assembler_sparc.hpp revision 113
/*
* Copyright 1997-2007 Sun Microsystems, Inc. All Rights Reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* under the terms of the GNU General Public License version 2 only, as
* published by the Free Software Foundation.
*
* This code is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
* version 2 for more details (a copy is included in the LICENSE file that
* accompanied this code).
*
* You should have received a copy of the GNU General Public License version
* 2 along with this work; if not, write to the Free Software Foundation,
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
*
* Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
* CA 95054 USA or visit www.sun.com if you need additional information or
* have any questions.
*
*/
class BiasedLockingCounters;
#define ST_RESERVED_FOR_USER_0 0x10
/* Written: David Ungar 4/19/97 */
// Contains all the definitions needed for sparc assembly code generation.
// Register aliases for parts of the system:
// 64 bit values can be kept in g1-g5, o1-o5 and o7 and all 64 bits are safe
// across context switches in V8+ ABI. Of course, there are no 64 bit regs
// in V8 ABI. All 64 bits are preserved in V9 ABI for all registers.
// g2-g4 are scratch registers called "application globals". Their
// meaning is reserved to the "compilation system"--which means us!
// They are are not supposed to be touched by ordinary C code, although
// highly-optimized C code might steal them for temps. They are safe
// across thread switches, and the ABI requires that they be safe
// across function calls.
//
// g1 and g3 are touched by more modules. V8 allows g1 to be clobbered
// across func calls, and V8+ also allows g5 to be clobbered across
// func calls. Also, g1 and g5 can get touched while doing shared
// library loading.
//
// We must not touch g7 (it is the thread-self register) and g6 is
// reserved for certain tools. g0, of course, is always zero.
//
// (Sources: SunSoft Compilers Group, thread library engineers.)
// %%%% The interpreter should be revisited to reduce global scratch regs.
// This global always holds the current JavaThread pointer:
// The following globals are part of the Java calling convention:
// The following globals are used for the new C1 & interpreter calling convention:
// This local is used to preserve G2_thread in the interpreter and in stubs:
// These globals are used as scratch registers in the interpreter:
// These globals are used as short-lived scratch registers in the compiler:
// The compiler requires that G5_megamorphic_method is G5_inline_cache_klass,
// because a single patchable "set" instruction (NativeMovConstReg,
// or NativeMovConstPatching for compiler1) instruction
// serves to set up either quantity, depending on whether the compiled
// call site is an inline cache or is megamorphic. See the function
// CompiledIC::set_to_megamorphic.
//
// On the other hand, G5_inline_cache_klass must differ from G5_method,
// because both registers are needed for an inline cache that calls
// an interpreted method.
//
// Note that G5_method is only the method-self for the interpreter,
// and is logically unrelated to G5_megamorphic_method.
//
// Invariants on G2_thread (the JavaThread pointer):
// - it should not be used for any other purpose anywhere
// - it must be re-initialized by StubRoutines::call_stub()
// - it must be preserved around every use of call_VM
// JavaThread, such as the card-marking base or perhaps pointers into
// Eden. It's something of a waste to use them as scratch temporaries,
// since they are not supposed to be volatile. (Of course, if we find
// that Java doesn't benefit from application globals, then we can just
// use them as ordinary temporaries.)
//
// it makes sense to use them routinely for procedure linkage,
// whenever the On registers are not applicable. Examples: G5_method,
// G5_inline_cache_klass, and a double handful of miscellaneous compiler
// stubs. This means that compiler stubs, etc., should be kept to a
// maximum of two or three G-register arguments.
// stub frames
REGISTER_DECLARATION(Register, Lentry_args , L0); // pointer to args passed to callee (interpreter) not stub itself
// Interpreter frames
#ifdef CC_INTERP
REGISTER_DECLARATION(Register, I5_savedSP , I5); // Saved SP before bumping for locals. This is simply
// a copy SP, so in 64-bit it's a biased value. The bias
// is added and removed as needed in the frame code.
// Interface to signature handler
#else
// must match Llocals in asm interpreter
// When calling out from the interpreter we record SP so that we can remove any extra stack
// space allocated during adapter transitions. This register is only live from the point
// of the call until we return.
REGISTER_DECLARATION(Register, I5_savedSP , I5); // Saved SP before bumping for locals. This is simply
// a copy SP, so in 64-bit it's a biased value. The bias
// is added and removed as needed in the frame code.
REGISTER_DECLARATION(Register, IdispatchTables , I4); // Base address of the bytecode dispatch tables
REGISTER_DECLARATION(Register, IdispatchAddress , I3); // Register which saves the dispatch address for each bytecode
#endif /* CC_INTERP */
// NOTE: Lscratch2 and LcpoolCache point to the same registers in
// the interpreter code. If Lscratch2 needs to be used for some
// purpose than LcpoolCache should be restore after that for
// the interpreter to work right
// (These assignments must be compatible with L7_thread_cache; see above.)
// Since Lbcp points into the middle of the method object,
// it is temporarily converted into a "bcx" during GC.
// Exception processing
// These registers are passed into exception handlers.
// All exception handlers require the exception object being thrown.
// In addition, an nmethod's exception handler must be passed
// the address of the call site within the nmethod, to allow
// proper selection of the applicable catch block.
// (Interpreter frames use their own bcp() for this purpose.)
//
// The Oissuing_pc value is not always needed. When jumping to a
// handler that is known to be interpreted, the Oissuing_pc value can be
// omitted. An actual catch block in compiled code receives (from its
// nmethod's exception handler) the thrown exception in the Oexception,
// but it doesn't need the Oissuing_pc.
//
// If an exception handler (either interpreted or compiled)
// discovers there is no applicable catch block, it updates
// the Oissuing_pc to the continuation PC of its own caller,
// pops back to that caller's stack frame, and executes that
// caller's exception handler. Obviously, this process will
// iterate until the control stack is popped back to a method
// containing an applicable catch block. A key invariant is
// that the Oissuing_pc value is always a value local to
// the method whose exception handler is currently executing.
//
// Note: The issuing PC value is __not__ a raw return address (I7 value).
// It is a "return pc", the address __following__ the call.
// Raw return addresses are converted to issuing PCs by frame::pc(),
// or by stubs. Issuing PCs can be used directly with PC range tables.
//
// These must occur after the declarations above
#ifndef DONT_USE_REGISTER_DEFINES
#ifdef CC_INTERP
#else
#endif /* ! CC_INTERP */
#endif
// Address is an abstraction used to represent a memory location.
