197929ea7SJunchao Zhang static char help[] = " Test VecScatterRemap() on various vecscatter. \n\
297929ea7SJunchao Zhang We may do optimization based on index patterns. After index remapping by VecScatterRemap(), we need to \n\
397929ea7SJunchao Zhang make sure the vecscatter works as expected with the optimizaiton. \n\
497929ea7SJunchao Zhang VecScatterRemap() does not support all kinds of vecscatters. In addition, it only supports remapping \n\
56aad120cSJose E. Roman entries where we read the data (i.e., todata in parallel scatter, fromdata in sequential scatter). This test \n\
66aad120cSJose E. Roman tests VecScatterRemap on parallel to parallel (PtoP) vecscatter, sequential general to sequential \n\
797929ea7SJunchao Zhang general (SGToSG) vecscatter and sequential general to sequential stride 1 (SGToSS_Stride1) vecscatter.\n\n";
897929ea7SJunchao Zhang
997929ea7SJunchao Zhang #include <petscvec.h>
1097929ea7SJunchao Zhang
main(int argc,char ** argv)11d71ae5a4SJacob Faibussowitsch int main(int argc, char **argv)
12d71ae5a4SJacob Faibussowitsch {
1397929ea7SJunchao Zhang PetscInt i, n, *ix, *iy, *tomap, start;
1497929ea7SJunchao Zhang Vec x, y;
1597929ea7SJunchao Zhang PetscMPIInt nproc, rank;
1697929ea7SJunchao Zhang IS isx, isy;
1797929ea7SJunchao Zhang const PetscInt *ranges;
1897929ea7SJunchao Zhang VecScatter vscat;
1997929ea7SJunchao Zhang
2097929ea7SJunchao Zhang PetscFunctionBegin;
21327415f7SBarry Smith PetscFunctionBeginUser;
22*c8025a54SPierre Jolivet PetscCall(PetscInitialize(&argc, &argv, NULL, help));
239566063dSJacob Faibussowitsch PetscCallMPI(MPI_Comm_size(PETSC_COMM_WORLD, &nproc));
249566063dSJacob Faibussowitsch PetscCallMPI(MPI_Comm_rank(PETSC_COMM_WORLD, &rank));
2597929ea7SJunchao Zhang
2608401ef6SPierre Jolivet PetscCheck(nproc == 2, PETSC_COMM_WORLD, PETSC_ERR_WRONG_MPI_SIZE, "This test must run with exactly two MPI ranks");
2797929ea7SJunchao Zhang
2897929ea7SJunchao Zhang /* ====================================================================
2997929ea7SJunchao Zhang (1) test VecScatterRemap on a parallel to parallel (PtoP) vecscatter
3097929ea7SJunchao Zhang ====================================================================
3197929ea7SJunchao Zhang */
3297929ea7SJunchao Zhang
3397929ea7SJunchao Zhang n = 64; /* long enough to trigger memcpy optimizations both in local scatter and remote scatter */
3497929ea7SJunchao Zhang
3597929ea7SJunchao Zhang /* create two MPI vectors x, y of length n=64, N=128 */
3677433607SBarry Smith PetscCall(VecCreateFromOptions(PETSC_COMM_WORLD, NULL, 1, n, PETSC_DECIDE, &x));
379566063dSJacob Faibussowitsch PetscCall(VecDuplicate(x, &y));
3897929ea7SJunchao Zhang
3997929ea7SJunchao Zhang /* Initialize x as {0~127} */
409566063dSJacob Faibussowitsch PetscCall(VecGetOwnershipRanges(x, &ranges));
419566063dSJacob Faibussowitsch for (i = ranges[rank]; i < ranges[rank + 1]; i++) PetscCall(VecSetValue(x, i, (PetscScalar)i, INSERT_VALUES));
429566063dSJacob Faibussowitsch PetscCall(VecAssemblyBegin(x));
439566063dSJacob Faibussowitsch PetscCall(VecAssemblyEnd(x));
4497929ea7SJunchao Zhang
4597929ea7SJunchao Zhang /* create two general index sets isx = {0~127} and isy = {32~63,64~95,96~127,0~31}. isx is sequential, but we use
4697929ea7SJunchao Zhang it as general and let PETSc detect the pattern and optimize it. indices in isy are set to make the vecscatter
