mips.md.svn-base

PSP用开发必装库GCC4

	    (minus:SI (match_operand:SI 1 "register_operand" "d")
		      (match_operand:SI 2 "register_operand" "d"))))]
  "TARGET_64BIT"
  "subu\t%0,%1,%2"
  [(set_attr "type" "arith")
   (set_attr "mode" "DI")])

;;
;;  ....................
;;
;;	MULTIPLICATION
;;
;;  ....................
;;

(define_expand "mul<mode>3"
  [(set (match_operand:SCALARF 0 "register_operand")
	(mult:SCALARF (match_operand:SCALARF 1 "register_operand")
		      (match_operand:SCALARF 2 "register_operand")))]
  ""
  "")

(define_insn "*mul<mode>3"
  [(set (match_operand:SCALARF 0 "register_operand" "=f")
	(mult:SCALARF (match_operand:SCALARF 1 "register_operand" "f")
		      (match_operand:SCALARF 2 "register_operand" "f")))]
  "!TARGET_4300_MUL_FIX"
  "mul.<fmt>\t%0,%1,%2"
  [(set_attr "type" "fmul")
   (set_attr "mode" "<MODE>")])

;; Early VR4300 silicon has a CPU bug where multiplies with certain
;; operands may corrupt immediately following multiplies. This is a
;; simple fix to insert NOPs.

(define_insn "*mul<mode>3_r4300"
  [(set (match_operand:SCALARF 0 "register_operand" "=f")
	(mult:SCALARF (match_operand:SCALARF 1 "register_operand" "f")
		      (match_operand:SCALARF 2 "register_operand" "f")))]
  "TARGET_4300_MUL_FIX"
  "mul.<fmt>\t%0,%1,%2\;nop"
  [(set_attr "type" "fmul")
   (set_attr "mode" "<MODE>")
   (set_attr "length" "8")])

(define_insn "mulv2sf3"
  [(set (match_operand:V2SF 0 "register_operand" "=f")
	(mult:V2SF (match_operand:V2SF 1 "register_operand" "f")
		   (match_operand:V2SF 2 "register_operand" "f")))]
  "TARGET_PAIRED_SINGLE_FLOAT"
  "mul.ps\t%0,%1,%2"
  [(set_attr "type" "fmul")
   (set_attr "mode" "SF")])

;; The original R4000 has a cpu bug.  If a double-word or a variable
;; shift executes while an integer multiplication is in progress, the
;; shift may give an incorrect result.  Avoid this by keeping the mflo
;; with the mult on the R4000.
;;
;; From "MIPS R4000PC/SC Errata, Processor Revision 2.2 and 3.0"
;; (also valid for MIPS R4000MC processors):
;;
;; "16. R4000PC, R4000SC: Please refer to errata 28 for an update to
;;	this errata description.
;;	The following code sequence causes the R4000 to incorrectly
;;	execute the Double Shift Right Arithmetic 32 (dsra32)
;;	instruction.  If the dsra32 instruction is executed during an
;;	integer multiply, the dsra32 will only shift by the amount in
;;	specified in the instruction rather than the amount plus 32
;;	bits.
;;	instruction 1:		mult	rs,rt		integer multiply
;;	instruction 2-12:	dsra32	rd,rt,rs	doubleword shift
;;							right arithmetic + 32
;;	Workaround: A dsra32 instruction placed after an integer
;;	multiply should not be one of the 11 instructions after the
;;	multiply instruction."
;;
;; and:
;;
;; "28. R4000PC, R4000SC: The text from errata 16 should be replaced by
;;	the following description.
;;	All extended shifts (shift by n+32) and variable shifts (32 and
;;	64-bit versions) may produce incorrect results under the
;;	following conditions:
;;	1) An integer multiply is currently executing
;;	2) These types of shift instructions are executed immediately
;;	   following an integer divide instruction.
;;	Workaround:
;;	1) Make sure no integer multiply is running wihen these
;;	   instruction are executed.  If this cannot be predicted at
;;	   compile time, then insert a "mfhi" to R0 instruction
;;	   immediately after the integer multiply instruction.  This
;;	   will cause the integer multiply to complete before the shift
;;	   is executed.
;;	2) Separate integer divide and these two classes of shift
;;	   instructions by another instruction or a noop."
;;
;; These processors have PRId values of 0x00004220 and 0x00004300,
;; respectively.

