diff mbox series

Unshare DR_STEP before gimplifying it

Message ID mptpng8if8n.fsf@arm.com
State New
Headers show
Series Unshare DR_STEP before gimplifying it | expand

Commit Message

Richard Sandiford Dec. 28, 2019, 4:31 p.m. UTC 
In this testcase we use an unmasked SVE loop with an Advanced SIMD
epilogue (because we don't yet support fully-masked downward loops).
The main loop uses a gather load for the strided access while the
epilogue loop builds the access from scalars instead.  In both cases
we gimplify expressions based on the DR_STEP and insert them in the
loop preheader.

The problem was that the gather load code didn't copy the DR_STEP before
gimplifying it, meaning that the epilogue loop tried to reuse a result
from the (non-dominating) main loop preheader.

It looks at first glance like there could be other instances of this too,
but this patch just deals with the gather/scatter case.

Tested on aarch64-linux-gnu and x86_64-linux-gnu.  OK to install?

Richard


2019-12-28  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
	* tree-vect-stmts.c (vect_get_strided_load_store_ops): Copy
	DR_STEP before gimplifying it.

gcc/testsuite/
	* gcc.dg/vect/vect-strided-epilogue-1.c: New test.

Comments

Richard Biener Dec. 28, 2019, 4:39 p.m. UTC | #1 
On December 28, 2019 5:31:20 PM GMT+01:00, Richard Sandiford <richard.sandiford@arm.com> wrote:
>In this testcase we use an unmasked SVE loop with an Advanced SIMD
>epilogue (because we don't yet support fully-masked downward loops).
>The main loop uses a gather load for the strided access while the
>epilogue loop builds the access from scalars instead.  In both cases
>we gimplify expressions based on the DR_STEP and insert them in the
>loop preheader.
>
>The problem was that the gather load code didn't copy the DR_STEP
>before
>gimplifying it, meaning that the epilogue loop tried to reuse a result
>from the (non-dominating) main loop preheader.
>
>It looks at first glance like there could be other instances of this
>too,
>but this patch just deals with the gather/scatter case.
>
>Tested on aarch64-linux-gnu and x86_64-linux-gnu.  OK to install?

Ok. 

Richard. 

>Richard
>
>
>2019-12-28  Richard Sandiford  <richard.sandiford@arm.com>
>
>gcc/
>	* tree-vect-stmts.c (vect_get_strided_load_store_ops): Copy
>	DR_STEP before gimplifying it.
>
>gcc/testsuite/
>	* gcc.dg/vect/vect-strided-epilogue-1.c: New test.
>
>Index: gcc/tree-vect-stmts.c
>===================================================================
>--- gcc/tree-vect-stmts.c	2019-12-28 16:28:16.344143831 +0000
>+++ gcc/tree-vect-stmts.c	2019-12-28 16:29:52.003499967 +0000
>@@ -2993,7 +2993,7 @@ vect_get_strided_load_store_ops (stmt_ve
>   gimple_seq stmts;
> 
>   tree bump = size_binop (MULT_EXPR,
>-			  fold_convert (sizetype, DR_STEP (dr)),
>+			  fold_convert (sizetype, unshare_expr (DR_STEP (dr))),
> 			  size_int (TYPE_VECTOR_SUBPARTS (vectype)));
>  *dataref_bump = force_gimple_operand (bump, &stmts, true, NULL_TREE);
>   if (stmts)
>@@ -3005,7 +3005,7 @@ vect_get_strided_load_store_ops (stmt_ve
>   offset_type = TREE_TYPE (gs_info->offset_vectype);
> 
>/* Calculate X = DR_STEP / SCALE and convert it to the appropriate
>type.  */
>-  tree step = size_binop (EXACT_DIV_EXPR, DR_STEP (dr),
>+  tree step = size_binop (EXACT_DIV_EXPR, unshare_expr (DR_STEP (dr)),
> 			  ssize_int (gs_info->scale));
>   step = fold_convert (offset_type, step);
>   step = force_gimple_operand (step, &stmts, true, NULL_TREE);
>Index: gcc/testsuite/gcc.dg/vect/vect-strided-epilogue-1.c
>===================================================================
>--- /dev/null	2019-09-17 11:41:18.176664108 +0100
>+++ gcc/testsuite/gcc.dg/vect/vect-strided-epilogue-1.c	2019-12-28
>16:29:52.003499967 +0000
>@@ -0,0 +1,8 @@
>+/* { dg-do compile } */
>+
>+void
>+f (int *x, short *y, int z)
>+{
>+  for (int i = 0; i < 0x82; ++i)
>+    x[-i] += x[z * i];
>+}
diff mbox series

Patch

Index: gcc/tree-vect-stmts.c
===================================================================
--- gcc/tree-vect-stmts.c	2019-12-28 16:28:16.344143831 +0000
+++ gcc/tree-vect-stmts.c	2019-12-28 16:29:52.003499967 +0000
@@ -2993,7 +2993,7 @@  vect_get_strided_load_store_ops (stmt_ve
   gimple_seq stmts;
 
   tree bump = size_binop (MULT_EXPR,
-			  fold_convert (sizetype, DR_STEP (dr)),
+			  fold_convert (sizetype, unshare_expr (DR_STEP (dr))),
 			  size_int (TYPE_VECTOR_SUBPARTS (vectype)));
   *dataref_bump = force_gimple_operand (bump, &stmts, true, NULL_TREE);
   if (stmts)
@@ -3005,7 +3005,7 @@  vect_get_strided_load_store_ops (stmt_ve
   offset_type = TREE_TYPE (gs_info->offset_vectype);
 
   /* Calculate X = DR_STEP / SCALE and convert it to the appropriate type.  */
-  tree step = size_binop (EXACT_DIV_EXPR, DR_STEP (dr),
+  tree step = size_binop (EXACT_DIV_EXPR, unshare_expr (DR_STEP (dr)),
 			  ssize_int (gs_info->scale));
   step = fold_convert (offset_type, step);
   step = force_gimple_operand (step, &stmts, true, NULL_TREE);
Index: gcc/testsuite/gcc.dg/vect/vect-strided-epilogue-1.c
===================================================================
--- /dev/null	2019-09-17 11:41:18.176664108 +0100
+++ gcc/testsuite/gcc.dg/vect/vect-strided-epilogue-1.c	2019-12-28 16:29:52.003499967 +0000
@@ -0,0 +1,8 @@ 
+/* { dg-do compile } */
+
+void
+f (int *x, short *y, int z)
+{
+  for (int i = 0; i < 0x82; ++i)
+    x[-i] += x[z * i];
+}