4797929ea7SJunchao Zhang have both local scatter and remote scatter (i.e., MPI communication)
4897929ea7SJunchao Zhang */
499566063dSJacob Faibussowitsch PetscCall(PetscMalloc2(n, &ix, n, &iy));
5097929ea7SJunchao Zhang start = ranges[rank];
5197929ea7SJunchao Zhang for (i = ranges[rank]; i < ranges[rank + 1]; i++) ix[i - start] = i;
529566063dSJacob Faibussowitsch PetscCall(ISCreateGeneral(PETSC_COMM_WORLD, n, ix, PETSC_COPY_VALUES, &isx));
5397929ea7SJunchao Zhang
549371c9d4SSatish Balay if (rank == 0) {
559371c9d4SSatish Balay for (i = 0; i < n; i++) iy[i] = i + 32;
569371c9d4SSatish Balay } else
579371c9d4SSatish Balay for (i = 0; i < n / 2; i++) {
589371c9d4SSatish Balay iy[i] = i + 96;
599371c9d4SSatish Balay iy[i + n / 2] = i;
609371c9d4SSatish Balay }
6197929ea7SJunchao Zhang
629566063dSJacob Faibussowitsch PetscCall(ISCreateGeneral(PETSC_COMM_WORLD, n, iy, PETSC_COPY_VALUES, &isy));
6397929ea7SJunchao Zhang
6469d47153SPierre Jolivet /* create a vecscatter that shifts x to the tail by quarter periodically and puts the results in y */
659566063dSJacob Faibussowitsch PetscCall(VecScatterCreate(x, isx, y, isy, &vscat));
669566063dSJacob Faibussowitsch PetscCall(VecScatterBegin(vscat, x, y, INSERT_VALUES, SCATTER_FORWARD));
679566063dSJacob Faibussowitsch PetscCall(VecScatterEnd(vscat, x, y, INSERT_VALUES, SCATTER_FORWARD));
6897929ea7SJunchao Zhang
6997929ea7SJunchao Zhang /* view y to check the result. y should be {Q3,Q0,Q1,Q2} of x, that is {96~127,0~31,32~63,64~95} */
709566063dSJacob Faibussowitsch PetscCall(PetscPrintf(PETSC_COMM_WORLD, "Before VecScatterRemap on PtoP, MPI vector y is:\n"));
719566063dSJacob Faibussowitsch PetscCall(VecView(y, PETSC_VIEWER_STDOUT_WORLD));
7297929ea7SJunchao Zhang
7397929ea7SJunchao Zhang /* now call the weird subroutine VecScatterRemap to slightly change the vecscatter. It changes where we read vector
7497929ea7SJunchao Zhang x entries to send out, but does not change the communication pattern (i.e., send/recv pairs and msg lengths).
7597929ea7SJunchao Zhang
766aad120cSJose E. Roman We create tomap as {32~63,0~31}. Originally, we read from indices {0~64} of the local x to send out. The remap
7797929ea7SJunchao Zhang does indices[i] = tomap[indices[i]]. Therefore, after the remap, we read from indices {32~63,0~31} of the local x.
7897929ea7SJunchao Zhang isy is unchanged. So, we will shift x to {Q2,Q1,Q0,Q3}, that is {64~95,32~63,0~31,96~127}
7997929ea7SJunchao Zhang */
809566063dSJacob Faibussowitsch PetscCall(PetscMalloc1(n, &tomap));
819371c9d4SSatish Balay for (i = 0; i < n / 2; i++) {
829371c9d4SSatish Balay tomap[i] = i + n / 2;
839371c9d4SSatish Balay tomap[i + n / 2] = i;
849371c9d4SSatish Balay };
859566063dSJacob Faibussowitsch PetscCall(VecScatterRemap(vscat, tomap, NULL));
869566063dSJacob Faibussowitsch PetscCall(VecScatterBegin(vscat, x, y, INSERT_VALUES, SCATTER_FORWARD));
879566063dSJacob Faibussowitsch PetscCall(VecScatterEnd(vscat, x, y, INSERT_VALUES, SCATTER_FORWARD));
8897929ea7SJunchao Zhang
8997929ea7SJunchao Zhang /* view y to check the result. y should be {64~95,32~63,0~31,96~127} */
909566063dSJacob Faibussowitsch PetscCall(PetscPrintf(PETSC_COMM_WORLD, "After VecScatterRemap on PtoP, MPI vector y is:\n"));
919566063dSJacob Faibussowitsch PetscCall(VecView(y, PETSC_VIEWER_STDOUT_WORLD));