(define_expand "mul<mode>3"
  [(set (match_operand:GPR 0 "register_operand")
	(mult:GPR (match_operand:GPR 1 "register_operand")
		  (match_operand:GPR 2 "register_operand")))]
  ""
{
  if (GENERATE_MULT3_<MODE>)
    emit_insn (gen_mul<mode>3_mult3 (operands[0], operands[1], operands[2]));
  else if (!TARGET_FIX_R4000)
    emit_insn (gen_mul<mode>3_internal (operands[0], operands[1],
					operands[2]));
  else
    emit_insn (gen_mul<mode>3_r4000 (operands[0], operands[1], operands[2]));
  DONE;
})

(define_insn "mulsi3_mult3"
  [(set (match_operand:SI 0 "register_operand" "=d,l")
	(mult:SI (match_operand:SI 1 "register_operand" "d,d")
		 (match_operand:SI 2 "register_operand" "d,d")))
   (clobber (match_scratch:SI 3 "=h,h"))
   (clobber (match_scratch:SI 4 "=l,X"))]
  "GENERATE_MULT3_SI"
{
  if (which_alternative == 1)
    return "mult\t%1,%2";
  if (TARGET_MAD
      || TARGET_MIPS5400
      || TARGET_MIPS5500
      || TARGET_MIPS7000
      || TARGET_MIPS9000
      || ISA_MIPS32
      || ISA_MIPS32R2
      || ISA_MIPS64)
    return "mul\t%0,%1,%2";
  return "mult\t%0,%1,%2";
}
  [(set_attr "type" "imul")
   (set_attr "mode" "SI")])

(define_insn "muldi3_mult3"
  [(set (match_operand:DI 0 "register_operand" "=d")
	(mult:DI (match_operand:DI 1 "register_operand" "d")
		 (match_operand:DI 2 "register_operand" "d")))
   (clobber (match_scratch:DI 3 "=h"))
   (clobber (match_scratch:DI 4 "=l"))]
  "TARGET_64BIT && GENERATE_MULT3_DI"
  "dmult\t%0,%1,%2"
  [(set_attr "type" "imul")
   (set_attr "mode" "DI")])

;; If a register gets allocated to LO, and we spill to memory, the reload
;; will include a move from LO to a GPR.  Merge it into the multiplication
;; if it can set the GPR directly.
;;
;; Operand 0: LO
;; Operand 1: GPR (1st multiplication operand)
;; Operand 2: GPR (2nd multiplication operand)
;; Operand 3: HI
;; Operand 4: GPR (destination)
(define_peephole2
  [(parallel
       [(set (match_operand:SI 0 "register_operand")
	     (mult:SI (match_operand:SI 1 "register_operand")
		      (match_operand:SI 2 "register_operand")))
        (clobber (match_operand:SI 3 "register_operand"))
        (clobber (scratch:SI))])
   (set (match_operand:SI 4 "register_operand")
	(unspec [(match_dup 0) (match_dup 3)] UNSPEC_MFHILO))]
  "GENERATE_MULT3_SI && peep2_reg_dead_p (2, operands[0])"
  [(parallel
       [(set (match_dup 4)
	     (mult:SI (match_dup 1)
		      (match_dup 2)))
        (clobber (match_dup 3))
        (clobber (match_dup 0))])])

(define_insn "mul<mode>3_internal"
  [(set (match_operand:GPR 0 "register_operand" "=l")
	(mult:GPR (match_operand:GPR 1 "register_operand" "d")
		  (match_operand:GPR 2 "register_operand" "d")))
   (clobber (match_scratch:GPR 3 "=h"))]
  "!TARGET_FIX_R4000"
  "<d>mult\t%1,%2"
  [(set_attr "type" "imul")
   (set_attr "mode" "<MODE>")])

(define_insn "mul<mode>3_r4000"
  [(set (match_operand:GPR 0 "register_operand" "=d")
	(mult:GPR (match_operand:GPR 1 "register_operand" "d")
		  (match_operand:GPR 2 "register_operand" "d")))
   (clobber (match_scratch:GPR 3 "=h"))
   (clobber (match_scratch:GPR 4 "=l"))]
  "TARGET_FIX_R4000"
  "<d>mult\t%1,%2\;mflo\t%0"
  [(set_attr "type" "imul")
   (set_attr "mode" "<MODE>")
   (set_attr "length" "8")])

;; On the VR4120 and VR4130, it is better to use "mtlo $0; macc" instead
;; of "mult; mflo".  They have the same latency, but the first form gives
;; us an extra cycle to compute the operands.

