CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.
-
Upload
noel-berry -
Category
Documents
-
view
218 -
download
0
Transcript of CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.
![Page 1: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/1.jpg)
CUDA in a Nutshell
Roger Hoang
![Page 2: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/2.jpg)
But first…
• Cluster and Lab Details• Contact Information
![Page 3: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/3.jpg)
Zoidberg
• 2 node Rocks cluster + head node– 1 Tesla C2050
• Full-speed double precision• Fermi architecture• CUDA Compute Capability 2.0
• From the outside world…– vrpad.cse.unr.edu– Port 22222
![Page 4: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/4.jpg)
HPCVIS
• Harris’s lab in SEM248• Every machine there has at least one CUDA
(1.0 <= X < 2.0) capable card in it– Easier to debug– Probably hard to find an open machine…– That < 2.0 means certain features are missing
• Certain idiosyncrasies are different (Warp size, caching, memory access)
![Page 5: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/5.jpg)
User Accounts
• More in-depth tutorial-esque junk starts next time
• Need a user ID from each of you• Also need your e-mail
![Page 6: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/6.jpg)
About Me
• Most likely in SEM248 if I’m on campus– Usually on campus from [10AM-12PM] until [4PM-
8PM] every weekday• Mobile: 775-342-4168• E-mail/Gtalk: [email protected]• AIM: rvhoang
![Page 7: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/7.jpg)
Introduction to CUDA
• Super brief history lesson on GPGPU• CUDA
A lot of these slides can be found on the companion site for the book.http://www.elsevierdirect.com/companion.jsp?ISBN=9780123814722
![Page 8: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/8.jpg)
In the beginning…
or as far back as I care to remember.
![Page 9: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/9.jpg)
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE408, University of Illinois, Urbana-Champaign
Host
Vertex ControlVertex Cache
VS/T&L
Triangle Setup
Raster
Shader
ROP
FBI
TextureCache Frame
BufferMemory
CPU
GPUHost Interface
A Fixed Function GPU Pipeline
![Page 10: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/10.jpg)
10
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE408, University of Illinois, Urbana-Champaign
3D Applicationor Game
3D API:OpenGL or Direct3D
ProgrammableVertex
Processor
PrimitiveAssembly
Rasterization & Interpolation
3D API Commands
Transformed Vertices
Assembled Polygons, Lines, and
Points
GPU Command &
Data Stream
ProgrammableFragmentProcessor
RasterizedPre-transformed
Fragments
TransformedFragments
RasterOperation
s
Framebuffer
Pixel UpdatesGPU
Front End
Pre-transformed Vertices
Vertex Index Stream
Pixel Location Stream
CPU – GPU Boundary
CPU
GPU
An example of separate vertex processor and fragment processor in a programmable graphics pipeline
Programmable Vertex and Pixel Processors
![Page 11: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/11.jpg)
11
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
What is (Historical) GPGPU ?
• General Purpose computation using GPU and graphics API in applications other than 3D graphics– GPU accelerates critical path of application
• Data parallel algorithms leverage GPU attributes– Large data arrays, streaming throughput– Fine-grain SIMD parallelism– Low-latency floating point (FP) computation
• Applications – see //GPGPU.org– Game effects (FX) physics, image processing– Physical modeling, computational engineering, matrix algebra,
convolution, correlation, sorting
![Page 12: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/12.jpg)
12
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
Previous GPGPU Constraints• Dealing with graphics API
– Working with the corner cases of the graphics API
• Addressing modes– Limited texture size/dimension
• Shader capabilities– Limited outputs
• Instruction sets– Lack of Integer & bit ops
• Communication limited– Between pixels– Scatter a[i] = p
![Page 13: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/13.jpg)
13
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE408, University of Illinois, Urbana-Champaign
Input Registers
Fragment Program
Output Registers
Constants
Texture
Temp Registers
per threadper Shaderper Context
FB Memory
The restricted input and output capabilities of a shader programming model.
