Skip to content

Commit

Permalink
[NDTensors] [ITensors] [ITensorGPU] Bump NDTensors v0.1.46, ITensors …
Browse files Browse the repository at this point in the history 
…v0.3.26, CUDA v4.0, ITensorGPU v0.1.1 (#1053)
  • Loading branch information
@mtfishman
mtfishman authored Feb 7, 2023
1 parent 3f43e24 commit e2a5726
Show file tree
Hide file tree
Showing 6 changed files with 77 additions and 74 deletions.
11 changes: 7 additions & 4 deletions ITensorGPU/Project.toml
View file
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
name = "ITensorGPU"
uuid = "d89171c1-af8f-46b3-badf-d2a472317c15"
authors = ["Katharine Hyatt", "Matthew Fishman <mfishman@flatironinstitute.org>"]
version = "0.1.0"
version = "0.1.1"

[deps]
Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
Expand All @@ -11,23 +11,26 @@ Functors = "d9f16b24-f501-4c13-a1f2-28368ffc5196"
HDF5 = "f67ccb44-e63f-5c2f-98bd-6dc0ccc4ba2f"
ITensors = "9136182c-28ba-11e9-034c-db9fb085ebd5"
LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
NDTensors = "23ae76d9-e61a-49c4-8f12-3f1a16adf9cf"
Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
SimpleTraits = "699a6c99-e7fa-54fc-8d76-47d257e15c1d"
StaticArrays = "90137ffa-7385-5640-81b9-e52037218182"
Strided = "5e0ebb24-38b0-5f93-81fe-25c709ecae67"
TimerOutputs = "a759f4b9-e2f1-59dc-863e-4aeb61b1ea8f"
cuTENSOR = "011b41b2-24ef-40a8-b3eb-fa098493e9e1"

[compat]
Adapt = "3.3, 3.4"
CUDA = "3.5.0"
Adapt = "3.5"
CUDA = "4.0"
Combinatorics = "1.0.2"
Functors = "0.2, 0.3, 0.4"
HDF5 = "0.15.7, 0.16"
ITensors = "0.3.19"
ITensors = "0.3.26"
SimpleTraits = "0.9.4"
StaticArrays = "1.2.13"
Strided = "1.1.2"
TimerOutputs = "0.5.13"
cuTENSOR = "1.0.1"
julia = "1.6"

[extras]
Expand Down
58 changes: 29 additions & 29 deletions ITensorGPU/src/ITensorGPU.jl
View file
Original file line number Diff line number Diff line change
@@ -1,37 +1,26 @@
module ITensorGPU

using Adapt
using CUDA
using CUDA.Adapt
using CUDA.CUTENSOR
using CUDA.CUBLAS
using CUDA.CUSOLVER
using Functors
using ITensors
using LinearAlgebra
using Random, Strided
using TimerOutputs
using NDTensors
using Random
using SimpleTraits
using StaticArrays
using ITensors
using ITensors.NDTensors
using Strided
using TimerOutputs
using cuTENSOR

using NDTensors: setdata, setstorage, cpu, IsWrappedArray, parenttype

import Adapt: adapt_structure
import Base: *, permutedims!
import CUDA: CuArray, CuMatrix, CuVector, cu
import CUDA.CUTENSOR: cutensorContractionPlan_t, cutensorAlgo_t
import CUDA.Adapt: adapt_structure
import CUDA.Mem: pin
#=
const devs = Ref{Vector{CUDAdrv.CuDevice}}()
const dev_rows = Ref{Int}(0)
const dev_cols = Ref{Int}(0)
function __init__()
voltas = filter(dev->occursin("V100", CUDAdrv.name(dev)), collect(CUDAdrv.devices()))
pascals = filter(dev->occursin("P100", CUDAdrv.name(dev)), collect(CUDAdrv.devices()))
devs[] = voltas[1:1]
#devs[] = pascals[1:2]
CUBLASMG.cublasMgDeviceSelect(CUBLASMG.mg_handle(), length(devs[]), devs[])
dev_rows[] = 1
dev_cols[] = 1
end
=#
import ITensors:
randn!,
compute_contraction_labels,
Expand All @@ -50,7 +39,7 @@ import ITensors:
permute,
BroadcastStyle,
Indices
import ITensors.NDTensors:
import NDTensors:
Atrans,
Btrans,
CombinerTensor,
Expand Down Expand Up @@ -88,10 +77,10 @@ import ITensors.NDTensors:
set_ndims,
similartype,
zero_contraction_output
import cuTENSOR: cutensorContractionPlan_t, cutensorAlgo_t

