17.16 Describing Relative Costs of Operations

These macros let you describe the relative speed of various operations on the target machine.

Macro: REGISTER_MOVE_COST (mode, from, to)

A C expression for the cost of moving data of mode mode from a register in class from to one in class to. The classes are expressed using the enumeration values such as GENERAL_REGS. A value of 2 is the default; other values are interpreted relative to that.

It is not required that the cost always equal 2 when from is the same as to; on some machines it is expensive to move between registers if they are not general registers.

If reload sees an insn consisting of a single set between two hard registers, and if REGISTER_MOVE_COST applied to their classes returns a value of 2, reload does not check to ensure that the constraints of the insn are met. Setting a cost of other than 2 will allow reload to verify that the constraints are met. You should do this if the ‘movm’ pattern’s constraints do not allow such copying.

These macros are obsolete, new ports should use the target hook TARGET_REGISTER_MOVE_COST instead.

Target Hook: int TARGET_REGISTER_MOVE_COST (machine_mode mode, reg_class_t from, reg_class_t to)

This target hook should return the cost of moving data of mode mode from a register in class from to one in class to. The classes are expressed using the enumeration values such as GENERAL_REGS. A value of 2 is the default; other values are interpreted relative to that.

It is not required that the cost always equal 2 when from is the same as to; on some machines it is expensive to move between registers if they are not general registers.

If reload sees an insn consisting of a single set between two hard registers, and if TARGET_REGISTER_MOVE_COST applied to their classes returns a value of 2, reload does not check to ensure that the constraints of the insn are met. Setting a cost of other than 2 will allow reload to verify that the constraints are met. You should do this if the ‘movm’ pattern’s constraints do not allow such copying.

The default version of this function returns 2.

Macro: MEMORY_MOVE_COST (mode, class, in)

A C expression for the cost of moving data of mode mode between a register of class class and memory; in is zero if the value is to be written to memory, nonzero if it is to be read in. This cost is relative to those in REGISTER_MOVE_COST. If moving between registers and memory is more expensive than between two registers, you should define this macro to express the relative cost.

If you do not define this macro, GCC uses a default cost of 4 plus the cost of copying via a secondary reload register, if one is needed. If your machine requires a secondary reload register to copy between memory and a register of class but the reload mechanism is more complex than copying via an intermediate, define this macro to reflect the actual cost of the move.

GCC defines the function memory_move_secondary_cost if secondary reloads are needed. It computes the costs due to copying via a secondary register. If your machine copies from memory using a secondary register in the conventional way but the default base value of 4 is not correct for your machine, define this macro to add some other value to the result of that function. The arguments to that function are the same as to this macro.

These macros are obsolete, new ports should use the target hook TARGET_MEMORY_MOVE_COST instead.

Target Hook: int TARGET_MEMORY_MOVE_COST (machine_mode mode, reg_class_t rclass, bool in)

This target hook should return the cost of moving data of mode mode between a register of class rclass and memory; in is false if the value is to be written to memory, true if it is to be read in. This cost is relative to those in TARGET_REGISTER_MOVE_COST. If moving between registers and memory is more expensive than between two registers, you should add this target hook to express the relative cost.

If you do not add this target hook, GCC uses a default cost of 4 plus the cost of copying via a secondary reload register, if one is needed. If your machine requires a secondary reload register to copy between memory and a register of rclass but the reload mechanism is more complex than copying via an intermediate, use this target hook to reflect the actual cost of the move.

GCC defines the function memory_move_secondary_cost if secondary reloads are needed. It computes the costs due to copying via a secondary register. If your machine copies from memory using a secondary register in the conventional way but the default base value of 4 is not correct for your machine, use this target hook to add some other value to the result of that function. The arguments to that function are the same as to this target hook.

Macro: BRANCH_COST (speed_p, predictable_p)

A C expression for the cost of a branch instruction. A value of 1 is the default; other values are interpreted relative to that. Parameter speed_p is true when the branch in question should be optimized for speed. When it is false, BRANCH_COST should return a value optimal for code size rather than performance. predictable_p is true for well-predicted branches. On many architectures the BRANCH_COST can be reduced then.

