diff mbox series

[libfortran] Fix PR 88235, buffer overrun in matmul

Message ID 5fcb79b6-415f-7b0e-7ff3-2bbd5610ad68@tkoenig.net
State New
Headers show
Series [libfortran] Fix PR 88235, buffer overrun in matmul | expand

Commit Message

Thomas König April 6, 2018, 4:52 p.m. UTC 
Hello world,

the attached patch fixes a buffer overrun in matmul, an 8 regression.
No test case since this was only detectable with the address sanitizer
or with valgrind.

Regression-tested on trunk. OK?

Regards

	Thomas

2018-04-06  Thomas Koenig  <tkoenig@gcc.gnu.org>

         PR libfortran/85253
         * m4/matmul_internal.m4: If ycount == 1, add one more row to
         the internal buffer.
         * generated/matmul_c10.c: Regenerated.
         * generated/matmul_c16.c: Regenerated.
         * generated/matmul_c4.c: Regenerated.
         * generated/matmul_c8.c: Regenerated.
         * generated/matmul_i1.c: Regenerated.
         * generated/matmul_i16.c: Regenerated.
         * generated/matmul_i2.c: Regenerated.
         * generated/matmul_i4.c: Regenerated.
         * generated/matmul_i8.c: Regenerated.
         * generated/matmul_r10.c: Regenerated.
         * generated/matmul_r16.c: Regenerated.
         * generated/matmul_r4.c: Regenerated.
         * generated/matmul_r8.c: Regenerated.
         * generated/matmulavx128_c10.c: Regenerated.
         * generated/matmulavx128_c16.c: Regenerated.
         * generated/matmulavx128_c4.c: Regenerated.
         * generated/matmulavx128_c8.c: Regenerated.
         * generated/matmulavx128_i1.c: Regenerated.
         * generated/matmulavx128_i16.c: Regenerated.
         * generated/matmulavx128_i2.c: Regenerated.
         * generated/matmulavx128_i4.c: Regenerated.
         * generated/matmulavx128_i8.c: Regenerated.
         * generated/matmulavx128_r10.c: Regenerated.
         * generated/matmulavx128_r16.c: Regenerated.
         * generated/matmulavx128_r4.c: Regenerated.
         * generated/matmulavx128_r8.c: Regenerated.

Comments

Steve Kargl April 6, 2018, 6:37 p.m. UTC | #1 
On Fri, Apr 06, 2018 at 06:52:36PM +0200, Thomas König wrote:
> 
> the attached patch fixes a buffer overrun in matmul, an 8 regression.
> No test case since this was only detectable with the address sanitizer
> or with valgrind.
> 
> Regression-tested on trunk. OK?
> 

