// Copyright 2012 the V8 project authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#include "src/mips64/codegen-mips64.h"
#if V8_TARGET_ARCH_MIPS64
#include <memory>
#include "src/codegen.h"
#include "src/macro-assembler.h"
#include "src/mips64/simulator-mips64.h"
namespace v8 {
namespace internal {
#define __ masm.
#if defined(V8_HOST_ARCH_MIPS)
MemCopyUint8Function CreateMemCopyUint8Function(Isolate* isolate,
MemCopyUint8Function stub) {
#if defined(USE_SIMULATOR)
return stub;
#else
size_t actual_size;
byte* buffer =
static_cast<byte*>(base::OS::Allocate(3 * KB, &actual_size, true));
if (buffer == nullptr) return stub;
// This code assumes that cache lines are 32 bytes and if the cache line is
// larger it will not work correctly.
MacroAssembler masm(isolate, buffer, static_cast<int>(actual_size),
CodeObjectRequired::kNo);
{
Label lastb, unaligned, aligned, chkw,
loop16w, chk1w, wordCopy_loop, skip_pref, lastbloop,
leave, ua_chk16w, ua_loop16w, ua_skip_pref, ua_chkw,
ua_chk1w, ua_wordCopy_loop, ua_smallCopy, ua_smallCopy_loop;
// The size of each prefetch.
uint32_t pref_chunk = 32;
// The maximum size of a prefetch, it must not be less than pref_chunk.
// If the real size of a prefetch is greater than max_pref_size and
// the kPrefHintPrepareForStore hint is used, the code will not work
// correctly.
uint32_t max_pref_size = 128;
DCHECK(pref_chunk < max_pref_size);
// pref_limit is set based on the fact that we never use an offset
// greater then 5 on a store pref and that a single pref can
// never be larger then max_pref_size.
uint32_t pref_limit = (5 * pref_chunk) + max_pref_size;
int32_t pref_hint_load = kPrefHintLoadStreamed;
int32_t pref_hint_store = kPrefHintPrepareForStore;
uint32_t loadstore_chunk = 4;
// The initial prefetches may fetch bytes that are before the buffer being
// copied. Start copies with an offset of 4 so avoid this situation when
// using kPrefHintPrepareForStore.
DCHECK(pref_hint_store != kPrefHintPrepareForStore ||
pref_chunk * 4 >= max_pref_size);
// If the size is less than 8, go to lastb. Regardless of size,
// copy dst pointer to v0 for the retuen value.
__ slti(a6, a2, 2 * loadstore_chunk);
__ bne(a6, zero_reg, &lastb);
__ mov(v0, a0); // In delay slot.
// If src and dst have different alignments, go to unaligned, if they
// have the same alignment (but are not actually aligned) do a partial
// load/store to make them aligned. If they are both already aligned
// we can start copying at aligned.
__ xor_(t8, a1, a0);
__ andi(t8, t8, loadstore_chunk - 1); // t8 is a0/a1 word-displacement.
__ bne(t8, zero_reg, &unaligned);
__ subu(a3, zero_reg, a0); // In delay slot.
__ andi(a3, a3, loadstore_chunk - 1); // Copy a3 bytes to align a0/a1.
__ beq(a3, zero_reg, &aligned); // Already aligned.
__ subu(a2, a2, a3); // In delay slot. a2 is the remining bytes count.
if (kArchEndian == kLittle) {
__ lwr(t8, MemOperand(a1));
__ addu(a1, a1, a3);
__ swr(t8, MemOperand(a0));
__ addu(a0, a0, a3);
} else {
__ lwl(t8, MemOperand(a1));
__ addu(a1, a1, a3);
__ swl(t8, MemOperand(a0));
__ addu(a0, a0, a3);
}
// Now dst/src are both aligned to (word) aligned addresses. Set a2 to
// count how many bytes we have to copy after all the 64 byte chunks are
// copied and a3 to the dst pointer after all the 64 byte chunks have been
// copied. We will loop, incrementing a0 and a1 until a0 equals a3.
__ bind(&aligned);
__ andi(t8, a2, 0x3f);
__ beq(a2, t8, &chkw); // Less than 64?
__ subu(a3, a2, t8); // In delay slot.
__ addu(a3, a0, a3); // Now a3 is the final dst after loop.
// When in the loop we prefetch with kPrefHintPrepareForStore hint,
// in this case the a0+x should be past the "a4-32" address. This means:
// for x=128 the last "safe" a0 address is "a4-160". Alternatively, for
// x=64 the last "safe" a0 address is "a4-96". In the current version we
// will use "pref hint, 128(a0)", so "a4-160" is the limit.
