blob: 9ae10f2ba930edbd7569e3cbc2598643f34cb1d1 [file] [log] [blame]
%def binop(preinstr="", result="a0", chkzero="0", instr=""):
/*
* Generic 32-bit binary operation. Provide an "instr" line that
* specifies an instruction that performs "result = a0 op a1".
* This could be a MIPS instruction or a function call. (If the result
* comes back in a register other than a0, you can override "result".)
*
* If "chkzero" is set to 1, we perform a divide-by-zero check on
* vCC (a1). Useful for integer division and modulus. Note that we
* *don't* check for (INT_MIN / -1) here, because the CPU handles it
* correctly.
*
* For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
* xor-int, shl-int, shr-int, ushr-int
*/
/* binop vAA, vBB, vCC */
FETCH(a0, 1) # a0 <- CCBB
GET_OPA(rOBJ) # rOBJ <- AA
srl a3, a0, 8 # a3 <- CC
and a2, a0, 255 # a2 <- BB
GET_VREG(a1, a3) # a1 <- vCC
GET_VREG(a0, a2) # a0 <- vBB
.if $chkzero
# is second operand zero?
beqz a1, common_errDivideByZero
.endif
FETCH_ADVANCE_INST(2) # advance rPC, load rINST
$preinstr # optional op
$instr # $result <- op, a0-a3 changed
GET_INST_OPCODE(t0) # extract opcode from rINST
SET_VREG_GOTO($result, rOBJ, t0) # vAA <- $result
%def binop2addr(preinstr="", result="a0", chkzero="0", instr=""):
/*
* Generic 32-bit "/2addr" binary operation. Provide an "instr" line
* that specifies an instruction that performs "result = a0 op a1".
* This could be an MIPS instruction or a function call.
*
* If "chkzero" is set to 1, we perform a divide-by-zero check on
* vCC (a1). Useful for integer division and modulus.
*
* For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
* rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
* shl-int/2addr, shr-int/2addr, ushr-int/2addr
*/
/* binop/2addr vA, vB */
GET_OPA4(rOBJ) # rOBJ <- A+
GET_OPB(a3) # a3 <- B
GET_VREG(a0, rOBJ) # a0 <- vA
GET_VREG(a1, a3) # a1 <- vB
.if $chkzero
# is second operand zero?
beqz a1, common_errDivideByZero
.endif
FETCH_ADVANCE_INST(1) # advance rPC, load rINST
$preinstr # optional op
$instr # $result <- op, a0-a3 changed
GET_INST_OPCODE(t0) # extract opcode from rINST
SET_VREG_GOTO($result, rOBJ, t0) # vA <- $result
%def binopLit16(preinstr="", result="a0", chkzero="0", instr=""):
/*
* Generic 32-bit "lit16" binary operation. Provide an "instr" line
* that specifies an instruction that performs "result = a0 op a1".
* This could be an MIPS instruction or a function call. (If the result
* comes back in a register other than a0, you can override "result".)
*
* If "chkzero" is set to 1, we perform a divide-by-zero check on
* vCC (a1). Useful for integer division and modulus.
*
* For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
* rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
*/
/* binop/lit16 vA, vB, +CCCC */
FETCH_S(a1, 1) # a1 <- ssssCCCC (sign-extended)
GET_OPB(a2) # a2 <- B
GET_OPA4(rOBJ) # rOBJ <- A+
GET_VREG(a0, a2) # a0 <- vB
.if $chkzero
# cmp a1, 0; is second operand zero?
beqz a1, common_errDivideByZero
.endif
FETCH_ADVANCE_INST(2) # advance rPC, load rINST
$preinstr # optional op
$instr # $result <- op, a0-a3 changed
GET_INST_OPCODE(t0) # extract opcode from rINST
SET_VREG_GOTO($result, rOBJ, t0) # vA <- $result
%def binopLit8(preinstr="", result="a0", chkzero="0", instr=""):
/*
* Generic 32-bit "lit8" binary operation. Provide an "instr" line
* that specifies an instruction that performs "result = a0 op a1".
* This could be an MIPS instruction or a function call. (If the result
* comes back in a register other than a0, you can override "result".)
*
* If "chkzero" is set to 1, we perform a divide-by-zero check on
* vCC (a1). Useful for integer division and modulus.