//
// Note: A register location is represented via a Register, not
// via an address for efficiency & simplicity reasons.
class Address VALUE_OBJ_CLASS_SPEC {
private:
#ifdef _LP64
int _hi32; // bits 63::32
int _low32; // bits 31::0
#endif
int _hi;
int _disp;
switch (rt) {
case relocInfo::external_word_type:
return external_word_Relocation::spec(a);
case relocInfo::internal_word_type:
return internal_word_Relocation::spec(a);
#ifdef _LP64
case relocInfo::opt_virtual_call_type:
return opt_virtual_call_Relocation::spec();
case relocInfo::static_call_type:
return static_call_Relocation::spec();
case relocInfo::runtime_call_type:
return runtime_call_Relocation::spec();
#endif
return RelocationHolder();
default:
return RelocationHolder();
}
}
public:
{
_base = b;
#ifdef _LP64
#endif
}
{
_base = b;
#ifdef _LP64
#endif
}
{
_base = b;
#ifdef _LP64
// [RGV] Put in Assert to force me to check usage of this constructor
assert( h == 0, "Check usage of this constructor" );
_hi32 = h;
_low32 = d;
_hi = h;
_disp = d;
#else
_hi = h;
_disp = d;
#endif
}
Address()
: _rspec(RelocationHolder())
{
#ifdef _LP64
_hi32 = 0;
_low32 = 0;
#endif
_hi = 0;
_disp = 0;
}
// fancier constructors
enum addr_type {
extra_in_argument, // in the In registers
extra_out_argument // in the Outs
};
// accessors
#ifdef _LP64
#endif
#ifdef _LP64
#else
#endif
}
Address a = (*this);
if (hi_disp != 0) {
}
return a;
}
Address after_save() const {
Address a = (*this);
return a;
}
Address after_restore() const {
Address a = (*this);
return a;
}
friend class Assembler;
};
}
// Argument is an abstraction used to represent an outgoing
// actual argument or an incoming formal parameter, whether
// it resides in memory or in a register, in a manner consistent
// with the SPARC Application Binary Interface, or ABI. This is
// often referred to as the native or C calling convention.
class Argument VALUE_OBJ_CLASS_SPEC {
private:
int _number;
bool _is_in;
public:
#ifdef _LP64
enum {
};
#else
enum {
};
#endif
// creation
// locating register-based arguments:
#ifdef _LP64
// locating Floating Point register-based arguments:
FloatRegister as_float_register() const {
}
FloatRegister as_double_register() const {
}
#endif
Register as_register() const {
}
// locating memory-based arguments
Address as_address() const {
return address_in_frame();
}
// When applied to a register-based argument, give the corresponding address
// into the 6-word area "into which callee may store register arguments"
// (This is a different place than the corresponding register-save area location.)
Address address_in_frame() const {
_number );
}
// debugging
const char* name() const;
friend class Assembler;
};
// The SPARC Assembler: Pure assembler doing NO optimizations on the instruction
// level; i.e., what you write
// is what you get. The Assembler is generating code into a CodeBuffer.
class Assembler : public AbstractAssembler {
protected:
static void print_instruction(int inst);
friend class AbstractAssembler;
// code patchers need various routines like inv_wdisp()
friend class NativeInstruction;
friend class NativeGeneralJump;
friend class Relocation;
friend class Label;
public:
// op carries format info; see page 62 & 267
enum ops {
branch_op = 0, // also sethi (fmt2)
};
enum op2s {
bpr_op2 = 3,
fb_op2 = 6,
fbp_op2 = 5,
br_op2 = 2,
bp_op2 = 1,
sethi_op2 = 4
};
enum op3s {
// selected op3s
add_op3 = 0x00,
and_op3 = 0x01,
or_op3 = 0x02,
xor_op3 = 0x03,
sub_op3 = 0x04,
andn_op3 = 0x05,
orn_op3 = 0x06,
xnor_op3 = 0x07,
addc_op3 = 0x08,
mulx_op3 = 0x09,
umul_op3 = 0x0a,
smul_op3 = 0x0b,
subc_op3 = 0x0c,
udivx_op3 = 0x0d,
udiv_op3 = 0x0e,
sdiv_op3 = 0x0f,
addcc_op3 = 0x10,
andcc_op3 = 0x11,
orcc_op3 = 0x12,
xorcc_op3 = 0x13,
subcc_op3 = 0x14,
andncc_op3 = 0x15,
orncc_op3 = 0x16,
xnorcc_op3 = 0x17,
addccc_op3 = 0x18,
umulcc_op3 = 0x1a,
smulcc_op3 = 0x1b,
subccc_op3 = 0x1c,
udivcc_op3 = 0x1e,
sdivcc_op3 = 0x1f,
taddcc_op3 = 0x20,
tsubcc_op3 = 0x21,
taddcctv_op3 = 0x22,
tsubcctv_op3 = 0x23,
mulscc_op3 = 0x24,
sll_op3 = 0x25,
sllx_op3 = 0x25,
srl_op3 = 0x26,
srlx_op3 = 0x26,
sra_op3 = 0x27,
srax_op3 = 0x27,
rdreg_op3 = 0x28,
membar_op3 = 0x28,
flushw_op3 = 0x2b,
movcc_op3 = 0x2c,
sdivx_op3 = 0x2d,
popc_op3 = 0x2e,
movr_op3 = 0x2f,
sir_op3 = 0x30,
wrreg_op3 = 0x30,
saved_op3 = 0x31,
fpop1_op3 = 0x34,
fpop2_op3 = 0x35,
impdep1_op3 = 0x36,
impdep2_op3 = 0x37,
jmpl_op3 = 0x38,
rett_op3 = 0x39,
trap_op3 = 0x3a,
flush_op3 = 0x3b,
save_op3 = 0x3c,
restore_op3 = 0x3d,
done_op3 = 0x3e,
retry_op3 = 0x3e,
lduw_op3 = 0x00,
ldub_op3 = 0x01,
lduh_op3 = 0x02,
ldd_op3 = 0x03,
stw_op3 = 0x04,
stb_op3 = 0x05,
sth_op3 = 0x06,
std_op3 = 0x07,
ldsw_op3 = 0x08,
ldsb_op3 = 0x09,
ldsh_op3 = 0x0a,
ldx_op3 = 0x0b,
ldstub_op3 = 0x0d,
stx_op3 = 0x0e,
swap_op3 = 0x0f,
lduwa_op3 = 0x10,
ldxa_op3 = 0x1b,
stwa_op3 = 0x14,
stxa_op3 = 0x1e,
ldf_op3 = 0x20,
ldfsr_op3 = 0x21,
ldqf_op3 = 0x22,
lddf_op3 = 0x23,
stf_op3 = 0x24,
stfsr_op3 = 0x25,
stqf_op3 = 0x26,
stdf_op3 = 0x27,
prefetch_op3 = 0x2d,
ldc_op3 = 0x30,
ldcsr_op3 = 0x31,
lddc_op3 = 0x33,
stc_op3 = 0x34,
stcsr_op3 = 0x35,
stdcq_op3 = 0x36,
stdc_op3 = 0x37,
casa_op3 = 0x3c,
casxa_op3 = 0x3e,
alt_bit_op3 = 0x10,
cc_bit_op3 = 0x10
};
enum opfs {
// selected opfs
fmovs_opf = 0x01,
fmovd_opf = 0x02,
fnegs_opf = 0x05,
fnegd_opf = 0x06,
fadds_opf = 0x41,
faddd_opf = 0x42,
fsubs_opf = 0x45,
fsubd_opf = 0x46,
fmuls_opf = 0x49,
fmuld_opf = 0x4a,
fdivs_opf = 0x4d,
fdivd_opf = 0x4e,
fcmps_opf = 0x51,
fcmpd_opf = 0x52,
fstox_opf = 0x81,
fdtox_opf = 0x82,
fxtos_opf = 0x84,
fxtod_opf = 0x88,
fitos_opf = 0xc4,
fdtos_opf = 0xc6,
fitod_opf = 0xc8,
fstod_opf = 0xc9,
fstoi_opf = 0xd1,
fdtoi_opf = 0xd2
};
enum Condition {
// for FBfcc & FBPfcc instruction
f_never = 0,
f_notEqual = 1,
f_notZero = 1,
f_lessOrGreater = 2,
f_unorderedOrLess = 3,
f_less = 4,
f_unorderedOrGreater = 5,
f_greater = 6,
f_unordered = 7,
f_always = 8,
f_equal = 9,
f_zero = 9,
f_unorderedOrEqual = 10,
f_greaterOrEqual = 11,
f_lessOrEqual = 13,
f_unorderedOrLessOrEqual = 14,
f_ordered = 15,
// V8 coproc, pp 123 v8 manual
cp_always = 8,
cp_never = 0,
cp_3 = 7,
cp_2 = 6,
cp_2or3 = 5,
cp_1 = 4,
cp_1or3 = 3,
cp_1or2 = 2,
cp_1or2or3 = 1,
cp_0 = 9,
cp_0or3 = 10,
cp_0or2 = 11,
cp_0or2or3 = 12,
cp_0or1 = 13,
cp_0or1or3 = 14,
cp_0or1or2 = 15,
// for integers
never = 0,
equal = 1,
zero = 1,
lessEqual = 2,
less = 3,
lessEqualUnsigned = 4,
lessUnsigned = 5,
carrySet = 5,
negative = 6,
overflowSet = 7,
always = 8,
notEqual = 9,
notZero = 9,
greater = 10,
greaterEqual = 11,
greaterUnsigned = 12,
greaterEqualUnsigned = 13,
carryClear = 13,
positive = 14,
overflowClear = 15
};
enum CC {
// ptr_cc is the correct condition code for a pointer or intptr_t:
};
enum PrefetchFcn {
};
public:
// Helper functions for groups of instructions
enum Membar_mask_bits { // page 184, v9
LoadLoad = 1 << 0,
};
// test if x is within signed immediate range for nbits
// test if -4096 <= x <= 4095
enum ASIs { // page 72, v9
ASI_PRIMARY = 0x80,
ASI_PRIMARY_LITTLE = 0x88
// add more from book as needed
};
protected:
// helpers
// x is supposed to fit in a field "nbits" wide
// and be sign-extended. Check the range.
