lpsolver 0.1.0 → 0.2.1

data/ext/lpsolver-highs/highs/pdlp/cupdlp/cupdlp_linalg.c

@@ -0,0 +1,802 @@
+
+#include "pdlp/cupdlp/cupdlp_linalg.h"
+
+/**
+ * The function `ScatterCol` performs a scatter operation on a specific
+ * column of a matrix.
+ *
+ * @param pdhg A pointer to a structure of type pdhg.
+ * @param iCol The parameter "iCol" represents the index of the column in the
+ * matrix that we want to scatter.
+ * @param multiplier The multiplier is a scalar value that is multiplied with
+ * each element in the column of the matrix. It is used to scale the values
+ * before adding them to the target array.
+ * @param target The "target" parameter is a pointer to a cupdlp_float array
+ * where the scattered column values will be added.
+ */
+void ScatterCol(CUPDLPwork *w, cupdlp_int iCol, cupdlp_float multiplier,
+                cupdlp_float *target) {
+  CUPDLPcsc *matrix = w->problem->data->csc_matrix;
+
+  for (cupdlp_int p = matrix->colMatBeg[iCol]; p < matrix->colMatBeg[iCol + 1];
+       ++p)
+    target[matrix->colMatIdx[p]] += matrix->colMatElem[p] * multiplier;
+}
+
+void ScatterRow(CUPDLPwork *w, cupdlp_int iRow, cupdlp_float multiplier,
+                cupdlp_float *target) {
+  CUPDLPcsr *matrix = w->problem->data->csr_matrix;
+
+  for (cupdlp_int p = matrix->rowMatBeg[iRow]; p < matrix->rowMatBeg[iRow + 1];
+       ++p)
+    target[matrix->rowMatIdx[p]] += matrix->rowMatElem[p] * multiplier;
+}
+
+void AxCPU(CUPDLPwork *w, cupdlp_float *ax, const cupdlp_float *x) {
+  // #if PDHG_USE_TIMERS
+  //     ++w->timers->nAxCalls;
+  //     cupdlp_float dStartTime = getTimeStamp();
+  // #endif
+
+  CUPDLPproblem *lp = w->problem;
+
+  /* no indentity currently
+
+  FILL_ZERO(ax, lp->data->nRows);
+
+  // [A I]*x
+  for (cupdlp_int iSeq = ncols, iRow = 0; iSeq < lp->nSeq; ++iSeq, ++iRow)
+  {
+      if ((pdhg->lower[iSeq] > -INFINITY) && (pdhg->upper[iSeq] < INFINITY))
+      {
+          ax[iRow] = scaling->rowScale ? scaling->rowScale[iRow] * x[iSeq] :
+  x[iSeq];
+      }
+      else
+      {
+          ax[iRow] = 0.0;
+      }
+  }
+  */
+
+  memset(ax, 0, sizeof(cupdlp_float) * lp->data->nRows);
+
+  for (cupdlp_int iCol = 0; iCol < lp->data->nCols; ++iCol) {
+    ScatterCol(w, iCol, x[iCol], ax);
+  }
+
+  // #if PDHG_USE_TIMERS
+  //     w->timers->dAxTime += getTimeStamp() - dStartTime;
+  // #endif
+}
+
+void ATyCPU(CUPDLPwork *w, cupdlp_float *aty, const cupdlp_float *y) {
+  // #if PDHG_USE_TIMERS
+  //     ++w->timers->nAtyCalls;
+  //     cupdlp_float dStartTime = getTimeStamp();
+  // #endif
+
+  CUPDLPproblem *lp = w->problem;
+
+  /* no indentity currently
+  FILL_ZERO(aty, lp->nSeq);
+
+  // [A I]'*y
+  for (cupdlp_int iSeq = ncols, iRow = 0; iSeq < lp->nSeq; ++iSeq, ++iRow)
+  {
+      ScatterRow(pdhg, iRow, y[iRow], aty);
+