if (pref_hint_store == kPrefHintPrepareForStore) {
__ addu(a4, a0, a2); // a4 is the "past the end" address.
__ Subu(t9, a4, pref_limit); // t9 is the "last safe pref" address.
}
__ Pref(pref_hint_load, MemOperand(a1, 0 * pref_chunk));
__ Pref(pref_hint_load, MemOperand(a1, 1 * pref_chunk));
__ Pref(pref_hint_load, MemOperand(a1, 2 * pref_chunk));
__ Pref(pref_hint_load, MemOperand(a1, 3 * pref_chunk));
if (pref_hint_store != kPrefHintPrepareForStore) {
__ Pref(pref_hint_store, MemOperand(a0, 1 * pref_chunk));
__ Pref(pref_hint_store, MemOperand(a0, 2 * pref_chunk));
__ Pref(pref_hint_store, MemOperand(a0, 3 * pref_chunk));
}
__ bind(&loop16w);
__ lw(a4, MemOperand(a1));
if (pref_hint_store == kPrefHintPrepareForStore) {
__ sltu(v1, t9, a0); // If a0 > t9, don't use next prefetch.
__ Branch(USE_DELAY_SLOT, &skip_pref, gt, v1, Operand(zero_reg));
}
__ lw(a5, MemOperand(a1, 1, loadstore_chunk)); // Maybe in delay slot.
__ Pref(pref_hint_store, MemOperand(a0, 4 * pref_chunk));
__ Pref(pref_hint_store, MemOperand(a0, 5 * pref_chunk));
__ bind(&skip_pref);
__ lw(a6, MemOperand(a1, 2, loadstore_chunk));
__ lw(a7, MemOperand(a1, 3, loadstore_chunk));
__ lw(t0, MemOperand(a1, 4, loadstore_chunk));
__ lw(t1, MemOperand(a1, 5, loadstore_chunk));
__ lw(t2, MemOperand(a1, 6, loadstore_chunk));
__ lw(t3, MemOperand(a1, 7, loadstore_chunk));
__ Pref(pref_hint_load, MemOperand(a1, 4 * pref_chunk));
__ sw(a4, MemOperand(a0));
__ sw(a5, MemOperand(a0, 1, loadstore_chunk));
__ sw(a6, MemOperand(a0, 2, loadstore_chunk));
__ sw(a7, MemOperand(a0, 3, loadstore_chunk));
__ sw(t0, MemOperand(a0, 4, loadstore_chunk));
__ sw(t1, MemOperand(a0, 5, loadstore_chunk));
__ sw(t2, MemOperand(a0, 6, loadstore_chunk));
__ sw(t3, MemOperand(a0, 7, loadstore_chunk));
__ lw(a4, MemOperand(a1, 8, loadstore_chunk));
__ lw(a5, MemOperand(a1, 9, loadstore_chunk));
__ lw(a6, MemOperand(a1, 10, loadstore_chunk));
__ lw(a7, MemOperand(a1, 11, loadstore_chunk));
__ lw(t0, MemOperand(a1, 12, loadstore_chunk));
__ lw(t1, MemOperand(a1, 13, loadstore_chunk));
__ lw(t2, MemOperand(a1, 14, loadstore_chunk));
__ lw(t3, MemOperand(a1, 15, loadstore_chunk));
__ Pref(pref_hint_load, MemOperand(a1, 5 * pref_chunk));
__ sw(a4, MemOperand(a0, 8, loadstore_chunk));
__ sw(a5, MemOperand(a0, 9, loadstore_chunk));
__ sw(a6, MemOperand(a0, 10, loadstore_chunk));
__ sw(a7, MemOperand(a0, 11, loadstore_chunk));
__ sw(t0, MemOperand(a0, 12, loadstore_chunk));
__ sw(t1, MemOperand(a0, 13, loadstore_chunk));
__ sw(t2, MemOperand(a0, 14, loadstore_chunk));
__ sw(t3, MemOperand(a0, 15, loadstore_chunk));
__ addiu(a0, a0, 16 * loadstore_chunk);
__ bne(a0, a3, &loop16w);
__ addiu(a1, a1, 16 * loadstore_chunk); // In delay slot.
__ mov(a2, t8);
// Here we have src and dest word-aligned but less than 64-bytes to go.
// Check for a 32 bytes chunk and copy if there is one. Otherwise jump
// down to chk1w to handle the tail end of the copy.
__ bind(&chkw);
__ Pref(pref_hint_load, MemOperand(a1, 0 * pref_chunk));
__ andi(t8, a2, 0x1f);
__ beq(a2, t8, &chk1w); // Less than 32?
__ nop(); // In delay slot.
__ lw(a4, MemOperand(a1));
__ lw(a5, MemOperand(a1, 1, loadstore_chunk));
__ lw(a6, MemOperand(a1, 2, loadstore_chunk));
__ lw(a7, MemOperand(a1, 3, loadstore_chunk));
__ lw(t0, MemOperand(a1, 4, loadstore_chunk));
__ lw(t1, MemOperand(a1, 5, loadstore_chunk));
__ lw(t2, MemOperand(a1, 6, loadstore_chunk));
__ lw(t3, MemOperand(a1, 7, loadstore_chunk));
__ addiu(a1, a1, 8 * loadstore_chunk);
__ sw(a4, MemOperand(a0));
__ sw(a5, MemOperand(a0, 1, loadstore_chunk));
__ sw(a6, MemOperand(a0, 2, loadstore_chunk));
__ sw(a7, MemOperand(a0, 3, loadstore_chunk));
__ sw(t0, MemOperand(a0, 4, loadstore_chunk));
__ sw(t1, MemOperand(a0, 5, loadstore_chunk));
__ sw(t2, MemOperand(a0, 6, loadstore_chunk));
__ sw(t3, MemOperand(a0, 7, loadstore_chunk));
__ addiu(a0, a0, 8 * loadstore_chunk);
// Here we have less than 32 bytes to copy. Set up for a loop to copy
// one word at a time. Set a2 to count how many bytes we have to copy
// after all the word chunks are copied and a3 to the dst pointer after
// all the word chunks have been copied. We will loop, incrementing a0
// and a1 untill a0 equals a3.
__ bind(&chk1w);
__ andi(a2, t8, loadstore_chunk - 1);
__ beq(a2, t8, &lastb);
__ subu(a3, t8, a2); // In delay slot.
__ addu(a3, a0, a3);
__ bind(&wordCopy_loop);
__ lw(a7, MemOperand(a1));
__ addiu(a0, a0, loadstore_chunk);
__ addiu(a1, a1, loadstore_chunk);
__ bne(a0, a3, &wordCopy_loop);
__ sw(a7, MemOperand(a0, -1, loadstore_chunk)); // In delay slot.
__ bind(&lastb);
__ Branch(&leave, le, a2, Operand(zero_reg));
__ addu(a3, a0, a2);
__ bind(&lastbloop);
__ lb(v1, MemOperand(a1));
__ addiu(a0, a0, 1);
__ addiu(a1, a1, 1);
__ bne(a0, a3, &lastbloop);
__ sb(v1, MemOperand(a0, -1)); // In delay slot.
__ bind(&leave);
__ jr(ra);
__ nop();
// Unaligned case. Only the dst gets aligned so we need to do partial
// loads of the source followed by normal stores to the dst (once we
// have aligned the destination).