using ITensors.NDTensors: setdata, setstorage, cpu, IsWrappedArray, parenttype

import Base.*, Base.permutedims!
#const ContractionPlans = Dict{String, Tuple{cutensorAlgo_t, cutensorContractionPlan_t}}()
const ContractionPlans = Dict{String,cutensorAlgo_t}()

include("cuarray/set_types.jl")
include("traits.jl")
Expand All @@ -105,10 +94,21 @@ include("tensor/cudiag.jl")
include("cuitensor.jl")
include("mps/cumps.jl")

#const ContractionPlans = Dict{String, Tuple{cutensorAlgo_t, cutensorContractionPlan_t}}()
const ContractionPlans = Dict{String,cutensorAlgo_t}()

export cu,
cpu, cuITensor, randomCuITensor, cuMPS, randomCuMPS, productCuMPS, randomCuMPO, cuMPO

## TODO: Is this needed?
## const devs = Ref{Vector{CUDAdrv.CuDevice}}()
## const dev_rows = Ref{Int}(0)
## const dev_cols = Ref{Int}(0)
## function __init__()
## voltas = filter(dev->occursin("V100", CUDAdrv.name(dev)), collect(CUDAdrv.devices()))
## pascals = filter(dev->occursin("P100", CUDAdrv.name(dev)), collect(CUDAdrv.devices()))
## devs[] = voltas[1:1]
## #devs[] = pascals[1:2]
## CUBLASMG.cublasMgDeviceSelect(CUBLASMG.mg_handle(), length(devs[]), devs[])
## dev_rows[] = 1
## dev_cols[] = 1
## end

end #module
54 changes: 27 additions & 27 deletions ITensorGPU/src/tensor/cudense.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -261,7 +261,7 @@ function _contract!(
for (ii, ic) in enumerate(Cinds)
ctcinds[ii] = findfirst(x -> x == ic, ind_dict)
end
id_op = CUDA.CUTENSOR.CUTENSOR.CUTENSOR_OP_IDENTITY
id_op = cuTENSOR.CUTENSOR_OP_IDENTITY
dict_key = ""
for cc in zip(ctcinds, Cdims)
dict_key *= string(cc[1]) * "," * string(cc[2]) * ","
Expand All @@ -277,7 +277,7 @@ function _contract!(
dict_val = ContractionPlans[dict_key]
algo = dict_val
#plan = dict_val[2]
Cdata = CUDA.CUTENSOR.contraction!(
Cdata = cuTENSOR.contraction!(
α,
Adata,
Vector{Char}(ctainds),
Expand All @@ -300,13 +300,13 @@ function _contract!(
best_plan = nothing
best_algo = nothing
max_algos = Ref{Int32}(C_NULL)
CUDA.CUTENSOR.cutensorContractionMaxAlgos(max_algos)
cuTENSOR.cutensorContractionMaxAlgos(max_algos)
# fix once the other options are documented
#algos = collect(Cint(CUDA.CUTENSOR.CUTENSOR_ALGO_GETT):Cint(max_algos[] - 1))
algos = collect(Cint(CUDA.CUTENSOR.CUTENSOR_ALGO_GETT):Cint(-1))
#algos = collect(Cint(cuTENSOR.CUTENSOR_ALGO_GETT):Cint(max_algos[] - 1))
algos = collect(Cint(cuTENSOR.CUTENSOR_ALGO_GETT):Cint(-1))
for algo in reverse(algos)
try
Cdata, this_time, bytes, gctime, memallocs = @timed CUDA.CUTENSOR.contraction!(
Cdata, this_time, bytes, gctime, memallocs = @timed cuTENSOR.contraction!(
α,
Adata,
Vector{Char}(ctainds),
Expand All @@ -319,12 +319,12 @@ function _contract!(
Vector{Char}(ctcinds),
id_op,
id_op;
algo=CUDA.CUTENSOR.cutensorAlgo_t(algo),
algo=cuTENSOR.cutensorAlgo_t(algo),
)
if this_time < best_time
best_time = this_time
#best_plan = this_plan
best_algo = CUDA.CUTENSOR.cutensorAlgo_t(algo)
best_algo = cuTENSOR.cutensorAlgo_t(algo)
end
catch err
@warn "Algorithm $algo not supported"
Expand All @@ -333,7 +333,7 @@ function _contract!(
ContractionPlans[dict_key] = best_algo
end
else
Cdata = CUDA.CUTENSOR.contraction!(
Cdata = cuTENSOR.contraction!(
α,
Adata,
Vector{Char}(ctainds),
Expand All @@ -352,9 +352,9 @@ function _contract!(
end