*
* For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
* rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
* shl-int/lit8, shr-int/lit8, ushr-int/lit8
*/
/* binop/lit8 vAA, vBB, +CC */
FETCH_S(a3, 1) # a3 <- ssssCCBB (sign-extended for CC)
GET_OPA(rOBJ) # rOBJ <- AA
and a2, a3, 255 # a2 <- BB
GET_VREG(a0, a2) # a0 <- vBB
sra a1, a3, 8 # a1 <- ssssssCC (sign extended)
.if $chkzero
# is second operand zero?
beqz a1, common_errDivideByZero
.endif
FETCH_ADVANCE_INST(2) # advance rPC, load rINST
$preinstr # optional op
$instr # $result <- op, a0-a3 changed
GET_INST_OPCODE(t0) # extract opcode from rINST
SET_VREG_GOTO($result, rOBJ, t0) # vAA <- $result
%def binopWide(preinstr="", result0="a0", result1="a1", chkzero="0", arg0="a0", arg1="a1", arg2="a2", arg3="a3", instr=""):
/*
* Generic 64-bit binary operation. Provide an "instr" line that
* specifies an instruction that performs "result = a0-a1 op a2-a3".
* This could be a MIPS instruction or a function call. (If the result
* comes back in a register pair other than a0-a1, you can override "result".)
*
* If "chkzero" is set to 1, we perform a divide-by-zero check on
* vCC (a2-a3). Useful for integer division and modulus.
*
* for: add-long, sub-long, div-long, rem-long, and-long, or-long,
* xor-long
*
* IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
*/
/* binop vAA, vBB, vCC */
FETCH(a0, 1) # a0 <- CCBB
GET_OPA(rOBJ) # rOBJ <- AA
and a2, a0, 255 # a2 <- BB
srl a3, a0, 8 # a3 <- CC
EAS2(a2, rFP, a2) # a2 <- &fp[BB]
EAS2(t1, rFP, a3) # a3 <- &fp[CC]
LOAD64($arg0, $arg1, a2) # a0/a1 <- vBB/vBB+1
LOAD64($arg2, $arg3, t1) # a2/a3 <- vCC/vCC+1
.if $chkzero
or t0, $arg2, $arg3 # second arg (a2-a3) is zero?
beqz t0, common_errDivideByZero
.endif
FETCH_ADVANCE_INST(2) # advance rPC, load rINST
$preinstr # optional op
$instr # result <- op, a0-a3 changed
GET_INST_OPCODE(t0) # extract opcode from rINST
SET_VREG64_GOTO($result0, $result1, rOBJ, t0) # vAA/vAA+1 <- $result0/$result1
%def binopWide2addr(preinstr="", result0="a0", result1="a1", chkzero="0", arg0="a0", arg1="a1", arg2="a2", arg3="a3", instr=""):
/*
* Generic 64-bit "/2addr" binary operation. Provide an "instr" line
* that specifies an instruction that performs "result = a0-a1 op a2-a3".
* This could be a MIPS instruction or a function call. (If the result
* comes back in a register pair other than a0-a1, you can override "result".)
*
* If "chkzero" is set to 1, we perform a divide-by-zero check on
* vB (a2-a3). Useful for integer division and modulus.
*
* For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
* and-long/2addr, or-long/2addr, xor-long/2addr
*/
/* binop/2addr vA, vB */
GET_OPA4(rOBJ) # rOBJ <- A+
GET_OPB(a1) # a1 <- B
EAS2(a1, rFP, a1) # a1 <- &fp[B]
EAS2(t0, rFP, rOBJ) # t0 <- &fp[A]
LOAD64($arg2, $arg3, a1) # a2/a3 <- vB/vB+1
LOAD64($arg0, $arg1, t0) # a0/a1 <- vA/vA+1
.if $chkzero
or t0, $arg2, $arg3 # second arg (a2-a3) is zero?
beqz t0, common_errDivideByZero
.endif
FETCH_ADVANCE_INST(1) # advance rPC, load rINST
$preinstr # optional op
$instr # result <- op, a0-a3 changed
GET_INST_OPCODE(t0) # extract opcode from rINST
SET_VREG64_GOTO($result0, $result1, rOBJ, t0) # vA/vA+1 <- $result0/$result1
%def unop(preinstr="", result0="a0", instr=""):
/*
* Generic 32-bit unary operation. Provide an "instr" line that
* specifies an instruction that performs "result0 = op a0".
