When code performance is an issue, it is sometimes useful to get absolute
performance measurements in order to objectivise what is "slow" or
"fast". GFlops.jl leverages the power of Cassette.jl to automatically count
the number of floating-point operations in a piece of code. When combined with
the accuracy of BenchmarkTools, this allows for easy and absolute performance
measurements.
This package is registered and can therefore be simply be installed with
pkg> add GFlopsThis simple example shows how to track the number of operations in a vector summation:
julia> using GFlops
julia> x = rand(1000);
julia> @count_ops sum($x)
Flop Counter: 999 flop
┌─────┬─────────┐
│ │ Float64 │
├─────┼─────────┤
│ add │ 999 │
└─────┴─────────┘
julia> @gflops sum($x);
8.86 GFlops, 12.76% peak (9.99e+02 flop, 1.13e-07 s, 0 alloc: 0 bytes)GFlops.jl internally tracks several types of Floating-Point operations, for
both 32-bit and 64-bit operands. Pretty-printing a Flop Counter only
shows non-zero entries, but any individual counter can be accessed:
julia> function mixed_dot(x, y)
acc = 0.0
@inbounds @simd for i in eachindex(x, y)
acc += x[i] * y[i]
end
acc
end
mixed_dot (generic function with 1 method)
julia> x = rand(Float32, 1000); y = rand(Float32, 1000);
julia> cnt = @count_ops mixed_dot($x, $y)
Flop Counter: 1000 flop
┌─────┬─────────┬─────────┐
│ │ Float32 │ Float64 │
├─────┼─────────┼─────────┤
│ add │ 0 │ 1000 │
│ mul │ 1000 │ 0 │
└─────┴─────────┴─────────┘
julia> fieldnames(GFlops.Counter)
(:fma32, :fma64, :muladd32, :muladd64, :add32, :add64, :sub32, ...)
julia> cnt.add64
1000
julia> @gflops mixed_dot($x, $y);
9.91 GFlops, 13.36% peak (2.00e+03 flop, 2.02e-07 s, 0 alloc: 0 bytes)On systems which support them, FMAs and MulAdds compute two operations (an
addition and a multiplication) in one instruction. @count_ops counts each
individual FMA/MulAdd as one operation, which makes it easier to interpret
counters. However, @gflops will count two floating-point operations for each
FMA, in accordance to the way high-performance benchmarks usually behave:
julia> x = 0.5; coeffs = rand(10);
# 9 MulAdds but 18 flop
julia> cnt = @count_ops evalpoly($x, $coeffs)
Flop Counter: 18 flop
┌────────┬─────────┐
│ │ Float64 │
├────────┼─────────┤
│ muladd │ 9 │
└────────┴─────────┘
julia> @gflops evalpoly($x, $coeffs);
0.87 GFlops, 1.63% peak (1.80e+01 flop, 2.06e-08 s, 0 alloc: 0 bytes)GFlops.jl does not see what happens outside the realm of Julia code. It
especially does not see operations performed in external libraries such as BLAS
calls:
julia> using LinearAlgebra
julia> @count_ops dot($x, $y)
Flop Counter: 0 flopThis is a known issue; we'll try and find a way to circumvent the problem.