5523 #endif
5524 BIND(NEXT_1);
5525 ldrh(tmp1, Address(post(src, 2)));
5526 tst(tmp1, 0xff00);
5527 br(NE, DONE);
5528 strb(tmp1, Address(post(dst, 1)));
5529 subs(len, len, 1);
5530 br(GT, NEXT_1);
5531
5532 BIND(DONE);
5533 sub(result, result, len); // Return index where we stopped
5534 // Return len == 0 if we processed all
5535 // characters
5536 }
5537
5538
5539 // Inflate byte[] array to char[].
5540 void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
5541 FloatRegister vtmp1, FloatRegister vtmp2, FloatRegister vtmp3,
5542 Register tmp4) {
5543 Label big, done;
5544
5545 assert_different_registers(src, dst, len, tmp4, rscratch1);
5546
5547 fmovd(vtmp1 , zr);
5548 lsrw(rscratch1, len, 3);
5549
5550 cbnzw(rscratch1, big);
5551
5552 // Short string: less than 8 bytes.
5553 {
5554 Label loop, around, tiny;
5555
5556 subsw(len, len, 4);
5557 andw(len, len, 3);
5558 br(LO, tiny);
5559
5560 // Use SIMD to do 4 bytes.
5561 ldrs(vtmp2, post(src, 4));
5562 zip1(vtmp3, T8B, vtmp2, vtmp1);
5563 strd(vtmp3, post(dst, 8));
5564
5565 cbzw(len, done);
5566
5567 // Do the remaining bytes by steam.
5568 bind(loop);
5569 ldrb(tmp4, post(src, 1));
5570 strh(tmp4, post(dst, 2));
5571 subw(len, len, 1);
5572
5573 bind(tiny);
5574 cbnz(len, loop);
5575
5576 bind(around);
5577 b(done);
5578 }
5579
5580 // Unpack the bytes 8 at a time.
5581 bind(big);
5582 andw(len, len, 7);
5583
5584 {
5585 Label loop, around;
5586
5587 bind(loop);
5588 ldrd(vtmp2, post(src, 8));
5589 sub(rscratch1, rscratch1, 1);
5590 zip1(vtmp3, T16B, vtmp2, vtmp1);
5591 st1(vtmp3, T8H, post(dst, 16));
5592 cbnz(rscratch1, loop);
5593
5594 bind(around);
5595 }
5596
5597 // Do the tail of up to 8 bytes.
5598 sub(src, src, 8);
5599 add(src, src, len, ext::uxtw, 0);
5600 ldrd(vtmp2, Address(src));
5601 sub(dst, dst, 16);
5602 add(dst, dst, len, ext::uxtw, 1);
5603 zip1(vtmp3, T16B, vtmp2, vtmp1);
5604 st1(vtmp3, T8H, Address(dst));
5605
5606 bind(done);
5607 }
5608
5609 // Compress char[] array to byte[].
5610 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
5611 FloatRegister tmp1Reg, FloatRegister tmp2Reg,
5612 FloatRegister tmp3Reg, FloatRegister tmp4Reg,
5613 Register result) {
5614 encode_iso_array(src, dst, len, result,
5615 tmp1Reg, tmp2Reg, tmp3Reg, tmp4Reg);
5616 cmp(len, zr);
5617 csel(result, result, zr, EQ);
5618 }
5619
5620 // get_thread() can be called anywhere inside generated code so we
5621 // need to save whatever non-callee save context might get clobbered
5622 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
5623 // the call setup code.
5624 //
|
5523 #endif
5524 BIND(NEXT_1);
5525 ldrh(tmp1, Address(post(src, 2)));
5526 tst(tmp1, 0xff00);
5527 br(NE, DONE);
5528 strb(tmp1, Address(post(dst, 1)));
5529 subs(len, len, 1);
5530 br(GT, NEXT_1);
5531
5532 BIND(DONE);
5533 sub(result, result, len); // Return index where we stopped
5534 // Return len == 0 if we processed all
5535 // characters
5536 }
5537
5538
5539 // Inflate byte[] array to char[].
