You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

976 lines
40 KiB

%def binop(preinstr="", result="r0", chkzero="0", instr=""):
/*
* Generic 32-bit binary operation. Provide an "instr" line that
* specifies an instruction that performs "result = r0 op r1".
* This could be an ARM instruction or a function call. (If the result
* comes back in a register other than r0, you can override "result".)
*
* If "chkzero" is set to 1, we perform a divide-by-zero check on
* vCC (r1). Useful for integer division and modulus. Note that we
* *don't* check for (INT_MIN / -1) here, because the ARM math lib
* 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, add-float, sub-float,
* mul-float, div-float, rem-float
*/
/* binop vAA, vBB, vCC */
FETCH r0, 1 @ r0<- CCBB
mov r9, rINST, lsr #8 @ r9<- AA
mov r3, r0, lsr #8 @ r3<- CC
and r2, r0, #255 @ r2<- BB
GET_VREG r1, r3 @ r1<- vCC
GET_VREG r0, r2 @ r0<- vBB
.if $chkzero
cmp r1, #0 @ is second operand zero?
beq common_errDivideByZero
.endif
FETCH_ADVANCE_INST 2 @ advance rPC, load rINST
$preinstr @ optional op; may set condition codes
$instr @ $result<- op, r0-r3 changed
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG $result, r9 @ vAA<- $result
GOTO_OPCODE ip @ jump to next instruction
/* 11-14 instructions */
%def binop2addr(preinstr="", result="r0", chkzero="0", instr=""):
/*
* Generic 32-bit "/2addr" binary operation. Provide an "instr" line
* that specifies an instruction that performs "result = r0 op r1".
* This could be an ARM instruction or a function call. (If the result
* comes back in a register other than r0, you can override "result".)
*
* If "chkzero" is set to 1, we perform a divide-by-zero check on
* vCC (r1). 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, add-float/2addr,
* sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
*/
/* binop/2addr vA, vB */
mov r3, rINST, lsr #12 @ r3<- B
ubfx r9, rINST, #8, #4 @ r9<- A
GET_VREG r1, r3 @ r1<- vB
GET_VREG r0, r9 @ r0<- vA
.if $chkzero
cmp r1, #0 @ is second operand zero?
beq common_errDivideByZero
.endif
FETCH_ADVANCE_INST 1 @ advance rPC, load rINST
$preinstr @ optional op; may set condition codes
$instr @ $result<- op, r0-r3 changed
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG $result, r9 @ vAA<- $result
GOTO_OPCODE ip @ jump to next instruction
/* 10-13 instructions */
%def binopLit16(result="r0", chkzero="0", instr=""):
/*
* Generic 32-bit "lit16" binary operation. Provide an "instr" line
* that specifies an instruction that performs "result = r0 op r1".
* This could be an ARM instruction or a function call. (If the result
* comes back in a register other than r0, you can override "result".)
*
* If "chkzero" is set to 1, we perform a divide-by-zero check on
* vCC (r1). 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 r1, 1 @ r1<- ssssCCCC (sign-extended)
mov r2, rINST, lsr #12 @ r2<- B
ubfx r9, rINST, #8, #4 @ r9<- A
GET_VREG r0, r2 @ r0<- vB
.if $chkzero
cmp r1, #0 @ is second operand zero?
beq common_errDivideByZero
.endif
FETCH_ADVANCE_INST 2 @ advance rPC, load rINST
$instr @ $result<- op, r0-r3 changed
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG $result, r9 @ vAA<- $result
GOTO_OPCODE ip @ jump to next instruction
/* 10-13 instructions */
%def binopLit8(extract="asr r1, r3, #8", result="r0", chkzero="0", instr=""):
/*
* Generic 32-bit "lit8" binary operation. Provide an "instr" line
* that specifies an instruction that performs "result = r0 op r1".
* This could be an ARM instruction or a function call. (If the result
* comes back in a register other than r0, you can override "result".)
*
* You can override "extract" if the extraction of the literal value
* from r3 to r1 is not the default "asr r1, r3, #8". The extraction
* can be omitted completely if the shift is embedded in "instr".
*
* If "chkzero" is set to 1, we perform a divide-by-zero check on
* vCC (r1). 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 r3, 1 @ r3<- ssssCCBB (sign-extended for CC)
mov r9, rINST, lsr #8 @ r9<- AA
and r2, r3, #255 @ r2<- BB
GET_VREG r0, r2 @ r0<- vBB
$extract @ optional; typically r1<- ssssssCC (sign extended)
.if $chkzero
@cmp r1, #0 @ is second operand zero?
beq common_errDivideByZero
.endif
FETCH_ADVANCE_INST 2 @ advance rPC, load rINST
$instr @ $result<- op, r0-r3 changed
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG $result, r9 @ vAA<- $result
GOTO_OPCODE ip @ jump to next instruction
/* 10-12 instructions */
%def binopWide(preinstr="", result0="r0", result1="r1", chkzero="0", instr=""):
/*
* Generic 64-bit binary operation. Provide an "instr" line that
* specifies an instruction that performs "result = r0-r1 op r2-r3".
