[02/10] autovectorizer: Add basic support for convert optabs

Message ID 20240710140602.1707875-3-victor.donascimento@arm.com
State New
Headers
Series Make `dot_prod' a convert-type optab |

Commit Message

Victor Do Nascimento July 10, 2024, 2:05 p.m. UTC
  Given the shift from modeling dot products as direct optabs to
treating them as conversion optabs, we make necessary changes to the
autovectorizer code to ensure that given the relevant tree code,
together with the input and output data modes, we can retrieve the
relevant optab and subsequently the insn_code for it.

gcc/ChangeLog:

	* gimple-match-exports.cc (directly_supported_p): Add overload
	for conversion-type optabs.
	* gimple-match.h (directly_supported_p): Add new function
	prototype.
	* optabs.cc (expand_widen_pattern_expr): Make the
	DOT_PROD_EXPR tree code use `find_widening_optab_handler' to
	retrieve icode.
	* tree-vect-loop.cc (vect_is_emulated_mixed_dot_prod): make it
	call conversion-type overloaded `directly_supported_p'.
	* tree-vect-patterns.cc (vect_supportable_conv_optab_p): New.
	(vect_recog_dot_prod_pattern): s/direct/conv/ in call to
	`vect_supportable_direct_optab_p'.
---
 gcc/gimple-match-exports.cc | 18 ++++++++++++++++
 gcc/gimple-match.h          |  2 ++
 gcc/optabs.cc               |  3 ++-
 gcc/tree-vect-loop.cc       |  1 +
 gcc/tree-vect-patterns.cc   | 43 +++++++++++++++++++++++++++++++++++--
 5 files changed, 64 insertions(+), 3 deletions(-)
  

Comments

Tamar Christina July 11, 2024, 7:15 a.m. UTC | #1
Hi Victor,

> -----Original Message-----
> From: Victor Do Nascimento <victor.donascimento@arm.com>
> Sent: Wednesday, July 10, 2024 3:06 PM
> To: gcc-patches@gcc.gnu.org
> Cc: Richard Sandiford <Richard.Sandiford@arm.com>; Richard Earnshaw
> <Richard.Earnshaw@arm.com>; Victor Do Nascimento
> <Victor.DoNascimento@arm.com>
> Subject: [PATCH 02/10] autovectorizer: Add basic support for convert optabs
> 
> Given the shift from modeling dot products as direct optabs to
> treating them as conversion optabs, we make necessary changes to the
> autovectorizer code to ensure that given the relevant tree code,
> together with the input and output data modes, we can retrieve the
> relevant optab and subsequently the insn_code for it.
> 
> gcc/ChangeLog:
> 
> 	* gimple-match-exports.cc (directly_supported_p): Add overload
> 	for conversion-type optabs.
> 	* gimple-match.h (directly_supported_p): Add new function
> 	prototype.
> 	* optabs.cc (expand_widen_pattern_expr): Make the
> 	DOT_PROD_EXPR tree code use `find_widening_optab_handler' to
> 	retrieve icode.
> 	* tree-vect-loop.cc (vect_is_emulated_mixed_dot_prod): make it
> 	call conversion-type overloaded `directly_supported_p'.
> 	* tree-vect-patterns.cc (vect_supportable_conv_optab_p): New.
> 	(vect_recog_dot_prod_pattern): s/direct/conv/ in call to
> 	`vect_supportable_direct_optab_p'.
> ---
>  gcc/gimple-match-exports.cc | 18 ++++++++++++++++
>  gcc/gimple-match.h          |  2 ++
>  gcc/optabs.cc               |  3 ++-
>  gcc/tree-vect-loop.cc       |  1 +
>  gcc/tree-vect-patterns.cc   | 43 +++++++++++++++++++++++++++++++++++--
>  5 files changed, 64 insertions(+), 3 deletions(-)
> 
> diff --git a/gcc/gimple-match-exports.cc b/gcc/gimple-match-exports.cc
> index aacf3ff0414..c079fa1fb19 100644
> --- a/gcc/gimple-match-exports.cc
> +++ b/gcc/gimple-match-exports.cc
> @@ -1381,6 +1381,24 @@ directly_supported_p (code_helper code, tree type,
> optab_subtype query_type)
>  	  && direct_internal_fn_supported_p (ifn, type, OPTIMIZE_FOR_SPEED));
>  }
> 
> +/* As above, overloading the function for conversion-type optabs.  */
> +bool
> +directly_supported_p (code_helper code, tree type_out, tree type_in,
> +		      optab_subtype query_type)
> +{
> +
> +  if (code.is_tree_code ())
> +    {
> +      convert_optab optab = optab_for_tree_code (tree_code (code), type_in,
> +						query_type);
> +      return (optab != unknown_optab
> +	      && convert_optab_handler (optab, TYPE_MODE (type_out),
> +					TYPE_MODE (type_in)) !=
> CODE_FOR_nothing);
> +    }
> +  gcc_unreachable ();
> +}
> +

