openmp: Optimize triangular loop logical iterator to actual iterators computation using search for quadratic equation root(s)

This patch implements the optimized logical to actual iterators
computation for triangular loops.

I have a rough implementation using integers, but this one uses floating
point.  There is a small problem that -fopenmp programs aren't linked with
-lm, so it does it only if the hw has sqrt optab (and uses ifn rather than
__builtin_sqrt because it obviously doesn't need errno handling etc.).

Do you think it is ok this way, or should I use the integral computation
using inlined isqrt (we have inequation of the form
start >= x * t10 + t11 * (((x - 1) * x) / 2)
where t10 and t11 are signed long long values and start unsigned long long,
and the division by 2 actually is a problem for accuracy in some cases, so
if we do it in integral, we need to do actually
      long long t12 = 2 * t10 - t11;
      unsigned long long t13 = t12 * t12 + start * 8 * t11;
      unsigned long long isqrt_ = isqrtull (t13);
      long long x = (((long long) isqrt_ - t12) / t11) >> 1;
with careful overflow checking on all the computations before isqrtull
(and on overflows use the fallback implementation).

2020-07-09  Jakub Jelinek  <jakub@redhat.com>

	* omp-general.h (struct omp_for_data): Add min_inner_iterations
	and factor members.
	* omp-general.c (omp_extract_for_data): Initialize them and remember
	them in OMP_CLAUSE_COLLAPSE_COUNT if needed and restore from there.
	* omp-expand.c (expand_omp_for_init_counts): Fix up computation of
	counts[fd->last_nonrect] if fd->loop.n2 is INTEGER_CST.
	(expand_omp_for_init_vars): For
	fd->first_nonrect + 1 == fd->last_nonrect loops with for now
	INTEGER_CST fd->loop.n2 find quadratic equation roots instead of
	using fallback method when possible.

	* testsuite/libgomp.c/loop-19.c: New test.
	* testsuite/libgomp.c/loop-20.c: New test.

1 files changed, 84 insertions, 0 deletions

diff --git a/libgomp/testsuite/libgomp.c/loop-20.c b/libgomp/testsuite/libgomp.c/loop-20.c
new file mode 100644
index 0000000..c3265ac
--- /dev/null
+++ b/libgomp/testsuite/libgomp.c/loop-20.c
@@ -0,0 +1,84 @@
+/* { dg-do run } */
+
+extern void abort (void);
+
+unsigned long long int x, i, j;
+volatile unsigned long long int a, b, c, d, e, f, g, h;
+int k[4][206];
+
+int
+main ()
+{
+  long long int niters;
+  for (j = ~0ULL / 2 - 32; j < ((~0ULL / 2) + 6); j++)
+    k[0][j - ~0ULL / 2 + 64] = 1;
+  a = 1; b = 2; c = 1; d = 0; e = ~0ULL / 2 - 32; f = ((~0ULL / 2) + 6); g = 0; h = 1;
+  niters = 0; i = -100; j = -100; x = -100;
+  #pragma omp parallel for collapse(2) lastprivate (i, j, x) reduction(+:niters)
+  for (i = 1; i < 2; i++)
+    for (j = ~0ULL / 2 - 32; j < i * ((~0ULL / 2) + 6); j++)
+      {
+	if (i != 1 || j < ~0ULL / 2 - 32 || j >= ((~0ULL / 2) + 6) || k[0][j - ~0ULL / 2 + 64] != 1)
+	  abort ();
+	k[0][j - ~0ULL / 2 + 64]++;
+	x = i * 1024 + (j & 1023);
+	niters++;
+      }
+  if (i != 2 || j != ((~0ULL / 2) + 6) || x != 1028 || niters != 38)
+    abort ();
+  niters = 0; i = -100; j = -100; x = -100;
+  #pragma omp parallel for collapse(2) lastprivate (i, j, x) reduction(+:niters)
+  for (i = a; i < b; i += c)
+    for (j = d * i + e; j < g + i * f; j += h)
+      {
+	if (i != 1 || j < ~0ULL / 2 - 32 || j >= ((~0ULL / 2) + 6) || k[0][j - ~0ULL / 2 + 64] != 2)
+	  abort ();
+	k[0][j - ~0ULL / 2 + 64]++;
+	x = i * 1024 + (j & 1023);
+	niters++;
+      }
+  if (i != 2 || j != ((~0ULL / 2) + 6) || x != 1028 || niters != 38)
+    abort ();
+  for (j = ~0ULL / 2 - 32; j < ((~0ULL / 2) + 6); j++)
+    if (k[0][j - ~0ULL / 2 + 64] == 3)
+      k[0][j - ~0ULL / 2 + 64] = 0;
+    else
+      abort ();
+  for (i = 1; i < 4; i++)
+    for (j = 64ULL * i; j < i * 32ULL + 110; j++)
+      k[i][j] = 1;
+  a = 1; b = 4; c = 1; d = 64ULL; e = 0; f = 32ULL; g = 110ULL; h = 1;
+  niters = 0; i = -100; j = -100; x = -100;
+  #pragma omp parallel for collapse(2) lastprivate (i, j, x) reduction(+:niters)
+  for (i = 1; i < 4; i++)
+    for (j = 64ULL * i; j < i * 32ULL + 110; j++)
+      {
+	if (i < 1 || i >= 4 || j < 64ULL * i || j >= i * 32ULL + 110 || k[i][j] != 1)
+	  abort ();
+	k[i][j]++;
+	x = i * 1024 + (j & 1023);
+	niters++;
+      }
+  if (i != 4 || j != 206 || x != 3277 || niters != 138)
+    abort ();
+  niters = 0; i = -100; j = -100; x = -100;
+  #pragma omp parallel for collapse(2) lastprivate (i, j, x) reduction(+:niters)
+  for (i = a; i < b; i += c)
+    for (j = d * i + e; j < g + i * f; j += h)
+      {
+	if (i < 1 || i >= 4 || j < 64ULL * i || j >= i * 32ULL + 110 || k[i][j] != 2)
+	  abort ();
+	k[i][j]++;
+	x = i * 1024 + (j & 1023);
+	niters++;
+      }
+  if (i != 4 || j != 206 || x != 3277 || niters != 138)
+    abort ();
+  for (i = 1; i < 4; i++)
+    for (j = 64ULL * i; j < i * 32ULL + 110; j++)
+      if (k[i][j] == 3)
+	k[i][j] = 0;
+      else
+	abort ();
+  return 0;
+}