* This could be a MIPS instruction or a function call.
*
* for: int-to-byte, int-to-char, int-to-short,
* neg-int, not-int, neg-float
*/
/* unop vA, vB */
GET_OPB(a3) # a3 <- B
GET_OPA4(t0) # t0 <- A+
GET_VREG(a0, a3) # a0 <- vB
FETCH_ADVANCE_INST(1) # advance rPC, load rINST
$preinstr # optional op
$instr # a0 <- op, a0-a3 changed
GET_INST_OPCODE(t1) # extract opcode from rINST
SET_VREG_GOTO($result0, t0, t1) # vA <- result0
%def unopNarrower(load="LOAD64_F(fa0, fa0f, a3)", instr=""):
/*
* Generic 64bit-to-32bit floating-point unary operation. Provide an "instr"
* line that specifies an instruction that performs "fv0 = op fa0".
*
* For: double-to-float
*/
/* unop vA, vB */
GET_OPB(a3) # a3 <- B
GET_OPA4(rOBJ) # rOBJ <- A+
EAS2(a3, rFP, a3) # a3 <- &fp[B]
$load
FETCH_ADVANCE_INST(1) # advance rPC, load rINST
$instr
GET_INST_OPCODE(t0) # extract opcode from rINST
SET_VREG_F_GOTO(fv0, rOBJ, t0) # vA <- fv0
%def unopWide(preinstr="", result0="a0", result1="a1", instr=""):
/*
* Generic 64-bit unary operation. Provide an "instr" line that
* specifies an instruction that performs "result0/result1 = op a0/a1".
* This could be MIPS instruction or a function call.
*
* For: neg-long, not-long, neg-double,
*/
/* unop vA, vB */
GET_OPA4(rOBJ) # rOBJ <- A+
GET_OPB(a3) # a3 <- B
EAS2(a3, rFP, a3) # a3 <- &fp[B]
LOAD64(a0, a1, a3) # a0/a1 <- vA
FETCH_ADVANCE_INST(1) # advance rPC, load rINST
$preinstr # optional op
$instr # a0/a1 <- op, a2-a3 changed
GET_INST_OPCODE(t0) # extract opcode from rINST
SET_VREG64_GOTO($result0, $result1, rOBJ, t0) # vA/vA+1 <- a0/a1
%def unopWider(preinstr="", result0="a0", result1="a1", instr=""):
/*
* Generic 32bit-to-64bit unary operation. Provide an "instr" line
* that specifies an instruction that performs "result0/result1 = op a0".
*
* For: int-to-long
*/
/* unop vA, vB */
GET_OPA4(rOBJ) # rOBJ <- A+
GET_OPB(a3) # a3 <- B
GET_VREG(a0, a3) # a0 <- vB
FETCH_ADVANCE_INST(1) # advance rPC, load rINST
$preinstr # optional op
$instr # result <- op, a0-a3 changed
GET_INST_OPCODE(t0) # extract opcode from rINST
SET_VREG64_GOTO($result0, $result1, rOBJ, t0) # vA/vA+1 <- a0/a1
%def op_add_int():
% binop(instr="addu a0, a0, a1")
%def op_add_int_2addr():
% binop2addr(instr="addu a0, a0, a1")
%def op_add_int_lit16():
% binopLit16(instr="addu a0, a0, a1")
%def op_add_int_lit8():
% binopLit8(instr="addu a0, a0, a1")
%def op_add_long():
/*
* The compiler generates the following sequence for
* [v1 v0] = [a1 a0] + [a3 a2];
* addu v0,a2,a0
* addu a1,a3,a1
* sltu v1,v0,a2
* addu v1,v1,a1
*/
% binopWide(result0="v0", result1="v1", preinstr="addu v0, a2, a0", instr="addu a1, a3, a1; sltu v1, v0, a2; addu v1, v1, a1")
%def op_add_long_2addr():
/*
* See op_add_long.S for details
*/
% binopWide2addr(result0="v0", result1="v1", preinstr="addu v0, a2, a0", instr="addu a1, a3, a1; sltu v1, v0, a2; addu v1, v1, a1")
%def op_and_int():