9297929ea7SJunchao Zhang
9397929ea7SJunchao Zhang /* destroy everything before we recreate them in different types */
949566063dSJacob Faibussowitsch PetscCall(PetscFree2(ix, iy));
959566063dSJacob Faibussowitsch PetscCall(VecDestroy(&x));
969566063dSJacob Faibussowitsch PetscCall(VecDestroy(&y));
979566063dSJacob Faibussowitsch PetscCall(ISDestroy(&isx));
989566063dSJacob Faibussowitsch PetscCall(ISDestroy(&isy));
999566063dSJacob Faibussowitsch PetscCall(PetscFree(tomap));
1009566063dSJacob Faibussowitsch PetscCall(VecScatterDestroy(&vscat));
10197929ea7SJunchao Zhang
10297929ea7SJunchao Zhang /* ==========================================================================================
10397929ea7SJunchao Zhang (2) test VecScatterRemap on a sequential general to sequential general (SGToSG) vecscatter
10497929ea7SJunchao Zhang ==========================================================================================
10597929ea7SJunchao Zhang */
10697929ea7SJunchao Zhang n = 64; /* long enough to trigger memcpy optimizations in local scatter */
10797929ea7SJunchao Zhang
10897929ea7SJunchao Zhang /* create two seq vectors x, y of length n */
10977433607SBarry Smith PetscCall(VecCreateFromOptions(PETSC_COMM_SELF, NULL, 1, n, n, &x));
1109566063dSJacob Faibussowitsch PetscCall(VecDuplicate(x, &y));
11197929ea7SJunchao Zhang
11297929ea7SJunchao Zhang /* Initialize x as {0~63} */
1139566063dSJacob Faibussowitsch for (i = 0; i < n; i++) PetscCall(VecSetValue(x, i, (PetscScalar)i, INSERT_VALUES));
1149566063dSJacob Faibussowitsch PetscCall(VecAssemblyBegin(x));
1159566063dSJacob Faibussowitsch PetscCall(VecAssemblyEnd(x));
11697929ea7SJunchao Zhang
11797929ea7SJunchao Zhang /* create two general index sets isx = isy = {0~63}, which are sequential, but we use them as
11897929ea7SJunchao Zhang general and let PETSc detect the pattern and optimize it */
1199566063dSJacob Faibussowitsch PetscCall(PetscMalloc2(n, &ix, n, &iy));
12097929ea7SJunchao Zhang for (i = 0; i < n; i++) ix[i] = i;
1219566063dSJacob Faibussowitsch PetscCall(ISCreateGeneral(PETSC_COMM_SELF, n, ix, PETSC_COPY_VALUES, &isx));
1229566063dSJacob Faibussowitsch PetscCall(ISDuplicate(isx, &isy));
12397929ea7SJunchao Zhang
12497929ea7SJunchao Zhang /* create a vecscatter that just copies x to y */
1259566063dSJacob Faibussowitsch PetscCall(VecScatterCreate(x, isx, y, isy, &vscat));
1269566063dSJacob Faibussowitsch PetscCall(VecScatterBegin(vscat, x, y, INSERT_VALUES, SCATTER_FORWARD));
1279566063dSJacob Faibussowitsch PetscCall(VecScatterEnd(vscat, x, y, INSERT_VALUES, SCATTER_FORWARD));
12897929ea7SJunchao Zhang
12997929ea7SJunchao Zhang /* view y to check the result. y should be {0~63} */
1309566063dSJacob Faibussowitsch PetscCall(PetscPrintf(PETSC_COMM_WORLD, "\nBefore VecScatterRemap on SGToSG, SEQ vector y is:\n"));
1319566063dSJacob Faibussowitsch PetscCall(VecView(y, PETSC_VIEWER_STDOUT_WORLD));
13297929ea7SJunchao Zhang
13397929ea7SJunchao Zhang /* now call the weird subroutine VecScatterRemap to slightly change the vecscatter.
13497929ea7SJunchao Zhang
1356aad120cSJose E. Roman Create tomap as {32~63,0~31}. Originally, we read from indices {0~64} of seq x to write to y. The remap
13697929ea7SJunchao Zhang does indices[i] = tomap[indices[i]]. Therefore, after the remap, we read from indices{32~63,0~31} of seq x.
