Previous: , Up: Match and Simplify   [Contents][Index]

25.2 The Language

The language to write expression simplifications in resembles other domain-specific languages GCC uses. Thus it is lispy. Lets start with an example from the match.pd file:

  (bit_and @0 integer_all_onesp)

This example contains all required parts of an expression simplification. A simplification is wrapped inside a (simplify ...) expression. That contains at least two operands - an expression that is matched with the GIMPLE or GENERIC IL and a replacement expression that is returned if the match was successful.

Expressions have an operator ID, bit_and in this case. Expressions can be lower-case tree codes with _expr stripped off or builtin function code names in all-caps, like BUILT_IN_SQRT.

@n denotes a so-called capture. It captures the operand and lets you refer to it in other places of the match-and-simplify. In the above example it is refered to in the replacement expression. Captures are @ followed by a number or an identifier.

  (bit_xor @0 @0)
  { build_zero_cst (type); })

In this example @0 is mentioned twice which constrains the matched expression to have two equal operands. Usually matches are constraint to equal types. If operands may be constants and conversions are involved matching by value might be preferred in which case use @@0 to denote a by value match and the specific operand you want to refer to in the result part. This example also introduces operands written in C code. These can be used in the expression replacements and are supposed to evaluate to a tree node which has to be a valid GIMPLE operand (so you cannot generate expressions in C code).

  (trunc_mod integer_zerop@0 @1)
  (if (!integer_zerop (@1))

Here @0 captures the first operand of the trunc_mod expression which is also predicated with integer_zerop. Expression operands may be either expressions, predicates or captures. Captures can be unconstrained or capture expresions or predicates.

This example introduces an optional operand of simplify, the if-expression. This condition is evaluated after the expression matched in the IL and is required to evaluate to true to enable the replacement expression in the second operand position. The expression operand of the if is a standard C expression which may contain references to captures. The if has an optional third operand which may contain the replacement expression that is enabled when the condition evaluates to false.

A if expression can be used to specify a common condition for multiple simplify patterns, avoiding the need to repeat that multiple times:

(if (!TYPE_SATURATING (type)
     && !FLOAT_TYPE_P (type) && !FIXED_POINT_TYPE_P (type))
    (minus (plus @0 @1) @0)
    (minus (minus @0 @1) @0)
    (negate @1)))

Note that ifs in outer position do not have the optional else clause but instead have multiple then clauses.

Ifs can be nested.

There exists a switch expression which can be used to chain conditions avoiding nesting ifs too much:

 (simple_comparison @0 REAL_CST@1)
  /* a CMP (-0) -> a CMP 0  */
   (cmp @0 { build_real (TREE_TYPE (@1), dconst0); }))
  /* x != NaN is always true, other ops are always false.  */
       && ! HONOR_SNANS (@1))
   { constant_boolean_node (cmp == NE_EXPR, type); })))

Is equal to

 (simple_comparison @0 REAL_CST@1)
  /* a CMP (-0) -> a CMP 0  */
   (cmp @0 { build_real (TREE_TYPE (@1), dconst0); })
   /* x != NaN is always true, other ops are always false.  */
        && ! HONOR_SNANS (@1))
    { constant_boolean_node (cmp == NE_EXPR, type); }))))

which has the second if in the else operand of the first. The switch expression takes if expressions as operands (which may not have else clauses) and as a last operand a replacement expression which should be enabled by default if no other condition evaluated to true.

Captures can also be used for capturing results of sub-expressions.

  (pointer_plus (addr@2 @0) INTEGER_CST_P@1)
  (if (is_gimple_min_invariant (@2)))
    HOST_WIDE_INT off;
    tree base = get_addr_base_and_unit_offset (@0, &off);
    off += tree_to_uhwi (@1);
    /* Now with that we should be able to simply write
       (addr (mem_ref (addr @base) (plus @off @1)))  */
    build1 (ADDR_EXPR, type,
            build2 (MEM_REF, TREE_TYPE (TREE_TYPE (@2)),
                    build_fold_addr_expr (base),
                    build_int_cst (ptr_type_node, off)));

In the above example, @2 captures the result of the expression (addr @0). For outermost expression only its type can be captured, and the keyword type is reserved for this purpose. The above example also gives a way to conditionalize patterns to only apply to GIMPLE or GENERIC by means of using the pre-defined preprocessor macros GIMPLE and GENERIC and using preprocessor directives.

