-\chapterauthor{Raphaël Couturier}{Femto-ST Institute, University of Franche-Comte}
+\chapterauthor{Raphaël Couturier}{Femto-ST Institute, University of Franche-Comte, France}
\chapter{Presentation of the GPU architecture and of the CUDA environment}
\label{chapter1}
\section{Introduction}\label{ch1:intro}
-
This chapter introduces the Graphics Processing Unit (GPU) architecture and all
the concepts needed to understand how GPUs work and can be used to speed up the
execution of some algorithms. First of all this chapter gives a brief history of
-the development of Graphics card until they have been used in order to make
+the development of the graphics cards up to the point when they started being used in order to make
general purpose computation. Then the architecture of a GPU is
illustrated. There are many fundamental differences between a GPU and a
tradition processor. In order to benefit from the power of a GPU, a CUDA
-\section{Brief history of Video Card}
+\section{Brief history of the video card}
-Video cards or Graphics cards have been introduced in personal computers to
+Video cards or graphics cards have been introduced in personal computers to
produce high quality graphics faster than classical Central Processing Units
-(CPU) and to alleviate CPU from this task. In general, display tasks are very
+(CPU) and to free the CPU from this task. In general, display tasks are very
repetitive and very specific. Hence, some manufacturers have produced more and
-more sophisticated video cards, providing 2D accelerations then 3D accelerations,
+more sophisticated video cards, providing 2D accelerations, then 3D accelerations,
then some light transforms. Video cards own their own memory to perform their
computation. For at least two decades, every personal computer has had a video
card which is simple for desktop computers or which provides many accelerations
-for game and/or graphic oriented computers. In the latter case, graphic cards
+for game and/or graphic-oriented computers. In the latter case, graphic cards
may be more expensive than a CPU.
Since 2000, video cards have allowed users to apply arithmetic operations
-simultaneously on a sequence of pixels, also later called stream processing. In
-this case, the information of the pixels (color, location and other information) are
+simultaneously on a sequence of pixels, later called stream processing. In
+this case, the information of the pixels (color, location and other information) is
combined in order to produce a pixel color that can be displayed on a screen.
Simultaneous computations are provided by shaders which calculate rendering
effects on graphics hardware with a high degree of flexibility. These shaders
-handles the stream data with pipelines.
+handle the stream data with pipelines.
Some researchers tried to apply those operations on other data, representing
something different from pixels, and consequently this resulted in the first
uses of video cards for performing general purpose computation. The programming
-model was not easy to use at all and was very dependent of the hardware
+model was not easy to use at all and was very dependent on the hardware
constraints. More precisely it consisted in using either DirectX of OpenGL
functions providing an interface to some classical operations for videos
-operations (memory transfers, texture manipulation, ...). Floating point
+operations (memory transfers, texture manipulation, etc.). Floating point
operations were most of the time unimaginable. Obviously when something went
-wrong, programmers had no way (and neither the tools) to detect it.
+wrong, programmers had no way (and no tools) to detect it.
\section{GPGPU}
In order to benefit from the computing power of more recent video cards, CUDA
-was first proposed in 2007 by NVidia. It unifies the programming model for some
-of their most performant video cards. Cuda~\cite{ch1:cuda} has quickly been
+was first proposed in 2007 by NVIDIA. It unifies the programming model for some
+of their most efficient video cards. CUDA~\cite{ch1:cuda} has quickly been
considered by the scientific community as a great advance for general purpose
graphics processing unit (GPGPU) computing. Of course other programming models
have been proposed. The other well-known alternative is OpenCL which aims at
-proposing an alternative to Cuda and which is multi-platform and portable. This
+proposing an alternative to CUDA and which is multiplatform and portable. This
is a great advantage since it is even possible to execute OpenCL programs on
traditional CPUs. The main drawback is that it is less tight with the hardware
-and consequently sometimes provides less efficient programs. Moreover, Cuda
+and consequently sometimes provides less efficient programs. Moreover, CUDA
benefits from more mature compilation and optimization procedures. Other less
-known environments have been proposed, but most of them have been stopped, for
-example we can cite: FireStream by ATI which is not maintained anymore and
-replaced by OpenCL, BrookGPU by Standford University~\cite{ch1:Buck:2004:BGS}.
