From 0dc77a0c4942d3b264f8f8cfc2c509ecc02c3634 Mon Sep 17 00:00:00 2001
From: Richard Biener <rguenther@suse.de>
Date: Mon, 6 Dec 2021 15:13:49 +0100
Subject: [PATCH] tree-optimization/103581 - fix masked gather on x86

The recent fix to PR103527 exposed an issue with how the various
special casing for AVX512 masks in vect_build_gather_load_calls
are handled.  The following makes that more obvious, fixing the
miscompile of 403.gcc.

2021-12-06  Richard Biener  <rguenther@suse.de>

	PR tree-optimization/103581
	* tree-vect-stmts.c (vect_build_gather_load_calls): Properly
	guard all the AVX512 mask cases.

	* gcc.dg/vect/pr103581.c: New testcase.
---
 gcc/testsuite/gcc.dg/vect/pr103581.c | 59 ++++++++++++++++++++++++++++
 gcc/tree-vect-stmts.c                |  4 +-
 2 files changed, 61 insertions(+), 2 deletions(-)
 create mode 100644 gcc/testsuite/gcc.dg/vect/pr103581.c

diff --git a/gcc/testsuite/gcc.dg/vect/pr103581.c b/gcc/testsuite/gcc.dg/vect/pr103581.c
new file mode 100644
index 00000000000..d072748de31
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/vect/pr103581.c
@@ -0,0 +1,59 @@
+/* { dg-additional-options "-mavx2 -mtune-ctrl=use_gather" { target avx2_runtime } } */
+
+#include "tree-vect.h"
+
+#define MASKGATHER(SUFF, TYPE1, TYPE2) \
+TYPE1 * __attribute__((noipa)) \
+maskgather ## SUFF (int n, TYPE2 *indices, TYPE1 *data) \
+{ \
+  TYPE1 *out = __builtin_malloc (sizeof (TYPE1) * n); \
+  for (int i = 0; i < n; ++i) \
+    { \
+      TYPE2 d = indices[i]; \
+      if (d > 1) \
+        out[i] = data[d]; \
+    } \
+  return out; \
+}
+
+MASKGATHER(udiusi, unsigned long long, unsigned int)
+MASKGATHER(usiusi, unsigned int, unsigned int)
+MASKGATHER(udiudi, unsigned long long, unsigned long long)
+MASKGATHER(usiudi, unsigned int, unsigned long long)
+
+int
+main()
+{
+  check_vect ();
+
+    unsigned int idx4[32], data4[32];
+  unsigned long long idx8[32], data8[32];
+  for (int i = 0; i < 32; ++i)
+    {
+      idx4[i] = i;
+      idx8[i] = i;
+      data4[i] = i;
+      data8[i] = i;
+    }
+  unsigned long long *resudiusi = maskgatherudiusi (16, idx4, data8);
+  unsigned int *resusiusi = maskgatherusiusi (16, idx4, data4);
+  unsigned long long *resudiudi = maskgatherudiudi (16, idx8, data8);
+  unsigned int *resusiudi = maskgatherusiudi (16, idx8, data4);
+  for (int i = 0; i < 16; ++i)
+    {
+      unsigned int d = idx4[i];
+      if (d > 1)
+        {
+          if (resudiusi[i] != data4[d])
+            __builtin_abort ();
+          if (resudiudi[i] != data4[d])
+            __builtin_abort ();
+          if (resusiudi[i] != data4[d])
+            __builtin_abort ();
+          if (resusiusi[i] != data4[d])
+            __builtin_abort ();
+        }
+    }
+  return 0;
+}
+
diff --git a/gcc/tree-vect-stmts.c b/gcc/tree-vect-stmts.c
index 84c6d9777db..8c427174b37 100644
--- a/gcc/tree-vect-stmts.c
+++ b/gcc/tree-vect-stmts.c
@@ -2785,7 +2785,7 @@ vect_build_gather_load_calls (vec_info *vinfo, stmt_vec_info stmt_info,
 
       ncopies *= 2;
 
-      if (mask && masktype == real_masktype)
+      if (mask && VECTOR_TYPE_P (real_masktype))
 	{
 	  for (int i = 0; i < count; ++i)
 	    sel[i] = i | (count / 2);
@@ -2882,7 +2882,7 @@ vect_build_gather_load_calls (vec_info *vinfo, stmt_vec_info stmt_info,
 		  mask_op = var;
 		}
 	    }
-	  if (modifier == NARROW && masktype != real_masktype)
+	  if (modifier == NARROW && !VECTOR_TYPE_P (real_masktype))
 	    {
 	      var = vect_get_new_ssa_name (mask_halftype, vect_simple_var);
 	      gassign *new_stmt