L1 Gauss Seidel preconditioner #191
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change | ||||||
|---|---|---|---|---|---|---|---|---|
| @@ -0,0 +1,35 @@ | ||||||||
| ######################################### | ||||||||
| ## CUDA L1 Gauss Seidel Preconditioner ## | ||||||||
| ######################################### | ||||||||
|
|
||||||||
| # PIRACY ALERT: this code exhibit piratic nature because both `adapt` and its arguments are foreign objects. | ||||||||
| # Therefore, `adapt` behavior is going to be different depending on whether `Thunderbolt` and `CuThunderboltExt` are loaded or not. | ||||||||
| # Reference: https://juliatesting.github.io/Aqua.jl/stable/piracies/ | ||||||||
| # Note: the problem is with `AbstractSparseMatrix` as the default behavior of `adapt` is to return the same object, whatever the backend is. | ||||||||
| # Adapt.adapt(::CUDABackend, A::CUSPARSE.AbstractCuSparseMatrix) = A | ||||||||
| # Adapt.adapt(::CUDABackend,A::AbstractSparseMatrix) = A |> cu | ||||||||
| # Adapt.adapt(::CUDABackend, x::Vector) = x |> cu # not needed | ||||||||
| # Adapt.adapt(::CUDABackend, x::CuVector) = x # not needed | ||||||||
|
|
||||||||
| # TODO: remove this function if back compatibility is not needed | ||||||||
|
There was a problem hiding this comment. Backward compat with what? |
||||||||
| Preconditioners.convert_to_backend(::CUDABackend, A::AbstractSparseMatrix) = adapt(CUDABackend(), A) | ||||||||
|
|
||||||||
|
|
||||||||
| # For some reason, these properties are not automatically defined for Device Arrays, | ||||||||
| # TODO: remove the following code when https://github.com/JuliaGPU/CUDA.jl/pull/2738 is merged | ||||||||
| #SparseArrays.rowvals(A::CUSPARSE.CuSparseDeviceMatrixCSC{Tv,Ti,1}) where {Tv,Ti} = A.rowVal | ||||||||
| #SparseArrays.getcolptr(A::CUSPARSE.CuSparseDeviceMatrixCSC{Tv,Ti,1}) where {Tv,Ti} = A.colPtr | ||||||||
| #SparseArrays.getnzval(A::CUSPARSE.CuSparseDeviceMatrixCSC{Tv,Ti,1}) where {Tv,Ti} = A.nzVal | ||||||||
| #SparseMatricesCSR.getnzval(A::CUSPARSE.CuSparseDeviceMatrixCSR{Tv,Ti,1}) where {Tv,Ti} = A.nzVal | ||||||||
|
|
||||||||
| # PIRACY ALERT: the following code is commented out to avoid piracy | ||||||||
| # SparseMatricesCSR.colvals(A::CUSPARSE.CuSparseDeviceMatrixCSR{Tv,Ti,1}) where {Tv,Ti} = A.colVal | ||||||||
| # SparseMatricesCSR.getrowptr(A::CUSPARSE.CuSparseDeviceMatrixCSR{Tv,Ti,1}) where {Tv,Ti} = A.rowPtr | ||||||||
|
Comment on lines
+6
to
+7
There was a problem hiding this comment. these two lines should be added in JuliaGPU/CUDA.jl#2720, right ? There was a problem hiding this comment. Actually I am not sure about this one. Probably not, because SparseMatricesCSR is not an interface package. I would discuss this in the linked PR.
Comment on lines
+5
to
+7
There was a problem hiding this comment.
Suggested change
|
||||||||
|
|
||||||||
| # workaround for the issue with SparseMatricesCSR | ||||||||
| # TODO: find a more robust solution to dispatch the correct function | ||||||||
| Preconditioners.colvals(A::CUSPARSE.CuSparseDeviceMatrixCSR{Tv,Ti,1}) where {Tv,Ti} = A.colVal | ||||||||
| Preconditioners.getrowptr(A::CUSPARSE.CuSparseDeviceMatrixCSR{Tv,Ti,1}) where {Tv,Ti} = A.rowPtr | ||||||||
|
|
||||||||
| Preconditioners.sparsemat_format_type(::CUSPARSE.CuSparseDeviceMatrixCSC{Tv,Ti,1}) where {Tv,Ti} = CSCFormat | ||||||||
Abdelrahman912 marked this conversation as resolved.