Here are additional macros which do not specify precise relative costs, but only that certain actions are more expensive than GCC would ordinarily expect.

Macro: SLOW_BYTE_ACCESS

Define this macro as a C expression which is nonzero if accessing less than a word of memory (i.e. a char or a short) is no faster than accessing a word of memory, i.e., if such access require more than one instruction or if there is no difference in cost between byte and (aligned) word loads.

When this macro is not defined, the compiler will access a field by finding the smallest containing object; when it is defined, a fullword load will be used if alignment permits. Unless bytes accesses are faster than word accesses, using word accesses is preferable since it may eliminate subsequent memory access if subsequent accesses occur to other fields in the same word of the structure, but to different bytes.

Macro: SLOW_UNALIGNED_ACCESS (mode, alignment)

Define this macro to be the value 1 if memory accesses described by the mode and alignment parameters have a cost many times greater than aligned accesses, for example if they are emulated in a trap handler. This macro is invoked only for unaligned accesses, i.e. when alignment < GET_MODE_ALIGNMENT (mode).

When this macro is nonzero, the compiler will act as if STRICT_ALIGNMENT were nonzero when generating code for block moves. This can cause significantly more instructions to be produced. Therefore, do not set this macro nonzero if unaligned accesses only add a cycle or two to the time for a memory access.

If the value of this macro is always zero, it need not be defined. If this macro is defined, it should produce a nonzero value when STRICT_ALIGNMENT is nonzero.

Macro: MOVE_RATIO (speed)

The threshold of number of scalar memory-to-memory move insns, below which a sequence of insns should be generated instead of a string move insn or a library call. Increasing the value will always make code faster, but eventually incurs high cost in increased code size.

Note that on machines where the corresponding move insn is a define_expand that emits a sequence of insns, this macro counts the number of such sequences.

The parameter speed is true if the code is currently being optimized for speed rather than size.

If you don’t define this, a reasonable default is used.

Target Hook: bool TARGET_USE_BY_PIECES_INFRASTRUCTURE_P (unsigned HOST_WIDE_INT size, unsigned int alignment, enum by_pieces_operation op, bool speed_p)

GCC will attempt several strategies when asked to copy between two areas of memory, or to set, clear or store to memory, for example when copying a struct. The by_pieces infrastructure implements such memory operations as a sequence of load, store or move insns. Alternate strategies are to expand the movmem or setmem optabs, to emit a library call, or to emit unit-by-unit, loop-based operations.

This target hook should return true if, for a memory operation with a given size and alignment, using the by_pieces infrastructure is expected to result in better code generation. Both size and alignment are measured in terms of storage units.

The parameter op is one of: CLEAR_BY_PIECES, MOVE_BY_PIECES, SET_BY_PIECES, STORE_BY_PIECES or COMPARE_BY_PIECES. These describe the type of memory operation under consideration.

The parameter speed_p is true if the code is currently being optimized for speed rather than size.

Returning true for higher values of size can improve code generation for speed if the target does not provide an implementation of the movmem or setmem standard names, if the movmem or setmem implementation would be more expensive than a sequence of insns, or if the overhead of a library call would dominate that of the body of the memory operation.

Returning true for higher values of size may also cause an increase in code size, for example where the number of insns emitted to perform a move would be greater than that of a library call.

Target Hook: int TARGET_COMPARE_BY_PIECES_BRANCH_RATIO (machine_mode mode)

When expanding a block comparison in MODE, gcc can try to reduce the number of branches at the expense of more memory operations. This hook allows the target to override the default choice. It should return the factor by which branches should be reduced over the plain expansion with one comparison per mode-sized piece. A port can also prevent a particular mode from being used for block comparisons by returning a negative number from this hook.