__ bind(&unaligned);
__ andi(a3, a3, loadstore_chunk - 1); // Copy a3 bytes to align a0/a1.
__ beq(a3, zero_reg, &ua_chk16w);
__ subu(a2, a2, a3); // In delay slot.
if (kArchEndian == kLittle) {
__ lwr(v1, MemOperand(a1));
__ lwl(v1,
MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
__ addu(a1, a1, a3);
__ swr(v1, MemOperand(a0));
__ addu(a0, a0, a3);
} else {
__ lwl(v1, MemOperand(a1));
__ lwr(v1,
MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
__ addu(a1, a1, a3);
__ swl(v1, MemOperand(a0));
__ addu(a0, a0, a3);
}
// Now the dst (but not the source) is aligned. Set a2 to count how many
// bytes we have to copy after all the 64 byte chunks are copied and a3 to
// the dst pointer after all the 64 byte chunks have been copied. We will
// loop, incrementing a0 and a1 until a0 equals a3.
__ bind(&ua_chk16w);
__ andi(t8, a2, 0x3f);
__ beq(a2, t8, &ua_chkw);
__ subu(a3, a2, t8); // In delay slot.
__ addu(a3, a0, a3);
if (pref_hint_store == kPrefHintPrepareForStore) {
__ addu(a4, a0, a2);
__ Subu(t9, a4, pref_limit);
}
__ Pref(pref_hint_load, MemOperand(a1, 0 * pref_chunk));
__ Pref(pref_hint_load, MemOperand(a1, 1 * pref_chunk));
__ Pref(pref_hint_load, MemOperand(a1, 2 * pref_chunk));
if (pref_hint_store != kPrefHintPrepareForStore) {
__ Pref(pref_hint_store, MemOperand(a0, 1 * pref_chunk));
__ Pref(pref_hint_store, MemOperand(a0, 2 * pref_chunk));
__ Pref(pref_hint_store, MemOperand(a0, 3 * pref_chunk));
}
__ bind(&ua_loop16w);
if (kArchEndian == kLittle) {
__ Pref(pref_hint_load, MemOperand(a1, 3 * pref_chunk));
__ lwr(a4, MemOperand(a1));
__ lwr(a5, MemOperand(a1, 1, loadstore_chunk));
__ lwr(a6, MemOperand(a1, 2, loadstore_chunk));
if (pref_hint_store == kPrefHintPrepareForStore) {
__ sltu(v1, t9, a0);
__ Branch(USE_DELAY_SLOT, &ua_skip_pref, gt, v1, Operand(zero_reg));
}
__ lwr(a7, MemOperand(a1, 3, loadstore_chunk)); // Maybe in delay slot.
__ Pref(pref_hint_store, MemOperand(a0, 4 * pref_chunk));
__ Pref(pref_hint_store, MemOperand(a0, 5 * pref_chunk));
__ bind(&ua_skip_pref);
__ lwr(t0, MemOperand(a1, 4, loadstore_chunk));
__ lwr(t1, MemOperand(a1, 5, loadstore_chunk));
__ lwr(t2, MemOperand(a1, 6, loadstore_chunk));
__ lwr(t3, MemOperand(a1, 7, loadstore_chunk));
__ lwl(a4,
MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(a5,
MemOperand(a1, 2, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(a6,
MemOperand(a1, 3, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(a7,
MemOperand(a1, 4, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(t0,
MemOperand(a1, 5, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(t1,
MemOperand(a1, 6, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(t2,
MemOperand(a1, 7, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(t3,
MemOperand(a1, 8, loadstore_chunk, MemOperand::offset_minus_one));