Yes.
diff mbox series

Patch

Index: m4/matmul_internal.m4
===================================================================
--- m4/matmul_internal.m4	(Revision 259152)
+++ m4/matmul_internal.m4	(Arbeitskopie)
@@ -234,7 +234,7 @@  sinclude(`matmul_asm_'rtype_code`.m4')dnl
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmul_c10.c
===================================================================
--- generated/matmul_c10.c	(Revision 259152)
+++ generated/matmul_c10.c	(Arbeitskopie)
@@ -318,7 +318,7 @@  matmul_c10_avx (gfc_array_c10 * const restrict ret
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -870,7 +870,7 @@  matmul_c10_avx2 (gfc_array_c10 * const restrict re
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1422,7 +1422,7 @@  matmul_c10_avx512f (gfc_array_c10 * const restrict
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1988,7 +1988,7 @@  matmul_c10_vanilla (gfc_array_c10 * const restrict
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -2614,7 +2614,7 @@  matmul_c10 (gfc_array_c10 * const restrict retarra
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmul_c16.c
===================================================================
--- generated/matmul_c16.c	(Revision 259152)
+++ generated/matmul_c16.c	(Arbeitskopie)
@@ -318,7 +318,7 @@  matmul_c16_avx (gfc_array_c16 * const restrict ret
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -870,7 +870,7 @@  matmul_c16_avx2 (gfc_array_c16 * const restrict re
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1422,7 +1422,7 @@  matmul_c16_avx512f (gfc_array_c16 * const restrict
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1988,7 +1988,7 @@  matmul_c16_vanilla (gfc_array_c16 * const restrict
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -2614,7 +2614,7 @@  matmul_c16 (gfc_array_c16 * const restrict retarra
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmul_c4.c
===================================================================
--- generated/matmul_c4.c	(Revision 259152)
+++ generated/matmul_c4.c	(Arbeitskopie)
@@ -318,7 +318,7 @@  matmul_c4_avx (gfc_array_c4 * const restrict retar
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -870,7 +870,7 @@  matmul_c4_avx2 (gfc_array_c4 * const restrict reta
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1422,7 +1422,7 @@  matmul_c4_avx512f (gfc_array_c4 * const restrict r
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1988,7 +1988,7 @@  matmul_c4_vanilla (gfc_array_c4 * const restrict r
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -2614,7 +2614,7 @@  matmul_c4 (gfc_array_c4 * const restrict retarray,
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmul_c8.c
===================================================================
--- generated/matmul_c8.c	(Revision 259152)
+++ generated/matmul_c8.c	(Arbeitskopie)
@@ -318,7 +318,7 @@  matmul_c8_avx (gfc_array_c8 * const restrict retar
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -870,7 +870,7 @@  matmul_c8_avx2 (gfc_array_c8 * const restrict reta
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1422,7 +1422,7 @@  matmul_c8_avx512f (gfc_array_c8 * const restrict r
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1988,7 +1988,7 @@  matmul_c8_vanilla (gfc_array_c8 * const restrict r
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -2614,7 +2614,7 @@  matmul_c8 (gfc_array_c8 * const restrict retarray,
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmul_i1.c
===================================================================
--- generated/matmul_i1.c	(Revision 259152)
+++ generated/matmul_i1.c	(Arbeitskopie)
@@ -318,7 +318,7 @@  matmul_i1_avx (gfc_array_i1 * const restrict retar
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -870,7 +870,7 @@  matmul_i1_avx2 (gfc_array_i1 * const restrict reta
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1422,7 +1422,7 @@  matmul_i1_avx512f (gfc_array_i1 * const restrict r
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1988,7 +1988,7 @@  matmul_i1_vanilla (gfc_array_i1 * const restrict r
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -2614,7 +2614,7 @@  matmul_i1 (gfc_array_i1 * const restrict retarray,
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmul_i16.c
===================================================================
--- generated/matmul_i16.c	(Revision 259152)
+++ generated/matmul_i16.c	(Arbeitskopie)
@@ -318,7 +318,7 @@  matmul_i16_avx (gfc_array_i16 * const restrict ret
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -870,7 +870,7 @@  matmul_i16_avx2 (gfc_array_i16 * const restrict re
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1422,7 +1422,7 @@  matmul_i16_avx512f (gfc_array_i16 * const restrict
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1988,7 +1988,7 @@  matmul_i16_vanilla (gfc_array_i16 * const restrict
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -2614,7 +2614,7 @@  matmul_i16 (gfc_array_i16 * const restrict retarra
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmul_i2.c
===================================================================
--- generated/matmul_i2.c	(Revision 259152)
+++ generated/matmul_i2.c	(Arbeitskopie)
@@ -318,7 +318,7 @@  matmul_i2_avx (gfc_array_i2 * const restrict retar
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -870,7 +870,7 @@  matmul_i2_avx2 (gfc_array_i2 * const restrict reta
