openmp: Allow non-rectangular loops with pointer iterators

This patch handles pointer iterators for non-rectangular loops.  They are
more limited than integral iterators of non-rectangular loops, in particular
only var-outer, var-outer + a2, a2 + var-outer or var-outer - a2 can appear
in lb or ub where a2 is some integral loop invariant expression, so no e.g.
multiplication etc.

2021-10-27  Jakub Jelinek  <jakub@redhat.com>

gcc/
	* omp-expand.c (expand_omp_for_init_counts): Handle non-rectangular
	iterators with pointer types.
	(expand_omp_for_init_vars, extract_omp_for_update_vars): Likewise.
gcc/c-family/
	* c-omp.c (c_omp_check_loop_iv_r): Don't clear 3rd bit for
	POINTER_PLUS_EXPR.
	(c_omp_check_nonrect_loop_iv): Handle POINTER_PLUS_EXPR.
	(c_omp_check_loop_iv): Set kind even if the iterator is non-integral.
gcc/testsuite/
	* c-c++-common/gomp/loop-8.c: New test.
	* c-c++-common/gomp/loop-9.c: New test.
libgomp/
	* testsuite/libgomp.c/loop-26.c: New test.
	* testsuite/libgomp.c/loop-27.c: New test.

2 files changed, 376 insertions, 0 deletions

diff --git a/libgomp/testsuite/libgomp.c/loop-26.c b/libgomp/testsuite/libgomp.c/loop-26.c
new file mode 100644
index 0000000..5b30213
--- /dev/null
+++ b/libgomp/testsuite/libgomp.c/loop-26.c
@@ -0,0 +1,196 @@
+/* { dg-do run } */
+
+extern void abort (void);
+
+signed char v[5][7][9][21][4][42][3];
+int a[84];
+int *volatile zero = &a[42];
+int *volatile two = &a[42 + 2];
+int *volatile three = &a[42 + 3];
+int *volatile five = &a[42 + 5];
+int *volatile seven = &a[42 + 7];
+int *volatile nine = &a[42 + 9];
+int *volatile eleven = &a[42 + 11];
+int *volatile minusone = &a[42 - 1];
+volatile int zeroi = 0, onei = 1, twoi = 2, threei = 3, fivei = 5;
+
+int
+main ()
+{
+  for (int i = 0; i < 5; i++)
+  for (int j = 0; j < 7; j++)
+  for (int k = 0; k < 9; k++)
+  for (int l = j; l < 5 + j; l++)
+  for (int m = 7; m < 11; m++)
+  for (int n = 0; n < l - 2; n++)
+  for (int o = 0; o < 3; o++)
+    v[i][j][k][l][m - 7][n][o] = 1;
+
+  int niters = 0;
+  #pragma omp parallel
+  #pragma omp for collapse(7) reduction(+:niters)
+  for (int i = 0; i < 5; i++)
+  for (int *j = &a[42]; j < &a[42 + 7]; j++)
+  for (int *k = &a[42]; k < &a[42 + 9]; k++)
+  for (int *l = j; l < 5 + j; l++)
+  for (int *m = &a[42 + 7]; m < &a[42 + 11]; m++)
+  for (int *n = &a[42]; n < l - 2; n++)
+  for (int *o = &a[42]; o < &a[42 + 3]; o++)
+    {
+      niters++;
+      if (i < 0 || i >= 5
+	  || j - &a[42] < 0 || j - &a[42] >= 7
+	  || k - &a[42] < 0 || k - &a[42] >= 9
+	  || l - &a[42] < 0 || l >= j + 5
+	  || m - &a[42] < 7 || m - &a[42] >= 11
+	  || n - &a[42] < 0 || n >= l - 2
+	  || o - &a[42] < 0 || o - &a[42] >= 3)
+	abort ();