* This could be an ARM instruction or a function call. (If the result
* comes back in a register other than r0, you can override "result".)
*
* If "chkzero" is set to 1, we perform a divide-by-zero check on
* vCC (r1). Useful for integer division and modulus.
*
* for: add-long, sub-long, div-long, rem-long, and-long, or-long,
* xor-long, add-double, sub-double, mul-double, div-double,
* rem-double
*
* IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
*/
/* binop vAA, vBB, vCC */
FETCH r0, 1 @ r0<- CCBB
mov rINST, rINST, lsr #8 @ rINST<- AA
and r2, r0, #255 @ r2<- BB
mov r3, r0, lsr #8 @ r3<- CC
VREG_INDEX_TO_ADDR r9, rINST @ r9<- &fp[AA]
VREG_INDEX_TO_ADDR r2, r2 @ r2<- &fp[BB]
VREG_INDEX_TO_ADDR r3, r3 @ r3<- &fp[CC]
GET_VREG_WIDE_BY_ADDR r0, r1, r2 @ r0/r1<- vBB/vBB+1
GET_VREG_WIDE_BY_ADDR r2, r3, r3 @ r2/r3<- vCC/vCC+1
.if $chkzero
orrs ip, r2, r3 @ second arg (r2-r3) is zero?
beq common_errDivideByZero
.endif
CLEAR_SHADOW_PAIR rINST, lr, ip @ Zero out the shadow regs
FETCH_ADVANCE_INST 2 @ advance rPC, load rINST
$preinstr @ optional op; may set condition codes
$instr @ result<- op, r0-r3 changed
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG_WIDE_BY_ADDR $result0,$result1,r9 @ vAA/vAA+1<, $result0/$result1
GOTO_OPCODE ip @ jump to next instruction
/* 14-17 instructions */
%def binopWide2addr(preinstr="", result0="r0", result1="r1", chkzero="0", instr=""):
/*
* Generic 64-bit "/2addr" binary operation. Provide an "instr" line
* that specifies an instruction that performs "result = r0-r1 op r2-r3".
* This could be an ARM instruction or a function call. (If the result
* comes back in a register other than r0, you can override "result".)
*
* If "chkzero" is set to 1, we perform a divide-by-zero check on
* vCC (r1). 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, add-double/2addr,
* sub-double/2addr, mul-double/2addr, div-double/2addr,
* rem-double/2addr
*/
/* binop/2addr vA, vB */
mov r1, rINST, lsr #12 @ r1<- B
ubfx rINST, rINST, #8, #4 @ rINST<- A
VREG_INDEX_TO_ADDR r1, r1 @ r1<- &fp[B]
VREG_INDEX_TO_ADDR r9, rINST @ r9<- &fp[A]
GET_VREG_WIDE_BY_ADDR r2, r3, r1 @ r2/r3<- vBB/vBB+1
GET_VREG_WIDE_BY_ADDR r0, r1, r9 @ r0/r1<- vAA/vAA+1
.if $chkzero
orrs ip, r2, r3 @ second arg (r2-r3) is zero?
beq common_errDivideByZero
.endif
CLEAR_SHADOW_PAIR rINST, ip, lr @ Zero shadow regs
FETCH_ADVANCE_INST 1 @ advance rPC, load rINST
$preinstr @ optional op; may set condition codes
$instr @ result<- op, r0-r3 changed
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG_WIDE_BY_ADDR $result0,$result1,r9 @ vAA/vAA+1<- $result0/$result1
GOTO_OPCODE ip @ jump to next instruction
/* 12-15 instructions */
%def unop(preinstr="", instr=""):
/*
* Generic 32-bit unary operation. Provide an "instr" line that
* specifies an instruction that performs "result = op r0".
* This could be an ARM instruction or a function call.
*
* for: neg-int, not-int, neg-float, int-to-float, float-to-int,
* int-to-byte, int-to-char, int-to-short
*/
/* unop vA, vB */
mov r3, rINST, lsr #12 @ r3<- B
ubfx r9, rINST, #8, #4 @ r9<- A
GET_VREG r0, r3 @ r0<- vB
$preinstr @ optional op; may set condition codes
FETCH_ADVANCE_INST 1 @ advance rPC, load rINST
$instr @ r0<- op, r0-r3 changed
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG r0, r9 @ vAA<- r0
GOTO_OPCODE ip @ jump to next instruction
/* 8-9 instructions */
%def unopNarrower(preinstr="", instr=""):
/*
* Generic 64bit-to-32bit unary operation. Provide an "instr" line
* that specifies an instruction that performs "result = op r0/r1", where
* "result" is a 32-bit quantity in r0.
*
* For: long-to-float
*
* (This would work for long-to-int, but that instruction is actually
* an exact match for op_move.)
