The CUDA platform is built around a large-scale
similarity, where latency in memory access can be hidden waiting for other
computations due to this parallelism, it may be that many attempts to thread write
on the same storage space, it should be managed using nuclear memory functions,
which guarantees that the following steps include updating the memory location.
Performing without any interruptions: 1) Getting value; 2) Modified; And 3) writing it back are trying two threads both to lead
similar memory spaces for race situations and undefined results.
CUDA is a programming model created by
NVIDIA gives the developer access to GPU computing resources following
through an Application Programming Interface (API) the standard CUDA
terminology. We will see GPU as the device and CPU as the host programming
language extends to C / C ++ Language.
GPU programming is different from model normal
CPU models, and data must be clearly moved flat model between host and
device there is a multiple grid available for programmers’ thread blocks
are threads on the current structure classes of 32 Threads multiplied by
the name of Vars.
The longest the black arrow in
Fig 1 represents the whole beam, while it the small normal vector
background is the direction vector. The upcoming, scatterers are projecting
on the image line, and side and high distances are mapped as a weight
component using an analytical expression or a lookup table. From
radial distance, the closest index in the output vector determined and the
weighted dimension is added to it index the speed of sound is believed to be stable, and the
following formula is used to convert radial distance (R) from Time (T): The
factor of two accounts for T: (R) = 2R/C0 After the two-way journey
has been estimated all scatterers. The output vector is understood with pulse
wave last RF Line yield.
The COLE algorithm 6 is graphically
portrayed in Figure 1. aims to simulate a backstroked RF signal from the
target collection of 3-D point scatterers. Each scattered is one
position and one dimension that controls the power Backscattered signal an
RF Line is defined by its unit vectors in radial, lateral, and advanced
Figure 1 Illustration of the main parts of the COLE
algorithm. Every point the scatter is presented on an imaging line with an
approximate dimension. Depends on lateral and high distances. After launch, RF
signal the scatterers are here by convolving with a pulse wave drew with
proportional area for their full scattering dimension. The projected dimension
is reflected in the strength of the circle arc.
algorithm is very fast, but not efficient come free; There are many famous
limitations in the algorithm. Simulation accuracy for scatterers located nearby
area is close to the probe surface, there is no model of shore waves and
point-split function is considered to split. In addition, the use of 1-D
convection is implied the pulse wave is stable during the spread of the pulse.
In practice, this is not true for the promotion of a Broadband Sense depending
on frequency dependent attenuation
The accuracy of this convolutional model
has been examined 7 by creating a physical point-scattered phantom and for
the same simulation compared to the real scan compared to cola algorithms was
done in field II 3.
Real ultrasound recording has recently
been used as part of the simulation pipeline to get more realistic synthetic
Currently, the two most popular GPU
computing platforms source OpenCL (Khronos Group) and proprietary calculate
Unified Device Architecture (CUDA) (NVIDIA Corp, Santa Clara, CA, United
States). We chose as the CUDA platform due to good selection of GPU
implementation development tools and library with useful functionality, such as
GPU-Accelerated Quick Fourier Conversion (FFTs).
It will be difficult to cover all the
technical details involved CPU and GPU implementation in one adequate paper, so
we decided to follow the recommendation in 2 and complete source code is
hosted simulator project on Github, under the name “OpenBCSim” as a
repository 4 an approved open source license. In addition to the simulator
himself, this repository includes scripts to create points the gates are
referred to in this letter, as well scripts used for performance benchmarking.