Enable STV if no stack realignment is neededhjl/pr69692/master

Add *movdi_to_v2di for STV to load a DI into an xmm register.

3 files changed, 63 insertions, 20 deletions

diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index 1215ae45a4b..5d29feb1b03 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -3588,14 +3588,21 @@ convert_scalars_to_vector ()
   bitmap_obstack_release (NULL);
   df_process_deferred_rescans ();
 
-  /* Conversion means we may have 128bit register spills/fills
-     which require aligned stack.  */
-  if (converted_insns)
-    {
-      if (crtl->stack_alignment_needed < 128)
-	crtl->stack_alignment_needed = 128;
-      if (crtl->stack_alignment_estimated < 128)
-	crtl->stack_alignment_estimated = 128;
+  /* Conversion means we may have 64-bit register spills/fills which
+     require aligned stack.  No stack realignment is needed to get
+     64-bit stack alignment since the incoming stack boundary >= 64
+     bits. */
+  if (converted_insns && crtl->stack_alignment_estimated < 64)
+    {
+      unsigned int incoming_stack_boundary
+	= (crtl->parm_stack_boundary > ix86_incoming_stack_boundary
+	   ? crtl->parm_stack_boundary : ix86_incoming_stack_boundary);
+      if (incoming_stack_boundary < 64)
+	gcc_unreachable ();
+
+      /* Increase stack_alignment_estimated up to incoming_stack_boundary
+	 won't affect stack realignment.  */
+      crtl->stack_alignment_estimated = 64;
     }
 
   return 0;
@@ -3661,7 +3668,21 @@ public:
   /* opt_pass methods: */
   virtual bool gate (function *)
     {
-      return !TARGET_64BIT && TARGET_STV && TARGET_SSE2 && optimize > 1;
+      if (!TARGET_64BIT && TARGET_STV && TARGET_SSE2 && optimize > 1)
+	{
+	  /* STV may need 64-bit register spills/fills which require
+	     aligned stack. */
+	  if (crtl->stack_alignment_estimated >= 64)
+	    return true;
+
+	  /* No stack realignment is needed to get 64-bit stack
+	     alignment if incoming stack boundary >= 64 bits. */
+	  unsigned int incoming_stack_boundary
+	    = (crtl->parm_stack_boundary > ix86_incoming_stack_boundary
+	       ? crtl->parm_stack_boundary : ix86_incoming_stack_boundary);
+	  return incoming_stack_boundary >= 64;
+	}
+      return false;
     }
 
   virtual unsigned int execute (function *)
@@ -5453,13 +5474,6 @@ ix86_option_override_internal (bool main_args_p,
     opts->x_target_flags |= MASK_VZEROUPPER;
   if (!(opts_set->x_target_flags & MASK_STV))
     opts->x_target_flags |= MASK_STV;
-  /* Disable STV if -mpreferred-stack-boundary={2,3} or
-     -mincoming-stack-boundary={2,3} - the needed
-     stack realignment will be extra cost the pass doesn't take into
-     account and the pass can't realign the stack.  */
-  if (ix86_preferred_stack_boundary < 128
-      || ix86_incoming_stack_boundary < 128)
-    opts->x_target_flags &= ~MASK_STV;
   if (!ix86_tune_features[X86_TUNE_AVX256_UNALIGNED_LOAD_OPTIMAL]
       && !(opts_set->x_target_flags & MASK_AVX256_SPLIT_UNALIGNED_LOAD))
     opts->x_target_flags |= MASK_AVX256_SPLIT_UNALIGNED_LOAD;
@@ -29330,10 +29344,7 @@ ix86_minimum_alignment (tree exp, machine_mode mode,
   if ((mode == DImode || (type && TYPE_MODE (type) == DImode))
       && (!type || !TYPE_USER_ALIGN (type))
       && (!decl || !DECL_USER_ALIGN (decl)))
-    {
-      gcc_checking_assert (!TARGET_STV);
-      return 32;
-    }
+    return 32;
 
   return align;
 }
diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index 045a85f0ac3..201670a7c4b 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -1075,6 +1075,16 @@
    (set_attr "prefix" "maybe_vex")
    (set_attr "mode" "TI")])
 
+;; Used by STV to load a DI into an xmm register.
+(define_insn "*movdi_to_v2di"
+  [(set (match_operand:V2DI 0 "register_operand" "=x")
+        (subreg:V2DI (match_operand:DI 1 "nonimmediate_operand" "xm") 0))]
+  "!TARGET_64BIT && TARGET_SSE2"
+  "%vmovq\t{%1, %0|%0, %1}"
+  [(set_attr "type" "ssemov")
+   (set_attr "prefix" "maybe_vex")
+   (set_attr "mode" "DI")])
+
 ;; Move a DI from a 32-bit register pair (e.g. %edx:%eax) to an xmm.
 ;; We'd rather avoid this entirely; if the 32-bit reg pair was loaded
 ;; from memory, we'd prefer to load the memory directly into the %xmm
diff --git a/gcc/testsuite/gcc.target/i386/pr69692.c b/gcc/testsuite/gcc.target/i386/pr69692.c
new file mode 100644
index 00000000000..63b6f352e56
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr69692.c
@@ -0,0 +1,22 @@
+/* { dg-do compile { target { ia32 } } } */
+/* { dg-options "-O2 -msse2 -mno-accumulate-outgoing-args -mpreferred-stack-boundary=2" } */
+
+struct bar
+{
+  int i[16] __attribute__ ((aligned(16)));
+};
+
+extern void fn2 (void);
+long long a, b;
+
+struct bar
+fn1 (struct bar x)
+{
+  long long c = a;
+  a = b ^ a;
+  fn2 ();
+  a = c;
+  return x;
+}
+
+/* { dg-final { scan-assembler "pxor\[^\n\]*xmm" } } */