Since the argument is a code_helper, why not provide the implementation for internal_fn as well,
similar to the existing directly_supported_p.  You can get the optab from an ifn with direct_internal_fn_optab.

>  /* A wrapper around the internal-fn.cc versions of get_conditional_internal_fn
>     for a code_helper CODE operating on type TYPE.  */
> 
> diff --git a/gcc/gimple-match.h b/gcc/gimple-match.h
> index d710fcbace2..0333a5db00a 100644
> --- a/gcc/gimple-match.h
> +++ b/gcc/gimple-match.h
> @@ -419,6 +419,8 @@ code_helper canonicalize_code (code_helper, tree);
> 
>  #ifdef GCC_OPTABS_TREE_H
>  bool directly_supported_p (code_helper, tree, optab_subtype = optab_default);
> +bool directly_supported_p (code_helper, tree, tree,
> +			   optab_subtype = optab_default);
>  #endif
> 
>  internal_fn get_conditional_internal_fn (code_helper, tree);
> diff --git a/gcc/optabs.cc b/gcc/optabs.cc
> index 185c5b1a705..32737fb80e8 100644
> --- a/gcc/optabs.cc
> +++ b/gcc/optabs.cc
> @@ -317,7 +317,8 @@ expand_widen_pattern_expr (const_sepops ops, rtx op0,
> rtx op1, rtx wide_op,
>      widen_pattern_optab
>        = optab_for_tree_code (ops->code, TREE_TYPE (oprnd0), optab_default);
>    if (ops->code == WIDEN_MULT_PLUS_EXPR
> -      || ops->code == WIDEN_MULT_MINUS_EXPR)
> +      || ops->code == WIDEN_MULT_MINUS_EXPR
> +      || ops->code == DOT_PROD_EXPR)
>      icode = find_widening_optab_handler (widen_pattern_optab,
>  					 TYPE_MODE (TREE_TYPE (ops->op2)),
>  					 tmode0);
> diff --git a/gcc/tree-vect-loop.cc b/gcc/tree-vect-loop.cc
> index a64b5082bd1..7e4c1e0f52e 100644
> --- a/gcc/tree-vect-loop.cc
> +++ b/gcc/tree-vect-loop.cc
> @@ -5289,6 +5289,7 @@ vect_is_emulated_mixed_dot_prod (stmt_vec_info
> stmt_info)
> 
>    gcc_assert (STMT_VINFO_REDUC_VECTYPE_IN (stmt_info));
>    return !directly_supported_p (DOT_PROD_EXPR,
> +				STMT_VINFO_VECTYPE (stmt_info),
>  				STMT_VINFO_REDUC_VECTYPE_IN (stmt_info),
>  				optab_vector_mixed_sign);
>  }
> diff --git a/gcc/tree-vect-patterns.cc b/gcc/tree-vect-patterns.cc
> index 86e893a1c43..c4dd627aa90 100644
> --- a/gcc/tree-vect-patterns.cc
> +++ b/gcc/tree-vect-patterns.cc
> @@ -248,6 +248,45 @@ vect_supportable_direct_optab_p (vec_info *vinfo, tree
> otype, tree_code code,
>    return true;
>  }
> 
> +/* Return true if the target supports a vector version of CODE,
> +   where CODE is known to map to a conversion optab with the given SUBTYPE.
> +   ITYPE specifies the type of (some of) the scalar inputs and OTYPE
> +   specifies the type of the scalar result.
> +
> +   When returning true, set *VECOTYPE_OUT to the vector version of OTYPE.
> +   Also set *VECITYPE_OUT to the vector version of ITYPE if VECITYPE_OUT
> +   is nonnull.  */
> +
> +static bool
> +vect_supportable_conv_optab_p (vec_info *vinfo, tree otype, tree_code code,
> +				 tree itype, tree *vecotype_out,
> +				 tree *vecitype_out = NULL,
> +				 enum optab_subtype subtype = optab_default)
> +{
> +  tree vecitype = get_vectype_for_scalar_type (vinfo, itype);
> +  tree vecotype = get_vectype_for_scalar_type (vinfo, otype);
> +  if (!vecitype || !vecotype)
> +    return false;
> +
> +  optab optab = optab_for_tree_code (code, vecitype, subtype);
> +  if (!optab)
> +    return false;
> +
> +  insn_code icode = convert_optab_handler (optab, TYPE_MODE (vecotype),
> +					   TYPE_MODE (vecitype));
> +
> +  if (icode == CODE_FOR_nothing
> +      || insn_data[icode].operand[0].mode != TYPE_MODE (vecotype)
> +      || insn_data[icode].operand[1].mode != TYPE_MODE (vecitype))
> +    return false;
> +
> +  *vecotype_out = vecotype;
> +  if (vecitype_out)
> +    *vecitype_out = vecitype;
> +  return true;
> +}
> +

