CUDA Parallel Programming Tutorial

Richard Membarth [e-mail]

Description

CUDA Parallel Programming TutorialRichard [e-mail]-Software-Co-DesignUniversity of Erlangen-Nuremberg19.03.2009Friedrich-Alexander University of Erlangen-NurembergRichard Membarth1Outline◮ Tasks for CUDA◮ CUDA programming model◮ Getting started◮ Example codesFriedrich-Alexander University of Erlangen-NurembergRichard Membarth2Tasks for CUDA◮ Provide ability to run code on GPU◮ Manage resources◮ Partition data to fit on cores◮ Schedule blocks to coresFriedrich-Alexander University of Erlangen-NurembergRichard Membarth3Data Partitioning◮ Partition data in smallerblocks that can be processedby one core◮ Up to 512 threads in oneblock◮ All blocks define the grid◮ All blocks execute sameprogram (kernel)◮ Independent blocks◮ Only ONE kernel at a timeFriedrich-Alexander University of Erlangen-NurembergRichard Membarth4Memory HierarchyMemory types (fastest memoryfirst):◮ Registers◮ Shared memory◮ Device memory (texture,constant, local, global)Friedrich-Alexander University of Erlangen-NurembergRichard Membarth5Tesla Architecture◮ 30 cores, 240 ALUs (1 mul-add)◮ (1 mul-add + 1 mul): 240 * (2+1) * 1.3 GHz = 936 GFLOPS◮ 4.0 GB GDDR3, 102 GB/s Mem BW, 4GB/s PCIe BW to CPUFriedrich-Alexander University of Erlangen-NurembergRichard Membarth6CUDA: Extended C◮ Function qualifiers◮ Variable qualifiers◮ Built-in keywords◮ Intrinsics◮ Function callsFriedrich-Alexander University of ...

Informations

• Publié par: Obbi
• Nombre de lectures: 34
• Langue: English

Extrait

CUDA Parallel Programming Tutorial

Richard Membarth

[e-mail]

Hardware-Software-Co-Design University of Erlangen-Nuremberg

Friedrich-Alexander University of Erlangen-Nuremberg Richard Membarth

19.03.2009

1

Outline

◮

◮

◮

◮

Tasks for CUDA

CUDA programming

Getting started

Example codes

Friedrich-Alexander University of Erlangen-Nuremberg Richard Membarth

model

2

Tasks for CUDA

◮

◮

◮

◮

Provide ability to run code on

Manage resources

Partition data to t on cores

Schedule blocks to cores

Friedrich-Alexander University of Erlangen-Nuremberg Richard Membarth

GPU

3

Data Partitionin

◮

◮

◮

◮

◮

◮

Partition data in smaller blocks that can be processed by one core

Up to 512 threads in one block

All blocks dene the grid

All blocks execute same program (kernel)

Independent blocks

Only ONE kernel at a time

Friedrich-Alexander University of Erlangen-Nuremberg Richard Membarth

4

Memor

Hierarch

Memory types (fastest memory rst):

◮ Registers

◮ Shared memory

◮ Device memory (texture, constant, local, global)

Friedrich-Alexander University of Erlangen-Nuremberg Richard Membarth

5

Tesla Architecture

◮

◮

◮

30 cores, 240 ALUs (1 mul-add)

(1 mul-add + 1 mul): 240 * (2+1) * 1.3 GHz = 936 GFLOPS

4.0 GB GDDR3, 102 GB/s Mem BW, 4GB/s PCIe BW to CPU

Friedrich-Alexander University of Erlangen-Nuremberg Richard Membarth

6

CUDA: Extended

◮

◮

◮

◮

◮

Function qualiers

Variable qualiers

Built-in keywords

Intrinsics

Function calls

Friedrich-Alexander University of Erlangen-Nuremberg Richard Membarth

C

7

Function Qualiers

◮ Functions: device , host , global

__ __ global void f i l t e r ( int * in, int ... }

◮ Default: host ◮ No function pointers ◮ No recursion ◮ No static variables ◮ No variable number of arguments ◮ No return value

Friedrich-Alexander University of Erlangen-Nuremberg Richard Membarth

* out )

{

8

Variable Qualiers

◮

◮

Variables: device , constant ,

shared

c o n s t a n t float  ;t r i x [ 1 0 ]m a _ _ _ _ = {1 . 0 f , ... } s h a r e d int [ 3 2 ] [ 2 ] ; _ _ _ _

Default: Variables reside in registers

Friedrich-Alexander University of Erlangen-Nuremberg Richard Membarth

9

Built-In Variables

◮

◮

◮

◮

◮

Available inside of kernel code

Thread index within current block:

threadIdx.x , threadIdx.y , threadIdx.z

Block index within grid:

blockIdx.x , blockIdx.y

Dimension of grid, block:

gridDim.x , gridDim.y blockDim.x , blockDim.y blockDim.z

Warp size:

warpSize

Friedrich-Alexander University of Erlangen-Nuremberg Richard Membarth

,

10

Intrinsics

◮ void syncthreads(); __ ◮ Synchronizes in all thread of current block

◮ Use in conditional code may lead to deadlocks

◮ Intrinsics for most mathematical functions exists, e.g.

sinf(x), cosf(x), expf(x), ... __ __ __ ◮ Texture functions

Friedrich-Alexander University of Erlangen-Nuremberg Richard Membarth

11