+      if ((pdhg->lower[iSeq] > -INFINITY) && (pdhg->upper[iSeq] < INFINITY))
+      {
+          aty[iSeq] = (scaling->rowScale ? scaling->rowScale[iRow] * y[iRow] :
+  y[iRow]);
+      }
+      else
+      {
+          aty[iSeq] = 0.0;
+      }
+  }
+  */
+
+  memset(aty, 0, sizeof(cupdlp_float) * lp->data->nCols);
+  for (cupdlp_int iRow = 0; iRow < lp->data->nRows; ++iRow) {
+    ScatterRow(w, iRow, y[iRow], aty);
+  }
+
+  // #if PDHG_USE_TIMERS
+  //     w->timers->dAtyTime += getTimeStamp() - dStartTime;
+  // #endif
+}
+
+double nrm2(cupdlp_int n, const double *x, cupdlp_int incx) {
+#ifdef USE_MY_BLAS
+  assert(incx == 1);
+
+  double nrm = 0.0;
+
+  for (int i = 0; i < n; ++i) {
+    nrm += x[i] * x[i];
+  }
+
+  return sqrt(nrm);
+#else
+  return dnrm2(n, x, incx);
+#endif
+}
+
+double nrminf(cupdlp_int n, const double *x, cupdlp_int incx) {
+#ifdef USE_MY_BLAS
+  assert(incx == 1);
+
+  double nrm = 0.0;
+
+  for (int i = 0; i < n; ++i) {
+    double tmp = fabs(x[i]);
+    if (tmp > nrm) nrm = tmp;
+  }
+
+  return nrm;
+#else
+  return dnrm2(n, x, incx);
+#endif
+}
+
+static cupdlp_int nrminfindex(cupdlp_int n, const double *x, cupdlp_int incx) {
+#ifdef USE_MY_BLAS
+  assert(incx == 1);
+
+  double nrm = 0.0;
+  cupdlp_int index = 0;
+
+  for (int i = 0; i < n; ++i) {
+    double tmp = fabs(x[i]);
+    if (tmp > nrm) {
+      nrm = tmp;
+      index = i;
+    }
+  }
+
+  return index;
+#else
+  return dnrminfindex(n, x, incx);
+#endif
+}
+
+double twoNorm(double *x, cupdlp_int n) { return nrm2(n, x, 1); }
+
+double twoNormSquared(double *x, cupdlp_int n) { return pow(twoNorm(x, n), 2); }
+
+double infNorm(double *x, cupdlp_int n) { return nrminf(n, x, 1); }
+
+/*------------------------ new added --------------------*/
+double GenNorm(double *x, cupdlp_int n, cupdlp_float p) {
+  if (p == 2.0) {
+    return twoNorm(x, n);
+  } else if (p == INFINITY) {
+    return infNorm(x, n);
+  } else {
+    double nrm = 0.0;
+
+    for (int i = 0; i < n; ++i) {
+      nrm += pow(fabs(x[i]), p);
+    }
+
+    return pow(nrm, 1.0 / p);
+  }
+}
+
+/* x = x .* y*/
+void cupdlp_cdot(cupdlp_float *x, const cupdlp_float *y, const cupdlp_int len) {
+  for (int i = 0; i < len; i++) {
+    x[i] *= y[i];
+  }
+}
+
+/* x = x ./ y*/
+void cupdlp_cdiv(cupdlp_float *x, const cupdlp_float *y, const cupdlp_int len) {
+  for (int i = 0; i < len; i++) {
+    x[i] /= y[i];
+  }
+}
+
+/* xout = weight * x */
+// void cupdlp_scaleVector(cupdlp_float *xout, cupdlp_float *x, cupdlp_float
+// weight, const cupdlp_int len)
+// {
+//     for (int i = 0; i < len; i++)
+//     {
+//         xout[i] = weight * x[i];
+//     }
+// }
+
+/* xout = max(x, lb), lb is vector */
+void cupdlp_projLowerBound(cupdlp_float *x, const cupdlp_float *lb,
+                           const cupdlp_int len) {
+  for (int i = 0; i < len; i++) {
+    x[i] = x[i] > lb[i] ? x[i] : lb[i];