Aside from the creation of the vector modes, isn't the rest of this essentially
what directly_supported_p does?

Thanks for working on this,
Tamar

> +
>  /* Round bit precision PRECISION up to a full element.  */
> 
>  static unsigned int
> @@ -1266,13 +1305,13 @@ vect_recog_dot_prod_pattern (vec_info *vinfo,
>      half_type = signed_type_for (half_type);
> 
>    tree half_vectype;
> -  if (!vect_supportable_direct_optab_p (vinfo, type, DOT_PROD_EXPR, half_type,
> +  if (!vect_supportable_conv_optab_p (vinfo, type, DOT_PROD_EXPR, half_type,
>  					type_out, &half_vectype, subtype))
>      {
>        /* We can emulate a mixed-sign dot-product using a sequence of
>  	 signed dot-products; see vect_emulate_mixed_dot_prod for details.  */
>        if (subtype != optab_vector_mixed_sign
> -	  || !vect_supportable_direct_optab_p (vinfo, signed_type_for (type),
> +	  || !vect_supportable_conv_optab_p (vinfo, signed_type_for (type),
>  					       DOT_PROD_EXPR, half_type,
>  					       type_out, &half_vectype,
>  					       optab_vector))
> --
> 2.34.1
  

Patch

diff --git a/gcc/gimple-match-exports.cc b/gcc/gimple-match-exports.cc
index aacf3ff0414..c079fa1fb19 100644
--- a/gcc/gimple-match-exports.cc
+++ b/gcc/gimple-match-exports.cc
@@ -1381,6 +1381,24 @@  directly_supported_p (code_helper code, tree type, optab_subtype query_type)
 	  && direct_internal_fn_supported_p (ifn, type, OPTIMIZE_FOR_SPEED));
 }
 
+/* As above, overloading the function for conversion-type optabs.  */
+bool
+directly_supported_p (code_helper code, tree type_out, tree type_in,
+		      optab_subtype query_type)
+{
+
+  if (code.is_tree_code ())
+    {
+      convert_optab optab = optab_for_tree_code (tree_code (code), type_in,
+						query_type);
+      return (optab != unknown_optab
+	      && convert_optab_handler (optab, TYPE_MODE (type_out),
+					TYPE_MODE (type_in)) != CODE_FOR_nothing);
+    }
+  gcc_unreachable ();
+}
+
+
 /* A wrapper around the internal-fn.cc versions of get_conditional_internal_fn
    for a code_helper CODE operating on type TYPE.  */
 
diff --git a/gcc/gimple-match.h b/gcc/gimple-match.h
index d710fcbace2..0333a5db00a 100644
--- a/gcc/gimple-match.h
+++ b/gcc/gimple-match.h
@@ -419,6 +419,8 @@  code_helper canonicalize_code (code_helper, tree);
 
