[patch][vect]: simplify is_linear_load_p using vec_perm_builder and reject invalid [PR126592]

--qcw4+AymWSgpFk/J
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline

is_linear_load_p is supposed to test for valid permutations of lanes of complex
arithmetic.  It was written as a manual loop that iteratively discredited
what a permutation could be and returned the permute.

However the code is a bit hard to prove correct and as PR1265992 points out it
accepts at least one invalid permute [0 2 2 2].

To fix this and simplify the code to prevent other issues I have rewritten it to
use vec_perm_indices and use the convenient helper series_p ();

Bootstrapped Regtested on aarch64-none-linux-gnu,
arm-none-linux-gnueabihf, x86_64-pc-linux-gnu
-m32, -m64 and no issues.

Pushed.

Thanks,
Tamar

gcc/ChangeLog:

	PR tree-optimization/126592
	* tree-vect-slp-patterns.cc (is_linear_load_p): Rewrite using
	vec_perm_indices.

gcc/testsuite/ChangeLog:

	PR tree-optimization/126592
	* gcc.target/aarch64/pr126592.c: New test.

---
diff --git a/gcc/testsuite/gcc.target/aarch64/pr126592.c b/gcc/testsuite/gcc.target/aarch64/pr126592.c
new file mode 100644
index 0000000000000000000000000000000000000000..00e82c1a7954de9a96535d4027ff8fd8f781c197
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/pr126592.c
@@ -0,0 +1,53 @@
+/* { dg-do run { target arm_v8_3a_complex_neon_hw } } */
+/* { dg-require-effective-target aarch64_little_endian } */
+/* { dg-options "-O3 -march=armv8.3-a" } */
+
+#define N 64
+double a[N], b[N], c[N];
+
+__attribute__((noipa)) void
+mul (double *__restrict cc, double *__restrict aa, double *__restrict bb, int n)
+{
+  for (int i = 0; i < n; i += 4)
+    {
+      cc[i] = aa[i] * bb[i] - aa[i + 1] * bb[i + 1];
+      cc[i + 1] = aa[i + 2] * bb[i + 1] + aa[i + 1] * bb[i];
+      cc[i + 2] = aa[i + 2] * bb[i + 2] - aa[i + 3] * bb[i + 3];
+      cc[i + 3] = aa[i + 2] * bb[i + 3] + aa[i + 3] * bb[i + 2];
+    }
+}
+
+__attribute__((optimize (0))) void
+ref (double *__restrict cc, double *__restrict aa, double *__restrict bb, int n)
+{
+  for (int i = 0; i < n; i += 4)
+    {
+      cc[i] = aa[i] * bb[i] - aa[i + 1] * bb[i + 1];
+      cc[i + 1] = aa[i + 2] * bb[i + 1] + aa[i + 1] * bb[i];
+      cc[i + 2] = aa[i + 2] * bb[i + 2] - aa[i + 3] * bb[i + 3];
+      cc[i + 3] = aa[i + 2] * bb[i + 3] + aa[i + 3] * bb[i + 2];
+    }
+}
+
+int
+main (void)
+{
+  double e[N];
+
+  for (int i = 0; i < N; ++i)
+    {
+      a[i] = i + 1;
+      b[i] = i * 3 + 1;
+    }
+
+  mul (c, a, b, N);
+  ref (e, a, b, N);
+
+  for (int i = 0; i < N; ++i)
+    if (c[i] != e[i])
+      __builtin_abort ();
+
+  return 0;
+}
+
+/* { dg-final { scan-assembler-not {fcmla\t} } } */
diff --git a/gcc/tree-vect-slp-patterns.cc b/gcc/tree-vect-slp-patterns.cc
index 2136329645cad8f26f672ac5ad7d7f225bb08b7f..a40e3d29fdcb54ccf79da76f273faa8443f7654c 100644
--- a/gcc/tree-vect-slp-patterns.cc
+++ b/gcc/tree-vect-slp-patterns.cc
@@ -138,47 +138,34 @@ is_linear_load_p (load_permutation_t loads)
   if (loads.length() == 0)
     return PERM_UNKNOWN;
 
