re PR target/85323 (SSE/AVX/AVX512 shift by 0 not optimized away)

	PR target/85323
	* config/i386/i386.c: Include tree-vector-builder.h.
	(ix86_vector_shift_count): New function.
	(ix86_fold_builtin): Fold shift builtins by scalar count.
	(ix86_gimple_fold_builtin): Likewise.

	* gcc.target/i386/pr85323-1.c: New test.
	* gcc.target/i386/pr85323-2.c: New test.
	* gcc.target/i386/pr85323-3.c: New test.

From-SVN: r260311

1 files changed, 340 insertions, 0 deletions

diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index 0c7a6b7..f802add 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -91,6 +91,7 @@ along with GCC; see the file COPYING3.  If not see
 #include "ipa-prop.h"
 #include "ipa-fnsummary.h"
 #include "wide-int-bitmask.h"
+#include "tree-vector-builder.h"
 
 /* This file should be included last.  */
 #include "target-def.h"
@@ -33365,6 +33366,28 @@ fold_builtin_cpu (tree fndecl, tree *args)
   gcc_unreachable ();
 }
 
+/* Return the shift count of a vector by scalar shift builtin second argument
+   ARG1.  */
+static tree
+ix86_vector_shift_count (tree arg1)
+{
+  if (tree_fits_uhwi_p (arg1))
+    return arg1;
+  else if (TREE_CODE (arg1) == VECTOR_CST && CHAR_BIT == 8)
+    {
+      /* The count argument is weird, passed in as various 128-bit
+	 (or 64-bit) vectors, the low 64 bits from it are the count.  */
+      unsigned char buf[16];
+      int len = native_encode_expr (arg1, buf, 16);
+      if (len == 0)
+	return NULL_TREE;
+      tree t = native_interpret_expr (uint64_type_node, buf, len);
+      if (t && tree_fits_uhwi_p (t))
+	return t;
+    }
+  return NULL_TREE;
+}
+
 static tree
 ix86_fold_builtin (tree fndecl, int n_args,
 		   tree *args, bool ignore ATTRIBUTE_UNUSED)
@@ -33373,6 +33396,8 @@ ix86_fold_builtin (tree fndecl, int n_args,
     {
       enum ix86_builtins fn_code = (enum ix86_builtins)
 				   DECL_FUNCTION_CODE (fndecl);
+      enum rtx_code rcode;
+
       switch (fn_code)
 	{
 	case IX86_BUILTIN_CPU_IS:
@@ -33553,6 +33578,168 @@ ix86_fold_builtin (tree fndecl, int n_args,
 	    }
 	  break;
 