;; Operand 0: LO
;; Operand 1: GPR (1st multiplication operand)
;; Operand 2: GPR (2nd multiplication operand)
;; Operand 3: HI
;; Operand 4: GPR (destination)
(define_peephole2
  [(parallel
       [(set (match_operand:SI 0 "register_operand")
	     (mult:SI (match_operand:SI 1 "register_operand")
		      (match_operand:SI 2 "register_operand")))
        (clobber (match_operand:SI 3 "register_operand"))])
   (set (match_operand:SI 4 "register_operand")
	(unspec:SI [(match_dup 0) (match_dup 3)] UNSPEC_MFHILO))]
  "ISA_HAS_MACC && !GENERATE_MULT3_SI"
  [(set (match_dup 0)
	(const_int 0))
   (parallel
       [(set (match_dup 0)
	     (plus:SI (mult:SI (match_dup 1)
			       (match_dup 2))
		      (match_dup 0)))
	(set (match_dup 4)
	     (plus:SI (mult:SI (match_dup 1)
			       (match_dup 2))
		      (match_dup 0)))
        (clobber (match_dup 3))])])

;; Multiply-accumulate patterns

;; For processors that can copy the output to a general register:
;;
;; The all-d alternative is needed because the combiner will find this
;; pattern and then register alloc/reload will move registers around to
;; make them fit, and we don't want to trigger unnecessary loads to LO.
;;
;; The last alternative should be made slightly less desirable, but adding
;; "?" to the constraint is too strong, and causes values to be loaded into
;; LO even when that's more costly.  For now, using "*d" mostly does the
;; trick.
(define_insn "*mul_acc_si"
  [(set (match_operand:SI 0 "register_operand" "=l,*d,*d")
	(plus:SI (mult:SI (match_operand:SI 1 "register_operand" "d,d,d")
			  (match_operand:SI 2 "register_operand" "d,d,d"))
		 (match_operand:SI 3 "register_operand" "0,l,*d")))
   (clobber (match_scratch:SI 4 "=h,h,h"))
   (clobber (match_scratch:SI 5 "=X,3,l"))
   (clobber (match_scratch:SI 6 "=X,X,&d"))]
  "(TARGET_MIPS3900
   || ISA_HAS_MADD_MSUB)
   && !TARGET_MIPS16"
{
  static const char *const madd[] = { "madd\t%1,%2", "madd\t%0,%1,%2" };
  if (which_alternative == 2)
    return "#";
  if (ISA_HAS_MADD_MSUB && which_alternative != 0)
    return "#";
  return madd[which_alternative];
}
  [(set_attr "type"	"imadd,imadd,multi")
   (set_attr "mode"	"SI")
   (set_attr "length"	"4,4,8")])

;; Split the above insn if we failed to get LO allocated.
(define_split
  [(set (match_operand:SI 0 "register_operand")
	(plus:SI (mult:SI (match_operand:SI 1 "register_operand")
			  (match_operand:SI 2 "register_operand"))
		 (match_operand:SI 3 "register_operand")))
   (clobber (match_scratch:SI 4))
   (clobber (match_scratch:SI 5))
   (clobber (match_scratch:SI 6))]
  "reload_completed && !TARGET_DEBUG_D_MODE
   && GP_REG_P (true_regnum (operands[0]))
   && GP_REG_P (true_regnum (operands[3]))"
  [(parallel [(set (match_dup 6)
		   (mult:SI (match_dup 1) (match_dup 2)))
	      (clobber (match_dup 4))
	      (clobber (match_dup 5))])
   (set (match_dup 0) (plus:SI (match_dup 6) (match_dup 3)))]
  "")

;; Splitter to copy result of MADD to a general register
(define_split
  [(set (match_operand:SI                   0 "register_operand")
        (plus:SI (mult:SI (match_operand:SI 1 "register_operand")
                          (match_operand:SI 2 "register_operand"))
                 (match_operand:SI          3 "register_operand")))
   (clobber (match_scratch:SI               4))
   (clobber (match_scratch:SI               5))
   (clobber (match_scratch:SI               6))]
  "reload_completed && !TARGET_DEBUG_D_MODE
   && GP_REG_P (true_regnum (operands[0]))
   && true_regnum (operands[3]) == LO_REGNUM"
  [(parallel [(set (match_dup 3)
                   (plus:SI (mult:SI (match_dup 1) (match_dup 2))
                            (match_dup 3)))
              (clobber (match_dup 4))
              (clobber (match_dup 5))
              (clobber (match_dup 6))])
   (set (match_dup 0) (unspec:SI [(match_dup 5) (match_dup 4)] UNSPEC_MFHILO))]
  "")