5540 void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
5541 FloatRegister vtmp1, FloatRegister vtmp2, FloatRegister vtmp3,
5542 Register tmp4) {
5543 Label big, done, after_init, to_stub;
5544
5545 assert_different_registers(src, dst, len, tmp4, rscratch1);
5546
5547 fmovd(vtmp1, zr);
5548 lsrw(tmp4, len, 3);
5549 bind(after_init);
5550 cbnzw(tmp4, big);
5551 // Short string: less than 8 bytes.
5552 {
5553 Label loop, tiny;
5554
5555 cmpw(len, 4);
5556 br(LT, tiny);
5557 // Use SIMD to do 4 bytes.
5558 ldrs(vtmp2, post(src, 4));
5559 zip1(vtmp3, T8B, vtmp2, vtmp1);
5560 subw(len, len, 4);
5561 strd(vtmp3, post(dst, 8));
5562
5563 cbzw(len, done);
5564
5565 // Do the remaining bytes by steam.
5566 bind(loop);
5567 ldrb(tmp4, post(src, 1));
5568 strh(tmp4, post(dst, 2));
5569 subw(len, len, 1);
5570
5571 bind(tiny);
5572 cbnz(len, loop);
5573
5574 b(done);
5575 }
5576
5577 if (SoftwarePrefetchHintDistance >= 0) {
5578 bind(to_stub);
5579 RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_byte_array_inflate());
5580 assert(stub.target() != NULL, "large_byte_array_inflate stub has not been generated");
5581 trampoline_call(stub);
5582 b(after_init);
5583 }
5584
5585 // Unpack the bytes 8 at a time.
5586 bind(big);
5587 {
5588 Label loop, around, loop_last, loop_start;
5589
5590 if (SoftwarePrefetchHintDistance >= 0) {
5591 const int large_loop_threshold = (64 + 16)/8;
5592 ldrd(vtmp2, post(src, 8));
5593 andw(len, len, 7);
5594 cmp(tmp4, large_loop_threshold);
5595 br(GE, to_stub);
5596 b(loop_start);
5597
5598 bind(loop);
5599 ldrd(vtmp2, post(src, 8));
5600 bind(loop_start);
5601 subs(tmp4, tmp4, 1);
5602 br(EQ, loop_last);
5603 zip1(vtmp2, T16B, vtmp2, vtmp1);
5604 ldrd(vtmp3, post(src, 8));
5605 st1(vtmp2, T8H, post(dst, 16));
5606 subs(tmp4, tmp4, 1);
5607 zip1(vtmp3, T16B, vtmp3, vtmp1);
5608 st1(vtmp3, T8H, post(dst, 16));
5609 br(NE, loop);
5610 b(around);
5611 bind(loop_last);
5612 zip1(vtmp2, T16B, vtmp2, vtmp1);
5613 st1(vtmp2, T8H, post(dst, 16));
5614 bind(around);
5615 cbz(len, done);
5616 } else {
5617 andw(len, len, 7);
5618 bind(loop);
5619 ldrd(vtmp2, post(src, 8));
5620 sub(tmp4, tmp4, 1);
5621 zip1(vtmp3, T16B, vtmp2, vtmp1);
5622 st1(vtmp3, T8H, post(dst, 16));
5623 cbnz(tmp4, loop);
5624 }
5625 }
5626
5627 // Do the tail of up to 8 bytes.
5628 add(src, src, len);
5629 ldrd(vtmp3, Address(src, -8));
5630 add(dst, dst, len, ext::uxtw, 1);
5631 zip1(vtmp3, T16B, vtmp3, vtmp1);
5632 strq(vtmp3, Address(dst, -16));
5633
5634 bind(done);
5635 }
5636
5637 // Compress char[] array to byte[].
5638 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
5639 FloatRegister tmp1Reg, FloatRegister tmp2Reg,
5640 FloatRegister tmp3Reg, FloatRegister tmp4Reg,
5641 Register result) {
5642 encode_iso_array(src, dst, len, result,
5643 tmp1Reg, tmp2Reg, tmp3Reg, tmp4Reg);
5644 cmp(len, zr);
5645 csel(result, result, zr, EQ);
5646 }
5647
5648 // get_thread() can be called anywhere inside generated code so we
5649 // need to save whatever non-callee save context might get clobbered
5650 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
5651 // the call setup code.
5652 //
|