GiggleLiu

name	sci-brainstorm
description	Use when exploring a research idea, starting from a paper, concept, or question — conducts deep literature survey via arxiv and web search, discovers cross-field connections, then challenges the idea through Socratic and adversarial discussion

Scientific Research Brainstorming

Research-first brainstorming. Every phase begins with autonomous search — never ask the human what the literature says; go find out. Staged posture: Socratic while the idea is forming, adversarial once it has shape. Produces a research direction document.

Entry

GiggleLiu

using JuMP, HiGHS, Graphs, LinearAlgebra, Test

# Find one valid assignment with integer programming
function lattice_sites(num_sites::Vector{Int}, multiplicity::Vector{Int}, num_atoms::Vector{Int})
    model = Model(HiGHS.Optimizer)
    JuMP.set_silent(model)
    @variable(model, x[1:length(num_sites), 1:length(num_atoms)] >= 0, Int)
    for (ia, na) in enumerate(num_atoms)
        @constraint(model, sum(x[i, ia] * num_sites[i] for i in 1:length(num_sites)) == na)
    end

GiggleLiu

using GenericTensorNetworks

# create a random King's subgraph
graph = random_diagonal_coupled_graph(7, 7, 0.8)
# create an independent set problem
problem = IndependentSet(graph)
# convert to tensor network
pb = GenericTensorNetwork(problem)

# solve the MIS size

GiggleLiu

### A Pluto.jl notebook ###
# v0.19.9

using Markdown
using InteractiveUtils

# ╔═╡ 195eb152-17e7-11ed-32ab-ab200fb170af
using Yao, YaoPlots, Polynomials

# ╔═╡ f6f10336-205c-4581-9418-de6fdd9b6418

GiggleLiu

# To run this script, you need to install required packages in a Julia REPL:
# ```julia
# using Pkg; Pkg.add(["GenericTensorNetworks", "CUDA", "Comonicon", "DelimitedFiles"])
# ```
#
# Then open a terminal, go to the file folder and type
# ```bash
# $ julia fib2d.jl 20
# ```
#

GiggleLiu

using LinearAlgebra: mul!

"""
    strassen!(C, A, B, s0=8)

Compute matrix multiplication `C = A * B` using the strassen algorithm.
`s0` is the critical size to stop Strassen recursion, it can be slow if it recurse too deep,
then one needs to set this value larger.

!!!note

GiggleLiu

# requires OMEinsum version >= 0.7
using Distributed
using OMEinsum, CUDA

println("find $(length(devices())) GPU devices")
const procs = addprocs(length(devices())-nprocs()+1)
const gpus = collect(devices())
const process_device_map = Dict(zip(procs, gpus))
@info process_device_map

GiggleLiu

using TropicalNumbers, VectorizationBase
using Test
using LoopVectorization, Octavian
using VectorizationBase: OffsetPrecalc, StaticBool, Bit, static, NativeTypes, Index, gep_quote, VectorIndex

function distance(a::AbstractArray{<:Tropical}, b::AbstractArray{<:Tropical})
    sum(abs.(content.(a) .- content.(b)))
end

LoopVectorization.check_args(::Type{T}, ::Type{T}) where T<:Tropical = true

GiggleLiu

# v0.12.17

using Markdown
using InteractiveUtils

# ╔═╡ f178d9fe-49f4-11eb-0863-21e2e3934592
using Yao, YaoPlots

# ╔═╡ d59b39ec-49f5-11eb-3d4a-fb69f2505d3c
using SymEngine

GiggleLiu

using CUDA
using CUDA.WMMA
using CUDA.WMMA: ColMajor, load_a, load_b, load_c, mma, store_d
using StaticArrays
a = rand(Float16, 16, 16)
b = rand(Float16, 16, 16)
c = zeros(Float16, 16, 16)
d = zeros(Float16, 16, 16)

a_dev = CuArray(a)