"value out of range");
}
}
static void assert_unsigned_const(int x, int nbits) {
}
// fields: note bits numbered from LSB = 0,
// fields known by inclusive bit range
}
// inverse of u_field
return int(r);
}
// signed version: extract from field and sign-extend
}
// given a field that ranges from hi_bit to lo_bit (inclusive,
// LSB = 0), and an unsigned value for the field,
// shift it into the field
#ifdef ASSERT
"value out of range");
int r = x << lo_bit;
return r;
}
#else
// make sure this is inlined as it will reduce code size significantly
#endif
static int mov_cc( CC c, bool useFloat ) { return u_field(useFloat ? 0 : 1, 18, 18) | u_field(c, 12, 11); }
static int fd( FloatRegister r, FloatRegisterImpl::Width fwa) { return u_field(r->encoding(fwa), 29, 25); };
static int fs1(FloatRegister r, FloatRegisterImpl::Width fwa) { return u_field(r->encoding(fwa), 18, 14); };
static int fs2(FloatRegister r, FloatRegisterImpl::Width fwa) { return u_field(r->encoding(fwa), 4, 0); };
// some float instructions use this encoding on the op3 field
int r;
switch(w) {
case FloatRegisterImpl::S: r = op + 0; break;
default: ShouldNotReachHere(); break;
}
return op3(r);
}
// compute inverse of simm
}
// signed immediate, in low bits, nbits long
assert_signed_range(x, nbits);
}
// compute inverse of wdisp16
}
// word offset, 14 bits at LSend, 2 bits at B21, B20
return r;
}
// word displacement in low-order nbits bits
return (r << 2) + pos;
}
return r;
}
// Extract the top 32 bits in a 64 bit word
return r;
}
// given a sethi instruction, extract the constant, left-justified
static int inv_hi22( int x ) {
return x << 10;
}
// create an imm22 field, given a 32-bit left-justified constant
static int hi22( int x ) {
int r = int( juint(x) >> 10 );
return r;
}
// create a low10 __value__ (not a field) for a given a 32-bit constant
static int low10( int x ) {
return x & ((1 << 10) - 1);
}
// instruction only in v9
static void v9_only() { assert( VM_Version::v9_instructions_work(), "This instruction only works on SPARC V9"); }
// instruction only in v8
static void v8_only() { assert( VM_Version::v8_instructions_work(), "This instruction only works on SPARC V8"); }
// instruction deprecated in v9
static void v9_dep() { } // do nothing for now
// some float instructions only exist for single prec. on v8
// v8 has no CC field
protected:
// Simple delay-slot scheme:
// In order to check the programmer, the assembler keeps track of deley slots.
// It forbids CTIs in delay slots (conservative, but should be OK).
// Also, when putting an instruction into a delay slot, you must say
// asm->delayed()->add(...), in order to check that you don't omit
// delay-slot instructions.
// To implement this, we use a simple FSA
#ifdef ASSERT
#define CHECK_DELAY
#endif
#ifdef CHECK_DELAY
#endif
public:
// Tells assembler next instruction must NOT be in delay slot.
// Use at start of multinstruction macros.
void assert_not_delayed() {
// This is a separate overloading to avoid creation of string constants
// in non-asserted code--with some compilers this pollutes the object code.
#ifdef CHECK_DELAY
assert_not_delayed("next instruction should not be a delay slot");
#endif
}
void assert_not_delayed(const char* msg) {
#ifdef CHECK_DELAY
#endif
}
protected:
// Delay slot helpers
// cti is called when emitting control-transfer instruction,
// BEFORE doing the emitting.
// Only effective when assertion-checking is enabled.
void cti() {
#ifdef CHECK_DELAY
assert_not_delayed("cti should not be in delay slot");
#endif
}
// called when emitting cti with a delay slot, AFTER emitting
void has_delay_slot() {
#ifdef CHECK_DELAY
assert_not_delayed("just checking");
#endif
}
public:
// Tells assembler you know that next instruction is delayed
#ifdef CHECK_DELAY
#endif
return this;
}
void flush() {
#ifdef CHECK_DELAY
#endif
}
inline void emit_long(int); // shadows AbstractAssembler::emit_long
inline void emit_data(int, RelocationHolder const&);
// helper for above fcns
inline void check_delay();
public:
// instructions, refer to page numbers in the SPARC Architecture Manual, V9
// pp 135 (addc was addx in v8)
inline void add( Register s1, int simm13a, Register d, relocInfo::relocType rtype = relocInfo::none);
void addcc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(add_op3 | cc_bit_op3) | rs1(s1) | rs2(s2) ); }
void addcc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(add_op3 | cc_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void addc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(addc_op3 ) | rs1(s1) | rs2(s2) ); }
void addc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(addc_op3 ) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void addccc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(addc_op3 | cc_bit_op3) | rs1(s1) | rs2(s2) ); }
void addccc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(addc_op3 | cc_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
// pp 136
inline void bpr( RCondition c, bool a, Predict p, Register s1, address d, relocInfo::relocType rt = relocInfo::none );
protected: // use MacroAssembler::br instead
// pp 138
// pp 141
inline void fbp( Condition c, bool a, CC cc, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
public:
// pp 144
// pp 146
inline void bp( Condition c, bool a, CC cc, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
// pp 121 (V8)
// pp 149
// pp 150
// These instructions compare the contents of s2 with the contents of
// memory at address in s1. If the values are equal, the contents of memory
// at address s1 is swapped with the data in d. If the values are not equal,
// the the contents of memory at s1 is loaded into d, without the swap.