*/
/* unop vA, vB */
mov r3, rINST, lsr #12 @ r3<- B
ubfx r9, rINST, #8, #4 @ r9<- A
VREG_INDEX_TO_ADDR r3, r3 @ r3<- &fp[B]
GET_VREG_WIDE_BY_ADDR r0, r1, r3 @ r0/r1<- vB/vB+1
FETCH_ADVANCE_INST 1 @ advance rPC, load rINST
$preinstr @ optional op; may set condition codes
$instr @ r0<- op, r0-r3 changed
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG r0, r9 @ vA<- r0
GOTO_OPCODE ip @ jump to next instruction
/* 9-10 instructions */
%def unopWide(preinstr="", instr=""):
/*
* Generic 64-bit unary operation. Provide an "instr" line that
* specifies an instruction that performs "result = op r0/r1".
* This could be an ARM instruction or a function call.
*
* For: neg-long, not-long, neg-double, long-to-double, double-to-long
*/
/* unop vA, vB */
mov r3, rINST, lsr #12 @ r3<- B
ubfx rINST, rINST, #8, #4 @ rINST<- A
VREG_INDEX_TO_ADDR r3, r3 @ r3<- &fp[B]
VREG_INDEX_TO_ADDR r9, rINST @ r9<- &fp[A]
GET_VREG_WIDE_BY_ADDR r0, r1, r3 @ r0/r1<- vAA
CLEAR_SHADOW_PAIR rINST, ip, lr @ Zero shadow regs
FETCH_ADVANCE_INST 1 @ advance rPC, load rINST
$preinstr @ optional op; may set condition codes
$instr @ r0/r1<- op, r2-r3 changed
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG_WIDE_BY_ADDR r0, r1, r9 @ vAA<- r0/r1
GOTO_OPCODE ip @ jump to next instruction
/* 10-11 instructions */
%def unopWider(preinstr="", instr=""):
/*
* Generic 32bit-to-64bit unary operation. Provide an "instr" line
* that specifies an instruction that performs "result = op r0", where
* "result" is a 64-bit quantity in r0/r1.
*
* For: int-to-long, int-to-double, float-to-long, float-to-double
*/
/* unop vA, vB */
mov r3, rINST, lsr #12 @ r3<- B
ubfx rINST, rINST, #8, #4 @ rINST<- A
GET_VREG r0, r3 @ r0<- vB
VREG_INDEX_TO_ADDR r9, rINST @ r9<- &fp[A]
$preinstr @ optional op; may set condition codes
CLEAR_SHADOW_PAIR rINST, ip, lr @ Zero shadow regs
FETCH_ADVANCE_INST 1 @ advance rPC, load rINST
$instr @ r0<- op, r0-r3 changed
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG_WIDE_BY_ADDR r0, r1, r9 @ vA/vA+1<- r0/r1
GOTO_OPCODE ip @ jump to next instruction
/* 9-10 instructions */
%def op_add_int():
% binop(instr="add r0, r0, r1")
%def op_add_int_2addr():
% binop2addr(instr="add r0, r0, r1")
%def op_add_int_lit16():
% binopLit16(instr="add r0, r0, r1")
%def op_add_int_lit8():
% binopLit8(extract="", instr="add r0, r0, r3, asr #8")
%def op_add_long():
% binopWide(preinstr="adds r0, r0, r2", instr="adc r1, r1, r3")
%def op_add_long_2addr():
% binopWide2addr(preinstr="adds r0, r0, r2", instr="adc r1, r1, r3")
%def op_and_int():
% binop(instr="and r0, r0, r1")
%def op_and_int_2addr():
% binop2addr(instr="and r0, r0, r1")
%def op_and_int_lit16():
% binopLit16(instr="and r0, r0, r1")
%def op_and_int_lit8():
% binopLit8(extract="", instr="and r0, r0, r3, asr #8")
%def op_and_long():
% binopWide(preinstr="and r0, r0, r2", instr="and r1, r1, r3")
%def op_and_long_2addr():
% binopWide2addr(preinstr="and r0, r0, r2", instr="and r1, r1, r3")
%def op_cmp_long():
/*
* Compare two 64-bit values. Puts 0, 1, or -1 into the destination
* register based on the results of the comparison.
*/
/* cmp-long vAA, vBB, vCC */
FETCH r0, 1 @ r0<- CCBB
mov r9, rINST, lsr #8 @ r9<- AA
and r2, r0, #255 @ r2<- BB
mov r3, r0, lsr #8 @ r3<- CC
VREG_INDEX_TO_ADDR r2, r2 @ r2<- &fp[BB]
VREG_INDEX_TO_ADDR r3, r3 @ r3<- &fp[CC]
GET_VREG_WIDE_BY_ADDR r0, r1, r2 @ r0/r1<- vBB/vBB+1
GET_VREG_WIDE_BY_ADDR r2, r3, r3 @ r2/r3<- vCC/vCC+1
cmp r0, r2
sbcs ip, r1, r3 @ Sets correct CCs for checking LT (but not EQ/NE)
mov ip, #0
mvnlt ip, #0 @ -1
cmpeq r0, r2 @ For correct EQ/NE, we may need to repeat the first CMP
orrne ip, #1
FETCH_ADVANCE_INST 2 @ advance rPC, load rINST
SET_VREG ip, r9 @ vAA<- ip
GET_INST_OPCODE ip @ extract opcode from rINST
GOTO_OPCODE ip @ jump to next instruction
%def op_div_int():
/*
* Specialized 32-bit binary operation
*
* Performs "r0 = r0 div r1". The selection between sdiv or the gcc helper
* depends on the compile time value of __ARM_ARCH_EXT_IDIV__ (defined for
* ARMv7 CPUs that have hardware division support).