+  }
+}
+
+/* xout = min(x, ub), ub is vector */
+void cupdlp_projUpperBound(cupdlp_float *x, const cupdlp_float *ub,
+                           const cupdlp_int len) {
+  for (int i = 0; i < len; i++) {
+    x[i] = x[i] < ub[i] ? x[i] : ub[i];
+  }
+}
+
+/* xout = max(x, lb), lb is number */
+void cupdlp_projSameLowerBound(cupdlp_float *x, const cupdlp_float lb,
+                               const cupdlp_int len) {
+  for (int i = 0; i < len; i++) {
+    x[i] = x[i] > lb ? x[i] : lb;
+  }
+}
+
+/* xout = min(x, ub), ub is number */
+void cupdlp_projSameUpperBound(cupdlp_float *x, const cupdlp_float ub,
+                               const cupdlp_int len) {
+  for (int i = 0; i < len; i++) {
+    x[i] = x[i] < ub ? x[i] : ub;
+  }
+}
+
+/* xout = max(x, 0) */
+void cupdlp_projPositive(cupdlp_float *x, const cupdlp_int len) {
+  cupdlp_projSameLowerBound(x, 0.0, len);
+}
+
+/* xout = min(x, 0) */
+void cupdlp_projNegative(cupdlp_float *x, const cupdlp_int len) {
+  cupdlp_projSameUpperBound(x, 0.0, len);
+}
+
+// Unused warning
+/* ||x - y||_2^2 */
+// cupdlp_float cupdlp_diffTwoNormSquared(cupdlp_float *x, cupdlp_float *y,
+// const cupdlp_int len)
+// static cupdlp_float diffTwoNormSquared(cupdlp_float *x, cupdlp_float *y,
+//                                 const cupdlp_int len) {
+//   cupdlp_float res = 0.0;
+//   for (int i = 0; i < len; i++) {
+//     cupdlp_float tmp = x[i] - y[i];
+//     res += tmp * tmp;
+//   }
+//   return res;
+// }
+
+// Unused warning
+/* ||x - y||_2 */
+// cupdlp_float cupdlp_diffTwoNorm(cupdlp_float *x, cupdlp_float *y, const
+// cupdlp_int len)
+// static cupdlp_float diffTwoNorm(cupdlp_float *x, cupdlp_float *y,
+//                          const cupdlp_int len) {
+//   // return sqrt(cupdlp_diffTwoNormSquared(x, y, len));
+//   return sqrt(diffTwoNormSquared(x, y, len));
+// }
+
+// Unused warning
+/* ||x - y||_inf */
+// cupdlp_float cupdlp_diffInfNorm(cupdlp_float *x, cupdlp_float *y, const
+// cupdlp_int len)
+// static cupdlp_float diffInfNorm(cupdlp_float *x, cupdlp_float *y,
+//                          const cupdlp_int len) {
+//   cupdlp_float res = 0.0;
+//   for (int i = 0; i < len; i++) {
+//     cupdlp_float tmp = fabs(x[i] - y[i]);
+//     if (tmp > res) res = tmp;
+//   }
+//   return res;
+// }
+
+// Unused warning
+/* (x1 - x2)' (y1 - y2) */
+// cupdlp_float cupdlp_diffDotDiff(cupdlp_float *x1, cupdlp_float *x2,
+// cupdlp_float *y1, cupdlp_float *y2, const cupdlp_int len)
+// static cupdlp_float diffDotDiff(cupdlp_float *x1, cupdlp_float *x2, cupdlp_float *y1,
+//                          cupdlp_float *y2, const cupdlp_int len) {
+//   cupdlp_float x1y1 = dot(len, x1, 1, y1, 1);
+//   cupdlp_float x2y2 = dot(len, x2, 1, y2, 1);
+//   cupdlp_float x1y2 = dot(len, x1, 1, y2, 1);
+//   cupdlp_float x2y1 = dot(len, x2, 1, y1, 1);
