Skip to content

Enable multiple precisions in CeedVector #948

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
wants to merge 12 commits into
base: main
Choose a base branch
from
Draft

Enable multiple precisions in CeedVector #948

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
7016e0b
WIP multiprecision vectors: Change Vector read/write/set backend func…
nbeams Apr 4, 2022
775eb4c
WIP multiprecision vectors: adding multiprecision storage and some fc…
nbeams Apr 6, 2022
3b67d1a
WIP hip-ref: use backend fcn for sync, streamline logic
nbeams Apr 20, 2022
88a3338
WIP: make style
nbeams Apr 20, 2022
3e94c0d
Add specific FP32/FP64 Vector access functions to interface
nbeams Apr 20, 2022
fd75fdf
WIP, experimental: add user-level function for checking array status …
nbeams Apr 22, 2022
80606a9
WIP: use Generic vector access functions and void pointers in other b…
nbeams Apr 26, 2022
5c8e206
include: cleaner CEED_NUM_PRECISIONS
jedbrown May 3, 2022
df27fe0
WIP: backends/cuda: fix type mismatch
jedbrown May 3, 2022
ed90cc3
Merge branch 'main' into natalie/multiprec-vec
nbeams May 26, 2022
36d404f
Merge branch 'main' into natalie/multiprec-vec
nbeams Aug 26, 2022
112a04a
Resolve conflicts when updating from main
nbeams Aug 26, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
85 changes: 49 additions & 36 deletions backends/cuda-ref/ceed-cuda-vector.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -112,8 +112,9 @@ static inline int CeedVectorSyncD2H_Cuda(const CeedVector vec) {
//------------------------------------------------------------------------------
// Sync arrays
//------------------------------------------------------------------------------
static int CeedVectorSyncArray_Cuda(const CeedVector vec,
CeedMemType mem_type) {
static int CeedVectorSyncArrayGeneric_Cuda(const CeedVector vec,
CeedMemType mem_type,
CeedScalarType prec) {
int ierr;
// Check whether device/host sync is needed
bool need_sync = false;
Expand All @@ -123,8 +124,8 @@ static int CeedVectorSyncArray_Cuda(const CeedVector vec,
return CEED_ERROR_SUCCESS;