} else {
__ Pref(pref_hint_load, MemOperand(a1, 3 * pref_chunk));
__ lwl(a4, MemOperand(a1));
__ lwl(a5, MemOperand(a1, 1, loadstore_chunk));
__ lwl(a6, MemOperand(a1, 2, loadstore_chunk));
if (pref_hint_store == kPrefHintPrepareForStore) {
__ sltu(v1, t9, a0);
__ Branch(USE_DELAY_SLOT, &ua_skip_pref, gt, v1, Operand(zero_reg));
}
__ lwl(a7, MemOperand(a1, 3, loadstore_chunk)); // Maybe in delay slot.
__ Pref(pref_hint_store, MemOperand(a0, 4 * pref_chunk));
__ Pref(pref_hint_store, MemOperand(a0, 5 * pref_chunk));
__ bind(&ua_skip_pref);
__ lwl(t0, MemOperand(a1, 4, loadstore_chunk));
__ lwl(t1, MemOperand(a1, 5, loadstore_chunk));
__ lwl(t2, MemOperand(a1, 6, loadstore_chunk));
__ lwl(t3, MemOperand(a1, 7, loadstore_chunk));
__ lwr(a4,
MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(a5,
MemOperand(a1, 2, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(a6,
MemOperand(a1, 3, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(a7,
MemOperand(a1, 4, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(t0,
MemOperand(a1, 5, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(t1,
MemOperand(a1, 6, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(t2,
MemOperand(a1, 7, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(t3,
MemOperand(a1, 8, loadstore_chunk, MemOperand::offset_minus_one));
}
__ Pref(pref_hint_load, MemOperand(a1, 4 * pref_chunk));
__ sw(a4, MemOperand(a0));
__ sw(a5, MemOperand(a0, 1, loadstore_chunk));
__ sw(a6, MemOperand(a0, 2, loadstore_chunk));
__ sw(a7, MemOperand(a0, 3, loadstore_chunk));
__ sw(t0, MemOperand(a0, 4, loadstore_chunk));
__ sw(t1, MemOperand(a0, 5, loadstore_chunk));
__ sw(t2, MemOperand(a0, 6, loadstore_chunk));
__ sw(t3, MemOperand(a0, 7, loadstore_chunk));
if (kArchEndian == kLittle) {
__ lwr(a4, MemOperand(a1, 8, loadstore_chunk));
__ lwr(a5, MemOperand(a1, 9, loadstore_chunk));
__ lwr(a6, MemOperand(a1, 10, loadstore_chunk));
__ lwr(a7, MemOperand(a1, 11, loadstore_chunk));
__ lwr(t0, MemOperand(a1, 12, loadstore_chunk));
__ lwr(t1, MemOperand(a1, 13, loadstore_chunk));
__ lwr(t2, MemOperand(a1, 14, loadstore_chunk));
__ lwr(t3, MemOperand(a1, 15, loadstore_chunk));
__ lwl(a4,
MemOperand(a1, 9, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(a5,
MemOperand(a1, 10, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(a6,
MemOperand(a1, 11, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(a7,
MemOperand(a1, 12, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(t0,
MemOperand(a1, 13, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(t1,
MemOperand(a1, 14, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(t2,
MemOperand(a1, 15, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(t3,
MemOperand(a1, 16, loadstore_chunk, MemOperand::offset_minus_one));
} else {
__ lwl(a4, MemOperand(a1, 8, loadstore_chunk));
__ lwl(a5, MemOperand(a1, 9, loadstore_chunk));
__ lwl(a6, MemOperand(a1, 10, loadstore_chunk));
__ lwl(a7, MemOperand(a1, 11, loadstore_chunk));
__ lwl(t0, MemOperand(a1, 12, loadstore_chunk));
__ lwl(t1, MemOperand(a1, 13, loadstore_chunk));
__ lwl(t2, MemOperand(a1, 14, loadstore_chunk));
__ lwl(t3, MemOperand(a1, 15, loadstore_chunk));
__ lwr(a4,
MemOperand(a1, 9, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(a5,
MemOperand(a1, 10, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(a6,
MemOperand(a1, 11, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(a7,
MemOperand(a1, 12, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(t0,
MemOperand(a1, 13, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(t1,
MemOperand(a1, 14, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(t2,
MemOperand(a1, 15, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(t3,
MemOperand(a1, 16, loadstore_chunk, MemOperand::offset_minus_one));
}
__ Pref(pref_hint_load, MemOperand(a1, 5 * pref_chunk));
__ sw(a4, MemOperand(a0, 8, loadstore_chunk));
__ sw(a5, MemOperand(a0, 9, loadstore_chunk));
__ sw(a6, MemOperand(a0, 10, loadstore_chunk));
__ sw(a7, MemOperand(a0, 11, loadstore_chunk));
__ sw(t0, MemOperand(a0, 12, loadstore_chunk));
__ sw(t1, MemOperand(a0, 13, loadstore_chunk));
__ sw(t2, MemOperand(a0, 14, loadstore_chunk));
__ sw(t3, MemOperand(a0, 15, loadstore_chunk));
__ addiu(a0, a0, 16 * loadstore_chunk);
__ bne(a0, a3, &ua_loop16w);
__ addiu(a1, a1, 16 * loadstore_chunk); // In delay slot.
__ mov(a2, t8);
// Here less than 64-bytes. Check for
// a 32 byte chunk and copy if there is one. Otherwise jump down to
// ua_chk1w to handle the tail end of the copy.
__ bind(&ua_chkw);
__ Pref(pref_hint_load, MemOperand(a1));
__ andi(t8, a2, 0x1f);
__ beq(a2, t8, &ua_chk1w);
__ nop(); // In delay slot.
if (kArchEndian == kLittle) {
__ lwr(a4, MemOperand(a1));
__ lwr(a5, MemOperand(a1, 1, loadstore_chunk));
__ lwr(a6, MemOperand(a1, 2, loadstore_chunk));
__ lwr(a7, MemOperand(a1, 3, loadstore_chunk));
__ lwr(t0, MemOperand(a1, 4, loadstore_chunk));
__ lwr(t1, MemOperand(a1, 5, loadstore_chunk));
__ lwr(t2, MemOperand(a1, 6, loadstore_chunk));
__ lwr(t3, MemOperand(a1, 7, loadstore_chunk));
__ lwl(a4,
MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(a5,
MemOperand(a1, 2, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(a6,
MemOperand(a1, 3, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(a7,
MemOperand(a1, 4, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(t0,
MemOperand(a1, 5, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(t1,
MemOperand(a1, 6, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(t2,
MemOperand(a1, 7, loadstore_chunk, MemOperand::offset_minus_one));
__ lwl(t3,
MemOperand(a1, 8, loadstore_chunk, MemOperand::offset_minus_one));
} else {
__ lwl(a4, MemOperand(a1));
__ lwl(a5, MemOperand(a1, 1, loadstore_chunk));
__ lwl(a6, MemOperand(a1, 2, loadstore_chunk));
__ lwl(a7, MemOperand(a1, 3, loadstore_chunk));
__ lwl(t0, MemOperand(a1, 4, loadstore_chunk));
__ lwl(t1, MemOperand(a1, 5, loadstore_chunk));
__ lwl(t2, MemOperand(a1, 6, loadstore_chunk));
__ lwl(t3, MemOperand(a1, 7, loadstore_chunk));
__ lwr(a4,
MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(a5,
MemOperand(a1, 2, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(a6,
MemOperand(a1, 3, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(a7,
MemOperand(a1, 4, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(t0,
MemOperand(a1, 5, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(t1,
MemOperand(a1, 6, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(t2,
MemOperand(a1, 7, loadstore_chunk, MemOperand::offset_minus_one));
__ lwr(t3,
MemOperand(a1, 8, loadstore_chunk, MemOperand::offset_minus_one));
}
__ addiu(a1, a1, 8 * loadstore_chunk);
__ sw(a4, MemOperand(a0));
__ sw(a5, MemOperand(a0, 1, loadstore_chunk));
__ sw(a6, MemOperand(a0, 2, loadstore_chunk));
__ sw(a7, MemOperand(a0, 3, loadstore_chunk));
__ sw(t0, MemOperand(a0, 4, loadstore_chunk));
__ sw(t1, MemOperand(a0, 5, loadstore_chunk));
__ sw(t2, MemOperand(a0, 6, loadstore_chunk));
__ sw(t3, MemOperand(a0, 7, loadstore_chunk));
__ addiu(a0, a0, 8 * loadstore_chunk);
// Less than 32 bytes to copy. Set up for a loop to
// copy one word at a time.