% binop(instr="and a0, a0, a1")
%def op_and_int_2addr():
% binop2addr(instr="and a0, a0, a1")
%def op_and_int_lit16():
% binopLit16(instr="and a0, a0, a1")
%def op_and_int_lit8():
% binopLit8(instr="and a0, a0, a1")
%def op_and_long():
% binopWide(preinstr="and a0, a0, a2", instr="and a1, a1, a3")
%def op_and_long_2addr():
% binopWide2addr(preinstr="and a0, a0, a2", instr="and a1, a1, a3")
%def op_cmp_long():
/*
* Compare two 64-bit values
* x = y return 0
* x < y return -1
* x > y return 1
*
* I think I can improve on the ARM code by the following observation
* slt t0, x.hi, y.hi; # (x.hi < y.hi) ? 1:0
* sgt t1, x.hi, y.hi; # (y.hi > x.hi) ? 1:0
* subu v0, t0, t1 # v0= -1:1:0 for [ < > = ]
*/
/* cmp-long vAA, vBB, vCC */
FETCH(a0, 1) # a0 <- CCBB
GET_OPA(rOBJ) # rOBJ <- AA
and a2, a0, 255 # a2 <- BB
srl a3, a0, 8 # a3 <- CC
EAS2(a2, rFP, a2) # a2 <- &fp[BB]
EAS2(a3, rFP, a3) # a3 <- &fp[CC]
LOAD64(a0, a1, a2) # a0/a1 <- vBB/vBB+1
LOAD64(a2, a3, a3) # a2/a3 <- vCC/vCC+1
FETCH_ADVANCE_INST(2) # advance rPC, load rINST
slt t0, a1, a3 # compare hi
sgt t1, a1, a3
subu v0, t1, t0 # v0 <- (-1, 1, 0)
bnez v0, .L${opcode}_finish
# at this point x.hi==y.hi
sltu t0, a0, a2 # compare lo
sgtu t1, a0, a2
subu v0, t1, t0 # v0 <- (-1, 1, 0) for [< > =]
.L${opcode}_finish:
GET_INST_OPCODE(t0) # extract opcode from rINST
SET_VREG_GOTO(v0, rOBJ, t0) # vAA <- v0
%def op_div_int():
#ifdef MIPS32REVGE6
% binop(instr="div a0, a0, a1", chkzero="1")
#else
% binop(preinstr="div zero, a0, a1", instr="mflo a0", chkzero="1")
#endif
%def op_div_int_2addr():
#ifdef MIPS32REVGE6
% binop2addr(instr="div a0, a0, a1", chkzero="1")
#else
% binop2addr(preinstr="div zero, a0, a1", instr="mflo a0", chkzero="1")
#endif
%def op_div_int_lit16():
#ifdef MIPS32REVGE6
% binopLit16(instr="div a0, a0, a1", chkzero="1")
#else
% binopLit16(preinstr="div zero, a0, a1", instr="mflo a0", chkzero="1")
#endif
%def op_div_int_lit8():
#ifdef MIPS32REVGE6
% binopLit8(instr="div a0, a0, a1", chkzero="1")
#else
% binopLit8(preinstr="div zero, a0, a1", instr="mflo a0", chkzero="1")
#endif
%def op_div_long():
% binopWide(result0="v0", result1="v1", instr="JAL(__divdi3)", chkzero="1")
%def op_div_long_2addr():
% binopWide2addr(result0="v0", result1="v1", instr="JAL(__divdi3)", chkzero="1")
%def op_int_to_byte():
% unop(instr="SEB(a0, a0)")
%def op_int_to_char():
% unop(preinstr="", instr="and a0, 0xffff")
%def op_int_to_long():
% unopWider(instr="sra a1, a0, 31")
%def op_int_to_short():
% unop(instr="SEH(a0, a0)")
%def op_long_to_int():
/* we ignore the high word, making this equivalent to a 32-bit reg move */
% op_move()
%def op_mul_int():
% binop(instr="mul a0, a0, a1")
%def op_mul_int_2addr():
% binop2addr(instr="mul a0, a0, a1")
%def op_mul_int_lit16():
% binopLit16(instr="mul a0, a0, a1")
%def op_mul_int_lit8():
% binopLit8(instr="mul a0, a0, a1")
%def op_mul_long():
/*
* Signed 64-bit integer multiply.