 #ifdef GCC_OPTABS_TREE_H
 bool directly_supported_p (code_helper, tree, optab_subtype = optab_default);
+bool directly_supported_p (code_helper, tree, tree,
+			   optab_subtype = optab_default);
 #endif
 
 internal_fn get_conditional_internal_fn (code_helper, tree);
diff --git a/gcc/optabs.cc b/gcc/optabs.cc
index 185c5b1a705..32737fb80e8 100644
--- a/gcc/optabs.cc
+++ b/gcc/optabs.cc
@@ -317,7 +317,8 @@  expand_widen_pattern_expr (const_sepops ops, rtx op0, rtx op1, rtx wide_op,
     widen_pattern_optab
       = optab_for_tree_code (ops->code, TREE_TYPE (oprnd0), optab_default);
   if (ops->code == WIDEN_MULT_PLUS_EXPR
-      || ops->code == WIDEN_MULT_MINUS_EXPR)
+      || ops->code == WIDEN_MULT_MINUS_EXPR
+      || ops->code == DOT_PROD_EXPR)
     icode = find_widening_optab_handler (widen_pattern_optab,
 					 TYPE_MODE (TREE_TYPE (ops->op2)),
 					 tmode0);
diff --git a/gcc/tree-vect-loop.cc b/gcc/tree-vect-loop.cc
index a64b5082bd1..7e4c1e0f52e 100644
--- a/gcc/tree-vect-loop.cc
+++ b/gcc/tree-vect-loop.cc
@@ -5289,6 +5289,7 @@  vect_is_emulated_mixed_dot_prod (stmt_vec_info stmt_info)
 
   gcc_assert (STMT_VINFO_REDUC_VECTYPE_IN (stmt_info));
   return !directly_supported_p (DOT_PROD_EXPR,
+				STMT_VINFO_VECTYPE (stmt_info),
 				STMT_VINFO_REDUC_VECTYPE_IN (stmt_info),
 				optab_vector_mixed_sign);
 }
diff --git a/gcc/tree-vect-patterns.cc b/gcc/tree-vect-patterns.cc
index 86e893a1c43..c4dd627aa90 100644
--- a/gcc/tree-vect-patterns.cc
+++ b/gcc/tree-vect-patterns.cc
@@ -248,6 +248,45 @@  vect_supportable_direct_optab_p (vec_info *vinfo, tree otype, tree_code code,
   return true;
 }
 
+/* Return true if the target supports a vector version of CODE,
+   where CODE is known to map to a conversion optab with the given SUBTYPE.
+   ITYPE specifies the type of (some of) the scalar inputs and OTYPE
+   specifies the type of the scalar result.
+
+   When returning true, set *VECOTYPE_OUT to the vector version of OTYPE.
+   Also set *VECITYPE_OUT to the vector version of ITYPE if VECITYPE_OUT
+   is nonnull.  */
+
+static bool
+vect_supportable_conv_optab_p (vec_info *vinfo, tree otype, tree_code code,
+				 tree itype, tree *vecotype_out,
+				 tree *vecitype_out = NULL,
+				 enum optab_subtype subtype = optab_default)
+{
+  tree vecitype = get_vectype_for_scalar_type (vinfo, itype);
+  tree vecotype = get_vectype_for_scalar_type (vinfo, otype);
+  if (!vecitype || !vecotype)
+    return false;
+
+  optab optab = optab_for_tree_code (code, vecitype, subtype);
+  if (!optab)
+    return false;
+
+  insn_code icode = convert_optab_handler (optab, TYPE_MODE (vecotype),
+					   TYPE_MODE (vecitype));
+
+  if (icode == CODE_FOR_nothing
+      || insn_data[icode].operand[0].mode != TYPE_MODE (vecotype)
+      || insn_data[icode].operand[1].mode != TYPE_MODE (vecitype))
+    return false;
+
+  *vecotype_out = vecotype;
+  if (vecitype_out)
+    *vecitype_out = vecitype;
+  return true;
+}
+
+
 /* Round bit precision PRECISION up to a full element.  */
 
 static unsigned int
@@ -1266,13 +1305,13 @@  vect_recog_dot_prod_pattern (vec_info *vinfo,
     half_type = signed_type_for (half_type);
 
   tree half_vectype;
-  if (!vect_supportable_direct_optab_p (vinfo, type, DOT_PROD_EXPR, half_type,
+  if (!vect_supportable_conv_optab_p (vinfo, type, DOT_PROD_EXPR, half_type,
 					type_out, &half_vectype, subtype))
     {
       /* We can emulate a mixed-sign dot-product using a sequence of
 	 signed dot-products; see vect_emulate_mixed_dot_prod for details.  */
       if (subtype != optab_vector_mixed_sign
-	  || !vect_supportable_direct_optab_p (vinfo, signed_type_for (type),
+	  || !vect_supportable_conv_optab_p (vinfo, signed_type_for (type),
 					       DOT_PROD_EXPR, half_type,
 					       type_out, &half_vectype,
 					       optab_vector))