6477 // Perform above steps with lane comparison expression as INDEX >= 48 && INDEX < 64
6478 // and broadcasting third 128 bit lane.
6479 evpcmpb(ktmp, k0, shuffle, xtmp1, Assembler::nlt, true, vlen_enc);
6480 vpsllq(xtmp2, xtmp2, 0x1, vlen_enc);
6481 evpcmpb(ktmp, ktmp, shuffle, xtmp2, Assembler::lt, true, vlen_enc);
6482 evshufi64x2(xtmp3, src, src, 0xFF, vlen_enc);
6483 evpshufb(dst, ktmp, xtmp3, shuffle, true, vlen_enc);
6484 }
6485
6486 void C2_MacroAssembler::vector_rearrange_int_float(BasicType bt, XMMRegister dst,
6487 XMMRegister shuffle, XMMRegister src, int vlen_enc) {
6488 if (vlen_enc == AVX_128bit) {
6489 vpermilps(dst, src, shuffle, vlen_enc);
6490 } else if (bt == T_INT) {
6491 vpermd(dst, shuffle, src, vlen_enc);
6492 } else {
6493 assert(bt == T_FLOAT, "");
6494 vpermps(dst, shuffle, src, vlen_enc);
6495 }
6496 }
|
6477 // Perform above steps with lane comparison expression as INDEX >= 48 && INDEX < 64
6478 // and broadcasting third 128 bit lane.
6479 evpcmpb(ktmp, k0, shuffle, xtmp1, Assembler::nlt, true, vlen_enc);
6480 vpsllq(xtmp2, xtmp2, 0x1, vlen_enc);
6481 evpcmpb(ktmp, ktmp, shuffle, xtmp2, Assembler::lt, true, vlen_enc);
6482 evshufi64x2(xtmp3, src, src, 0xFF, vlen_enc);
6483 evpshufb(dst, ktmp, xtmp3, shuffle, true, vlen_enc);
6484 }
6485
6486 void C2_MacroAssembler::vector_rearrange_int_float(BasicType bt, XMMRegister dst,
6487 XMMRegister shuffle, XMMRegister src, int vlen_enc) {
6488 if (vlen_enc == AVX_128bit) {
6489 vpermilps(dst, src, shuffle, vlen_enc);
6490 } else if (bt == T_INT) {
6491 vpermd(dst, shuffle, src, vlen_enc);
6492 } else {
6493 assert(bt == T_FLOAT, "");
6494 vpermps(dst, shuffle, src, vlen_enc);
6495 }
6496 }
6497
6498 #ifdef _LP64
6499 void C2_MacroAssembler::load_nklass_compact_c2(Register dst, Register obj, Register index, Address::ScaleFactor scale, int disp) {
6500 C2LoadNKlassStub* stub = new (Compile::current()->comp_arena()) C2LoadNKlassStub(dst);
6501 Compile::current()->output()->add_stub(stub);
6502
6503 // Note: Don't clobber obj anywhere in that method!
6504
6505 // The incoming address is pointing into obj-start + klass_offset_in_bytes. We need to extract
6506 // obj-start, so that we can load from the object's mark-word instead. Usually the address
6507 // comes as obj-start in obj and klass_offset_in_bytes in disp. However, sometimes C2
6508 // emits code that pre-computes obj-start + klass_offset_in_bytes into a register, and
6509 // then passes that register as obj and 0 in disp. The following code extracts the base
6510 // and offset to load the mark-word.
6511 int offset = oopDesc::mark_offset_in_bytes() + disp - oopDesc::klass_offset_in_bytes();
6512 movq(dst, Address(obj, index, scale, offset));
6513 testb(dst, markWord::monitor_value);
6514 jcc(Assembler::notZero, stub->entry());
6515 bind(stub->continuation());
6516 shrq(dst, markWord::klass_shift);
6517 }
6518 #endif
|