![Page 14: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/14.jpg)
14
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE408, University of Illinois, Urbana-Champaign
L2
FB
SP SP
L1
TF
Th
rea
d P
roc
es
so
r
Vtx Thread Issue
Setup / Rstr / ZCull
Geom Thread Issue Pixel Thread Issue
Data Assembler
Host
SP SP
L1
TF
SP SP
L1
TF
SP SP
L1
TF
SP SP
L1
TF
SP SP
L1
TF
SP SP
L1
TF
SP SP
L1
TF
L2
FB
L2
FB
L2
FB
L2
FB
L2
FB
Unified Graphics Pipeline
![Page 15: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/15.jpg)
15
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
CUDA
• “Compute Unified Device Architecture”• General purpose programming model
– User kicks off batches of threads on the GPU– GPU = dedicated super-threaded, massively data parallel co-processor
• Targeted software stack– Compute oriented drivers, language, and tools
• Driver for loading computation programs into GPU– Standalone Driver - Optimized for computation – Interface designed for compute – graphics-free API– Data sharing with OpenGL buffer objects – Guaranteed maximum download & readback speeds– Explicit GPU memory management
![Page 16: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/16.jpg)
16© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010ECE 498AL, University of Illinois, Urbana-Champaign
Parallel Computing on a GPU
• 8-series GPUs deliver 25 to 200+ GFLOPSon compiled parallel C applications– Available in laptops, desktops, and clusters
• GPU parallelism is doubling every year• Programming model scales transparently
• Programmable in C with CUDA tools• Multithreaded SPMD model uses application
data parallelism and thread parallelism
GeForce 8800
Tesla S870
Tesla D870
![Page 17: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/17.jpg)
17
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
Overview
• CUDA programming model – basic concepts and data types
• CUDA application programming interface - basic
• Simple examples to illustrate basic concepts and functionalities
• Performance features will be covered later
![Page 18: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/18.jpg)
18© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010ECE 498AL, University of Illinois, Urbana-Champaign
CUDA – C with no shader limitations!• Integrated host+device app C program
– Serial or modestly parallel parts in host C code– Highly parallel parts in device SPMD kernel C code
Serial Code (host)
. . .
. . .
Parallel Kernel (device)
KernelA<<< nBlk, nTid >>>(args);
Serial Code (host)
Parallel Kernel (device)
KernelB<<< nBlk, nTid >>>(args);
![Page 19: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/19.jpg)
19
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
CUDA Devices and Threads• A compute device
– Is a coprocessor to the CPU or host– Has its own DRAM (device memory)– Runs many threads in parallel– Is typically a GPU but can also be another type of parallel processing
device
• Data-parallel portions of an application are expressed as device kernels which run on many threads
• Differences between GPU and CPU threads – GPU threads are extremely lightweight
• Very little creation overhead
– GPU needs 1000s of threads for full efficiency• Multi-core CPU needs only a few
![Page 20: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/20.jpg)
20
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
L2
FB
SP SP
L1
TF
Th
rea
d P
roc
es
so
r
Vtx Thread Issue
Setup / Rstr / ZCull
Geom Thread Issue Pixel Thread Issue
Input Assembler
Host
SP SP
L1
TF
SP SP
L1
TF
SP SP
L1
TF
SP SP
L1
TF
SP SP
L1
TF
SP SP
L1
TF
SP SP
L1
TF
L2
FB
L2
FB
L2
FB
L2
FB
L2
FB
• The future of GPUs is programmable processing• So – build the architecture around the processor
G80 – Graphics Mode
![Page 21: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/21.jpg)
21
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
G80 CUDA mode – A Device Example• Processors execute computing threads• New operating mode/HW interface for
computing
Load/store
Global Memory
Thread Execution Manager
Input Assembler
Host
Texture Texture Texture Texture Texture Texture Texture TextureTexture
Parallel DataCache
Parallel DataCache
Parallel DataCache
Parallel DataCache
Parallel DataCache
Parallel DataCache
Parallel DataCache
Parallel DataCache
Load/store Load/store Load/store Load/store Load/store
![Page 22: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/22.jpg)
22
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
Extended C• Declspecs
– global, device, shared, local, constant
• Keywords– threadIdx, blockIdx
• Intrinsics– __syncthreads
• Runtime API– Memory, symbol,
execution management
• Function launch
__device__ float filter[N];
__global__ void convolve (float *image) {
__shared__ float region[M]; ...
region[threadIdx] = image[i];
