Re: [guss-hackers] machine description format

[Johan Rydberg]

On Mon, 2 Jun 2003 18:09:52 +0200
Johan Rydberg <address@hidden> wrote:

: (define_insn "add+el+gl"
:    (+ (f-opcode 0x01))
:    "DATASIZE32"
:    (set (match_operand:SI 0 "general_modrm")
:         (plus:SI (match_dup:SI 0)
:                  (match_operand:SI 1 "general_register")))
: )

Another idea is that force the match_operand RTX to have a third
argument, specifying several possible versions of the operand.
Each version of a operand can have its own simulator semantic
implementation.  This will increase the performance of the 
interpreter.

 (set (match_operand:SI 0 "general_modrm" "m,r")
       (plus:SI (match_dup:SI 0)
                (match_operand:SI 1 "general_register" "r")))

The example above can result in two semantic implementations.
One (the first) that where the first operand can be a memory
address, and one where the operand is a simple register.
Which version of the semantic implementation that should be
used it determined when the insn in decoded and general_modrm
sees what type of mod/rm byte is located in the insn stream.

The third argument specifies what type of operand that is 
allowed for the insn, not which operands that is supported by
general_modrm.  

brgds,
Johan Rydberg