+	case IX86_BUILTIN_PSLLD:
+	case IX86_BUILTIN_PSLLD128:
+	case IX86_BUILTIN_PSLLD128_MASK:
+	case IX86_BUILTIN_PSLLD256:
+	case IX86_BUILTIN_PSLLD256_MASK:
+	case IX86_BUILTIN_PSLLD512:
+	case IX86_BUILTIN_PSLLDI:
+	case IX86_BUILTIN_PSLLDI128:
+	case IX86_BUILTIN_PSLLDI128_MASK:
+	case IX86_BUILTIN_PSLLDI256:
+	case IX86_BUILTIN_PSLLDI256_MASK:
+	case IX86_BUILTIN_PSLLDI512:
+	case IX86_BUILTIN_PSLLQ:
+	case IX86_BUILTIN_PSLLQ128:
+	case IX86_BUILTIN_PSLLQ128_MASK:
+	case IX86_BUILTIN_PSLLQ256:
+	case IX86_BUILTIN_PSLLQ256_MASK:
+	case IX86_BUILTIN_PSLLQ512:
+	case IX86_BUILTIN_PSLLQI:
+	case IX86_BUILTIN_PSLLQI128:
+	case IX86_BUILTIN_PSLLQI128_MASK:
+	case IX86_BUILTIN_PSLLQI256:
+	case IX86_BUILTIN_PSLLQI256_MASK:
+	case IX86_BUILTIN_PSLLQI512:
+	case IX86_BUILTIN_PSLLW:
+	case IX86_BUILTIN_PSLLW128:
+	case IX86_BUILTIN_PSLLW128_MASK:
+	case IX86_BUILTIN_PSLLW256:
+	case IX86_BUILTIN_PSLLW256_MASK:
+	case IX86_BUILTIN_PSLLW512_MASK:
+	case IX86_BUILTIN_PSLLWI:
+	case IX86_BUILTIN_PSLLWI128:
+	case IX86_BUILTIN_PSLLWI128_MASK:
+	case IX86_BUILTIN_PSLLWI256:
+	case IX86_BUILTIN_PSLLWI256_MASK:
+	case IX86_BUILTIN_PSLLWI512_MASK:
+	  rcode = ASHIFT;
+	  goto do_shift;
+	case IX86_BUILTIN_PSRAD:
+	case IX86_BUILTIN_PSRAD128:
+	case IX86_BUILTIN_PSRAD128_MASK:
+	case IX86_BUILTIN_PSRAD256:
+	case IX86_BUILTIN_PSRAD256_MASK:
+	case IX86_BUILTIN_PSRAD512:
+	case IX86_BUILTIN_PSRADI:
+	case IX86_BUILTIN_PSRADI128:
+	case IX86_BUILTIN_PSRADI128_MASK:
+	case IX86_BUILTIN_PSRADI256:
+	case IX86_BUILTIN_PSRADI256_MASK:
+	case IX86_BUILTIN_PSRADI512:
+	case IX86_BUILTIN_PSRAQ128_MASK:
+	case IX86_BUILTIN_PSRAQ256_MASK:
+	case IX86_BUILTIN_PSRAQ512:
+	case IX86_BUILTIN_PSRAQI128_MASK:
+	case IX86_BUILTIN_PSRAQI256_MASK:
+	case IX86_BUILTIN_PSRAQI512:
+	case IX86_BUILTIN_PSRAW:
+	case IX86_BUILTIN_PSRAW128:
+	case IX86_BUILTIN_PSRAW128_MASK:
+	case IX86_BUILTIN_PSRAW256:
+	case IX86_BUILTIN_PSRAW256_MASK:
+	case IX86_BUILTIN_PSRAW512:
+	case IX86_BUILTIN_PSRAWI:
+	case IX86_BUILTIN_PSRAWI128:
+	case IX86_BUILTIN_PSRAWI128_MASK:
+	case IX86_BUILTIN_PSRAWI256:
+	case IX86_BUILTIN_PSRAWI256_MASK:
+	case IX86_BUILTIN_PSRAWI512:
+	  rcode = ASHIFTRT;
+	  goto do_shift;
+	case IX86_BUILTIN_PSRLD:
+	case IX86_BUILTIN_PSRLD128:
+	case IX86_BUILTIN_PSRLD128_MASK:
+	case IX86_BUILTIN_PSRLD256:
+	case IX86_BUILTIN_PSRLD256_MASK:
+	case IX86_BUILTIN_PSRLD512:
+	case IX86_BUILTIN_PSRLDI:
+	case IX86_BUILTIN_PSRLDI128:
+	case IX86_BUILTIN_PSRLDI128_MASK:
+	case IX86_BUILTIN_PSRLDI256:
+	case IX86_BUILTIN_PSRLDI256_MASK:
+	case IX86_BUILTIN_PSRLDI512:
+	case IX86_BUILTIN_PSRLQ:
+	case IX86_BUILTIN_PSRLQ128:
+	case IX86_BUILTIN_PSRLQ128_MASK:
+	case IX86_BUILTIN_PSRLQ256:
+	case IX86_BUILTIN_PSRLQ256_MASK:
+	case IX86_BUILTIN_PSRLQ512:
+	case IX86_BUILTIN_PSRLQI:
+	case IX86_BUILTIN_PSRLQI128:
+	case IX86_BUILTIN_PSRLQI128_MASK:
+	case IX86_BUILTIN_PSRLQI256:
+	case IX86_BUILTIN_PSRLQI256_MASK:
+	case IX86_BUILTIN_PSRLQI512:
+	case IX86_BUILTIN_PSRLW:
+	case IX86_BUILTIN_PSRLW128:
+	case IX86_BUILTIN_PSRLW128_MASK:
+	case IX86_BUILTIN_PSRLW256:
+	case IX86_BUILTIN_PSRLW256_MASK:
+	case IX86_BUILTIN_PSRLW512:
+	case IX86_BUILTIN_PSRLWI:
+	case IX86_BUILTIN_PSRLWI128:
+	case IX86_BUILTIN_PSRLWI128_MASK:
+	case IX86_BUILTIN_PSRLWI256:
+	case IX86_BUILTIN_PSRLWI256_MASK:
+	case IX86_BUILTIN_PSRLWI512:
+	  rcode = LSHIFTRT;
+	  goto do_shift;
+
+	do_shift:
+	  gcc_assert (n_args >= 2);
+	  if (TREE_CODE (args[0]) != VECTOR_CST)
+	    break;
+	  if (n_args > 2)
+	    {
+	      /* This is masked shift.  Only optimize if the mask is all
+		 ones.  */
+	      if (!tree_fits_uhwi_p (args[n_args - 1])
+		  || TREE_SIDE_EFFECTS (args[n_args - 2]))
+		break;
+	      unsigned HOST_WIDE_INT mask = tree_to_uhwi (args[n_args - 1]);
+	      unsigned elems = TYPE_VECTOR_SUBPARTS (TREE_TYPE (args[0]));
+	      if ((mask | (HOST_WIDE_INT_M1U << elems)) != HOST_WIDE_INT_M1U)
+		break;
+	    }
+	  if (tree tem = ix86_vector_shift_count (args[1]))
+	    {
+	      unsigned HOST_WIDE_INT count = tree_to_uhwi (tem);
+	      if (count == 0)
+		return args[0];
+	      if (count >= TYPE_PRECISION (TREE_TYPE (TREE_TYPE (args[0]))))
+		{
+		  if (rcode != ASHIFTRT)
+		    return build_zero_cst (TREE_TYPE (args[0]));
+		  count = TYPE_PRECISION (TREE_TYPE (TREE_TYPE (args[0]))) - 1;
+		}
+	      tree countt = build_int_cst (integer_type_node, count);
+	      tree_vector_builder builder;
+	      builder.new_unary_operation (TREE_TYPE (args[0]), args[0],
+					   false);
+	      unsigned int cnt = builder.encoded_nelts ();
+	      for (unsigned int i = 0; i < cnt; ++i)
+		{
+		  tree elt = VECTOR_CST_ELT (args[0], i);
+		  if (TREE_CODE (elt) != INTEGER_CST || TREE_OVERFLOW (elt))
+		    return NULL_TREE;
+		  tree type = TREE_TYPE (elt);
+		  if (rcode == LSHIFTRT)
+		    elt = fold_convert (unsigned_type_for (type), elt);
+		  elt = const_binop (rcode == ASHIFT
+				     ? LSHIFT_EXPR : RSHIFT_EXPR, TREE_TYPE (elt),
+				     elt, countt);
+		  if (!elt || TREE_CODE (elt) != INTEGER_CST)
+		    return NULL_TREE;
+		  if (rcode == LSHIFTRT)
+		    elt = fold_convert (type, elt);
+		  builder.quick_push (elt);
+		}
+	      return builder.build ();
+	    }
+	  break;
+
 	default:
 	  break;
 	}
@@ -33578,6 +33765,8 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
   enum ix86_builtins fn_code = (enum ix86_builtins) DECL_FUNCTION_CODE (fndecl);
   tree decl = NULL_TREE;
   tree arg0, arg1;
+  enum rtx_code rcode;
+  unsigned HOST_WIDE_INT count;
 