-  unsigned load, i;
-  complex_perm_kinds_t candidates[4]
-    = { PERM_ODDODD
-      , PERM_EVENEVEN
-      , PERM_EVENODD
-      , PERM_ODDEVEN
-      };
+  if (loads.length () == 1)
+    return loads[0] == 0 ? PERM_EVENEVEN : PERM_ODDODD;
 
-  int valid_patterns = 4;
-  FOR_EACH_VEC_ELT (loads, i, load)
+  vec_perm_builder builder;
+  builder.new_vector (loads.length (), loads.length (), 1);
+  for (unsigned load : loads)
     {
-      unsigned adj_load = load % 2;
-      if (candidates[0] != PERM_UNKNOWN && adj_load != 1)
-	{
-	  candidates[0] = PERM_UNKNOWN;
-	  valid_patterns--;
-	}
-      if (candidates[1] != PERM_UNKNOWN && adj_load != 0)
-	{
-	  candidates[1] = PERM_UNKNOWN;
-	  valid_patterns--;
-	}
-      if (candidates[2] != PERM_UNKNOWN && load != i)
-	{
-	  candidates[2] = PERM_UNKNOWN;
-	  valid_patterns--;
-	}
-      if (candidates[3] != PERM_UNKNOWN
-	  && load != (i % 2 == 0 ? i + 1 : i - 1))
-	{
-	  candidates[3] = PERM_UNKNOWN;
-	  valid_patterns--;
-	}
-
-      if (valid_patterns == 0)
+      if (load >= loads.length ())
 	return PERM_UNKNOWN;
+      builder.quick_push (load);
     }
 
-  for (i = 0; i < sizeof(candidates); i++)
-    if (candidates[i] != PERM_UNKNOWN)
-      return candidates[i];
+  vec_perm_indices indices (builder, 1, loads.length ());
+
+  if (indices.series_p (0, 2, 1, 2)
+      && indices.series_p (1, 2, 1, 2))
+    return PERM_ODDODD;
+
+  if (indices.series_p (0, 2, 0, 2)
+      && indices.series_p (1, 2, 0, 2))
+    return PERM_EVENEVEN;
+
+  if (indices.series_p (0, 1, 0, 1))
+    return PERM_EVENODD;
+
+  if (indices.series_p (0, 2, 1, 2)
+      && indices.series_p (1, 2, 0, 2))
+    return PERM_ODDEVEN;
 
   return PERM_UNKNOWN;
 }


-- 

--qcw4+AymWSgpFk/J
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment; filename="rb20770.patch"