*
* div-int
*
*/
FETCH r0, 1 @ r0<- CCBB
mov r9, rINST, lsr #8 @ r9<- AA
mov r3, r0, lsr #8 @ r3<- CC
and r2, r0, #255 @ r2<- BB
GET_VREG r1, r3 @ r1<- vCC
GET_VREG r0, r2 @ r0<- vBB
cmp r1, #0 @ is second operand zero?
beq common_errDivideByZero
FETCH_ADVANCE_INST 2 @ advance rPC, load rINST
#ifdef __ARM_ARCH_EXT_IDIV__
sdiv r0, r0, r1 @ r0<- op
#else
bl __aeabi_idiv @ r0<- op, r0-r3 changed
#endif
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG r0, r9 @ vAA<- r0
GOTO_OPCODE ip @ jump to next instruction
/* 11-14 instructions */
%def op_div_int_2addr():
/*
* Specialized 32-bit binary operation
*
* Performs "r0 = r0 div r1". The selection between sdiv or the gcc helper
* depends on the compile time value of __ARM_ARCH_EXT_IDIV__ (defined for
* ARMv7 CPUs that have hardware division support).
*
* div-int/2addr
*
*/
mov r3, rINST, lsr #12 @ r3<- B
ubfx r9, rINST, #8, #4 @ r9<- A
GET_VREG r1, r3 @ r1<- vB
GET_VREG r0, r9 @ r0<- vA
cmp r1, #0 @ is second operand zero?
beq common_errDivideByZero
FETCH_ADVANCE_INST 1 @ advance rPC, load rINST
#ifdef __ARM_ARCH_EXT_IDIV__
sdiv r0, r0, r1 @ r0<- op
#else
bl __aeabi_idiv @ r0<- op, r0-r3 changed
#endif
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG r0, r9 @ vAA<- r0
GOTO_OPCODE ip @ jump to next instruction
/* 10-13 instructions */
%def op_div_int_lit16():
/*
* Specialized 32-bit binary operation
*
* Performs "r0 = r0 div r1". The selection between sdiv or the gcc helper
* depends on the compile time value of __ARM_ARCH_EXT_IDIV__ (defined for
* ARMv7 CPUs that have hardware division support).
*
* div-int/lit16
*
*/
FETCH_S r1, 1 @ r1<- ssssCCCC (sign-extended)
mov r2, rINST, lsr #12 @ r2<- B
ubfx r9, rINST, #8, #4 @ r9<- A
GET_VREG r0, r2 @ r0<- vB
cmp r1, #0 @ is second operand zero?
beq common_errDivideByZero
FETCH_ADVANCE_INST 2 @ advance rPC, load rINST
#ifdef __ARM_ARCH_EXT_IDIV__
sdiv r0, r0, r1 @ r0<- op
#else
bl __aeabi_idiv @ r0<- op, r0-r3 changed
#endif
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG r0, r9 @ vAA<- r0
GOTO_OPCODE ip @ jump to next instruction
/* 10-13 instructions */
%def op_div_int_lit8():
/*
* Specialized 32-bit binary operation
*
* Performs "r0 = r0 div r1". The selection between sdiv or the gcc helper
* depends on the compile time value of __ARM_ARCH_EXT_IDIV__ (defined for
* ARMv7 CPUs that have hardware division support).
*
* div-int/lit8
*
*/
FETCH_S r3, 1 @ r3<- ssssCCBB (sign-extended for CC
mov r9, rINST, lsr #8 @ r9<- AA
and r2, r3, #255 @ r2<- BB
GET_VREG r0, r2 @ r0<- vBB
movs r1, r3, asr #8 @ r1<- ssssssCC (sign extended)
@cmp r1, #0 @ is second operand zero?
beq common_errDivideByZero
FETCH_ADVANCE_INST 2 @ advance rPC, load rINST
#ifdef __ARM_ARCH_EXT_IDIV__
sdiv r0, r0, r1 @ r0<- op
#else
bl __aeabi_idiv @ r0<- op, r0-r3 changed
#endif
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG r0, r9 @ vAA<- r0
GOTO_OPCODE ip @ jump to next instruction
/* 10-12 instructions */
%def op_div_long():
% binopWide(instr="bl __aeabi_ldivmod", chkzero="1")
%def op_div_long_2addr():
% binopWide2addr(instr="bl __aeabi_ldivmod", chkzero="1")
%def op_int_to_byte():
% unop(instr="sxtb r0, r0")
%def op_int_to_char():
% unop(instr="uxth r0, r0")
%def op_int_to_long():
% unopWider(instr="mov r1, r0, asr #31")
%def op_int_to_short():
% unop(instr="sxth r0, r0")
%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():
/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
% binop(instr="mul r0, r1, r0")
%def op_mul_int_2addr():
/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
% binop2addr(instr="mul r0, r1, r0")
%def op_mul_int_lit16():
/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
% binopLit16(instr="mul r0, r1, r0")
%def op_mul_int_lit8():
/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
% binopLit8(instr="mul r0, r1, r0")
%def op_mul_long():
/*
* Signed 64-bit integer multiply.