void casa( Register s1, Register s2, Register d, int ia = -1 ) { v9_only(); emit_long( op(ldst_op) | rd(d) | op3(casa_op3 ) | rs1(s1) | (ia == -1 ? immed(true) : imm_asi(ia)) | rs2(s2)); }
void casxa( Register s1, Register s2, Register d, int ia = -1 ) { v9_only(); emit_long( op(ldst_op) | rd(d) | op3(casxa_op3) | rs1(s1) | (ia == -1 ? immed(true) : imm_asi(ia)) | rs2(s2)); }
// pp 152
void udiv( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(udiv_op3 ) | rs1(s1) | rs2(s2)); }
void udiv( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(udiv_op3 ) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void sdiv( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(sdiv_op3 ) | rs1(s1) | rs2(s2)); }
void sdiv( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(sdiv_op3 ) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void udivcc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(udiv_op3 | cc_bit_op3) | rs1(s1) | rs2(s2)); }
void udivcc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(udiv_op3 | cc_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void sdivcc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(sdiv_op3 | cc_bit_op3) | rs1(s1) | rs2(s2)); }
void sdivcc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(sdiv_op3 | cc_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
// pp 155
// pp 156
void fadd( FloatRegisterImpl::Width w, FloatRegister s1, FloatRegister s2, FloatRegister d ) { emit_long( op(arith_op) | fd(d, w) | op3(fpop1_op3) | fs1(s1, w) | opf(0x40 + w) | fs2(s2, w)); }
void fsub( FloatRegisterImpl::Width w, FloatRegister s1, FloatRegister s2, FloatRegister d ) { emit_long( op(arith_op) | fd(d, w) | op3(fpop1_op3) | fs1(s1, w) | opf(0x44 + w) | fs2(s2, w)); }
// pp 157
void fcmp( FloatRegisterImpl::Width w, CC cc, FloatRegister s1, FloatRegister s2) { v8_no_cc(cc); emit_long( op(arith_op) | cmpcc(cc) | op3(fpop2_op3) | fs1(s1, w) | opf(0x50 + w) | fs2(s2, w)); }
void fcmpe( FloatRegisterImpl::Width w, CC cc, FloatRegister s1, FloatRegister s2) { v8_no_cc(cc); emit_long( op(arith_op) | cmpcc(cc) | op3(fpop2_op3) | fs1(s1, w) | opf(0x54 + w) | fs2(s2, w)); }
// pp 159
void ftox( FloatRegisterImpl::Width w, FloatRegister s, FloatRegister d ) { v9_only(); emit_long( op(arith_op) | fd(d, w) | op3(fpop1_op3) | opf(0x80 + w) | fs2(s, w)); }
void ftoi( FloatRegisterImpl::Width w, FloatRegister s, FloatRegister d ) { emit_long( op(arith_op) | fd(d, w) | op3(fpop1_op3) | opf(0xd0 + w) | fs2(s, w)); }
// pp 160
void ftof( FloatRegisterImpl::Width sw, FloatRegisterImpl::Width dw, FloatRegister s, FloatRegister d ) { emit_long( op(arith_op) | fd(d, dw) | op3(fpop1_op3) | opf(0xc0 + sw + dw*4) | fs2(s, sw)); }
// pp 161
void fxtof( FloatRegisterImpl::Width w, FloatRegister s, FloatRegister d ) { v9_only(); emit_long( op(arith_op) | fd(d, w) | op3(fpop1_op3) | opf(0x80 + w*4) | fs2(s, w)); }
void fitof( FloatRegisterImpl::Width w, FloatRegister s, FloatRegister d ) { emit_long( op(arith_op) | fd(d, w) | op3(fpop1_op3) | opf(0xc0 + w*4) | fs2(s, w)); }
// pp 162
void fmov( FloatRegisterImpl::Width w, FloatRegister s, FloatRegister d ) { v8_s_only(w); emit_long( op(arith_op) | fd(d, w) | op3(fpop1_op3) | opf(0x00 + w) | fs2(s, w)); }
void fneg( FloatRegisterImpl::Width w, FloatRegister s, FloatRegister d ) { v8_s_only(w); emit_long( op(arith_op) | fd(d, w) | op3(fpop1_op3) | opf(0x04 + w) | fs2(s, w)); }
// page 144 sparc v8 architecture (double prec works on v8 if the source and destination registers are the same). fnegs is the only instruction available
// on v8 to do negation of single, double and quad precision floats.
void fneg( FloatRegisterImpl::Width w, FloatRegister sd ) { if (VM_Version::v9_instructions_work()) emit_long( op(arith_op) | fd(sd, w) | op3(fpop1_op3) | opf(0x04 + w) | fs2(sd, w)); else emit_long( op(arith_op) | fd(sd, w) | op3(fpop1_op3) | opf(0x05) | fs2(sd, w)); }
void fabs( FloatRegisterImpl::Width w, FloatRegister s, FloatRegister d ) { v8_s_only(w); emit_long( op(arith_op) | fd(d, w) | op3(fpop1_op3) | opf(0x08 + w) | fs2(s, w)); }
// page 144 sparc v8 architecture (double prec works on v8 if the source and destination registers are the same). fabss is the only instruction available
void fabs( FloatRegisterImpl::Width w, FloatRegister sd ) { if (VM_Version::v9_instructions_work()) emit_long( op(arith_op) | fd(sd, w) | op3(fpop1_op3) | opf(0x08 + w) | fs2(sd, w)); else emit_long( op(arith_op) | fd(sd, w) | op3(fpop1_op3) | opf(0x09) | fs2(sd, w)); }
// pp 163
void fmul( FloatRegisterImpl::Width w, FloatRegister s1, FloatRegister s2, FloatRegister d ) { emit_long( op(arith_op) | fd(d, w) | op3(fpop1_op3) | fs1(s1, w) | opf(0x48 + w) | fs2(s2, w)); }
void fmul( FloatRegisterImpl::Width sw, FloatRegisterImpl::Width dw, FloatRegister s1, FloatRegister s2, FloatRegister d ) { emit_long( op(arith_op) | fd(d, dw) | op3(fpop1_op3) | fs1(s1, sw) | opf(0x60 + sw + dw*4) | fs2(s2, sw)); }
void fdiv( FloatRegisterImpl::Width w, FloatRegister s1, FloatRegister s2, FloatRegister d ) { emit_long( op(arith_op) | fd(d, w) | op3(fpop1_op3) | fs1(s1, w) | opf(0x4c + w) | fs2(s2, w)); }
// pp 164
void fsqrt( FloatRegisterImpl::Width w, FloatRegister s, FloatRegister d ) { emit_long( op(arith_op) | fd(d, w) | op3(fpop1_op3) | opf(0x28 + w) | fs2(s, w)); }
// pp 165
// pp 167
// pp 168
void illtrap( int const22a) { if (const22a != 0) v9_only(); emit_long( op(branch_op) | u_field(const22a, 21, 0) ); }
// v8 unimp == illtrap(0)
// pp 169
void impdep1( int id1, int const19a ) { v9_only(); emit_long( op(arith_op) | fcn(id1) | op3(impdep1_op3) | u_field(const19a, 18, 0)); }
void impdep2( int id1, int const19a ) { v9_only(); emit_long( op(arith_op) | fcn(id1) | op3(impdep2_op3) | u_field(const19a, 18, 0)); }
// pp 149 (v8)
void cpop1( int opc, int cr1, int cr2, int crd ) { v8_only(); emit_long( op(arith_op) | fcn(crd) | op3(impdep1_op3) | u_field(cr1, 18, 14) | opf(opc) | u_field(cr2, 4, 0)); }
void cpop2( int opc, int cr1, int cr2, int crd ) { v8_only(); emit_long( op(arith_op) | fcn(crd) | op3(impdep2_op3) | u_field(cr1, 18, 14) | opf(opc) | u_field(cr2, 4, 0)); }
// pp 170
void jmpl( Register s1, int simm13a, Register d, RelocationHolder const& rspec = RelocationHolder() );
// 171
// pp 94 (v8)
// 173
void ldfa( FloatRegisterImpl::Width w, Register s1, Register s2, int ia, FloatRegister d ) { v9_only(); emit_long( op(ldst_op) | fd(d, w) | alt_op3(ldf_op3 | alt_bit_op3, w) | rs1(s1) | imm_asi(ia) | rs2(s2) ); }
void ldfa( FloatRegisterImpl::Width w, Register s1, int simm13a, FloatRegister d ) { v9_only(); emit_long( op(ldst_op) | fd(d, w) | alt_op3(ldf_op3 | alt_bit_op3, w) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
// pp 175, lduw is ld on v8
// pp 177
void ldsba( Register s1, Register s2, int ia, Register d ) { emit_long( op(ldst_op) | rd(d) | op3(ldsb_op3 | alt_bit_op3) | rs1(s1) | imm_asi(ia) | rs2(s2) ); }
void ldsba( Register s1, int simm13a, Register d ) { emit_long( op(ldst_op) | rd(d) | op3(ldsb_op3 | alt_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void ldsha( Register s1, Register s2, int ia, Register d ) { emit_long( op(ldst_op) | rd(d) | op3(ldsh_op3 | alt_bit_op3) | rs1(s1) | imm_asi(ia) | rs2(s2) ); }