function Base.:+(B::CuDenseTensor, A::CuDenseTensor)
opC = CUTENSOR.CUTENSOR_OP_IDENTITY
opA = CUTENSOR.CUTENSOR_OP_IDENTITY
opAC = CUTENSOR.CUTENSOR_OP_ADD
opC = cuTENSOR.CUTENSOR_OP_IDENTITY
opA = cuTENSOR.CUTENSOR_OP_IDENTITY
opAC = cuTENSOR.CUTENSOR_OP_ADD
Ais = inds(A)
Bis = inds(B)
ind_dict = Vector{Index}()
Expand All @@ -375,7 +375,7 @@ function Base.:+(B::CuDenseTensor, A::CuDenseTensor)
end
ctcinds = copy(ctbinds)
C = CUDA.zeros(eltype(Bdata), dims(Bis)...)
CUTENSOR.elementwiseBinary!(
cuTENSOR.elementwiseBinary!(
one(eltype(Adata)),
reshapeAdata,
ctainds,
Expand All @@ -393,9 +393,9 @@ function Base.:+(B::CuDenseTensor, A::CuDenseTensor)
end

function Base.:+(B::CuDense, Bis::IndexSet, A::CuDense, Ais::IndexSet)
opA = CUTENSOR.CUTENSOR_OP_IDENTITY
opC = CUTENSOR.CUTENSOR_OP_IDENTITY
opAC = CUTENSOR.CUTENSOR_OP_ADD
opA = cuTENSOR.CUTENSOR_OP_IDENTITY
opC = cuTENSOR.CUTENSOR_OP_IDENTITY
opAC = cuTENSOR.CUTENSOR_OP_ADD
ind_dict = Vector{Index}()
for (idx, i) in enumerate(Ais)
push!(ind_dict, i)
Expand All @@ -415,17 +415,17 @@ function Base.:+(B::CuDense, Bis::IndexSet, A::CuDense, Ais::IndexSet)
ctcinds = copy(ctbinds)
C = CUDA.zeros(eltype(Bdata), dims(Bis)...)
Cis = Bis
C = CUTENSOR.elementwiseBinary!(
C = cuTENSOR.elementwiseBinary!(
1, reshapeAdata, ctainds, opA, 1, reshapeBdata, ctbinds, opC, C, ctcinds, opAC
)
copyto!(data(B), vec(C))
return C
end

function Base.:-(B::CuDenseTensor, A::CuDenseTensor)
opC = CUTENSOR.CUTENSOR_OP_IDENTITY
opA = CUTENSOR.CUTENSOR_OP_IDENTITY
opAC = CUTENSOR.CUTENSOR_OP_ADD
opC = cuTENSOR.CUTENSOR_OP_IDENTITY
opA = cuTENSOR.CUTENSOR_OP_IDENTITY
opAC = cuTENSOR.CUTENSOR_OP_ADD
Ais = inds(A)
Bis = inds(B)
ind_dict = Vector{Index}()
Expand All @@ -446,7 +446,7 @@ function Base.:-(B::CuDenseTensor, A::CuDenseTensor)
end
ctcinds = copy(ctbinds)
C = CUDA.zeros(eltype(Bdata), dims(Bis))
CUTENSOR.elementwiseBinary!(
cuTENSOR.elementwiseBinary!(
-one(eltype(Adata)),
reshapeAdata,
ctainds,
Expand All @@ -464,9 +464,9 @@ function Base.:-(B::CuDenseTensor, A::CuDenseTensor)
end