+known environments have been proposed, but most of them have been discontinued, for
+example we can cite, FireStream by ATI which is not maintained anymore and
+has been replaced by OpenCL, BrookGPU by Standford University~\cite{ch1:Buck:2004:BGS}.
Another environment based on pragma (insertion of pragma directives inside the
-code to help the compiler to generate efficient code) is call OpenACC. For a
-comparison with OpenCL, interested readers may refer to~\cite{ch1:CMR:12}.
+code to help the compiler to generate efficient code) is called OpenACC. For a
+comparison with OpenCL, interested readers may refer to~\cite{ch1:Dongarra}.
\section{Architecture of current GPUs}
-The architecture \index{architecture of a GPU} of current GPUs is constantly
+The architecture \index{GPU!architecture of a} of current GPUs is constantly
evolving. Nevertheless some trends remain constant throughout this evolution.
-Processing units composing a GPU are far more simple than a traditional CPU but
+Processing units composing a GPU are far simpler than a traditional CPU and
it is much easier to integrate many computing units inside a GPU card than to do
-so with many cores inside a CPU. This is due to the fact that the cores of a GPU are
-simpler than the cores of a CPU. In 2012, the most powerful GPUs own more than 500
-cores and the most powerful CPUs have 8
+so with many cores inside a CPU. In 2012, the most powerful GPUs contained more than 500
+cores and the most powerful CPUs had 8
cores. Figure~\ref{ch1:fig:comparison_cpu_gpu} shows the number of cores inside
-a CPU and inside a GPU. In fact, in a current NVidia GPU, there are
-multiprocessors which have 32 cores (for example on Fermi cards). The core clock
-of CPU is generally around 3GHz and the one of GPU is about 1.5GHz. Although
-the core clock of GPU cores is slower, the amount of cores inside a GPU provides
+a CPU and inside a GPU. In fact, in a current NVIDIA GPU, there are
+multiprocessors which have 32 cores (for example, on Fermi cards). The core clock
+of a CPU is generally around 3GHz and the one of a GPU is about 1.5GHz. Although
+the core clock of GPU cores is slower, the number of cores inside a GPU provides
more computational power. This measure is commonly represented by the number of
floating point operation per seconds. Nowadays the most powerful GPUs provide more
-than 1TFlops, i.e. $10^{12}$ floating point operations per second.
-Nevertheless GPUs are very efficient to perform some operations but not all
-kinds of operations. They are very efficient to execute repetitive work in which
+than 1TFlops, i.e., $10^{12}$ floating point operations per second.
+Nevertheless GPUs are very efficient at executing repetitive work in which
only the data change. It is important to keep in mind that multiprocessors
inside a GPU have 32 cores. Later we will see that these 32 cores need to do the
same work to get maximum performance.
\end{figure}
On the most powerful GPU cards, called Fermi, multiprocessors are called streaming
-multiprocessors (SM). Each SM contains 32 cores and is able to perform 32
-floating points or integer operations on 32 bits numbers per clock or 16 floating
-points on 64 bits number per clock. SMs have their own registers, execution
-pipelines and caches. On Fermi architecture, there are 64Kb shared memory + L1
-cache and 32,536 32bits registers per SM. More precisely the programmer can
-decide what amount of shared memory and L1 cache SM can use. The constraint is
-that the sum of both amounts should be less or equal to 64Kb.
-
-Threads are used to benefit from the important number of cores of a GPU. Those
-threads are different from traditional threads for CPU. In
-chapter~\ref{chapter2}, some examples of GPU programming will explicit the
-details of the GPU threads. However, threads are gathered into blocks of 32
-threads, called ``warps''. Those warps are important when designing an algorithm
+multiprocessors (SMs). Each SM contains 32 cores and is able to perform 32
+floating points or integer operations per clock on 32 bit numbers or 16 floating
+points per clock on 64 bit numbers. SMs have their own registers, execution
+pipelines and caches. On Fermi architecture, there are 64Kb shared memory plus L1
+cache and 32,536 32 bit registers per SM. More precisely the programmer can
+decide what amounts of shared memory and L1 cache SM are to be used. The constraint is
+that the sum of both amounts should be less than or equal to 64Kb.