+
+//   return x1y1 - x1y2 - x2y1 + x2y2;
+// }
+
+/* x = x .* y */
+// void cupdlp_cdot_fb(cupdlp_float *x, const cupdlp_bool *y, const cupdlp_int
+// len)
+// {
+//     for (int i = 0; i < len; i++)
+//     {
+//         x[i] *= y[i];
+//     }
+// }
+
+/*------------------------ new added --------------------*/
+
+double dot(cupdlp_int n, const cupdlp_float *x, cupdlp_int incx, const cupdlp_float *y,
+           cupdlp_int incy) {
+#ifdef USE_MY_BLAS
+  assert(incx == 1 && incy == 1);
+
+  double dres = 0.0;
+
+  for (int i = 0; i < n; ++i) {
+    dres += x[i] * y[i];
+  }
+
+  return dres;
+#else
+  return ddot(n, x, incx, y, incy);
+#endif
+}
+
+double Dotprod(const cupdlp_float *x, const cupdlp_float *y, cupdlp_int n) {
+  return dot(n, x, 1, y, 1);
+}
+
+double Dotprod_Neumaier(const cupdlp_float *x, const cupdlp_float *y, cupdlp_int n) {
+  return dot(n, x, 1, y, 1);
+}
+
+/* x = x + weight * y */
+void AddToVector(cupdlp_float *x, const cupdlp_float weight,
+                 const cupdlp_float *y, const cupdlp_int n) {
+#ifdef USE_MY_BLAS
+
+  for (int i = 0; i < n; ++i) {
+    x[i] += weight * y[i];
+  }
+
+#else
+  return ddot(n, x, incx, y, incy);
+#endif
+}
+
+/* x = weight * x */
+void ScaleVector(cupdlp_float weight, cupdlp_float *x, cupdlp_int n) {
+#ifdef USE_MY_BLAS
+
+  for (int i = 0; i < n; ++i) {
+    x[i] *= weight;
+  }
+
+#else
+  return ddot(n, x, incx, y, incy);
+#endif
+}
+
+
+static void cupdlp_filter_lower_bound(cupdlp_float *x, const cupdlp_float *lb,
+                               const cupdlp_float bound, const cupdlp_int len) {
+  for (int i = 0; i < len; i++) {
+    x[i] = lb[i] > bound ? lb[i] : 0.0;
+  }
+}
+
+static void cupdlp_filter_upper_bound(cupdlp_float *x, const cupdlp_float *ub,
+                               const cupdlp_float bound, const cupdlp_int len) {
+  for (int i = 0; i < len; i++) {
+    x[i] = ub[i] < bound ? ub[i] : 0.0;
+  }
+}
+
+static void cupdlp_init_vector(cupdlp_float *x, const cupdlp_float val,
+                        const cupdlp_int len) {
+  for (int i = 0; i < len; i++) {
+    x[i] = val;
+  }
+}
+
+#ifndef CUPDLP_CPU
+
+static void Ax_single_gpu(CUPDLPwork *w, cusparseDnVecDescr_t vecX,
+                   cusparseDnVecDescr_t vecAx) {
+  cupdlp_float begin = getTimeStamp();
+  cupdlp_float alpha = 1.0;
+  cupdlp_float beta = 0.0;
+
+  switch (w->problem->data->matrix_format) {
+    case CSR_CSC:
+      cuda_csr_Ax(w->cusparsehandle, w->problem->data->csr_matrix->cuda_csr,
+                  vecX, vecAx, w->dBuffer_csr_Ax, alpha, beta);
+      break;
+    case CSC:
+      cupdlp_printf("Error: Ax_single_gpu requires CSR matrix\n");
+      exit(1);
+      break;
+    case CSR:
+      cuda_csr_Ax(w->cusparsehandle, w->problem->data->csr_matrix->cuda_csr,
+                  vecX, vecAx, w->dBuffer_csr_Ax, alpha, beta);