+      if (v[i][j - &a[42]][k - &a[42]][l - &a[42]][m - &a[42 + 7]][n - &a[42]][o - &a[42]] != 1)
+	abort ();
+      v[i][j - &a[42]][k - &a[42]][l - &a[42]][m - &a[42 + 7]][n - &a[42]][o - &a[42]]++;
+    }
+
+  if (niters != 58860)
+    abort ();
+  int niters2 = 0;
+  #pragma omp parallel
+  #pragma omp for collapse(7) reduction(+:niters2)
+  for (int *i = zero; i < five; i += onei)
+  for (int *j = seven - onei; j >= zero; j -= onei)
+  for (int *k = nine - onei; k >= zero; k += -onei)
+  for (int *l = j + zeroi; l < fivei + j; l += onei)
+  for (int *m = eleven - onei; m >= seven; m -= onei)
+  for (int *n = l - threei; n >= zero; n -= onei)
+  for (int *o = zero; o < three; o += onei)
+    {
+      niters2++;
+      if (i - &a[42] < 0 || i - &a[42] >= 5
+	  || j - &a[42] < 0 || j - &a[42] >= 7
+	  || k - &a[42] < 0 || k - &a[42] >= 9
+	  || l < j || l >= j + 5
+	  || m - &a[42] < 7 || m - &a[42] >= 11
+	  || n - &a[42] < 0 || n >= l - 2
+	  || o - &a[42] < 0 || o - &a[42] >= 3)
+	abort ();
+      if (v[i - &a[42]][j - &a[42]][k - &a[42]][l - &a[42]][m - &a[42 + 7]][n - &a[42]][o - &a[42]] != 2)
+	abort ();
+      v[i - &a[42]][j - &a[42]][k - &a[42]][l - &a[42]][m - &a[42 + 7]][n - &a[42]][o - &a[42]]++;
+    }
+
+  if (niters2 != 58860)
+    abort ();
+
+  for (int i = 0; i < 5; i++)
+  for (int j = 0; j < 7; j++)
+  for (int k = 0; k < 9; k++)
+  for (int l = j; l < 5 + j; l++)
+  for (int m = 7; m < 11; m++)
+  for (int n = 0; n < l - 2; n++)
+  for (int o = 0; o < 3; o++)
+    if (v[i][j][k][l][m - 7][n][o] != 3)
+      abort ();
+
+  int niters3 = 0;
+  #pragma omp parallel
+  #pragma omp for collapse(5) reduction(+:niters3)
+  for (int *i = &a[42 + 4]; i >= &a[42 + 0]; i--)
+  for (int *j = &a[42 + 6]; j >= &a[42 + 0]; --j)
+  for (int *l = j + 4; l >= j; l--)
+  for (int *n = l - 3; n >= &a[42]; --n)
+  for (int *o = &a[42 + 2]; o >= &a[42 + 0]; o--)
+    {
+      niters3++;
+      if (i - &a[42] < 0 || i - &a[42] >= 5
+	  || j - &a[42] < 0 || j - &a[42] >= 7
+	  || l < j || l >= j + 5
+	  || n - &a[42] < 0 || n >= l - 2
+	  || o - &a[42] < 0 || o - &a[42] >= 3)
+	abort ();
+      if (v[i - &a[42]][j - &a[42]][0][l - &a[42]][0][n - &a[42]][o - &a[42]] != 3)
+	abort ();
+      v[i - &a[42]][j - &a[42]][0][l - &a[42]][0][n - &a[42]][o - &a[42]]++;
+    }
+
+  if (niters3 != 1635)
+    abort ();
+
+  int niters4 = 0;
+  #pragma omp parallel
+  #pragma omp for collapse(5) reduction(+:niters4)
+  for (int *i = zero; i < five; i += onei)
+  for (int *j = zero; j <= seven - onei; j += onei)
+  for (int *l = zeroi + j; l < j + fivei; l += onei)
+  for (int *n = zero; n <= l - threei; n += onei)
+  for (int o = zeroi; o < threei; o += onei)
+    {
+      niters4++;
