diff mbox series

: Fix PR testsuite/116271, gcc.dg/vect/tsvc/vect-tsvc-s176.c fails

Message ID 19da5a41-fbbb-45dd-9df7-3768fb88b2cc@riscy-ip.com
State New
Headers show
Series : Fix PR testsuite/116271, gcc.dg/vect/tsvc/vect-tsvc-s176.c fails | expand

Commit Message

Joern Wolfgang Rennecke Aug. 9, 2024, 12:01 a.m. UTC 
In the previous patch to reduce iteration counts, I have overlooked 
that, in the inner loop of s176, the array index i+m-j-1
turns negativeat for higher iterations of the middle loop for small m.
m and the iteration end of the middle loop should stay the same.
Fix PR testsuite/116271, gcc.dg/vect/tsvc/vect-tsvc-s176.c fails

2024-08-07  Joern Rennecke  <joern.rennecke@riscy-ip.com>

gcc/testsuite:
	PR testsuite/116271
	* gcc.dg/vect/tsvc/vect-tsvc-s176.c [TRUNCATE_TEST]: Make sure
	that m stays the same as the loop bound of the middle loop.
	* gcc.dg/vect/tsvc/tsvc.h (get_expected_result): <s176> [TRUNCATE_TEST]:	Adjust expected value.

Comments

Richard Biener Aug. 20, 2024, 7:59 a.m. UTC | #1 
On Fri, Aug 9, 2024 at 2:03 AM Joern Wolfgang Rennecke
<joern.rennecke@riscy-ip.com> wrote:
>
> In the previous patch to reduce iteration counts, I have overlooked
> that, in the inner loop of s176, the array index i+m-j-1
> turns negativeat for higher iterations of the middle loop for small m.
> m and the iteration end of the middle loop should stay the same.

OK.
diff mbox series

Patch

Index: gcc.dg/vect/tsvc/tsvc.h
===================================================================
--- gcc.dg/vect/tsvc/tsvc.h	(revision 6682)
+++ gcc.dg/vect/tsvc/tsvc.h	(working copy)
@@ -1727,7 +1727,7 @@  real_t get_expected_result(const char * name)
 #ifndef TRUNCATE_TEST
         return 32021.121094f;
 #else /* TRUNCATE_TEST */
-        return 32024.082031f;
+        return 32023.751953f;
 #endif /* TRUNCATE_TEST */
 #endif /* iterations */
     } else {
Index: gcc.dg/vect/tsvc/vect-tsvc-s176.c
===================================================================
--- gcc.dg/vect/tsvc/vect-tsvc-s176.c	(revision 6682)
+++ gcc.dg/vect/tsvc/vect-tsvc-s176.c	(working copy)
@@ -15,18 +15,20 @@  real_t s176(struct args_t * func_args)
 
     initialise_arrays(__func__);
 
-    int m = LEN_1D/2;
 #ifdef TRUNCATE_TEST
-    /* Do something equivalent to if (1) which the compiler is unlikely to
-       figure out.
-       FUNC_ARGS is in the caller's frame, so it shouldn't be between A and B.
-     */
-    if ((void *)func_args <= (void *)a || (void *)func_args >= (void *)b)
-	m = 32;
+/* Reduce the iteration counts without changing what is a variable and
+   what is a constant expression.
+   32000/25 == 640, i.e. it still has a nice power of two factor, but is
+   not a power of two itself, and still somewhat large-ish, so hopefully
+   this won't perturb the vectorizer decisions much.  */
+#define M_CONST LEN_1D/50
+#else
+#define M_CONST LEN_1D/2
 #endif
+    int m = M_CONST;
     
     for (int nl = 0; nl < 4*(10*iterations/LEN_1D); nl++) {
-        for (int j = 0; j < (LEN_1D/2); j++) {
+        for (int j = 0; j < (M_CONST); j++) {
             for (int i = 0; i < m; i++) {
                 a[i] += b[i+m-j-1] * c[j];
             }