This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Fix up vectorizable_condition for comparisons of 2 booleans (PR tree-optimization/78938)


On Thu, 5 Jan 2017, Jakub Jelinek wrote:

> Hi!
> 
> As mentioned in the PR, while vectorizable_comparison has code to deal
> with comparison of 2 booleans by transorming those into one or two
> BIT_*_EXPR operations that work equally well on normal vectors as well
> as the AVX512 bitset masks, vectorizable_comparison lacks that and we
> ICE during expansion because of that.
> The following patch teaches vectorizable_condition to do that too.
> 
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

Ok.

Richard.

> 2017-01-05  Jakub Jelinek  <jakub@redhat.com>
> 
> 	PR tree-optimization/78938
> 	* tree-vect-stmts.c (vectorizable_condition): For non-masked COND_EXPR
> 	where comp_vectype is VECTOR_BOOLEAN_TYPE_P, use
> 	BIT_{NOT,XOR,AND,IOR}_EXPR on the comparison operands instead of
> 	{EQ,NE,GE,GT,LE,LT}_EXPR directly inside of VEC_COND_EXPR.  Formatting
> 	fixes.
> 
> 	* gcc.dg/vect/pr78938.c: New test.
> 
> --- gcc/tree-vect-stmts.c.jj	2017-01-01 12:45:39.000000000 +0100
> +++ gcc/tree-vect-stmts.c	2017-01-05 15:54:41.075218409 +0100
> @@ -7731,7 +7731,8 @@ vectorizable_condition (gimple *stmt, gi
>  {
>    tree scalar_dest = NULL_TREE;
>    tree vec_dest = NULL_TREE;
> -  tree cond_expr, then_clause, else_clause;
> +  tree cond_expr, cond_expr0 = NULL_TREE, cond_expr1 = NULL_TREE;
> +  tree then_clause, else_clause;
>    stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
>    tree comp_vectype = NULL_TREE;
>    tree vec_cond_lhs = NULL_TREE, vec_cond_rhs = NULL_TREE;
> @@ -7741,7 +7742,7 @@ vectorizable_condition (gimple *stmt, gi
>    loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
>    enum vect_def_type dt, dts[4];
>    int ncopies;
> -  enum tree_code code;
> +  enum tree_code code, cond_code, bitop1 = NOP_EXPR, bitop2 = NOP_EXPR;
>    stmt_vec_info prev_stmt_info = NULL;
>    int i, j;
>    bb_vec_info bb_vinfo = STMT_VINFO_BB_VINFO (stmt_info);
> @@ -7825,11 +7826,76 @@ vectorizable_condition (gimple *stmt, gi
>    if (vec_cmp_type == NULL_TREE)
>      return false;
>  
> +  cond_code = TREE_CODE (cond_expr);
> +  if (!masked)
> +    {
> +      cond_expr0 = TREE_OPERAND (cond_expr, 0);
> +      cond_expr1 = TREE_OPERAND (cond_expr, 1);
> +    }
> +
> +  if (!masked && VECTOR_BOOLEAN_TYPE_P (comp_vectype))
> +    {
> +      /* Boolean values may have another representation in vectors
> +	 and therefore we prefer bit operations over comparison for
> +	 them (which also works for scalar masks).  We store opcodes
> +	 to use in bitop1 and bitop2.  Statement is vectorized as
> +	 BITOP2 (rhs1 BITOP1 rhs2) or rhs1 BITOP2 (BITOP1 rhs2)
> +	 depending on bitop1 and bitop2 arity.  */
> +      switch (cond_code)
> +	{
> +	case GT_EXPR:
> +	  bitop1 = BIT_NOT_EXPR;
> +	  bitop2 = BIT_AND_EXPR;
> +	  break;
> +	case GE_EXPR:
> +	  bitop1 = BIT_NOT_EXPR;
> +	  bitop2 = BIT_IOR_EXPR;
> +	  break;
> +	case LT_EXPR:
> +	  bitop1 = BIT_NOT_EXPR;
> +	  bitop2 = BIT_AND_EXPR;
> +	  std::swap (cond_expr0, cond_expr1);
> +	  break;
> +	case LE_EXPR:
> +	  bitop1 = BIT_NOT_EXPR;
> +	  bitop2 = BIT_IOR_EXPR;
> +	  std::swap (cond_expr0, cond_expr1);
> +	  break;
> +	case NE_EXPR:
> +	  bitop1 = BIT_XOR_EXPR;
> +	  break;
> +	case EQ_EXPR:
> +	  bitop1 = BIT_XOR_EXPR;
> +	  bitop2 = BIT_NOT_EXPR;
> +	  break;
> +	default:
> +	  return false;
> +	}
> +      cond_code = SSA_NAME;
> +    }
> +
>    if (!vec_stmt)
>      {
>        STMT_VINFO_TYPE (stmt_info) = condition_vec_info_type;
> +      if (bitop1 != NOP_EXPR)
> +	{
> +	  machine_mode mode = TYPE_MODE (comp_vectype);
> +	  optab optab;
> +
> +	  optab = optab_for_tree_code (bitop1, comp_vectype, optab_default);
> +	  if (!optab || optab_handler (optab, mode) == CODE_FOR_nothing)
> +	    return false;
> +
> +	  if (bitop2 != NOP_EXPR)
> +	    {
> +	      optab = optab_for_tree_code (bitop2, comp_vectype,
