| 344 | |
| 345 | template <int NumCtrls> |
| 346 | void cpu_statevec_anyCtrlSwap_subC(Qureg qureg, vector<int> ctrls, vector<int> ctrlStates, int targ, int targState) { |
| 347 | |
| 348 | assert_numCtrlsMatchesNumCtrlStatesAndTemplateParam(ctrls.size(), ctrlStates.size(), NumCtrls); |
| 349 | |
| 350 | // each control qubit halves the number of iterations, each of which modifies one of the two target qubit states |
| 351 | qindex numIts = qureg.numAmpsPerNode / powerOf2(1 + ctrls.size()); |
| 352 | |
| 353 | // received amplitudes may begin at an arbitrary offset in the buffer |
| 354 | qindex offset = getBufferRecvInd(); |
| 355 | |
| 356 | auto sortedQubits = util_getSorted(ctrls, {targ}); |
| 357 | auto qubitStateMask = util_getBitMask(ctrls, ctrlStates, {targ}, {targState}); |
| 358 | |
| 359 | // use template param to compile-time unroll loop in insertBits() |
| 360 | SET_VAR_AT_COMPILE_TIME(int, numCtrlBits, NumCtrls, ctrls.size()); |
| 361 | int numQubitBits = numCtrlBits + 1; |
| 362 | |
| 363 | #pragma omp parallel for if(qureg.isMultithreaded) |
| 364 | for (qindex n=0; n<numIts; n++) { |
| 365 | |
| 366 | // i = nth local index where ctrls and targ are in specified states |
| 367 | qindex i = insertBitsWithMaskedValues(n, sortedQubits.data(), numQubitBits, qubitStateMask); |
| 368 | |
| 369 | // j = index of nth received amplitude from pair rank in buffer |
| 370 | qindex j = n + offset; |
| 371 | |
| 372 | // unpack the continuous sub-buffer among the strided local amplitudes |
| 373 | qureg.cpuAmps[i] = qureg.cpuCommBuffer[j]; |
| 374 | } |
| 375 | } |
| 376 | |
| 377 | |
| 378 | INSTANTIATE_FUNC_OPTIMISED_FOR_NUM_CTRLS( void, cpu_statevec_anyCtrlSwap_subA, (Qureg qureg, vector<int> ctrls, vector<int> ctrlStates, int targ1, int targ2) ) |
nothing calls this directly
no test coverage detected