* a1 a0
* x a3 a2
* -------------
* a2a1 a2a0
* a3a0
* a3a1 (<= unused)
* ---------------
* v1 v0
*/
/* mul-long vAA, vBB, vCC */
FETCH(a0, 1) # a0 <- CCBB
and t0, a0, 255 # a2 <- BB
srl t1, a0, 8 # a3 <- CC
EAS2(t0, rFP, t0) # t0 <- &fp[BB]
LOAD64(a0, a1, t0) # a0/a1 <- vBB/vBB+1
EAS2(t1, rFP, t1) # t0 <- &fp[CC]
LOAD64(a2, a3, t1) # a2/a3 <- vCC/vCC+1
mul v1, a3, a0 # v1= a3a0
#ifdef MIPS32REVGE6
mulu v0, a2, a0 # v0= a2a0
muhu t1, a2, a0
#else
multu a2, a0
mfhi t1
mflo v0 # v0= a2a0
#endif
mul t0, a2, a1 # t0= a2a1
addu v1, v1, t1 # v1+= hi(a2a0)
addu v1, v1, t0 # v1= a3a0 + a2a1;
GET_OPA(a0) # a0 <- AA
FETCH_ADVANCE_INST(2) # advance rPC, load rINST
b .L${opcode}_finish
%def op_mul_long_helper_code():
.Lop_mul_long_finish:
GET_INST_OPCODE(t0) # extract opcode from rINST
SET_VREG64_GOTO(v0, v1, a0, t0) # vAA/vAA+1 <- v0(low)/v1(high)
%def op_mul_long_2addr():
/*
* See op_mul_long.S for more details
*/
/* mul-long/2addr vA, vB */
GET_OPA4(rOBJ) # rOBJ <- A+
EAS2(t0, rFP, rOBJ) # t0 <- &fp[A]
LOAD64(a0, a1, t0) # vAA.low / high
GET_OPB(t1) # t1 <- B
EAS2(t1, rFP, t1) # t1 <- &fp[B]
LOAD64(a2, a3, t1) # vBB.low / high
mul v1, a3, a0 # v1= a3a0
#ifdef MIPS32REVGE6
mulu v0, a2, a0 # v0= a2a0
muhu t1, a2, a0
#else
multu a2, a0
mfhi t1
mflo v0 # v0= a2a0
#endif
mul t2, a2, a1 # t2= a2a1
addu v1, v1, t1 # v1= a3a0 + hi(a2a0)
addu v1, v1, t2 # v1= v1 + a2a1;
FETCH_ADVANCE_INST(1) # advance rPC, load rINST
GET_INST_OPCODE(t1) # extract opcode from rINST
SET_VREG64_GOTO(v0, v1, rOBJ, t1) # vA/vA+1 <- v0(low)/v1(high)
%def op_neg_int():
% unop(instr="negu a0, a0")
%def op_neg_long():
% unopWide(result0="v0", result1="v1", preinstr="negu v0, a0", instr="negu v1, a1; sltu a0, zero, v0; subu v1, v1, a0")
%def op_not_int():
% unop(instr="not a0, a0")
%def op_not_long():
% unopWide(preinstr="not a0, a0", instr="not a1, a1")
%def op_or_int():
% binop(instr="or a0, a0, a1")
%def op_or_int_2addr():
% binop2addr(instr="or a0, a0, a1")
%def op_or_int_lit16():
% binopLit16(instr="or a0, a0, a1")
%def op_or_int_lit8():
% binopLit8(instr="or a0, a0, a1")
%def op_or_long():
% binopWide(preinstr="or a0, a0, a2", instr="or a1, a1, a3")
%def op_or_long_2addr():
% binopWide2addr(preinstr="or a0, a0, a2", instr="or a1, a1, a3")
%def op_rem_int():
#ifdef MIPS32REVGE6
% binop(instr="mod a0, a0, a1", chkzero="1")
#else
% binop(preinstr="div zero, a0, a1", instr="mfhi a0", chkzero="1")
#endif
%def op_rem_int_2addr():
#ifdef MIPS32REVGE6
% binop2addr(instr="mod a0, a0, a1", chkzero="1")
#else
% binop2addr(preinstr="div zero, a0, a1", instr="mfhi a0", chkzero="1")
#endif
%def op_rem_int_lit16():
#ifdef MIPS32REVGE6
% binopLit16(instr="mod a0, a0, a1", chkzero="1")
#else
% binopLit16(preinstr="div zero, a0, a1", instr="mfhi a0", chkzero="1")
#endif
%def op_rem_int_lit8():
#ifdef MIPS32REVGE6
% binopLit8(instr="mod a0, a0, a1", chkzero="1")
#else
% binopLit8(preinstr="div zero, a0, a1", instr="mfhi a0", chkzero="1")
