Re: [PATCH V3 05/11] bpf: new GCC port

[jose dot marchesi at oracle dot com (Jose E. Marchesi)]

Hi Richard.

    > +(define_insn "mulsidi3"
    > +  [(set (match_operand:DI	   0 "register_operand" "=r,r")
    > +        (sign_extend:DI
    > +         (mult:SI (match_operand:SI 1 "register_operand" "0,0")
    > +                  (match_operand:SI 2 "reg_or_imm_operand" "r,I"))))]
    > +  ""
    > +  "mul32\t%0,%2"
    > +  [(set_attr "type" "alu32")])
    
    Sorry, Segher was right and I was wrong: mulsidi3 is instead:
    
      [(set (match_operand:DI	   0 "register_operand" "=r,r")
            (mult:DI
             (sign_extend:DI (match_operand:SI 1 "register_operand" "0,0"))
             (sign_extend:DI (match_operand:SI 2 "reg_or_imm_operand" "r,I"))))]
    
    i.e. extend the operands rather than the result.  So the define_insn
    shouldn't be called mulsidi3 after all.
    
    Are you sure this is a sign extension though?  From a quick look at the
    kernel sources, I got the impression it was a zero extension instead.

You are right.  This is from linux/kernel/bpf/core.c:

#define ALU(OPCODE, OP)			\
	ALU64_##OPCODE##_X:		\
		DST = DST OP SRC;	\
		CONT;			\
	ALU_##OPCODE##_X:		\
		DST = (u32) DST OP (u32) SRC;	\
		CONT;			\
	ALU64_##OPCODE##_K:		\
		DST = DST OP IMM;		\
		CONT;			\
	ALU_##OPCODE##_K:		\
		DST = (u32) DST OP (u32) IMM;	\
		CONT;

[...]

	ALU(MUL,  *)

So, mul32 zero-extends arguments and then multiplies, leaving the result
in the 64-bit register.  This is an unsigned widening multiplication,
that according to the internal manuals should be handled with a pattern
like:

(define_insn "umulsidi3"
  [(set (match_operand:DI	   0 "register_operand" "=r,r")
        (mult:DI
          (zero_extend:DI (match_operand:SI 1 "register_operand" "0,0"))
          (zero_extend:DI (match_operand:SI 2 "reg_or_imm_operand" "r,I"))))]
  ""
  "mul32\t%0,%2"
  [(set_attr "type" "alu32")])