__ bind(&ua_chk1w);
__ andi(a2, t8, loadstore_chunk - 1);
__ beq(a2, t8, &ua_smallCopy);
__ subu(a3, t8, a2); // In delay slot.
__ addu(a3, a0, a3);
__ bind(&ua_wordCopy_loop);
if (kArchEndian == kLittle) {
__ lwr(v1, MemOperand(a1));
__ lwl(v1,
MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
} else {
__ lwl(v1, MemOperand(a1));
__ lwr(v1,
MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one));
}
__ addiu(a0, a0, loadstore_chunk);
__ addiu(a1, a1, loadstore_chunk);
__ bne(a0, a3, &ua_wordCopy_loop);
__ sw(v1, MemOperand(a0, -1, loadstore_chunk)); // In delay slot.
// Copy the last 8 bytes.
__ bind(&ua_smallCopy);
__ beq(a2, zero_reg, &leave);
__ addu(a3, a0, a2); // In delay slot.
__ bind(&ua_smallCopy_loop);
__ lb(v1, MemOperand(a1));
__ addiu(a0, a0, 1);
__ addiu(a1, a1, 1);
__ bne(a0, a3, &ua_smallCopy_loop);
__ sb(v1, MemOperand(a0, -1)); // In delay slot.
__ jr(ra);
__ nop();
}
CodeDesc desc;
masm.GetCode(&desc);
DCHECK(!RelocInfo::RequiresRelocation(desc));
Assembler::FlushICache(isolate, buffer, actual_size);
base::OS::ProtectCode(buffer, actual_size);
return FUNCTION_CAST<MemCopyUint8Function>(buffer);
#endif
}
#endif
UnaryMathFunctionWithIsolate CreateSqrtFunction(Isolate* isolate) {
#if defined(USE_SIMULATOR)
return nullptr;
#else
size_t actual_size;
byte* buffer =
static_cast<byte*>(base::OS::Allocate(1 * KB, &actual_size, true));
if (buffer == nullptr) return nullptr;
MacroAssembler masm(isolate, buffer, static_cast<int>(actual_size),
CodeObjectRequired::kNo);
__ MovFromFloatParameter(f12);
__ sqrt_d(f0, f12);
__ MovToFloatResult(f0);
__ Ret();
CodeDesc desc;
masm.GetCode(&desc);
DCHECK(!RelocInfo::RequiresRelocation(desc));
Assembler::FlushICache(isolate, buffer, actual_size);
base::OS::ProtectCode(buffer, actual_size);
return FUNCTION_CAST<UnaryMathFunctionWithIsolate>(buffer);
#endif
}
#undef __
// -------------------------------------------------------------------------
// Platform-specific RuntimeCallHelper functions.
void StubRuntimeCallHelper::BeforeCall(MacroAssembler* masm) const {
masm->EnterFrame(StackFrame::INTERNAL);
DCHECK(!masm->has_frame());
masm->set_has_frame(true);
}
void StubRuntimeCallHelper::AfterCall(MacroAssembler* masm) const {
masm->LeaveFrame(StackFrame::INTERNAL);
DCHECK(masm->has_frame());
masm->set_has_frame(false);
}
// -------------------------------------------------------------------------
// Code generators
#define __ ACCESS_MASM(masm)
void ElementsTransitionGenerator::GenerateMapChangeElementsTransition(
MacroAssembler* masm,
Register receiver,
Register key,
Register value,
Register target_map,
AllocationSiteMode mode,
Label* allocation_memento_found) {
Register scratch_elements = a4;
DCHECK(!AreAliased(receiver, key, value, target_map,
scratch_elements));
if (mode == TRACK_ALLOCATION_SITE) {
__ JumpIfJSArrayHasAllocationMemento(
receiver, scratch_elements, allocation_memento_found);
}
// Set transitioned map.
__ sd(target_map, FieldMemOperand(receiver, HeapObject::kMapOffset));
__ RecordWriteField(receiver,
HeapObject::kMapOffset,
target_map,
t1,
kRAHasNotBeenSaved,
kDontSaveFPRegs,
EMIT_REMEMBERED_SET,
OMIT_SMI_CHECK);
}
void ElementsTransitionGenerator::GenerateSmiToDouble(
MacroAssembler* masm,
Register receiver,
Register key,
Register value,
Register target_map,
AllocationSiteMode mode,
Label* fail) {
// Register ra contains the return address.
Label loop, entry, convert_hole, gc_required, only_change_map, done;
Register elements = a4;
Register length = a5;
Register array = a6;
Register array_end = array;
// target_map parameter can be clobbered.
Register scratch1 = target_map;
Register scratch2 = t1;
Register scratch3 = a7;
// Verify input registers don't conflict with locals.
DCHECK(!AreAliased(receiver, key, value, target_map,
elements, length, array, scratch2));
Register scratch = t2;
if (mode == TRACK_ALLOCATION_SITE) {
__ JumpIfJSArrayHasAllocationMemento(receiver, elements, fail);
}
// Check for empty arrays, which only require a map transition and no changes
// to the backing store.
__ ld(elements, FieldMemOperand(receiver, JSObject::kElementsOffset));
__ LoadRoot(at, Heap::kEmptyFixedArrayRootIndex);
__ Branch(&only_change_map, eq, at, Operand(elements));
__ push(ra);
__ ld(length, FieldMemOperand(elements, FixedArray::kLengthOffset));
// elements: source FixedArray
// length: number of elements (smi-tagged)
// Allocate new FixedDoubleArray.
__ SmiScale(scratch, length, kDoubleSizeLog2);
__ Daddu(scratch, scratch, FixedDoubleArray::kHeaderSize);
__ Allocate(scratch, array, t3, scratch2, &gc_required, DOUBLE_ALIGNMENT);
__ Dsubu(array, array, kHeapObjectTag);
// array: destination FixedDoubleArray, not tagged as heap object
// Set destination FixedDoubleArray's length and map.