#endif
%def op_rem_long():
% binopWide(result0="v0", result1="v1", instr="JAL(__moddi3)", chkzero="1")
%def op_rem_long_2addr():
% binopWide2addr(result0="v0", result1="v1", instr="JAL(__moddi3)", chkzero="1")
%def op_rsub_int():
/* this op is "rsub-int", but can be thought of as "rsub-int/lit16" */
% binopLit16(instr="subu a0, a1, a0")
%def op_rsub_int_lit8():
% binopLit8(instr="subu a0, a1, a0")
%def op_shl_int():
% binop(instr="sll a0, a0, a1")
%def op_shl_int_2addr():
% binop2addr(instr="sll a0, a0, a1")
%def op_shl_int_lit8():
% binopLit8(instr="sll a0, a0, a1")
%def op_shl_long():
/*
* Long integer shift. This is different from the generic 32/64-bit
* binary operations because vAA/vBB are 64-bit but vCC (the shift
* distance) is 32-bit. Also, Dalvik requires us to mask off the low
* 6 bits of the shift distance.
*/
/* shl-long vAA, vBB, vCC */
FETCH(a0, 1) # a0 <- CCBB
GET_OPA(t2) # t2 <- AA
and a3, a0, 255 # a3 <- BB
srl a0, a0, 8 # a0 <- CC
EAS2(a3, rFP, a3) # a3 <- &fp[BB]
GET_VREG(a2, a0) # a2 <- vCC
LOAD64(a0, a1, a3) # a0/a1 <- vBB/vBB+1
FETCH_ADVANCE_INST(2) # advance rPC, load rINST
GET_INST_OPCODE(t0) # extract opcode from rINST
andi v1, a2, 0x20 # shift< shift & 0x20
sll v0, a0, a2 # rlo<- alo << (shift&31)
bnez v1, .L${opcode}_finish
not v1, a2 # rhi<- 31-shift (shift is 5b)
srl a0, 1
srl a0, v1 # alo<- alo >> (32-(shift&31))
sll v1, a1, a2 # rhi<- ahi << (shift&31)
or v1, a0 # rhi<- rhi | alo
SET_VREG64_GOTO(v0, v1, t2, t0) # vAA/vAA+1 <- v0/v1
%def op_shl_long_helper_code():
.Lop_shl_long_finish:
SET_VREG64_GOTO(zero, v0, t2, t0) # vAA/vAA+1 <- rlo/rhi
%def op_shl_long_2addr():
/*
* Long integer shift, 2addr version. vA is 64-bit value/result, vB is
* 32-bit shift distance.
*/
/* shl-long/2addr vA, vB */
GET_OPA4(rOBJ) # rOBJ <- A+
GET_OPB(a3) # a3 <- B
GET_VREG(a2, a3) # a2 <- vB
EAS2(t2, rFP, rOBJ) # t2 <- &fp[A]
LOAD64(a0, a1, t2) # a0/a1 <- vA/vA+1
FETCH_ADVANCE_INST(1) # advance rPC, load rINST
GET_INST_OPCODE(t0) # extract opcode from rINST
andi v1, a2, 0x20 # shift< shift & 0x20
sll v0, a0, a2 # rlo<- alo << (shift&31)
bnez v1, .L${opcode}_finish
not v1, a2 # rhi<- 31-shift (shift is 5b)
srl a0, 1
srl a0, v1 # alo<- alo >> (32-(shift&31))
sll v1, a1, a2 # rhi<- ahi << (shift&31)
or v1, a0 # rhi<- rhi | alo
SET_VREG64_GOTO(v0, v1, rOBJ, t0) # vA/vA+1 <- v0/v1
%def op_shl_long_2addr_helper_code():
.Lop_shl_long_2addr_finish:
SET_VREG64_GOTO(zero, v0, rOBJ, t0) # vA/vA+1 <- rlo/rhi
%def op_shr_int():
% binop(instr="sra a0, a0, a1")
%def op_shr_int_2addr():
% binop2addr(instr="sra a0, a0, a1")
%def op_shr_int_lit8():
% binopLit8(instr="sra a0, a0, a1")
%def op_shr_long():
/*
* Long integer shift. This is different from the generic 32/64-bit
* binary operations because vAA/vBB are 64-bit but vCC (the shift
* distance) is 32-bit. Also, Dalvik requires us to mask off the low
* 6 bits of the shift distance.