+
+Threads are used to benefit from the large number of cores of a GPU. These
+threads are different from traditional threads for a CPU. In
+Chapter~\ref{chapter2}, some examples of GPU programming will explain the
+details of the GPU threads. Threads are gathered into blocks of 32
+threads, called ``warps''. These warps are important when designing an algorithm
for GPU.
-Another big difference between CPU and GPU is the latency of memory. In CPU,
+Another big difference between a CPU and a GPU is the latency of memory. In a CPU,
everything is optimized to obtain a low latency architecture. This is possible
-through the use of cache memories. Moreover, nowadays CPUs perform many
+through the use of cache memories. Moreover, nowadays CPUs carry out many
performance optimizations such as speculative execution which roughly speaking
-consists in executing a small part of code in advance even if later this work
-reveals itself to be useless. On the contrary, GPUs do not have low latency
+consists of executing a small part of the code in advance even if later this work
+reveals itself to be useless. GPUs do not have low latency
memory. In comparison GPUs have small cache memories. Nevertheless the
architecture of GPUs is optimized for throughput computation and it takes into
account the memory latency.
\begin{figure}[b!]
\centerline{\includegraphics[scale=0.7]{Chapters/chapter1/figures/low_latency_vs_high_throughput.pdf}}
-\caption{Comparison of low latency of CPU and high throughput of GPU.}
+\caption{Comparison of low latency of a CPU and high throughput of a GPU.}
\label{ch1:fig:latency_throughput}
\end{figure}
memory latency between a CPU and a GPU. In a CPU, tasks ``ti'' are executed one
by one with a short memory latency to get the data to process. After some tasks,
there is a context switch that allows the CPU to run concurrent applications
-and/or multi-threaded applications. Memory latencies are longer in a GPU, the
-the principle to obtain a high throughput is to have many tasks to
-compute. Later we will see that those tasks are called threads with CUDA. With
+and/or multi-threaded applications. Memory latencies are longer in a GPU. Thhe
+ principle to obtain a high throughput is to have many tasks to
+compute. Later we will see that these tasks are called threads with CUDA. With
this principle, as soon as a task is finished the next one is ready to be
-executed while the wait for data for the previous task is overlapped by
-computation of other tasks.
+executed while the wait for data for the previous task is overlapped by the
+computation of other tasks.
\section{Kinds of parallelism}
-Many kinds of parallelism are amiable according to the type of hardware.
+Many kinds of parallelism are available according to the type of hardware.
Roughly speaking, there are three classes of parallelism: instruction-level
-parallelism, data parallelism and task parallelism.
+parallelism, data parallelism, and task parallelism.
-Instruction-level parallelism consists in re-ordering some instructions in order
+Instruction-level parallelism consists in reordering some instructions in order
to execute some of them in parallel without changing the result of the code.
-In modern CPUs, instruction pipelines allow processor to execute instructions
+In modern CPUs, instruction pipelines allow the processor to execute instructions
faster. With a pipeline a processor can execute multiple instructions
-simultaneously due to the fact that the output of a task is the input of the
+simultaneously because the output of a task is the input of the
next one.
Data parallelism consists in executing the same program with different data on
-different computing units. Of course, no dependency should exist between the
+different computing units. Of course, no dependency should exist among the
data. For example, it is easy to parallelize loops without dependency using the
data parallelism paradigm. This paradigm is linked with the Single Instructions
Multiple Data (SIMD) architecture. This is the kind of parallelism provided by
GPUs.
-Task parallelism is the common parallelism achieved out on clusters and grids and
+Task parallelism is the common parallelism achieved on clusters and grids and
high performance architectures where different tasks are executed by different
computing units.