*
* Consider WXxYZ (r1r0 x r3r2) with a long multiply:
* WX
* x YZ
* --------
* ZW ZX
* YW YX
*
* The low word of the result holds ZX, the high word holds
* (ZW+YX) + (the high overflow from ZX). YW doesn't matter because
* it doesn't fit in the low 64 bits.
*
* Unlike most ARM math operations, multiply instructions have
* restrictions on using the same register more than once (Rd and Rm
* cannot be the same).
*/
/* mul-long vAA, vBB, vCC */
FETCH r0, 1 @ r0<- CCBB
and r2, r0, #255 @ r2<- BB
mov r3, r0, lsr #8 @ r3<- CC
VREG_INDEX_TO_ADDR r2, r2 @ r2<- &fp[BB]
VREG_INDEX_TO_ADDR r3, r3 @ r3<- &fp[CC]
GET_VREG_WIDE_BY_ADDR r0, r1, r2 @ r0/r1<- vBB/vBB+1
GET_VREG_WIDE_BY_ADDR r2, r3, r3 @ r2/r3<- vCC/vCC+1
mul ip, r2, r1 @ ip<- ZxW
umull r1, lr, r2, r0 @ r1/lr <- ZxX
mla r2, r0, r3, ip @ r2<- YxX + (ZxW)
mov r0, rINST, lsr #8 @ r0<- AA
add r2, r2, lr @ r2<- lr + low(ZxW + (YxX))
CLEAR_SHADOW_PAIR r0, lr, ip @ Zero out the shadow regs
VREG_INDEX_TO_ADDR r0, r0 @ r0<- &fp[AA]
FETCH_ADVANCE_INST 2 @ advance rPC, load rINST
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG_WIDE_BY_ADDR r1, r2 , r0 @ vAA/vAA+1<- r1/r2
GOTO_OPCODE ip @ jump to next instruction
%def op_mul_long_2addr():
/*
* Signed 64-bit integer multiply, "/2addr" version.
*
* See op_mul_long for an explanation.
*
* We get a little tight on registers, so to avoid looking up &fp[A]
* again we stuff it into rINST.
*/
/* mul-long/2addr vA, vB */
mov r1, rINST, lsr #12 @ r1<- B
ubfx r9, rINST, #8, #4 @ r9<- A
VREG_INDEX_TO_ADDR r1, r1 @ r1<- &fp[B]
VREG_INDEX_TO_ADDR rINST, r9 @ rINST<- &fp[A]
GET_VREG_WIDE_BY_ADDR r2, r3, r1 @ r2/r3<- vBB/vBB+1
GET_VREG_WIDE_BY_ADDR r0, r1, rINST @ r0/r1<- vAA/vAA+1
mul ip, r2, r1 @ ip<- ZxW
umull r1, lr, r2, r0 @ r1/lr <- ZxX
mla r2, r0, r3, ip @ r2<- YxX + (ZxW)
mov r0, rINST @ r0<- &fp[A] (free up rINST)
FETCH_ADVANCE_INST 1 @ advance rPC, load rINST
add r2, r2, lr @ r2<- r2 + low(ZxW + (YxX))
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG_WIDE_BY_ADDR r1, r2, r0 @ vAA/vAA+1<- r1/r2
GOTO_OPCODE ip @ jump to next instruction
%def op_neg_int():
% unop(instr="rsb r0, r0, #0")
%def op_neg_long():
% unopWide(preinstr="rsbs r0, r0, #0", instr="rsc r1, r1, #0")
%def op_not_int():
% unop(instr="mvn r0, r0")
%def op_not_long():
% unopWide(preinstr="mvn r0, r0", instr="mvn r1, r1")
%def op_or_int():
% binop(instr="orr r0, r0, r1")
%def op_or_int_2addr():
% binop2addr(instr="orr r0, r0, r1")
%def op_or_int_lit16():
% binopLit16(instr="orr r0, r0, r1")
%def op_or_int_lit8():
% binopLit8(extract="", instr="orr r0, r0, r3, asr #8")
%def op_or_long():
% binopWide(preinstr="orr r0, r0, r2", instr="orr r1, r1, r3")
%def op_or_long_2addr():
% binopWide2addr(preinstr="orr r0, r0, r2", instr="orr r1, r1, r3")
%def op_rem_int():
/*
* Specialized 32-bit binary operation
*
* Performs "r1 = r0 rem r1". The selection between sdiv block or the gcc helper
* depends on the compile time value of __ARM_ARCH_EXT_IDIV__ (defined for
* ARMv7 CPUs that have hardware division support).
