Skip to content

java.lang.ExceptionInInitializerError: unable to load from #4

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
databig wants to merge 1 commit into
base: master
Choose a base branch
from
Open

java.lang.ExceptionInInitializerError: unable to load from #4

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
55 changes: 19 additions & 36 deletions src/main/scala/gust/linalg/cuda/CuLU.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -122,17 +122,7 @@ object CuLU extends UFunc {
// passing the pointer instead of the whole matrix allows us to
// do C-style pointer manipulation
def LUSingleBlockFloat(M: Int, N: Int, ADataPointer: CuPointer, pOffset: Int): Unit = {
// val d_A = CuMatrix.create[Float](A.rows, A.cols)
// // creating a copy of the matrix
// // JCuda.cudaMemcpy2D(d_A.data.toCuPointer, A.majorStride * A.elemSize,
// // A.data.toCuPointer,
// // A.majorStride * A.elemSize, A.cols * A.elemSize, A.rows, cudaMemcpyKind.cudaMemcpyDeviceToDevice)
// d_A := A

// val M = A.rows
// val N = A.cols
val minDim = if (M < N) M else N
//val ADataPointer = d_A.offsetPointer
val intRes = Array(0)
val intResPtr = jcuda.Pointer.to(intRes)
val A_ii = Array(0.0f)
Expand All @@ -145,24 +135,18 @@ object CuLU extends UFunc {

val pivotRow = i + intRes(0) - 1 // -1 because of cublas' 1-based indexing
val ip1 = i + 1
P(i + pOffset) = pivotRow + pOffset
P(i + pOffset) = pivotRow

if (pivotRow != i) {
// If you put N instead of A.cols, you have to uncomment the lines in LUBlocked
JCublas2.cublasSswap(handle, N, ADataPointer.withByteOffset(d_A.linearIndex(pivotRow, 0) * es),
lda, ADataPointer.withByteOffset(d_A.linearIndex(i, 0) * es), lda)
}

JCuda.cudaMemcpy2D(AiiPtr, d_A.majorStride * d_A.elemSize,
ADataPointer.withByteOffset(d_A.linearIndex(i, i) * es),
d_A.majorStride * es, d_A.elemSize, 1, cudaMemcpyKind.cudaMemcpyDeviceToHost)
// this one works as well:
// curesult = JCublas2.cublasGetVector(1, es.toInt, ADataPointer.withByteOffset(d_A.linearIndex(i, i) * es),
// 1, AiiPtr, 1)
JCublas2.cublasGetVector(1, es.toInt, ADataPointer.withByteOffset(d_A.linearIndex(i, i) * es),
1, AiiPtr, 1)

if (Math.abs(A_ii(0)) < 1e-20) {
println("Matrix is singular")
return
throw new MatrixSingularException
}

if (ip1 < M) {
Expand Down Expand Up @@ -190,20 +174,20 @@ object CuLU extends UFunc {
val minSize = if (M < N) M else N
val ADataPointer = d_A.offsetPointer

if (blockSize >= minSize) {
if (blockSize >= minSize || blockSize == 1) {
LUSingleBlockFloat(M, N, ADataPointer, 0)
return
}

cfor(0)(_ < minSize, _ + blockSize) { i => {

val realBlockSize = if (minSize - i < blockSize) minSize - i else blockSize
LUSingleBlockFloat(M - i, realBlockSize, ADataPointer.withByteOffset(d_A.linearIndex(i, i) * es), i)

// uncomment these lines if necessary
val mm = if (M < i + realBlockSize) M else i + realBlockSize

cfor(i)(_ < mm - 1, _ + 1) { p => {
// P(p) = P(p) + i // ???
cfor(i)(_ < mm , _ + 1) { p => {
P(p) = P(p) + i // ???
if (P(p) != p) {
JCublas2.cublasSswap(handle, i, ADataPointer.withByteOffset(d_A.linearIndex(p, 0) * es),
lda, ADataPointer.withByteOffset(d_A.linearIndex(P(p), 0) * es), lda)
Expand Down Expand Up @@ -286,24 +270,23 @@ object CuLU extends UFunc {

val pivotRow = i + intRes(0) - 1 // -1 because of cublas' 1-based indexing
val ip1 = i + 1
P(i + pOffset) = pivotRow + pOffset
P(i + pOffset) = pivotRow

if (pivotRow != i) {
// A.cols <-> N + uncomment in LUBlocked
JCublas2.cublasDswap(handle, N, ADataPointer.withByteOffset(d_A.linearIndex(pivotRow, 0) * es),
lda, ADataPointer.withByteOffset(d_A.linearIndex(i, 0) * es), lda)
}

JCuda.cudaMemcpy2D(AiiPtr, d_A.majorStride * d_A.elemSize,
ADataPointer.withByteOffset(d_A.linearIndex(i, i) * es),
d_A.majorStride * es, d_A.elemSize, 1, cudaMemcpyKind.cudaMemcpyDeviceToHost)
// this one works as well:
// curesult = JCublas2.cublasGetVector(1, es.toInt, ADataPointer.withByteOffset(d_A.linearIndex(i, i) * es),
// 1, AiiPtr, 1)
// JCuda.cudaMemcpy2D(AiiPtr, d_A.majorStride * d_A.elemSize,
// ADataPointer.withByteOffset(d_A.linearIndex(i, i) * es),
// d_A.majorStride * es, d_A.elemSize, 1, cudaMemcpyKind.cudaMemcpyDeviceToHost)
// // this one works as well:
JCublas2.cublasGetVector(1, es.toInt, ADataPointer.withByteOffset(d_A.linearIndex(i, i) * es),
1, AiiPtr, 1)

if (Math.abs(A_ii(0)) < 1e-20) {
println("Matrix is singular")
return
throw new MatrixSingularException
}

if (ip1 < M) {
Expand Down Expand Up @@ -331,7 +314,7 @@ object CuLU extends UFunc {
val minSize = if (M < N) M else N
val ADataPointer = d_A.offsetPointer

if (blockSize >= minSize) {
if (blockSize >= minSize || blockSize == 1) {
LUSingleBlockDouble(M, N, ADataPointer, 0)
return
}
Expand All @@ -342,8 +325,8 @@ object CuLU extends UFunc {

val mm = if (M < i + realBlockSize) M else i + realBlockSize

cfor(i)(_ < mm - 1, _ + 1) { p => {
//P(p) = P(p) + i // ???
cfor(i)(_ < mm , _ + 1) { p => {
P(p) = P(p) + i // ???
if (P(p) != p) {
JCublas2.cublasDswap(handle, i, ADataPointer.withByteOffset(d_A.linearIndex(p, 0) * es),
lda, ADataPointer.withByteOffset(d_A.linearIndex(P(p), 0) * es), lda)
Expand Down