+      break;
+    default:
+      cupdlp_printf("Error: Unknown matrix format in Ax_single_gpu\n");
+      exit(1);
+  }
+  w->timers->DeviceMatVecProdTime += getTimeStamp() - begin;
+}
+
+static void Ax_multi_gpu(CUPDLPdata *d, cupdlp_float *ax, const cupdlp_float *x) {
+  cupdlp_printf("Error: Ax_multi_gpu not implemented\n");
+  exit(1);
+}
+
+static void ATy_single_gpu(CUPDLPwork *w, cusparseDnVecDescr_t vecY,
+                    cusparseDnVecDescr_t vecATy) {
+  cupdlp_float begin = getTimeStamp();
+
+  cupdlp_float alpha = 1.0;
+  cupdlp_float beta = 0.0;
+
+  switch (w->problem->data->matrix_format) {
+    case CSR_CSC:
+      cuda_csc_ATy(w->cusparsehandle, w->problem->data->csc_matrix->cuda_csc,
+                   vecY, vecATy, w->dBuffer_csc_ATy, alpha, beta);
+      break;
+    case CSC:
+      cuda_csc_ATy(w->cusparsehandle, w->problem->data->csc_matrix->cuda_csc,
+                   vecY, vecATy, w->dBuffer_csc_ATy, alpha, beta);
+      break;
+    case CSR:
+      cupdlp_printf("Error: ATy_single_gpu requires CSC matrix\n");
+      exit(1);
+      break;
+    default:
+      printf("Error: Unknown matrix format in Ax_single_gpu\n");
+      exit(1);
+  }
+
+  w->timers->DeviceMatVecProdTime += getTimeStamp() - begin;
+}
+
+static void ATy_multi_gpu(CUPDLPdata *d, cupdlp_float *aty, const cupdlp_float *y) {
+  cupdlp_printf("Error: ATy_multi_gpu not implemented\n");
+  exit(1);
+}
+
+#endif
+
+void Ax(CUPDLPwork *w, CUPDLPvec *ax, const CUPDLPvec *x) {
+  cupdlp_float begin = getTimeStamp();
+
+  CUPDLPdata *d = w->problem->data;
+  switch (d->device) {
+    case CPU:
+      AxCPU(w, ax->data, x->data);
+      break;
+    case SINGLE_GPU:
+
+#ifndef CUPDLP_CPU
+      Ax_single_gpu(w, x->cuda_vec, ax->cuda_vec);
+#else
+      printf("GPU not supported in CPU build\n");
+      exit(1);
+#endif
+      break;
+    case MULTI_GPU:
+#ifndef CUPDLP_CPU
+      Ax_multi_gpu(d, ax->data, x->data);
+#else
+      printf("GPU not supported in CPU build\n");
+      exit(1);
+#endif
+      break;
+    default:
+      printf("Error: Unknown device type in Ax\n");
+      exit(1);
+  }
+
+#if PDHG_USE_TIMERS
+  w->timers->dAxTime += getTimeStamp() - begin;
+  w->timers->nAxCalls++;
+#endif
+}
+
+void ATy(CUPDLPwork *w, CUPDLPvec *aty, const CUPDLPvec *y)
+
+{
+  cupdlp_float begin = getTimeStamp();
+
+  CUPDLPdata *d = w->problem->data;
+  switch (d->device) {
+    case CPU:
+      ATyCPU(w, aty->data, y->data);
+      break;
+    case SINGLE_GPU:
+#ifndef CUPDLP_CPU
+      ATy_single_gpu(w, y->cuda_vec, aty->cuda_vec);
+#else
+      printf("GPU not supported in CPU build\n");
+      exit(1);
+#endif
+      break;
+    case MULTI_GPU:
+#ifndef CUPDLP_CPU
+      ATy_multi_gpu(d, aty->data, y->data);
+#else
+      printf("GPU not supported in CPU build\n");
+      exit(1);
+#endif
+      break;