Macro: MOVE_MAX_PIECES

A C expression used by move_by_pieces to determine the largest unit a load or store used to copy memory is. Defaults to MOVE_MAX.

Macro: STORE_MAX_PIECES

A C expression used by store_by_pieces to determine the largest unit a store used to memory is. Defaults to MOVE_MAX_PIECES, or two times the size of HOST_WIDE_INT, whichever is smaller.

Macro: COMPARE_MAX_PIECES

A C expression used by compare_by_pieces to determine the largest unit a load or store used to compare memory is. Defaults to MOVE_MAX_PIECES.

Macro: CLEAR_RATIO (speed)

The threshold of number of scalar move insns, below which a sequence of insns should be generated to clear memory instead of a string clear insn or a library call. Increasing the value will always make code faster, but eventually incurs high cost in increased code size.

The parameter speed is true if the code is currently being optimized for speed rather than size.

If you don’t define this, a reasonable default is used.

Macro: SET_RATIO (speed)

The threshold of number of scalar move insns, below which a sequence of insns should be generated to set memory to a constant value, instead of a block set insn or a library call. Increasing the value will always make code faster, but eventually incurs high cost in increased code size.

The parameter speed is true if the code is currently being optimized for speed rather than size.

If you don’t define this, it defaults to the value of MOVE_RATIO.

Macro: USE_LOAD_POST_INCREMENT (mode)

A C expression used to determine whether a load postincrement is a good thing to use for a given mode. Defaults to the value of HAVE_POST_INCREMENT.

Macro: USE_LOAD_POST_DECREMENT (mode)

A C expression used to determine whether a load postdecrement is a good thing to use for a given mode. Defaults to the value of HAVE_POST_DECREMENT.

Macro: USE_LOAD_PRE_INCREMENT (mode)

A C expression used to determine whether a load preincrement is a good thing to use for a given mode. Defaults to the value of HAVE_PRE_INCREMENT.

Macro: USE_LOAD_PRE_DECREMENT (mode)

A C expression used to determine whether a load predecrement is a good thing to use for a given mode. Defaults to the value of HAVE_PRE_DECREMENT.

Macro: USE_STORE_POST_INCREMENT (mode)

A C expression used to determine whether a store postincrement is a good thing to use for a given mode. Defaults to the value of HAVE_POST_INCREMENT.

Macro: USE_STORE_POST_DECREMENT (mode)

A C expression used to determine whether a store postdecrement is a good thing to use for a given mode. Defaults to the value of HAVE_POST_DECREMENT.

Macro: USE_STORE_PRE_INCREMENT (mode)

This macro is used to determine whether a store preincrement is a good thing to use for a given mode. Defaults to the value of HAVE_PRE_INCREMENT.

Macro: USE_STORE_PRE_DECREMENT (mode)

This macro is used to determine whether a store predecrement is a good thing to use for a given mode. Defaults to the value of HAVE_PRE_DECREMENT.

Macro: NO_FUNCTION_CSE

Define this macro to be true if it is as good or better to call a constant function address than to call an address kept in a register.

Macro: LOGICAL_OP_NON_SHORT_CIRCUIT

Define this macro if a non-short-circuit operation produced by ‘fold_range_test ()’ is optimal. This macro defaults to true if BRANCH_COST is greater than or equal to the value 2.

Target Hook: bool TARGET_OPTAB_SUPPORTED_P (int op, machine_mode mode1, machine_mode mode2, optimization_type opt_type)

Return true if the optimizers should use optab op with modes mode1 and mode2 for optimization type opt_type. The optab is known to have an associated .md instruction whose C condition is true. mode2 is only meaningful for conversion optabs; for direct optabs it is a copy of mode1.

For example, when called with op equal to rint_optab and mode1 equal to DFmode, the hook should say whether the optimizers should use optab rintdf2.

The default hook returns true for all inputs.

Target Hook: bool TARGET_RTX_COSTS (rtx x, machine_mode mode, int outer_code, int opno, int *total, bool speed)

This target hook describes the relative costs of RTL expressions.