__ LoadRoot(scratch2, Heap::kFixedDoubleArrayMapRootIndex);
__ sd(length, MemOperand(array, FixedDoubleArray::kLengthOffset));
// Update receiver's map.
__ sd(scratch2, MemOperand(array, HeapObject::kMapOffset));
__ sd(target_map, FieldMemOperand(receiver, HeapObject::kMapOffset));
__ RecordWriteField(receiver,
HeapObject::kMapOffset,
target_map,
scratch2,
kRAHasBeenSaved,
kDontSaveFPRegs,
OMIT_REMEMBERED_SET,
OMIT_SMI_CHECK);
// Replace receiver's backing store with newly created FixedDoubleArray.
__ Daddu(scratch1, array, Operand(kHeapObjectTag));
__ sd(scratch1, FieldMemOperand(receiver, JSObject::kElementsOffset));
__ RecordWriteField(receiver,
JSObject::kElementsOffset,
scratch1,
scratch2,
kRAHasBeenSaved,
kDontSaveFPRegs,
EMIT_REMEMBERED_SET,
OMIT_SMI_CHECK);
// Prepare for conversion loop.
__ Daddu(scratch1, elements,
Operand(FixedArray::kHeaderSize - kHeapObjectTag));
__ Daddu(scratch3, array, Operand(FixedDoubleArray::kHeaderSize));
__ SmiScale(array_end, length, kDoubleSizeLog2);
__ Daddu(array_end, array_end, scratch3);
// Repurpose registers no longer in use.
Register hole_lower = elements;
Register hole_upper = length;
__ li(hole_lower, Operand(kHoleNanLower32));
__ li(hole_upper, Operand(kHoleNanUpper32));
// scratch1: begin of source FixedArray element fields, not tagged
// hole_lower: kHoleNanLower32
// hole_upper: kHoleNanUpper32
// array_end: end of destination FixedDoubleArray, not tagged
// scratch3: begin of FixedDoubleArray element fields, not tagged
__ Branch(&entry);
__ bind(&only_change_map);
__ sd(target_map, FieldMemOperand(receiver, HeapObject::kMapOffset));
__ RecordWriteField(receiver,
HeapObject::kMapOffset,
target_map,
scratch2,
kRAHasBeenSaved,
kDontSaveFPRegs,
OMIT_REMEMBERED_SET,
OMIT_SMI_CHECK);
__ Branch(&done);
// Call into runtime if GC is required.
__ bind(&gc_required);
__ ld(ra, MemOperand(sp, 0));
__ Branch(USE_DELAY_SLOT, fail);
__ daddiu(sp, sp, kPointerSize); // In delay slot.
// Convert and copy elements.
__ bind(&loop);
__ ld(scratch2, MemOperand(scratch1));
__ Daddu(scratch1, scratch1, kPointerSize);
// scratch2: current element
__ JumpIfNotSmi(scratch2, &convert_hole);
__ SmiUntag(scratch2);
// Normal smi, convert to double and store.
__ mtc1(scratch2, f0);
__ cvt_d_w(f0, f0);
__ sdc1(f0, MemOperand(scratch3));
__ Branch(USE_DELAY_SLOT, &entry);
__ daddiu(scratch3, scratch3, kDoubleSize); // In delay slot.
// Hole found, store the-hole NaN.
__ bind(&convert_hole);
if (FLAG_debug_code) {
// Restore a "smi-untagged" heap object.
__ Or(scratch2, scratch2, Operand(1));
__ LoadRoot(at, Heap::kTheHoleValueRootIndex);
__ Assert(eq, kObjectFoundInSmiOnlyArray, at, Operand(scratch2));
}
// mantissa
__ sw(hole_lower, MemOperand(scratch3, Register::kMantissaOffset));
// exponent
__ sw(hole_upper, MemOperand(scratch3, Register::kExponentOffset));
__ Daddu(scratch3, scratch3, kDoubleSize);
__ bind(&entry);
__ Branch(&loop, lt, scratch3, Operand(array_end));
__ bind(&done);
__ pop(ra);
}
void ElementsTransitionGenerator::GenerateDoubleToObject(
MacroAssembler* masm,
Register receiver,
Register key,
Register value,
Register target_map,
AllocationSiteMode mode,
Label* fail) {
// Register ra contains the return address.
Label entry, loop, convert_hole, gc_required, only_change_map;
Register elements = a4;
Register array = a6;
Register length = a5;
Register scratch = t1;
// Verify input registers don't conflict with locals.
DCHECK(!AreAliased(receiver, key, value, target_map,
elements, array, length, scratch));
if (mode == TRACK_ALLOCATION_SITE) {
__ JumpIfJSArrayHasAllocationMemento(receiver, elements, fail);
}
// Check for empty arrays, which only require a map transition and no changes
// to the backing store.
__ ld(elements, FieldMemOperand(receiver, JSObject::kElementsOffset));
__ LoadRoot(at, Heap::kEmptyFixedArrayRootIndex);
__ Branch(&only_change_map, eq, at, Operand(elements));
__ MultiPush(
value.bit() | key.bit() | receiver.bit() | target_map.bit() | ra.bit());
__ ld(length, FieldMemOperand(elements, FixedArray::kLengthOffset));
// elements: source FixedArray
// length: number of elements (smi-tagged)
// Allocate new FixedArray.
// Re-use value and target_map registers, as they have been saved on the
// stack.
Register array_size = value;
Register allocate_scratch = target_map;
__ SmiScale(array_size, length, kPointerSizeLog2);
__ Daddu(array_size, array_size, FixedDoubleArray::kHeaderSize);
__ Allocate(array_size, array, allocate_scratch, scratch, &gc_required,
NO_ALLOCATION_FLAGS);
__ Dsubu(array, array, kHeapObjectTag);
// array: destination FixedArray, not tagged as heap object
// Set destination FixedDoubleArray's length and map.
