< prev index next >

hotspot/src/cpu/x86/vm/stubGenerator_x86_32.cpp

Print this page 
rev 7347 : 8078113: 8011102 changes may cause incorrect results
Summary: replace Vzeroupper instruction in stubs with zeroing only used ymm registers.
Reviewed-by: kvn
Contributed-by: sandhya.viswanathan@intel.com


 820       __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
 821       __ movq(xmm3, Address(from, 24));
 822       __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
 823       __ movq(xmm4, Address(from, 32));
 824       __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
 825       __ movq(xmm5, Address(from, 40));
 826       __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
 827       __ movq(xmm6, Address(from, 48));
 828       __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
 829       __ movq(xmm7, Address(from, 56));
 830       __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
 831     }
 832 
 833     __ addl(from, 64);
 834   __ BIND(L_copy_64_bytes);
 835     __ subl(qword_count, 8);
 836     __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
 837 
 838     if (UseUnalignedLoadStores && (UseAVX >= 2)) {
 839       // clean upper bits of YMM registers
 840       __ vzeroupper();

 841     }
 842     __ addl(qword_count, 8);
 843     __ jccb(Assembler::zero, L_exit);
 844     //
 845     // length is too short, just copy qwords
 846     //
 847   __ BIND(L_copy_8_bytes);
 848     __ movq(xmm0, Address(from, 0));
 849     __ movq(Address(from, to_from, Address::times_1), xmm0);
 850     __ addl(from, 8);
 851     __ decrement(qword_count);
 852     __ jcc(Assembler::greater, L_copy_8_bytes);
 853   __ BIND(L_exit);
 854   }
 855 
 856   // Copy 64 bytes chunks
 857   //
 858   // Inputs:
 859   //   from        - source array address
 860   //   to_from     - destination array address - from




 820       __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
 821       __ movq(xmm3, Address(from, 24));
 822       __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
 823       __ movq(xmm4, Address(from, 32));
 824       __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
 825       __ movq(xmm5, Address(from, 40));
 826       __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
 827       __ movq(xmm6, Address(from, 48));
 828       __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
 829       __ movq(xmm7, Address(from, 56));
 830       __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
 831     }
 832 
 833     __ addl(from, 64);
 834   __ BIND(L_copy_64_bytes);
 835     __ subl(qword_count, 8);
 836     __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
 837 
 838     if (UseUnalignedLoadStores && (UseAVX >= 2)) {
 839       // clean upper bits of YMM registers
 840       __ vpxor(xmm0, xmm0);
 841       __ vpxor(xmm1, xmm1);
 842     }
 843     __ addl(qword_count, 8);
 844     __ jccb(Assembler::zero, L_exit);
 845     //
 846     // length is too short, just copy qwords
 847     //
 848   __ BIND(L_copy_8_bytes);
 849     __ movq(xmm0, Address(from, 0));
 850     __ movq(Address(from, to_from, Address::times_1), xmm0);
 851     __ addl(from, 8);
 852     __ decrement(qword_count);
 853     __ jcc(Assembler::greater, L_copy_8_bytes);
 854   __ BIND(L_exit);
 855   }
 856 
 857   // Copy 64 bytes chunks
 858   //
 859   // Inputs:
 860   //   from        - source array address
 861   //   to_from     - destination array address - from
< prev index next >