obtained gcc-4.6.4.tar.bz2 from upstream website;upstream

verified gcc-4.6.4.tar.bz2.sig;
imported gcc-4.6.4 source tree from verified upstream tarball.

downloading a git-generated archive based on the 'upstream' tag
should provide you with a source tree that is binary identical
to the one extracted from the above tarball.

if you have obtained the source via the command 'git clone',
however, do note that line-endings of files in your working
directory might differ from line-endings of the respective
files in the upstream repository.

1 files changed, 55 insertions, 0 deletions

diff --git a/gcc/testsuite/gcc.dg/tree-ssa/prefetch-6.c b/gcc/testsuite/gcc.dg/tree-ssa/prefetch-6.c
new file mode 100644
index 000000000..7a1de5a45
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/tree-ssa/prefetch-6.c
@@ -0,0 +1,55 @@
+/* { dg-do compile { target i?86-*-* x86_64-*-* } } */
+/* { dg-require-effective-target ilp32 } */
+/* { dg-require-effective-target sse2 } */
+/* { dg-options "-O2 -fprefetch-loop-arrays -march=athlon -msse2 -mfpmath=sse --param simultaneous-prefetches=100 --param min-insn-to-prefetch-ratio=6 -fdump-tree-aprefetch-details" } */
+
+#define N 1000
+#define K 900
+
+double a[N][N];
+
+double test(void)
+{
+  unsigned i, j;
+  double sum = 0;
+
+  /* Here, we should use non-temporal prefetch instruction.  */
+  for (i = 0; i < K; i++)
+    for (j = 0; j < K; j++)
+      sum += a[i][j];
+
+  /* Here, we should not use non-temporal prefetch instruction, since the
+     value of a[i+10][j] is reused in L2 cache.  */
+  for (i = 0; i < K; i++)
+    for (j = 0; j < K; j++)
+      sum += a[i][j] * a[i + 10][j];
+
+  /* Here, we should use non-temporal prefetch instruction, since the
+     value of a[i+100][j] is too far to be reused in L2 cache.  */
+  for (i = 0; i < K; i++)
+    for (j = 0; j < K; j++)
+      sum += a[i][j] * a[i + 100][j];
+
+  /* Here, temporal prefetches should be used, since the volume of the
+     memory accesses is smaller than L2 cache.  */
+  for (i = 0; i < 100; i++)
+    for (j = 0; j < 100; j++)
+      sum += a[i][j] * a[i + 100][j];
+
+  /* Temporal prefetches should be used here (even though the accesses to
+     a[j][i] are independent, the same cache line is almost always hit
+     every N iterations).  */
+  for (i = 0; i < N; i++)
+    for (j = 0; j < N; j++)
+      sum += a[j][i];
+
+  return sum;
+}
+
+/* { dg-final { scan-tree-dump-times "Issued prefetch" 5 "aprefetch" } } */
+/* { dg-final { scan-tree-dump-times "Issued nontemporal prefetch" 3 "aprefetch" } } */
+
+/* { dg-final { scan-assembler-times "prefetcht" 5 } } */
+/* { dg-final { scan-assembler-times "prefetchnta" 3 } } */
+
+/* { dg-final { cleanup-tree-dump "aprefetch" } } */