+    default:
+      printf("Error: Unknown device type in ATy\n");
+      exit(1);
+  }
+#if PDHG_USE_TIMERS
+  w->timers->dAtyTime += getTimeStamp() - begin;
+  w->timers->nAtyCalls++;
+#endif
+}
+
+/*-------------- Apis compatible with both CPU and GPU -------------------*/
+// only implemented the APis need to be used on GPU
+
+// functions in cublas
+
+cupdlp_int cupdlp_axpy(CUPDLPwork *w, const cupdlp_int n,
+                       const cupdlp_float *alpha, const cupdlp_float *x,
+                       cupdlp_float *y) {
+#ifndef CUPDLP_CPU
+#ifndef SFLOAT
+  CHECK_CUBLAS(cublasDaxpy(w->cublashandle, n, alpha, x, 1, y, 1));
+#else
+  CHECK_CUBLAS(cublasSaxpy(w->cublashandle, n, alpha, x, 1, y, 1));
+#endif
+#else
+  // AddToVector(x, *alpha, y, n);
+  AddToVector(y, *alpha, x, n);
+#endif
+  return 0;
+}
+
+cupdlp_int cupdlp_dot(CUPDLPwork *w, const cupdlp_int n, const cupdlp_float *x,
+                      const cupdlp_float *y, cupdlp_float *res) {
+#ifndef CUPDLP_CPU
+#ifndef SFLOAT
+  CHECK_CUBLAS(cublasDdot(w->cublashandle, n, x, 1, y, 1, res));
+#else
+  CHECK_CUBLAS(cublasSdot(w->cublashandle, n, x, 1, y, 1 res));
+#endif
+#else
+  *res = dot(n, x, 1, y, 1);
+#endif
+  return 0;
+}
+
+cupdlp_int cupdlp_twoNorm(CUPDLPwork *w, const cupdlp_int n,
+                          const cupdlp_float *x, cupdlp_float *res) {
+#ifndef CUPDLP_CPU
+#ifndef SFLOAT
+  CHECK_CUBLAS(cublasDnrm2(w->cublashandle, n, x, 1, res));
+#else
+  CHECK_CUBLAS(cublasSnrm2(w->cublashandle, n, x, 1, res));
+#endif
+#else
+  *res = nrm2(n, x, 1);
+#endif
+  return 0;
+}
+
+cupdlp_int cupdlp_infNormIndex(CUPDLPwork *w, const cupdlp_int n,
+                               const cupdlp_float *x, cupdlp_int *res) {
+#ifndef CUPDLP_CPU
+#ifndef SFLOAT
+  CHECK_CUBLAS(cublasIdamax(w->cublashandle, n, x, 1, res));
+#else
+  CHECK_CUBLAS(cublasIsamax(w->cublashandle, n, x, 1, res));
+#endif
+#else
+  *res = nrminfindex(n, x, 1);
+#endif
+  return 0;
+}
+
+cupdlp_int cupdlp_scaleVector(CUPDLPwork *w, const cupdlp_float weight,
+                              cupdlp_float *x, const cupdlp_int n) {
+#ifndef CUPDLP_CPU
+#ifndef SFLOAT
+  CHECK_CUBLAS(cublasDscal(w->cublashandle, n, &weight, x, 1));
+#else
+  CHECK_CUBLAS(cublasSscal(w->cublashandle, n, &weight, x, 1));
+#endif
+#else
+  ScaleVector(weight, x, n);
+#endif
+  return 0;
+}
+
+void cupdlp_twoNormSquared(CUPDLPwork *w, const cupdlp_int n,
+                           const cupdlp_float *x, cupdlp_float *res) {
+  cupdlp_dot(w, n, x, x, res);
+}
+
+/* ||x - y||_2^2 */
+void cupdlp_diffTwoNormSquared(CUPDLPwork *w, const cupdlp_float *x,
+                               const cupdlp_float *y, const cupdlp_int len,
+                               cupdlp_float *res) {
+  CUPDLP_COPY_VEC(w->buffer2, x, cupdlp_float, len);
+  cupdlp_float alpha = -1.0;
