Skip to content

fix: Restore HPCG_MEMMGMT functionality by fixing memory allocation #2

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

Open
wants to merge 2 commits into
base: develop
Choose a base branch
from
Open

fix: Restore HPCG_MEMMGMT functionality by fixing memory allocation #2

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4112c21
fix: Restore HPCG_MEMMGMT functionality by fixing memory allocation/d…
dmikushin Oct 10, 2025
1a17cb6
fix: Fix remaining deviceDefrag errors in ExtractDiagonal and Permute…
dmikushin Oct 10, 2025
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
4 changes: 2 additions & 2 deletions src/GenerateCoarseProblem.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -124,8 +124,8 @@ void GenerateCoarseProblem(const SparseMatrix & Af) {
local_int_t* d_f2cOperator;
local_int_t* d_c2fOperator;

GPU_CHECK(gpuMalloc((void**)&d_f2cOperator, sizeof(local_int_t) * localNumberOfRows));
GPU_CHECK(gpuMalloc((void**)&d_c2fOperator, sizeof(local_int_t) * nxf * nyf * nzf));
GPU_CHECK(deviceMalloc((void**)&d_f2cOperator, sizeof(local_int_t) * localNumberOfRows));
GPU_CHECK(deviceMalloc((void**)&d_c2fOperator, sizeof(local_int_t) * nxf * nyf * nzf));

GPU_CHECK(gpuMemset(d_c2fOperator, -1, sizeof(local_int_t) * nxf * nyf * nzf));

Expand Down
18 changes: 9 additions & 9 deletions src/GenerateProblem.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -406,13 +406,13 @@ void GenerateProblem(SparseMatrix & A, Vector * b, Vector * x, Vector * xexact)
if(xexact != NULL) HIPInitializeVector(*xexact, localNumberOfRows);

// Allocate structures
GPU_CHECK(gpuMalloc((void**)&A.d_mtxIndG, std::max(sizeof(double), sizeof(global_int_t)) * localNumberOfRows * numberOfNonzerosPerRow));
GPU_CHECK(gpuMalloc((void**)&A.d_matrixValues, sizeof(double) * localNumberOfRows * numberOfNonzerosPerRow));
GPU_CHECK(gpuMalloc((void**)&A.d_mtxIndL, sizeof(local_int_t) * localNumberOfRows * numberOfNonzerosPerRow));
GPU_CHECK(gpuMalloc((void**)&A.d_nonzerosInRow, sizeof(char) * localNumberOfRows));
GPU_CHECK(gpuMalloc((void**)&A.d_matrixDiagonal, sizeof(local_int_t) * localNumberOfRows));
GPU_CHECK(gpuMalloc((void**)&A.d_rowHash, sizeof(local_int_t) * localNumberOfRows));
GPU_CHECK(gpuMalloc((void**)&A.d_localToGlobalMap, sizeof(global_int_t) * localNumberOfRows));
GPU_CHECK(deviceMalloc((void**)&A.d_mtxIndG, std::max(sizeof(double), sizeof(global_int_t)) * localNumberOfRows * numberOfNonzerosPerRow));
GPU_CHECK(deviceMalloc((void**)&A.d_matrixValues, sizeof(double) * localNumberOfRows * numberOfNonzerosPerRow));
GPU_CHECK(deviceMalloc((void**)&A.d_mtxIndL, sizeof(local_int_t) * localNumberOfRows * numberOfNonzerosPerRow));
GPU_CHECK(deviceMalloc((void**)&A.d_nonzerosInRow, sizeof(char) * localNumberOfRows));
GPU_CHECK(deviceMalloc((void**)&A.d_matrixDiagonal, sizeof(local_int_t) * localNumberOfRows));
GPU_CHECK(deviceMalloc((void**)&A.d_rowHash, sizeof(local_int_t) * localNumberOfRows));
GPU_CHECK(deviceMalloc((void**)&A.d_localToGlobalMap, sizeof(global_int_t) * localNumberOfRows));

// Determine blocksize
unsigned int blocksize = 512 / numberOfNonzerosPerRow;
Expand Down Expand Up @@ -511,8 +511,8 @@ void CopyProblemToHost(SparseMatrix& A, Vector* b, Vector* x, Vector* xexact)
GPU_CHECK(gpuMemcpy(mtxDiag, A.d_matrixDiagonal, sizeof(local_int_t) * A.localNumberOfRows, gpuMemcpyDeviceToHost));
GPU_CHECK(gpuMemcpy(A.localToGlobalMap.data(), A.d_localToGlobalMap, sizeof(global_int_t) * A.localNumberOfRows, gpuMemcpyDeviceToHost));

GPU_CHECK(gpuFree(A.d_nonzerosInRow));
GPU_CHECK(gpuFree(A.d_matrixDiagonal));
GPU_CHECK(deviceFree(A.d_nonzerosInRow));
GPU_CHECK(deviceFree(A.d_matrixDiagonal));