void ldsha( Register s1, int simm13a, Register d ) { emit_long( op(ldst_op) | rd(d) | op3(ldsh_op3 | alt_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void ldswa( Register s1, Register s2, int ia, Register d ) { v9_only(); emit_long( op(ldst_op) | rd(d) | op3(ldsw_op3 | alt_bit_op3) | rs1(s1) | imm_asi(ia) | rs2(s2) ); }
void ldswa( Register s1, int simm13a, Register d ) { v9_only(); emit_long( op(ldst_op) | rd(d) | op3(ldsw_op3 | alt_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void lduba( Register s1, Register s2, int ia, Register d ) { emit_long( op(ldst_op) | rd(d) | op3(ldub_op3 | alt_bit_op3) | rs1(s1) | imm_asi(ia) | rs2(s2) ); }
void lduba( Register s1, int simm13a, Register d ) { emit_long( op(ldst_op) | rd(d) | op3(ldub_op3 | alt_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void lduha( Register s1, Register s2, int ia, Register d ) { emit_long( op(ldst_op) | rd(d) | op3(lduh_op3 | alt_bit_op3) | rs1(s1) | imm_asi(ia) | rs2(s2) ); }
void lduha( Register s1, int simm13a, Register d ) { emit_long( op(ldst_op) | rd(d) | op3(lduh_op3 | alt_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void lduwa( Register s1, Register s2, int ia, Register d ) { emit_long( op(ldst_op) | rd(d) | op3(lduw_op3 | alt_bit_op3) | rs1(s1) | imm_asi(ia) | rs2(s2) ); }
void lduwa( Register s1, int simm13a, Register d ) { emit_long( op(ldst_op) | rd(d) | op3(lduw_op3 | alt_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void ldxa( Register s1, Register s2, int ia, Register d ) { v9_only(); emit_long( op(ldst_op) | rd(d) | op3(ldx_op3 | alt_bit_op3) | rs1(s1) | imm_asi(ia) | rs2(s2) ); }
void ldxa( Register s1, int simm13a, Register d ) { v9_only(); emit_long( op(ldst_op) | rd(d) | op3(ldx_op3 | alt_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void ldda( Register s1, Register s2, int ia, Register d ) { v9_dep(); emit_long( op(ldst_op) | rd(d) | op3(ldd_op3 | alt_bit_op3) | rs1(s1) | imm_asi(ia) | rs2(s2) ); }
void ldda( Register s1, int simm13a, Register d ) { v9_dep(); emit_long( op(ldst_op) | rd(d) | op3(ldd_op3 | alt_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
// pp 179
// pp 180
void ldstuba( Register s1, Register s2, int ia, Register d ) { emit_long( op(ldst_op) | rd(d) | op3(ldstub_op3 | alt_bit_op3) | rs1(s1) | imm_asi(ia) | rs2(s2) ); }
void ldstuba( Register s1, int simm13a, Register d ) { emit_long( op(ldst_op) | rd(d) | op3(ldstub_op3 | alt_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
// pp 181
void and3( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(and_op3 ) | rs1(s1) | rs2(s2) ); }
void and3( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(and_op3 ) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void andcc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(and_op3 | cc_bit_op3) | rs1(s1) | rs2(s2) ); }
void andcc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(and_op3 | cc_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void andn( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(andn_op3 ) | rs1(s1) | rs2(s2) ); }
void andn( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(andn_op3 ) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void andncc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(andn_op3 | cc_bit_op3) | rs1(s1) | rs2(s2) ); }
void andncc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(andn_op3 | cc_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void or3( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(or_op3 ) | rs1(s1) | rs2(s2) ); }
void or3( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(or_op3 ) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void orcc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(or_op3 | cc_bit_op3) | rs1(s1) | rs2(s2) ); }
void orcc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(or_op3 | cc_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void orn( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(orn_op3) | rs1(s1) | rs2(s2) ); }
void orn( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(orn_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void orncc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(orn_op3 | cc_bit_op3) | rs1(s1) | rs2(s2) ); }
void orncc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(orn_op3 | cc_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void xor3( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(xor_op3 ) | rs1(s1) | rs2(s2) ); }
void xor3( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(xor_op3 ) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void xorcc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(xor_op3 | cc_bit_op3) | rs1(s1) | rs2(s2) ); }
void xorcc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(xor_op3 | cc_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void xnor( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(xnor_op3 ) | rs1(s1) | rs2(s2) ); }
void xnor( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(xnor_op3 ) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void xnorcc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(xnor_op3 | cc_bit_op3) | rs1(s1) | rs2(s2) ); }
void xnorcc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(xnor_op3 | cc_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
// pp 183
void membar( Membar_mask_bits const7a ) { v9_only(); emit_long( op(arith_op) | op3(membar_op3) | rs1(O7) | immed(true) | u_field( int(const7a), 6, 0)); }
// pp 185
void fmov( FloatRegisterImpl::Width w, Condition c, bool floatCC, CC cca, FloatRegister s2, FloatRegister d ) { v9_only(); emit_long( op(arith_op) | fd(d, w) | op3(fpop2_op3) | cond_mov(c) | opf_cc(cca, floatCC) | opf_low6(w) | fs2(s2, w)); }
// pp 189
void fmov( FloatRegisterImpl::Width w, RCondition c, Register s1, FloatRegister s2, FloatRegister d ) { v9_only(); emit_long( op(arith_op) | fd(d, w) | op3(fpop2_op3) | rs1(s1) | rcond(c) | opf_low5(4 + w) | fs2(s2, w)); }
// pp 191
void movcc( Condition c, bool floatCC, CC cca, Register s2, Register d ) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(movcc_op3) | mov_cc(cca, floatCC) | cond_mov(c) | rs2(s2) ); }
void movcc( Condition c, bool floatCC, CC cca, int simm11a, Register d ) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(movcc_op3) | mov_cc(cca, floatCC) | cond_mov(c) | immed(true) | simm(simm11a, 11) ); }
// pp 195
void movr( RCondition c, Register s1, Register s2, Register d ) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(movr_op3) | rs1(s1) | rcond(c) | rs2(s2) ); }
void movr( RCondition c, Register s1, int simm10a, Register d ) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(movr_op3) | rs1(s1) | rcond(c) | immed(true) | simm(simm10a, 10) ); }
// pp 196
void mulx( Register s1, Register s2, Register d ) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(mulx_op3 ) | rs1(s1) | rs2(s2) ); }
void mulx( Register s1, int simm13a, Register d ) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(mulx_op3 ) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void sdivx( Register s1, Register s2, Register d ) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(sdivx_op3) | rs1(s1) | rs2(s2) ); }