-\section{CUDA Multithreading}
+\section{CUDA multithreading}
The data parallelism of CUDA is more precisely based on the Single Instruction
-Multiple Thread (SIMT) model. This is due to the fact that a programmer accesses
-to the cores by the intermediate of threads. In the CUDA model, all cores
+Multiple Thread (SIMT) model, because a programmer accesses
+ the cores by the intermediate of threads. In the CUDA model, all cores
execute the same set of instructions but with different data. This model has
similarities with the vector programming model proposed for vector machines through
-the 1970s into the 90s, notably the various Cray platforms. On the CUDA
+the 1970s and into the 90s, notably the various Cray platforms. On the CUDA
architecture, the performance is led by the use of a huge number of threads
-(from thousands up to to millions). The particularity of the model is that there
+(from thousands up to millions). The particularity of the model is that there
is no context switching as in CPUs and each thread has its own registers. In
-practice, threads are executed by SM and are gathered into groups of 32
-threads. Those groups are called ``warps''. Each SM alternatively executes
-``active warps'' and warps becoming temporarily inactive due to waiting of data
+practice, threads are executed by SM and gathered into groups of 32
+threads, called warps. Each SM alternatively executes
+active warps and warps becoming temporarily inactive due to waiting of data
(as shown in Figure~\ref{ch1:fig:latency_throughput}).
+\begin{figure}[b!]
+\centerline{\includegraphics[scale=0.65]{Chapters/chapter1/figures/scalability.pdf}}
+\caption{Scalability of GPU.}
+\label{ch1:fig:scalability}
+\end{figure}
+
The key to scalability in the CUDA model is the use of a huge number of threads.
-In practice, threads are not only gathered in warps but also in thread blocks. A
+In practice, threads are gathered not only in warps but also in thread blocks. A
thread block is executed by only one SM and it cannot migrate. The typical size of
-a thread block is a number power of two (for example: 64, 128, 256 or 512).
+a thread block is a power of two (for example, 64, 128, 256, or 512).
In this case, without changing anything inside a CUDA code, it is possible to
-run your code with a small CUDA device or the most performing Tesla CUDA cards.
+run code with a small CUDA device or the best performing Tesla CUDA cards.
Blocks are executed in any order depending on the number of SMs available. So
-the programmer must conceive its code having this issue in mind. This
+the programmer must conceive code having this issue in mind. This
independence between thread blocks provides the scalability of CUDA codes.
-\begin{figure}[b!]
-\centerline{\includegraphics[scale=0.65]{Chapters/chapter1/figures/scalability.pdf}}
-\caption{Scalability of GPU.}
-\label{ch1:fig:scalability}
-\end{figure}
+
A kernel is a function which contains a block of instructions that are executed
-by the threads of a GPU. When the problem considered is a two dimensional or three
-dimensional problem, it is possible to group thread blocks into a grid. In
-practice, the number of thread blocks and the size of thread blocks is given as
+by the threads of a GPU. When the problem considered is a two-dimensional or three-dimensional problem, it is possible to group thread blocks into a grid. In
+practice, the number of thread blocks and the size of thread blocks are given as
parameters to each kernel. Figure~\ref{ch1:fig:scalability} illustrates an
example of a kernel composed of 8 thread blocks. Then this kernel is executed on
a small device containing only 2 SMs. So in this case, blocks are executed 2
by 2 in any order. If the kernel is executed on a larger CUDA device containing
4 SMs, blocks are executed 4 by 4 simultaneously. The execution times should be
approximately twice faster in the latter case. Of course, that depends on other
-parameters that will be described later.
+parameters that will be described later (in this chapter and other chapters).
+
-Thread blocks provide a way to cooperation in the sense that threads of the same
+Thread blocks provide a way to cooperate in the sense that threads of the same
block cooperatively load and store blocks of memory they all
use. Synchronizations of threads in the same block are possible (but not between
threads of different blocks). Threads of the same block can also share results
-in order to compute a single result. In chapter~\ref{chapter2}, some examples
-will explicit that.
+in order to compute a single result. In Chapter~\ref{chapter2}, some examples
+will explain that.
