Enhancement #706: Improve Memory Layout and Allocation of MathVector - NemoMath - Open Source Collaboration Platform

Enhancement #706

Improve Memory Layout and Allocation of MathVector

Added by M. Rolf over 12 years ago. Updated over 11 years ago.

Status:

Resolved

Start date:

11/09/2011

Priority:

Normal

Due date:

Assignee:

M. Rolf

% Done:

100%

Category:

Target version:

NemoMath 0.4

Description

Explore memory layout without several layers of pointers, but store
everything (refCount,size,data) in one memory block. Compare std::string implementation.

Possible benefits (?):

Faster allocation/deallocation
Faster dereferencing
Smaller memory footprint (in particular for small vectors)
Maybe better caching behavior (memory that's used together belongs together!)

Drawbacks:
very ugly!

History

#1 Updated by M. Rolf almost 12 years ago

Status changed from New to In Progress
% Done changed from 0 to 40

#2 Updated by M. Rolf almost 12 years ago

Benchmark results for first implementation:

Benchsuite [MathVectorCreate]
    Performing each case 10000000 times with operation-size 10

    Benchcase [MathVectorCreate:Create]
            Operations per second:    9.71064e+06

    Benchcase [MathVectorCreate:CreateCopyModify]
            Operations per second:    4.81884e+06

Compared to previous implementation:

Benchsuite [MathVectorCreate]
    Performing each case 10000000 times with operation-size 10

    Benchcase [MathVectorCreate:Create]
            Operations per second:    5.36829e+06

    Benchcase [MathVectorCreate:CreateCopyModify]
            Operations per second:    2.64776e+06

Compared to Eigen2:

Benchsuite [MathVectorCreate]
    Performing each case 10000000 times with operation-size 10

    Benchcase [MathVectorCreate:Create:Eigen]
            Operations per second:    9.52886e+06

    Benchcase [MathVectorCreate:CreateCopyModify:Eigen]
            Operations per second:    3.77345e+06