Skip to content

More stuff #6

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.

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

Merged
merged 3 commits into from
Jan 20, 2021
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
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
5 changes: 4 additions & 1 deletion .github/workflows/ci.yml
Original file line number Diff line number Diff line change
Expand Up @@ -70,7 +70,10 @@ jobs:
using Documenter: doctest
using BLASBenchmarks
doctest(BLASBenchmarks)'
env:
JULIA_NUM_THREADS: 2
- run: julia --project=docs docs/make.jl
env:
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
DOCUMENTER_KEY: ${{ secrets.DOCUMENTER_KEY }}
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
JULIA_NUM_THREADS: 2
1 change: 1 addition & 0 deletions docs/make.jl
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,7 @@ makedocs(;
pages=[
"Home" => "index.md",
"Usage" => "usage.md",
"Memory Required for Large Matrices" => "memory-required.md",
"Public API" => "public-api.md",
"Internals (Private)" => "internals.md",
],
Expand Down

This file was deleted.

This file was deleted.

8 changes: 0 additions & 8 deletions docs/src/index.md
Original file line number Diff line number Diff line change
Expand Up @@ -8,11 +8,3 @@ BLASBenchmarks is a Julia package for benchmarking BLAS libraries.

The source code for this package is available in the
[GitHub repository](https://github.com/chriselrod/BLASBenchmarks.jl).

## Float64

![Float64](../assets/gemm_Float64_10_30_haswell_AVX2__multithreaded_logscale.svg)

## Float32

![Float32](../assets/gemm_Float32_10_30_haswell_AVX2__multithreaded_logscale.svg)
161 changes: 161 additions & 0 deletions docs/src/memory-required.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,161 @@
```@meta
CurrentModule = BLASBenchmarks
```

# Memory Required for Large Matrices

These tables show how much memory is required for four matrices of the given size and type. (We need four matrices: `A`, `B`, `C1`, and `C2`.)

## Float64

| Matrix Size | Memory |
| ----------- | ------ |
| 1k by 1k | 0.03 GiB |
| 2k by 2k | 0.12 GiB |
| 3k by 3k | 0.27 GiB |
| 4k by 4k | 0.48 GiB |
| 5k by 5k | 0.75 GiB |
| 6k by 6k | 1.07 GiB |
| 7k by 7k | 1.46 GiB |
| 8k by 8k | 1.91 GiB |
| 9k by 9k | 2.41 GiB |
| 10k by 10k | 2.98 GiB |
| 11k by 11k | 3.61 GiB |
| 12k by 12k | 4.29 GiB |
| 13k by 13k | 5.04 GiB |
| 14k by 14k | 5.84 GiB |
| 15k by 15k | 6.71 GiB |
| 16k by 16k | 7.63 GiB |
| 17k by 17k | 8.61 GiB |
| 18k by 18k | 9.66 GiB |
| 19k by 19k | 10.76 GiB |
| 20k by 20k | 11.92 GiB |
| 30k by 30k | 26.82 GiB |
| 40k by 40k | 47.68 GiB |
| 50k by 50k | 74.51 GiB |
| 60k by 60k | 107.29 GiB |
| 70k by 70k | 146.03 GiB |
| 80k by 80k | 190.73 GiB |
| 90k by 90k | 241.4 GiB |
| 100k by 100k | 298.02 GiB |