switch (mem_type) {
case CEED_MEM_HOST: return CeedVectorSyncD2H_Cuda(vec);
case CEED_MEM_DEVICE: return CeedVectorSyncH2D_Cuda(vec);
case CEED_MEM_HOST: return CeedVectorSyncD2H_Cuda(vec, prec);
case CEED_MEM_DEVICE: return CeedVectorSyncH2D_Cuda(vec, prec);
}
return CEED_ERROR_UNSUPPORTED;
}
Expand Down Expand Up @@ -182,7 +183,8 @@ static inline int CeedVectorHasArrayOfType_Cuda(const CeedVector vec,
// Check if has borrowed array of given type
//------------------------------------------------------------------------------
static inline int CeedVectorHasBorrowedArrayOfType_Cuda(const CeedVector vec,
CeedMemType mem_type, bool *has_borrowed_array_of_type) {
CeedMemType mem_type, CeedScalarType prec,
bool *has_borrowed_array_of_type) {
int ierr;
CeedVector_Cuda *impl;
ierr = CeedVectorGetData(vec, &impl); CeedChkBackend(ierr);
Expand All @@ -203,7 +205,7 @@ static inline int CeedVectorHasBorrowedArrayOfType_Cuda(const CeedVector vec,
// Set array from host
//------------------------------------------------------------------------------
static int CeedVectorSetArrayHost_Cuda(const CeedVector vec,
const CeedCopyMode copy_mode, CeedScalar *array) {
const CeedCopyMode copy_mode, void *array) {
int ierr;
CeedVector_Cuda *impl;
ierr = CeedVectorGetData(vec, &impl); CeedChkBackend(ierr);
Expand Down Expand Up @@ -245,7 +247,7 @@ static int CeedVectorSetArrayHost_Cuda(const CeedVector vec,
// Set array from device
//------------------------------------------------------------------------------
static int CeedVectorSetArrayDevice_Cuda(const CeedVector vec,
const CeedCopyMode copy_mode, CeedScalar *array) {
const CeedCopyMode copy_mode, void *array) {
int ierr;
Ceed ceed;
ierr = CeedVectorGetCeed(vec, &ceed); CeedChkBackend(ierr);
Expand Down Expand Up @@ -289,9 +291,11 @@ static int CeedVectorSetArrayDevice_Cuda(const CeedVector vec,
// Set the array used by a vector,
// freeing any previously allocated array if applicable
//------------------------------------------------------------------------------
static int CeedVectorSetArray_Cuda(const CeedVector vec,
const CeedMemType mem_type,
const CeedCopyMode copy_mode, CeedScalar *array) {
static int CeedVectorSetArrayGeneric_Cuda(const CeedVector vec,
const CeedMemType mem_type,
const CeedScalarType prec,
const CeedCopyMode copy_mode,
void *array) {
int ierr;
Ceed ceed;
ierr = CeedVectorGetCeed(vec, &ceed); CeedChkBackend(ierr);
Expand Down Expand Up @@ -368,8 +372,8 @@ static int CeedVectorSetValue_Cuda(CeedVector vec, CeedScalar val) {
//------------------------------------------------------------------------------
// Vector Take Array
//------------------------------------------------------------------------------
static int CeedVectorTakeArray_Cuda(CeedVector vec, CeedMemType mem_type,
CeedScalar **array) {
static int CeedVectorTakeArrayGeneric_Cuda(CeedVector vec, CeedMemType mem_type,
CeedScalarType prec, void **array) {
int ierr;
Ceed ceed;
ierr = CeedVectorGetCeed(vec, &ceed); CeedChkBackend(ierr);
Expand Down Expand Up @@ -401,7 +405,9 @@ static int CeedVectorTakeArray_Cuda(CeedVector vec, CeedMemType mem_type,
// If a different memory type is most up to date, this will perform a copy
//------------------------------------------------------------------------------
static int CeedVectorGetArrayCore_Cuda(const CeedVector vec,
const CeedMemType mem_type, CeedScalar **array) {
const CeedMemType mem_type,
const CeedScalarType prec,
void **array) {
int ierr;
Ceed ceed;
ierr = CeedVectorGetCeed(vec, &ceed); CeedChkBackend(ierr);
Expand All @@ -426,21 +432,26 @@ static int CeedVectorGetArrayCore_Cuda(const CeedVector vec,
//------------------------------------------------------------------------------
// Get read-only access to a vector via the specified mem_type
//------------------------------------------------------------------------------
static int CeedVectorGetArrayRead_Cuda(const CeedVector vec,
const CeedMemType mem_type, const CeedScalar **array) {
return CeedVectorGetArrayCore_Cuda(vec, mem_type, (CeedScalar **)array);
static int CeedVectorGetArrayReadGeneric_Cuda(const CeedVector vec,
const CeedMemType mem_type,
const CeedScalarType prec,
const void **array) {
return CeedVectorGetArrayCore_Cuda(vec, mem_type, prec, (void **)array);
}

//------------------------------------------------------------------------------
// Get read/write access to a vector via the specified mem_type
// Get read/write access to a vector via the specified mem_type and precision
//------------------------------------------------------------------------------
static int CeedVectorGetArray_Cuda(const CeedVector vec,
const CeedMemType mem_type, CeedScalar **array) {
const CeedMemType mem_type,
const CeedScalarType prec,
void **array) {
int ierr;
CeedVector_Cuda *impl;
ierr = CeedVectorGetData(vec, &impl); CeedChkBackend(ierr);

ierr = CeedVectorGetArrayCore_Cuda(vec, mem_type, array); CeedChkBackend(ierr);
ierr = CeedVectorGetArrayCore_Cuda(vec, mem_type, prec, array);
CeedChkBackend(ierr);

ierr = CeedVectorSetAllInvalid_Cuda(vec); CeedChkBackend(ierr);
switch (mem_type) {
Expand All @@ -456,10 +467,12 @@ static int CeedVectorGetArray_Cuda(const CeedVector vec,
}