\section{Memory hierarchy}
-The memory hierarchy of GPUs\index{memory~hierarchy} is different from the CPUs
-one. In practice, there are registers\index{memory~hierarchy!registers}, local
-memory\index{memory~hierarchy!local~memory}, shared
-memory\index{memory~hierarchy!shared~memory}, cache
-memory\index{memory~hierarchy!cache~memory} and global
-memory\index{memory~hierarchy!global~memory}.
+The memory hierarchy of GPUs\index{memory hierarchy} is different from that of CPUs. In practice, there are registers\index{memory hierarchy!registers}, local
+memory\index{memory hierarchy!local memory}, shared
+memory\index{memory hierarchy!shared memory}, cache
+memory\index{memory hierarchy!cache memory}, and global
+memory\index{memory hierarchy!global memory}.
As previously mentioned each thread can access its own registers. It is
Likewise each thread can access local memory which, in practice, is much slower
than registers. Local memory is automatically used by the compiler when all the
registers are occupied. So the best idea is to optimize the use of registers
-even if this implies to reduce the number of threads per block.
+even if this involves reducing the number of threads per block.
\begin{figure}[hbtp!]
\centerline{\includegraphics[scale=0.60]{Chapters/chapter1/figures/memory_hierarchy.pdf}}
Shared memory allows cooperation between threads of the same block. This kind
-of memory is fast because it requires to be manipulated manually and its size is
-limited. It is accessible during the execution of a kernel. So the principle is
+of memory is fast but it needs to be manipulated manually and its size is
+limited. It is accessible during the execution of a kernel. So the idea is
to fill the shared memory at the start of the kernel with global data that are
-used very frequently, then threads can access it for their computation. They
+used very frequently, then threads can access it for their computation. Threads
can obviously change the content of this shared memory either with computation
-or load of other data and they can store its content in the global memory. So
-shared memory can be seen as a cache memory manageable manually. This requires
-obviously an effort from the programmer.
+or by loading other data and they can store its content in the global memory. So
+shared memory can be seen as a cache memory manageable manually. This
+obviously requires an effort from the programmer.
On recent cards, the programmer may decide what amount of cache memory and
-shared memory is attributed to a kernel. The cache memory is a L1 cache which is
+shared memory is attributed to a kernel. The cache memory is an L1 cache which is
directly managed by the GPU. Sometimes, this cache provides very efficient
result and sometimes the use of shared memory is a better solution.
Figure~\ref{ch1:fig:memory_hierarchy} illustrates the memory hierarchy of a
-GPU. Threads are represented on the top of the figure. They can access to their
-own registers and their local memory. Threads of the same block can access to
-the shared memory of this block. The cache memory is not represented here but it
-is local to a thread. Then each block can access to the global memory of the
+GPU. Threads are represented on the top of the figure. They can have access to their
+own registers and their local memory. Threads of the same block can access
+the shared memory of that block. The cache memory is not represented here but it
+is local to a thread. Then each block can access the global memory of the
GPU.
\section{Conclusion}
In this chapter, a brief presentation of the video card, which has later been
used to perform computation, has been given. The architecture of a GPU has been
-illustrated focusing on the particularity of GPUs in term of parallelism, memory
-latency and threads. In order to design an efficient algorithm for GPU, it is
-essential to have all these parameters in mind.
-
-%%http://people.maths.ox.ac.uk/gilesm/pp10/lec2_2x2.pdf
-%%https://people.maths.ox.ac.uk/erban/papers/paperCUDA.pdf
-%%http://forum.wttsnxt.com/my_forum/viewtopic.php?f=5&t=9519
-%%http://www.cs.nyu.edu/manycores/cuda_many_cores.pdf
-%%http://www.cc.gatech.edu/~vetter/keeneland/tutorial-2011-04-14/02-cuda-overview.pdf
-%%http://people.maths.ox.ac.uk/~gilesm/cuda/
+illustrated focusing on the particularity of GPUs in terms of parallelism, memory
+latency, and threads. In order to design an efficient algorithm for GPU, it is
+essential to keep all these parameters in mind.
\putbib[Chapters/chapter1/biblio]