// Initialize pointers
A.matrixDiagonal[0] = A.matrixValues[0] + mtxDiag[0];
Expand Down
20 changes: 10 additions & 10 deletions src/MultiColoring.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -245,7 +245,7 @@ void JPLColoring(SparseMatrix& A)
{
local_int_t m = A.localNumberOfRows;

GPU_CHECK(gpuMalloc((void**)&A.perm, sizeof(local_int_t) * m));
GPU_CHECK(deviceMalloc((void**)&A.perm, sizeof(local_int_t) * m));
GPU_CHECK(gpuMemset(A.perm, -1, sizeof(local_int_t) * m));

A.nblocks = 0;
Expand Down Expand Up @@ -321,15 +321,15 @@ void JPLColoring(SparseMatrix& A)

A.ublocks = A.nblocks - 1;

GPU_CHECK(gpuFree(A.d_rowHash));
GPU_CHECK(deviceFree(A.d_rowHash));

local_int_t* tmp_color;
local_int_t* tmp_perm;
local_int_t* perm;

GPU_CHECK(gpuMalloc((void**)&tmp_color, sizeof(local_int_t) * m));
GPU_CHECK(gpuMalloc((void**)&tmp_perm, sizeof(local_int_t) * m));
GPU_CHECK(gpuMalloc((void**)&perm, sizeof(local_int_t) * m));
GPU_CHECK(deviceMalloc((void**)&tmp_color, sizeof(local_int_t) * m));
GPU_CHECK(deviceMalloc((void**)&tmp_perm, sizeof(local_int_t) * m));
GPU_CHECK(deviceMalloc((void**)&perm, sizeof(local_int_t) * m));

kernel_identity<1024><<<(m - 1) / 1024 + 1, 1024>>>(m, perm);

Expand All @@ -344,9 +344,9 @@ void JPLColoring(SparseMatrix& A)
int endbit = 32 - __builtin_clz(A.nblocks);

GPU_CHECK(rocprim::radix_sort_pairs(buf, size, keys, vals, m, startbit, endbit));
GPU_CHECK(gpuMalloc(&buf, size));
GPU_CHECK(deviceMalloc(&buf, size));
GPU_CHECK(rocprim::radix_sort_pairs(buf, size, keys, vals, m, startbit, endbit));
GPU_CHECK(gpuFree(buf));
GPU_CHECK(deviceFree(buf));

kernel_create_perm<1024><<<(m - 1) / 1024 + 1, 1024>>>(m, vals.current(), A.perm);
#else
Expand All @@ -367,9 +367,9 @@ void JPLColoring(SparseMatrix& A)
kernel_create_perm<1024><<<(m - 1) / 1024 + 1, 1024>>>(m, perm, A.perm);
#endif

GPU_CHECK(gpuFree(tmp_color));
GPU_CHECK(gpuFree(tmp_perm));
GPU_CHECK(gpuFree(perm));
GPU_CHECK(deviceFree(tmp_color));
GPU_CHECK(deviceFree(tmp_perm));
GPU_CHECK(deviceFree(perm));

#ifndef HPCG_REFERENCE
--A.ublocks;
Expand Down
4 changes: 2 additions & 2 deletions src/Permute.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -237,10 +237,10 @@ __global__ void kernel_permute(local_int_t size,
void PermuteVector(local_int_t size, Vector& v, const local_int_t* perm)
{
double* buffer;
GPU_CHECK(gpuMalloc((void**)&buffer, sizeof(double) * v.localLength));
GPU_CHECK(deviceMalloc((void**)&buffer, sizeof(double) * v.localLength));

kernel_permute<1024><<<(size - 1) / 1024 + 1, 1024>>>(size, perm, v.d_values, buffer);

GPU_CHECK(gpuFree(v.d_values));
GPU_CHECK(deviceFree(v.d_values));
v.d_values = buffer;
}
6 changes: 3 additions & 3 deletions src/SparseMatrix.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -330,7 +330,7 @@ void ConvertToELL(SparseMatrix& A)
else LAUNCH_TO_ELL_COL(27, 4)

// Free old matrix indices
GPU_CHECK(gpuFree(A.d_mtxIndL));
GPU_CHECK(deviceFree(A.d_mtxIndL));

#ifndef HPCG_NO_MPI
GPU_CHECK(gpuMemcpy(&A.halo_rows, d_halo_rows, sizeof(local_int_t), gpuMemcpyDeviceToHost));
Expand Down Expand Up @@ -408,8 +408,8 @@ void ExtractDiagonal(SparseMatrix& A)
local_int_t m = A.localNumberOfRows;

// Allocate memory to extract diagonal entries
GPU_CHECK(gpuMalloc((void**)&A.diag_idx, sizeof(local_int_t) * m));
GPU_CHECK(gpuMalloc((void**)&A.inv_diag, sizeof(double) * m));
GPU_CHECK(deviceMalloc((void**)&A.diag_idx, sizeof(local_int_t) * m));
GPU_CHECK(deviceMalloc((void**)&A.inv_diag, sizeof(double) * m));

// Extract diagonal entries
kernel_extract_diag_index<1024><<<(m - 1) / 1024 + 1, 1024>>>(
Expand Down