//------------------------------------------------------------------------------
// Get write access to a vector via the specified mem_type
// Get write access to a vector via the specified mem_type and precision
//------------------------------------------------------------------------------
static int CeedVectorGetArrayWrite_Cuda(const CeedVector vec,
const CeedMemType mem_type, CeedScalar **array) {
static int CeedVectorGetArrayWriteGeneric_Cuda(const CeedVector vec,
const CeedMemType mem_type,
const CeedScalarType prec,
void **array) {
int ierr;
CeedVector_Cuda *impl;
ierr = CeedVectorGetData(vec, &impl); CeedChkBackend(ierr);
Expand All @@ -469,7 +482,7 @@ static int CeedVectorGetArrayWrite_Cuda(const CeedVector vec,
CeedChkBackend(ierr);
if (!has_array_of_type) {
// Allocate if array is not yet allocated
ierr = CeedVectorSetArray(vec, mem_type, CEED_COPY_VALUES, NULL);
ierr = CeedVectorSetArrayGeneric(vec, mem_type, prec, CEED_COPY_VALUES, NULL);
CeedChkBackend(ierr);
} else {
// Select dirty array
Expand All @@ -488,7 +501,7 @@ static int CeedVectorGetArrayWrite_Cuda(const CeedVector vec,
}
}

return CeedVectorGetArray_Cuda(vec, mem_type, array);
return CeedVectorGetArrayGeneric_Cuda(vec, mem_type, prec, array);
}

//------------------------------------------------------------------------------
Expand Down Expand Up @@ -755,21 +768,21 @@ int CeedVectorCreate_Cuda(CeedSize n, CeedVector vec) {
ierr = CeedSetBackendFunction(ceed, "Vector", vec, "HasBorrowedArrayOfType",
CeedVectorHasBorrowedArrayOfType_Cuda);
CeedChkBackend(ierr);
ierr = CeedSetBackendFunction(ceed, "Vector", vec, "SetArray",
CeedVectorSetArray_Cuda); CeedChkBackend(ierr);
ierr = CeedSetBackendFunction(ceed, "Vector", vec, "TakeArray",
CeedVectorTakeArray_Cuda); CeedChkBackend(ierr);
ierr = CeedSetBackendFunction(ceed, "Vector", vec, "SetArrayGeneric",
CeedVectorSetArrayGeneric_Cuda); CeedChkBackend(ierr);
ierr = CeedSetBackendFunction(ceed, "Vector", vec, "TakeArrayGeneric",
CeedVectorTakeArrayGeneric_Cuda); CeedChkBackend(ierr);
ierr = CeedSetBackendFunction(ceed, "Vector", vec, "SetValue",
(int (*)())(CeedVectorSetValue_Cuda));
CeedChkBackend(ierr);
ierr = CeedSetBackendFunction(ceed, "Vector", vec, "SyncArray",
CeedVectorSyncArray_Cuda); CeedChkBackend(ierr);
ierr = CeedSetBackendFunction(ceed, "Vector", vec, "GetArray",
CeedVectorGetArray_Cuda); CeedChkBackend(ierr);
ierr = CeedSetBackendFunction(ceed, "Vector", vec, "GetArrayRead",
CeedVectorGetArrayRead_Cuda); CeedChkBackend(ierr);
ierr = CeedSetBackendFunction(ceed, "Vector", vec, "GetArrayWrite",
CeedVectorGetArrayWrite_Cuda); CeedChkBackend(ierr);
ierr = CeedSetBackendFunction(ceed, "Vector", vec, "SyncArrayGeneric",
CeedVectorSyncArrayGeneric_Cuda); CeedChkBackend(ierr);
ierr = CeedSetBackendFunction(ceed, "Vector", vec, "GetArrayGeneric",
CeedVectorGetArrayGeneric_Cuda); CeedChkBackend(ierr);
ierr = CeedSetBackendFunction(ceed, "Vector", vec, "GetArrayReadGeneric",
CeedVectorGetArrayReadGeneric_Cuda); CeedChkBackend(ierr);
ierr = CeedSetBackendFunction(ceed, "Vector", vec, "GetArrayWriteGeneric",
CeedVectorGetArrayWriteGeneric_Cuda); CeedChkBackend(ierr);
ierr = CeedSetBackendFunction(ceed, "Vector", vec, "Norm",
CeedVectorNorm_Cuda); CeedChkBackend(ierr);
ierr = CeedSetBackendFunction(ceed, "Vector", vec, "Reciprocal",
Expand Down
32 changes: 20 additions & 12 deletions backends/hip-ref/ceed-hip-ref-basis.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,12 +29,16 @@ int CeedBasisApply_Hip(CeedBasis basis, const CeedInt num_elem,
const int max_block_size = 64;