+  cupdlp_axpy(w, len, &alpha, y, w->buffer2);
+  cupdlp_twoNormSquared(w, len, w->buffer2, res);
+}
+
+/* ||x - y||_2 */
+void cupdlp_diffTwoNorm(CUPDLPwork *w, const cupdlp_float *x,
+                        const cupdlp_float *y, const cupdlp_int len,
+                        cupdlp_float *res) {
+  CUPDLP_COPY_VEC(w->buffer2, x, cupdlp_float, len);
+  cupdlp_float alpha = -1.0;
+  cupdlp_axpy(w, len, &alpha, y, w->buffer2);
+  cupdlp_twoNorm(w, len, w->buffer2, res);
+}
+
+/* (x1 - x2)' (y1 - y2) */
+void cupdlp_diffDotDiff(CUPDLPwork *w, const cupdlp_float *x1,
+                        const cupdlp_float *x2, const cupdlp_float *y1,
+                        const cupdlp_float *y2, const cupdlp_int len,
+                        cupdlp_float *res) {
+  CUPDLP_COPY_VEC(w->buffer2, x1, cupdlp_float, len);
+  cupdlp_float alpha = -1.0;
+  cupdlp_axpy(w, len, &alpha, x2, w->buffer2);
+  CUPDLP_COPY_VEC(w->buffer3, y1, cupdlp_float, len);
+  cupdlp_axpy(w, len, &alpha, y2, w->buffer3);
+  // reduce step
+  cupdlp_dot(w, len, w->buffer2, w->buffer3, res);
+}
+
+// functions not in cublas
+
+/* element wise dot: x = x .* y*/
+void cupdlp_edot(cupdlp_float *x, const cupdlp_float *y, const cupdlp_int len) {
+#ifndef CUPDLP_CPU
+  cupdlp_edot_cuda(x, y, len);
+#else
+  cupdlp_cdot(x, y, len);
+#endif
+}
+
+/* element wise div: x = x ./ y*/
+void cupdlp_ediv(cupdlp_float *x, const cupdlp_float *y, const cupdlp_int len) {
+#ifndef CUPDLP_CPU
+  cupdlp_ediv_cuda(x, y, len);
+#else
+  cupdlp_cdiv(x, y, len);
+#endif
+}
+
+void cupdlp_projlb(cupdlp_float *x, const cupdlp_float *lb,
+                   const cupdlp_int len) {
+#ifndef CUPDLP_CPU
+  cupdlp_projlb_cuda(x, lb, len);
+#else
+  cupdlp_projLowerBound(x, lb, len);
+#endif
+}
+
+void cupdlp_projub(cupdlp_float *x, const cupdlp_float *ub,
+                   const cupdlp_int len) {
+#ifndef CUPDLP_CPU
+  cupdlp_projub_cuda(x, ub, len);
+#else
+  cupdlp_projUpperBound(x, ub, len);
+#endif
+}
+
+void cupdlp_projSamelb(cupdlp_float *x, const cupdlp_float lb,
+                       const cupdlp_int len) {
+#ifndef CUPDLP_CPU
+  cupdlp_projSamelb_cuda(x, lb, len);
+#else
+  cupdlp_projSameLowerBound(x, lb, len);
+#endif
+}
+
+void cupdlp_projSameub(cupdlp_float *x, const cupdlp_float ub,
+                       const cupdlp_int len) {
+#ifndef CUPDLP_CPU
+  cupdlp_projSameub_cuda(x, ub, len);
+#else
+  cupdlp_projSameUpperBound(x, ub, len);
+#endif
+}
+
+/* xout = max(x, 0) */
+void cupdlp_projPos(cupdlp_float *x, const cupdlp_int len) {
+  cupdlp_projSamelb(x, 0.0, len);
+}
+
+/* xout = min(x, 0) */
+void cupdlp_projNeg(cupdlp_float *x, const cupdlp_int len) {
+  cupdlp_projSameub(x, 0.0, len);
+}
+
+// void cupdlp_haslb(cupdlp_float *haslb, const cupdlp_float *lb,
+//                   const cupdlp_float bound, const cupdlp_int len) {
+// #ifndef CUPDLP_CPU