*
* NOTE: idivmod returns quotient in r0 and remainder in r1
*
* rem-int
*
*/
FETCH r0, 1 @ r0<- CCBB
mov r9, rINST, lsr #8 @ r9<- AA
mov r3, r0, lsr #8 @ r3<- CC
and r2, r0, #255 @ r2<- BB
GET_VREG r1, r3 @ r1<- vCC
GET_VREG r0, r2 @ r0<- vBB
cmp r1, #0 @ is second operand zero?
beq common_errDivideByZero
FETCH_ADVANCE_INST 2 @ advance rPC, load rINST
#ifdef __ARM_ARCH_EXT_IDIV__
sdiv r2, r0, r1
mls r1, r1, r2, r0 @ r1<- op, r0-r2 changed
#else
bl __aeabi_idivmod @ r1<- op, r0-r3 changed
#endif
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG r1, r9 @ vAA<- r1
GOTO_OPCODE ip @ jump to next instruction
/* 11-14 instructions */
%def op_rem_int_2addr():
/*
* Specialized 32-bit binary operation
*
* Performs "r1 = r0 rem r1". The selection between sdiv block or the gcc helper
* depends on the compile time value of __ARM_ARCH_EXT_IDIV__ (defined for
* ARMv7 CPUs that have hardware division support).
*
* NOTE: idivmod returns quotient in r0 and remainder in r1
*
* rem-int/2addr
*
*/
mov r3, rINST, lsr #12 @ r3<- B
ubfx r9, rINST, #8, #4 @ r9<- A
GET_VREG r1, r3 @ r1<- vB
GET_VREG r0, r9 @ r0<- vA
cmp r1, #0 @ is second operand zero?
beq common_errDivideByZero
FETCH_ADVANCE_INST 1 @ advance rPC, load rINST
#ifdef __ARM_ARCH_EXT_IDIV__
sdiv r2, r0, r1
mls r1, r1, r2, r0 @ r1<- op
#else
bl __aeabi_idivmod @ r1<- op, r0-r3 changed
#endif
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG r1, r9 @ vAA<- r1
GOTO_OPCODE ip @ jump to next instruction
/* 10-13 instructions */
%def op_rem_int_lit16():
/*
* Specialized 32-bit binary operation
*
* Performs "r1 = r0 rem r1". The selection between sdiv block or the gcc helper
* depends on the compile time value of __ARM_ARCH_EXT_IDIV__ (defined for
* ARMv7 CPUs that have hardware division support).
*
* NOTE: idivmod returns quotient in r0 and remainder in r1
*
* rem-int/lit16
*
*/
FETCH_S r1, 1 @ r1<- ssssCCCC (sign-extended)
mov r2, rINST, lsr #12 @ r2<- B
ubfx r9, rINST, #8, #4 @ r9<- A
GET_VREG r0, r2 @ r0<- vB
cmp r1, #0 @ is second operand zero?
beq common_errDivideByZero
FETCH_ADVANCE_INST 2 @ advance rPC, load rINST
#ifdef __ARM_ARCH_EXT_IDIV__
sdiv r2, r0, r1
mls r1, r1, r2, r0 @ r1<- op
#else
bl __aeabi_idivmod @ r1<- op, r0-r3 changed
#endif
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG r1, r9 @ vAA<- r1
GOTO_OPCODE ip @ jump to next instruction
/* 10-13 instructions */
%def op_rem_int_lit8():
/*
* Specialized 32-bit binary operation
*
* Performs "r1 = r0 rem r1". The selection between sdiv block or the gcc helper
* depends on the compile time value of __ARM_ARCH_EXT_IDIV__ (defined for
* ARMv7 CPUs that have hardware division support).
*
* NOTE: idivmod returns quotient in r0 and remainder in r1
*
* rem-int/lit8
*
*/
FETCH_S r3, 1 @ r3<- ssssCCBB (sign-extended for CC)
mov r9, rINST, lsr #8 @ r9<- AA
and r2, r3, #255 @ r2<- BB
GET_VREG r0, r2 @ r0<- vBB
movs r1, r3, asr #8 @ r1<- ssssssCC (sign extended)
@cmp r1, #0 @ is second operand zero?