__syncthreads() ...
image[j] = result;}
// Allocate GPU memoryvoid *myimage = cudaMalloc(bytes)
// 100 blocks, 10 threads per blockconvolve<<<100, 10>>> (myimage);
![Page 23: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/23.jpg)
23
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
Arrays of Parallel Threads
• A CUDA kernel is executed by an array of threads– All threads run the same code (SPMD)– Each thread has an ID that it uses to compute
memory addresses and make control decisions
76543210
…float x = input[threadID];float y = func(x);output[threadID] = y;…
threadID
![Page 24: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/24.jpg)
24
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
…float x = input[threadID];float y = func(x);output[threadID] = y;…
threadID
Thread Block 0
……float x = input[threadID];float y = func(x);output[threadID] = y;…
Thread Block 1
…float x = input[threadID];float y = func(x);output[threadID] = y;…
Thread Block N - 1
Thread Blocks: Scalable Cooperation• Divide monolithic thread array into multiple
blocks– Threads within a block cooperate via shared memory,
atomic operations and barrier synchronization– Threads in different blocks cannot cooperate
76543210 76543210 76543210
![Page 25: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/25.jpg)
25© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010ECE 498AL, University of Illinois, Urbana-Champaign
Host
Kernel 1
Kernel 2
Device
Grid 1
Block(0, 0)
Block(1, 0)
Block(0, 1)
Block(1, 1)
Grid 2
Courtesy: NDVIA
Figure 3.2. An Example of CUDA Thread Organization.
Block (1, 1)
Thread(0,1,0)
Thread(1,1,0)
Thread(2,1,0)
Thread(3,1,0)
Thread(0,0,0)
Thread(1,0,0)
Thread(2,0,0)
Thread(3,0,0)
(0,0,1) (1,0,1) (2,0,1) (3,0,1)
Block IDs and Thread IDs
• Each thread uses IDs to decide what data to work on– Block ID: 1D or 2D– Thread ID: 1D, 2D, or 3D
• Simplifies memoryaddressing when processingmultidimensional data– Image processing– Solving PDEs on volumes– …
![Page 26: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/26.jpg)
26
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
CUDA Memory Model Overview
• Global memory– Main means of
communicating R/W Data between host and device
– Contents visible to all threads
– Long latency access• We will focus on global
memory for now– Constant and texture
memory will come later
Grid
Global Memory
Block (0, 0)
Shared Memory
Thread (0, 0)
Registers
Thread (1, 0)
Registers
Block (1, 0)
Shared Memory
Thread (0, 0)
Registers
Thread (1, 0)
Registers
Host
![Page 27: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/27.jpg)
27
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
CUDA API Highlights:Easy and Lightweight
• The API is an extension to the ANSI C programming language Low learning curve
• The hardware is designed to enable lightweight runtime and driver High performance
![Page 28: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/28.jpg)
28© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010ECE 498AL, University of Illinois, Urbana-Champaign
CUDA Device Memory Allocation• cudaMalloc()
– Allocates object in the device Global Memory
– Requires two parameters• Address of a pointer to the
allocated object• Size of of allocated object
• cudaFree()– Frees object from device
Global Memory• Pointer to freed object
Grid
GlobalMemory
Block (0, 0)
Shared Memory
Thread (0, 0)
Registers
Thread (1, 0)
Registers
Block (1, 0)
Shared Memory
Thread (0, 0)
Registers
Thread (1, 0)
Registers
Host
![Page 29: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/29.jpg)
29© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010ECE 498AL, University of Illinois, Urbana-Champaign
CUDA Device Memory Allocation (cont.)
• Code example: – Allocate a 64 * 64 single precision float array– Attach the allocated storage to Md– “d” is often used to indicate a device data
structure
TILE_WIDTH = 64;Float* Mdint size = TILE_WIDTH * TILE_WIDTH * sizeof(float);
cudaMalloc((void**)&Md, size);cudaFree(Md);
![Page 30: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/30.jpg)
30
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
CUDA Host-Device Data Transfer• cudaMemcpy()
– memory data transfer– Requires four parameters
• Pointer to destination • Pointer to source• Number of bytes copied• Type of transfer
– Host to Host– Host to Device– Device to Host– Device to Device
• Asynchronous transfer
Grid
GlobalMemory
Block (0, 0)
Shared Memory
Thread (0, 0)
Registers
Thread (1, 0)
Registers
Block (1, 0)
Shared Memory
Thread (0, 0)
Registers
Thread (1, 0)
Registers
Host
![Page 31: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/31.jpg)
31© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010ECE 498AL, University of Illinois, Urbana-Champaign
CUDA Host-Device Data Transfer(cont.)
• Code example: – Transfer a 64 * 64 single precision float array– M is in host memory and Md is in device memory– cudaMemcpyHostToDevice and
cudaMemcpyDeviceToHost are symbolic constants
cudaMemcpy(Md, M, size, cudaMemcpyHostToDevice);
cudaMemcpy(M, Md, size, cudaMemcpyDeviceToHost);
![Page 32: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/32.jpg)
32© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010ECE 498AL, University of Illinois, Urbana-Champaign
CUDA Keywords
![Page 33: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/33.jpg)
33© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010ECE 498AL, University of Illinois, Urbana-Champaign
CUDA Function Declarations
hosthost__host__ float HostFunc()
hostdevice__global__ void KernelFunc()
devicedevice__device__ float DeviceFunc()
Only callable from the:
Executed on the:
• __global__ defines a kernel function– Must return void
• __device__ and __host__ can be used together
![Page 34: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/34.jpg)
34
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
CUDA Function Declarations (cont.)