+      if (i - &a[42] < 0 || i - &a[42] >= 5
+	  || j - &a[42] < 0 || j - &a[42] >= 7
+	  || l < j || l >= j + 5
+	  || n - &a[42] < 0 || n >= l - 2
+	  || o < 0 || o >= 3)
+	abort ();
+      if (v[i - &a[42]][j - &a[42]][0][l - &a[42]][0][n - &a[42]][o] != 4)
+	abort ();
+      v[i - &a[42]][j - &a[42]][0][l - &a[42]][0][n - &a[42]][o]++;
+    }
+
+  if (niters4 != 1635)
+    abort ();
+
+  for (int i = 0; i < 5; i++)
+  for (int j = 0; j < 7; j++)
+  for (int l = j; l < j + 5; l++)
+  for (int n = 0; n < l - 2; n++)
+  for (int o = 0; o < 3; o++)
+    if (v[i][j][0][l][0][n][o] != 5)
+      abort ();
+
+  int niters5 = 0;
+  #pragma omp parallel
+  #pragma omp for collapse(3) reduction(+:niters5)
+  for (int *j = &a[42 + 6]; j >= &a[42]; --j)
+  for (int *l = j + 0; l <= j + 4; l++)
+  for (int *n = l - 3; n > &a[42 - 1]; --n)
+    {
+      niters5++;
+      if (j - &a[42] < 0 || j - &a[42] >= 7
+	  || l < j || l >= j + 5
+	  || n < &a[42] || n >= l - 2)
+	abort ();
+      if (v[0][j - &a[42]][0][l - &a[42]][0][n - &a[42]][0] != 5)
+	abort ();
+      v[0][j - &a[42]][0][l - &a[42]][0][n - &a[42]][0]++;
+    }
+
+  if (niters5 != 109)
+    abort ();
+
+  int niters6 = 0;
+  #pragma omp parallel
+  #pragma omp for collapse(3) reduction(+:niters6)
+  for (int *j = seven - onei; j > minusone; j -= onei)
+  for (int *l = j + threei + onei; l >= j; l += -onei)
+  for (int *n = l - threei; n > minusone; n -= onei)
+    {
+      niters6++;
+      if (j - &a[42] < 0 || j - &a[42] >= 7
+	  || l < j || l >= j + 5
+	  || n < &a[42] || n >= l - 2)
+	abort ();
+      if (v[0][j - &a[42]][0][l - &a[42]][0][n - &a[42]][0] != 6)
+	abort ();
+      v[0][j - &a[42]][0][l - &a[42]][0][n - &a[42]][0]++;
+    }
+
+  if (niters6 != 109)
+    abort ();
+
+  for (int j = 0; j < 7; j++)
+  for (int l = j; l < j + 5; l++)
+  for (int n = 0; n < l - 2; n++)
+    if (v[0][j][0][l][0][n][0] != 7)
+      abort ();
+  return 0;
+}
diff --git a/libgomp/testsuite/libgomp.c/loop-27.c b/libgomp/testsuite/libgomp.c/loop-27.c
new file mode 100644
index 0000000..3353b87
--- /dev/null
+++ b/libgomp/testsuite/libgomp.c/loop-27.c
@@ -0,0 +1,180 @@
+/* { dg-do run } */
+
+extern void abort (void);
+
+int x;
+short *i, *j;
+int ii, jj;
+short *volatile a;
+short *volatile b;
+volatile int c, e, f, g;
+short *volatile d;
+int k[11][20];
+short v[84];
+
+int
+main ()
+{
+  int niters;
+  for (ii = 1; ii <= 10; ii++)
+    for (jj = 1; jj <= ii + 5; jj++)
+      k[ii][jj] = 1;
+  a = &v[42 + 1]; b = &v[42 + 11]; c = 1; d = &v[42 + 1]; e = 6; f = 1;
+  niters = 0; i = &v[0]; j = &v[0]; x = -100;
+  #pragma omp parallel for collapse(2) lastprivate (i, j, x) reduction(+:niters)
+  for (i = &v[42 + 1]; i <= &v[42 + 10]; i++)