beq common_errDivideByZero
FETCH_ADVANCE_INST 2 @ advance rPC, load rINST
#ifdef __ARM_ARCH_EXT_IDIV__
sdiv r2, r0, r1
mls r1, r1, r2, r0 @ r1<- op
#else
bl __aeabi_idivmod @ r1<- op, r0-r3 changed
#endif
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG r1, r9 @ vAA<- r1
GOTO_OPCODE ip @ jump to next instruction
/* 10-12 instructions */
%def op_rem_long():
/* ldivmod returns quotient in r0/r1 and remainder in r2/r3 */
% binopWide(instr="bl __aeabi_ldivmod", result0="r2", result1="r3", chkzero="1")
%def op_rem_long_2addr():
/* ldivmod returns quotient in r0/r1 and remainder in r2/r3 */
% binopWide2addr(instr="bl __aeabi_ldivmod", result0="r2", result1="r3", chkzero="1")
%def op_rsub_int():
/* this op is "rsub-int", but can be thought of as "rsub-int/lit16" */
% binopLit16(instr="rsb r0, r0, r1")
%def op_rsub_int_lit8():
% binopLit8(extract="", instr="rsb r0, r0, r3, asr #8")
%def op_shl_int():
% binop(preinstr="and r1, r1, #31", instr="mov r0, r0, asl r1")
%def op_shl_int_2addr():
% binop2addr(preinstr="and r1, r1, #31", instr="mov r0, r0, asl r1")
%def op_shl_int_lit8():
% binopLit8(extract="ubfx r1, r3, #8, #5", instr="mov r0, r0, asl r1")
%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 r0, 1 @ r0<- CCBB
mov r9, rINST, lsr #8 @ r9<- AA
and r3, r0, #255 @ r3<- BB
mov r0, r0, lsr #8 @ r0<- CC
VREG_INDEX_TO_ADDR r3, r3 @ r3<- &fp[BB]
GET_VREG r2, r0 @ r2<- vCC
GET_VREG_WIDE_BY_ADDR r0, r1, r3 @ r0/r1<- vBB/vBB+1
CLEAR_SHADOW_PAIR r9, lr, ip @ Zero out the shadow regs
and r2, r2, #63 @ r2<- r2 & 0x3f
VREG_INDEX_TO_ADDR r9, r9 @ r9<- &fp[AA]
mov r1, r1, asl r2 @ r1<- r1 << r2
rsb r3, r2, #32 @ r3<- 32 - r2
orr r1, r1, r0, lsr r3 @ r1<- r1 | (r0 << (32-r2))
subs ip, r2, #32 @ ip<- r2 - 32
movpl r1, r0, asl ip @ if r2 >= 32, r1<- r0 << (r2-32)
FETCH_ADVANCE_INST 2 @ advance rPC, load rINST
mov r0, r0, asl r2 @ r0<- r0 << r2
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG_WIDE_BY_ADDR r0, r1, r9 @ vAA/vAA+1<- r0/r1
GOTO_OPCODE ip @ jump to next instruction
%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 */
mov r3, rINST, lsr #12 @ r3<- B
ubfx r9, rINST, #8, #4 @ r9<- A
GET_VREG r2, r3 @ r2<- vB
CLEAR_SHADOW_PAIR r9, lr, ip @ Zero out the shadow regs
VREG_INDEX_TO_ADDR r9, r9 @ r9<- &fp[A]
and r2, r2, #63 @ r2<- r2 & 0x3f
GET_VREG_WIDE_BY_ADDR r0, r1, r9 @ r0/r1<- vAA/vAA+1
mov r1, r1, asl r2 @ r1<- r1 << r2
rsb r3, r2, #32 @ r3<- 32 - r2
orr r1, r1, r0, lsr r3 @ r1<- r1 | (r0 << (32-r2))
subs ip, r2, #32 @ ip<- r2 - 32
FETCH_ADVANCE_INST 1 @ advance rPC, load rINST
movpl r1, r0, asl ip @ if r2 >= 32, r1<- r0 << (r2-32)
mov r0, r0, asl r2 @ r0<- r0 << r2
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG_WIDE_BY_ADDR r0, r1, r9 @ vAA/vAA+1<- r0/r1
GOTO_OPCODE ip @ jump to next instruction
%def op_shr_int():
% binop(preinstr="and r1, r1, #31", instr="mov r0, r0, asr r1")
%def op_shr_int_2addr():
% binop2addr(preinstr="and r1, r1, #31", instr="mov r0, r0, asr r1")
%def op_shr_int_lit8():
% binopLit8(extract="ubfx r1, r3, #8, #5", instr="mov r0, r0, asr r1")
%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 r0, 1 @ r0<- CCBB
mov r9, rINST, lsr #8 @ r9<- AA
and r3, r0, #255 @ r3<- BB
mov r0, r0, lsr #8 @ r0<- CC
VREG_INDEX_TO_ADDR r3, r3 @ r3<- &fp[BB]
GET_VREG r2, r0 @ r2<- vCC
GET_VREG_WIDE_BY_ADDR r0, r1, r3 @ r0/r1<- vBB/vBB+1
CLEAR_SHADOW_PAIR r9, lr, ip @ Zero out the shadow regs
and r2, r2, #63 @ r0<- r0 & 0x3f
VREG_INDEX_TO_ADDR r9, r9 @ r9<- &fp[AA]
mov r0, r0, lsr r2 @ r0<- r2 >> r2
rsb r3, r2, #32 @ r3<- 32 - r2
orr r0, r0, r1, asl r3 @ r0<- r0 | (r1 << (32-r2))
subs ip, r2, #32 @ ip<- r2 - 32