• __device__ functions cannot have their address taken
• For functions executed on the device:– No recursion– No static variable declarations inside the function– No variable number of arguments
![Page 35: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/35.jpg)
35
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
Calling a Kernel Function – Thread Creation
• A kernel function must be called with an execution configuration:
__global__ void KernelFunc(...);
dim3 DimGrid(100, 50); // 5000 thread blocks
dim3 DimBlock(4, 8, 8); // 256 threads per block
size_t SharedMemBytes = 64; // 64 bytes of shared memory
KernelFunc<<< DimGrid, DimBlock, SharedMemBytes >>>(...);
• Any call to a kernel function is asynchronous from CUDA 1.0 on, explicit synch needed for blocking
![Page 36: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/36.jpg)
36
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
A Simple Running ExampleMatrix Multiplication
• A simple matrix multiplication example that illustrates the basic features of memory and thread management in CUDA programs– Leave shared memory usage until later– Local, register usage– Thread ID usage– Memory data transfer API between host and
device– Assume square matrix for simplicity
![Page 37: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/37.jpg)
37
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
Programming Model:Square Matrix Multiplication Example
• P = M * N of size WIDTH x WIDTH
• Without tiling:– One thread calculates one element
of P– M and N are loaded WIDTH times
from global memoryM
N
P
WID
TH
WID
TH
WIDTH WIDTH
![Page 38: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/38.jpg)
38
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
M2,0
M1,1
M1,0M0,0
M0,1
M3,0
M2,1 M3,1
Memory Layout of a Matrix in C
M2,0M1,0M0,0 M3,0 M1,1M0,1 M2,1 M3,1 M1,2M0,2 M2,2 M3,2
M1,2M0,2 M2,2 M3,2
M1,3M0,3 M2,3 M3,3
M1,3M0,3 M2,3 M3,3
M
![Page 39: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/39.jpg)
39
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
Step 1: Matrix MultiplicationA Simple Host Version in C
M
N
P
WID
TH
WID
TH
WIDTH WIDTH
// Matrix multiplication on the (CPU) host in double precisionvoid MatrixMulOnHost(float* M, float* N, float* P, int Width){ for (int i = 0; i < Width; ++i) for (int j = 0; j < Width; ++j) { double sum = 0; for (int k = 0; k < Width; ++k) { double a = M[i * width + k]; double b = N[k * width + j]; sum += a * b; } P[i * Width + j] = sum; }}
i
k
k
j
![Page 40: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/40.jpg)
40
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
void MatrixMulOnDevice(float* M, float* N, float* P, int Width)
{ int size = Width * Width * sizeof(float); float* Md, Nd, Pd; …
1. // Allocate and Load M, N to device memory cudaMalloc(&Md, size); cudaMemcpy(Md, M, size, cudaMemcpyHostToDevice);
cudaMalloc(&Nd, size); cudaMemcpy(Nd, N, size, cudaMemcpyHostToDevice);
// Allocate P on the device cudaMalloc(&Pd, size);
Step 2: Input Matrix Data Transfer(Host-side Code)
![Page 41: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/41.jpg)
41
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
Step 3: Output Matrix Data Transfer(Host-side Code)
2. // Kernel invocation code – to be shown later …
3. // Read P from the device cudaMemcpy(P, Pd, size, cudaMemcpyDeviceToHost);
// Free device matrices cudaFree(Md); cudaFree(Nd); cudaFree (Pd); }
![Page 42: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/42.jpg)
42
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
Step 4: Kernel Function
// Matrix multiplication kernel – per thread code
__global__ void MatrixMulKernel(float* Md, float* Nd, float* Pd, int Width)
{ // Pvalue is used to store the element of the matrix // that is computed by the thread float Pvalue = 0;
![Page 43: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/43.jpg)
43
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
Nd
Md Pd
WID
TH
WID
TH
WIDTH WIDTH
Step 4: Kernel Function (cont.)