function Base.:-(A::CuDense, Ais::IndexSet, B::CuDense, Bis::IndexSet)
opA = CUTENSOR.CUTENSOR_OP_IDENTITY
opC = CUTENSOR.CUTENSOR_OP_IDENTITY
opAC = CUTENSOR.CUTENSOR_OP_ADD
opA = cuTENSOR.CUTENSOR_OP_IDENTITY
opC = cuTENSOR.CUTENSOR_OP_IDENTITY
opAC = cuTENSOR.CUTENSOR_OP_ADD
ind_dict = Vector{Index}()
for (idx, i) in enumerate(Ais)
push!(ind_dict, i)
Expand All @@ -486,7 +486,7 @@ function Base.:-(A::CuDense, Ais::IndexSet, B::CuDense, Bis::IndexSet)
ctcinds = copy(ctbinds)
C = CUDA.zeros(eltype(Bdata), dims(Bis)...)
Cis = Bis
C = CUTENSOR.elementwiseBinary!(
C = cuTENSOR.elementwiseBinary!(
one(eltype(Adata)),
reshapeAdata,
ctainds,
Expand Down Expand Up @@ -523,7 +523,7 @@ function Base.permute!(B::CuDenseTensor, A::CuDenseTensor)
for (ii, ib) in enumerate(Bis)
ctbinds[ii] = findfirst(x -> x == ib, ind_dict)
end
CUDA.CUTENSOR.permutation!(
cuTENSOR.permutation!(
one(eltype(Adata)),
reshapeAdata,
Vector{Char}(ctainds),
Expand Down Expand Up @@ -560,7 +560,7 @@ function Base.permute!(B::CuDense, Bis::IndexSet, A::CuDense, Ais::IndexSet)
ctbinds[ii] = findfirst(x -> x == ib, ind_dict)
end

