[PATCH] PR fortran/83900 -- follow-up to set type

[PATCH] PR fortran/83900 -- follow-up to set type

[Steve Kargl]

[-- Attachment #1: Type: text/plain, Size: 542 bytes --]

The attache patch is a follow-up to my patch for
PR fortran/83900, which removed a bogus assert.
This allowed gfc_simplify_matmul to do its job,
except it did not properly set the type of the
result.  The attach fixes that issue. 

Regression tested on x86_64-*-freebsd.
OK to commit?


2018-01-20  Steven G. Kargl  <kargl@gcc.gnu.org>

	PR fortran/83900
    * simplify.c (gfc_simplify_matmul): Set return type correctly.

2018-01-20  Steven G. Kargl  <kargl@gcc.gnu.org>

	PR fortran/83900
	* gfortran.dg/matmul_18.f90: New test.

-- 
Steve

[-- Attachment #2: z.diff --]
[-- Type: text/x-diff, Size: 1478 bytes --]

Index: gcc/fortran/simplify.c
===================================================================
--- gcc/fortran/simplify.c	(revision 256912)
+++ gcc/fortran/simplify.c	(working copy)
@@ -4590,9 +4590,23 @@ gfc_simplify_matmul (gfc_expr *matrix_a, gfc_expr *mat
       || !is_constant_array_expr (matrix_b))
     return NULL;
 
-  result = gfc_get_array_expr (matrix_a->ts.type,
-			       matrix_a->ts.kind,
-			       &matrix_a->where);
+  /* MATMUL should do mixed-mode arithmetic.  Set the result type.  */
+  if (matrix_a->ts.type != matrix_b->ts.type)
+    {
+      gfc_expr e;
+      e.expr_type = EXPR_OP;
+      gfc_clear_ts (&e.ts);
+      e.value.op.op = INTRINSIC_NONE;
+      e.value.op.op1 = matrix_a;
+      e.value.op.op2 = matrix_b;
+      gfc_type_convert_binary (&e, 1);
+      result = gfc_get_array_expr (e.ts.type, e.ts.kind, &matrix_a->where);
+    }
+  else
+    {
+      result = gfc_get_array_expr (matrix_a->ts.type, matrix_a->ts.kind,
+				   &matrix_a->where);
+    }
 
   if (matrix_a->rank == 1 && matrix_b->rank == 2)
     {
Index: gcc/testsuite/gfortran.dg/matmul_18.f90
===================================================================
--- gcc/testsuite/gfortran.dg/matmul_18.f90	(nonexistent)
+++ gcc/testsuite/gfortran.dg/matmul_18.f90	(working copy)
@@ -0,0 +1,8 @@
+! { dg-do run }
+program p
+   integer, parameter :: a(3,2) = 1
+   real, parameter :: b(2,3) = 2
+   real d(3,3)
+   d = 4
+   if (any(d /= matmul(a,b))) call abort
+end

Re: [PATCH] PR fortran/83900 -- follow-up to set type

[Jerry DeLisle]

On 01/20/2018 12:19 PM, Steve Kargl wrote:
> The attache patch is a follow-up to my patch for
> PR fortran/83900, which removed a bogus assert.
> This allowed gfc_simplify_matmul to do its job,
> except it did not properly set the type of the
> result.  The attach fixes that issue. 
> 
> Regression tested on x86_64-*-freebsd.
> OK to commit?
> 
> 
> 2018-01-20  Steven G. Kargl  <kargl@gcc.gnu.org>
> 
> 	PR fortran/83900
>     * simplify.c (gfc_simplify_matmul): Set return type correctly.
> 
> 2018-01-20  Steven G. Kargl  <kargl@gcc.gnu.org>
> 
> 	PR fortran/83900
> 	* gfortran.dg/matmul_18.f90: New test.
> 

OK, thanks for fix.

Jerry