// Read vectors
const CeedScalar *d_u;
CeedScalar *d_v;
const void *d_u;
void *d_v;
if (eval_mode != CEED_EVAL_WEIGHT) {
ierr = CeedVectorGetArrayRead(u, CEED_MEM_DEVICE, &d_u); CeedChkBackend(ierr);
ierr = CeedVectorGetArrayReadGeneric(u, CEED_MEM_DEVICE, CEED_SCALAR_TYPE,
&d_u);
CeedChkBackend(ierr);
}
ierr = CeedVectorGetArrayWrite(v, CEED_MEM_DEVICE, &d_v); CeedChkBackend(ierr);
ierr = CeedVectorGetArrayWriteGeneric(v, CEED_MEM_DEVICE, CEED_SCALAR_TYPE,
&d_v);
CeedChkBackend(ierr);

// Clear v for transpose operation
if (t_mode == CEED_TRANSPOSE) {
Expand Down Expand Up @@ -93,9 +97,9 @@ int CeedBasisApply_Hip(CeedBasis basis, const CeedInt num_elem,

// Restore vectors
if (eval_mode != CEED_EVAL_WEIGHT) {
ierr = CeedVectorRestoreArrayRead(u, &d_u); CeedChkBackend(ierr);
ierr = CeedVectorRestoreArrayReadGeneric(u, &d_u); CeedChkBackend(ierr);
}
ierr = CeedVectorRestoreArray(v, &d_v); CeedChkBackend(ierr);
ierr = CeedVectorRestoreArrayGeneric(v, &d_v); CeedChkBackend(ierr);
return CEED_ERROR_SUCCESS;
}

Expand All @@ -121,12 +125,16 @@ int CeedBasisApplyNonTensor_Hip(CeedBasis basis, const CeedInt num_elem,
num_elem/elemsPerBlock*elemsPerBlock<num_elem)?1:0);

// Read vectors
const CeedScalar *d_u;
CeedScalar *d_v;
const void *d_u;
void *d_v;
if (eval_mode != CEED_EVAL_WEIGHT) {
ierr = CeedVectorGetArrayRead(u, CEED_MEM_DEVICE, &d_u); CeedChkBackend(ierr);
ierr = CeedVectorGetArrayReadGeneric(u, CEED_MEM_DEVICE, CEED_SCALAR_TYPE,
&d_u);
CeedChkBackend(ierr);
}
ierr = CeedVectorGetArrayWrite(v, CEED_MEM_DEVICE, &d_v); CeedChkBackend(ierr);
ierr = CeedVectorGetArrayWriteGeneric(v, CEED_MEM_DEVICE, CEED_SCALAR_TYPE,
&d_v);
CeedChkBackend(ierr);

// Clear v for transpose operation
if (t_mode == CEED_TRANSPOSE) {
Expand Down Expand Up @@ -183,9 +191,9 @@ int CeedBasisApplyNonTensor_Hip(CeedBasis basis, const CeedInt num_elem,

// Restore vectors
if (eval_mode != CEED_EVAL_WEIGHT) {
ierr = CeedVectorRestoreArrayRead(u, &d_u); CeedChkBackend(ierr);
ierr = CeedVectorRestoreArrayReadGeneric(u, &d_u); CeedChkBackend(ierr);
}
ierr = CeedVectorRestoreArray(v, &d_v); CeedChkBackend(ierr);
ierr = CeedVectorRestoreArrayGeneric(v, &d_v); CeedChkBackend(ierr);
return CEED_ERROR_SUCCESS;
}

Expand Down
Loading