[PATCH GCC][3/4]Add support for constant operand in pattern (convert (op:s (convert@2 @0) (convert?@3 @1)))

Bin.Cheng amker.cheng@gmail.com
Wed Oct 26 14:01:00 GMT 2016


On Tue, Oct 25, 2016 at 1:00 PM, Richard Biener
<richard.guenther@gmail.com> wrote:
> On Tue, Oct 25, 2016 at 1:21 PM, Bin Cheng <Bin.Cheng@arm.com> wrote:
>> Hi,
>> This is an update patch for https://gcc.gnu.org/ml/gcc-patches/2016-10/msg00738.html .  In this version, existing pattern (convert (op:s (convert@2 @0) (convert?@3 @1))) is extended.  It allows narrowing of arithmetic operation which has constant integer as its second operand.  It also simplifies next patch handling cond_expr.
>> Bootstrap and test on x86_64 and AArch64 for whole patch set.  Is it OK?
>
> +        && types_match (@0, type)
> +        && (types_match (@0, @1)
> +            /* Or the second operand must be constant integer.  */
> +            || (@3 == @1
> +                && types_match (@1, @2)
> +                && TREE_CODE (@1) == INTEGER_CST)))
>
> So this fails to match the pattern if we get into it via valueization
> and get, say,
> (short)((int)a + (int)7).  I believe for plus and minus we're always safe so
> I suggest to simply do
>
>   && (types_match (@0, @1)
>         || TREE_CODE (@1) == INTEGER_CST)
>
>        (if (TYPE_OVERFLOW_WRAPS (TREE_TYPE (@0)))
> -       (convert (op @0 @1))
> +       (convert (op @0 (convert:type @1)))
>
> :type shouldn't be necessary -- it also shows the outer (convert ..)
> is not required,
> please remove it while you're here.
>
>         (with { tree utype = unsigned_type_for (TREE_TYPE (@0)); }
> -        (convert (op (convert:utype @0) (convert:utype @1))))))))
> +        (convert (op (convert:utype @0)
> +                     (convert:utype (convert:type @1)))))))))
>
> Why do you need the intermediate conversion?

Thanks for reviewing, updated patch attached.  Is it OK?

Thanks,
bin
-------------- next part --------------
diff --git a/gcc/match.pd b/gcc/match.pd
index 01d088d..dbe11bc 100644
--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -3328,7 +3328,7 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
    operation and convert the result to the desired type.  */
 (for op (plus minus)
   (simplify
-    (convert (op:s (convert@2 @0) (convert@3 @1)))
+    (convert (op:s (convert@2 @0) (convert?@3 @1)))
     (if (INTEGRAL_TYPE_P (type)
 	 /* We check for type compatibility between @0 and @1 below,
 	    so there's no need to check that @1/@3 are integral types.  */
@@ -3344,12 +3344,16 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
 	 && TYPE_PRECISION (type) == GET_MODE_PRECISION (TYPE_MODE (type))
 	 /* The inner conversion must be a widening conversion.  */
 	 && TYPE_PRECISION (TREE_TYPE (@2)) > TYPE_PRECISION (TREE_TYPE (@0))
-	 && types_match (@0, @1)
-	 && types_match (@0, type))
+	 && types_match (@0, type)
+	 && (types_match (@0, @1)
+	     /* Or the second operand is const integer or converted const
+		integer from valueize.  */
+	     || TREE_CODE (@1) == INTEGER_CST))
       (if (TYPE_OVERFLOW_WRAPS (TREE_TYPE (@0)))
-	(convert (op @0 @1))
+	(op @0 (convert @1))
 	(with { tree utype = unsigned_type_for (TREE_TYPE (@0)); }
-	 (convert (op (convert:utype @0) (convert:utype @1))))))))
+	 (convert (op (convert:utype @0)
+		      (convert:utype @1))))))))
 
 /* This is another case of narrowing, specifically when there's an outer
    BIT_AND_EXPR which masks off bits outside the type of the innermost
diff --git a/gcc/testsuite/gcc.dg/fold-narrowbopcst-1.c b/gcc/testsuite/gcc.dg/fold-narrowbopcst-1.c
new file mode 100644
index 0000000..8a33677
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/fold-narrowbopcst-1.c
@@ -0,0 +1,14 @@
+/* { dg-do compile } */
+/* { dg-options "-O1 -fdump-tree-optimized" } */
+
+int foo1 (unsigned char a[], unsigned int x)
+{
+  unsigned int i;
+  for (i = 0; i < 1000; i++)
+    {
+      x = a[i];
+      a[i] = (unsigned char)(x >= 100 ? x - 100 : 0);
+    }
+  return x;
+}
+/* { dg-final { scan-tree-dump " = _.* \\+ 156" "optimized" } } */


More information about the Gcc-patches mailing list