> +					   optab_default);
> +	      if (!optab || optab_handler (optab, mode) == CODE_FOR_nothing)
> +		return false;
> +	    }
> +	}
>        return expand_vec_cond_expr_p (vectype, comp_vectype,
> -				     TREE_CODE (cond_expr));
> +				     cond_code);
>      }
>  
>    /* Transform.  */
> @@ -7858,11 +7924,11 @@ vectorizable_condition (gimple *stmt, gi
>  	      auto_vec<vec<tree>, 4> vec_defs;
>  
>  	      if (masked)
> -		  ops.safe_push (cond_expr);
> +		ops.safe_push (cond_expr);
>  	      else
>  		{
> -		  ops.safe_push (TREE_OPERAND (cond_expr, 0));
> -		  ops.safe_push (TREE_OPERAND (cond_expr, 1));
> +		  ops.safe_push (cond_expr0);
> +		  ops.safe_push (cond_expr1);
>  		}
>                ops.safe_push (then_clause);
>                ops.safe_push (else_clause);
> @@ -7886,17 +7952,15 @@ vectorizable_condition (gimple *stmt, gi
>  		}
>  	      else
>  		{
> -		  vec_cond_lhs =
> -		    vect_get_vec_def_for_operand (TREE_OPERAND (cond_expr, 0),
> -						  stmt, comp_vectype);
> -		  vect_is_simple_use (TREE_OPERAND (cond_expr, 0),
> -				      loop_vinfo, &gtemp, &dts[0]);
> -
> -		  vec_cond_rhs =
> -		    vect_get_vec_def_for_operand (TREE_OPERAND (cond_expr, 1),
> -						  stmt, comp_vectype);
> -		  vect_is_simple_use (TREE_OPERAND (cond_expr, 1),
> -				      loop_vinfo, &gtemp, &dts[1]);
> +		  vec_cond_lhs
> +		    = vect_get_vec_def_for_operand (cond_expr0,
> +						    stmt, comp_vectype);
> +		  vect_is_simple_use (cond_expr0, loop_vinfo, &gtemp, &dts[0]);
> +
> +		  vec_cond_rhs
> +		    = vect_get_vec_def_for_operand (cond_expr1,
> +						    stmt, comp_vectype);
> +		  vect_is_simple_use (cond_expr1, loop_vinfo, &gtemp, &dts[1]);
>  		}
>  	      if (reduc_index == 1)
>  		vec_then_clause = reduc_def;
> @@ -7953,8 +8017,37 @@ vectorizable_condition (gimple *stmt, gi
>  	  else
>  	    {
>  	      vec_cond_rhs = vec_oprnds1[i];
> -	      vec_compare = build2 (TREE_CODE (cond_expr), vec_cmp_type,
> -				    vec_cond_lhs, vec_cond_rhs);
> +	      if (bitop1 == NOP_EXPR)
> +		vec_compare = build2 (cond_code, vec_cmp_type,
> +				      vec_cond_lhs, vec_cond_rhs);
> +	      else
> +		{
> +		  new_temp = make_ssa_name (vec_cmp_type);
> +		  if (bitop1 == BIT_NOT_EXPR)
> +		    new_stmt = gimple_build_assign (new_temp, bitop1,
> +						    vec_cond_rhs);
> +		  else
> +		    new_stmt
> +		      = gimple_build_assign (new_temp, bitop1, vec_cond_lhs,
> +					     vec_cond_rhs);
> +		  vect_finish_stmt_generation (stmt, new_stmt, gsi);
> +		  if (bitop2 == NOP_EXPR)
> +		    vec_compare = new_temp;
> +		  else if (bitop2 == BIT_NOT_EXPR)
> +		    {
> +		      /* Instead of doing ~x ? y : z do x ? z : y.  */
> +		      vec_compare = new_temp;
> +		      std::swap (vec_then_clause, vec_else_clause);
> +		    }
> +		  else
> +		    {
> +		      vec_compare = make_ssa_name (vec_cmp_type);
> +		      new_stmt
> +			= gimple_build_assign (vec_compare, bitop2,
> +					       vec_cond_lhs, new_temp);
> +		      vect_finish_stmt_generation (stmt, new_stmt, gsi);
> +		    }
> +		}
>  	    }
>            new_temp = make_ssa_name (vec_dest);
>            new_stmt = gimple_build_assign (new_temp, VEC_COND_EXPR,
> --- gcc/testsuite/gcc.dg/vect/pr78938.c.jj	2017-01-05 16:06:37.854958230 +0100
> +++ gcc/testsuite/gcc.dg/vect/pr78938.c	2017-01-05 16:07:04.770609976 +0100
> @@ -0,0 +1,18 @@
> +/* PR tree-optimization/78938 */
> +/* { dg-do compile } */
> +/* { dg-additional-options "-mavx512bw" { target i?86-*-* x86_64-*-* } } */
> +
> +short int v;
> +
> +int
> +foo (char x, int *b)
> +{
> +  int a = 1;
> +  for (; x < 1; x++)
> +    {
> +      int c = *b != v;
> +      int d = x != 0;
> +      a = c == d ? 2 : 0;
> +    }
> +  return a;
> +}
> 
> 	Jakub
> 
> 

-- 
Richard Biener <rguenther@suse.de>
SUSE LINUX GmbH, GF: Felix Imendoerffer, Jane Smithard, Graham Norton, HRB 21284 (AG Nuernberg)


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]