13797929ea7SJunchao Zhang */
1389566063dSJacob Faibussowitsch PetscCall(PetscMalloc1(n, &tomap));
1399371c9d4SSatish Balay for (i = 0; i < n / 2; i++) {
1409371c9d4SSatish Balay tomap[i] = i + n / 2;
1419371c9d4SSatish Balay tomap[i + n / 2] = i;
1429371c9d4SSatish Balay };
1439566063dSJacob Faibussowitsch PetscCall(VecScatterRemap(vscat, tomap, NULL));
1449566063dSJacob Faibussowitsch PetscCall(VecScatterBegin(vscat, x, y, INSERT_VALUES, SCATTER_FORWARD));
1459566063dSJacob Faibussowitsch PetscCall(VecScatterEnd(vscat, x, y, INSERT_VALUES, SCATTER_FORWARD));
14697929ea7SJunchao Zhang
14797929ea7SJunchao Zhang /* view y to check the result. y should be {32~63,0~31} */
1489566063dSJacob Faibussowitsch PetscCall(PetscPrintf(PETSC_COMM_WORLD, "After VecScatterRemap on SGToSG, SEQ vector y is:\n"));
1499566063dSJacob Faibussowitsch PetscCall(VecView(y, PETSC_VIEWER_STDOUT_WORLD));
15097929ea7SJunchao Zhang
15197929ea7SJunchao Zhang /* destroy everything before we recreate them in different types */
1529566063dSJacob Faibussowitsch PetscCall(PetscFree2(ix, iy));
1539566063dSJacob Faibussowitsch PetscCall(VecDestroy(&x));
1549566063dSJacob Faibussowitsch PetscCall(VecDestroy(&y));
1559566063dSJacob Faibussowitsch PetscCall(ISDestroy(&isx));
1569566063dSJacob Faibussowitsch PetscCall(ISDestroy(&isy));
1579566063dSJacob Faibussowitsch PetscCall(PetscFree(tomap));
1589566063dSJacob Faibussowitsch PetscCall(VecScatterDestroy(&vscat));
15997929ea7SJunchao Zhang
16097929ea7SJunchao Zhang /* ===================================================================================================
16197929ea7SJunchao Zhang (3) test VecScatterRemap on a sequential general to sequential stride 1 (SGToSS_Stride1) vecscatter
16297929ea7SJunchao Zhang ===================================================================================================
16397929ea7SJunchao Zhang */
16497929ea7SJunchao Zhang n = 64; /* long enough to trigger memcpy optimizations in local scatter */
16597929ea7SJunchao Zhang
16697929ea7SJunchao Zhang /* create two seq vectors x of length n, and y of length n/2 */
16777433607SBarry Smith PetscCall(VecCreateFromOptions(PETSC_COMM_SELF, NULL, 1, n, n, &x));
16877433607SBarry Smith PetscCall(VecCreateFromOptions(PETSC_COMM_SELF, NULL, 1, n / 2, n / 2, &y));
16997929ea7SJunchao Zhang
17097929ea7SJunchao Zhang /* Initialize x as {0~63} */
1719566063dSJacob Faibussowitsch for (i = 0; i < n; i++) PetscCall(VecSetValue(x, i, (PetscScalar)i, INSERT_VALUES));
1729566063dSJacob Faibussowitsch PetscCall(VecAssemblyBegin(x));
1739566063dSJacob Faibussowitsch PetscCall(VecAssemblyEnd(x));
17497929ea7SJunchao Zhang
17597929ea7SJunchao Zhang /* create a general index set isx = {0:63:2}, which actually is a stride IS with first=0, n=32, step=2,
17697929ea7SJunchao Zhang but we use it as general and let PETSc detect the pattern and optimize it. */
1779566063dSJacob Faibussowitsch PetscCall(PetscMalloc2(n / 2, &ix, n / 2, &iy));
17897929ea7SJunchao Zhang for (i = 0; i < n / 2; i++) ix[i] = i * 2;
1799566063dSJacob Faibussowitsch PetscCall(ISCreateGeneral(PETSC_COMM_SELF, n / 2, ix, PETSC_COPY_VALUES, &isx));
18097929ea7SJunchao Zhang
18197929ea7SJunchao Zhang /* create a stride1 index set isy = {0~31}. We intentionally set the step to 1 to trigger optimizations */
1829566063dSJacob Faibussowitsch PetscCall(ISCreateStride(PETSC_COMM_SELF, 32, 0, 1, &isy));
18397929ea7SJunchao Zhang
18497929ea7SJunchao Zhang /* create a vecscatter that just copies even entries of x to y */
1859566063dSJacob Faibussowitsch PetscCall(VecScatterCreate(x, isx, y, isy, &vscat));
1869566063dSJacob Faibussowitsch PetscCall(VecScatterBegin(vscat, x, y, INSERT_VALUES, SCATTER_FORWARD));
1879566063dSJacob Faibussowitsch PetscCall(VecScatterEnd(vscat, x, y, INSERT_VALUES, SCATTER_FORWARD));
18897929ea7SJunchao Zhang
18997929ea7SJunchao Zhang /* view y to check the result. y should be {0:63:2} */
1909566063dSJacob Faibussowitsch PetscCall(PetscPrintf(PETSC_COMM_WORLD, "\nBefore VecScatterRemap on SGToSS_Stride1, SEQ vector y is:\n"));
1919566063dSJacob Faibussowitsch PetscCall(VecView(y, PETSC_VIEWER_STDOUT_WORLD));
19297929ea7SJunchao Zhang
19397929ea7SJunchao Zhang /* now call the weird subroutine VecScatterRemap to slightly change the vecscatter.