## Float32

| Matrix Size | Memory |
| ----------- | ------ |
| 1k by 1k | 0.01 GiB |
| 2k by 2k | 0.06 GiB |
| 3k by 3k | 0.13 GiB |
| 4k by 4k | 0.24 GiB |
| 5k by 5k | 0.37 GiB |
| 6k by 6k | 0.54 GiB |
| 7k by 7k | 0.73 GiB |
| 8k by 8k | 0.95 GiB |
| 9k by 9k | 1.21 GiB |
| 10k by 10k | 1.49 GiB |
| 11k by 11k | 1.8 GiB |
| 12k by 12k | 2.15 GiB |
| 13k by 13k | 2.52 GiB |
| 14k by 14k | 2.92 GiB |
| 15k by 15k | 3.35 GiB |
| 16k by 16k | 3.81 GiB |
| 17k by 17k | 4.31 GiB |
| 18k by 18k | 4.83 GiB |
| 19k by 19k | 5.38 GiB |
| 20k by 20k | 5.96 GiB |
| 30k by 30k | 13.41 GiB |
| 40k by 40k | 23.84 GiB |
| 50k by 50k | 37.25 GiB |
| 60k by 60k | 53.64 GiB |
| 70k by 70k | 73.02 GiB |
| 80k by 80k | 95.37 GiB |
| 90k by 90k | 120.7 GiB |
| 100k by 100k | 149.01 GiB |

## Int64

| Matrix Size | Memory |
| ----------- | ------ |
| 1k by 1k | 0.03 GiB |
| 2k by 2k | 0.12 GiB |
| 3k by 3k | 0.27 GiB |
| 4k by 4k | 0.48 GiB |
| 5k by 5k | 0.75 GiB |
| 6k by 6k | 1.07 GiB |
| 7k by 7k | 1.46 GiB |
| 8k by 8k | 1.91 GiB |
| 9k by 9k | 2.41 GiB |
| 10k by 10k | 2.98 GiB |
| 11k by 11k | 3.61 GiB |
| 12k by 12k | 4.29 GiB |
| 13k by 13k | 5.04 GiB |
| 14k by 14k | 5.84 GiB |
| 15k by 15k | 6.71 GiB |
| 16k by 16k | 7.63 GiB |
| 17k by 17k | 8.61 GiB |
| 18k by 18k | 9.66 GiB |
| 19k by 19k | 10.76 GiB |
| 20k by 20k | 11.92 GiB |
| 30k by 30k | 26.82 GiB |
| 40k by 40k | 47.68 GiB |
| 50k by 50k | 74.51 GiB |
| 60k by 60k | 107.29 GiB |
| 70k by 70k | 146.03 GiB |
| 80k by 80k | 190.73 GiB |
| 90k by 90k | 241.4 GiB |
| 100k by 100k | 298.02 GiB |

## Int32

| Matrix Size | Memory |
| ----------- | ------ |
| 1k by 1k | 0.01 GiB |
| 2k by 2k | 0.06 GiB |
| 3k by 3k | 0.13 GiB |
| 4k by 4k | 0.24 GiB |
| 5k by 5k | 0.37 GiB |
| 6k by 6k | 0.54 GiB |
| 7k by 7k | 0.73 GiB |
| 8k by 8k | 0.95 GiB |
| 9k by 9k | 1.21 GiB |
| 10k by 10k | 1.49 GiB |
| 11k by 11k | 1.8 GiB |
| 12k by 12k | 2.15 GiB |
| 13k by 13k | 2.52 GiB |
| 14k by 14k | 2.92 GiB |
| 15k by 15k | 3.35 GiB |
| 16k by 16k | 3.81 GiB |
| 17k by 17k | 4.31 GiB |
| 18k by 18k | 4.83 GiB |
| 19k by 19k | 5.38 GiB |
| 20k by 20k | 5.96 GiB |
| 30k by 30k | 13.41 GiB |
| 40k by 40k | 23.84 GiB |
| 50k by 50k | 37.25 GiB |
| 60k by 60k | 53.64 GiB |
| 70k by 70k | 73.02 GiB |
| 80k by 80k | 95.37 GiB |
| 90k by 90k | 120.7 GiB |
| 100k by 100k | 149.01 GiB |

