
 	   || kind == scalar_to_vec)
 	  || (where == vect_body && kind == vector_stmt))
-	rs6000_vect_nonmem = true;
+	cost_data->vect_nonmem = true;
     }
 
   return retval;
@@ -5415,7 +5414,7 @@ rs6000_finish_cost (void *data, unsigned *prologue_cost,
   if (cost_data->loop_info)
     {
       loop_vec_info vec_info = loop_vec_info_for_loop (cost_data->loop_info);
-      if (!rs6000_vect_nonmem
+      if (!cost_data->vect_nonmem
 	  && LOOP_VINFO_VECT_FACTOR (vec_info) == 2
 	  && LOOP_REQUIRES_VERSIONING (vec_info))
 	cost_data->cost[vect_body] += 10000;
-- 
2.17.1

Re: [PATCH] rs6000: Move rs6000_vect_nonmem into target cost_data

[Segher Boessenkool]

Hi!

On Sat, May 08, 2021 at 04:19:34PM +0800, Kewen.Lin wrote:
> This patch is to move rs6000_vect_nonmem (target cost_data
> related information) into target cost_data struct.

> 	* config/rs6000/rs6000.c (rs6000_vect_nonmem): Renamed to
> 	vect_nonmem and moved into...
> 	(struct rs6000_cost_data): ...here.
> 	(rs6000_init_cost): Use vect_nonmem of cost_data instead.
> 	(rs6000_add_stmt_cost): Likewise.
> 	(rs6000_finish_cost): Likewise.

This is okay for trunk.  Thanks!


Segher