void sdivx( Register s1, int simm13a, Register d ) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(sdivx_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void udivx( Register s1, Register s2, Register d ) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(udivx_op3) | rs1(s1) | rs2(s2) ); }
void udivx( Register s1, int simm13a, Register d ) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(udivx_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
// pp 197
void umul( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(umul_op3 ) | rs1(s1) | rs2(s2) ); }
void umul( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(umul_op3 ) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void smul( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(smul_op3 ) | rs1(s1) | rs2(s2) ); }
void smul( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(smul_op3 ) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void umulcc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(umul_op3 | cc_bit_op3) | rs1(s1) | rs2(s2) ); }
void umulcc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(umul_op3 | cc_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void smulcc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(smul_op3 | cc_bit_op3) | rs1(s1) | rs2(s2) ); }
void smulcc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(smul_op3 | cc_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
// pp 199
void mulscc( Register s1, Register s2, Register d ) { v9_dep(); emit_long( op(arith_op) | rd(d) | op3(mulscc_op3) | rs1(s1) | rs2(s2) ); }
void mulscc( Register s1, int simm13a, Register d ) { v9_dep(); emit_long( op(arith_op) | rd(d) | op3(mulscc_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
// pp 201
// pp 202
void popc( Register s, Register d) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(popc_op3) | rs2(s)); }
void popc( int simm13a, Register d) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(popc_op3) | immed(true) | simm(simm13a, 13)); }
// pp 203
void prefetcha( Register s1, Register s2, int ia, PrefetchFcn f ) { v9_only(); emit_long( op(ldst_op) | fcn(f) | op3(prefetch_op3 | alt_bit_op3) | rs1(s1) | imm_asi(ia) | rs2(s2) ); }
void prefetcha( Register s1, int simm13a, PrefetchFcn f ) { v9_only(); emit_long( op(ldst_op) | fcn(f) | op3(prefetch_op3 | alt_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
// pp 208
// not implementing read privileged register
inline void rdy( Register d) { v9_dep(); emit_long( op(arith_op) | rd(d) | op3(rdreg_op3) | u_field(0, 18, 14)); }
inline void rdccr( Register d) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(rdreg_op3) | u_field(2, 18, 14)); }
inline void rdasi( Register d) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(rdreg_op3) | u_field(3, 18, 14)); }
inline void rdtick( Register d) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(rdreg_op3) | u_field(4, 18, 14)); } // Spoon!
inline void rdpc( Register d) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(rdreg_op3) | u_field(5, 18, 14)); }
inline void rdfprs( Register d) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(rdreg_op3) | u_field(6, 18, 14)); }
// pp 213
// pp 214
void save( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(save_op3) | rs1(s1) | rs2(s2) ); }
void save( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(save_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void restore( Register s1 = G0, Register s2 = G0, Register d = G0 ) { emit_long( op(arith_op) | rd(d) | op3(restore_op3) | rs1(s1) | rs2(s2) ); }
void restore( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(restore_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
// pp 216
// pp 217
// pp 218
void sll( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(sll_op3) | rs1(s1) | sx(0) | rs2(s2) ); }
void sll( Register s1, int imm5a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(sll_op3) | rs1(s1) | sx(0) | immed(true) | u_field(imm5a, 4, 0) ); }
void srl( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(srl_op3) | rs1(s1) | sx(0) | rs2(s2) ); }
void srl( Register s1, int imm5a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(srl_op3) | rs1(s1) | sx(0) | immed(true) | u_field(imm5a, 4, 0) ); }
void sra( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(sra_op3) | rs1(s1) | sx(0) | rs2(s2) ); }
void sra( Register s1, int imm5a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(sra_op3) | rs1(s1) | sx(0) | immed(true) | u_field(imm5a, 4, 0) ); }
void sllx( Register s1, Register s2, Register d ) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(sll_op3) | rs1(s1) | sx(1) | rs2(s2) ); }
void sllx( Register s1, int imm6a, Register d ) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(sll_op3) | rs1(s1) | sx(1) | immed(true) | u_field(imm6a, 5, 0) ); }
void srlx( Register s1, Register s2, Register d ) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(srl_op3) | rs1(s1) | sx(1) | rs2(s2) ); }
void srlx( Register s1, int imm6a, Register d ) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(srl_op3) | rs1(s1) | sx(1) | immed(true) | u_field(imm6a, 5, 0) ); }
void srax( Register s1, Register s2, Register d ) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(sra_op3) | rs1(s1) | sx(1) | rs2(s2) ); }
void srax( Register s1, int imm6a, Register d ) { v9_only(); emit_long( op(arith_op) | rd(d) | op3(sra_op3) | rs1(s1) | sx(1) | immed(true) | u_field(imm6a, 5, 0) ); }
// pp 220
void sir( int simm13a ) { emit_long( op(arith_op) | fcn(15) | op3(sir_op3) | immed(true) | simm(simm13a, 13)); }
// pp 221
// pp 222
// pp 224
void stfa( FloatRegisterImpl::Width w, FloatRegister d, Register s1, Register s2, int ia ) { v9_only(); emit_long( op(ldst_op) | fd(d, w) | alt_op3(stf_op3 | alt_bit_op3, w) | rs1(s1) | imm_asi(ia) | rs2(s2) ); }
void stfa( FloatRegisterImpl::Width w, FloatRegister d, Register s1, int simm13a ) { v9_only(); emit_long( op(ldst_op) | fd(d, w) | alt_op3(stf_op3 | alt_bit_op3, w) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
// p 226
// pp 177
void stba( Register d, Register s1, Register s2, int ia ) { emit_long( op(ldst_op) | rd(d) | op3(stb_op3 | alt_bit_op3) | rs1(s1) | imm_asi(ia) | rs2(s2) ); }
void stba( Register d, Register s1, int simm13a ) { emit_long( op(ldst_op) | rd(d) | op3(stb_op3 | alt_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void stha( Register d, Register s1, Register s2, int ia ) { emit_long( op(ldst_op) | rd(d) | op3(sth_op3 | alt_bit_op3) | rs1(s1) | imm_asi(ia) | rs2(s2) ); }
void stha( Register d, Register s1, int simm13a ) { emit_long( op(ldst_op) | rd(d) | op3(sth_op3 | alt_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void stwa( Register d, Register s1, Register s2, int ia ) { emit_long( op(ldst_op) | rd(d) | op3(stw_op3 | alt_bit_op3) | rs1(s1) | imm_asi(ia) | rs2(s2) ); }
void stwa( Register d, Register s1, int simm13a ) { emit_long( op(ldst_op) | rd(d) | op3(stw_op3 | alt_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void stxa( Register d, Register s1, Register s2, int ia ) { v9_only(); emit_long( op(ldst_op) | rd(d) | op3(stx_op3 | alt_bit_op3) | rs1(s1) | imm_asi(ia) | rs2(s2) ); }