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1422,7 +1422,7 @@  matmul_i2_avx512f (gfc_array_i2 * const restrict r
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1988,7 +1988,7 @@  matmul_i2_vanilla (gfc_array_i2 * const restrict r
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -2614,7 +2614,7 @@  matmul_i2 (gfc_array_i2 * const restrict retarray,
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmul_i4.c
===================================================================
--- generated/matmul_i4.c	(Revision 259152)
+++ generated/matmul_i4.c	(Arbeitskopie)
@@ -318,7 +318,7 @@  matmul_i4_avx (gfc_array_i4 * const restrict retar
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -870,7 +870,7 @@  matmul_i4_avx2 (gfc_array_i4 * const restrict reta
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1422,7 +1422,7 @@  matmul_i4_avx512f (gfc_array_i4 * const restrict r
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1988,7 +1988,7 @@  matmul_i4_vanilla (gfc_array_i4 * const restrict r
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -2614,7 +2614,7 @@  matmul_i4 (gfc_array_i4 * const restrict retarray,
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmul_i8.c
===================================================================
--- generated/matmul_i8.c	(Revision 259152)
+++ generated/matmul_i8.c	(Arbeitskopie)
@@ -318,7 +318,7 @@  matmul_i8_avx (gfc_array_i8 * const restrict retar
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -870,7 +870,7 @@  matmul_i8_avx2 (gfc_array_i8 * const restrict reta
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1422,7 +1422,7 @@  matmul_i8_avx512f (gfc_array_i8 * const restrict r
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1988,7 +1988,7 @@  matmul_i8_vanilla (gfc_array_i8 * const restrict r
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -2614,7 +2614,7 @@  matmul_i8 (gfc_array_i8 * const restrict retarray,
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmul_r10.c
===================================================================
--- generated/matmul_r10.c	(Revision 259152)
+++ generated/matmul_r10.c	(Arbeitskopie)
@@ -318,7 +318,7 @@  matmul_r10_avx (gfc_array_r10 * const restrict ret
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -870,7 +870,7 @@  matmul_r10_avx2 (gfc_array_r10 * const restrict re
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1422,7 +1422,7 @@  matmul_r10_avx512f (gfc_array_r10 * const restrict
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1988,7 +1988,7 @@  matmul_r10_vanilla (gfc_array_r10 * const restrict
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -2614,7 +2614,7 @@  matmul_r10 (gfc_array_r10 * const restrict retarra
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmul_r16.c
===================================================================
--- generated/matmul_r16.c	(Revision 259152)
+++ generated/matmul_r16.c	(Arbeitskopie)
@@ -318,7 +318,7 @@  matmul_r16_avx (gfc_array_r16 * const restrict ret
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -870,7 +870,7 @@  matmul_r16_avx2 (gfc_array_r16 * const restrict re
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1422,7 +1422,7 @@  matmul_r16_avx512f (gfc_array_r16 * const restrict
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1988,7 +1988,7 @@  matmul_r16_vanilla (gfc_array_r16 * const restrict
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -2614,7 +2614,7 @@  matmul_r16 (gfc_array_r16 * const restrict retarra
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmul_r4.c
===================================================================
--- generated/matmul_r4.c	(Revision 259152)
+++ generated/matmul_r4.c	(Arbeitskopie)
@@ -318,7 +318,7 @@  matmul_r4_avx (gfc_array_r4 * const restrict retar
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -870,7 +870,7 @@  matmul_r4_avx2 (gfc_array_r4 * const restrict reta
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1422,7 +1422,7 @@  matmul_r4_avx512f (gfc_array_r4 * const restrict r
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1988,7 +1988,7 @@  matmul_r4_vanilla (gfc_array_r4 * const restrict r
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -2614,7 +2614,7 @@  matmul_r4 (gfc_array_r4 * const restrict retarray,
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmul_r8.c
===================================================================
--- generated/matmul_r8.c	(Revision 259152)
+++ generated/matmul_r8.c	(Arbeitskopie)
@@ -318,7 +318,7 @@  matmul_r8_avx (gfc_array_r8 * const restrict retar
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -870,7 +870,7 @@  matmul_r8_avx2 (gfc_array_r8 * const restrict reta
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1422,7 +1422,7 @@  matmul_r8_avx512f (gfc_array_r8 * const restrict r
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -1988,7 +1988,7 @@  matmul_r8_vanilla (gfc_array_r8 * const restrict r
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -2614,7 +2614,7 @@  matmul_r8 (gfc_array_r8 * const restrict retarray,
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmulavx128_c10.c
===================================================================
--- generated/matmulavx128_c10.c	(Revision 259152)
+++ generated/matmulavx128_c10.c	(Arbeitskopie)
@@ -283,7 +283,7 @@  matmul_c10_avx128_fma3 (gfc_array_c10 * const rest
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -836,7 +836,7 @@  matmul_c10_avx128_fma4 (gfc_array_c10 * const rest