__ LoadRoot(scratch, Heap::kFixedArrayMapRootIndex);
__ sd(length, MemOperand(array, FixedDoubleArray::kLengthOffset));
__ sd(scratch, MemOperand(array, HeapObject::kMapOffset));
// Prepare for conversion loop.
Register src_elements = elements;
Register dst_elements = target_map;
Register dst_end = length;
Register heap_number_map = scratch;
__ Daddu(src_elements, src_elements,
Operand(FixedDoubleArray::kHeaderSize - kHeapObjectTag));
__ Daddu(dst_elements, array, Operand(FixedArray::kHeaderSize));
__ SmiScale(dst_end, dst_end, kPointerSizeLog2);
__ Daddu(dst_end, dst_elements, dst_end);
// Allocating heap numbers in the loop below can fail and cause a jump to
// gc_required. We can't leave a partly initialized FixedArray behind,
// so pessimistically fill it with holes now.
Label initialization_loop, initialization_loop_entry;
__ LoadRoot(scratch, Heap::kTheHoleValueRootIndex);
__ Branch(&initialization_loop_entry);
__ bind(&initialization_loop);
__ sd(scratch, MemOperand(dst_elements));
__ Daddu(dst_elements, dst_elements, Operand(kPointerSize));
__ bind(&initialization_loop_entry);
__ Branch(&initialization_loop, lt, dst_elements, Operand(dst_end));
__ Daddu(dst_elements, array, Operand(FixedArray::kHeaderSize));
__ Daddu(array, array, Operand(kHeapObjectTag));
__ LoadRoot(heap_number_map, Heap::kHeapNumberMapRootIndex);
// Using offsetted addresses.
// dst_elements: begin of destination FixedArray element fields, not tagged
// src_elements: begin of source FixedDoubleArray element fields, not tagged,
// points to the exponent
// dst_end: end of destination FixedArray, not tagged
// array: destination FixedArray
// heap_number_map: heap number map
__ Branch(&entry);
// Call into runtime if GC is required.
__ bind(&gc_required);
__ MultiPop(
value.bit() | key.bit() | receiver.bit() | target_map.bit() | ra.bit());
__ Branch(fail);
__ bind(&loop);
Register upper_bits = key;
__ lw(upper_bits, MemOperand(src_elements, Register::kExponentOffset));
__ Daddu(src_elements, src_elements, kDoubleSize);
// upper_bits: current element's upper 32 bit
// src_elements: address of next element
__ Branch(&convert_hole, eq, a1, Operand(kHoleNanUpper32));
// Non-hole double, copy value into a heap number.
Register heap_number = receiver;
Register scratch2 = value;
Register scratch3 = t2;
__ AllocateHeapNumber(heap_number, scratch2, scratch3, heap_number_map,
&gc_required);
// heap_number: new heap number
// Load current element, src_elements point to next element.
__ ld(scratch2, MemOperand(src_elements, -kDoubleSize));
__ sd(scratch2, FieldMemOperand(heap_number, HeapNumber::kValueOffset));
__ mov(scratch2, dst_elements);
__ sd(heap_number, MemOperand(dst_elements));
__ Daddu(dst_elements, dst_elements, kPointerSize);
__ RecordWrite(array,
scratch2,
heap_number,
kRAHasBeenSaved,
kDontSaveFPRegs,
EMIT_REMEMBERED_SET,
OMIT_SMI_CHECK);
__ Branch(&entry);
// Replace the-hole NaN with the-hole pointer.
__ bind(&convert_hole);
__ LoadRoot(scratch2, Heap::kTheHoleValueRootIndex);
__ sd(scratch2, MemOperand(dst_elements));
__ Daddu(dst_elements, dst_elements, kPointerSize);
__ bind(&entry);
__ Branch(&loop, lt, dst_elements, Operand(dst_end));
__ MultiPop(receiver.bit() | target_map.bit() | value.bit() | key.bit());
// Replace receiver's backing store with newly created and filled FixedArray.
__ sd(array, FieldMemOperand(receiver, JSObject::kElementsOffset));
__ RecordWriteField(receiver,
JSObject::kElementsOffset,
array,
scratch,
kRAHasBeenSaved,
kDontSaveFPRegs,
EMIT_REMEMBERED_SET,
OMIT_SMI_CHECK);
__ pop(ra);
__ bind(&only_change_map);
// Update receiver's map.
__ sd(target_map, FieldMemOperand(receiver, HeapObject::kMapOffset));
__ RecordWriteField(receiver,
HeapObject::kMapOffset,
target_map,
scratch,
kRAHasNotBeenSaved,
kDontSaveFPRegs,
OMIT_REMEMBERED_SET,
OMIT_SMI_CHECK);
}
void StringCharLoadGenerator::Generate(MacroAssembler* masm,
Register string,
Register index,
Register result,
Label* call_runtime) {
// Fetch the instance type of the receiver into result register.
__ ld(result, FieldMemOperand(string, HeapObject::kMapOffset));
__ lbu(result, FieldMemOperand(result, Map::kInstanceTypeOffset));
// We need special handling for indirect strings.
Label check_sequential;
__ And(at, result, Operand(kIsIndirectStringMask));
__ Branch(&check_sequential, eq, at, Operand(zero_reg));
// Dispatch on the indirect string shape: slice or cons.
Label cons_string;
__ And(at, result, Operand(kSlicedNotConsMask));
__ Branch(&cons_string, eq, at, Operand(zero_reg));
// Handle slices.
Label indirect_string_loaded;
__ ld(result, FieldMemOperand(string, SlicedString::kOffsetOffset));
__ ld(string, FieldMemOperand(string, SlicedString::kParentOffset));
__ dsra32(at, result, 0);
__ Daddu(index, index, at);
__ jmp(&indirect_string_loaded);
// Handle cons strings.
// Check whether the right hand side is the empty string (i.e. if
// this is really a flat string in a cons string). If that is not
// the case we would rather go to the runtime system now to flatten
// the string.