*/
/* shr-long vAA, vBB, vCC */
FETCH(a0, 1) # a0 <- CCBB
GET_OPA(t3) # t3 <- AA
and a3, a0, 255 # a3 <- BB
srl a0, a0, 8 # a0 <- CC
EAS2(a3, rFP, a3) # a3 <- &fp[BB]
GET_VREG(a2, a0) # a2 <- vCC
LOAD64(a0, a1, a3) # a0/a1 <- vBB/vBB+1
FETCH_ADVANCE_INST(2) # advance rPC, load rINST
GET_INST_OPCODE(t0) # extract opcode from rINST
andi v0, a2, 0x20 # shift & 0x20
sra v1, a1, a2 # rhi<- ahi >> (shift&31)
bnez v0, .L${opcode}_finish
srl v0, a0, a2 # rlo<- alo >> (shift&31)
not a0, a2 # alo<- 31-shift (shift is 5b)
sll a1, 1
sll a1, a0 # ahi<- ahi << (32-(shift&31))
or v0, a1 # rlo<- rlo | ahi
SET_VREG64_GOTO(v0, v1, t3, t0) # vAA/VAA+1 <- v0/v1
%def op_shr_long_helper_code():
.Lop_shr_long_finish:
sra a3, a1, 31 # a3<- sign(ah)
SET_VREG64_GOTO(v1, a3, t3, t0) # vAA/VAA+1 <- rlo/rhi
%def op_shr_long_2addr():
/*
* Long integer shift, 2addr version. vA is 64-bit value/result, vB is
* 32-bit shift distance.
*/
/* shr-long/2addr vA, vB */
GET_OPA4(t2) # t2 <- A+
GET_OPB(a3) # a3 <- B
GET_VREG(a2, a3) # a2 <- vB
EAS2(t0, rFP, t2) # t0 <- &fp[A]
LOAD64(a0, a1, t0) # a0/a1 <- vA/vA+1
FETCH_ADVANCE_INST(1) # advance rPC, load rINST
GET_INST_OPCODE(t0) # extract opcode from rINST
andi v0, a2, 0x20 # shift & 0x20
sra v1, a1, a2 # rhi<- ahi >> (shift&31)
bnez v0, .L${opcode}_finish
srl v0, a0, a2 # rlo<- alo >> (shift&31)
not a0, a2 # alo<- 31-shift (shift is 5b)
sll a1, 1
sll a1, a0 # ahi<- ahi << (32-(shift&31))
or v0, a1 # rlo<- rlo | ahi
SET_VREG64_GOTO(v0, v1, t2, t0) # vA/vA+1 <- v0/v1
%def op_shr_long_2addr_helper_code():
.Lop_shr_long_2addr_finish:
sra a3, a1, 31 # a3<- sign(ah)
SET_VREG64_GOTO(v1, a3, t2, t0) # vA/vA+1 <- rlo/rhi
%def op_sub_int():
% binop(instr="subu a0, a0, a1")
%def op_sub_int_2addr():
% binop2addr(instr="subu a0, a0, a1")
%def op_sub_long():
/*
* For little endian the code sequence looks as follows:
* subu v0,a0,a2
* subu v1,a1,a3
* sltu a0,a0,v0
* subu v1,v1,a0
*/
% binopWide(result0="v0", result1="v1", preinstr="subu v0, a0, a2", instr="subu v1, a1, a3; sltu a0, a0, v0; subu v1, v1, a0")
%def op_sub_long_2addr():
/*
* See op_sub_long.S for more details
*/
% binopWide2addr(result0="v0", result1="v1", preinstr="subu v0, a0, a2", instr="subu v1, a1, a3; sltu a0, a0, v0; subu v1, v1, a0")
%def op_ushr_int():
% binop(instr="srl a0, a0, a1")
%def op_ushr_int_2addr():
% binop2addr(instr="srl a0, a0, a1 ")
%def op_ushr_int_lit8():
% binopLit8(instr="srl a0, a0, a1")
%def op_ushr_long():
/*
* Long integer shift. This is different from the generic 32/64-bit
* binary operations because vAA/vBB are 64-bit but vCC (the shift
* distance) is 32-bit. Also, Dalvik requires us to mask off the low
* 6 bits of the shift distance.