movpl r0, r1, asr ip @ if r2 >= 32, r0<-r1 >> (r2-32)
FETCH_ADVANCE_INST 2 @ advance rPC, load rINST
mov r1, r1, asr r2 @ r1<- r1 >> r2
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG_WIDE_BY_ADDR r0, r1, r9 @ vAA/vAA+1<- r0/r1
GOTO_OPCODE ip @ jump to next instruction
%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 */
mov r3, rINST, lsr #12 @ r3<- B
ubfx r9, rINST, #8, #4 @ r9<- A
GET_VREG r2, r3 @ r2<- vB
CLEAR_SHADOW_PAIR r9, lr, ip @ Zero out the shadow regs
VREG_INDEX_TO_ADDR r9, r9 @ r9<- &fp[A]
and r2, r2, #63 @ r2<- r2 & 0x3f
GET_VREG_WIDE_BY_ADDR r0, r1, r9 @ r0/r1<- vAA/vAA+1
mov r0, r0, lsr r2 @ r0<- r2 >> r2
rsb r3, r2, #32 @ r3<- 32 - r2
orr r0, r0, r1, asl r3 @ r0<- r0 | (r1 << (32-r2))
subs ip, r2, #32 @ ip<- r2 - 32
FETCH_ADVANCE_INST 1 @ advance rPC, load rINST
movpl r0, r1, asr ip @ if r2 >= 32, r0<-r1 >> (r2-32)
mov r1, r1, asr r2 @ r1<- r1 >> r2
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG_WIDE_BY_ADDR r0, r1, r9 @ vAA/vAA+1<- r0/r1
GOTO_OPCODE ip @ jump to next instruction
%def op_sub_int():
% binop(instr="sub r0, r0, r1")
%def op_sub_int_2addr():
% binop2addr(instr="sub r0, r0, r1")
%def op_sub_long():
% binopWide(preinstr="subs r0, r0, r2", instr="sbc r1, r1, r3")
%def op_sub_long_2addr():
% binopWide2addr(preinstr="subs r0, r0, r2", instr="sbc r1, r1, r3")
%def op_ushr_int():
% binop(preinstr="and r1, r1, #31", instr="mov r0, r0, lsr r1")
%def op_ushr_int_2addr():
% binop2addr(preinstr="and r1, r1, #31", instr="mov r0, r0, lsr r1")
%def op_ushr_int_lit8():
% binopLit8(extract="ubfx r1, r3, #8, #5", instr="mov r0, r0, lsr r1")
%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 r0, 1 @ r0<- CCBB
mov r9, rINST, lsr #8 @ r9<- AA
and r3, r0, #255 @ r3<- BB
mov r0, r0, lsr #8 @ r0<- CC
VREG_INDEX_TO_ADDR r3, r3 @ r3<- &fp[BB]
GET_VREG r2, r0 @ r2<- vCC
GET_VREG_WIDE_BY_ADDR r0, r1, r3 @ r0/r1<- vBB/vBB+1
CLEAR_SHADOW_PAIR r9, lr, ip @ Zero out the shadow regs
and r2, r2, #63 @ r0<- r0 & 0x3f
VREG_INDEX_TO_ADDR r9, r9 @ r9<- &fp[AA]
mov r0, r0, lsr r2 @ r0<- r2 >> r2
rsb r3, r2, #32 @ r3<- 32 - r2
orr r0, r0, r1, asl r3 @ r0<- r0 | (r1 << (32-r2))
subs ip, r2, #32 @ ip<- r2 - 32
movpl r0, r1, lsr ip @ if r2 >= 32, r0<-r1 >>> (r2-32)
FETCH_ADVANCE_INST 2 @ advance rPC, load rINST
mov r1, r1, lsr r2 @ r1<- r1 >>> r2
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG_WIDE_BY_ADDR r0, r1, r9 @ vAA/vAA+1<- r0/r1
GOTO_OPCODE ip @ jump to next instruction
%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 */
mov r3, rINST, lsr #12 @ r3<- B
ubfx r9, rINST, #8, #4 @ r9<- A
GET_VREG r2, r3 @ r2<- vB
CLEAR_SHADOW_PAIR r9, lr, ip @ Zero out the shadow regs
VREG_INDEX_TO_ADDR r9, r9 @ r9<- &fp[A]
and r2, r2, #63 @ r2<- r2 & 0x3f
GET_VREG_WIDE_BY_ADDR r0, r1, r9 @ r0/r1<- vAA/vAA+1
mov r0, r0, lsr r2 @ r0<- r2 >> r2
rsb r3, r2, #32 @ r3<- 32 - r2
orr r0, r0, r1, asl r3 @ r0<- r0 | (r1 << (32-r2))
subs ip, r2, #32 @ ip<- r2 - 32
FETCH_ADVANCE_INST 1 @ advance rPC, load rINST
movpl r0, r1, lsr ip @ if r2 >= 32, r0<-r1 >>> (r2-32)
mov r1, r1, lsr r2 @ r1<- r1 >>> r2
GET_INST_OPCODE ip @ extract opcode from rINST
SET_VREG_WIDE_BY_ADDR r0, r1, r9 @ vAA/vAA+1<- r0/r1
GOTO_OPCODE ip @ jump to next instruction
%def op_xor_int():
% binop(instr="eor r0, r0, r1")
%def op_xor_int_2addr():
% binop2addr(instr="eor r0, r0, r1")
%def op_xor_int_lit16():
% binopLit16(instr="eor r0, r0, r1")
%def op_xor_int_lit8():
% binopLit8(extract="", instr="eor r0, r0, r3, asr #8")
%def op_xor_long():
% binopWide(preinstr="eor r0, r0, r2", instr="eor r1, r1, r3")
%def op_xor_long_2addr():
% binopWide2addr(preinstr="eor r0, r0, r2", instr="eor r1, r1, r3")