CUDA.CUTENSOR.permutation!(
cuTENSOR.permutation!(
one(eltype(Adata)),
reshapeAdata,
Vector{Char}(ctainds),
Expand Down
18 changes: 9 additions & 9 deletions ITensorGPU/src/tensor/dense.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -110,7 +110,7 @@ function _big_contract!(
for (ii, ic) in enumerate(Cinds)
ctcinds[ii] = findfirst(x -> x == ic, ind_dict)
end
id_op = CUDA.CUTENSOR.CUTENSOR_OP_IDENTITY
id_op = cuTENSOR.CUTENSOR_OP_IDENTITY
dict_key = ""
for cc in zip(ctcinds, Cdims)
dict_key *= string(cc[1]) * "," * string(cc[2]) * ","
Expand All @@ -126,7 +126,7 @@ function _big_contract!(
if haskey(ContractionPlans, dict_key)
dict_val = ContractionPlans[dict_key]
algo = dict_val
Cdata = CUDA.CUTENSOR.contraction!(α, Adata, Vector{Char}(ctainds), id_op, Bdata, Vector{Char}(ctbinds), id_op, β, Cdata, Vector{Char}(ctcinds), id_op, id_op; algo=algo)
Cdata = cuTENSOR.contraction!(α, Adata, Vector{Char}(ctainds), id_op, Bdata, Vector{Char}(ctbinds), id_op, β, Cdata, Vector{Char}(ctcinds), id_op, id_op; algo=algo)
synchronize()
else
# loop through all algos
Expand All @@ -136,17 +136,17 @@ function _big_contract!(
best_plan = nothing
best_algo = nothing
max_algos = Ref{Int32}(C_NULL)
CUDA.CUTENSOR.cutensorContractionMaxAlgos(max_algos)
cuTENSOR.cutensorContractionMaxAlgos(max_algos)
# fix once the other options are documented
#algos = collect(Cint(CUDA.CUTENSOR.CUTENSOR_ALGO_GETT):Cint(max_algos[] - 1))
algos = collect(Cint(CUDA.CUTENSOR.CUTENSOR_ALGO_GETT):Cint(-1))
#algos = collect(Cint(cuTENSOR.CUTENSOR_ALGO_GETT):Cint(max_algos[] - 1))
algos = collect(Cint(cuTENSOR.CUTENSOR_ALGO_GETT):Cint(-1))
for algo in reverse(algos)
try
Cdata, this_time, bytes, gctime, memallocs = @timed CUDA.CUTENSOR.contraction!(α, Adata, Vector{Char}(ctainds), id_op, Bdata, Vector{Char}(ctbinds), id_op, β, Cdata, Vector{Char}(ctcinds), id_op, id_op; algo=CUDA.CUTENSOR.cutensorAlgo_t(algo))
Cdata, this_time, bytes, gctime, memallocs = @timed cuTENSOR.contraction!(α, Adata, Vector{Char}(ctainds), id_op, Bdata, Vector{Char}(ctbinds), id_op, β, Cdata, Vector{Char}(ctcinds), id_op, id_op; algo=cuTENSOR.cutensorAlgo_t(algo))
synchronize()
if this_time < best_time
best_time = this_time
best_algo = CUDA.CUTENSOR.cutensorAlgo_t(algo)
best_algo = cuTENSOR.cutensorAlgo_t(algo)
end
catch err
@warn "Algorithm $algo not supported"
Expand Down Expand Up @@ -180,12 +180,12 @@ function _big_contract!(
#@assert !any(isnan.(AC))
#@assert !any(isnan.(BC))
#@assert !any(isnan.(CC))
#CC = CUDA.CUTENSOR.contraction!(α, AC, ctainds, id_op, BC, ctbinds, id_op, β, CC, ctcinds, id_op, id_op)
#CC = cuTENSOR.contraction!(α, AC, ctainds, id_op, BC, ctbinds, id_op, β, CC, ctcinds, id_op, id_op)
#synchronize()
#@assert !any(isnan.(AC))
#@assert !any(isnan.(BC))
#@assert !any(isnan.(CC))
Cdata = CUDA.CUTENSOR.contraction!(
Cdata = cuTENSOR.contraction!(
α, Adata, ctainds, id_op, Bdata, ctbinds, id_op, β, Cdata, ctcinds, id_op, id_op
)
synchronize()
Expand Down
4 changes: 2 additions & 2 deletions NDTensors/Project.toml
View file
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
name = "NDTensors"
uuid = "23ae76d9-e61a-49c4-8f12-3f1a16adf9cf"
authors = ["Matthew Fishman <mfishman@flatironinstitute.org>"]
version = "0.1.45"
version = "0.1.46"

[deps]
Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
Expand All @@ -19,7 +19,7 @@ TimerOutputs = "a759f4b9-e2f1-59dc-863e-4aeb61b1ea8f"
TupleTools = "9d95972d-f1c8-5527-a6e0-b4b365fa01f6"

[compat]
Adapt = "3.3, 3.4"
Adapt = "3.5"
Compat = "2.1, 3, 4"
Dictionaries = "0.3.5"
Functors = "0.2, 0.3, 0.4"
Expand Down
Loading

4 comments on commit e2a5726

@mtfishman
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@JuliaRegistrator register subdir=NDTensors

@JuliaRegistrator
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Registration pull request created: JuliaRegistries/General/77235

After the above pull request is merged, it is recommended that a tag is created on this repository for the registered package version.

This will be done automatically if the Julia TagBot GitHub Action is installed, or can be done manually through the github interface, or via:

git tag -a NDTensors-v0.1.46 -m "<description of version>" e2a5726d4f37eba146af246537109a0d2971f34b
git push origin NDTensors-v0.1.46

@mtfishman
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@JuliaRegistrator register

@JuliaRegistrator
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Registration pull request created: JuliaRegistries/General/77236

After the above pull request is merged, it is recommended that a tag is created on this repository for the registered package version.

This will be done automatically if the Julia TagBot GitHub Action is installed, or can be done manually through the github interface, or via:

git tag -a v0.3.26 -m "<description of version>" e2a5726d4f37eba146af246537109a0d2971f34b
git push origin v0.3.26

Also, note the warning: Version 0.3.26 skips over 0.3.25
This can be safely ignored. However, if you want to fix this you can do so. Call register() again after making the fix. This will update the Pull request.

Please sign in to comment.