void stxa( Register d, Register s1, int simm13a ) { v9_only(); emit_long( op(ldst_op) | rd(d) | op3(stx_op3 | alt_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void stda( Register d, Register s1, Register s2, int ia ) { emit_long( op(ldst_op) | rd(d) | op3(std_op3 | alt_bit_op3) | rs1(s1) | imm_asi(ia) | rs2(s2) ); }
void stda( Register d, Register s1, int simm13a ) { emit_long( op(ldst_op) | rd(d) | op3(std_op3 | alt_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
// pp 97 (v8)
// pp 230
void sub( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(sub_op3 ) | rs1(s1) | rs2(s2) ); }
void sub( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(sub_op3 ) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void subcc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(sub_op3 | cc_bit_op3 ) | rs1(s1) | rs2(s2) ); }
void subcc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(sub_op3 | cc_bit_op3 ) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void subc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(subc_op3 ) | rs1(s1) | rs2(s2) ); }
void subc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(subc_op3 ) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void subccc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(subc_op3 | cc_bit_op3) | rs1(s1) | rs2(s2) ); }
void subccc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(subc_op3 | cc_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
// pp 231
// pp 232
void swapa( Register s1, Register s2, int ia, Register d ) { v9_dep(); emit_long( op(ldst_op) | rd(d) | op3(swap_op3 | alt_bit_op3) | rs1(s1) | imm_asi(ia) | rs2(s2) ); }
void swapa( Register s1, int simm13a, Register d ) { v9_dep(); emit_long( op(ldst_op) | rd(d) | op3(swap_op3 | alt_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
// pp 234, note op in book is wrong, see pp 268
void taddcc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(taddcc_op3 ) | rs1(s1) | rs2(s2) ); }
void taddcc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(taddcc_op3 ) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void taddcctv( Register s1, Register s2, Register d ) { v9_dep(); emit_long( op(arith_op) | rd(d) | op3(taddcctv_op3) | rs1(s1) | rs2(s2) ); }
void taddcctv( Register s1, int simm13a, Register d ) { v9_dep(); emit_long( op(arith_op) | rd(d) | op3(taddcctv_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
// pp 235
void tsubcc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(tsubcc_op3 ) | rs1(s1) | rs2(s2) ); }
void tsubcc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(tsubcc_op3 ) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
void tsubcctv( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(tsubcctv_op3) | rs1(s1) | rs2(s2) ); }
void tsubcctv( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(tsubcctv_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
// pp 237
void trap( Condition c, CC cc, Register s1, Register s2 ) { v8_no_cc(cc); emit_long( op(arith_op) | cond(c) | op3(trap_op3) | rs1(s1) | trapcc(cc) | rs2(s2)); }
void trap( Condition c, CC cc, Register s1, int trapa ) { v8_no_cc(cc); emit_long( op(arith_op) | cond(c) | op3(trap_op3) | rs1(s1) | trapcc(cc) | immed(true) | u_field(trapa, 6, 0)); }
// simple uncond. trap
// pp 239 omit write priv register for now
inline void wry( Register d) { v9_dep(); emit_long( op(arith_op) | rs1(d) | op3(wrreg_op3) | u_field(0, 29, 25)); }
inline void wrccr(Register s) { v9_only(); emit_long( op(arith_op) | rs1(s) | op3(wrreg_op3) | u_field(2, 29, 25)); }
rs1(s) |
inline void wrasi( Register d) { v9_only(); emit_long( op(arith_op) | rs1(d) | op3(wrreg_op3) | u_field(3, 29, 25)); }
inline void wrfprs( Register d) { v9_only(); emit_long( op(arith_op) | rs1(d) | op3(wrreg_op3) | u_field(6, 29, 25)); }
// Creation
#ifdef CHECK_DELAY
#endif
}
// Testing
#ifndef PRODUCT
void test_v9();
void test_v8_onlys();
#endif
};
class RegistersForDebugging : public StackObj {
public:
float f[32];
double d[32];
void print(outputStream* s);
// gen asm code to save regs
static void save_registers(MacroAssembler* a);
// restore global registers in case C code disturbed them
};
// MacroAssembler extends Assembler by a few frequently used macros.
//
// Most of the standard SPARC synthetic ops are defined here.
// Instructions for which a 'better' code sequence exists depending
// on arguments should also go in here.
class MacroAssembler: public Assembler {
protected:
// Support for VM calls
// This is the base routine called by the different versions of call_VM_leaf. The interpreter
// additional registers when doing a VM call).
#ifdef CC_INTERP
#define VIRTUAL
#else
#define VIRTUAL virtual
#endif
VIRTUAL void call_VM_leaf_base(Register thread_cache, address entry_point, int number_of_arguments);
//
// It is imperative that all calls into the VM are handled via the call_VM macros.
// They make sure that the stack linkage is setup correctly. call_VM's correspond
//
// This is the base routine called by the different versions of call_VM. The interpreter
// additional registers when doing a VM call).
//
// A non-volatile java_thread_cache register should be specified so
// that the G2_thread value can be preserved across the call.
// (If java_thread_cache is noreg, then a slow get_thread call
// will re-initialize the G2_thread.) call_VM_base returns the register that contains the
// thread.
//
// If no last_java_sp is specified (noreg) than SP will be used instead.
virtual void call_VM_base(
int number_of_arguments, // the number of arguments (w/o thread) to pop after call
bool check_exception=true // flag which indicates if exception should be checked
);
// This routine should emit JVMTI PopFrame and ForceEarlyReturn handling code.
// The implementation is only non-empty for the InterpreterMacroAssembler,
// as only the interpreter handles and ForceEarlyReturn PopFrame requests.
public:
// Support for NULL-checks
//
// Generates code that causes a NULL OS exception if the content of reg is NULL.
// If the accessed location is M[reg + offset] and the offset is known, provide the
// offset. No explicit code generation is needed if the offset is within a certain
// range (0 <= offset <= page_size).
//
// %%%%%% Currently not done for SPARC
// support for delayed instructions
// branches that use right instruction for v8 vs. v9
inline void br( Condition c, bool a, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
inline void fb( Condition c, bool a, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
// compares register with zero and branches (V9 and V8 instructions)
// Compares a pointer register with zero and branches on (not)null.
// Does a test & branch on 32-bit systems and a register-branch on 64-bit.
inline void bp( Condition c, bool a, CC cc, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
// Branch that tests xcc in LP64 and icc in !LP64
inline void brx( Condition c, bool a, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
// unconditional short branch
// Branch that tests fp condition codes
inline void fbp( Condition c, bool a, CC cc, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
// get PC the best way
// Sparc shorthands(pp 85, V8 manual, pp 289 V9 manual)
// Emits nothing on V8
#ifdef PRODUCT
} else {
}
}
#else
#endif /* PRODUCT */
// Required platform-specific helpers for Label::patch_instructions.