+//   cupdlp_haslb_cuda(haslb, lb, bound, len);
+// #else
+//   cupdlp_hasLower(haslb, lb, bound, len);
+// #endif
+// }
+
+// void cupdlp_hasub(cupdlp_float *hasub, const cupdlp_float *ub,
+//                   const cupdlp_float bound, const cupdlp_int len) {
+// #ifndef CUPDLP_CPU
+//   cupdlp_hasub_cuda(hasub, ub, bound, len);
+// #else
+//   cupdlp_hasUpper(hasub, ub, bound, len);
+// #endif
+// }
+
+void cupdlp_filterlb(cupdlp_float *x, const cupdlp_float *lb,
+                     const cupdlp_float bound, const cupdlp_int len) {
+#ifndef CUPDLP_CPU
+  cupdlp_filterlb_cuda(x, lb, bound, len);
+#else
+  cupdlp_filter_lower_bound(x, lb, bound, len);
+#endif
+}
+
+void cupdlp_filterub(cupdlp_float *x, const cupdlp_float *ub,
+                     const cupdlp_float bound, const cupdlp_int len) {
+#ifndef CUPDLP_CPU
+  cupdlp_filterub_cuda(x, ub, bound, len);
+#else
+  cupdlp_filter_upper_bound(x, ub, bound, len);
+#endif
+}
+
+void cupdlp_initvec(cupdlp_float *x, const cupdlp_float val,
+                    const cupdlp_int len) {
+#ifndef CUPDLP_CPU
+  cupdlp_initvec_cuda(x, val, len);
+#else
+  cupdlp_init_vector(x, val, len);
+#endif
+}
+
+/*
+void cupdlp_sub(cupdlp_float *xout, const cupdlp_float *x1,
+                const cupdlp_float *x2, const cupdlp_int len) {
+#ifndef CUPDLP_CPU
+  cupdlp_sub_cuda(xout, x1, x2, len);
+#else
+  CUPDLP_COPY_VEC(xout, x1, cupdlp_float, len);
+  cupdlp_float alpha = -1.0;
+  cupdlp_axpy(NULL, len, &alpha, x2, xout);
+#endif
+}
+*/
+
+void cupdlp_compute_interaction_and_movement(CUPDLPwork *w,
+                                             cupdlp_float *dMovement,
+                                             cupdlp_float *dInteraction) {
+  CUPDLPiterates *iterates = w->iterates;
+  cupdlp_int nCols = w->problem->nCols;
+  cupdlp_int nRows = w->problem->nRows;
+  cupdlp_float beta = sqrt(w->stepsize->dBeta);
+  cupdlp_float dX = 0.0;
+  cupdlp_float dY = 0.0;
+
+  cupdlp_int iter = w->timers->nIter;
+  CUPDLPvec *x = iterates->x[iter % 2];
+  CUPDLPvec *y = iterates->y[iter % 2];
+  CUPDLPvec *aty = iterates->aty[iter % 2];
+  CUPDLPvec *xUpdate = iterates->x[(iter + 1) % 2];
+  CUPDLPvec *yUpdate = iterates->y[(iter + 1) % 2];
+  CUPDLPvec *atyUpdate = iterates->aty[(iter + 1) % 2];
+
+#if !defined(CUPDLP_CPU) && USE_KERNELS
+  cupdlp_movement_interaction_cuda(&dX, &dY, dInteraction, w->buffer2,
+      xUpdate->data, x->data, yUpdate->data, y->data, atyUpdate->data, aty->data, nRows, nCols);
+#else
+    cupdlp_diffTwoNormSquared(w, x->data, xUpdate->data, nCols, &dX);
+    cupdlp_diffTwoNormSquared(w, y->data, yUpdate->data, nRows, &dY);
+    //      Δx' (AΔy)
+    cupdlp_diffDotDiff(w, x->data, xUpdate->data, aty->data, atyUpdate->data,
+                       nCols, dInteraction);
+#endif
+  *dMovement = dX * 0.5 * beta + dY / (2.0 * beta);
+}
+