(define_insn "*macc"
  [(set (match_operand:SI 0 "register_operand" "=l,d")
	(plus:SI (mult:SI (match_operand:SI 1 "register_operand" "d,d")
			  (match_operand:SI 2 "register_operand" "d,d"))
		 (match_operand:SI 3 "register_operand" "0,l")))
   (clobber (match_scratch:SI 4 "=h,h"))
   (clobber (match_scratch:SI 5 "=X,3"))]
  "ISA_HAS_MACC"
{
  if (which_alternative == 1)
    return "macc\t%0,%1,%2";
  else if (TARGET_MIPS5500)
    return "madd\t%1,%2";
  else
    /* The VR4130 assumes that there is a two-cycle latency between a macc
       that "writes" to $0 and an instruction that reads from it.  We avoid
       this by assigning to $1 instead.  */
    return "%[macc\t%@,%1,%2%]";
}
  [(set_attr "type" "imadd")
   (set_attr "mode" "SI")])

(define_insn "*msac"
  [(set (match_operand:SI 0 "register_operand" "=l,d")
        (minus:SI (match_operand:SI 1 "register_operand" "0,l")
                  (mult:SI (match_operand:SI 2 "register_operand" "d,d")
                           (match_operand:SI 3 "register_operand" "d,d"))))
   (clobber (match_scratch:SI 4 "=h,h"))
   (clobber (match_scratch:SI 5 "=X,1"))]
  "ISA_HAS_MSAC"
{
  if (which_alternative == 1)
    return "msac\t%0,%2,%3";
  else if (TARGET_MIPS5500)
    return "msub\t%2,%3";
  else
    return "msac\t$0,%2,%3";
}
  [(set_attr "type"     "imadd")
   (set_attr "mode"     "SI")])

;; An msac-like instruction implemented using negation and a macc.
(define_insn_and_split "*msac_using_macc"
  [(set (match_operand:SI 0 "register_operand" "=l,d")
        (minus:SI (match_operand:SI 1 "register_operand" "0,l")
                  (mult:SI (match_operand:SI 2 "register_operand" "d,d")
                           (match_operand:SI 3 "register_operand" "d,d"))))
   (clobber (match_scratch:SI 4 "=h,h"))
   (clobber (match_scratch:SI 5 "=X,1"))
   (clobber (match_scratch:SI 6 "=d,d"))]
  "ISA_HAS_MACC && !ISA_HAS_MSAC"
  "#"
  "&& reload_completed"
  [(set (match_dup 6)
	(neg:SI (match_dup 3)))
   (parallel
       [(set (match_dup 0)
	     (plus:SI (mult:SI (match_dup 2)
			       (match_dup 6))
		      (match_dup 1)))
	(clobber (match_dup 4))
	(clobber (match_dup 5))])]
  ""
  [(set_attr "type"     "imadd")
   (set_attr "length"	"8")])

;; Patterns generated by the define_peephole2 below.

(define_insn "*macc2"
  [(set (match_operand:SI 0 "register_operand" "=l")
	(plus:SI (mult:SI (match_operand:SI 1 "register_operand" "d")
			  (match_operand:SI 2 "register_operand" "d"))
		 (match_dup 0)))
   (set (match_operand:SI 3 "register_operand" "=d")
	(plus:SI (mult:SI (match_dup 1)
			  (match_dup 2))
		 (match_dup 0)))
   (clobber (match_scratch:SI 4 "=h"))]
  "ISA_HAS_MACC && reload_completed"
  "macc\t%3,%1,%2"
  [(set_attr "type"	"imadd")
   (set_attr "mode"	"SI")])

(define_insn "*msac2"
  [(set (match_operand:SI 0 "register_operand" "=l")
	(minus:SI (match_dup 0)
		  (mult:SI (match_operand:SI 1 "register_operand" "d")
			   (match_operand:SI 2 "register_operand" "d"))))
   (set (match_operand:SI 3 "register_operand" "=d")
	(minus:SI (match_dup 0)
		  (mult:SI (match_dup 1)
			   (match_dup 2))))
   (clobber (match_scratch:SI 4 "=h"))]
  "ISA_HAS_MSAC && reload_completed"
  "msac\t%3,%1,%2"
  [(set_attr "type"	"imadd")
   (set_attr "mode"	"SI")])

;; Convert macc $0,<r1>,<r2> & mflo <r3> into macc <r3>,<r1>,<r2>