// They _shadow_ the declarations in AbstractAssembler, which are undefined.
#ifndef PRODUCT
#endif
// sethi Macro handles optimizations and relocations
void sethi( intptr_t imm22a, Register d, bool ForceRelocatable = false, RelocationHolder const& rspec = RelocationHolder());
static int worst_case_size_of_set();
// set may be either setsw or setuw (high 32 bits may be zero or sign)
// sign-extend 32 to 64
// Functions for isolating 64 bit atomic swaps for LP64
// cas_ptr will perform cas for 32 bit VM's and casx for 64 bit VM's
#ifdef _LP64
#else
#endif
}
// Functions for isolating 64 bit shifts for LP64
// little-endian
// copy & clear upper word
// clear upper word
// membar psuedo instruction. takes into account target memory model.
// returns if membar generates anything.
// mov pseudo instructions
else assert_not_delayed(); // Put something useful in the delay slot!
}
else nop();
}
// address pseudos: make these names unlike instruction names to avoid confusion
// ring buffer traceable jumps
// argument pseudos:
// handy macros:
}
// --------------------------------------------------
// Functions for isolating 64 bit loads for LP64
// ld_ptr will perform ld for 32 bit VM's and ldx for 64 bit VM's
// st_ptr will perform st for 32 bit VM's and stx for 64 bit VM's
// ld_long will perform ld for 32 bit VM's and ldx for 64 bit VM's
// st_long will perform st for 32 bit VM's and stx for 64 bit VM's
// --------------------------------------------------
public:
// traps as per trap.h (SPARC ABI?)
void breakpoint_trap();
void flush_windows_trap();
void clean_windows_trap();
void get_psr_trap();
void set_psr_trap();
void flush_windows();
// Support for serializing memory accesses between threads
void enter();
void leave();
// Manipulation of C++ bools
// These are idioms to flag the need for care with accessing bools but on
// this platform we assume byte size
// klass oop manipulations if compressed
// oop manipulations
void encode_heap_oop(Register r) {
encode_heap_oop(r, r);
}
void decode_heap_oop(Register r) {
decode_heap_oop(r, r);
}
void encode_heap_oop_not_null(Register r);
void decode_heap_oop_not_null(Register r);
// Support for managing the JavaThread pointer (i.e.; the reference to
// thread-local information).
void get_thread(); // load G2_thread
void verify_thread(); // verify G2_thread contents
// Support for last Java frame (but use call_VM instead where possible)
void reset_last_Java_frame(void);
// Call into the VM.
// Passes the thread pointer (in O0) as a prepended argument.
// Makes sure oop return values are visible to the GC.
void call_VM(Register oop_result, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
void call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions = true);
void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true);
// these overloadings are not presently used on SPARC:
void call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
void call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true);
void call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
void call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true);
void call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2, Register arg_3);
// vm result is currently getting hijacked to for oop preservation
// if call_VM_base was called with check_exceptions=false, then call
// check_and_forward_exception to handle exceptions when it is safe
private:
// For V8
#ifdef ASSERT
// For V8 debugging. Uses V8 instruction sequence and checks
// result with V9 insturctions rdccr and wrccr.
// Uses Gscatch and Gscatch2
#endif // ASSERT
public:
// Stores
void store_check(Register tmp, Register obj); // store check for obj - register is destroyed afterwards
void store_check(Register tmp, Register obj, Register offset); // store check for obj - register is destroyed afterwards
// pushes double TOS element of FPU stack on CPU stack; pops from FPU stack
void push_fTOS();
// pops double TOS element from CPU stack and pushes on FPU stack
void pop_fTOS();
void empty_FPU_stack();
void push_IU_state();
void pop_IU_state();
void push_FPU_state();
void pop_FPU_state();
void push_CPU_state();
void pop_CPU_state();
// if heap base register is used - reinit it with the correct value
void reinit_heapbase();
// Debugging
// only if +VerifyOops
// only if +VerifyFPU
void unimplemented(const char* what = "") { char* b = new char[1024]; sprintf(b, "unimplemented: %s", what); stop(b); }
void print_CPU_state();
// oops in code
// nop padding
// declare a safepoint
void safepoint();
// factor out part of stop into subroutine to save space
void stop_subroutine();
// factor out part of verify_oop into subroutine to save space
void verify_oop_subroutine();
// side-door communication with signalHandler in os_solaris.cpp
#ifndef PRODUCT
static void test();
#endif
// convert an incoming arglist to varargs format; put the pointer in d
int total_frame_size_in_bytes(int extraWords);
// used when extraWords known statically
void save_frame(int extraWords);
void save_frame_c1(int size_in_bytes);
// make a frame, and simultaneously pass up one or two register value
// into the new register window
void save_frame_and_mov(int extraWords, Register s1, Register d1, Register s2 = Register(), Register d2 = Register());
// give no. (outgoing) params, calc # of words will need on frame
// used to calculate frame size dynamically
// result is in bytes and must be negated for save inst
// calc and also save
// implementations of bytecodes used by both interpreter and compiler
#ifdef _LP64
#endif
void save_all_globals_into_locals();
void restore_globals_from_locals();
// These set the icc condition code to equal if the lock succeeded
// and notEqual if it failed and requires a slow case
// Biased locking support
// Upon entry, lock_reg must point to the lock record on the stack,
// obj_reg must contain the target object, and mark_reg must contain
// the target object's header.
// Destroys mark_reg if an attempt is made to bias an anonymously
// biased lock. In this case a failure will go either to the slow
// case or fall through with the notEqual condition code set with
// the expectation that the slow case in the runtime will be called.
// In the fall-through case where the CAS-based lock is done,
// mark_reg is not destroyed.
// Upon entry, the base register of mark_addr must contain the oop.
// Destroys temp_reg.
// If allow_delay_slot_filling is set to true, the next instruction
// emitted after this one will go in an annulled delay slot if the
// biased locking exit case failed.
void biased_locking_exit(Address mark_addr, Register temp_reg, Label& done, bool allow_delay_slot_filling = false);
// allocation
void eden_allocate(
int con_size_in_bytes, // object size in bytes if known at compile time
);
void tlab_allocate(
int con_size_in_bytes, // object size in bytes if known at compile time
);
// Stack overflow checking
// Note: this clobbers G3_scratch
void bang_stack_with_offset(int offset) {
// stack grows down, caller passes positive offset
}
// Writes to stack successive pages until offset reached to check for
// stack overflow + shadow pages. Clobbers tsp and scratch registers.
void verify_tlab();
// Helper functions for statistics gathering.
// Conditionally (non-atomically) increments passed counter address, preserving condition codes.
// Unconditional increment.
};
/**
* class SkipIfEqual:
*
* Instantiating this class will result in assembly code being output that will
* jump around any code emitted between the creation of the instance and it's
* automatic destruction at the end of a scope block, depending on the value of
* the flag passed to the constructor, which will be checked at run-time.
*/
class SkipIfEqual : public StackObj {
private:
public:
// 'temp' is a temp register that this object can use (and trash)
~SkipIfEqual();
};
#ifdef ASSERT
// On RISC, there's no benefit to verifying instruction boundaries.
inline bool AbstractAssembler::pd_check_instruction_mark() { return false; }
#endif