+    for (j = &v[42 + 1]; j <= i + 5; j++)
+      {
+	if (i < &v[42 + 1] || i > &v[42 + 10] || j < &v[42 + 1] || j > i + 5 || k[i - &v[42]][j - &v[42]] != 1)
+	  abort ();
+	k[i - &v[42]][j - &v[42]]++;
+	x = (i - &v[42]) * 1024 + ((j - &v[42]) & 1023);
+	niters++;
+      }
+  if (i != &v[42 + 11] || j != &v[42 + 16] || x != 10255 || niters != 105)
+    abort ();
+  niters = 0; i = &v[0]; j = &v[0]; x = -100;
+  #pragma omp parallel for collapse(2) lastprivate (i, j, x) reduction(+:niters)
+  for (i = a; i < b; i += c)
+    for (j = d; j < e + i; j += f)
+      {
+	if (i < &v[42 + 1] || i > &v[42 + 10] || j < &v[42 + 1] || j > i + 5 || k[i - &v[42]][j - &v[42]] != 2)
+	  abort ();
+	k[i - &v[42]][j - &v[42]]++;
+	x = (i - &v[42]) * 1024 + ((j - &v[42]) & 1023);
+	niters++;
+      }
+  if (i != &v[42 + 11] || j != &v[42 + 16] || x != 10255 || niters != 105)
+    abort ();
+  for (ii = 1; ii <= 10; ii++)
+    for (jj = 1; jj <= ii + 5; jj++)
+      if (k[ii][jj] == 3)
+	k[ii][jj] = 0;
+      else
+	abort ();
+  for (ii = 0; ii < 11; ii++)
+    for (jj = 0; jj < 20; jj++)
+      if (k[ii][jj] != 0)
+	abort ();
+  for (ii = 0; ii < 10; ii++)
+    for (jj = 0; jj < ii; jj++)
+      k[ii][jj] = 1;
+  a = &v[42]; b = &v[42 + 10]; c = 1; d = &v[42]; e = 0; f = 1;
+  niters = 0; i = &v[0]; j = &v[0]; x = -100;
+  #pragma omp parallel for collapse(2) lastprivate (i, j, x) reduction(+:niters)
+  for (i = &v[42]; i < &v[42 + 10]; i++)
+    for (j = &v[42]; j < i; j++)
+      {
+	if (i < &v[42] || i >= &v[42 + 10] || j < &v[42] || j >= i || k[i - &v[42]][j - &v[42]] != 1)
+	  abort ();
+	k[i - &v[42]][j - &v[42]]++;
+	x = (i - &v[42]) * 1024 + ((j - &v[42]) & 1023);
+	niters++;
+      }
+  if (i != &v[42 + 10] || j != &v[42 + 9] || x != 9224 || niters != 45)
+    abort ();
+  niters = 0; i = &v[0]; j = &v[0]; x = -100;
+  #pragma omp parallel for collapse(2) lastprivate (i, j, x) reduction(+:niters)
+  for (i = a; i < b; i += c)
+    for (j = d; j < i - e; j += f)
+      {
+	if (i < &v[42] || i >= &v[42 + 10] || j < &v[42] || j >= i || k[i - &v[42]][j - &v[42]] != 2)
+	  abort ();
+	k[i - &v[42]][j - &v[42]]++;
+	x = (i - &v[42]) * 1024 + ((j - &v[42]) & 1023);
+	niters++;
+      }
+  if (i != &v[42 + 10] || j != &v[42 + 9] || x != 9224 || niters != 45)
+    abort ();
+  for (ii = 0; ii < 10; ii++)
+    for (jj = 0; jj < ii; jj++)
+      if (k[ii][jj] == 3)
+	k[ii][jj] = 0;
+      else
+	abort ();
+  for (ii = 0; ii < 11; ii++)
+    for (jj = 0; jj < 20; jj++)
+      if (k[ii][jj] != 0)
+	abort ();
+  for (ii = 0; ii < 10; ii++)
+    for (jj = ii + 1; jj < ii + 4; jj++)
+      k[ii][jj] = 1;
+  a = &v[42]; b = &v[42 + 10]; c = 1; e = 1; f = 1; g = -4;
+  niters = 0; i = &v[0]; j = &v[0]; x = -100;