diff --git a/gcc/testsuite/gcc.target/aarch64/pr126592.c b/gcc/testsuite/gcc.target/aarch64/pr126592.c
new file mode 100644
index 0000000000000000000000000000000000000000..00e82c1a7954de9a96535d4027ff8fd8f781c197
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/pr126592.c
@@ -0,0 +1,53 @@
+/* { dg-do run { target arm_v8_3a_complex_neon_hw } } */
+/* { dg-require-effective-target aarch64_little_endian } */
+/* { dg-options "-O3 -march=armv8.3-a" } */
+
+#define N 64
+double a[N], b[N], c[N];
+
+__attribute__((noipa)) void
+mul (double *__restrict cc, double *__restrict aa, double *__restrict bb, int n)
+{
+  for (int i = 0; i < n; i += 4)
+    {
+      cc[i] = aa[i] * bb[i] - aa[i + 1] * bb[i + 1];
+      cc[i + 1] = aa[i + 2] * bb[i + 1] + aa[i + 1] * bb[i];
+      cc[i + 2] = aa[i + 2] * bb[i + 2] - aa[i + 3] * bb[i + 3];
+      cc[i + 3] = aa[i + 2] * bb[i + 3] + aa[i + 3] * bb[i + 2];
+    }
+}
+
+__attribute__((optimize (0))) void
+ref (double *__restrict cc, double *__restrict aa, double *__restrict bb, int n)
+{
+  for (int i = 0; i < n; i += 4)
+    {
+      cc[i] = aa[i] * bb[i] - aa[i + 1] * bb[i + 1];
+      cc[i + 1] = aa[i + 2] * bb[i + 1] + aa[i + 1] * bb[i];
+      cc[i + 2] = aa[i + 2] * bb[i + 2] - aa[i + 3] * bb[i + 3];
+      cc[i + 3] = aa[i + 2] * bb[i + 3] + aa[i + 3] * bb[i + 2];
+    }
+}
+
+int
+main (void)
+{
+  double e[N];
+
+  for (int i = 0; i < N; ++i)
+    {
+      a[i] = i + 1;
+      b[i] = i * 3 + 1;
+    }
+
+  mul (c, a, b, N);
+  ref (e, a, b, N);
+
+  for (int i = 0; i < N; ++i)
+    if (c[i] != e[i])
+      __builtin_abort ();
+
+  return 0;
+}
+
+/* { dg-final { scan-assembler-not {fcmla\t} } } */
diff --git a/gcc/tree-vect-slp-patterns.cc b/gcc/tree-vect-slp-patterns.cc
index 2136329645cad8f26f672ac5ad7d7f225bb08b7f..a40e3d29fdcb54ccf79da76f273faa8443f7654c 100644
--- a/gcc/tree-vect-slp-patterns.cc
+++ b/gcc/tree-vect-slp-patterns.cc
@@ -138,47 +138,34 @@ is_linear_load_p (load_permutation_t loads)
   if (loads.length() == 0)
     return PERM_UNKNOWN;
 
-  unsigned load, i;
-  complex_perm_kinds_t candidates[4]
-    = { PERM_ODDODD
-      , PERM_EVENEVEN
-      , PERM_EVENODD
-      , PERM_ODDEVEN
-      };
+  if (loads.length () == 1)
+    return loads[0] == 0 ? PERM_EVENEVEN : PERM_ODDODD;
 
-  int valid_patterns = 4;
-  FOR_EACH_VEC_ELT (loads, i, load)
+  vec_perm_builder builder;
+  builder.new_vector (loads.length (), loads.length (), 1);
+  for (unsigned load : loads)
     {
-      unsigned adj_load = load % 2;
-      if (candidates[0] != PERM_UNKNOWN && adj_load != 1)
-	{
-	  candidates[0] = PERM_UNKNOWN;
-	  valid_patterns--;
-	}
-      if (candidates[1] != PERM_UNKNOWN && adj_load != 0)
-	{
-	  candidates[1] = PERM_UNKNOWN;
-	  valid_patterns--;
-	}
-      if (candidates[2] != PERM_UNKNOWN && load != i)
-	{
-	  candidates[2] = PERM_UNKNOWN;
-	  valid_patterns--;
-	}
-      if (candidates[3] != PERM_UNKNOWN
-	  && load != (i % 2 == 0 ? i + 1 : i - 1))
-	{
-	  candidates[3] = PERM_UNKNOWN;
-	  valid_patterns--;
-	}
-
-      if (valid_patterns == 0)
+      if (load >= loads.length ())
 	return PERM_UNKNOWN;
+      builder.quick_push (load);
     }
 
-  for (i = 0; i < sizeof(candidates); i++)
-    if (candidates[i] != PERM_UNKNOWN)
-      return candidates[i];
+  vec_perm_indices indices (builder, 1, loads.length ());
+
+  if (indices.series_p (0, 2, 1, 2)
+      && indices.series_p (1, 2, 1, 2))
+    return PERM_ODDODD;
+
+  if (indices.series_p (0, 2, 0, 2)
+      && indices.series_p (1, 2, 0, 2))
+    return PERM_EVENEVEN;
+
+  if (indices.series_p (0, 1, 0, 1))
+    return PERM_EVENODD;
+
+  if (indices.series_p (0, 2, 1, 2)
+      && indices.series_p (1, 2, 0, 2))
+    return PERM_ODDEVEN;
 
   return PERM_UNKNOWN;
 }


--qcw4+AymWSgpFk/J--