*/
/* ushr-long vAA, vBB, vCC */
FETCH(a0, 1) # a0 <- CCBB
GET_OPA(rOBJ) # rOBJ <- AA
and a3, a0, 255 # a3 <- BB
srl a0, a0, 8 # a0 <- CC
EAS2(a3, rFP, a3) # a3 <- &fp[BB]
GET_VREG(a2, a0) # a2 <- vCC
LOAD64(a0, a1, a3) # a0/a1 <- vBB/vBB+1
FETCH_ADVANCE_INST(2) # advance rPC, load rINST
GET_INST_OPCODE(t0) # extract opcode from rINST
andi v0, a2, 0x20 # shift & 0x20
srl v1, a1, a2 # rhi<- ahi >> (shift&31)
bnez v0, .L${opcode}_finish
srl v0, a0, a2 # rlo<- alo >> (shift&31)
not a0, a2 # alo<- 31-n (shift is 5b)
sll a1, 1
sll a1, a0 # ahi<- ahi << (32-(shift&31))
or v0, a1 # rlo<- rlo | ahi
SET_VREG64_GOTO(v0, v1, rOBJ, t0) # vAA/vAA+1 <- v0/v1
%def op_ushr_long_helper_code():
.Lop_ushr_long_finish:
SET_VREG64_GOTO(v1, zero, rOBJ, t0) # vAA/vAA+1 <- rlo/rhi
%def op_ushr_long_2addr():
/*
* Long integer shift, 2addr version. vA is 64-bit value/result, vB is
* 32-bit shift distance.
*/
/* ushr-long/2addr vA, vB */
GET_OPA4(t3) # t3 <- A+
GET_OPB(a3) # a3 <- B
GET_VREG(a2, a3) # a2 <- vB
EAS2(t0, rFP, t3) # t0 <- &fp[A]
LOAD64(a0, a1, t0) # a0/a1 <- vA/vA+1
FETCH_ADVANCE_INST(1) # advance rPC, load rINST
GET_INST_OPCODE(t0) # extract opcode from rINST
andi v0, a2, 0x20 # shift & 0x20
srl v1, a1, a2 # rhi<- ahi >> (shift&31)
bnez v0, .L${opcode}_finish
srl v0, a0, a2 # rlo<- alo >> (shift&31)
not a0, a2 # alo<- 31-n (shift is 5b)
sll a1, 1
sll a1, a0 # ahi<- ahi << (32-(shift&31))
or v0, a1 # rlo<- rlo | ahi
SET_VREG64_GOTO(v0, v1, t3, t0) # vA/vA+1 <- v0/v1
%def op_ushr_long_2addr_helper_code():
.Lop_ushr_long_2addr_finish:
SET_VREG64_GOTO(v1, zero, t3, t0) # vA/vA+1 <- rlo/rhi
%def op_xor_int():
% binop(instr="xor a0, a0, a1")
%def op_xor_int_2addr():
% binop2addr(instr="xor a0, a0, a1")
%def op_xor_int_lit16():
% binopLit16(instr="xor a0, a0, a1")
%def op_xor_int_lit8():
% binopLit8(instr="xor a0, a0, a1")
%def op_xor_long():
% binopWide(preinstr="xor a0, a0, a2", instr="xor a1, a1, a3")
%def op_xor_long_2addr():
% binopWide2addr(preinstr="xor a0, a0, a2", instr="xor a1, a1, a3")