3560 arrays_hashcode_elvload(vtmp[idx], Address(ary1, index, Address::times(elsize), 8 * idx * elsize), eltype);
3561 }
3562 // vresult = vresult * vnext + ary1[index+8*idx:index+8*idx+7];
3563 for (int idx = 0; idx < 4; idx++) {
3564 vpmulld(vresult[idx], vresult[idx], vnext, Assembler::AVX_256bit);
3565 arrays_hashcode_elvcast(vtmp[idx], eltype);
3566 vpaddd(vresult[idx], vresult[idx], vtmp[idx], Assembler::AVX_256bit);
3567 }
3568 // index += 32;
3569 addl(index, 32);
3570 // index < bound;
3571 cmpl(index, bound);
3572 jcc(Assembler::less, UNROLLED_VECTOR_LOOP_BEGIN);
3573 // }
3574
3575 lea(ary1, Address(ary1, bound, Address::times(elsize)));
3576 subl(cnt1, bound);
3577 // release bound
3578
3579 // vresult *= IntVector.fromArray(I256, power_of_31_backwards, 1);
3580 for (int idx = 0; idx < 4; idx++) {
3581 lea(tmp2, ExternalAddress(StubRoutines::x86::arrays_hashcode_powers_of_31() + ((8 * idx + 1) * sizeof(jint))));
3582 arrays_hashcode_elvload(vcoef[idx], Address(tmp2, 0), T_INT);
3583 vpmulld(vresult[idx], vresult[idx], vcoef[idx], Assembler::AVX_256bit);
3584 }
3585 // result += vresult.reduceLanes(ADD);
3586 for (int idx = 0; idx < 4; idx++) {
3587 reduceI(Op_AddReductionVI, 256/(sizeof(jint) * 8), result, result, vresult[idx], vtmp[(idx * 2 + 0) % 4], vtmp[(idx * 2 + 1) % 4]);
3588 }
3589
3590 // } else if (cnt1 < 32) {
3591
3592 bind(SHORT_UNROLLED_BEGIN);
3593 // int i = 1;
3594 movl(index, 1);
3595 cmpl(index, cnt1);
3596 jcc(Assembler::greaterEqual, SHORT_UNROLLED_LOOP_EXIT);
3597
3598 // for (; i < cnt1 ; i += 2) {
3599 bind(SHORT_UNROLLED_LOOP_BEGIN);
3600 movl(tmp3, 961);
3601 imull(result, tmp3);
3602 arrays_hashcode_elload(tmp2, Address(ary1, index, Address::times(elsize), -elsize), eltype);
|
3560 arrays_hashcode_elvload(vtmp[idx], Address(ary1, index, Address::times(elsize), 8 * idx * elsize), eltype);
3561 }
3562 // vresult = vresult * vnext + ary1[index+8*idx:index+8*idx+7];
3563 for (int idx = 0; idx < 4; idx++) {
3564 vpmulld(vresult[idx], vresult[idx], vnext, Assembler::AVX_256bit);
3565 arrays_hashcode_elvcast(vtmp[idx], eltype);
3566 vpaddd(vresult[idx], vresult[idx], vtmp[idx], Assembler::AVX_256bit);
3567 }
3568 // index += 32;
3569 addl(index, 32);
3570 // index < bound;
3571 cmpl(index, bound);
3572 jcc(Assembler::less, UNROLLED_VECTOR_LOOP_BEGIN);
3573 // }
3574
3575 lea(ary1, Address(ary1, bound, Address::times(elsize)));
3576 subl(cnt1, bound);
3577 // release bound
3578
3579 // vresult *= IntVector.fromArray(I256, power_of_31_backwards, 1);
3580 lea(tmp2, ExternalAddress(StubRoutines::x86::arrays_hashcode_powers_of_31() + (0 * sizeof(jint))));
3581 for (int idx = 0; idx < 4; idx++) {
3582 arrays_hashcode_elvload(vcoef[idx], Address(tmp2, (int)((8 * idx + 1) * sizeof(jint))), T_INT);
3583 vpmulld(vresult[idx], vresult[idx], vcoef[idx], Assembler::AVX_256bit);
3584 }
3585 // result += vresult.reduceLanes(ADD);
3586 for (int idx = 0; idx < 4; idx++) {
3587 reduceI(Op_AddReductionVI, 256/(sizeof(jint) * 8), result, result, vresult[idx], vtmp[(idx * 2 + 0) % 4], vtmp[(idx * 2 + 1) % 4]);
3588 }
3589
3590 // } else if (cnt1 < 32) {
3591
3592 bind(SHORT_UNROLLED_BEGIN);
3593 // int i = 1;
3594 movl(index, 1);
3595 cmpl(index, cnt1);
3596 jcc(Assembler::greaterEqual, SHORT_UNROLLED_LOOP_EXIT);
3597
3598 // for (; i < cnt1 ; i += 2) {
3599 bind(SHORT_UNROLLED_LOOP_BEGIN);
3600 movl(tmp3, 961);
3601 imull(result, tmp3);
3602 arrays_hashcode_elload(tmp2, Address(ary1, index, Address::times(elsize), -elsize), eltype);
|