The cost may depend on the precise form of the expression, which is available for examination in x, and the fact that x appears as operand opno of an expression with rtx code outer_code. That is, the hook can assume that there is some rtx y such that ‘GET_CODE (y) == outer_code’ and such that either (a) ‘XEXP (y, opno) == x’ or (b) ‘XVEC (y, opno)’ contains x.

mode is x’s machine mode, or for cases like const_int that do not have a mode, the mode in which x is used.

In implementing this hook, you can use the construct COSTS_N_INSNS (n) to specify a cost equal to n fast instructions.

On entry to the hook, *total contains a default estimate for the cost of the expression. The hook should modify this value as necessary. Traditionally, the default costs are COSTS_N_INSNS (5) for multiplications, COSTS_N_INSNS (7) for division and modulus operations, and COSTS_N_INSNS (1) for all other operations.

When optimizing for code size, i.e. when speed is false, this target hook should be used to estimate the relative size cost of an expression, again relative to COSTS_N_INSNS.

The hook returns true when all subexpressions of x have been processed, and false when rtx_cost should recurse.

Target Hook: int TARGET_ADDRESS_COST (rtx address, machine_mode mode, addr_space_t as, bool speed)

This hook computes the cost of an addressing mode that contains address. If not defined, the cost is computed from the address expression and the TARGET_RTX_COST hook.

For most CISC machines, the default cost is a good approximation of the true cost of the addressing mode. However, on RISC machines, all instructions normally have the same length and execution time. Hence all addresses will have equal costs.

In cases where more than one form of an address is known, the form with the lowest cost will be used. If multiple forms have the same, lowest, cost, the one that is the most complex will be used.

For example, suppose an address that is equal to the sum of a register and a constant is used twice in the same basic block. When this macro is not defined, the address will be computed in a register and memory references will be indirect through that register. On machines where the cost of the addressing mode containing the sum is no higher than that of a simple indirect reference, this will produce an additional instruction and possibly require an additional register. Proper specification of this macro eliminates this overhead for such machines.

This hook is never called with an invalid address.

On machines where an address involving more than one register is as cheap as an address computation involving only one register, defining TARGET_ADDRESS_COST to reflect this can cause two registers to be live over a region of code where only one would have been if TARGET_ADDRESS_COST were not defined in that manner. This effect should be considered in the definition of this macro. Equivalent costs should probably only be given to addresses with different numbers of registers on machines with lots of registers.

Target Hook: unsigned int TARGET_MAX_NOCE_IFCVT_SEQ_COST (edge e)

This hook returns a value in the same units as TARGET_RTX_COSTS, giving the maximum acceptable cost for a sequence generated by the RTL if-conversion pass when conditional execution is not available. The RTL if-conversion pass attempts to convert conditional operations that would require a branch to a series of unconditional operations and movmodecc insns. This hook returns the maximum cost of the unconditional instructions and the movmodecc insns. RTL if-conversion is cancelled if the cost of the converted sequence is greater than the value returned by this hook.

e is the edge between the basic block containing the conditional branch to the basic block which would be executed if the condition were true.

The default implementation of this hook uses the max-rtl-if-conversion-[un]predictable parameters if they are set, and uses a multiple of BRANCH_COST otherwise.

Target Hook: bool TARGET_NOCE_CONVERSION_PROFITABLE_P (rtx_insn *seq, struct noce_if_info *if_info)

This hook returns true if the instruction sequence seq is a good candidate as a replacement for the if-convertible sequence described in if_info.

Target Hook: bool TARGET_NO_SPECULATION_IN_DELAY_SLOTS_P (void)

This predicate controls the use of the eager delay slot filler to disallow speculatively executed instructions being placed in delay slots. Targets such as certain MIPS architectures possess both branches with and without delay slots. As the eager delay slot filler can decrease performance, disabling it is beneficial when ordinary branches are available. Use of delay slot branches filled using the basic filler is often still desirable as the delay slot can hide a pipeline bubble.