   switch (fn_code)
     {
@@ -33657,6 +33846,157 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
 	}
       break;
 
+    case IX86_BUILTIN_PSLLD:
+    case IX86_BUILTIN_PSLLD128:
+    case IX86_BUILTIN_PSLLD128_MASK:
+    case IX86_BUILTIN_PSLLD256:
+    case IX86_BUILTIN_PSLLD256_MASK:
+    case IX86_BUILTIN_PSLLD512:
+    case IX86_BUILTIN_PSLLDI:
+    case IX86_BUILTIN_PSLLDI128:
+    case IX86_BUILTIN_PSLLDI128_MASK:
+    case IX86_BUILTIN_PSLLDI256:
+    case IX86_BUILTIN_PSLLDI256_MASK:
+    case IX86_BUILTIN_PSLLDI512:
+    case IX86_BUILTIN_PSLLQ:
+    case IX86_BUILTIN_PSLLQ128:
+    case IX86_BUILTIN_PSLLQ128_MASK:
+    case IX86_BUILTIN_PSLLQ256:
+    case IX86_BUILTIN_PSLLQ256_MASK:
+    case IX86_BUILTIN_PSLLQ512:
+    case IX86_BUILTIN_PSLLQI:
+    case IX86_BUILTIN_PSLLQI128:
+    case IX86_BUILTIN_PSLLQI128_MASK:
+    case IX86_BUILTIN_PSLLQI256:
+    case IX86_BUILTIN_PSLLQI256_MASK:
+    case IX86_BUILTIN_PSLLQI512:
+    case IX86_BUILTIN_PSLLW:
+    case IX86_BUILTIN_PSLLW128:
+    case IX86_BUILTIN_PSLLW128_MASK:
+    case IX86_BUILTIN_PSLLW256:
+    case IX86_BUILTIN_PSLLW256_MASK:
+    case IX86_BUILTIN_PSLLW512_MASK:
+    case IX86_BUILTIN_PSLLWI:
+    case IX86_BUILTIN_PSLLWI128:
+    case IX86_BUILTIN_PSLLWI128_MASK:
+    case IX86_BUILTIN_PSLLWI256:
+    case IX86_BUILTIN_PSLLWI256_MASK:
+    case IX86_BUILTIN_PSLLWI512_MASK:
+      rcode = ASHIFT;
+      goto do_shift;
+    case IX86_BUILTIN_PSRAD:
+    case IX86_BUILTIN_PSRAD128:
+    case IX86_BUILTIN_PSRAD128_MASK:
+    case IX86_BUILTIN_PSRAD256:
+    case IX86_BUILTIN_PSRAD256_MASK:
+    case IX86_BUILTIN_PSRAD512:
+    case IX86_BUILTIN_PSRADI:
+    case IX86_BUILTIN_PSRADI128:
+    case IX86_BUILTIN_PSRADI128_MASK:
+    case IX86_BUILTIN_PSRADI256:
+    case IX86_BUILTIN_PSRADI256_MASK:
+    case IX86_BUILTIN_PSRADI512:
+    case IX86_BUILTIN_PSRAQ128_MASK:
+    case IX86_BUILTIN_PSRAQ256_MASK:
+    case IX86_BUILTIN_PSRAQ512:
+    case IX86_BUILTIN_PSRAQI128_MASK:
+    case IX86_BUILTIN_PSRAQI256_MASK:
+    case IX86_BUILTIN_PSRAQI512:
+    case IX86_BUILTIN_PSRAW:
+    case IX86_BUILTIN_PSRAW128:
+    case IX86_BUILTIN_PSRAW128_MASK:
+    case IX86_BUILTIN_PSRAW256:
+    case IX86_BUILTIN_PSRAW256_MASK:
+    case IX86_BUILTIN_PSRAW512:
+    case IX86_BUILTIN_PSRAWI:
+    case IX86_BUILTIN_PSRAWI128:
+    case IX86_BUILTIN_PSRAWI128_MASK:
+    case IX86_BUILTIN_PSRAWI256:
+    case IX86_BUILTIN_PSRAWI256_MASK:
+    case IX86_BUILTIN_PSRAWI512:
+      rcode = ASHIFTRT;
+      goto do_shift;
+    case IX86_BUILTIN_PSRLD:
+    case IX86_BUILTIN_PSRLD128:
+    case IX86_BUILTIN_PSRLD128_MASK:
+    case IX86_BUILTIN_PSRLD256:
+    case IX86_BUILTIN_PSRLD256_MASK:
+    case IX86_BUILTIN_PSRLD512:
+    case IX86_BUILTIN_PSRLDI:
+    case IX86_BUILTIN_PSRLDI128:
+    case IX86_BUILTIN_PSRLDI128_MASK:
+    case IX86_BUILTIN_PSRLDI256:
+    case IX86_BUILTIN_PSRLDI256_MASK:
+    case IX86_BUILTIN_PSRLDI512:
+    case IX86_BUILTIN_PSRLQ:
+    case IX86_BUILTIN_PSRLQ128:
+    case IX86_BUILTIN_PSRLQ128_MASK:
+    case IX86_BUILTIN_PSRLQ256:
+    case IX86_BUILTIN_PSRLQ256_MASK:
+    case IX86_BUILTIN_PSRLQ512:
+    case IX86_BUILTIN_PSRLQI:
+    case IX86_BUILTIN_PSRLQI128:
+    case IX86_BUILTIN_PSRLQI128_MASK:
+    case IX86_BUILTIN_PSRLQI256:
+    case IX86_BUILTIN_PSRLQI256_MASK:
+    case IX86_BUILTIN_PSRLQI512:
+    case IX86_BUILTIN_PSRLW:
+    case IX86_BUILTIN_PSRLW128:
+    case IX86_BUILTIN_PSRLW128_MASK:
+    case IX86_BUILTIN_PSRLW256:
+    case IX86_BUILTIN_PSRLW256_MASK:
+    case IX86_BUILTIN_PSRLW512:
+    case IX86_BUILTIN_PSRLWI:
+    case IX86_BUILTIN_PSRLWI128:
+    case IX86_BUILTIN_PSRLWI128_MASK:
+    case IX86_BUILTIN_PSRLWI256:
+    case IX86_BUILTIN_PSRLWI256_MASK:
+    case IX86_BUILTIN_PSRLWI512:
+      rcode = LSHIFTRT;
+      goto do_shift;
+
+    do_shift:
+      gcc_assert (n_args >= 2);
+      arg0 = gimple_call_arg (stmt, 0);
+      arg1 = gimple_call_arg (stmt, 1);
+      if (n_args > 2)
+	{
+	  /* This is masked shift.  Only optimize if the mask is all ones.  */
+	  tree argl = gimple_call_arg (stmt, n_args - 1);
+	  if (!tree_fits_uhwi_p (argl))
+	    break;
+	  unsigned HOST_WIDE_INT mask = tree_to_uhwi (argl);
+	  unsigned elems = TYPE_VECTOR_SUBPARTS (TREE_TYPE (arg0));
+	  if ((mask | (HOST_WIDE_INT_M1U << elems)) != HOST_WIDE_INT_M1U)
+	    break;
+	}
+      arg1 = ix86_vector_shift_count (arg1);
+      if (!arg1)
+	break;
+      count = tree_to_uhwi (arg1);
+      if (count == 0)
+	{
+	  /* Just return the first argument for shift by 0.  */
+	  location_t loc = gimple_location (stmt);
+	  gimple *g = gimple_build_assign (gimple_call_lhs (stmt), arg0);
+	  gimple_set_location (g, loc);
+	  gsi_replace (gsi, g, false);
+	  return true;
+	}
+      if (rcode != ASHIFTRT
+	  && count >= TYPE_PRECISION (TREE_TYPE (TREE_TYPE (arg0))))
+	{
+	  /* For shift counts equal or greater than precision, except for
+	     arithmetic right shift the result is zero.  */
+	  location_t loc = gimple_location (stmt);
+	  gimple *g = gimple_build_assign (gimple_call_lhs (stmt),
+					   build_zero_cst (TREE_TYPE (arg0)));
+	  gimple_set_location (g, loc);
+	  gsi_replace (gsi, g, false);
+	  return true;
+	}
+      break;
+
     default:
       break;
     }