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmulavx128_c16.c
===================================================================
--- generated/matmulavx128_c16.c	(Revision 259152)
+++ generated/matmulavx128_c16.c	(Arbeitskopie)
@@ -283,7 +283,7 @@  matmul_c16_avx128_fma3 (gfc_array_c16 * const rest
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -836,7 +836,7 @@  matmul_c16_avx128_fma4 (gfc_array_c16 * const rest
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmulavx128_c4.c
===================================================================
--- generated/matmulavx128_c4.c	(Revision 259152)
+++ generated/matmulavx128_c4.c	(Arbeitskopie)
@@ -283,7 +283,7 @@  matmul_c4_avx128_fma3 (gfc_array_c4 * const restri
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -836,7 +836,7 @@  matmul_c4_avx128_fma4 (gfc_array_c4 * const restri
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmulavx128_c8.c
===================================================================
--- generated/matmulavx128_c8.c	(Revision 259152)
+++ generated/matmulavx128_c8.c	(Arbeitskopie)
@@ -283,7 +283,7 @@  matmul_c8_avx128_fma3 (gfc_array_c8 * const restri
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -836,7 +836,7 @@  matmul_c8_avx128_fma4 (gfc_array_c8 * const restri
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmulavx128_i1.c
===================================================================
--- generated/matmulavx128_i1.c	(Revision 259152)
+++ generated/matmulavx128_i1.c	(Arbeitskopie)
@@ -283,7 +283,7 @@  matmul_i1_avx128_fma3 (gfc_array_i1 * const restri
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -836,7 +836,7 @@  matmul_i1_avx128_fma4 (gfc_array_i1 * const restri
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmulavx128_i16.c
===================================================================
--- generated/matmulavx128_i16.c	(Revision 259152)
+++ generated/matmulavx128_i16.c	(Arbeitskopie)
@@ -283,7 +283,7 @@  matmul_i16_avx128_fma3 (gfc_array_i16 * const rest
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -836,7 +836,7 @@  matmul_i16_avx128_fma4 (gfc_array_i16 * const rest
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmulavx128_i2.c
===================================================================
--- generated/matmulavx128_i2.c	(Revision 259152)
+++ generated/matmulavx128_i2.c	(Arbeitskopie)
@@ -283,7 +283,7 @@  matmul_i2_avx128_fma3 (gfc_array_i2 * const restri
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -836,7 +836,7 @@  matmul_i2_avx128_fma4 (gfc_array_i2 * const restri
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmulavx128_i4.c
===================================================================
--- generated/matmulavx128_i4.c	(Revision 259152)
+++ generated/matmulavx128_i4.c	(Arbeitskopie)
@@ -283,7 +283,7 @@  matmul_i4_avx128_fma3 (gfc_array_i4 * const restri
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -836,7 +836,7 @@  matmul_i4_avx128_fma4 (gfc_array_i4 * const restri
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmulavx128_i8.c
===================================================================
--- generated/matmulavx128_i8.c	(Revision 259152)
+++ generated/matmulavx128_i8.c	(Arbeitskopie)
@@ -283,7 +283,7 @@  matmul_i8_avx128_fma3 (gfc_array_i8 * const restri
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -836,7 +836,7 @@  matmul_i8_avx128_fma4 (gfc_array_i8 * const restri
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmulavx128_r10.c
===================================================================
--- generated/matmulavx128_r10.c	(Revision 259152)
+++ generated/matmulavx128_r10.c	(Arbeitskopie)
@@ -283,7 +283,7 @@  matmul_r10_avx128_fma3 (gfc_array_r10 * const rest
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -836,7 +836,7 @@  matmul_r10_avx128_fma4 (gfc_array_r10 * const rest
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmulavx128_r16.c
===================================================================
--- generated/matmulavx128_r16.c	(Revision 259152)
+++ generated/matmulavx128_r16.c	(Arbeitskopie)
@@ -283,7 +283,7 @@  matmul_r16_avx128_fma3 (gfc_array_r16 * const rest
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -836,7 +836,7 @@  matmul_r16_avx128_fma4 (gfc_array_r16 * const rest
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmulavx128_r4.c
===================================================================
--- generated/matmulavx128_r4.c	(Revision 259152)
+++ generated/matmulavx128_r4.c	(Arbeitskopie)
@@ -283,7 +283,7 @@  matmul_r4_avx128_fma3 (gfc_array_r4 * const restri
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -836,7 +836,7 @@  matmul_r4_avx128_fma4 (gfc_array_r4 * const restri
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
Index: generated/matmulavx128_r8.c
===================================================================
--- generated/matmulavx128_r8.c	(Revision 259152)
+++ generated/matmulavx128_r8.c	(Arbeitskopie)
@@ -283,7 +283,7 @@  matmul_r8_avx128_fma3 (gfc_array_r8 * const restri
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;
 
@@ -836,7 +836,7 @@  matmul_r8_avx128_fma4 (gfc_array_r8 * const restri
 
       /* Adjust size of t1 to what is needed.  */
       index_type t1_dim;
-      t1_dim = (a_dim1-1) * 256 + b_dim1;
+      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
       if (t1_dim > 65536)
 	t1_dim = 65536;