for (int k = 0; k < Width; ++k) { float Melement = Md[threadIdx.y*Width+k]; float Nelement = Nd[k*Width+threadIdx.x]; Pvalue += Melement * Nelement; }
Pd[threadIdx.y*Width+threadIdx.x] = Pvalue;}
ty
tx
ty
tx
k
k
![Page 44: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/44.jpg)
44
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
// Setup the execution configuration
dim3 dimGrid(1, 1); dim3 dimBlock(Width, Width);
// Launch the device computation threads! MatrixMulKernel<<<dimGrid, dimBlock>>>(Md, Nd, Pd, Width);
Step 5: Kernel Invocation(Host-side Code)
![Page 45: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/45.jpg)
45© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010ECE 498AL, University of Illinois, Urbana-Champaign
Only One Thread Block Used• One Block of threads compute
matrix Pd– Each thread computes one
element of Pd• Each thread
– Loads a row of matrix Md– Loads a column of matrix Nd– Perform one multiply and
addition for each pair of Md and Nd elements
– Compute to off-chip memory access ratio close to 1:1 (not very high)
• Size of matrix limited by the number of threads allowed in a thread block
Grid 1
Block 1
3 2 5 4
2
4
2
6
48
Thread(2, 2)
WIDTH
Md Pd
Nd
![Page 46: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/46.jpg)
46
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
Step 7: Handling Arbitrary Sized Square Matrices (will cover later)
• Have each 2D thread block to compute a (TILE_WIDTH)2 sub-matrix (tile) of the result matrix– Each has (TILE_WIDTH)2 threads
• Generate a 2D Grid of (WIDTH/TILE_WIDTH)2 blocksMd
Nd
Pd
WID
TH
WID
TH
WIDTH WIDTH
ty
tx
by
bx
You still need to put a loop around the kernel call for cases where WIDTH/TILE_WIDTH is greater than max grid size (64K)!
TILE_WIDTH
![Page 47: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/47.jpg)
47© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010ECE 498AL, University of Illinois, Urbana-Champaign
Some Useful Information on Tools
![Page 48: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/48.jpg)
48
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
Compiling a CUDA Program
NVCC
C/C++ CUDAApplication
PTX to TargetCompiler
G80 … GPU
Target code
PTX CodeVirtual
Physical
CPU Code
• Parallel Thread eXecution (PTX)
– Virtual Machine and ISA
– Programming model
– Execution resources and state
float4 me = gx[gtid];me.x += me.y * me.z;
ld.global.v4.f32 {$f1,$f3,$f5,$f7}, [$r9+0];mad.f32 $f1, $f5, $f3, $f1;
![Page 49: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/49.jpg)
49
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
Compilation
• Any source file containing CUDA language extensions must be compiled with NVCC
• NVCC is a compiler driver– Works by invoking all the necessary tools and
compilers like cudacc, g++, cl, ...
• NVCC outputs:– C code (host CPU Code)
• Must then be compiled with the rest of the application using another tool
– PTX• Object code directly• Or, PTX source, interpreted at runtime
![Page 50: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/50.jpg)
50
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
Linking
• Any executable with CUDA code requires two dynamic libraries:– The CUDA runtime library (cudart)– The CUDA core library (cuda)
![Page 51: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/51.jpg)
51
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
Debugging Using theDevice Emulation Mode
• An executable compiled in device emulation mode (nvcc -deviceemu) runs completely on the host using the CUDA runtime– No need of any device and CUDA driver– Each device thread is emulated with a host thread
• Running in device emulation mode, one can:– Use host native debug support (breakpoints, inspection, etc.)– Access any device-specific data from host code and vice-versa– Call any host function from device code (e.g. printf) and vice-versa– Detect deadlock situations caused by improper usage of
__syncthreads
![Page 52: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/52.jpg)
52
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
Device Emulation Mode Pitfalls• Emulated device threads execute sequentially,
so simultaneous accesses of the same memory location by multiple threads could produce different results.
• Dereferencing device pointers on the host or host pointers on the device can produce correct results in device emulation mode, but will generate an error in device execution mode
![Page 53: CUDA in a Nutshell Roger Hoang. But first… Cluster and Lab Details Contact Information.](https://reader034.fdocuments.in/reader034/viewer/2022051401/56649eb25503460f94bb9aec/html5/thumbnails/53.jpg)
53
© David Kirk/NVIDIA and Wen-mei W. Hwu, 2007-2010
ECE 498AL, University of Illinois, Urbana-Champaign
Floating Point
• Results of floating-point computations will slightly differ because of:– Different compiler outputs, instruction sets– Use of extended precision for intermediate results
• There are various options to force strict single precision on the host