Show resolved
Hide resolved
|
||||||||
| Preconditioners.sparsemat_format_type(::CUSPARSE.CuSparseDeviceMatrixCSR{Tv,Ti,1}) where {Tv,Ti} = CSRFormat | ||||||||
|
There was a problem hiding this comment. Can we remove these here for now and add a comment to the tests that prior to 2720 we need to add these dispatches? I do not want to artificially postpone this PR here further due to missing optional stuff in the dependencies. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,20 @@ | ||
| # remove the following code once PR (https://github.com/JuliaGPU/CUDA.jl/pull/2720) is merged ## | ||
| CUDA.CUSPARSE.CuSparseMatrixCSR{T}(Mat::SparseMatrixCSR) where {T} = | ||
| CUDA.CUSPARSE.CuSparseMatrixCSR{T}(CuVector{Cint}(Mat.rowptr), CuVector{Cint}(Mat.colval), | ||
| CuVector{T}(Mat.nzval), size(Mat)) | ||
|
|
||
|
|
||
| CUSPARSE.CuSparseMatrixCSC{T}(Mat::SparseMatrixCSR) where {T} = | ||
| CUSPARSE.CuSparseMatrixCSC{T}(CUSPARSE.CuSparseMatrixCSR(Mat)) | ||
|
|
||
| SparseMatricesCSR.SparseMatrixCSR(A::CUSPARSE.CuSparseMatrixCSR) = | ||
| SparseMatrixCSR(CUSPARSE.SparseMatrixCSC(A)) # no direct conversion (gpu_CSR -> cpu_CSC -> cpu_CSR) | ||
|
|
||
| Adapt.adapt_storage(::Type{CuArray}, xs::SparseMatrixCSR) = | ||
| CUSPARSE.CuSparseMatrixCSR(xs) | ||
|
|
||
| Adapt.adapt_storage(::Type{CuArray{T}}, xs::SparseMatrixCSR) where {T} = | ||
| CUSPARSE.CuSparseMatrixCSR{T}(xs) | ||
|
|
||
| Adapt.adapt_storage(::Type{Array}, mat::CUSPARSE.CuSparseMatrixCSR) = | ||
| SparseMatrixCSR(mat) | ||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,48 @@ | ||
| module Preconditioners | ||
|
|
||
| using SparseArrays, SparseMatricesCSR | ||
| using LinearSolve | ||
| import LinearSolve: \ | ||
| using Adapt | ||
| using UnPack | ||
| import KernelAbstractions: Backend, @kernel, @index, @ndrange, @groupsize, @print, functional, | ||
| CPU,synchronize | ||
| import SparseArrays: getcolptr,getnzval | ||
| import SparseMatricesCSR: getnzval | ||
| import LinearAlgebra: Symmetric | ||
|
|
||
| ## Generic Code # | ||
|
|
||
| # CSR and CSC are exact the same in symmetric matrices,so we need to hold symmetry info | ||
| # in order to be exploited in cases in which one format has better access pattern than the other. | ||
| abstract type AbstractMatrixSymmetry end | ||
| struct SymmetricMatrix <: AbstractMatrixSymmetry end | ||
| struct NonSymmetricMatrix <: AbstractMatrixSymmetry end | ||
termi-official marked this conversation as resolved.
Show resolved
Hide resolved
|
||
|
|
||
| abstract type AbstractMatrixFormat end | ||
| struct CSRFormat <: AbstractMatrixFormat end | ||
| struct CSCFormat <: AbstractMatrixFormat end | ||
|
|
||
|
|
||
| # Why using these traits? | ||
| # Since we are targeting multiple backends, but unfortunately, all the sparse matrix CSC/CSR on all | ||
| # backends don't share the same supertype (e.g. AbstractSparseMatrixCSC/AbstractSparseMatrixCSR) | ||
| # e.g. CUSPARSE.CuSparseDeviceMatrixCSC <:SparseArrays.AbstractSparseMatrixCSC → false | ||
| # So we need to define our own traits to identify the format of the sparse matrix | ||
| sparsemat_format_type(::SparseMatrixCSC) = CSCFormat | ||
| sparsemat_format_type(::SparseMatrixCSR) = CSRFormat | ||
|
|
||
| #TODO: remove once https://github.com/JuliaGPU/CUDA.jl/pull/2740 is merged | ||
| convert_to_backend(backend::Backend, A::AbstractSparseMatrix) = | ||
| adapt(backend, A) # fallback value, specific backends are to be extended in their corresponding extensions. | ||
|
|
||
| # Why? because we want to circumvent piracy when extending these functions for device backend (e.g. CuSparseDeviceMatrixCSR) | ||
| # TODO: find a more robust solution to dispatch the correct function | ||
| colvals(A::SparseMatrixCSR) = SparseMatricesCSR.colvals(A) | ||
| getrowptr(A::SparseMatrixCSR) = SparseMatricesCSR.getrowptr(A) | ||
|
|
||
| include("l1_gauss_seidel.jl") | ||
|
|
||
| export L1GSPrecBuilder | ||
|
|
||
| end | ||
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Can you add a compat entry for the new CUDA release?