+  #pragma omp parallel for collapse(2) lastprivate (i, j, x) reduction(+:niters)
+  for (i = &v[42]; i < &v[42 + 10]; i++)
+    for (j = i + 1; j < i + 4; j++)
+      {
+	if (i < &v[42] || i >= &v[42 + 10] || j < i + 1 || j >= i + 4 || k[i - &v[42]][j - &v[42]] != 1)
+	  abort ();
+	k[i - &v[42]][j - &v[42]]++;
+	x = (i - &v[42]) * 1024 + ((j - &v[42]) & 1023);
+	niters++;
+      }
+  if (i != &v[42 + 10] || j != &v[42 + 13] || x != 9228 || niters != 30)
+    abort ();
+  niters = 0; i = &v[0]; j = &v[0]; x = -100;
+  #pragma omp parallel for collapse(2) lastprivate (i, j, x) reduction(+:niters)
+  for (i = a; i < b; i += c)
+    for (j = i + e; j < i - g; j += f)
+      {
+	if (i < &v[42] || i >= &v[42 + 10] || j < i + 1 || j >= i + 4 || k[i - &v[42]][j - &v[42]] != 2)
+	  abort ();
+	k[i - &v[42]][j - &v[42]]++;
+	x = (i - &v[42]) * 1024 + ((j - &v[42]) & 1023);
+	niters++;
+      }
+  if (i != &v[42 + 10] || j != &v[42 + 13] || x != 9228 || niters != 30)
+    abort ();
+  for (ii = 0; ii < 10; ii++)
+    for (jj = ii + 1; jj < ii + 4; jj++)
+      if (k[ii][jj] == 3)
+	k[ii][jj] = 0;
+      else
+	abort ();
+  for (ii = 0; ii < 11; ii++)
+    for (jj = 0; jj < 20; jj++)
+      if (k[ii][jj] != 0)
+	abort ();
+  for (ii = 1; ii < 10; ii += 2)
+    for (jj = 1; jj < ii + 1; jj++)
+      k[ii][jj] = 1;
+  a = &v[42 + 1]; b = &v[42 + 10]; c = 2; d = &v[42 + 1]; e = 1; f = 1;
+  niters = 0; i = &v[0]; j = &v[0]; x = -100;
+  #pragma omp parallel for collapse(2) lastprivate (i, j, x) reduction(+:niters)
+  for (i = &v[42 + 1]; i < &v[42 + 10]; i += 2)
+    for (j = &v[42 + 1]; j < i + 1; j++)
+      {
+	if (i < &v[42 + 1] || i >= &v[42 + 10] || j < &v[42 + 1] || j >= i + 1 || k[i - &v[42]][j - &v[42]] != 1)
+	  abort ();
+	k[i - &v[42]][j - &v[42]]++;
+	x = (i - &v[42]) * 1024 + ((j - &v[42]) & 1023);
+	niters++;
+      }
+  if (i != &v[42 + 11] || j != &v[42 + 10] || x != 9225 || niters != 25)
+    abort ();
+  niters = 0; i = &v[0]; j = &v[0]; x = -100;
+  #pragma omp parallel for collapse(2) lastprivate (i, j, x) reduction(+:niters)
+  for (i = a; i < b; i += c)
+    for (j = d; j < i + e; j += f)
+      {
+	if (i < &v[42 + 1] || i >= &v[42 + 10] || j < &v[42 + 1] || j >= i + 1 || k[i - &v[42]][j - &v[42]] != 2)
+	  abort ();
+	k[i - &v[42]][j - &v[42]]++;
+	x = (i - &v[42]) * 1024 + ((j - &v[42]) & 1023);
+	niters++;
+      }
+  if (i != &v[42 + 11] || j != &v[42 + 10] || x != 9225 || niters != 25)
+    abort ();
+  for (ii = 1; ii < 10; ii += 2)
+    for (jj = 1; jj < ii + 1; jj++)
+      if (k[ii][jj] == 3)
+	k[ii][jj] = 0;
+      else
+	abort ();
+  for (ii = 0; ii < 11; ii++)
+    for (jj = 0; jj < 20; jj++)
+      if (k[ii][jj] != 0)
+	abort ();
+  return 0;
+}