  (bit_and:c integral_op_p@0 (bit_ior:c (bit_not @0) @1))
  (bit_and @1 @0))

Here we introduce flags on match expressions. The flag used above, c, denotes that the expression should be also matched commutated. Thus the above match expression is really the following four match expressions:

  (bit_and integral_op_p@0 (bit_ior (bit_not @0) @1))
  (bit_and (bit_ior (bit_not @0) @1) integral_op_p@0)
  (bit_and integral_op_p@0 (bit_ior @1 (bit_not @0)))
  (bit_and (bit_ior @1 (bit_not @0)) integral_op_p@0)

Usual canonicalizations you know from GENERIC expressions are applied before matching, so for example constant operands always come second in commutative expressions.

The second supported flag is s which tells the code generator to fail the pattern if the expression marked with s does have more than one use. For example in

  (pointer_plus (pointer_plus:s @0 @1) @3)
  (pointer_plus @0 (plus @1 @3)))

this avoids the association if (pointer_plus @0 @1) is used outside of the matched expression and thus it would stay live and not trivially removed by dead code elimination.

More features exist to avoid too much repetition.

(for op (plus pointer_plus minus bit_ior bit_xor)
    (op @0 integer_zerop)

A for expression can be used to repeat a pattern for each operator specified, substituting op. for can be nested and a for can have multiple operators to iterate.

(for opa (plus minus)
     opb (minus plus)
  (for opc (plus minus)

In this example the pattern will be repeated four times with opa, opb, opc being plus, minus, plus, plus, minus, minus, minus, plus, plus, minus, plus, minus.

To avoid repeating operator lists in for you can name them via

(define_operator_list pmm plus minus mult)

and use them in for operator lists where they get expanded.

(for opa (pmm trunc_div)

So this example iterates over plus, minus, mult and trunc_div.

Using operator lists can also remove the need to explicitely write a for. All operator list uses that appear in a simplify or match pattern in operator positions will implicitely be added to a new for. For example

 (SQRT (POW @0 @1))
 (POW (abs @0) (mult @1 { built_real (TREE_TYPE (@1), dconsthalf); })))

is the same as

  (SQRT (POW @0 @1))
  (POW (abs @0) (mult @1 { built_real (TREE_TYPE (@1), dconsthalf); }))))

fors and operator lists can include the special identifier null that matches nothing and can never be generated. This can be used to pad an operator list so that it has a standard form, even if there isn’t a suitable operator for every form.

Another building block are with expressions in the result expression which nest the generated code in a new C block followed by its argument:

 (convert (mult @0 @1))
 (with { tree utype = unsigned_type_for (type); }
  (convert (mult (convert:utype @0) (convert:utype @1)))))

This allows code nested in the with to refer to the declared variables. In the above case we use the feature to specify the type of a generated expression with the :type syntax where type needs to be an identifier that refers to the desired type. Usually the types of the generated result expressions are determined from the context, but sometimes like in the above case it is required that you specify them explicitely.

As intermediate conversions are often optional there is a way to avoid the need to repeat patterns both with and without such conversions. Namely you can mark a conversion as being optional with a ?:

 (eq (convert@0 @1) (convert? @2))
 (eq @1 (convert @2)))

which will match both (eq (convert @1) (convert @2)) and (eq (convert @1) @2). The optional converts are supposed to be all either present or not, thus (eq (convert? @1) (convert? @2)) will result in two patterns only. If you want to match all four combinations you have access to two additional conditional converts as in (eq (convert1? @1) (convert2? @2)).

Predicates available from the GCC middle-end need to be made available explicitely via define_predicates:

 integer_onep integer_zerop integer_all_onesp)

You can also define predicates using the pattern matching language and the match form:

(match negate_expr_p
      || may_negate_without_overflow_p (t))))
(match negate_expr_p
 (negate @0))

This shows that for match expressions there is t available which captures the outermost expression (something not possible in the simplify context). As you can see match has an identifier as first operand which is how you refer to the predicate in patterns. Multiple match for the same identifier add additional cases where the predicate matches.

Predicates can also match an expression in which case you need to provide a template specifying the identifier and where to get its operands from:

(match (logical_inverted_value @0)
 (eq @0 integer_zerop))
(match (logical_inverted_value @0)
 (bit_not truth_valued_p@0))

You can use the above predicate like

 (bit_and @0 (logical_inverted_value @0))
 { build_zero_cst (type); })

Which will match a bitwise and of an operand with its logical inverted value.

Previous: , Up: Match and Simplify   [Contents][Index]