Function localiser_statevec_getAmps

quest/src/core/localiser.cpp:499–537

void localiser_statevec_getAmps(qcomp* outAmps, Qureg qureg, qindex globalStartInd, qindex globalNumAmps) {
    
    // we do not assert state-vec, since the density matrix routine re-uses this function

    // when not distributed, all nodes merely perform direct local overwrite and finish
    if (!qureg.isDistributed) {
        accel_statevec_getAmps_sub(outAmps, qureg, globalStartInd, globalNumAmps);
        return;
    }

    // when distributed, each node will broadcast their overlap (which may be zero) with the global range
    int myRank = comm_getRank();
    int numNodes = comm_getNumNodes();

    // which they first overwrite into their local copies of out (may involve a GPU-to-CPU copy)
    if (util_areAnyVectorElemsWithinNode(myRank, qureg.numAmpsPerNode, globalStartInd, globalNumAmps)) {
        auto localInds = util_getLocalIndRangeOfVectorElemsWithinNode(myRank, qureg.numAmpsPerNode, globalStartInd, globalNumAmps);
        accel_statevec_getAmps_sub(&outAmps[localInds.localDuplicStartInd], qureg, localInds.localDistribStartInd, localInds.numElems);
    }

    // determine the overlap with each node (i.e. which amps, if any, they contribute)
    vector<qindex> globalRecvInds(numNodes);
    vector<qindex> localSendInds (numNodes);
    vector<qindex> numAmpsPerRank(numNodes, 0); // default = zero contributed amps

    for (int sendRank=0; sendRank<numNodes; sendRank++) {

        if (!util_areAnyVectorElemsWithinNode(sendRank, qureg.numAmpsPerNode, globalStartInd, globalNumAmps))
            continue;

        auto inds = util_getLocalIndRangeOfVectorElemsWithinNode(sendRank, qureg.numAmpsPerNode, globalStartInd, globalNumAmps);
        globalRecvInds[sendRank] = inds.localDuplicStartInd;
        localSendInds [sendRank] = inds.localDistribStartInd;
        numAmpsPerRank[sendRank] = inds.numElems;
    }

    // contributor nodes broadcast, all nodes receive, so that every node populates 'outAmps' fully
    comm_combineSubArrays(outAmps, globalRecvInds, localSendInds, numAmpsPerRank);
}


void localiser_densmatr_getAmps(qcomp** outAmps, Qureg qureg, qindex startRow, qindex startCol, qindex numRows, qindex numCols) {