3477 arrays_hashcode_elvload(vtmp[idx], Address(ary1, index, Address::times(elsize), 8 * idx * elsize), eltype);
3478 }
3479 // vresult = vresult * vnext + ary1[index+8*idx:index+8*idx+7];
3480 for (int idx = 0; idx < 4; idx++) {
3481 vpmulld(vresult[idx], vresult[idx], vnext, Assembler::AVX_256bit);
3482 arrays_hashcode_elvcast(vtmp[idx], eltype);
3483 vpaddd(vresult[idx], vresult[idx], vtmp[idx], Assembler::AVX_256bit);
3484 }
3485 // index += 32;
3486 addl(index, 32);
3487 // index < bound;
3488 cmpl(index, bound);
3489 jcc(Assembler::less, UNROLLED_VECTOR_LOOP_BEGIN);
3490 // }
3491
3492 lea(ary1, Address(ary1, bound, Address::times(elsize)));
3493 subl(cnt1, bound);
3494 // release bound
3495
3496 // vresult *= IntVector.fromArray(I256, power_of_31_backwards, 1);
3497 for (int idx = 0; idx < 4; idx++) {
3498 lea(tmp2, ExternalAddress(StubRoutines::x86::arrays_hashcode_powers_of_31() + ((8 * idx + 1) * sizeof(jint))));
3499 arrays_hashcode_elvload(vcoef[idx], Address(tmp2, 0), T_INT);
3500 vpmulld(vresult[idx], vresult[idx], vcoef[idx], Assembler::AVX_256bit);
3501 }
3502 // result += vresult.reduceLanes(ADD);
3503 for (int idx = 0; idx < 4; idx++) {
3504 reduceI(Op_AddReductionVI, 256/(sizeof(jint) * 8), result, result, vresult[idx], vtmp[(idx * 2 + 0) % 4], vtmp[(idx * 2 + 1) % 4]);
3505 }
3506
3507 // } else if (cnt1 < 32) {
3508
3509 bind(SHORT_UNROLLED_BEGIN);
3510 // int i = 1;
3511 movl(index, 1);
3512 cmpl(index, cnt1);
3513 jcc(Assembler::greaterEqual, SHORT_UNROLLED_LOOP_EXIT);
3514
3515 // for (; i < cnt1 ; i += 2) {
3516 bind(SHORT_UNROLLED_LOOP_BEGIN);
3517 movl(tmp3, 961);
3518 imull(result, tmp3);
3519 arrays_hashcode_elload(tmp2, Address(ary1, index, Address::times(elsize), -elsize), eltype);
|
3477 arrays_hashcode_elvload(vtmp[idx], Address(ary1, index, Address::times(elsize), 8 * idx * elsize), eltype);
3478 }
3479 // vresult = vresult * vnext + ary1[index+8*idx:index+8*idx+7];
3480 for (int idx = 0; idx < 4; idx++) {
3481 vpmulld(vresult[idx], vresult[idx], vnext, Assembler::AVX_256bit);
3482 arrays_hashcode_elvcast(vtmp[idx], eltype);
3483 vpaddd(vresult[idx], vresult[idx], vtmp[idx], Assembler::AVX_256bit);
3484 }
3485 // index += 32;
3486 addl(index, 32);
3487 // index < bound;
3488 cmpl(index, bound);
3489 jcc(Assembler::less, UNROLLED_VECTOR_LOOP_BEGIN);
3490 // }
3491
3492 lea(ary1, Address(ary1, bound, Address::times(elsize)));
3493 subl(cnt1, bound);
3494 // release bound
3495
3496 // vresult *= IntVector.fromArray(I256, power_of_31_backwards, 1);
3497 lea(tmp2, ExternalAddress(StubRoutines::x86::arrays_hashcode_powers_of_31() + (0 * sizeof(jint))));
3498 for (int idx = 0; idx < 4; idx++) {
3499 arrays_hashcode_elvload(vcoef[idx], Address(tmp2, (int)((8 * idx + 1) * sizeof(jint))), T_INT);
3500 vpmulld(vresult[idx], vresult[idx], vcoef[idx], Assembler::AVX_256bit);
3501 }
3502 // result += vresult.reduceLanes(ADD);
3503 for (int idx = 0; idx < 4; idx++) {
3504 reduceI(Op_AddReductionVI, 256/(sizeof(jint) * 8), result, result, vresult[idx], vtmp[(idx * 2 + 0) % 4], vtmp[(idx * 2 + 1) % 4]);
3505 }
3506
3507 // } else if (cnt1 < 32) {
3508
3509 bind(SHORT_UNROLLED_BEGIN);
3510 // int i = 1;
3511 movl(index, 1);
3512 cmpl(index, cnt1);
3513 jcc(Assembler::greaterEqual, SHORT_UNROLLED_LOOP_EXIT);
3514
3515 // for (; i < cnt1 ; i += 2) {
3516 bind(SHORT_UNROLLED_LOOP_BEGIN);
3517 movl(tmp3, 961);
3518 imull(result, tmp3);
3519 arrays_hashcode_elload(tmp2, Address(ary1, index, Address::times(elsize), -elsize), eltype);
|