19497929ea7SJunchao Zhang
1956aad120cSJose E. Roman Create tomap as {32~63,0~31}. Originally, we read from indices{0:63:2} of seq x to write to y. The remap
19697929ea7SJunchao Zhang does indices[i] = tomap[indices[i]]. Therefore, after the remap, we read from indices{32:63:2,0:31:2} of seq x.
19797929ea7SJunchao Zhang */
1989566063dSJacob Faibussowitsch PetscCall(PetscMalloc1(n, &tomap));
1999371c9d4SSatish Balay for (i = 0; i < n / 2; i++) {
2009371c9d4SSatish Balay tomap[i] = i + n / 2;
2019371c9d4SSatish Balay tomap[i + n / 2] = i;
2029371c9d4SSatish Balay };
2039566063dSJacob Faibussowitsch PetscCall(VecScatterRemap(vscat, tomap, NULL));
2049566063dSJacob Faibussowitsch PetscCall(VecScatterBegin(vscat, x, y, INSERT_VALUES, SCATTER_FORWARD));
2059566063dSJacob Faibussowitsch PetscCall(VecScatterEnd(vscat, x, y, INSERT_VALUES, SCATTER_FORWARD));
20697929ea7SJunchao Zhang
20797929ea7SJunchao Zhang /* view y to check the result. y should be {32:63:2,0:31:2} */
2089566063dSJacob Faibussowitsch PetscCall(PetscPrintf(PETSC_COMM_WORLD, "After VecScatterRemap on SGToSS_Stride1, SEQ vector y is:\n"));
2099566063dSJacob Faibussowitsch PetscCall(VecView(y, PETSC_VIEWER_STDOUT_WORLD));
21097929ea7SJunchao Zhang
21197929ea7SJunchao Zhang /* destroy everything before PetscFinalize */
2129566063dSJacob Faibussowitsch PetscCall(PetscFree2(ix, iy));
2139566063dSJacob Faibussowitsch PetscCall(VecDestroy(&x));
2149566063dSJacob Faibussowitsch PetscCall(VecDestroy(&y));
2159566063dSJacob Faibussowitsch PetscCall(ISDestroy(&isx));
2169566063dSJacob Faibussowitsch PetscCall(ISDestroy(&isy));
2179566063dSJacob Faibussowitsch PetscCall(PetscFree(tomap));
2189566063dSJacob Faibussowitsch PetscCall(VecScatterDestroy(&vscat));
21997929ea7SJunchao Zhang
2209566063dSJacob Faibussowitsch PetscCall(PetscFinalize());
221b122ec5aSJacob Faibussowitsch return 0;
22297929ea7SJunchao Zhang }
22397929ea7SJunchao Zhang
22497929ea7SJunchao Zhang /*TEST
22597929ea7SJunchao Zhang
22697929ea7SJunchao Zhang test:
22797929ea7SJunchao Zhang suffix: 1
22897929ea7SJunchao Zhang nsize: 2
22997929ea7SJunchao Zhang diff_args: -j
23097929ea7SJunchao Zhang requires: double
23197929ea7SJunchao Zhang TEST*/
232