## Generating These Tables

```julia
mem_req(s, T) = 4s^2*sizeof(T) / (1 << 30)

function print_table(::Type{T}, Ns = nothing) where {T}
println("| Matrix Size | Memory |")
println("| ----------- | ------ |")
if Ns isa Nothing
_Ns = sort(unique(vcat(collect(1:1:20), collect(20:10:100))))
else
_Ns = Ns
end
for N in _Ns
mem = mem_req(N * 1_000, T)
m = round(mem; digits = 2)
println("| $(N)k by $(N)k | $(m) GiB |")
end
return nothing
end
```
42 changes: 30 additions & 12 deletions src/plotting.jl
Original file line number Diff line number Diff line change
Expand Up @@ -26,13 +26,33 @@ function default_plot_directory()
return joinpath(_pkgdir(), "docs", "src", "assets")
end

"""
default_plot_filename(br::BenchmarkResult;
desc,
logscale)
"""
function default_plot_filename(br::BenchmarkResult{T};
desc::AbstractString,
logscale::Bool) where {T}
df = br.df
l, u = extrema(df.Size)
if logscale
desc *= "_logscale"
end
desc = (br.threaded ? "_multithreaded" : "_singlethreaded") * desc
suffix = pick_suffix(desc)
return "gemm_$(string(T))_$(l)_$(u)_$(suffix)"
end

"""
plot(br::BenchmarkResult;
desc = "",
logscale = true,
width = 1200,
height = 600,
plot_directory = default_plot_directory())
plot_directory = default_plot_directory(),
plot_filename = default_plot_filename(br; desc = desc, logscale = logscale),
file_extensions = ["svg", "png"])
"""
function plot(br::BenchmarkResult{T}; kwargs...) where {T}
_plot(br; kwargs...)
Expand All @@ -47,6 +67,8 @@ function _plot(
width::Real = 1200,
height::Real = 600,
plot_directory::AbstractString = default_plot_directory(),
plot_filename::AbstractString = default_plot_filename(br; desc = desc, logscale = logscale),
file_extensions = ["svg", "png"],
) where {T}
df = br.df
plt = if logscale
Expand All @@ -62,16 +84,12 @@ function _plot(
width = width, height = height
)
end
l, u = extrema(df.Size)
if logscale
desc *= "_logscale"
end
desc = (br.threaded ? "_multithreaded" : "_singlethreaded") * desc
suffix = pick_suffix(desc)
mkpath(plot_directory)
foo1 =
foo2 =
save(joinpath(plot_directory, "gemm_$(string(T))_$(l)_$(u)_$(suffix).svg"), plt)
save(joinpath(plot_directory, "gemm_$(string(T))_$(l)_$(u)_$(suffix).png"), plt)
return foo1, foo2
_filenames = String[]
for ext in file_extensions
_filename = joinpath(plot_directory, "$(plot_filename).$(ext)")
save(_filename, plt)
push!(_filenames, _filename)
end
return _filenames
end
40 changes: 39 additions & 1 deletion src/runbenchmark.jl
Original file line number Diff line number Diff line change
Expand Up @@ -102,10 +102,48 @@ Base.axes(ls::LogSpace) = axes(ls.r)
Base.eltype(::LogSpace) = Int

"""
all_libs()
"""
function all_libs()
libs = Symbol[
:BLIS,
:Gaius,
:MKL,
:Octavian,
:OpenBLAS,
:Tullio,
]
return libs
end

function _integer_libs()
libs_to_exclude = Symbol[:BLIS, :MKL, :OpenBLAS]
return sort(unique(setdiff(all_libs(), libs_to_exclude)))
end

"""
default_libs(T)
"""
function default_libs(::Type{T}) where {T}
if T <: Integer
return _integer_libs()
else
return all_libs()
end
end

"""
runbench(T = Float64;
libs = default_libs(T),
sizes = logspace(2, 4000, 200),
threaded::Bool = Threads.nthreads() > 1,
A_transform = identity,
B_transform = identity,
sleep_time = 0.0)
"""
function runbench(
::Type{T} = Float64;
libs = [:BLIS, :Gaius, :MKL, :Octavian, :OpenBLAS, :Tullio],
libs = default_libs(T),
sizes = logspace(2, 4000, 200),
threaded::Bool = Threads.nthreads() > 1,
A_transform = identity,
Expand Down