__ bind(&cons_string);
__ ld(result, FieldMemOperand(string, ConsString::kSecondOffset));
__ LoadRoot(at, Heap::kempty_stringRootIndex);
__ Branch(call_runtime, ne, result, Operand(at));
// Get the first of the two strings and load its instance type.
__ ld(string, FieldMemOperand(string, ConsString::kFirstOffset));
__ bind(&indirect_string_loaded);
__ ld(result, FieldMemOperand(string, HeapObject::kMapOffset));
__ lbu(result, FieldMemOperand(result, Map::kInstanceTypeOffset));
// Distinguish sequential and external strings. Only these two string
// representations can reach here (slices and flat cons strings have been
// reduced to the underlying sequential or external string).
Label external_string, check_encoding;
__ bind(&check_sequential);
STATIC_ASSERT(kSeqStringTag == 0);
__ And(at, result, Operand(kStringRepresentationMask));
__ Branch(&external_string, ne, at, Operand(zero_reg));
// Prepare sequential strings
STATIC_ASSERT(SeqTwoByteString::kHeaderSize == SeqOneByteString::kHeaderSize);
__ Daddu(string,
string,
SeqTwoByteString::kHeaderSize - kHeapObjectTag);
__ jmp(&check_encoding);
// Handle external strings.
__ bind(&external_string);
if (FLAG_debug_code) {
// Assert that we do not have a cons or slice (indirect strings) here.
// Sequential strings have already been ruled out.
__ And(at, result, Operand(kIsIndirectStringMask));
__ Assert(eq, kExternalStringExpectedButNotFound,
at, Operand(zero_reg));
}
// Rule out short external strings.
STATIC_ASSERT(kShortExternalStringTag != 0);
__ And(at, result, Operand(kShortExternalStringMask));
__ Branch(call_runtime, ne, at, Operand(zero_reg));
__ ld(string, FieldMemOperand(string, ExternalString::kResourceDataOffset));
Label one_byte, done;
__ bind(&check_encoding);
STATIC_ASSERT(kTwoByteStringTag == 0);
__ And(at, result, Operand(kStringEncodingMask));
__ Branch(&one_byte, ne, at, Operand(zero_reg));
// Two-byte string.
__ Dlsa(at, string, index, 1);
__ lhu(result, MemOperand(at));
__ jmp(&done);
__ bind(&one_byte);
// One_byte string.
__ Daddu(at, string, index);
__ lbu(result, MemOperand(at));
__ bind(&done);
}
#ifdef DEBUG
// nop(CODE_AGE_MARKER_NOP)
static const uint32_t kCodeAgePatchFirstInstruction = 0x00010180;
#endif
CodeAgingHelper::CodeAgingHelper(Isolate* isolate) {
USE(isolate);
DCHECK(young_sequence_.length() == kNoCodeAgeSequenceLength);
// Since patcher is a large object, allocate it dynamically when needed,
// to avoid overloading the stack in stress conditions.
// DONT_FLUSH is used because the CodeAgingHelper is initialized early in
// the process, before MIPS simulator ICache is setup.
std::unique_ptr<CodePatcher> patcher(
new CodePatcher(isolate, young_sequence_.start(),
young_sequence_.length() / Assembler::kInstrSize,
CodePatcher::DONT_FLUSH));
PredictableCodeSizeScope scope(patcher->masm(), young_sequence_.length());
patcher->masm()->PushStandardFrame(a1);
patcher->masm()->nop(Assembler::CODE_AGE_SEQUENCE_NOP);
patcher->masm()->nop(Assembler::CODE_AGE_SEQUENCE_NOP);
patcher->masm()->nop(Assembler::CODE_AGE_SEQUENCE_NOP);
}
#ifdef DEBUG
bool CodeAgingHelper::IsOld(byte* candidate) const {
return Memory::uint32_at(candidate) == kCodeAgePatchFirstInstruction;
}
#endif
bool Code::IsYoungSequence(Isolate* isolate, byte* sequence) {
bool result = isolate->code_aging_helper()->IsYoung(sequence);
DCHECK(result || isolate->code_aging_helper()->IsOld(sequence));
return result;
}
void Code::GetCodeAgeAndParity(Isolate* isolate, byte* sequence, Age* age,
MarkingParity* parity) {
if (IsYoungSequence(isolate, sequence)) {
*age = kNoAgeCodeAge;
*parity = NO_MARKING_PARITY;
} else {
Address target_address = Assembler::target_address_at(
sequence + Assembler::kInstrSize);
Code* stub = GetCodeFromTargetAddress(target_address);
GetCodeAgeAndParity(stub, age, parity);
}
}
void Code::PatchPlatformCodeAge(Isolate* isolate,
byte* sequence,
Code::Age age,
MarkingParity parity) {
uint32_t young_length = isolate->code_aging_helper()->young_sequence_length();
if (age == kNoAgeCodeAge) {
isolate->code_aging_helper()->CopyYoungSequenceTo(sequence);
Assembler::FlushICache(isolate, sequence, young_length);
} else {
Code* stub = GetCodeAgeStub(isolate, age, parity);
CodePatcher patcher(isolate, sequence,
young_length / Assembler::kInstrSize);
// Mark this code sequence for FindPlatformCodeAgeSequence().
patcher.masm()->nop(Assembler::CODE_AGE_MARKER_NOP);
// Load the stub address to t9 and call it,
// GetCodeAgeAndParity() extracts the stub address from this instruction.
patcher.masm()->li(
t9,
Operand(reinterpret_cast<uint64_t>(stub->instruction_start())),
ADDRESS_LOAD);
patcher.masm()->nop(); // Prevent jalr to jal optimization.
patcher.masm()->jalr(t9, a0);
patcher.masm()->nop(); // Branch delay slot nop.
patcher.masm()->nop(); // Pad the empty space.
}
}
#undef __
} // namespace internal
} // namespace v8
#endif // V8_TARGET_ARCH_MIPS64